From a273b77783612a28e8974ff430efa6d6e2e7b9c2 Mon Sep 17 00:00:00 2001 From: Morten Hjorth-Jensen Date: Sun, 3 Sep 2023 21:02:06 +0200 Subject: [PATCH] update --- .../2023/Project1/html/._Project1-bs000.html | 724 ++++++++++++++ .../2023/Project1/html/Project1-bs.html | 724 ++++++++++++++ doc/Projects/2023/Project1/html/Project1.html | 747 +++++++++++++++ .../2023/Project1/ipynb/Project1.ipynb | 882 ++++++++++++++++++ .../Project1/ipynb/ipynb-Project1-src.tar.gz | Bin 0 -> 193 bytes doc/Projects/2023/Project1/pdf/Project1.p.tex | 678 ++++++++++++++ doc/Projects/2023/Project1/pdf/Project1.pdf | Bin 0 -> 269564 bytes doc/Projects/2023/Project1/pdf/Project1.tex | 646 +++++++++++++ .../Projects/2023/Project1/Project1.do.txt | 130 ++- doc/src/Projects/2023/Project1/make.sh | 2 +- 10 files changed, 4466 insertions(+), 67 deletions(-) create mode 100644 doc/Projects/2023/Project1/html/._Project1-bs000.html create mode 100644 doc/Projects/2023/Project1/html/Project1-bs.html create mode 100644 doc/Projects/2023/Project1/html/Project1.html create mode 100644 doc/Projects/2023/Project1/ipynb/Project1.ipynb create mode 100644 doc/Projects/2023/Project1/ipynb/ipynb-Project1-src.tar.gz create mode 100644 doc/Projects/2023/Project1/pdf/Project1.p.tex create mode 100644 doc/Projects/2023/Project1/pdf/Project1.pdf create mode 100644 doc/Projects/2023/Project1/pdf/Project1.tex diff --git a/doc/Projects/2023/Project1/html/._Project1-bs000.html b/doc/Projects/2023/Project1/html/._Project1-bs000.html new file mode 100644 index 000000000..e4b893262 --- /dev/null +++ b/doc/Projects/2023/Project1/html/._Project1-bs000.html @@ -0,0 +1,724 @@ + + + + + + + +Project 1 on Machine Learning, deadline October 7, 2023 + + + + + + + + + + + + + + + + + + +
+
+ + Project 1 on Machine Learning, deadline October 7, 2023 +
+
+ +
+
+ +
+

 

 

 

+ + +
+
+

Project 1 on Machine Learning, deadline October 7, 2023

+
+ + +
+Data Analysis and Machine Learning FYS-STK3155/FYS4155 +
+ +
+University of Oslo, Norway +
+
+
+

September 3

+
+
+ + +
+

Regression analysis and resampling methods

+ +

The main aim of this project is to study in more detail various +regression methods, including the Ordinary Least Squares (OLS) method. +In addition to the scientific part, in this course we want also to +give you an experience in writing scientific reports. The format for +the delivery of your answers is namely that of a scientific report. At +for example +https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/EvaluationGrading/EvaluationForm.md +we detail how to write a report. Furthermore, at +https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/ReportExample/ +you can find examples of previous reports. How to write reports will +also be discussed during lectures and at the various lab sessions. +

+ +

A small recommendation when developing the codes here. Instead of +jumping on to the two-dimensional function described below, we +recommend to do the code development and testing with a simpler +one-dimensional function, similar to those discussed in the exercises +of week 35. A simple test, as discussed during the lectures the first +two weeks is to set the design matrix equal to the identity +matrix. Then your model should give a mean square error which is exactly equal to zero. +When you are sure that your codes function well, you can then replace +the one-dimensional test function with the two-dimensional Franke function +discussed here. +

+ +

The Franke function serves as a stepping stone towards the analysis of +real topographic data. The latter is the last part of this project. +

+

Description of two-dimensional function

+ +

We will first study how to fit polynomials to a specific +two-dimensional function called Franke's +function. This +is a function which has been widely used when testing various +interpolation and fitting algorithms. Furthermore, after having +established the model and the method, we will employ resamling +techniques such as cross-validation and/or bootstrap in order to perform a +proper assessment of our models. We will also study in detail the +so-called Bias-Variance trade off. +

+ +

The Franke function, which is a weighted sum of four exponentials reads as follows

+$$ +\begin{align*} +f(x,y) &= \frac{3}{4}\exp{\left(-\frac{(9x-2)^2}{4} - \frac{(9y-2)^2}{4}\right)}+\frac{3}{4}\exp{\left(-\frac{(9x+1)^2}{49}- \frac{(9y+1)}{10}\right)} \\ +&+\frac{1}{2}\exp{\left(-\frac{(9x-7)^2}{4} - \frac{(9y-3)^2}{4}\right)} -\frac{1}{5}\exp{\left(-(9x-4)^2 - (9y-7)^2\right) }. +\end{align*} +$$ + +

The function will be defined for \( x,y\in [0,1] \). Our first step will +be to perform an OLS regression analysis of this function, trying out +a polynomial fit with an \( x \) and \( y \) dependence of the form \( [x, y, +x^2, y^2, xy, \dots] \). We will also include bootstrap first as a +resampling technique. After that we will include the cross-validation +technique. As discussed in the lectures for weeks 35 and 36,, we can +use a uniform distribution to set up the arrays of values for \( x \) and +\( y \), or as in the example below just a set of fixed values for \( x \) and +\( y \) with a given step size. We will fit a function (for example a +polynomial) of \( x \) and \( y \). Thereafter we will repeat much of the +same procedure using the Ridge and Lasso regression methods, +introducing thus a dependence on the bias (penalty) \( \lambda \). +

+ +

Finally we are going to use (real) digital terrain data and try to +reproduce these data using the same methods. We will also try to go +beyond the second-order polynomials metioned above and explore +which polynomial fits the data best. +

+ +

The Python code for the Franke function is included here (it performs also a three-dimensional plot of it)

+ + +
+
+
+
+
+ 
from mpl_toolkits.mplot3d import Axes3D
+import matplotlib.pyplot as plt
+from matplotlib import cm
+from matplotlib.ticker import LinearLocator, FormatStrFormatter
+import numpy as np
+from random import random, seed
+
+fig = plt.figure()
+ax = fig.gca(projection='3d')
+
+# Make data.
+x = np.arange(0, 1, 0.05)
+y = np.arange(0, 1, 0.05)
+x, y = np.meshgrid(x,y)
+
+
+def FrankeFunction(x,y):
+    term1 = 0.75*np.exp(-(0.25*(9*x-2)**2) - 0.25*((9*y-2)**2))
+    term2 = 0.75*np.exp(-((9*x+1)**2)/49.0 - 0.1*(9*y+1))
+    term3 = 0.5*np.exp(-(9*x-7)**2/4.0 - 0.25*((9*y-3)**2))
+    term4 = -0.2*np.exp(-(9*x-4)**2 - (9*y-7)**2)
+    return term1 + term2 + term3 + term4
+
+
+z = FrankeFunction(x, y)
+
+# Plot the surface.
+surf = ax.plot_surface(x, y, z, cmap=cm.coolwarm,
+                       linewidth=0, antialiased=False)
+
+# Customize the z axis.
+ax.set_zlim(-0.10, 1.40)
+ax.zaxis.set_major_locator(LinearLocator(10))
+ax.zaxis.set_major_formatter(FormatStrFormatter('%.02f'))
+
+# Add a color bar which maps values to colors.
+fig.colorbar(surf, shrink=0.5, aspect=5)
+
+plt.show()
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+

Part a) : Ordinary Least Square (OLS) on the Franke function

+ +

We will generate our own dataset for a function +\( \mathrm{FrankeFunction}(x,y) \) with \( x,y \in [0,1] \). The function +\( f(x,y) \) is the Franke function. You should explore also the addition +of an added stochastic noise to this function using the normal +distribution \( N(0,1) \). +

+ +

Write your own code (using either a matrix inversion or a singular +value decomposition from e.g., numpy ) and perform a standard ordinary least square regression +analysis using polynomials in \( x \) and \( y \) up to fifth order. +

+ +

Evaluate the mean Squared error (MSE)

+ +$$ MSE(\boldsymbol{y},\tilde{\boldsymbol{y}}) = \frac{1}{n} +\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +$$ + +

and the \( R^2 \) score function. If \( \tilde{\boldsymbol{y}}_i \) is the predicted +value of the \( i-th \) sample and \( y_i \) is the corresponding true value, +then the score \( R^2 \) is defined as +

+ +$$ +R^2(\boldsymbol{y}, \tilde{\boldsymbol{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +$$ + +

where we have defined the mean value of \( \boldsymbol{y} \) as

+ +$$ +\bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +$$ + +

Plot the resulting scores (MSE and R$^2$) as functions of the polynomial degree (here up to polymial degree five). +Plot also the parameters \( \beta \) as you increase the order of the polynomial. Comment your results. +

+ +

Your code has to include a scaling/centering of the data (for example by +subtracting the mean value), and +a split of the data in training and test data. For this exercise you can +either write your own code or use for example the function for +splitting training data provided by the library Scikit-Learn (make +sure you have installed it). This function is called +\( train\_test\_split \). You should present a critical discussion of why and how you have scaled or not scaled the data. +

+ +

It is normal in essentially all Machine Learning studies to split the +data in a training set and a test set (eventually also an additional +validation set). There +is no explicit recipe for how much data should be included as training +data and say test data. An accepted rule of thumb is to use +approximately \( 2/3 \) to \( 4/5 \) of the data as training data. +

+ +

You can easily reuse the solutions to your exercises from week 35 and week 36. +See also the lecture slides from week 35 and week 36. +

+

Part b): Adding Ridge and Lasso Regression on the Franke function

+ +

Write your own code for the Ridge method, either using matrix +inversion or the singular value decomposition as done in the previous +exercise. +

+ +

Perform the same analysis as you did in the previous exercise but now for different values of \( \lambda \). Compare and +analyze your results with those obtained in parts b-d). Study the +dependence on \( \lambda \). +

+ +

This exercise is essentially a repeat of the previous two ones, but now +with Lasso regression. Write either your own code (difficult and optional) or, in this case, +you can also use the functionalities of Scikit-Learn (recommended). +Give a +critical discussion of the three methods and a judgement of which +model fits the data best. Perform here as well an analysis of the bias-variance trade-off using the bootstrap resampling technique and an analysis of the mean squared error using cross-validation. +

+

Part a): Paper and pencil part

+ +

This exercise deals with various mean values and variances in linear regression method (here it may be useful to look up chapter 3, equation (3.8) of Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer).

+ +

The assumption we have made is +that there exists a continuous function \( f(\boldsymbol{x}) \) and a normal distributed error \( \boldsymbol{\varepsilon}\sim N(0, \sigma^2) \) +which describes our data +

+$$ +\boldsymbol{y} = f(\boldsymbol{x})+\boldsymbol{\varepsilon} +$$ + +

We then approximate this function \( f(\boldsymbol{x}) \) with our model \( \boldsymbol{\tilde{y}} \) from the solution of the linear regression equations (ordinary least squares OLS), that is our +function \( f \) is approximated by \( \boldsymbol{\tilde{y}} \) where we minimized \( (\boldsymbol{y}-\boldsymbol{\tilde{y}})^2 \), with +

+$$ +\boldsymbol{\tilde{y}} = \boldsymbol{X}\boldsymbol{\beta}. +$$ + +

The matrix \( \boldsymbol{X} \) is the so-called design or feature matrix.

+ +

Show that the expectation value of \( \boldsymbol{y} \) for a given element \( i \)

+$$ +\mathbb{E}(y_i) =\sum_{j}x_{ij} \beta_j=\mathbf{X}_{i, \ast} \, \boldsymbol{\beta}, +$$ + +

and that +its variance is +

+$$ +\mbox{Var}(y_i) = \sigma^2. +$$ + +

Hence, \( y_i \sim N( \mathbf{X}_{i, \ast} \, \boldsymbol{\beta}, \sigma^2) \), that is \( \boldsymbol{y} \) follows a normal distribution with +mean value \( \boldsymbol{X}\boldsymbol{\beta} \) and variance \( \sigma^2 \). +

+ +

With the OLS expressions for the optimal parameters \( \boldsymbol{\hat{\beta}} \) show that

+$$ +\mathbb{E}(\boldsymbol{\hat{\beta}}) = \boldsymbol{\beta}. +$$ + +

Show finally that the variance of \( \boldsymbol{\beta} \) is

+$$ +\mbox{Var}(\boldsymbol{\hat{\beta}}) = \sigma^2 \, (\mathbf{X}^{T} \mathbf{X})^{-1}. +$$ + +

We can use the last expression when we define a so-called confidence interval for the parameters \( \beta \). . +A given parameter \( \beta_j \) is given by the diagonal matrix element of the above matrix. +

+

Part c): Bias-variance trade-off and resampling techniques

+ +

Our aim here is to study the bias-variance trade-off by implementing the bootstrap resampling technique.

+ +

With a code which does OLS and includes resampling techniques, +we will now discuss the bias-variance trade-off in the context of +continuous predictions such as regression. However, many of the +intuitions and ideas discussed here also carry over to classification +tasks and basically all Machine Learning algorithms. +

+ +

Before you perform an analysis of the bias-variance trade-off on your test data, make +first a figure similar to Fig. 2.11 of Hastie, Tibshirani, and +Friedman. Figure 2.11 of this reference displays only the test and training MSEs. The test MSE can be used to +indicate possible regions of low/high bias and variance. You will most likely not get an +equally smooth curve! +

+ +

With this result we move on to the bias-variance trade-off analysis.

+ +

Consider a +dataset \( \mathcal{L} \) consisting of the data +\( \mathbf{X}_\mathcal{L}=\{(y_j, \boldsymbol{x}_j), j=0\ldots n-1\} \). +

+ +

As in part a), we assume that the true data is generated from a noisy model

+ +$$ +\boldsymbol{y}=f(\boldsymbol{x}) + \boldsymbol{\epsilon}. +$$ + +

Here \( \epsilon \) is normally distributed with mean zero and standard +deviation \( \sigma^2 \). +

+ +

In our derivation of the ordinary least squares method we defined then +an approximation to the function \( f \) in terms of the parameters +\( \boldsymbol{\beta} \) and the design matrix \( \boldsymbol{X} \) which embody our model, +that is \( \boldsymbol{\tilde{y}}=\boldsymbol{X}\boldsymbol{\beta} \). +

+ +

The parameters \( \boldsymbol{\beta} \) are in turn found by optimizing the mean +squared error via the so-called cost function +

+ +$$ +C(\boldsymbol{X},\boldsymbol{\beta}) =\frac{1}{n}\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2=\mathbb{E}\left[(\boldsymbol{y}-\boldsymbol{\tilde{y}})^2\right]. +$$ + +

Here the expected value \( \mathbb{E} \) is the sample value.

+ +

Show that you can rewrite this in terms of a term which contains the variance of the model itself (the so-called variance term), a +term which measures the deviation from the true data and the mean value of the model (the bias term) and finally the variance of the noise. +That is, show that +

+$$ +\mathbb{E}\left[(\boldsymbol{y}-\boldsymbol{\tilde{y}})^2\right]=(\mathrm{Bias}[\tilde{y}])^2+\mathrm{var}[\tilde{f}]+\sigma^2, +$$ + +

with

+$$ +(\mathrm{Bias}[\tilde{y}])^2=\left(\boldsymbol{y}-\mathbb{E}\left[\boldsymbol{\tilde{y}}\right]\right)^2, +$$ + +

and

+$$ +\mathrm{var}[\tilde{f}]=\frac{1}{n}\sum_i(\tilde{y}_i-\mathbb{E}\left[\boldsymbol{\tilde{y}}\right])^2. +$$ + +

The answer to this exercise should be included in the theory part of the report. +Explain what the terms mean and discuss their interpretations. +

+ +

Perform then a bias-variance analysis of the Franke function by +studying the MSE value as function of the complexity of your model. +

+ +

Discuss the bias and variance trade-off as function +of your model complexity (the degree of the polynomial) and the number +of data points, and possibly also your training and test data using the bootstrap resampling method. +You can follow the code example in the jupyter-book at https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter3.html#the-bias-variance-tradeoff. +

+

Part d): Cross-validation as resampling techniques, adding more complexity

+ +

The aim here is to write your own code for another widely popular +resampling technique, the so-called cross-validation method. Again, +before you start with cross-validation approach, you should scale your +data if you think this is needed. +

+ +

Implement the \( k \)-fold cross-validation algorithm (write your own +code) and evaluate again the MSE function resulting +from the test folds. You can compare your own code with that from +Scikit-Learn if needed. +

+ +

Compare the MSE you get from your cross-validation code with the one +you got from your bootstrap code. Comment your results. Try \( 5-10 \) +folds. You can also compare your own cross-validation code with the +one provided by Scikit-Learn. +

+

Part g): Analysis of real data

+ +

With our codes functioning and having been tested properly on a +simpler function we are now ready to look at real data. We will +essentially repeat in this exercise what was done in exercises 1-5. However, we +need first to download the data and prepare properly the inputs to our +codes. We are going to download digital terrain data from the website +https://earthexplorer.usgs.gov/, +

+ +

Or, if you prefer, we have placed selected datafiles at https://github.com/CompPhysics/MachineLearning/tree/master/doc/Projects/2022/Project1/DataFiles

+ +

In order to obtain data for a specific region, you need to register as +a user (free) at this website and then decide upon which area you want +to fetch the digital terrain data from. In order to be able to read +the data properly, you need to specify that the format should be SRTM +Arc-Second Global and download the data as a GeoTIF file. The +files are then stored in tif format which can be imported into a +Python program using +

+ + + +
+
+
+
+
+ 
scipy.misc.imread
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +

Here is a simple part of a Python code which reads and plots the data +from such files +

+ + + +
+
+
+
+
+ 
import numpy as np
+from imageio import imread
+import matplotlib.pyplot as plt
+from mpl_toolkits.mplot3d import Axes3D
+from matplotlib import cm
+
+# Load the terrain
+terrain1 = imread('SRTM_data_Norway_1.tif')
+# Show the terrain
+plt.figure()
+plt.title('Terrain over Norway 1')
+plt.imshow(terrain1, cmap='gray')
+plt.xlabel('X')
+plt.ylabel('Y')
+plt.show()
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +

If you should have problems in downloading the digital terrain data, +we provide two examples under the data folder of project 1. One is +from a region close to Stavanger in Norway and the other Møsvatn +Austfjell, again in Norway. +Feel free to produce your own terrain data. +

+ +

Alternatively, if you would like to use another data set, feel free to do so. This could be data close to your reseach area or simply a data set you found interesting. See for example kaggle.com for examples.

+ +

Our final part deals with the parameterization of your digital terrain +data (or your own data). We will apply all three methods for linear regression, the same type (or higher order) of polynomial +approximation and cross-validation as resampling technique to evaluate which +model fits the data best. +

+ +

At the end, you should present a critical evaluation of your results +and discuss the applicability of these regression methods to the type +of data presented here (either the terrain data we propose or other data sets). +

+

Background literature

+ +
    +
  1. For a discussion and derivation of the variances and mean squared errors using linear regression, see the Lecture notes on ridge regression by Wessel N. van Wieringen
    2. +
  2. The textbook of Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer, chapters 3 and 7 are the most relevant ones for the analysis here.
    3. +
+

Introduction to numerical projects

+ +

Here follows a brief recipe and recommendation on how to answer the various questions when preparing your answers.

+ + +

Format for electronic delivery of report and programs

+ +

The preferred format for the report is a PDF file. You can also use DOC or postscript formats or as an ipython notebook file. As programming language we prefer that you choose between C/C++, Fortran2008, Julia or Python. The following prescription should be followed when preparing the report:

+ + +

Finally, +we encourage you to collaborate. Optimal working groups consist of +2-3 students. You can then hand in a common report. +

+

Software and needed installations

+ +

If you have Python installed (we recommend Python3) and you feel pretty familiar with installing different packages, +we recommend that you install the following Python packages via pip as +

+
    +
  1. pip install numpy scipy matplotlib ipython scikit-learn tensorflow sympy pandas pillow
    2. +
+

For Python3, replace pip with pip3.

+ +

See below for a discussion of tensorflow and scikit-learn.

+ +

For OSX users we recommend also, after having installed Xcode, to install brew. Brew allows +for a seamless installation of additional software via for example +

+
    +
  1. brew install python3
    2. +
+

For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution +you can use pip as well and simply install Python as +

+
    +
  1. sudo apt-get install python3 (or python for python2.7)
    2. +
+

etc etc.

+ +

If you don't want to install various Python packages with their dependencies separately, we recommend two widely used distrubutions which set up all relevant dependencies for Python, namely

+
    +
  1. Anaconda Anaconda is an open source distribution of the Python and R programming languages for large-scale data processing, predictive analytics, and scientific computing, that aims to simplify package management and deployment. Package versions are managed by the package management system conda
    2. +
  2. Enthought canopy is a Python distribution for scientific and analytic computing distribution and analysis environment, available for free and under a commercial license.
    3. +
+

Popular software packages written in Python for ML are

+ + +

These are all freely available at their respective GitHub sites. They +encompass communities of developers in the thousands or more. And the number +of code developers and contributors keeps increasing. +

+ +

+ +

+ +
+ + + + + + + diff --git a/doc/Projects/2023/Project1/html/Project1-bs.html b/doc/Projects/2023/Project1/html/Project1-bs.html new file mode 100644 index 000000000..e4b893262 --- /dev/null +++ b/doc/Projects/2023/Project1/html/Project1-bs.html @@ -0,0 +1,724 @@ + + + + + + + +Project 1 on Machine Learning, deadline October 7, 2023 + + + + + + + + + + + + + + + + + + +
+
+ + Project 1 on Machine Learning, deadline October 7, 2023 +
+
+ +
+
+ +
+

 

 

 

+ + +
+
+

Project 1 on Machine Learning, deadline October 7, 2023

+
+ + +
+Data Analysis and Machine Learning FYS-STK3155/FYS4155 +
+ +
+University of Oslo, Norway +
+
+
+

September 3

+
+
+ + +
+

Regression analysis and resampling methods

+ +

The main aim of this project is to study in more detail various +regression methods, including the Ordinary Least Squares (OLS) method. +In addition to the scientific part, in this course we want also to +give you an experience in writing scientific reports. The format for +the delivery of your answers is namely that of a scientific report. At +for example +https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/EvaluationGrading/EvaluationForm.md +we detail how to write a report. Furthermore, at +https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/ReportExample/ +you can find examples of previous reports. How to write reports will +also be discussed during lectures and at the various lab sessions. +

+ +

A small recommendation when developing the codes here. Instead of +jumping on to the two-dimensional function described below, we +recommend to do the code development and testing with a simpler +one-dimensional function, similar to those discussed in the exercises +of week 35. A simple test, as discussed during the lectures the first +two weeks is to set the design matrix equal to the identity +matrix. Then your model should give a mean square error which is exactly equal to zero. +When you are sure that your codes function well, you can then replace +the one-dimensional test function with the two-dimensional Franke function +discussed here. +

+ +

The Franke function serves as a stepping stone towards the analysis of +real topographic data. The latter is the last part of this project. +

+

Description of two-dimensional function

+ +

We will first study how to fit polynomials to a specific +two-dimensional function called Franke's +function. This +is a function which has been widely used when testing various +interpolation and fitting algorithms. Furthermore, after having +established the model and the method, we will employ resamling +techniques such as cross-validation and/or bootstrap in order to perform a +proper assessment of our models. We will also study in detail the +so-called Bias-Variance trade off. +

+ +

The Franke function, which is a weighted sum of four exponentials reads as follows

+$$ +\begin{align*} +f(x,y) &= \frac{3}{4}\exp{\left(-\frac{(9x-2)^2}{4} - \frac{(9y-2)^2}{4}\right)}+\frac{3}{4}\exp{\left(-\frac{(9x+1)^2}{49}- \frac{(9y+1)}{10}\right)} \\ +&+\frac{1}{2}\exp{\left(-\frac{(9x-7)^2}{4} - \frac{(9y-3)^2}{4}\right)} -\frac{1}{5}\exp{\left(-(9x-4)^2 - (9y-7)^2\right) }. +\end{align*} +$$ + +

The function will be defined for \( x,y\in [0,1] \). Our first step will +be to perform an OLS regression analysis of this function, trying out +a polynomial fit with an \( x \) and \( y \) dependence of the form \( [x, y, +x^2, y^2, xy, \dots] \). We will also include bootstrap first as a +resampling technique. After that we will include the cross-validation +technique. As discussed in the lectures for weeks 35 and 36,, we can +use a uniform distribution to set up the arrays of values for \( x \) and +\( y \), or as in the example below just a set of fixed values for \( x \) and +\( y \) with a given step size. We will fit a function (for example a +polynomial) of \( x \) and \( y \). Thereafter we will repeat much of the +same procedure using the Ridge and Lasso regression methods, +introducing thus a dependence on the bias (penalty) \( \lambda \). +

+ +

Finally we are going to use (real) digital terrain data and try to +reproduce these data using the same methods. We will also try to go +beyond the second-order polynomials metioned above and explore +which polynomial fits the data best. +

+ +

The Python code for the Franke function is included here (it performs also a three-dimensional plot of it)

+ + +
+
+
+
+
+ 
from mpl_toolkits.mplot3d import Axes3D
+import matplotlib.pyplot as plt
+from matplotlib import cm
+from matplotlib.ticker import LinearLocator, FormatStrFormatter
+import numpy as np
+from random import random, seed
+
+fig = plt.figure()
+ax = fig.gca(projection='3d')
+
+# Make data.
+x = np.arange(0, 1, 0.05)
+y = np.arange(0, 1, 0.05)
+x, y = np.meshgrid(x,y)
+
+
+def FrankeFunction(x,y):
+    term1 = 0.75*np.exp(-(0.25*(9*x-2)**2) - 0.25*((9*y-2)**2))
+    term2 = 0.75*np.exp(-((9*x+1)**2)/49.0 - 0.1*(9*y+1))
+    term3 = 0.5*np.exp(-(9*x-7)**2/4.0 - 0.25*((9*y-3)**2))
+    term4 = -0.2*np.exp(-(9*x-4)**2 - (9*y-7)**2)
+    return term1 + term2 + term3 + term4
+
+
+z = FrankeFunction(x, y)
+
+# Plot the surface.
+surf = ax.plot_surface(x, y, z, cmap=cm.coolwarm,
+                       linewidth=0, antialiased=False)
+
+# Customize the z axis.
+ax.set_zlim(-0.10, 1.40)
+ax.zaxis.set_major_locator(LinearLocator(10))
+ax.zaxis.set_major_formatter(FormatStrFormatter('%.02f'))
+
+# Add a color bar which maps values to colors.
+fig.colorbar(surf, shrink=0.5, aspect=5)
+
+plt.show()
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+

Part a) : Ordinary Least Square (OLS) on the Franke function

+ +

We will generate our own dataset for a function +\( \mathrm{FrankeFunction}(x,y) \) with \( x,y \in [0,1] \). The function +\( f(x,y) \) is the Franke function. You should explore also the addition +of an added stochastic noise to this function using the normal +distribution \( N(0,1) \). +

+ +

Write your own code (using either a matrix inversion or a singular +value decomposition from e.g., numpy ) and perform a standard ordinary least square regression +analysis using polynomials in \( x \) and \( y \) up to fifth order. +

+ +

Evaluate the mean Squared error (MSE)

+ +$$ MSE(\boldsymbol{y},\tilde{\boldsymbol{y}}) = \frac{1}{n} +\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +$$ + +

and the \( R^2 \) score function. If \( \tilde{\boldsymbol{y}}_i \) is the predicted +value of the \( i-th \) sample and \( y_i \) is the corresponding true value, +then the score \( R^2 \) is defined as +

+ +$$ +R^2(\boldsymbol{y}, \tilde{\boldsymbol{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +$$ + +

where we have defined the mean value of \( \boldsymbol{y} \) as

+ +$$ +\bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +$$ + +

Plot the resulting scores (MSE and R$^2$) as functions of the polynomial degree (here up to polymial degree five). +Plot also the parameters \( \beta \) as you increase the order of the polynomial. Comment your results. +

+ +

Your code has to include a scaling/centering of the data (for example by +subtracting the mean value), and +a split of the data in training and test data. For this exercise you can +either write your own code or use for example the function for +splitting training data provided by the library Scikit-Learn (make +sure you have installed it). This function is called +\( train\_test\_split \). You should present a critical discussion of why and how you have scaled or not scaled the data. +

+ +

It is normal in essentially all Machine Learning studies to split the +data in a training set and a test set (eventually also an additional +validation set). There +is no explicit recipe for how much data should be included as training +data and say test data. An accepted rule of thumb is to use +approximately \( 2/3 \) to \( 4/5 \) of the data as training data. +

+ +

You can easily reuse the solutions to your exercises from week 35 and week 36. +See also the lecture slides from week 35 and week 36. +

+

Part b): Adding Ridge and Lasso Regression on the Franke function

+ +

Write your own code for the Ridge method, either using matrix +inversion or the singular value decomposition as done in the previous +exercise. +

+ +

Perform the same analysis as you did in the previous exercise but now for different values of \( \lambda \). Compare and +analyze your results with those obtained in parts b-d). Study the +dependence on \( \lambda \). +

+ +

This exercise is essentially a repeat of the previous two ones, but now +with Lasso regression. Write either your own code (difficult and optional) or, in this case, +you can also use the functionalities of Scikit-Learn (recommended). +Give a +critical discussion of the three methods and a judgement of which +model fits the data best. Perform here as well an analysis of the bias-variance trade-off using the bootstrap resampling technique and an analysis of the mean squared error using cross-validation. +

+

Part a): Paper and pencil part

+ +

This exercise deals with various mean values and variances in linear regression method (here it may be useful to look up chapter 3, equation (3.8) of Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer).

+ +

The assumption we have made is +that there exists a continuous function \( f(\boldsymbol{x}) \) and a normal distributed error \( \boldsymbol{\varepsilon}\sim N(0, \sigma^2) \) +which describes our data +

+$$ +\boldsymbol{y} = f(\boldsymbol{x})+\boldsymbol{\varepsilon} +$$ + +

We then approximate this function \( f(\boldsymbol{x}) \) with our model \( \boldsymbol{\tilde{y}} \) from the solution of the linear regression equations (ordinary least squares OLS), that is our +function \( f \) is approximated by \( \boldsymbol{\tilde{y}} \) where we minimized \( (\boldsymbol{y}-\boldsymbol{\tilde{y}})^2 \), with +

+$$ +\boldsymbol{\tilde{y}} = \boldsymbol{X}\boldsymbol{\beta}. +$$ + +

The matrix \( \boldsymbol{X} \) is the so-called design or feature matrix.

+ +

Show that the expectation value of \( \boldsymbol{y} \) for a given element \( i \)

+$$ +\mathbb{E}(y_i) =\sum_{j}x_{ij} \beta_j=\mathbf{X}_{i, \ast} \, \boldsymbol{\beta}, +$$ + +

and that +its variance is +

+$$ +\mbox{Var}(y_i) = \sigma^2. +$$ + +

Hence, \( y_i \sim N( \mathbf{X}_{i, \ast} \, \boldsymbol{\beta}, \sigma^2) \), that is \( \boldsymbol{y} \) follows a normal distribution with +mean value \( \boldsymbol{X}\boldsymbol{\beta} \) and variance \( \sigma^2 \). +

+ +

With the OLS expressions for the optimal parameters \( \boldsymbol{\hat{\beta}} \) show that

+$$ +\mathbb{E}(\boldsymbol{\hat{\beta}}) = \boldsymbol{\beta}. +$$ + +

Show finally that the variance of \( \boldsymbol{\beta} \) is

+$$ +\mbox{Var}(\boldsymbol{\hat{\beta}}) = \sigma^2 \, (\mathbf{X}^{T} \mathbf{X})^{-1}. +$$ + +

We can use the last expression when we define a so-called confidence interval for the parameters \( \beta \). . +A given parameter \( \beta_j \) is given by the diagonal matrix element of the above matrix. +

+

Part c): Bias-variance trade-off and resampling techniques

+ +

Our aim here is to study the bias-variance trade-off by implementing the bootstrap resampling technique.

+ +

With a code which does OLS and includes resampling techniques, +we will now discuss the bias-variance trade-off in the context of +continuous predictions such as regression. However, many of the +intuitions and ideas discussed here also carry over to classification +tasks and basically all Machine Learning algorithms. +

+ +

Before you perform an analysis of the bias-variance trade-off on your test data, make +first a figure similar to Fig. 2.11 of Hastie, Tibshirani, and +Friedman. Figure 2.11 of this reference displays only the test and training MSEs. The test MSE can be used to +indicate possible regions of low/high bias and variance. You will most likely not get an +equally smooth curve! +

+ +

With this result we move on to the bias-variance trade-off analysis.

+ +

Consider a +dataset \( \mathcal{L} \) consisting of the data +\( \mathbf{X}_\mathcal{L}=\{(y_j, \boldsymbol{x}_j), j=0\ldots n-1\} \). +

+ +

As in part a), we assume that the true data is generated from a noisy model

+ +$$ +\boldsymbol{y}=f(\boldsymbol{x}) + \boldsymbol{\epsilon}. +$$ + +

Here \( \epsilon \) is normally distributed with mean zero and standard +deviation \( \sigma^2 \). +

+ +

In our derivation of the ordinary least squares method we defined then +an approximation to the function \( f \) in terms of the parameters +\( \boldsymbol{\beta} \) and the design matrix \( \boldsymbol{X} \) which embody our model, +that is \( \boldsymbol{\tilde{y}}=\boldsymbol{X}\boldsymbol{\beta} \). +

+ +

The parameters \( \boldsymbol{\beta} \) are in turn found by optimizing the mean +squared error via the so-called cost function +

+ +$$ +C(\boldsymbol{X},\boldsymbol{\beta}) =\frac{1}{n}\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2=\mathbb{E}\left[(\boldsymbol{y}-\boldsymbol{\tilde{y}})^2\right]. +$$ + +

Here the expected value \( \mathbb{E} \) is the sample value.

+ +

Show that you can rewrite this in terms of a term which contains the variance of the model itself (the so-called variance term), a +term which measures the deviation from the true data and the mean value of the model (the bias term) and finally the variance of the noise. +That is, show that +

+$$ +\mathbb{E}\left[(\boldsymbol{y}-\boldsymbol{\tilde{y}})^2\right]=(\mathrm{Bias}[\tilde{y}])^2+\mathrm{var}[\tilde{f}]+\sigma^2, +$$ + +

with

+$$ +(\mathrm{Bias}[\tilde{y}])^2=\left(\boldsymbol{y}-\mathbb{E}\left[\boldsymbol{\tilde{y}}\right]\right)^2, +$$ + +

and

+$$ +\mathrm{var}[\tilde{f}]=\frac{1}{n}\sum_i(\tilde{y}_i-\mathbb{E}\left[\boldsymbol{\tilde{y}}\right])^2. +$$ + +

The answer to this exercise should be included in the theory part of the report. +Explain what the terms mean and discuss their interpretations. +

+ +

Perform then a bias-variance analysis of the Franke function by +studying the MSE value as function of the complexity of your model. +

+ +

Discuss the bias and variance trade-off as function +of your model complexity (the degree of the polynomial) and the number +of data points, and possibly also your training and test data using the bootstrap resampling method. +You can follow the code example in the jupyter-book at https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter3.html#the-bias-variance-tradeoff. +

+

Part d): Cross-validation as resampling techniques, adding more complexity

+ +

The aim here is to write your own code for another widely popular +resampling technique, the so-called cross-validation method. Again, +before you start with cross-validation approach, you should scale your +data if you think this is needed. +

+ +

Implement the \( k \)-fold cross-validation algorithm (write your own +code) and evaluate again the MSE function resulting +from the test folds. You can compare your own code with that from +Scikit-Learn if needed. +

+ +

Compare the MSE you get from your cross-validation code with the one +you got from your bootstrap code. Comment your results. Try \( 5-10 \) +folds. You can also compare your own cross-validation code with the +one provided by Scikit-Learn. +

+

Part g): Analysis of real data

+ +

With our codes functioning and having been tested properly on a +simpler function we are now ready to look at real data. We will +essentially repeat in this exercise what was done in exercises 1-5. However, we +need first to download the data and prepare properly the inputs to our +codes. We are going to download digital terrain data from the website +https://earthexplorer.usgs.gov/, +

+ +

Or, if you prefer, we have placed selected datafiles at https://github.com/CompPhysics/MachineLearning/tree/master/doc/Projects/2022/Project1/DataFiles

+ +

In order to obtain data for a specific region, you need to register as +a user (free) at this website and then decide upon which area you want +to fetch the digital terrain data from. In order to be able to read +the data properly, you need to specify that the format should be SRTM +Arc-Second Global and download the data as a GeoTIF file. The +files are then stored in tif format which can be imported into a +Python program using +

+ + + +
+
+
+
+
+ 
scipy.misc.imread
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +

Here is a simple part of a Python code which reads and plots the data +from such files +

+ + + +
+
+
+
+
+ 
import numpy as np
+from imageio import imread
+import matplotlib.pyplot as plt
+from mpl_toolkits.mplot3d import Axes3D
+from matplotlib import cm
+
+# Load the terrain
+terrain1 = imread('SRTM_data_Norway_1.tif')
+# Show the terrain
+plt.figure()
+plt.title('Terrain over Norway 1')
+plt.imshow(terrain1, cmap='gray')
+plt.xlabel('X')
+plt.ylabel('Y')
+plt.show()
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +

If you should have problems in downloading the digital terrain data, +we provide two examples under the data folder of project 1. One is +from a region close to Stavanger in Norway and the other Møsvatn +Austfjell, again in Norway. +Feel free to produce your own terrain data. +

+ +

Alternatively, if you would like to use another data set, feel free to do so. This could be data close to your reseach area or simply a data set you found interesting. See for example kaggle.com for examples.

+ +

Our final part deals with the parameterization of your digital terrain +data (or your own data). We will apply all three methods for linear regression, the same type (or higher order) of polynomial +approximation and cross-validation as resampling technique to evaluate which +model fits the data best. +

+ +

At the end, you should present a critical evaluation of your results +and discuss the applicability of these regression methods to the type +of data presented here (either the terrain data we propose or other data sets). +

+

Background literature

+ +
    +
  1. For a discussion and derivation of the variances and mean squared errors using linear regression, see the Lecture notes on ridge regression by Wessel N. van Wieringen
    2. +
  2. The textbook of Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer, chapters 3 and 7 are the most relevant ones for the analysis here.
    3. +
+

Introduction to numerical projects

+ +

Here follows a brief recipe and recommendation on how to answer the various questions when preparing your answers.

+ + +

Format for electronic delivery of report and programs

+ +

The preferred format for the report is a PDF file. You can also use DOC or postscript formats or as an ipython notebook file. As programming language we prefer that you choose between C/C++, Fortran2008, Julia or Python. The following prescription should be followed when preparing the report:

+ + +

Finally, +we encourage you to collaborate. Optimal working groups consist of +2-3 students. You can then hand in a common report. +

+

Software and needed installations

+ +

If you have Python installed (we recommend Python3) and you feel pretty familiar with installing different packages, +we recommend that you install the following Python packages via pip as +

+
    +
  1. pip install numpy scipy matplotlib ipython scikit-learn tensorflow sympy pandas pillow
    2. +
+

For Python3, replace pip with pip3.

+ +

See below for a discussion of tensorflow and scikit-learn.

+ +

For OSX users we recommend also, after having installed Xcode, to install brew. Brew allows +for a seamless installation of additional software via for example +

+
    +
  1. brew install python3
    2. +
+

For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution +you can use pip as well and simply install Python as +

+
    +
  1. sudo apt-get install python3 (or python for python2.7)
    2. +
+

etc etc.

+ +

If you don't want to install various Python packages with their dependencies separately, we recommend two widely used distrubutions which set up all relevant dependencies for Python, namely

+
    +
  1. Anaconda Anaconda is an open source distribution of the Python and R programming languages for large-scale data processing, predictive analytics, and scientific computing, that aims to simplify package management and deployment. Package versions are managed by the package management system conda
    2. +
  2. Enthought canopy is a Python distribution for scientific and analytic computing distribution and analysis environment, available for free and under a commercial license.
    3. +
+

Popular software packages written in Python for ML are

+ + +

These are all freely available at their respective GitHub sites. They +encompass communities of developers in the thousands or more. And the number +of code developers and contributors keeps increasing. +

+ +

+ +

+ +
+ + + + + + + diff --git a/doc/Projects/2023/Project1/html/Project1.html b/doc/Projects/2023/Project1/html/Project1.html new file mode 100644 index 000000000..3e45c97e0 --- /dev/null +++ b/doc/Projects/2023/Project1/html/Project1.html @@ -0,0 +1,747 @@ + + + + + + + +Project 1 on Machine Learning, deadline October 7, 2023 + + + + + + + + + + + + + + +
+

Project 1 on Machine Learning, deadline October 7, 2023

+
+ + +
+Data Analysis and Machine Learning FYS-STK3155/FYS4155 +
+ +
+University of Oslo, Norway +
+
+
+

September 3

+
+
+

Regression analysis and resampling methods

+ +

The main aim of this project is to study in more detail various +regression methods, including the Ordinary Least Squares (OLS) method. +In addition to the scientific part, in this course we want also to +give you an experience in writing scientific reports. The format for +the delivery of your answers is namely that of a scientific report. At +for example +https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/EvaluationGrading/EvaluationForm.md +we detail how to write a report. Furthermore, at +https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/ReportExample/ +you can find examples of previous reports. How to write reports will +also be discussed during lectures and at the various lab sessions. +

+ +

A small recommendation when developing the codes here. Instead of +jumping on to the two-dimensional function described below, we +recommend to do the code development and testing with a simpler +one-dimensional function, similar to those discussed in the exercises +of week 35. A simple test, as discussed during the lectures the first +two weeks is to set the design matrix equal to the identity +matrix. Then your model should give a mean square error which is exactly equal to zero. +When you are sure that your codes function well, you can then replace +the one-dimensional test function with the two-dimensional Franke function +discussed here. +

+ +

The Franke function serves as a stepping stone towards the analysis of +real topographic data. The latter is the last part of this project. +

+

Description of two-dimensional function

+ +

We will first study how to fit polynomials to a specific +two-dimensional function called Franke's +function. This +is a function which has been widely used when testing various +interpolation and fitting algorithms. Furthermore, after having +established the model and the method, we will employ resamling +techniques such as cross-validation and/or bootstrap in order to perform a +proper assessment of our models. We will also study in detail the +so-called Bias-Variance trade off. +

+ +

The Franke function, which is a weighted sum of four exponentials reads as follows

+$$ +\begin{align*} +f(x,y) &= \frac{3}{4}\exp{\left(-\frac{(9x-2)^2}{4} - \frac{(9y-2)^2}{4}\right)}+\frac{3}{4}\exp{\left(-\frac{(9x+1)^2}{49}- \frac{(9y+1)}{10}\right)} \\ +&+\frac{1}{2}\exp{\left(-\frac{(9x-7)^2}{4} - \frac{(9y-3)^2}{4}\right)} -\frac{1}{5}\exp{\left(-(9x-4)^2 - (9y-7)^2\right) }. +\end{align*} +$$ + +

The function will be defined for \( x,y\in [0,1] \). Our first step will +be to perform an OLS regression analysis of this function, trying out +a polynomial fit with an \( x \) and \( y \) dependence of the form \( [x, y, +x^2, y^2, xy, \dots] \). We will also include bootstrap first as a +resampling technique. After that we will include the cross-validation +technique. As discussed in the lectures for weeks 35 and 36,, we can +use a uniform distribution to set up the arrays of values for \( x \) and +\( y \), or as in the example below just a set of fixed values for \( x \) and +\( y \) with a given step size. We will fit a function (for example a +polynomial) of \( x \) and \( y \). Thereafter we will repeat much of the +same procedure using the Ridge and Lasso regression methods, +introducing thus a dependence on the bias (penalty) \( \lambda \). +

+ +

Finally we are going to use (real) digital terrain data and try to +reproduce these data using the same methods. We will also try to go +beyond the second-order polynomials metioned above and explore +which polynomial fits the data best. +

+ +

The Python code for the Franke function is included here (it performs also a three-dimensional plot of it)

+ + +
+
+
+
+
+ 
from mpl_toolkits.mplot3d import Axes3D
+import matplotlib.pyplot as plt
+from matplotlib import cm
+from matplotlib.ticker import LinearLocator, FormatStrFormatter
+import numpy as np
+from random import random, seed
+
+fig = plt.figure()
+ax = fig.gca(projection='3d')
+
+# Make data.
+x = np.arange(0, 1, 0.05)
+y = np.arange(0, 1, 0.05)
+x, y = np.meshgrid(x,y)
+
+
+def FrankeFunction(x,y):
+    term1 = 0.75*np.exp(-(0.25*(9*x-2)**2) - 0.25*((9*y-2)**2))
+    term2 = 0.75*np.exp(-((9*x+1)**2)/49.0 - 0.1*(9*y+1))
+    term3 = 0.5*np.exp(-(9*x-7)**2/4.0 - 0.25*((9*y-3)**2))
+    term4 = -0.2*np.exp(-(9*x-4)**2 - (9*y-7)**2)
+    return term1 + term2 + term3 + term4
+
+
+z = FrankeFunction(x, y)
+
+# Plot the surface.
+surf = ax.plot_surface(x, y, z, cmap=cm.coolwarm,
+                       linewidth=0, antialiased=False)
+
+# Customize the z axis.
+ax.set_zlim(-0.10, 1.40)
+ax.zaxis.set_major_locator(LinearLocator(10))
+ax.zaxis.set_major_formatter(FormatStrFormatter('%.02f'))
+
+# Add a color bar which maps values to colors.
+fig.colorbar(surf, shrink=0.5, aspect=5)
+
+plt.show()
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+

Part a) : Ordinary Least Square (OLS) on the Franke function

+ +

We will generate our own dataset for a function +\( \mathrm{FrankeFunction}(x,y) \) with \( x,y \in [0,1] \). The function +\( f(x,y) \) is the Franke function. You should explore also the addition +of an added stochastic noise to this function using the normal +distribution \( N(0,1) \). +

+ +

Write your own code (using either a matrix inversion or a singular +value decomposition from e.g., numpy ) and perform a standard ordinary least square regression +analysis using polynomials in \( x \) and \( y \) up to fifth order. +

+ +

Evaluate the mean Squared error (MSE)

+ +$$ MSE(\boldsymbol{y},\tilde{\boldsymbol{y}}) = \frac{1}{n} +\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +$$ + +

and the \( R^2 \) score function. If \( \tilde{\boldsymbol{y}}_i \) is the predicted +value of the \( i-th \) sample and \( y_i \) is the corresponding true value, +then the score \( R^2 \) is defined as +

+ +$$ +R^2(\boldsymbol{y}, \tilde{\boldsymbol{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +$$ + +

where we have defined the mean value of \( \boldsymbol{y} \) as

+ +$$ +\bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +$$ + +

Plot the resulting scores (MSE and R$^2$) as functions of the polynomial degree (here up to polymial degree five). +Plot also the parameters \( \beta \) as you increase the order of the polynomial. Comment your results. +

+ +

Your code has to include a scaling/centering of the data (for example by +subtracting the mean value), and +a split of the data in training and test data. For this exercise you can +either write your own code or use for example the function for +splitting training data provided by the library Scikit-Learn (make +sure you have installed it). This function is called +\( train\_test\_split \). You should present a critical discussion of why and how you have scaled or not scaled the data. +

+ +

It is normal in essentially all Machine Learning studies to split the +data in a training set and a test set (eventually also an additional +validation set). There +is no explicit recipe for how much data should be included as training +data and say test data. An accepted rule of thumb is to use +approximately \( 2/3 \) to \( 4/5 \) of the data as training data. +

+ +

You can easily reuse the solutions to your exercises from week 35 and week 36. +See also the lecture slides from week 35 and week 36. +

+

Part b): Adding Ridge and Lasso Regression on the Franke function

+ +

Write your own code for the Ridge method, either using matrix +inversion or the singular value decomposition as done in the previous +exercise. +

+ +

Perform the same analysis as you did in the previous exercise but now for different values of \( \lambda \). Compare and +analyze your results with those obtained in parts b-d). Study the +dependence on \( \lambda \). +

+ +

This exercise is essentially a repeat of the previous two ones, but now +with Lasso regression. Write either your own code (difficult and optional) or, in this case, +you can also use the functionalities of Scikit-Learn (recommended). +Give a +critical discussion of the three methods and a judgement of which +model fits the data best. Perform here as well an analysis of the bias-variance trade-off using the bootstrap resampling technique and an analysis of the mean squared error using cross-validation. +

+

Part a): Paper and pencil part

+ +

This exercise deals with various mean values and variances in linear regression method (here it may be useful to look up chapter 3, equation (3.8) of Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer).

+ +

The assumption we have made is +that there exists a continuous function \( f(\boldsymbol{x}) \) and a normal distributed error \( \boldsymbol{\varepsilon}\sim N(0, \sigma^2) \) +which describes our data +

+$$ +\boldsymbol{y} = f(\boldsymbol{x})+\boldsymbol{\varepsilon} +$$ + +

We then approximate this function \( f(\boldsymbol{x}) \) with our model \( \boldsymbol{\tilde{y}} \) from the solution of the linear regression equations (ordinary least squares OLS), that is our +function \( f \) is approximated by \( \boldsymbol{\tilde{y}} \) where we minimized \( (\boldsymbol{y}-\boldsymbol{\tilde{y}})^2 \), with +

+$$ +\boldsymbol{\tilde{y}} = \boldsymbol{X}\boldsymbol{\beta}. +$$ + +

The matrix \( \boldsymbol{X} \) is the so-called design or feature matrix.

+ +

Show that the expectation value of \( \boldsymbol{y} \) for a given element \( i \)

+$$ +\mathbb{E}(y_i) =\sum_{j}x_{ij} \beta_j=\mathbf{X}_{i, \ast} \, \boldsymbol{\beta}, +$$ + +

and that +its variance is +

+$$ +\mbox{Var}(y_i) = \sigma^2. +$$ + +

Hence, \( y_i \sim N( \mathbf{X}_{i, \ast} \, \boldsymbol{\beta}, \sigma^2) \), that is \( \boldsymbol{y} \) follows a normal distribution with +mean value \( \boldsymbol{X}\boldsymbol{\beta} \) and variance \( \sigma^2 \). +

+ +

With the OLS expressions for the optimal parameters \( \boldsymbol{\hat{\beta}} \) show that

+$$ +\mathbb{E}(\boldsymbol{\hat{\beta}}) = \boldsymbol{\beta}. +$$ + +

Show finally that the variance of \( \boldsymbol{\beta} \) is

+$$ +\mbox{Var}(\boldsymbol{\hat{\beta}}) = \sigma^2 \, (\mathbf{X}^{T} \mathbf{X})^{-1}. +$$ + +

We can use the last expression when we define a so-called confidence interval for the parameters \( \beta \). . +A given parameter \( \beta_j \) is given by the diagonal matrix element of the above matrix. +

+

Part c): Bias-variance trade-off and resampling techniques

+ +

Our aim here is to study the bias-variance trade-off by implementing the bootstrap resampling technique.

+ +

With a code which does OLS and includes resampling techniques, +we will now discuss the bias-variance trade-off in the context of +continuous predictions such as regression. However, many of the +intuitions and ideas discussed here also carry over to classification +tasks and basically all Machine Learning algorithms. +

+ +

Before you perform an analysis of the bias-variance trade-off on your test data, make +first a figure similar to Fig. 2.11 of Hastie, Tibshirani, and +Friedman. Figure 2.11 of this reference displays only the test and training MSEs. The test MSE can be used to +indicate possible regions of low/high bias and variance. You will most likely not get an +equally smooth curve! +

+ +

With this result we move on to the bias-variance trade-off analysis.

+ +

Consider a +dataset \( \mathcal{L} \) consisting of the data +\( \mathbf{X}_\mathcal{L}=\{(y_j, \boldsymbol{x}_j), j=0\ldots n-1\} \). +

+ +

As in part a), we assume that the true data is generated from a noisy model

+ +$$ +\boldsymbol{y}=f(\boldsymbol{x}) + \boldsymbol{\epsilon}. +$$ + +

Here \( \epsilon \) is normally distributed with mean zero and standard +deviation \( \sigma^2 \). +

+ +

In our derivation of the ordinary least squares method we defined then +an approximation to the function \( f \) in terms of the parameters +\( \boldsymbol{\beta} \) and the design matrix \( \boldsymbol{X} \) which embody our model, +that is \( \boldsymbol{\tilde{y}}=\boldsymbol{X}\boldsymbol{\beta} \). +

+ +

The parameters \( \boldsymbol{\beta} \) are in turn found by optimizing the mean +squared error via the so-called cost function +

+ +$$ +C(\boldsymbol{X},\boldsymbol{\beta}) =\frac{1}{n}\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2=\mathbb{E}\left[(\boldsymbol{y}-\boldsymbol{\tilde{y}})^2\right]. +$$ + +

Here the expected value \( \mathbb{E} \) is the sample value.

+ +

Show that you can rewrite this in terms of a term which contains the variance of the model itself (the so-called variance term), a +term which measures the deviation from the true data and the mean value of the model (the bias term) and finally the variance of the noise. +That is, show that +

+$$ +\mathbb{E}\left[(\boldsymbol{y}-\boldsymbol{\tilde{y}})^2\right]=(\mathrm{Bias}[\tilde{y}])^2+\mathrm{var}[\tilde{f}]+\sigma^2, +$$ + +

with

+$$ +(\mathrm{Bias}[\tilde{y}])^2=\left(\boldsymbol{y}-\mathbb{E}\left[\boldsymbol{\tilde{y}}\right]\right)^2, +$$ + +

and

+$$ +\mathrm{var}[\tilde{f}]=\frac{1}{n}\sum_i(\tilde{y}_i-\mathbb{E}\left[\boldsymbol{\tilde{y}}\right])^2. +$$ + +

The answer to this exercise should be included in the theory part of the report. +Explain what the terms mean and discuss their interpretations. +

+ +

Perform then a bias-variance analysis of the Franke function by +studying the MSE value as function of the complexity of your model. +

+ +

Discuss the bias and variance trade-off as function +of your model complexity (the degree of the polynomial) and the number +of data points, and possibly also your training and test data using the bootstrap resampling method. +You can follow the code example in the jupyter-book at https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter3.html#the-bias-variance-tradeoff. +

+

Part d): Cross-validation as resampling techniques, adding more complexity

+ +

The aim here is to write your own code for another widely popular +resampling technique, the so-called cross-validation method. Again, +before you start with cross-validation approach, you should scale your +data if you think this is needed. +

+ +

Implement the \( k \)-fold cross-validation algorithm (write your own +code) and evaluate again the MSE function resulting +from the test folds. You can compare your own code with that from +Scikit-Learn if needed. +

+ +

Compare the MSE you get from your cross-validation code with the one +you got from your bootstrap code. Comment your results. Try \( 5-10 \) +folds. You can also compare your own cross-validation code with the +one provided by Scikit-Learn. +

+

Part g): Analysis of real data

+ +

With our codes functioning and having been tested properly on a +simpler function we are now ready to look at real data. We will +essentially repeat in this exercise what was done in exercises 1-5. However, we +need first to download the data and prepare properly the inputs to our +codes. We are going to download digital terrain data from the website +https://earthexplorer.usgs.gov/, +

+ +

Or, if you prefer, we have placed selected datafiles at https://github.com/CompPhysics/MachineLearning/tree/master/doc/Projects/2022/Project1/DataFiles

+ +

In order to obtain data for a specific region, you need to register as +a user (free) at this website and then decide upon which area you want +to fetch the digital terrain data from. In order to be able to read +the data properly, you need to specify that the format should be SRTM +Arc-Second Global and download the data as a GeoTIF file. The +files are then stored in tif format which can be imported into a +Python program using +

+ + + +
+
+
+
+
+ 
scipy.misc.imread
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +

Here is a simple part of a Python code which reads and plots the data +from such files +

+ + + +
+
+
+
+
+ 
import numpy as np
+from imageio import imread
+import matplotlib.pyplot as plt
+from mpl_toolkits.mplot3d import Axes3D
+from matplotlib import cm
+
+# Load the terrain
+terrain1 = imread('SRTM_data_Norway_1.tif')
+# Show the terrain
+plt.figure()
+plt.title('Terrain over Norway 1')
+plt.imshow(terrain1, cmap='gray')
+plt.xlabel('X')
+plt.ylabel('Y')
+plt.show()
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +

If you should have problems in downloading the digital terrain data, +we provide two examples under the data folder of project 1. One is +from a region close to Stavanger in Norway and the other Møsvatn +Austfjell, again in Norway. +Feel free to produce your own terrain data. +

+ +

Alternatively, if you would like to use another data set, feel free to do so. This could be data close to your reseach area or simply a data set you found interesting. See for example kaggle.com for examples.

+ +

Our final part deals with the parameterization of your digital terrain +data (or your own data). We will apply all three methods for linear regression, the same type (or higher order) of polynomial +approximation and cross-validation as resampling technique to evaluate which +model fits the data best. +

+ +

At the end, you should present a critical evaluation of your results +and discuss the applicability of these regression methods to the type +of data presented here (either the terrain data we propose or other data sets). +

+

Background literature

+ +
    +
  1. For a discussion and derivation of the variances and mean squared errors using linear regression, see the Lecture notes on ridge regression by Wessel N. van Wieringen
    2. +
  2. The textbook of Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer, chapters 3 and 7 are the most relevant ones for the analysis here.
    3. +
+

Introduction to numerical projects

+ +

Here follows a brief recipe and recommendation on how to answer the various questions when preparing your answers.

+ + +

Format for electronic delivery of report and programs

+ +

The preferred format for the report is a PDF file. You can also use DOC or postscript formats or as an ipython notebook file. As programming language we prefer that you choose between C/C++, Fortran2008, Julia or Python. The following prescription should be followed when preparing the report:

+ + +

Finally, +we encourage you to collaborate. Optimal working groups consist of +2-3 students. You can then hand in a common report. +

+

Software and needed installations

+ +

If you have Python installed (we recommend Python3) and you feel pretty familiar with installing different packages, +we recommend that you install the following Python packages via pip as +

+
    +
  1. pip install numpy scipy matplotlib ipython scikit-learn tensorflow sympy pandas pillow
    2. +
+

For Python3, replace pip with pip3.

+ +

See below for a discussion of tensorflow and scikit-learn.

+ +

For OSX users we recommend also, after having installed Xcode, to install brew. Brew allows +for a seamless installation of additional software via for example +

+
    +
  1. brew install python3
    2. +
+

For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution +you can use pip as well and simply install Python as +

+
    +
  1. sudo apt-get install python3 (or python for python2.7)
    2. +
+

etc etc.

+ +

If you don't want to install various Python packages with their dependencies separately, we recommend two widely used distrubutions which set up all relevant dependencies for Python, namely

+
    +
  1. Anaconda Anaconda is an open source distribution of the Python and R programming languages for large-scale data processing, predictive analytics, and scientific computing, that aims to simplify package management and deployment. Package versions are managed by the package management system conda
    2. +
  2. Enthought canopy is a Python distribution for scientific and analytic computing distribution and analysis environment, available for free and under a commercial license.
    3. +
+

Popular software packages written in Python for ML are

+ + +

These are all freely available at their respective GitHub sites. They +encompass communities of developers in the thousands or more. And the number +of code developers and contributors keeps increasing. +

+ + + + + diff --git a/doc/Projects/2023/Project1/ipynb/Project1.ipynb b/doc/Projects/2023/Project1/ipynb/Project1.ipynb new file mode 100644 index 000000000..eb44608a1 --- /dev/null +++ b/doc/Projects/2023/Project1/ipynb/Project1.ipynb @@ -0,0 +1,882 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "9b31c43c", + "metadata": { + "editable": true + }, + "source": [ + "\n", + "" + ] + }, + { + "cell_type": "markdown", + "id": "de414429", + "metadata": { + "editable": true + }, + "source": [ + "# Project 1 on Machine Learning, deadline October 7, 2023\n", + "**[Data Analysis and Machine Learning FYS-STK3155/FYS4155](http://www.uio.no/studier/emner/matnat/fys/FYS3155/index-eng.html)**, University of Oslo, Norway\n", + "\n", + "Date: **September 3**" + ] + }, + { + "cell_type": "markdown", + "id": "24422f72", + "metadata": { + "editable": true + }, + "source": [ + "## Regression analysis and resampling methods\n", + "\n", + "The main aim of this project is to study in more detail various\n", + "regression methods, including the Ordinary Least Squares (OLS) method.\n", + "In addition to the scientific part, in this course we want also to\n", + "give you an experience in writing scientific reports. The format for\n", + "the delivery of your answers is namely that of a scientific report. At\n", + "for example\n", + "\n", + "we detail how to write a report. Furthermore, at\n", + "\n", + "you can find examples of previous reports. How to write reports will\n", + "also be discussed during lectures and at the various lab sessions.\n", + "\n", + "**A small recommendation when developing the codes here**. Instead of\n", + "jumping on to the two-dimensional function described below, we\n", + "recommend to do the code development and testing with a simpler\n", + "one-dimensional function, similar to those discussed in the exercises\n", + "of week 35. A simple test, as discussed during the lectures the first\n", + "two weeks is to set the design matrix equal to the identity\n", + "matrix. Then your model should give a mean square error which is exactly equal to zero.\n", + "When you are sure that your codes function well, you can then replace\n", + "the one-dimensional test function with the two-dimensional **Franke** function\n", + "discussed here.\n", + "\n", + "The Franke function serves as a stepping stone towards the analysis of\n", + "real topographic data. The latter is the last part of this project." + ] + }, + { + "cell_type": "markdown", + "id": "2f57baf4", + "metadata": { + "editable": true + }, + "source": [ + "### Description of two-dimensional function\n", + "\n", + "We will first study how to fit polynomials to a specific\n", + "two-dimensional function called [Franke's\n", + "function](http://www.dtic.mil/dtic/tr/fulltext/u2/a081688.pdf). This\n", + "is a function which has been widely used when testing various\n", + "interpolation and fitting algorithms. Furthermore, after having\n", + "established the model and the method, we will employ resamling\n", + "techniques such as cross-validation and/or bootstrap in order to perform a\n", + "proper assessment of our models. We will also study in detail the\n", + "so-called Bias-Variance trade off.\n", + "\n", + "The Franke function, which is a weighted sum of four exponentials reads as follows" + ] + }, + { + "cell_type": "markdown", + "id": "1cd79adf", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\begin{align*}\n", + "f(x,y) &= \\frac{3}{4}\\exp{\\left(-\\frac{(9x-2)^2}{4} - \\frac{(9y-2)^2}{4}\\right)}+\\frac{3}{4}\\exp{\\left(-\\frac{(9x+1)^2}{49}- \\frac{(9y+1)}{10}\\right)} \\\\\n", + "&+\\frac{1}{2}\\exp{\\left(-\\frac{(9x-7)^2}{4} - \\frac{(9y-3)^2}{4}\\right)} -\\frac{1}{5}\\exp{\\left(-(9x-4)^2 - (9y-7)^2\\right) }.\n", + "\\end{align*}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "35c04be8", + "metadata": { + "editable": true + }, + "source": [ + "The function will be defined for $x,y\\in [0,1]$. Our first step will\n", + "be to perform an OLS regression analysis of this function, trying out\n", + "a polynomial fit with an $x$ and $y$ dependence of the form $[x, y,\n", + "x^2, y^2, xy, \\dots]$. We will also include bootstrap first as a\n", + "resampling technique. After that we will include the cross-validation\n", + "technique. As discussed in the lectures for weeks 35 and 36,, we can\n", + "use a uniform distribution to set up the arrays of values for $x$ and\n", + "$y$, or as in the example below just a set of fixed values for $x$ and\n", + "$y$ with a given step size. We will fit a function (for example a\n", + "polynomial) of $x$ and $y$. Thereafter we will repeat much of the\n", + "same procedure using the Ridge and Lasso regression methods,\n", + "introducing thus a dependence on the bias (penalty) $\\lambda$.\n", + "\n", + "Finally we are going to use (real) digital terrain data and try to\n", + "reproduce these data using the same methods. We will also try to go\n", + "beyond the second-order polynomials metioned above and explore \n", + "which polynomial fits the data best.\n", + "\n", + "The Python code for the Franke function is included here (it performs also a three-dimensional plot of it)" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "id": "d531f23c", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "%matplotlib inline\n", + "\n", + "from mpl_toolkits.mplot3d import Axes3D\n", + "import matplotlib.pyplot as plt\n", + "from matplotlib import cm\n", + "from matplotlib.ticker import LinearLocator, FormatStrFormatter\n", + "import numpy as np\n", + "from random import random, seed\n", + "\n", + "fig = plt.figure()\n", + "ax = fig.gca(projection='3d')\n", + "\n", + "# Make data.\n", + "x = np.arange(0, 1, 0.05)\n", + "y = np.arange(0, 1, 0.05)\n", + "x, y = np.meshgrid(x,y)\n", + "\n", + "\n", + "def FrankeFunction(x,y):\n", + " term1 = 0.75*np.exp(-(0.25*(9*x-2)**2) - 0.25*((9*y-2)**2))\n", + " term2 = 0.75*np.exp(-((9*x+1)**2)/49.0 - 0.1*(9*y+1))\n", + " term3 = 0.5*np.exp(-(9*x-7)**2/4.0 - 0.25*((9*y-3)**2))\n", + " term4 = -0.2*np.exp(-(9*x-4)**2 - (9*y-7)**2)\n", + " return term1 + term2 + term3 + term4\n", + "\n", + "\n", + "z = FrankeFunction(x, y)\n", + "\n", + "# Plot the surface.\n", + "surf = ax.plot_surface(x, y, z, cmap=cm.coolwarm,\n", + " linewidth=0, antialiased=False)\n", + "\n", + "# Customize the z axis.\n", + "ax.set_zlim(-0.10, 1.40)\n", + "ax.zaxis.set_major_locator(LinearLocator(10))\n", + "ax.zaxis.set_major_formatter(FormatStrFormatter('%.02f'))\n", + "\n", + "# Add a color bar which maps values to colors.\n", + "fig.colorbar(surf, shrink=0.5, aspect=5)\n", + "\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "3117b409", + "metadata": { + "editable": true + }, + "source": [ + "### Part a) : Ordinary Least Square (OLS) on the Franke function\n", + "\n", + "We will generate our own dataset for a function\n", + "$\\mathrm{FrankeFunction}(x,y)$ with $x,y \\in [0,1]$. The function\n", + "$f(x,y)$ is the Franke function. You should explore also the addition\n", + "of an added stochastic noise to this function using the normal\n", + "distribution $N(0,1)$.\n", + "\n", + "*Write your own code* (using either a matrix inversion or a singular\n", + "value decomposition from e.g., **numpy** ) and perform a standard **ordinary least square regression**\n", + "analysis using polynomials in $x$ and $y$ up to fifth order.\n", + "\n", + "Evaluate the mean Squared error (MSE)" + ] + }, + { + "cell_type": "markdown", + "id": "b10e86f2", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "MSE(\\boldsymbol{y},\\tilde{\\boldsymbol{y}}) = \\frac{1}{n}\n", + "\\sum_{i=0}^{n-1}(y_i-\\tilde{y}_i)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "71459fea", + "metadata": { + "editable": true + }, + "source": [ + "and the $R^2$ score function. If $\\tilde{\\boldsymbol{y}}_i$ is the predicted\n", + "value of the $i-th$ sample and $y_i$ is the corresponding true value,\n", + "then the score $R^2$ is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "f5687714", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "R^2(\\boldsymbol{y}, \\tilde{\\boldsymbol{y}}) = 1 - \\frac{\\sum_{i=0}^{n - 1} (y_i - \\tilde{y}_i)^2}{\\sum_{i=0}^{n - 1} (y_i - \\bar{y})^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "b7ddfef4", + "metadata": { + "editable": true + }, + "source": [ + "where we have defined the mean value of $\\boldsymbol{y}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "5ffbd03c", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\bar{y} = \\frac{1}{n} \\sum_{i=0}^{n - 1} y_i.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "04963148", + "metadata": { + "editable": true + }, + "source": [ + "Plot the resulting scores (MSE and R$^2$) as functions of the polynomial degree (here up to polymial degree five).\n", + "Plot also the parameters $\\beta$ as you increase the order of the polynomial. Comment your results.\n", + "\n", + "Your code has to include a scaling/centering of the data (for example by\n", + "subtracting the mean value), and\n", + "a split of the data in training and test data. For this exercise you can\n", + "either write your own code or use for example the function for\n", + "splitting training data provided by the library **Scikit-Learn** (make\n", + "sure you have installed it). This function is called\n", + "$train\\_test\\_split$. **You should present a critical discussion of why and how you have scaled or not scaled the data**.\n", + "\n", + "It is normal in essentially all Machine Learning studies to split the\n", + "data in a training set and a test set (eventually also an additional\n", + "validation set). There\n", + "is no explicit recipe for how much data should be included as training\n", + "data and say test data. An accepted rule of thumb is to use\n", + "approximately $2/3$ to $4/5$ of the data as training data.\n", + "\n", + "You can easily reuse the solutions to your exercises from week 35 and week 36.\n", + "See also the lecture slides from week 35 and week 36." + ] + }, + { + "cell_type": "markdown", + "id": "d17e5544", + "metadata": { + "editable": true + }, + "source": [ + "### Part b): Adding Ridge and Lasso Regression on the Franke function\n", + "\n", + "Write your own code for the Ridge method, either using matrix\n", + "inversion or the singular value decomposition as done in the previous\n", + "exercise.\n", + "\n", + "Perform the same analysis as you did in the previous exercise but now for different values of $\\lambda$. Compare and\n", + "analyze your results with those obtained in parts b-d). Study the\n", + "dependence on $\\lambda$.\n", + "\n", + "This exercise is essentially a repeat of the previous two ones, but now\n", + "with Lasso regression. Write either your own code (difficult and optional) or, in this case,\n", + "you can also use the functionalities of **Scikit-Learn** (recommended). \n", + "Give a\n", + "critical discussion of the three methods and a judgement of which\n", + "model fits the data best. Perform here as well an analysis of the bias-variance trade-off using the **bootstrap** resampling technique and an analysis of the mean squared error using cross-validation." + ] + }, + { + "cell_type": "markdown", + "id": "fbd4c145", + "metadata": { + "editable": true + }, + "source": [ + "### Part a): Paper and pencil part\n", + "\n", + "This exercise deals with various mean values and variances in linear regression method (here it may be useful to look up chapter 3, equation (3.8) of [Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer](https://www.springer.com/gp/book/9780387848570)).\n", + "\n", + "The assumption we have made is \n", + "that there exists a continuous function $f(\\boldsymbol{x})$ and a normal distributed error $\\boldsymbol{\\varepsilon}\\sim N(0, \\sigma^2)$\n", + "which describes our data" + ] + }, + { + "cell_type": "markdown", + "id": "ce635f60", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{y} = f(\\boldsymbol{x})+\\boldsymbol{\\varepsilon}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "d55bb605", + "metadata": { + "editable": true + }, + "source": [ + "We then approximate this function $f(\\boldsymbol{x})$ with our model $\\boldsymbol{\\tilde{y}}$ from the solution of the linear regression equations (ordinary least squares OLS), that is our\n", + "function $f$ is approximated by $\\boldsymbol{\\tilde{y}}$ where we minimized $(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}})^2$, with" + ] + }, + { + "cell_type": "markdown", + "id": "1a43accc", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\tilde{y}} = \\boldsymbol{X}\\boldsymbol{\\beta}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "8a708d42", + "metadata": { + "editable": true + }, + "source": [ + "The matrix $\\boldsymbol{X}$ is the so-called design or feature matrix. \n", + "\n", + "Show that the expectation value of $\\boldsymbol{y}$ for a given element $i$" + ] + }, + { + "cell_type": "markdown", + "id": "a5d55b46", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\mathbb{E}(y_i) =\\sum_{j}x_{ij} \\beta_j=\\mathbf{X}_{i, \\ast} \\, \\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "19e93313", + "metadata": { + "editable": true + }, + "source": [ + "and that\n", + "its variance is" + ] + }, + { + "cell_type": "markdown", + "id": "bb735793", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\mbox{Var}(y_i) = \\sigma^2.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ee65e390", + "metadata": { + "editable": true + }, + "source": [ + "Hence, $y_i \\sim N( \\mathbf{X}_{i, \\ast} \\, \\boldsymbol{\\beta}, \\sigma^2)$, that is $\\boldsymbol{y}$ follows a normal distribution with \n", + "mean value $\\boldsymbol{X}\\boldsymbol{\\beta}$ and variance $\\sigma^2$.\n", + "\n", + "With the OLS expressions for the optimal parameters $\\boldsymbol{\\hat{\\beta}}$ show that" + ] + }, + { + "cell_type": "markdown", + "id": "b29b4187", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\mathbb{E}(\\boldsymbol{\\hat{\\beta}}) = \\boldsymbol{\\beta}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "f8adffa4", + "metadata": { + "editable": true + }, + "source": [ + "Show finally that the variance of $\\boldsymbol{\\beta}$ is" + ] + }, + { + "cell_type": "markdown", + "id": "1f56f45d", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\mbox{Var}(\\boldsymbol{\\hat{\\beta}}) = \\sigma^2 \\, (\\mathbf{X}^{T} \\mathbf{X})^{-1}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "74e687b8", + "metadata": { + "editable": true + }, + "source": [ + "We can use the last expression when we define a so-called confidence interval for the parameters $\\beta$. .\n", + "A given parameter $\\beta_j$ is given by the diagonal matrix element of the above matrix." + ] + }, + { + "cell_type": "markdown", + "id": "b0b7d7bb", + "metadata": { + "editable": true + }, + "source": [ + "### Part c): Bias-variance trade-off and resampling techniques\n", + "\n", + "Our aim here is to study the bias-variance trade-off by implementing the **bootstrap** resampling technique.\n", + "\n", + "With a code which does OLS and includes resampling techniques, \n", + "we will now discuss the bias-variance trade-off in the context of\n", + "continuous predictions such as regression. However, many of the\n", + "intuitions and ideas discussed here also carry over to classification\n", + "tasks and basically all Machine Learning algorithms. \n", + "\n", + "Before you perform an analysis of the bias-variance trade-off on your test data, make\n", + "first a figure similar to Fig. 2.11 of Hastie, Tibshirani, and\n", + "Friedman. Figure 2.11 of this reference displays only the test and training MSEs. The test MSE can be used to \n", + "indicate possible regions of low/high bias and variance. You will most likely not get an\n", + "equally smooth curve!\n", + "\n", + "With this result we move on to the bias-variance trade-off analysis.\n", + "\n", + "Consider a\n", + "dataset $\\mathcal{L}$ consisting of the data\n", + "$\\mathbf{X}_\\mathcal{L}=\\{(y_j, \\boldsymbol{x}_j), j=0\\ldots n-1\\}$.\n", + "\n", + "As in part a), we assume that the true data is generated from a noisy model" + ] + }, + { + "cell_type": "markdown", + "id": "6f7509d8", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{y}=f(\\boldsymbol{x}) + \\boldsymbol{\\epsilon}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "8d06ec98", + "metadata": { + "editable": true + }, + "source": [ + "Here $\\epsilon$ is normally distributed with mean zero and standard\n", + "deviation $\\sigma^2$.\n", + "\n", + "In our derivation of the ordinary least squares method we defined then\n", + "an approximation to the function $f$ in terms of the parameters\n", + "$\\boldsymbol{\\beta}$ and the design matrix $\\boldsymbol{X}$ which embody our model,\n", + "that is $\\boldsymbol{\\tilde{y}}=\\boldsymbol{X}\\boldsymbol{\\beta}$.\n", + "\n", + "The parameters $\\boldsymbol{\\beta}$ are in turn found by optimizing the mean\n", + "squared error via the so-called cost function" + ] + }, + { + "cell_type": "markdown", + "id": "9e493a94", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "C(\\boldsymbol{X},\\boldsymbol{\\beta}) =\\frac{1}{n}\\sum_{i=0}^{n-1}(y_i-\\tilde{y}_i)^2=\\mathbb{E}\\left[(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}})^2\\right].\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "b6b7b71c", + "metadata": { + "editable": true + }, + "source": [ + "Here the expected value $\\mathbb{E}$ is the sample value. \n", + "\n", + "Show that you can rewrite this in terms of a term which contains the variance of the model itself (the so-called variance term), a\n", + "term which measures the deviation from the true data and the mean value of the model (the bias term) and finally the variance of the noise.\n", + "That is, show that" + ] + }, + { + "cell_type": "markdown", + "id": "eadda792", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\mathbb{E}\\left[(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}})^2\\right]=(\\mathrm{Bias}[\\tilde{y}])^2+\\mathrm{var}[\\tilde{f}]+\\sigma^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "cef4fa19", + "metadata": { + "editable": true + }, + "source": [ + "with" + ] + }, + { + "cell_type": "markdown", + "id": "38cc8dbb", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "(\\mathrm{Bias}[\\tilde{y}])^2=\\left(\\boldsymbol{y}-\\mathbb{E}\\left[\\boldsymbol{\\tilde{y}}\\right]\\right)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "a9dcfd8b", + "metadata": { + "editable": true + }, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "41198b5e", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\mathrm{var}[\\tilde{f}]=\\frac{1}{n}\\sum_i(\\tilde{y}_i-\\mathbb{E}\\left[\\boldsymbol{\\tilde{y}}\\right])^2.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "a252009f", + "metadata": { + "editable": true + }, + "source": [ + "The answer to this exercise should be included in the theory part of the report.\n", + "Explain what the terms mean and discuss their interpretations.\n", + "\n", + "Perform then a bias-variance analysis of the Franke function by\n", + "studying the MSE value as function of the complexity of your model.\n", + "\n", + "Discuss the bias and variance trade-off as function\n", + "of your model complexity (the degree of the polynomial) and the number\n", + "of data points, and possibly also your training and test data using the **bootstrap** resampling method.\n", + "You can follow the code example in the jupyter-book at ." + ] + }, + { + "cell_type": "markdown", + "id": "00854ec4", + "metadata": { + "editable": true + }, + "source": [ + "### Part d): Cross-validation as resampling techniques, adding more complexity\n", + "\n", + "The aim here is to write your own code for another widely popular\n", + "resampling technique, the so-called cross-validation method. Again,\n", + "before you start with cross-validation approach, you should scale your\n", + "data if you think this is needed.\n", + "\n", + "Implement the $k$-fold cross-validation algorithm (write your own\n", + "code) and evaluate again the MSE function resulting\n", + "from the test folds. You can compare your own code with that from\n", + "**Scikit-Learn** if needed. \n", + "\n", + "Compare the MSE you get from your cross-validation code with the one\n", + "you got from your **bootstrap** code. Comment your results. Try $5-10$\n", + "folds. You can also compare your own cross-validation code with the\n", + "one provided by **Scikit-Learn**." + ] + }, + { + "cell_type": "markdown", + "id": "0603ed7b", + "metadata": { + "editable": true + }, + "source": [ + "### Part g): Analysis of real data\n", + "\n", + "With our codes functioning and having been tested properly on a\n", + "simpler function we are now ready to look at real data. We will\n", + "essentially repeat in this exercise what was done in exercises 1-5. However, we\n", + "need first to download the data and prepare properly the inputs to our\n", + "codes. We are going to download digital terrain data from the website\n", + ",\n", + "\n", + "Or, if you prefer, we have placed selected datafiles at \n", + "\n", + "In order to obtain data for a specific region, you need to register as\n", + "a user (free) at this website and then decide upon which area you want\n", + "to fetch the digital terrain data from. In order to be able to read\n", + "the data properly, you need to specify that the format should be **SRTM\n", + "Arc-Second Global** and download the data as a **GeoTIF** file. The\n", + "files are then stored in *tif* format which can be imported into a\n", + "Python program using" + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "id": "293624f7", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "scipy.misc.imread" + ] + }, + { + "cell_type": "markdown", + "id": "2eba6a2c", + "metadata": { + "editable": true + }, + "source": [ + "Here is a simple part of a Python code which reads and plots the data\n", + "from such files" + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "id": "a7f37cdb", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "from imageio import imread\n", + "import matplotlib.pyplot as plt\n", + "from mpl_toolkits.mplot3d import Axes3D\n", + "from matplotlib import cm\n", + "\n", + "# Load the terrain\n", + "terrain1 = imread('SRTM_data_Norway_1.tif')\n", + "# Show the terrain\n", + "plt.figure()\n", + "plt.title('Terrain over Norway 1')\n", + "plt.imshow(terrain1, cmap='gray')\n", + "plt.xlabel('X')\n", + "plt.ylabel('Y')\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "74af5abe", + "metadata": { + "editable": true + }, + "source": [ + "If you should have problems in downloading the digital terrain data,\n", + "we provide two examples under the data folder of project 1. One is\n", + "from a region close to Stavanger in Norway and the other Møsvatn\n", + "Austfjell, again in Norway.\n", + "Feel free to produce your own terrain data.\n", + "\n", + "Alternatively, if you would like to use another data set, feel free to do so. This could be data close to your reseach area or simply a data set you found interesting. See for example [kaggle.com](https://www.kaggle.com/datasets) for examples.\n", + "\n", + "Our final part deals with the parameterization of your digital terrain\n", + "data (or your own data). We will apply all three methods for linear regression, the same type (or higher order) of polynomial\n", + "approximation and cross-validation as resampling technique to evaluate which\n", + "model fits the data best.\n", + "\n", + "At the end, you should present a critical evaluation of your results\n", + "and discuss the applicability of these regression methods to the type\n", + "of data presented here (either the terrain data we propose or other data sets)." + ] + }, + { + "cell_type": "markdown", + "id": "5ad507ad", + "metadata": { + "editable": true + }, + "source": [ + "## Background literature\n", + "\n", + "1. For a discussion and derivation of the variances and mean squared errors using linear regression, see the [Lecture notes on ridge regression by Wessel N. van Wieringen](https://arxiv.org/abs/1509.09169)\n", + "\n", + "2. The textbook of [Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer](https://www.springer.com/gp/book/9780387848570), chapters 3 and 7 are the most relevant ones for the analysis here." + ] + }, + { + "cell_type": "markdown", + "id": "194f7c51", + "metadata": { + "editable": true + }, + "source": [ + "## Introduction to numerical projects\n", + "\n", + "Here follows a brief recipe and recommendation on how to answer the various questions when preparing your answers. \n", + "\n", + " * Give a short description of the nature of the problem and the eventual numerical methods you have used.\n", + "\n", + " * Describe the algorithm you have used and/or developed. Here you may find it convenient to use pseudocoding. In many cases you can describe the algorithm in the program itself.\n", + "\n", + " * Include the source code of your program. Comment your program properly. You should have the code at your GitHub/GitLab link. You can also place the code in an appendix of your report.\n", + "\n", + " * If possible, try to find analytic solutions, or known limits in order to test your program when developing the code.\n", + "\n", + " * Include your results either in figure form or in a table. Remember to label your results. All tables and figures should have relevant captions and labels on the axes.\n", + "\n", + " * Try to evaluate the reliabilty and numerical stability/precision of your results. If possible, include a qualitative and/or quantitative discussion of the numerical stability, eventual loss of precision etc.\n", + "\n", + " * Try to give an interpretation of you results in your answers to the problems.\n", + "\n", + " * Critique: if possible include your comments and reflections about the exercise, whether you felt you learnt something, ideas for improvements and other thoughts you've made when solving the exercise. We wish to keep this course at the interactive level and your comments can help us improve it.\n", + "\n", + " * Try to establish a practice where you log your work at the computerlab. You may find such a logbook very handy at later stages in your work, especially when you don't properly remember what a previous test version of your program did. Here you could also record the time spent on solving the exercise, various algorithms you may have tested or other topics which you feel worthy of mentioning." + ] + }, + { + "cell_type": "markdown", + "id": "a1cf04b9", + "metadata": { + "editable": true + }, + "source": [ + "## Format for electronic delivery of report and programs\n", + "\n", + "The preferred format for the report is a PDF file. You can also use DOC or postscript formats or as an ipython notebook file. As programming language we prefer that you choose between C/C++, Fortran2008, Julia or Python. The following prescription should be followed when preparing the report:\n", + "\n", + " * Use Canvas to hand in your projects, log in at with your normal UiO username and password.\n", + "\n", + " * Upload **only** the report file or the link to your GitHub/GitLab or similar typo of repos! For the source code file(s) you have developed please provide us with your link to your GitHub/GitLab or similar domain. The report file should include all of your discussions and a list of the codes you have developed. Do not include library files which are available at the course homepage, unless you have made specific changes to them.\n", + "\n", + " * In your GitHub/GitLab or similar repository, please include a folder which contains selected results. These can be in the form of output from your code for a selected set of runs and input parameters.\n", + "\n", + "Finally, \n", + "we encourage you to collaborate. Optimal working groups consist of \n", + "2-3 students. You can then hand in a common report." + ] + }, + { + "cell_type": "markdown", + "id": "421f2a0c", + "metadata": { + "editable": true + }, + "source": [ + "## Software and needed installations\n", + "\n", + "If you have Python installed (we recommend Python3) and you feel pretty familiar with installing different packages, \n", + "we recommend that you install the following Python packages via **pip** as\n", + "1. pip install numpy scipy matplotlib ipython scikit-learn tensorflow sympy pandas pillow\n", + "\n", + "For Python3, replace **pip** with **pip3**.\n", + "\n", + "See below for a discussion of **tensorflow** and **scikit-learn**. \n", + "\n", + "For OSX users we recommend also, after having installed Xcode, to install **brew**. Brew allows \n", + "for a seamless installation of additional software via for example\n", + "1. brew install python3\n", + "\n", + "For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution\n", + "you can use **pip** as well and simply install Python as \n", + "1. sudo apt-get install python3 (or python for python2.7)\n", + "\n", + "etc etc. \n", + "\n", + "If you don't want to install various Python packages with their dependencies separately, we recommend two widely used distrubutions which set up all relevant dependencies for Python, namely\n", + "1. [Anaconda](https://docs.anaconda.com/) Anaconda is an open source distribution of the Python and R programming languages for large-scale data processing, predictive analytics, and scientific computing, that aims to simplify package management and deployment. Package versions are managed by the package management system **conda**\n", + "\n", + "2. [Enthought canopy](https://www.enthought.com/product/canopy/) is a Python distribution for scientific and analytic computing distribution and analysis environment, available for free and under a commercial license.\n", + "\n", + "Popular software packages written in Python for ML are\n", + "\n", + "* [Scikit-learn](http://scikit-learn.org/stable/), \n", + "\n", + "* [Tensorflow](https://www.tensorflow.org/),\n", + "\n", + "* [PyTorch](http://pytorch.org/) and \n", + "\n", + "* [Keras](https://keras.io/).\n", + "\n", + "These are all freely available at their respective GitHub sites. They \n", + "encompass communities of developers in the thousands or more. And the number\n", + "of code developers and contributors keeps increasing." + ] + } + ], + "metadata": {}, + "nbformat": 4, + "nbformat_minor": 5 +} diff --git a/doc/Projects/2023/Project1/ipynb/ipynb-Project1-src.tar.gz b/doc/Projects/2023/Project1/ipynb/ipynb-Project1-src.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..b06940a4ca2bf30e9e76918d1c847c0b091e9fb3 GIT binary patch literal 193 zcmV;y06za8iwFQB*z{xo1MSbv3W7io2XN0m#XN!R>Z<1;4;=!DUSQ2N*K)^pSID=| z52#boMIzYW<2uT=&DbpmLlY~n>qcjD|SQ^uq5f%cWQ5;1c zko8V_>5b)jJf*E0p^Q-P=EkwA`mkqt1)ljQju#ncN)){w`htd literal 0 HcmV?d00001 diff --git a/doc/Projects/2023/Project1/pdf/Project1.p.tex b/doc/Projects/2023/Project1/pdf/Project1.p.tex new file mode 100644 index 000000000..0f998ea00 --- /dev/null +++ b/doc/Projects/2023/Project1/pdf/Project1.p.tex @@ -0,0 +1,678 @@ +%% +%% Automatically generated file from DocOnce source +%% (https://github.com/doconce/doconce/) +%% doconce format latex Project1.do.txt --print_latex_style=trac --latex_admon=paragraph +%% +% #ifdef PTEX2TEX_EXPLANATION +%% +%% The file follows the ptex2tex extended LaTeX format, see +%% ptex2tex: https://code.google.com/p/ptex2tex/ +%% +%% Run +%% ptex2tex myfile +%% or +%% doconce ptex2tex myfile +%% +%% to turn myfile.p.tex into an ordinary LaTeX file myfile.tex. +%% (The ptex2tex program: https://code.google.com/p/ptex2tex) +%% Many preprocess options can be added to ptex2tex or doconce ptex2tex +%% +%% ptex2tex -DMINTED myfile +%% doconce ptex2tex myfile envir=minted +%% +%% ptex2tex will typeset code environments according to a global or local +%% .ptex2tex.cfg configure file. doconce ptex2tex will typeset code +%% according to options on the command line (just type doconce ptex2tex to +%% see examples). If doconce ptex2tex has envir=minted, it enables the +%% minted style without needing -DMINTED. +% #endif + +% #define PREAMBLE + +% #ifdef PREAMBLE +%-------------------- begin preamble ---------------------- + +\documentclass[% +oneside, % oneside: electronic viewing, twoside: printing +final, % draft: marks overfull hboxes, figures with paths +10pt]{article} + +\listfiles % print all files needed to compile this document + +\usepackage{relsize,makeidx,color,setspace,amsmath,amsfonts,amssymb} +\usepackage[table]{xcolor} +\usepackage{bm,ltablex,microtype} + +\usepackage[pdftex]{graphicx} + +\usepackage{ptex2tex} +% #ifdef MINTED +\usepackage{minted} +\usemintedstyle{default} +% #endif + +\usepackage[T1]{fontenc} +%\usepackage[latin1]{inputenc} +\usepackage{ucs} +\usepackage[utf8x]{inputenc} + +\usepackage{lmodern} % Latin Modern fonts derived from Computer Modern + +% Hyperlinks in PDF: +\definecolor{linkcolor}{rgb}{0,0,0.4} +\usepackage{hyperref} +\hypersetup{ + breaklinks=true, + colorlinks=true, + linkcolor=linkcolor, + urlcolor=linkcolor, + citecolor=black, + filecolor=black, + %filecolor=blue, + pdfmenubar=true, + pdftoolbar=true, + bookmarksdepth=3 % Uncomment (and tweak) for PDF bookmarks with more levels than the TOC + } +%\hyperbaseurl{} % hyperlinks are relative to this root + +\setcounter{tocdepth}{2} % levels in table of contents + +% prevent orhpans and widows +\clubpenalty = 10000 +\widowpenalty = 10000 + +% --- end of standard preamble for documents --- + + +% insert custom LaTeX commands... + +\raggedbottom +\makeindex +\usepackage[totoc]{idxlayout} % for index in the toc +\usepackage[nottoc]{tocbibind} % for references/bibliography in the toc + +%-------------------- end preamble ---------------------- + +\begin{document} + +% matching end for #ifdef PREAMBLE +% #endif + +\newcommand{\exercisesection}[1]{\subsection*{#1}} + + +% ------------------- main content ---------------------- + + + +% ----------------- title ------------------------- + +\thispagestyle{empty} + +\begin{center} +{\LARGE\bf +\begin{spacing}{1.25} +Project 1 on Machine Learning, deadline October 7, 2023 +\end{spacing} +} +\end{center} + +% ----------------- author(s) ------------------------- + +\begin{center} +{\bf \href{{http://www.uio.no/studier/emner/matnat/fys/FYS3155/index-eng.html}}{Data Analysis and Machine Learning FYS-STK3155/FYS4155}} +\end{center} + + \begin{center} +% List of all institutions: +\centerline{{\small University of Oslo, Norway}} +\end{center} + +% ----------------- end author(s) ------------------------- + +% --- begin date --- +\begin{center} +September 3 +\end{center} +% --- end date --- + +\vspace{1cm} + + +\subsection{Regression analysis and resampling methods} + +The main aim of this project is to study in more detail various +regression methods, including the Ordinary Least Squares (OLS) method. +In addition to the scientific part, in this course we want also to +give you an experience in writing scientific reports. The format for +the delivery of your answers is namely that of a scientific report. At +for example +\href{{https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/EvaluationGrading/EvaluationForm.md}}{\nolinkurl{https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/EvaluationGrading/EvaluationForm.md}} +we detail how to write a report. Furthermore, at +\href{{https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/ReportExample/}}{\nolinkurl{https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/ReportExample/}} +you can find examples of previous reports. How to write reports will +also be discussed during lectures and at the various lab sessions. + +\textbf{A small recommendation when developing the codes here}. Instead of +jumping on to the two-dimensional function described below, we +recommend to do the code development and testing with a simpler +one-dimensional function, similar to those discussed in the exercises +of week 35. A simple test, as discussed during the lectures the first +two weeks is to set the design matrix equal to the identity +matrix. Then your model should give a mean square error which is exactly equal to zero. +When you are sure that your codes function well, you can then replace +the one-dimensional test function with the two-dimensional \textbf{Franke} function +discussed here. + +The Franke function serves as a stepping stone towards the analysis of +real topographic data. The latter is the last part of this project. + +\paragraph{Description of two-dimensional function.} +We will first study how to fit polynomials to a specific +two-dimensional function called \href{{http://www.dtic.mil/dtic/tr/fulltext/u2/a081688.pdf}}{Franke's +function}. This +is a function which has been widely used when testing various +interpolation and fitting algorithms. Furthermore, after having +established the model and the method, we will employ resamling +techniques such as cross-validation and/or bootstrap in order to perform a +proper assessment of our models. We will also study in detail the +so-called Bias-Variance trade off. + +The Franke function, which is a weighted sum of four exponentials reads as follows +\begin{align*} +f(x,y) &= \frac{3}{4}\exp{\left(-\frac{(9x-2)^2}{4} - \frac{(9y-2)^2}{4}\right)}+\frac{3}{4}\exp{\left(-\frac{(9x+1)^2}{49}- \frac{(9y+1)}{10}\right)} \\ +&+\frac{1}{2}\exp{\left(-\frac{(9x-7)^2}{4} - \frac{(9y-3)^2}{4}\right)} -\frac{1}{5}\exp{\left(-(9x-4)^2 - (9y-7)^2\right) }. +\end{align*} + +The function will be defined for $x,y\in [0,1]$. Our first step will +be to perform an OLS regression analysis of this function, trying out +a polynomial fit with an $x$ and $y$ dependence of the form $[x, y, +x^2, y^2, xy, \dots]$. We will also include bootstrap first as a +resampling technique. After that we will include the cross-validation +technique. As discussed in the lectures for weeks 35 and 36,, we can +use a uniform distribution to set up the arrays of values for $x$ and +$y$, or as in the example below just a set of fixed values for $x$ and +$y$ with a given step size. We will fit a function (for example a +polynomial) of $x$ and $y$. Thereafter we will repeat much of the +same procedure using the Ridge and Lasso regression methods, +introducing thus a dependence on the bias (penalty) $\lambda$. + +Finally we are going to use (real) digital terrain data and try to +reproduce these data using the same methods. We will also try to go +beyond the second-order polynomials metioned above and explore +which polynomial fits the data best. + +The Python code for the Franke function is included here (it performs also a three-dimensional plot of it) + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +\bpycod +from mpl_toolkits.mplot3d import Axes3D +import matplotlib.pyplot as plt +from matplotlib import cm +from matplotlib.ticker import LinearLocator, FormatStrFormatter +import numpy as np +from random import random, seed + +fig = plt.figure() +ax = fig.gca(projection='3d') + +# Make data. +x = np.arange(0, 1, 0.05) +y = np.arange(0, 1, 0.05) +x, y = np.meshgrid(x,y) + + +def FrankeFunction(x,y): + term1 = 0.75*np.exp(-(0.25*(9*x-2)**2) - 0.25*((9*y-2)**2)) + term2 = 0.75*np.exp(-((9*x+1)**2)/49.0 - 0.1*(9*y+1)) + term3 = 0.5*np.exp(-(9*x-7)**2/4.0 - 0.25*((9*y-3)**2)) + term4 = -0.2*np.exp(-(9*x-4)**2 - (9*y-7)**2) + return term1 + term2 + term3 + term4 + + +z = FrankeFunction(x, y) + +# Plot the surface. +surf = ax.plot_surface(x, y, z, cmap=cm.coolwarm, + linewidth=0, antialiased=False) + +# Customize the z axis. +ax.set_zlim(-0.10, 1.40) +ax.zaxis.set_major_locator(LinearLocator(10)) +ax.zaxis.set_major_formatter(FormatStrFormatter('%.02f')) + +# Add a color bar which maps values to colors. +fig.colorbar(surf, shrink=0.5, aspect=5) + +plt.show() + + +\epycod + + +\paragraph{Part a) : Ordinary Least Square (OLS) on the Franke function.} +We will generate our own dataset for a function +$\mathrm{FrankeFunction}(x,y)$ with $x,y \in [0,1]$. The function +$f(x,y)$ is the Franke function. You should explore also the addition +of an added stochastic noise to this function using the normal +distribution $N(0,1)$. + +\emph{Write your own code} (using either a matrix inversion or a singular +value decomposition from e.g., \textbf{numpy} ) and perform a standard \textbf{ordinary least square regression} +analysis using polynomials in $x$ and $y$ up to fifth order. + +Evaluate the mean Squared error (MSE) + +\[ MSE(\bm{y},\tilde{\bm{y}}) = \frac{1}{n} +\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +\] + +and the $R^2$ score function. If $\tilde{\bm{y}}_i$ is the predicted +value of the $i-th$ sample and $y_i$ is the corresponding true value, +then the score $R^2$ is defined as + +\[ +R^2(\bm{y}, \tilde{\bm{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +\] + +where we have defined the mean value of $\bm{y}$ as + +\[ +\bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +\] + +Plot the resulting scores (MSE and R$^2$) as functions of the polynomial degree (here up to polymial degree five). +Plot also the parameters $\beta$ as you increase the order of the polynomial. Comment your results. + +Your code has to include a scaling/centering of the data (for example by +subtracting the mean value), and +a split of the data in training and test data. For this exercise you can +either write your own code or use for example the function for +splitting training data provided by the library \textbf{Scikit-Learn} (make +sure you have installed it). This function is called +$train\_test\_split$. \textbf{You should present a critical discussion of why and how you have scaled or not scaled the data}. + +It is normal in essentially all Machine Learning studies to split the +data in a training set and a test set (eventually also an additional +validation set). There +is no explicit recipe for how much data should be included as training +data and say test data. An accepted rule of thumb is to use +approximately $2/3$ to $4/5$ of the data as training data. + +You can easily reuse the solutions to your exercises from week 35 and week 36. +See also the lecture slides from week 35 and week 36. + +\paragraph{Part b): Adding Ridge and Lasso Regression on the Franke function.} +Write your own code for the Ridge method, either using matrix +inversion or the singular value decomposition as done in the previous +exercise. + +Perform the same analysis as you did in the previous exercise but now for different values of $\lambda$. Compare and +analyze your results with those obtained in parts b-d). Study the +dependence on $\lambda$. + +This exercise is essentially a repeat of the previous two ones, but now +with Lasso regression. Write either your own code (difficult and optional) or, in this case, +you can also use the functionalities of \textbf{Scikit-Learn} (recommended). +Give a +critical discussion of the three methods and a judgement of which +model fits the data best. Perform here as well an analysis of the bias-variance trade-off using the \textbf{bootstrap} resampling technique and an analysis of the mean squared error using cross-validation. + +\paragraph{Part a): Paper and pencil part.} +This exercise deals with various mean values and variances in linear regression method (here it may be useful to look up chapter 3, equation (3.8) of \href{{https://www.springer.com/gp/book/9780387848570}}{Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer}). + +The assumption we have made is +that there exists a continuous function $f(\bm{x})$ and a normal distributed error $\bm{\varepsilon}\sim N(0, \sigma^2)$ +which describes our data +\[ +\bm{y} = f(\bm{x})+\bm{\varepsilon} +\] + +We then approximate this function $f(\bm{x})$ with our model $\bm{\tilde{y}}$ from the solution of the linear regression equations (ordinary least squares OLS), that is our +function $f$ is approximated by $\bm{\tilde{y}}$ where we minimized $(\bm{y}-\bm{\tilde{y}})^2$, with +\[ +\bm{\tilde{y}} = \bm{X}\bm{\beta}. +\] +The matrix $\bm{X}$ is the so-called design or feature matrix. + +Show that the expectation value of $\bm{y}$ for a given element $i$ +\[ +\mathbb{E}(y_i) =\sum_{j}x_{ij} \beta_j=\mathbf{X}_{i, \ast} \, \bm{\beta}, +\] +and that +its variance is +\[ +\mbox{Var}(y_i) = \sigma^2. +\] +Hence, $y_i \sim N( \mathbf{X}_{i, \ast} \, \bm{\beta}, \sigma^2)$, that is $\bm{y}$ follows a normal distribution with +mean value $\bm{X}\bm{\beta}$ and variance $\sigma^2$. + +With the OLS expressions for the optimal parameters $\bm{\hat{\beta}}$ show that +\[ +\mathbb{E}(\bm{\hat{\beta}}) = \bm{\beta}. +\] +Show finally that the variance of $\bm{\beta}$ is +\[ +\mbox{Var}(\bm{\hat{\beta}}) = \sigma^2 \, (\mathbf{X}^{T} \mathbf{X})^{-1}. +\] + +We can use the last expression when we define a so-called confidence interval for the parameters $\beta$. . +A given parameter $\beta_j$ is given by the diagonal matrix element of the above matrix. + +\paragraph{Part c): Bias-variance trade-off and resampling techniques.} +Our aim here is to study the bias-variance trade-off by implementing the \textbf{bootstrap} resampling technique. + +With a code which does OLS and includes resampling techniques, +we will now discuss the bias-variance trade-off in the context of +continuous predictions such as regression. However, many of the +intuitions and ideas discussed here also carry over to classification +tasks and basically all Machine Learning algorithms. + +Before you perform an analysis of the bias-variance trade-off on your test data, make +first a figure similar to Fig.~2.11 of Hastie, Tibshirani, and +Friedman. Figure 2.11 of this reference displays only the test and training MSEs. The test MSE can be used to +indicate possible regions of low/high bias and variance. You will most likely not get an +equally smooth curve! + +With this result we move on to the bias-variance trade-off analysis. + +Consider a +dataset $\mathcal{L}$ consisting of the data +$\mathbf{X}_\mathcal{L}=\{(y_j, \boldsymbol{x}_j), j=0\ldots n-1\}$. + +As in part a), we assume that the true data is generated from a noisy model + +\[ +\bm{y}=f(\boldsymbol{x}) + \bm{\epsilon}. +\] + +Here $\epsilon$ is normally distributed with mean zero and standard +deviation $\sigma^2$. + +In our derivation of the ordinary least squares method we defined then +an approximation to the function $f$ in terms of the parameters +$\bm{\beta}$ and the design matrix $\bm{X}$ which embody our model, +that is $\bm{\tilde{y}}=\bm{X}\bm{\beta}$. + +The parameters $\bm{\beta}$ are in turn found by optimizing the mean +squared error via the so-called cost function + +\[ +C(\bm{X},\bm{\beta}) =\frac{1}{n}\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2=\mathbb{E}\left[(\bm{y}-\bm{\tilde{y}})^2\right]. +\] +Here the expected value $\mathbb{E}$ is the sample value. + +Show that you can rewrite this in terms of a term which contains the variance of the model itself (the so-called variance term), a +term which measures the deviation from the true data and the mean value of the model (the bias term) and finally the variance of the noise. +That is, show that +\[ +\mathbb{E}\left[(\bm{y}-\bm{\tilde{y}})^2\right]=(\mathrm{Bias}[\tilde{y}])^2+\mathrm{var}[\tilde{f}]+\sigma^2, +\] +with +\[ +(\mathrm{Bias}[\tilde{y}])^2=\left(\bm{y}-\mathbb{E}\left[\bm{\tilde{y}}\right]\right)^2, +\] +and +\[ +\mathrm{var}[\tilde{f}]=\frac{1}{n}\sum_i(\tilde{y}_i-\mathbb{E}\left[\bm{\tilde{y}}\right])^2. +\] +The answer to this exercise should be included in the theory part of the report. +Explain what the terms mean and discuss their interpretations. + +Perform then a bias-variance analysis of the Franke function by +studying the MSE value as function of the complexity of your model. + +Discuss the bias and variance trade-off as function +of your model complexity (the degree of the polynomial) and the number +of data points, and possibly also your training and test data using the \textbf{bootstrap} resampling method. +You can follow the code example in the jupyter-book at \href{{https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter3.html#the-bias-variance-tradeoff}}{\nolinkurl{https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter3.html\#the-bias-variance-tradeoff}}. + +\paragraph{Part d): Cross-validation as resampling techniques, adding more complexity.} +The aim here is to write your own code for another widely popular +resampling technique, the so-called cross-validation method. Again, +before you start with cross-validation approach, you should scale your +data if you think this is needed. + +Implement the $k$-fold cross-validation algorithm (write your own +code) and evaluate again the MSE function resulting +from the test folds. You can compare your own code with that from +\textbf{Scikit-Learn} if needed. + +Compare the MSE you get from your cross-validation code with the one +you got from your \textbf{bootstrap} code. Comment your results. Try $5-10$ +folds. You can also compare your own cross-validation code with the +one provided by \textbf{Scikit-Learn}. + +\paragraph{Part g): Analysis of real data.} +With our codes functioning and having been tested properly on a +simpler function we are now ready to look at real data. We will +essentially repeat in this exercise what was done in exercises 1-5. However, we +need first to download the data and prepare properly the inputs to our +codes. We are going to download digital terrain data from the website +\href{{https://earthexplorer.usgs.gov/}}{\nolinkurl{https://earthexplorer.usgs.gov/}}, + +Or, if you prefer, we have placed selected datafiles at \href{{https://github.com/CompPhysics/MachineLearning/tree/master/doc/Projects/2022/Project1/DataFiles}}{\nolinkurl{https://github.com/CompPhysics/MachineLearning/tree/master/doc/Projects/2022/Project1/DataFiles}} + +In order to obtain data for a specific region, you need to register as +a user (free) at this website and then decide upon which area you want +to fetch the digital terrain data from. In order to be able to read +the data properly, you need to specify that the format should be \textbf{SRTM +Arc-Second Global} and download the data as a \textbf{GeoTIF} file. The +files are then stored in \emph{tif} format which can be imported into a +Python program using + + + +\bpycod +scipy.misc.imread + +\epycod + + +Here is a simple part of a Python code which reads and plots the data +from such files + + + + + + + + + + + + + + + + + +\bpycod +import numpy as np +from imageio import imread +import matplotlib.pyplot as plt +from mpl_toolkits.mplot3d import Axes3D +from matplotlib import cm + +# Load the terrain +terrain1 = imread('SRTM_data_Norway_1.tif') +# Show the terrain +plt.figure() +plt.title('Terrain over Norway 1') +plt.imshow(terrain1, cmap='gray') +plt.xlabel('X') +plt.ylabel('Y') +plt.show() + +\epycod + + +If you should have problems in downloading the digital terrain data, +we provide two examples under the data folder of project 1. One is +from a region close to Stavanger in Norway and the other Møsvatn +Austfjell, again in Norway. +Feel free to produce your own terrain data. + +Alternatively, if you would like to use another data set, feel free to do so. This could be data close to your reseach area or simply a data set you found interesting. See for example \href{{https://www.kaggle.com/datasets}}{kaggle.com} for examples. + +Our final part deals with the parameterization of your digital terrain +data (or your own data). We will apply all three methods for linear regression, the same type (or higher order) of polynomial +approximation and cross-validation as resampling technique to evaluate which +model fits the data best. + +At the end, you should present a critical evaluation of your results +and discuss the applicability of these regression methods to the type +of data presented here (either the terrain data we propose or other data sets). + +\subsection{Background literature} + +\begin{enumerate} +\item For a discussion and derivation of the variances and mean squared errors using linear regression, see the \href{{https://arxiv.org/abs/1509.09169}}{Lecture notes on ridge regression by Wessel N. van Wieringen} + +\item The textbook of \href{{https://www.springer.com/gp/book/9780387848570}}{Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer}, chapters 3 and 7 are the most relevant ones for the analysis here. +\end{enumerate} + +\noindent +\subsection{Introduction to numerical projects} + +Here follows a brief recipe and recommendation on how to answer the various questions when preparing your answers. + +\begin{itemize} + \item Give a short description of the nature of the problem and the eventual numerical methods you have used. + + \item Describe the algorithm you have used and/or developed. Here you may find it convenient to use pseudocoding. In many cases you can describe the algorithm in the program itself. + + \item Include the source code of your program. Comment your program properly. You should have the code at your GitHub/GitLab link. You can also place the code in an appendix of your report. + + \item If possible, try to find analytic solutions, or known limits in order to test your program when developing the code. + + \item Include your results either in figure form or in a table. Remember to label your results. All tables and figures should have relevant captions and labels on the axes. + + \item Try to evaluate the reliabilty and numerical stability/precision of your results. If possible, include a qualitative and/or quantitative discussion of the numerical stability, eventual loss of precision etc. + + \item Try to give an interpretation of you results in your answers to the problems. + + \item Critique: if possible include your comments and reflections about the exercise, whether you felt you learnt something, ideas for improvements and other thoughts you've made when solving the exercise. We wish to keep this course at the interactive level and your comments can help us improve it. + + \item Try to establish a practice where you log your work at the computerlab. You may find such a logbook very handy at later stages in your work, especially when you don't properly remember what a previous test version of your program did. Here you could also record the time spent on solving the exercise, various algorithms you may have tested or other topics which you feel worthy of mentioning. +\end{itemize} + +\noindent +\subsection{Format for electronic delivery of report and programs} + +The preferred format for the report is a PDF file. You can also use DOC or postscript formats or as an ipython notebook file. As programming language we prefer that you choose between C/C++, Fortran2008, Julia or Python. The following prescription should be followed when preparing the report: + +\begin{itemize} + \item Use Canvas to hand in your projects, log in at \href{{https://www.uio.no/english/services/it/education/canvas/}}{\nolinkurl{https://www.uio.no/english/services/it/education/canvas/}} with your normal UiO username and password. + + \item Upload \textbf{only} the report file or the link to your GitHub/GitLab or similar typo of repos! For the source code file(s) you have developed please provide us with your link to your GitHub/GitLab or similar domain. The report file should include all of your discussions and a list of the codes you have developed. Do not include library files which are available at the course homepage, unless you have made specific changes to them. + + \item In your GitHub/GitLab or similar repository, please include a folder which contains selected results. These can be in the form of output from your code for a selected set of runs and input parameters. +\end{itemize} + +\noindent +Finally, +we encourage you to collaborate. Optimal working groups consist of +2-3 students. You can then hand in a common report. + +\subsection{Software and needed installations} + +If you have Python installed (we recommend Python3) and you feel pretty familiar with installing different packages, +we recommend that you install the following Python packages via \textbf{pip} as +\begin{enumerate} +\item pip install numpy scipy matplotlib ipython scikit-learn tensorflow sympy pandas pillow +\end{enumerate} + +\noindent +For Python3, replace \textbf{pip} with \textbf{pip3}. + +See below for a discussion of \textbf{tensorflow} and \textbf{scikit-learn}. + +For OSX users we recommend also, after having installed Xcode, to install \textbf{brew}. Brew allows +for a seamless installation of additional software via for example +\begin{enumerate} +\item brew install python3 +\end{enumerate} + +\noindent +For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution +you can use \textbf{pip} as well and simply install Python as +\begin{enumerate} +\item sudo apt-get install python3 (or python for python2.7) +\end{enumerate} + +\noindent +etc etc. + +If you don't want to install various Python packages with their dependencies separately, we recommend two widely used distrubutions which set up all relevant dependencies for Python, namely +\begin{enumerate} +\item \href{{https://docs.anaconda.com/}}{Anaconda} Anaconda is an open source distribution of the Python and R programming languages for large-scale data processing, predictive analytics, and scientific computing, that aims to simplify package management and deployment. Package versions are managed by the package management system \textbf{conda} + +\item \href{{https://www.enthought.com/product/canopy/}}{Enthought canopy} is a Python distribution for scientific and analytic computing distribution and analysis environment, available for free and under a commercial license. +\end{enumerate} + +\noindent +Popular software packages written in Python for ML are + +\begin{itemize} +\item \href{{http://scikit-learn.org/stable/}}{Scikit-learn}, + +\item \href{{https://www.tensorflow.org/}}{Tensorflow}, + +\item \href{{http://pytorch.org/}}{PyTorch} and + +\item \href{{https://keras.io/}}{Keras}. +\end{itemize} + +\noindent +These are all freely available at their respective GitHub sites. They +encompass communities of developers in the thousands or more. And the number +of code developers and contributors keeps increasing. + + +% ------------------- end of main content --------------- + +% #ifdef PREAMBLE +\end{document} +% #endif + diff --git a/doc/Projects/2023/Project1/pdf/Project1.pdf b/doc/Projects/2023/Project1/pdf/Project1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..71e34f984f7a32800985f3a3ec86dbbee5bd53df GIT binary patch literal 269564 zcma&MQS>*RteZG@SWqO(^np2Y?t&3q*vOcd#VT~$X?Q(S9L|E54dLn(^eOB{q zt|6zrsqNq2muX^Vjv8OR?0a0VazESwe&bnI;4VOK!7>g=UZL-kxYQ(!P*c4Q>(I6+ zE%<*{I~HivgiW{OY1G~*atkN}JWJVM1UBtE4*>!YE~RB?-+A>H$IZJt<1TWQypl0H zIw^UNnOL1GpGtQZ<`?+C@~=)eJYg*zfxE0~@@Y!Rpkm2h*R(Lj%?Bj>*AyjncAF&1 z7x%SF5^Os|+14F*nqh?=-aE|>Z0QU=hUocPv$!l#y`j0}=mWj`!}Qd2UwW%Ae`mg{ zcmE+$C2TqMMEf3vMDlSlgPKZ|Cs~=}dJfOzs+a_UwRgPdpc;KzbhKO05$U5wx9`}D4K zq(B)6w|cvr;9*5Vn8k!tzXpBzvRPnJTWKX>k$#1Z`Lx}=%I`I@4S+i?J*$Kh9=!_w zr1S!c@-rL9dBIu6Z987$`s4lV$f)k}D=D$?iAF#3og`5`KWrF+OL&U{CdqHdy;Ock zczsI$ZHEA88V`KeHehyUOysTL7~P4@xw=9y5e^q8nuXUTDDk)dx!9fCE0Nh zES02=yFH{jjTGkY@WevGqcpRjkoKD~<ow$xP6!`@+qDR_C{zMbavK^R=Zc$44lIHx*grl09&O%z zC3K(If<9zV2o-Vj$ZH**2@OGq?n( z)h3?Z`}Pb_DK#;<%SEflY>^sdFfICE9&$88DQ$Cfyk*+NKBVMYK{kro#uV2{lpEnz z6&YXUG_s)aeuq$EYn5ggGV`t%k=iB#1u=@iBE{q zUTK3x>EbCV<}z@LQ&;gU9>xNOFoHH{4JCrW{sI+uMb?6mwm#CKqUMgoU{4G)9LgUKwLzKFc>heiNUKg-x zp7L9FPFJ{wTs5~s>ditLuF3AJ^v^9!4b38-opv^-CbQ-?4L_QxkVNN{17E&wWN0M_ zGsj4Iro8vxRdYn!-z>4A1KJ)$cWc1L$lny@SdDaI)=HO3s278yKc6P+QmqEM!c;fe zocFp-6O*{&(GRj3UDg<(puyfGxK`>s5+V-c@RU(E?>{iQCD|vZmFw?D`NjX3Rn6Ef zp54{Z`>Q?};2-vdAIdVR;fqCU*)?>1vxrPGu!A4BNp(so!=_28E$&7!VGCWOvvQz} z7+Qg*tw+YX&APL-tpfoidkRrNF%v}@D&ka-sC>{;7R?U1_4#~}CSF?;rUGl1G;E?~ zSb}jV$h$t%v+yPz+2wv)q6zVUTqVYkBXgWXL=L(wu6!KuPEc>%@%H#2WOEE*W&oc^^jW%Oz+IjAQ4aH}saY z4a;r4lb&&&((lqhhhZGS)k?~jXCBZ#cNnivE(*ad2ERk}^Vt|1Dt4&DR^}UBp}&yO z`~tU|RMHe`!%=v9B(~&5g3gzOSSRht&&s=&R5J;aB=EsaXNFU9Iaa@|CuL`Unkcno zrr%OqTQ_-p@w%(}N%ua^w5el`a1p%TCy!`pJyKJV3cyouxAD|ou<~6xPDC*pYDh(8 z1>)=1sL)DD2E6R6>b;JL81?vE>vp**V;`j1DWg>|SoyLqs@en zylYy|iSyN;=D@w)>K5V&0Q=tvv2K4Gq0Q_~|38-b@Am(45Hsul0z@1f9RIgKbc(%Q zzdio%+ZQZ|KeV~-SzpQ?Sb)<~K4@4h`J>RwF*X>TDyoT;n)H<9`I@JtB~>z7BAJ0f z%wV){cK&9^!+k77Y%pYC*Xj5B-yz_%Z#l?}Jz43V5drjAcLWzflrYuUDHSv+Dod5z zQxoED7$K!fb{N_JxY72iT)HErN|~6J{fVxIwxMz}_3iXW!4WXf6TMg`6TvW}!ImT2mVW4vA{xdO#D`v!CedC!GxZrU073&2 z-fF~5N-zdU@nXw))m)kt5_;@URNa`mHG`ut=d$zB)j%JQg47(dG~ zv(J_2^2F7$*6yb0)Xp!R=pdzA2;4a>FJbh|#Bh-l`Wb*^C>w|B~ z7)Gte4uy|*z`Zz2-f4-}+xMB)TRngwoEmslW?|4{H%bfiE_PpIg}Hj@8D-|dCrE0^ z6MNlhCMnH;za12dRY{ZiLc~Roq%d3yfcgVwFF~5$I4UX;m9X4esOq1TZPUiMY!!m& zv=JPq=3HE=NlJ|1S&u@+g~Hw~MT1L7UnN$Jxq(3vjpYdGZd0_nw1Q#_JrpX%uTveN zRAp^qpIH4tt{lfE6WS$KJ`v+R=*uG9X?!n4+x#Q~m zWZdEb-^rwmxZK_ZkWE6MIp{7hV&V0 zRE4#Le3#o%3o@c&Jng@~%^V4p`FDYK3h`^7oXF#n3ZNc=gzh@r_?IuL2 zV`DKsunkV|(WyP9+hbS5UG7dDQVkh&b3Jq93yNy$h4R%x%baZ9sBD7dJYu5?h z$rc}&tVwvOVAv`!2TmTG`T_qth=jsP=~`$HMQU)WyL8UzL1mhSbmmRH23xBxq=$(I z6`iD4K=bAke6?#YNR$tI(L0x?_mmZXt9;(er4JTmvf_Jhe7DI z;f4QLJeyuMQmKz|6$_cHfbgNsHj{Y@&UdBNgk9>NnD~kI>*$36`N8h!wZNE!1PQ|- zpykpfGI8u2Af5RR+b5sdoy!D8k>Hx9EP`RfwM2Vqt&*CAVM1l+JGrBO*DTFEbP5=n znkc7+)}!DCzqRKC9vOE=P3gOY*!C}6Iw)U(|2r7bLrE8N#30^zcl+jp&FfFkLxsCv z>`B;V!(TllkrH&jC@~ZqkV8s7D;0E*fPz$Cd0elV#~m|g+cwi{uaGhN;wg1g4Z*oH zSrp`~n$2;g0vq^jbvy7DHKcIu+NYP)(zQLe2^yZ!m?QV^oPMH@8mB*>1Hg_&lwKwA zys=j?T?$9JSj;!_(Hr&}r=_;`{IoKPdC)kioB^E=Nsk?G3m!wdGqA0KOwSSc)JcKW zU;huV6so1P;O-WO1=8br7;EqVWm)^6)$(6IL_m$PS-!qY2-L&+E#+7dI-|i&?hx0) z@pWrmL5d#2x2v50mAA6*prQ|x#XD9~^NgSu3>%BUGZVE!!SMU1a}LLD+qwM#H37}K zTrKUi9g+Pp{|kf|iXEcp(M2$j%a#@^*jYZ1_uXa+!p34+xtBeYq=0p>1wW}9*Hqb^D%z9c7G8w-G*JbiMZ3PB4T*ALZ$Avdc7f8(RcoCP%Ac5blxd2j0i>D>fIK_SJe>Dq z%EnErGnAl;dzZ*~F%?`u;WU_MAm`*nJ<95mv5}?A`<~)72R`z4m_uvRtgN{)oH}I2 z;v60HT76mPq=Mm9*YNU=o=L#$fLPiP_I-5-vRl+A*_em$;y9w!NhaVsB~KhTzMKO_ z4)+m_`B%2&wMn}`ICz99I+pDWCHra7Px)n$ME8Y^q2Kh6T&$!@=IS{pNHw+FPo|Fs)#7j<*U@^6{orWl_7#)v0mj@H{$F( z3@Tz1oXqnA)rc<-hmxRa@TbqbvjZRblMFv<*TSeb7>SAuy>{6IJeN-+d^(Iu+m6h}`ZgqpyXEwvD2Zxu`llaS&qnD3qqonL#3Ev+g@BFj8bE~~34{n}32lR4u z|09sgdqyXO*3GH6UE90`L&uMkDQT?{+Se`1f>R)%tQDr0zpT)o;5qYrNvV<CjTSksHXwX7!fabvXAYnWg&@{uj{qGqeeU6*i%f4MI%Ea1R zfpUsN28hMc@U4~?0g%6C8?{>`1groil*O4IIGxHBx<5}BWqo{X%PZWt&kxfRFs&QK zMN9T-R~bcxdc2N#PUS@hhuv)b)C(0_P4gU>Hj>ebfcoiiiNLPkbzv7@(}R!!55b^C zH#dCQqJxfNzrcp+XH)A^C?cNk9;%tuiQ=kFY^lmv^XTXHT5Riqz4NvYnkvFf=s{E8 zq2(~`3OHg-G=hy6!JaXrSlPE-Vb&|-$oUs%-GKg(P4C!`4ed45mITW?5mbk0xqNkp zX|1i)RT%6Z2(q%d?eb#q@+ulxd-vH;}?-*M*^PRZ%{ct>Bv;KI8-KB5XnkV*KG3*imRQ z|MD(Iiu;%Xg2yk|{UV;y{{&?gmj4BmS(v&0uR&Q`$6a(tTx1P{%YRkVYB32le$&iM~+{1yLaz?Q99~Tw1tbigT=00`9 z^=amvJCysltNv-NN4L^h4WHrJp|0D%<*;<6SC+`6tMN{0kA!vk*!tG-=$zHK}XEHFEO-sM~Ct=hkvTw(O6HEDK{<*s+Gp1bzVeRtP^Mvg1BkzAbn!% z#s!lr_fqAk>7@$j5#8}HaJ;@3_a#%X4j6QlYN_!QJf@#Kw;_f|hsa7fQfDtho#u{G zXWfJTO9~tVr`=oNEGTC)&L>-9ZPUGZ*xuBoWc%))k7lkT*gfmvwd1{g%0c9Vj(FEZ9r5xxz}=bCuv4-ehtUugczd(vs-i0rOs3(% zu0ls?`H>|hc9?l7h@-ymH|i}lES^wliK5Ico<((Vab+P-&syO;US{P&RT+1IJ`pBM zFS(AP$RGuHpU=3I2;)zKGA<_0WzlJhi$hfUB}2Wa4=^r|_KZ@oBIvvf5de+N$9E zn#0UVDy?-De$6DS>wF@hyY(#W;budoZxD*=)XmFYdsk>_!mVcDXW|g=^ImPiED1@g ztuG6-(k<*aGsS~|TNukM!iu$NcVe2Uvv=s=qWtP-(dMc>`oG)kq95u>rh1Mo{cJks zqs!}J2s#$RWGc8VL_OzzxsUkeh<7WBq2ii01pP}Eyo#GhgS|ADkrq+t5jx*q1WB;X z)f~KoR7)YD2l26{>IFt4gN9SRg*Km-*1TJ)W%D><#$d^-XA5Wv6t?C6h=4>_!Gd9e zctSYFrH2n^QQ8hfNFf&KDy=tAc0onhed9~DDEOm3&mqMb7m19j*e&!qt{nQCM$4~do+Q!C++g72k|*Ww6^2o)bdaouO2znqi%lVs*$R$IAU!`OdOyNa1F$(o*liH5 zbZZRZz4C_Wp_EaIxOhKak>*S?ks*U9tAXbfu11N!hWWx{bHc>J#cMz~-_K98f)H6~ zG^rYuNrm&8dh=-rKUx*Ye5Lag#12+gGL~TR)rZV%UmsrDVti6>jF-;#06rH%TX`#GcxzwQiF_mqV zpYO2r>lX_Fm2?e+oOiW+*PV|3eTTPuP=QB2#}sfr;LI*G4Zg( zU<}7FbvG#cRRB@&Rcqix@;rh8zKe*g${61h(i%y`_*?uhGI?Cp$r<}`?8OAX^)C7Y z^CX7_+#IPk5Pz;SA$XU%!7pa}{m!jCqfYsEX!H)(Ewlgn_I%!+{h6WqR5GdjkBRc7 z3rY()^IpQoT;%Rq^j0|HZ1s8X;DQ$sky0yy$ge|Srgdc&$k-H9qLrqe$|uoK==pj= z5FZ7sigXl4T+mblGx+p>=Z%@kfl_sZa$x~BUbbDmmWe&xo(jc=ej-}TWb#lGJs z?ct8VLErD6_+=a#d&c|OnSBbh&ah(l6nhrESkjc7Jv33;<_<200EGblmXS}HatiD# zJdW3L*N-dam0db3L1o2$+oIIUUIqF0;fKygWj5cOXUF%%+^a$F)QTLvZILuD_pytb z_8%ngl=CV|Iq`F8ZO16}G^wJv>=kGU3txMMiM_n+UBwQ#m9;&(`iI88WV2=b{G1m} z4G8Sr9)vs>&!1Qx>1f_Cls2U1Scvl1`jI))q+I-a4u$aaekYfhoX0iBQJv4j*Q(`$ zcE?_k?y(nX8=MWek<}sy_(ye7Mu3*v>JnW2bym@Ps6bMY8Ih_!t>oMj3~d`%+6 z7yoe6f==&L6#td^dRMgviPdB^?SoVlF#`wJ7xh%KRZriwVZSmdNSP6y?t&5yms+gx zeJ|q_$L+rh?2|=7nS!R>d~R(~>quE-(Ll*l#+IEuLK-f-z{v_+?4gobBjf8T6*MM6U9$SOA-oF5vF~ zXF=tTnm%bEh#4$_Xw!&{#}#YB=92k|di?Pzc1p0?;j#oc2D?5mZ>C|385emzkDai5 zC35Nw>?iPZr*geR_A(^$Rad<92CyGs7l^f3=V3lvSZIG*{E;l@XLQ(%p4Z?g~OrtK-J%}eei!edIE?4 zNF6x;H;6!q6tGwZ;@^6;EwiFqb!*?PY}lm0{XainTsI6<|M8-XH}mrbPJpdfFyAF? zGN7|>1~IMi5F8b{7`*nV?5B(t+sLL&eA017TIgRo0deo4zxtHMBkz2XokM8L@o~nu*oF4Ic}St?5}akGX_nh@X@|M_x=l5w~J0*7usue|ZrmAn|Ii zxJxV`+&s=tlEcSUoG$_i0q$jw7mLRcR^cZA*Wup@9Xif<_p1mMez$Lb2+A8DhkTs( z(RR>3F*%n-SgqBChR=!9#Zj5HC7#p=U-7>m1nrVB+3zV#UKLt8gMNmQii>pehv~?Q zv(nL%c_Q+^`^9_6`HM>7L^DL-PEr5|YZqW8I*Zsl!PnxZ<{27O2OWy8Xvtzg&4mW9 zpuJm29ngOeda?)idC^fTlQ4xZjw5eTroy~|Jz@;G`u%uzqnkhw)MO`8+g9unn|z6E z(lMl2l+0ttKA@-WzgUyLO`cZG9^6F!lOE^z;KYNZyHUhABL=#@cMYcu9x-}cE;EhB zS%@S?4G{6F3eQd0J3+^ZRA+?D$?^CT(tH$p{^K$>iv!KGXE6QVm#AP>!u01cf4Tef zXq$;wV#FK|G2($wQ^c4I$^<-!QbeSS{g1;2m!@rp*dLd}oChP^n;=7b&k)_Lf=s9l z86+aPD~rt8)L7WmYITR)X#EQ52?xu~M%SZ-a$pPP$Py|34neraw@Jr?zTuubLUUmw zFD|E2ZD>RyK@kQcl;!u-^DWOx1fv~L^DiedvlNs$f@@O1GH{s??k{yAtub1t&gY+P zIoTugW?)D0ckVh~4Z{tZb%Duw^EwG9LWAbPo!c1UlIhWwRTAan%v<`8ZfLrJjJ4>v zrWn{a3__6`97Dco1m|&kY1|3CHDWM3U-;-e{}?a6!nZ7AHUr_X{v|Lu&8Z7(9kAvf zx2z?hN5N>G_7&!SA|9S4eDj?CZ4ZOvPTW8pld!^2K{fEMPX-+#kz~s--dh5 z-S6XHz?1ijtvXoh{-E--OPJ_0}vSr(jtkEnD^@|NUo$HHHJQ)m%Q=1+g=geMT|EZkZ8w{M_cNm#j2?V<(Qs2>h~|8 z25SnSg_ukl;PXp4?>R^t>WYS?wP?TnRLyo<%!|Kn+@D`>N_Yc5(Q>ngN=%F!=)M?> zE?j3$r=7G5k_UI#0H4^{gFgtsb5P@`;f5I0m`o;vL4~EQ-mdW*l_u{3AA}V4%7=np zO>|OdktnlvryXq9I@YfzGGYM6?H^=-JjluwF*h}8Qj|=~5SB70qVW9lRUVy8*3=;| zGL7Wp9zQ#&2I7X;7-d8z5)xG%iMU3>oeH|(p!=JhU7MMuAv^9zn-e(wK=w&ASQETS zutN!4cOj0Ikd$LWU_mZFIr}s$`<<7SR*t%=()V%Bvxo<724~Xq5=6+V!4EXqcA;*- zQfULcSMw=DhE*msj0Iz+@f`ooyk3L4!UutuUo z%RMc7qQy>e9M(PN>QxmM^yS`;%e7Y2!dDzGk-}*8fY&YS{O?>F8?)H7R!J?^oZBG( z02f2#nsq{NW#BM>x_2=bw1nSO@KN>3fFe@2e_4wZCd|LLVd4bj1-jY^5-Vd_QY(Ne zKqGKm&bqvEArzaS^1qLDF#ZDw)3j7K6e@Fi)h_eu80ZOc>p_a9dp}%aV#A1YDR9s8 z$WHHBWr7H=k;gf8B6_3wbe^B7VvS z$$lsSdxEJ>V+cQlc6S#l=6Pila=uh*jptU`6`e*K3`I`02aqiU|4u zt4eOK&6la5_8|JBeV~D;q%te|?i)q$x`?tK{aU-Bl@e{{8jd_y@AhS)+Dops3ySc3 zI|N2S;`?j!(mQY)*h*EfykF7sSg)So-e|(j*U6jf^jypdT>@)Il1fY~Rmkl7lUP$^ zzaSoAcCYoMH(~2_HPcaXp=rxy#>O{VDS9T-G}qU@ZpSj8;H!kq*N~G00MjYJDCEmU0HHCtshl;!oA}!%eR3 zTox@G`G_+r#p_aWO=3(abKP(vd>3J%(1gomRs9ZVv)eUaR7B5!N%_`Ps`4Lg9HtiO zbxedHu9%dc=3yWj<(#;lOjo-cmVPPCr!n>^Qr3Sr@ISV zmRryY7-J=mmfH*G&(I=$^P!q zO->WJh^9Rjvm0o(1CW@v(3bK7ME}RA*d@JSfARr?l}z2iH(?|F4-{}nD z{FMGABvI0Jf$)y2E{>2Vjr4&rEWQSC z+{m$Gaw6qfH*5o0|3^Ixeg$iuN3U}wa4R+%A@lLpN5YXsn!kGJtp;X^6V(586YB2* z+1k;>j@2CT<9>&R7}xvj-<+`vS^6Tf=Ek(n= zbQ`r+R+O!q&%lu~^O#1~d{&~jTaVs1OV{o?hYgjPb8UG2yXWKGc7y%^-Kc3AdMzbt zd+Xys1y+k+R{HXSrT3QOJ#4Ll$Q>tYjex*@EmO%M+}cu}F3!JmNlO`TzOBTMfQw;0 z8?EYP!YdlL)0F$%f4EOTNGu1CjS$|pl~jxHjI@A>Wm!R+bE9f1tKV*IDbAnQbcTII z=j2b6^_OezE7|<$nVH2T3)#OWpo}|iF{O5L^~O>wzRZjshDAtdQ+!P}joOf!5KP{) zKU)@<3*PHo>|s2Q2+?Z3lb*4V881^wP@AP{mWd0cx712NG=5MeyNdS4)#6a9GunYdr0WY&O(+Yjhs(;5Xc)y3Q7%JFsgd@W=&^vSs~kwBcM&75 zg4uW;3+4X(4X7p2jUW5S0;$9RCaFajt)?xM1iGTAdUxKh5$mI>!%c40K34-6CsxJ< zOcux(l>YlQ2blk_Sm`v|itsP+%o>AM`v)E^536|dO%mIsYNFg~D$`&TaAX&qm?;U^ zKBdO>RIP%E7e|T)+gf+8`+7Jlr+;HpahJ%1>ehN`O;uB-SBJUz0)wF!SrAu8@OQI z#o+xD%OXnjZ&}?hTevvdO*ltkpw-=?Wak2z&w4e z$w?E|%adzuO**0Z5jl2wg_C$amr_?L&^iS&Ht{UOL2vlEfhtz6n4o}lY&i+Cu?OGq zUK`NhM!AtU5>uy|FFGYZ;T@p~^9-3R+1Wj@==zhy2u3LG z*qT8_K+`N#lodEwR_ZY1$-24xP!wzHc)*I+S=KPwj{aDtFfWRUet4SZh2|r26k5FC z@}!9uYpqh7O`nGv@6v#eKHc4~LLU5GS-UW$VcM$U_sF_pT1xY4^40BsCu4cNJNJ9dd0vZR~%I z-j*?;lA&ubiYPf)P|L&mCB_>mly*1ML&bIwO8$ZF#;-f9-Y9B}6Uvxuk|BL>!`;+g z!Hx-&;v9z;?A0dSlNzw94g5H!R6*&QOfRZ!d;2w8h5`YW(S3u8G$tZMRsa_xiVm}9 zpY89Ly05hK4@o>_%IbyP2%Vp2J4qC(b52w3W=V${OTm72+=A3%NK#diUbym4+X&~k zDPFGsCS!M(s49e}=(F;dY45$;bqM5Mj#|?wLin2W+8U0zF%hDmJ83;9SyG`j__0i+ zx5=wFMc7ZYKsg#@>pW z${pF2xq1L#i2>WTvPnyyEU1qplfh}zR)qp7%Z=P&TrW@Ct2`MxJSV;pDe3g?h>)&@ z)}a${$gO8JK&7lCzM{D54VJ5H+;D}E5GYQ$TuxtHmy9&6uTv`>ymIp4q%!yN=hogb;e(NUg zVngL~d+4w4UV9JD>fmu>1egrmKuUrX`5S-5U?VNxl)_9d4c=zy%L-toZyr7-iQ9r3 z2GoqGIFNwu)&50_m%74?G>pZ9hg+bv1*y)q>cR$FA1kKl!>C32EE~ zTHTBVWmq+{wU}*lN%=V2z^#tgul2wGY~N(^MwZN-O<|#q?G;I0`5Rs+o3edYzmGF{ zTixztQp(d3GnL(4K7LLIn;KV)NitXz7d`_5YuVtHu zS?=^~vFTngiYO}&3^D6#q!L7&PeBqDxt-Qxn4_8YP1)b+O3&uS(DkcC%o4d8^nqJ) zXVqT$)Pdfg+s7kKDxlQS>A@~^u8TzW4aBK9E7XMk1s8i=gV9jq^*p5HnNu!=ck#r3 z)Lka`8S&B+9dg^4)@N^`Mq59g z>}cF}AHl?u0wUAZ2)0TSNnJRNx&X~Z$VqH$hn2yffNtf1IOm0IUsT@AfmGc1HOVej z&ewlnhOEzbD1^8HHha224B8zv)VQ|JXtLVXVfS5B^u#)Bl`oQf z`rCp>Af?$=RIR)BnvwoMH;wr((D1Kf_oa(<2J69k7#ytYc6lA8n1a%5ce+~kQ{+}& z-NwChZN&G~S)UFzW(|wT1gMI&j@`trPQe%5>NRi4TK#FCx97!;+uXz%r}KD|)OOKY~@EBB*0J@pv(N7q@8c+aB07PcJ6)%CC$A7OI!`CFmVUI0lI9M7C1+sYv zz`!w&im`Ynt9t8Bj!N^p_k~2hro$ZYIa%70G$wE^Ii#9B@OGZdgjdA=N3Du{j0gL4 zR}038@ltm|tAzM%7S6Xndmoy$DTl!3*!!vScRc&H{C>eAYB+{chxZjD+QR-)wiJyS z)_f6dz*%}sOP$kh*IVFI%$yi8-EC-uE`83F)b7Z_$Ud&(x_UkdzwT{^tD#pQPv^Ya z&eU1p z-`zg}OUIyOGtNWpvOpgUHdn^;nQet1i2odAhX1J#*#9?uz{Jk-f2$8xwPg~H|J&Ux zG!Q1nfHL@OC?vDL81nS^55gzk{E=_e2gQv}&MXtnDm8bff4{A|x;m5jTbox4a>9IN zkKR*VbNx@vsn`%hbcll^>GS&vaQ4#1Bs8r|BJ0=z2YODM#LdWTM7%n*OYYh8;r$H& zd~UCE<0Fj=k{hAO+`E*L6ibapiZ_$4<)1l-G9HeTfWz{QVZi86mWwhs9$E{bjwsWM z7vH;49+^HPOM&D!ioz5|p&>cy_ItT`JkG#5#{=U-l4L168LEyRje@=(p569PZVZmx zg+xxVLgCY05-rk-IaMFb%^9TsJ@-h?r5cGkUdrPYmomvBB$6bP8&n}rHWfybAfX;p zE{aGIlbAADCwCg18Pwx+^V3GC z0Es#!0j4YiEXOHD?!}uX`3R<$96~wXRe}Pa}$_wCd66ZbG3s}Elu&2s+2K~Q=w9Y!qSrjh0&QrDratMz9-~v;xUK$BoT*Xzf=nm5*n&h?*d`iI$IPHl%ck?;+;mDD#}| z&+>Kn0!7CQnDM1sUfr|PBS=!oNSj!5xPGZfVI^Zcujwt^SRg)3AR&v9L%tw2L&^ri zhuEAR<-8p{P~4u`Kt-FIsi&is2(@2CX?M%CP%~?QEg#fHdfY)eKw88$s1-qT;>|QD zL&BpS578PITg%TW#yfF(aA6Y=#BGtJa1E6Er)aJRr424fkKz-si+fgQJ?stphSPSzJGG_4^m#6kr@0L;|7HE_Z$N}W|OX8)61rjo>O zTrMD~iZ5KN$XbH;$l>W|avnXLd^$L5eE_@MSDE>^NY2qW0Z z%t=$BT@RK2mf;AG>C+a92ZU3#KF@UyqN_1n^*JDy`Bq^`$+^T%4pi&u4+vz$nE3^WzM zOE?{WOHw=mY^WI`1?g z-AT3cw~v@ExAoqH^2pCFZfV4<$*W0=1D+bps1*zt=ZG=a;kXAkw!}J~Pj6|u!1G7k z4tJr2MhUD{ptIWLoO{a#Zujbwf2y8}ziWzjve7wk=0Ah)8X4x}_bqnX&O_1{NSn}O zm~56F8c;h?{W#9iczj4|J$q`!`=|3?f*?-pFrryRj5jWdRNaF~)g!Jav0ogUwdSlk zPPsgl2K$#&H-mxH)PPdbAb#=$_zc7?#V)$B{UF?Om*;qeRDN+ZqX=$}d<=KiLZUZO zPq;cTS+tyKtUM$pqHi+CcpcR}_l0(M#PxJ`z?*w5g1M#=E(*a`sJqBX!Ef}FpFk1$ zJa*32K%)*0hv&FFxLiZ~x6^LMr#k0AYjq7n9D%)rNNxsnq@gEE$Y7XK)!nOAJc4CL z5YzEr@vUW)>2MnhcRI=H9>0S@jTY_WY!SZ&cOKJ#gpmej+h{dF`@mm2KGGi}YFDeI zyhm^@_BK}kMr7%Bx6gzPQu~MJqu^F<(Bw*P&`?L?Fgbp)!4$SzIu}gUZ(;;;-7ZiF zU>8$92C=R(6m{!>87r$vy7E?Vp7QJnDNm%wY5!;aJAbdp`w9FsG7~ym?;s-J8@w?% z|4%>zf@1*4rYSL?|62a`yraYNrh+fDtk;Z3@bcV8eXZkVlt{ri{ZBiQC&2ql!2G?R z7hM2VD@5F}r|NN=Ai9@_W30fsF=#trr(nj~rRNTQ)vkX*-Vt@m@daX0x)a__6lggf zKNn|Zw%e(1x0_|TBjOsP&ZLrYFun8;bno8Kc4!p+g9p)D3sh=P%xy}8-NFj3j{PE~ zpkV|QeZYRugfM+%)l{v^i4V1|GmcS`kDXTcE*jTkn_$LZ9?ZK)KG3t}5(?hUrR|y& z2-u}yEP@mCldn$Ybb-7fS57o=vD`fZ>GFySsPA#(68{+4G;XC!_@VJXo@fh(hFIMFjLMD#@u>24 zc6ZAT1#hlsVm#i}p(Be~Gr5eX+~6A{}lV#0@->UZzQ zH0slWnhrXlO_2q7@ZTKvSpzPb{VGEYehSxo`CwP^6FI4MXTyB^5mb@gm_J}}+rsJX z6Lz@gSA#J~r9TCo#FDkHxCI7{LNYNMg@`pgKd?pkYLzhc23BY79_qTn?Of@%;4!p1 zX0G>xgtlZ#uwpu4@=dg`=YS>GQlZs}L{|*S_NwoA+a+%+negws*5L)z?mSD;eWN=u zL3SgCkWq?18(8k0oPIXL*kNmRx-1rB@}H+}i-n-N_=Ju&q^^9g0|wV_^=cDKZFDdF z;3#^EOMs}bpd^qM$5LfD{vXE9AxaQnS)y(Cv~AnAZQHi(p0;hfQC_1?o+TQA zT2O-lG#b#LYXa{=q3S#2-d{wk@ma^YR{$C)$mkgX%e_Mgc^)Vs#!Wa8HKSyx+{c*cw?plFCtrDyX6 zVH$^6l;S>qLn^0v8G-q`^P-jUsV;Mmfm${f*v~6O=iUdd#t+K2_L71mZ=eMsVQj?d zLlbouBHMFzNTt;r0K6Q-SeV-~G_*NqoFt%sqRA(0GYzH7_k!0phY>l|0&i}rCB&~a zuTg)}XM(kv81{5g-?v4~?)u&SyJh~2O9=tQSB=zufh4NZUb}8{87Fwoclf>nFMgF6 zZubQF~j=9ML`lZ3|GUj<-HFCSjEW%;2W)y92PoTb$su9 zvr!oupA|l)BY7Gqji(5f*fK^i(uvr!nfm_Azqb4*u*^4f-NCvP)^fD!BGWNih@&F3 zv;v&I@CC=>zDET8C3AA)C+=-lfp_4e0W}c8{@uO;ppxxs>n5+G8qYSDin_g@4$!CY zZ8!+E7Z?(QpJE2_H&)$L@cv%~u-g@#0P~-v3LE`@jxrb-*ct!VQYBhlt9G*u;om5O zelB#3Uplg*vX8fDJI6Uv2I&ZW4KyeZ7h^{L;W%s22a#@Yf9q{J zYGBwGfp(`G>isiyy7E~ki&|2_RH0?<3N@b`8q#ZDkk(i+Oj|UPdRhdzG*YJZjpdEL zOQYtX4d&@jj|6&FiULL&k<6dJ8x(T)ZBY^Y^KIq!53m*}i7hJCDU53iOBJfig&T

gew$WK=mE;m4+SICw5saV*H1NO6&Y`i>;};t>yJ{|lUeApcA-b=h&+O#QHpu`1+ZIdF&V**#9)c0{9j?N}B3X&Je9!vm^yL^W z=EDJ|h5O5)^$RhCZY@R2*%;>Gr!RHd{nE>rHad$tc`qAJc40tSnTbyOn^dPc=a>Z| zz;l)+bZvMv)_E_d8AI`71Mcy`#p1-BM^0%WTy>=0B-k=aPc5G@(V{sPBlS=E&t+*v zy9+aGWjQG=d1EQ9u|*mC)uW=TWs-xT;41Uxs&jgZ&uvGb^i7RQM$wtky-3nEw#ktq zIVh1$Qz8lRwpy?Os?GM)setO1_^n6rZqU@XV2k0$l+!{Fl-31m;Jn7>=Kv1{hvcXh%DNMdzZM##{U_g!$w3Zxu4QVk6|Cq8 z66m0}nvVdaZq~V8iV~#3Fty4tt|$tMLO$8Y*$>{Cdxlu10~h`KWWHT1z3 z-#%tcnq7mtLPS+-jg3RnZQ%r1pQZ?li`nxLv~BVPVg)2CCi!QGqUY&n|8oH~9Fy4S z28gtM--*YbOk5TrM$N7|e|l(P4a3B)@1OB4@50Qd>Hi9CoP+Q|I2%MLOJE=z$5a>= zx4rUVkC@%kJeTd2(E0-K@PN6#NRQ?*o_6ncoP~56Ak@oFRQ}i(=t*OTQ(;gPzpev6 zU2!Y`tNHgqi&u!0dOWE8FN~Csppx-sMm$|{T1+Vj5Rq^Nme~HI_=Ga@9s^k~PY%~a z={=v}oRm>u3s|yrA@f=6_P(%?x4P~5owBV9VTaGQidS)T?o^lQtTrBRX<(YLQPYP2 zwk&%`*Y{iuogQ4x2P5X9?{Xmu4r5qrYeUy}jrG?dXz5kRBJue3d&i?0Aeoee*srh` zAD8<;yi+ljJ(#DU<4%I`+hBV+y+CizZEURtXJ7kMeA&JT$iP2310E{6 zWGm#6{J=FJ0uW@WBRzfjaLS0XK*omV0zS3yU^LS)JSvj)wHyVGK-5*d8y1=NjLxb0 zF?}Cds>H-o#PR*1IY^{c7AJDE@M2R57eV6ycQ68ilw&E$3LMDatK?oTMz;$ z8qZz{)%GuL9T#3(-z6IHoRLoejC!uwK1og_)olFdQa>dlK|r)rF}MOjAETx6#mhy3Hqdsp?RLP|U^ zNX{8;6usWwVi-qgwCFo6_I0(Q->^bL(oH`C45*jTEo_4Jw75zzz84a8^~nLnMRXhbJpxrIHfj@$u5a`;i&2Z0v}N&mua~ z2mL{y{7^9^NnSv24Jp9!e^alJ8xt zgW}1JUZB-9<-6Aq`UZF^h}+b(GZkJTb56j&1ciFvmJxibU%bjpttdH5DgxYU-`B?v z8oHqHG!B$ZGX2;cdn1x1ZataO?0JqWiYuC#y_bvzap(*p3oAK6B`*KVPS#QJ*FaqMNKgs7_{hT{sz(o23#q$(g3gR`%~-< zG4*n@{>^ZJbGCxOYsCdTuN8p2Em2$EQ-F`T>R(LpLmhgkpz0CuOg|=^GO9u74>No7 z<(yLK>XPr6|1=e}zlwOL>jybw+)L=?c`jL;*{V~|ugQ|@5uzCD5gdk;+ZSjb*xf+s zGXr|D*9e8kjc`dX5u6Cr5|OG0NA25lYLCZ{#><28{N5!B?GY0b;orusjCZT)$e2O9 zy4L27^+p-=<1TJ3f2}V}dzg1QBCC(Xz}hAWL70{9rQcG2-ehyPI~RgUT`X4#llj~PI@&gX{HuDg1 z51EQR%|vG%C2j}FK%1v>k`Iussz3o%W}_{Up%?dM^LXKgZQJ?uUsMhIe^E6|9RIsP zLtV=5P!#cB19pxES$~l@bGF;2Azcjq%D@mGg4D}g$tIi&xtv+T>+ANlvc^&}>1c{n zT)*#MBJNU0Po-(2reL{~E&j{b>-Xfw2DAEry{ds)W1FwWuy}xwPR&C#D11WoHFLRi zPfMp(1>1XP!}o&qQcJ~HU^8mu!lA+RJNFasgz9#u!dWF)CFkg-gE%VE&;=_spOUJF z<`0@))d773B8C+hHrxYBFr;0jLhzMLm~9v z==$kJ&PgDinapn@Kjs@4x;wzlvl`aHbx^NDE?(X*RtqcRb@k270CGJM1C^Y`%p(Jh zVd-0D=516x5|pAPjBD~(Agh)Y`f&$)B02QIOkqqZP(Oo?hRtL=-=HYo_Dx^rW~8=E z=kKv5v4BG)sg`x?=?4bxVa2e(P!~loI`4;5mvqoIw(3K2UNbV;LCNs9Uz~+k$)J_{ zqf;=)o0bx-Gsl&48e~24FJ*&tsueHC@>g~~4|RhS6%F)#q1U@_S#_N0)O!P;Ok{!C zU%p4qrw^3$xRDpH z!OUABU9Za;n<>-UPOKDyHHq1E3KjIxV+^j{R#UkC{T7Eb42NF6=sL|g)k{|z%+-`B z&;nFLUrzS+&0=ak&<35wW-cFpr-|(dzqK9)a^st8aq8Ovum-|EhuX^j?0oSkb~097 zB=F!}8*bK-lP6WCb(ixMBCQQRe|+H`~Z{ z@xpcGy`KVJxOq+zBz17qq1*?GcjjgL!r9n)vlB91z0K8-6?&hDj6&siw#So)2keqE&1Dikon|Af@~At_r3fLH(9JNBPBLxJZyv!4xudQb zspy7ZNko+zsV>(_V1L`PLeK&EXpZDR1U%{-ZY8x%OOc&)L51ktaw82!V_0rt3n#E2 ze38})3!*!ZiO1z8pAY{8uzrHG~7hj1hbMdj)^NKsz zjp*E788A&Lg0&-c4*m0dwz^Kx7?gpOPYV(^OGQ~wvtF2ZF1shd%*!a_#~D-DNoUT8DI9jU>}p=Si5cIF^6ArG z{4%9cjgdw-$P54ES8iGWdnQyE4%Ow zGbI~^lz1L}6FD$@tw!CJhH`y!nIGelpY|A1G5*eHX zr)P=3nUhNDV<2Ad_ob}bWYBdD?WtuiTraS95*uFX*@>>gcTXU>HxaMG2`0_~WJSbz z4_X!aAS~W`7`!;c5&Orwo1JZr7qh=F3B}IhoLV8sZeC%9Hq{-2u53mlDBN+m!@WNU zB2dT&P7wMlN>oZQA?Pi(Lsl2)Lmx_Fi68@logFV3|zD@m2u0YMobEO|# zYDn7Ch7cBIRX0pl8ceB#d!yFh$yxnebXuF|#$i#Gw2dp@Bf7iE(R~!xih26lOgN2` zYF+hSw*YuHN+>?X6qgiZ`-%`R9aTp7YP9hfRiM9HJNCeYKm(C6nG^%YoSn}rFz;LC zYRlwm(`p9@Ue?3|(WId@6BWN5bcu~|3?g+n@6lQVwf=-YuG+T?1yoKWHu(cUv@=G6 z=%F(#DN#6fEJy#RVM8KTl+}&VUmf#%u8D}ux^+Yy>aP>Ou!LdJA}n4n7rbuDhRy=W z_e=x|RoTo?DzXK{Q=f2TcL%?g1pjy~ND_7k2bRywiUt6-l{~PtFNncALWh17a8^96 zEDaZ2J90i+hb;VU12G@;E}A+6?3lM)aU-qL&7YzGVFXg+hrI?+FT^3-ckJ=R2w^$o z9_-8BotPQy#dU=YfXSjp9Fv)ttk?D{4A4zeSF$N`xp~_`?)Z2Oc|J)OVKHoT@=78` zWV+KcV!Yg(x*h1#iQ}Nah6izy0{!o~gwxwz)lGNWMd~U5f&^m7NpZuoT!8J!+g{)a zEBUN%BcIH|EGt|NYlj`v$>V`^RkKYQ1B+3~Xw&Ark^vWlvpOgKjRr6STBhb9a?OPW zzI0JuF12s9vgksi+i7nG)E-OQ6Xa8-znp-5%CDKOM>#8o?Jlq|{d=?~N;bsTFcYI# zg{j%V{uw+fx7cBTCNbJQg7(_UqbFEm1p2kQ5HgUJFN?0HA|h0VoV(aoX^8xOL_`ut z=zs>?4q$RpdRWPg^5k#IHR&mDQaPChWTl^K=OTC6omROxpg|)^1!Qd!ih|vVE-p|b ze>^GieCy}xgrk<9-K)~;=Oy1aCPJe!Cq&})a?)e8<`a(HLaJ{263;y`BN0L?6_0!> zsuo_oGJ3(4bIj`5CuKLRoKOp0EoXZ}5N9*!+m#{_-39CYg+sG7F-vB-Y9tuA%a@6S zH?Iv4bT@(?v%)*JQZAWQde$ zm1MZMcILl;p=L7+o2Ekak5h2c1+Q+2J-188u2Mqbms0Q%qAB_ooX7ydD|3!BS2OnN zBL;~PUD5&!EJ_p(th>Q(mHgb^USbbin$sxBsbOVlK%Ys`J)$eNQfgo#$Y zDV<~ls45Q@(V1&Tqpk%gwPV?FCg7X)ffpJQX|>5OekqzME9H+B-%}zBcXxZ=@xt%p zo&Qs*as0Rb6EnyEOrmX8mx|r|r_?^x%gv)6}eUqzveG~6)<>2gCUnOfN+NO+KJlCod z+Nk%G8W5@CQ5{lBE2*cA$`_Ug9iGLQv8T2*DNxdDOKPi`?Bb;hPdk}oD|TE9g+gK~ zOn0AM*|Bk^{n`^nmZJaBV2X6jPXNOZSpi}QX7q1_kH8v98ZIqVz(ZfT3%S-a+<5V< zqa**(Hay-?j(WsvMOr4DF((6qa>w%uak31gg@)Gs&O--@(Xd+!?Zg%d-Rg9#HhrQ> zt#4Se2o4x!NGxq=T(Zs$n^irkA{?LVf){UYC6m6@Ve8~_M7jLn(9=eigy1@th3#aF za&M8t7JK1G@nCzqIr{kvjEBlLXkcZ1T5qY^fX-H-o9YU}kT9O|kk1T~QBAT)1E22p zqSMIgBG=P1vL9n!DG-wPXy2rG{o>I3Hf?@$v&hCoQ)1TT{4d6dvBox zI78`NmKU}v9B}GI&BC642DuLU`gh_cxsEjIB_FwET|`s$4Yu|!c5>v2vmasQX|v;Z z0U8~@HW7DoZuE~Q?awqn=3M`J6kVVFBHfopo-s!M%yh<`#ERX48ZxVN;#GFvD++xj ztyat~u+wNfpN>y6t?)*N{#E0SGL_8`Ia(t}xx=pYdMd)W!@4%p{CKc^ya~nIS$O-Y z;eW;Ys7RJjq<>wluQ#Vlc2YN)7psNrDyHqSq8=bG<5h7eQGf2NvEcOMqyP*Ay%BV+ z(=G3F2(`6tVKxn?3lPIPqW)n8X|O2WmIWX+a00{OaqGuZn1UQ*eHBD`1c<92a|)9} zC{ZBwLxp$9?^26A-PJaT+ZN+LN?y;G1@jbl>W80I%IWSOS@KAGhc!ox^X&Y z^~)DPXb6?;zXWtwVwWi5@Zuu|0ckQUK`GLN*$f%Uy#~hwV0^k%OS_p)K(3d5i?MMz zN)p}+yHVnBdjWsavTu3w>zKgw9=0lYiuuv02~?=54O##t#!eIIakT8PoGd@cq>@4* z^QhNeZ=Rln70#_qB#dS1dl=@7L72fvnhp~n7d47_e`4&czeZ~80R6ce%Dujuy>|fS z>s5)J(n*Eoob4NyOqn@VL52B6;~gC@=o-8H3E8;~i2T#9p(~$t!e6uj z-$=^8#VCRHl~5G^*Dvs-S6?B3K4Pe0sE2+FD4=|8Qg&TKK4wGOWp@ux_jOD%$s8)chWg}O zs7J~lLA}vfS(7YNZDgG-p&5EDAoS(wHdrhv@pV|(GTH>_b(5y%nvn5ACjiu#)TS;q z5k7y~K}Wb3ArFS~OjbAzdCy0be(%+0fwn~11L-<~E|*7r9dv6WpyO8^fUE#8XDK0J zBfQUvRsB7hGPE}NLlU45MN`j)3lD|ic7O;b`dK5SHE)1zc+{ml%F|^J=!K@B-6R>u zvzWNw7iFzVb8CUHRg)&&i(u$Wno`M8wptq9qLNvb2`CE7Tj(o!NjZs^H$Q<1H3Hx> z9eUcYKuEo#M|Qvq&Oab%R0U!3e_4kB=g^+&guqLH&-kk)B;u!Ts3MBFA_qE)`U30M zF?nAq;@z)}Dy9w?odsP^EByfn=ByC$C(vGcGUnB3JL@s9IdPXR<<%74RUHvvn97cr zh1E?Hdd>Cx@hp!1g8Gf_2NAH|1@#<4DiQBIP^*|X0ZwuQl4W?KaaKn7jJSu^md;Ev zf!z$aGPmMswT`uN3R+vr5y=Hjz$!ifXoBKJmOk0*OjW{E941jscx?jpNS^@H=s=mg zSunWQ4ZY-=GRQZ48p@huI-?4I_dMVbSMaLLgOf{^cjVbP3?Qwhm^9sX-VcF+Byxt7 z1>S0jI#%1N4p9bgBow&;^Fy$N@9bzAZ5Ecerc%e$yYjbhz<1~C)?W|6xVUkLPB|H8QE-0M%=}$9wLq|#ft>;S(B~Ay z<{p-f+^AeRO{pDpMTl~aaYf?BNs!b$LESVy(B1G1YJc?0;!W@rECaw;#9_<<$dr(qDYL11>Ozp3W3Ka0X5zO)D1;#mecxg!d? zpVtY9uS&g@3@fTX8zt|$2p&N<5FX~Ey?kktq9?P0m2FDtf?cj>l#_Gne&*AY2vUP< zu`zlnIy+BcrU@nyqPf2Vp4_D83Xu<^(flGa60tO@0CFO!7OI>$YiG%pop6b_AuY?= z7MJ{IhOJ13g_$?e79Pd+N^^uV*A^tk={^t(OOPDd)59m|56Tuln!_KRW@$!$e%toG zs)Q*h-1q&(+tik28Fl%sp};LZe+EP4&VKqYdSBT*CLw8JzuNPAeJl6#IhjX#1*4-a z`5`M}v_0*(%`?o{ZjN6OAb))Fhhxeyr}UY^T>E>>ERauN&jknJM5;aT4~G&?M;z!S z?r-?fQ?0C?ZOx;PsOU@j(%@3U2@8{b0vpZ~RQ(7+&qPmy>D3E}=RFkKd#rhG#oGGI zSeFdqB7&JEBwm5DM<&7^v*rt}*c5bOLZww&CF6O!mD7ZKs`wp;I|*_G0ysc4fq6dj zfi&GLPpc6BfW9Zszj_eOfKs%d?0>cb0bA?k2nYb1p~b+1-2)|M(X;Xwi&w#ackF!- zNDR}FQ&-i>;$?~4iZq_y9f}ami%y5BJi^+&x2n>(WPRqNR}=1Xh@GkNlj(}v!XqA4 z@H@#`jp&3#!DBvt!N_q)UOD#DK&n+?h6;cFU{izl0FO_Le zz7F@NbML{uAO2GcGSIX9XV;E_k@90(}Jwc|AFCq#>A6Kv0_ssQ$j~%xT zgbRJ52WR$5@ad`~>BX(knZF>ZZfr;EUrT3-o!FJ@UL_eb8x#o-0 z#}QALvK7j+sP}X+#<##H?%>;b;dC8L4$qmvn_xme#@1byRx_PES4>7$BO?Pl*EkV@ z$rX~wbZL25>`p6<|K8_{GMzWChj1R3~_&d6SJG?E#@{BV$+5dcCtAbn9BzGDK%pS zbP#ahYze3?fcmG)an8Wl{P*zY2u~IHr6)AfLgYk&-(j`!#om56(~Fa8SV$575fi)l z6|VV>+{U8#X1~ZfGR@ll*5+ZbDI1`8AYh)0DZN}i&ER=JKJ}qiQXEY3zo^uU(wL+` zuQbo*bufTsi1mD)IZt@$QSACUyx{)zNt-bKfeUl@Q9)`8J3Q8@+OLEj5 z{edTCP#@-iVYDIwO=ldkzG&4ulm#Brr6eJ7GXN?f?^p(q`1Yk3jNnrK7g@s-JJzG| zHwjQ!g=uocjt+j}$W-|`(nSB3Iu4Z+9?)vkjQsT0Y2fmDNfVf*+$m@Ru!h6pJdt%p z+|U^s^tN?3TYcH)y@yQ~=}YfVk&O9OPIyJNRRs}|nB;3T!$Lk>r1qQgnJJ9A!-Y>C(%caRodHIZtwwoOJI)e8DAlB|j1aJ( z=BSlX)9}$1g1?d?%cZ%7S}a_edHhp3P_QL>;Bu&|6r_#LX{JccWW6#=^?<|F?fg zq5nVPv$HV#&mU{FGpO>?_6IHgb{GW2Slj~4_BI9V!XSMgJQF>Wvy`M1L>aTVMS9hWZK-;$)3cPo?|Ebo|Tj)m?IPe4Vw@<$< zk_zGkHh3NQDsKk9Psv0EHwi5Q8X$1}2P`lv{;tzK4jkeFrlB49%K`^PK4lTq|AzF- ziMD<3t8>?At)ue+b0E+52A*ZOm%)wn%Ww0FfZQJ*9s(Hw6_kGq0G_T6`1TeEY^NrU zPqt_Gq!Z)IM`ynVptF+!|3{u{Y#XnK4)F*OFc1C)-f#E!`fV167Z0xw6-pG)8vjKe z{inBaH=V%M@A~!z5AYc%4e<7C6dd5^=iAD}VU%8reQN!4_l{3(sTwY5mh~2*+t#!Hx1?_|*dNWmn#4Om87Mlxy>+RSp)kV=)kWYUf>V6heejs2jh9 zQyzJ3pRVJ}gzbxl*^C^f09Z`WkB+zf6@g5r3T|wSVYYcDEFA1cR2;ObrQz)X-_hS} zO~Ei*`@w0J7Ns`hCZWCmt{irI8p;4R@G`;SMtu2=2J|U zSWi}#?8Yqmt3b_fJp8s7jY@vcs$&Yzjp4je-n1{)y419UwYZ?u&C$<3YFW+qGnwHe~Wu>>hAOfSLENFDv85V9?+SrgDo=@N_aH7&%_>SZss&CGqG|Pv^i+ZH4X@=> zihhO`gmY@)Inlf-X=1xB-*L2(-?aS}ETY`pLfcrd&&fCznW(Zh2|=YhIJ3b8F?QH| z%l1)5)0|@TSXJS1ow8HEH5{)M zQO^}$7%v7pUWm9tOs>ZWEAUViAMmRDEzG$99xty^$%Oioq<0+0` z8x&(58Ilg?DXkr7M8o0NzIuk*T}qhcR-gtoW(=(qv(xh?yUo`7cb~_DWBbI6)pO2WPO3UL zay7H9IbHhElB`5!X33x+DcA#Ui_Rih zR&ZA+5IF`Ym(aWM(uC*WXQ);_vm0BEgnc<35_UpNo1AB^;tP22|Gmn)>EhWAh0r6d zcNJwlT3@oS`BK=_orPTcP8={2C@WGqcMks4bvy_c+ADU<65q?bn*6uESFI#di#R&d z;WBy>{=;30uz~ff!v?WqSO=-<;I50>iNzCdrhER z=jFykg1b!B%h5y22IY?uN_r^A+jOulg*D0HSk|L{V#Nqmb6te7frng2|2|z8lp4zT zUTUK@2-P;$M@;Mo%1{@|Hil&e4E{`a@!0&+k;Sgvd@O7NM$DCe_zw=9A|^KG-Y>T+ zl8{YZ)x;!{@e{wO_#DM*2+GShR~H9$?o%KntRt!3qZszhv7Q2QRw9`pe!9(D(huRb zU`J$vK9sW+%{$#^544W)WnP``L7_iaEw~+M8SEFJfxxt24dq5#lsqV)CyGA8=U=bMTFZnnjqY4A1} z1}jhmZYjTA9g<0yVs13(h2K0dbvwQo?{x~A9Wqb}D&%>Y%oE=qkaJ$v7QJbdR59nm zIIHLYLpqvaMFW+XIB?#_-RN`hmW*^zJmwv9&nz>uDC`ueC4c z4jc)yl_%|hE}x~eo}C`eNIZv%DR%^j@jz_W#Z#LOSIx1!G4MF`(!>C%vpgsNr7;#~ z=iy?H{uCF9&LEHh6uM0IDaF8|)grNuX<%H#Q9S1SNQOjiUdcT?r?VCa6h<473M-v! z4kzPT8SU);YPNaeMciVxW}#01U^u8tr{UD~^u0c^2%I{9Sz$|r^J~>WO+xMm!gj*Y zX1lKL6ZRH5mZI{^lg4)(jxNXtieHU=P(t)I&@`RcguJ-ZR$Dt24}M?6TtYE<9EYul z(8{FH(S|hHX|;@79TN!9U{57K(YLKe_=EXB6+M3%5p)hbt{68JjsWszfA2# zMU;Fr>#4ZqrJxj0i#ux|stEPMEL}o6fsjzUGw-*hpU!5x2Evy;pz2R z&O3I$4a)}OPJX_{3emCrAsd=38n%Kd?CPJDY5Ub7ZOF2+;KIU9QO%9Los4-LUxqb* zDxTmX^#Vb3-@q-D7D$4G?Ho(->%^}EBw)GJwTyHuuDx9K#-Q{7aYO1bCW~)!a+iM^PgG;x z%(`890@0<;VA#GOH(8Q^hMPt5y@rwjbvaEjm?h%uRDgaDr#r9fF81h75jFc{T+eIA zL@jxr%b>RDjvSshd)9kZs@z&iFlDZXu$!fnl<%p$1t&ga;PrTjpaDKx$-|kij>IEq zwbMQsE7HSmxN=-DiA;PyU#ZhL@rt6sk_-2tA+z)Flx#xg>F=8Q%%(^p45I;-O?*>r zfjBX#g_oE0@O_`e$tHdgI7OuL9OfBdwV^~MUF4c08omrnA}6*7LTP%mMox2w(jGs$bu3rwaTW;S2HG2-lO>d2fTl>Co7 zc{Fl|nRq^o&3U1Je{0Dr$Rq$bxJ&3pkgjbnpHkXs^kjfDGZRhWFD^Myb1&YZwwzyIAR zx2;3WL*$XAMLFuit)58iMPV8eW1}@&fkwLjlhNIyaZ;zobEMQqhs_^k;Hfn#TKg+dXeW>sJ5*)$ zd3T*=Ax;f(#WPdloz}br*6=x{;d!Gn?N&EPH#}|U4=>B^68uAv$6+dT3g`*ohc)EV z8V<&^n&ELt(NjA_9=`79`y9ugfuGBedQ5sZ=Yn`4-Uif#9{n>QJUPC;?EP6WV56ds zMsSlVn|ETGNF}KRm28}xsbEUIxkHmduGkJC@kplCRuyK6u1{D~5npl2X`Gs!mdW&$lUq=<5sGvpWxS6nfKSr-i*U zh5~Y`@51X(ef7zaDA#-W1ebNb8jx|uY}uqG%rd{Sei;L9#UcOjG_bMstcB1odKAulQ47D+?ur0^t5dh%whsHearI)5Xr@)=RL*0(g zI5b`(4yO51^S7$iFU3&ezQLRbw9!%23FFcjTl**b2rpODN#gw?15yVbyrfUE?n>vC zBXCK`mQSR6Fxic@fONcffDf+r0(?2OTBy#@JMvGEi!#KGTnjutY|sS&&H-W6`#Ws) zDeJ|DRfb1W<#e+4t-PrXM=cVPo;9cv9OJ}ryP2IXNi>W1g)W>v=;Udo-koSMPp98*V1A-QPASOF3<%fzM5ej^Ad@pOAm>=1PV%7$~lRo~2(Tp$#GJ=eZ03L%u)Wp6vQPPGwYTn)N# z0z#1+Vf`+hUUuz(9ts1(kTkTeVl~K6~7-z60L>{O#oS55c!2iNoUCd zeNXF>lM6{cCMb3TH!%(mcc@62moDxVyq^R8p5qVk*>tVPp;D`TUo9WZolH3C((w{- zY9?(*zBLI#+9v|fRaLICMf=-OTdEzCa3))s@!FJ)u|{gjsEc!<Ikz!_TL+ z`ff4$V?2APNpD%yBqXs-Zoej7HC@aPMC1q;&mYyN9~OKCdo8i#sS$%0G-VZ91Y2^O z)a{JS;M+R+RAS1F-~~kH(a~FS>y+3P52dxo=$yS^ty4XxF00sL;B_6XTzpq94f-C# zgQ)?-6fe*#dEff^&lqD~i^N z-!0dx^f>EoHFvpm=Z!02)1mWd1AAjwV-Jtq`!R`r#MX zOAeDDiVRwaO1AHavV26}l1Y<^eijmZh_)f(p*ShMC|<^GPbcHbYAR^h+3EP$-TEB= zlPD(bZOf$NY6{X-60670f^N7=v|T8MWK(EmS_H2Qr9#U1kGLYJ2o?<#ZPSf*60{_N zXgSi7ebNiq8=Z}G#>T>RMrZ!dKi0q|uMf)eC!>ej6M<~pFhdKC^F)GbhHL!>Qw2p|% ztvujWbA-1BN$Im#>&=$31Yfh6oWe`3OQ0WKmkqu8`$teABggkJO!PPIjg73oD;MtM z*+K7pRvbdSCu8%vHgQV^^bAYW5JqftP$yQx%5R+kYJ5WE@YvS3)xoPS(pJE0!;*X0 z;EDOCe#L9)3OV&Bz8j?`Swx4Gp1Q(2RjYW{_+hKmG*fAq4uNfSh7!Rbzv#iy7V_xd znvoga{5ep&g#uaVlvgU0)xXt>^%<=%?y8F=N!0dao*S;u_M*4lLs=787#MLMWJJIT z$y#=k*X^&pylbcJAk@7rHLS<6>$(`xPmv>-&pN3};H4KA*{4Vy{nqQK5+^&JrN3|X zP-@r>tmF!6VWU~U<(T`0;2C^B+ybReIzEYP+0vZdp9Q;>r`3Vhk2cZ?DVKFc8s60Q z!X;;54;ny&OZumgh^Eb&sKK*C?r0C>L?tWTJ-W`JVZ!?B&&$o^=>=G7igHw!??B<{{;2!+ zP>7w-EW=W~iVO4l{N(j6aqcQ|1^5E-*7;@X$A7f@`o@W}rdGRI6CWTuuIc8+_vhs4 zou!;7M5SU3^2e|7xxu*I2PeCFZS_eE;OR$0{;e+%Xy0%1=U-!%Ekbg!Ww%G;tzvE- zB~H(h!YtAKnnRkA1-FwQcY6a6~zt`N9;s&1r%KbB)oB z_KM+8cz}bBwp3@3I_!QVw;Sw7$52Ucc- z#^-zEC~j~K(3_Cmi4-$-MFkI#qg6Jl%bB`ayE`+)eM6Z zDrkY#OiTZ!VLV0ay$aI$LX22S@6x#Zz(A2&8)aFZeH_KBVkzY6n_7O{-8}jFXR?A> zX{I`R*dYc;D?MbN^G{W@{^M)?5~err+PR~Y-ndWwU|G-|OOz%`+}hA>sdf{VkvzY% z9GRuUvR{)p=!IJa2DM}m+(qaCJ~eA>ShwBh3w1TbN&5MPH$=Na$mxJrL8sJy330b04h>$up z*NxeLAQhho+f6GtrH3Fb&llJjW;{gGLUy^j_=H_P9QmI$OEERXO6w>$kyhW~x0@EL zaef|lI3u!Ye$GJE{-y$deGIRAUuERvrtz07^S}430^Md@BNE-AGgD~&$Y!%)uP&IP zJP|pE*2+!$2bvW$qwh>--M}2iI8|tdm-=9M#4 zYL3ptnx0Oi!WeV+Vh3*XlA6IrCRhu~k3#Ks?mMpTo+Tc;`Vno7QR}OFm>yR%B-**U zE53O=RVWZYgo5py&UYsO(BL>X`BZ4847pk$Jjf@=)wizmTlXoecHBd_uD#c z+qP}nwr$(CZQI@Vn@sK?_a-wxckS%#zm?RgXZbXuzFX{ua`{uRFXNo?%3tJnbJgX& zXmHHz)|rR#^K^@7tZUT}Gh?BB?0M5O&L@&vC0tQVi+rw2PAqi~lMn24=U&NP7waPh znlG^gZy1_nb<65L@$izE*HmGfKuz&@yxgsayTVV>A22OQyqsbcgkdqHbiT}w_M0uu ztUSs&8e~IR^l|)MQ|K~EB`7Y~f26=)JDgf8GI_pmv(@rA=U#uTrZ}6D=ga}4+THss zj8s$sEv~p&6B!S*z({dDlUsY7x4LN-q4f26jiLA)1trZd%+zf=!_UQr zovnFB%`Vw3TEY&p^4>t-uc&UKitaOf$**kiIE@g0e91zk(dHta?#`O{u*!_xa8U~b zaEVvM6rQr2F1{DtM5=dfQ}PqutVjyV^u-O%HUm-jlbPawAgX+J+_UN4kfQxOcBu2S z&Y!-_DorA6*7cx~|FSywE?Yu4GVKDk4$stbdRNFgij!-azIp!HEMOKXK{66Ro7`8g zG&d7i>)Z5__x$ct|FGgw5lXr5hSSf`s)a=ME7_}hX8=ME0SS%5$4}a);qBt#?w2W| z>6kKRTA`ev>5&b^W(l+Dz{BskZN6T-1NU{?(%t&G^QLqFb?MzwY{k!HV=_9iGwh8b zc;bhOv_OZYzoLp;^-{XzJ-*g%ezaTpc{Ja|O==3>z-iArs)9(p#|O@;(`^FQ^QSkp z0keE1g@=Hkfz?7U`@ti?nIGSK3xVkQ_~vl0L2@1d{#ubIK!7nErTDo?nI8Mh(7a*M z?`3=%V1sf|Iue&Hv3*_|_od}~sHOyBhBKZ1s0Ms^3V}U#e;Z9%x^BrVG11x>% zwf0Nx12)amQl4}$rn4K=VK$NofwfIo)+k!6eZ#-HqF;{w{_)&?GJMLfI@C(a>(&Cb z+0xg?U#!kD$lI*4jX9OshECEa5~SkB*T z1rxW~C8-92aDA!xvN7dfc6F6Em;wi)970tN9T!xPqGGE(d@zi$SFj7!qhpcyYA@_N z;s7rzf|YiIeE#tD8mKt~%**<=j~q0Q+XVLm$kdfGq*AcBM+b0}HxXO`$L>?zAATxy z8`%FHzQOjt0B`@{8!XI>|C7@A58q&7{J)ru{};Z&$jPtjKt&I z-+Jl5ASd>P{RUv^tA)R$crYL+iZD{83s4f064Dggr0^(I&-u1rJFl<5xsTqav+Ld` z8I~tG?WY;Mr~7*I%txuA`PTbXW=W9zYCt656c(11)P90MP$V=!AkfJ6c36LTv-W(s z!#JY$ZEhr3q%H{1MA6Z>)Ozuz@YpE5`2;8Z4~`b zhG3yVo4|qKSzvAeL-}DY_s+r_>z-N!Ppiaz@W;S?2?+;Yy@SB+rfoPVFoM7vNQjK5J-<@@wAQ!#`$VJ^XV=%XK=v+}0-fvQ4|l*j5{RCCKsN}kj{$>z zy0V~LNTzSP-Oxz(Kw>WvulzhAWuxt8|1+ybk} z6TuY14RBZh$xtw9Fi;yc(5wFj><`z#*edwXCiD-ut}F81X@Cnb(rR=cXdKkrGh%!$ z_Yf?CTbug;=-1ErTdOD-1SkW^5@=oU=3fPo-*F%KL1f>xRTMkn4X~ladhY#7kpTzrLB*Ga9@?Cz))raS&HXU&5r)QzOpG*soUqgTN zIY8a-OSz{&4W8R|KL*`zWgV@(IX^+eeVu_>@MiL@nG%i z;VyKtf;`LBDiL&e1TYdidl>!vGg2+BLLZ`6w>?AKtZh0-00>3Xs56wEwg-@(KqvBd zMp_UUq1gxQhqZq&kf3H8u<0I=Zh?n&=N zut!I`_Uof2!)o`J@$V0+-=O!Ne8Kpv+1WNj=tNz~t?kiy_q!@&ut-=Zf?*7lj7pMu z_sX*{TCuMes?WEZP>-|K^?fXO1a1C6C6~C8MO@)?m+yul<;C~~nO4v>Q|>QA@YF1= zE3I5U>vRgnBjf-@8s4)`w_xn+e78c*rdJJQs`A=E4YmT{0QGj zX?z?~PdNGtP40hv%AJaj(^WTD=jVQRf3|59m9B4*^9yu1owF?ZqdpPl;u$-8xHd~H z6T5M)ZByjDIUg2;7^licviZzGyKqd#KtQ|o_UCiLBn$fj`Amm*IU@N&QkmJxpB#!^ z6%cSvCzLlK9(0RY=;+d_EjG{~S+ovEmRJkK@1a8afQaSk9yzd@_t)p!BngS&4!1+m zRV`X+hr&RJ`M-3{EDNJHe@=~Z$q2bx3~%-SZM$p}(jJ&`E+uiMjGZe+J5xF9Ku=3o zK_$v=+;EgQxI{AexiF<0q13o}ep7cYNK z@u7+u6g1WX0`VoEW<+AVZ@W@$<>PaU66DM!Y5KzNg)~&Hwg<=O+krD_)V1nc^+Wwd zY(Ix32B&jhZ_7D`JB=<@ZPZhSSKFtV4ZC2hJ%V`!vdMDm;yv2|=Fw)%{12Y%B)z(F zT#r`2jaaB9T*~>!dwP1F`t`&fe{Tx86TSR~%Z==3#;&eC7Y|T1uy~xF-hrjPacAu8 zR~H=0Q&In5;w9*E>qMe_1jS-64~fO=#Fa~o(m?W08SBT4H6f_lz8Y3Zksk$MrXat1 z)0+RK@Nft_%a_Y?z@NeKp~Ime^~@7=(yVM(ok<74y6Q*}C!v=1t%)BP&hM;%-sD%| zU2Uc*v1CYD&a9l9&~t(XIKw*dRUVkhQLWkl&Q}c00!be(;q7{bjJY#O3&P zTuZ?P&)_<|E$EQo>ZYAO(tBP?%yepj2WjW78!PN0b?E#Y4{LwNfw;rh1B--J_ey$f zvw$>C(6`jN=tqIe=JnNIbo9YBp#E5r!CCL$f@0y;*ueEm_3iXO1FgV_!xN284alEj z+R2fQoJ1}}v@M%;r$wtDv#%=ZU$bY+g^B^NK@aBh|CGY^zY8`qPior>(#Ypi?9XuS zUe)l6-k~z0b@}Vh!W+wIDQ@f#;e_n_U9c0VP{eHXX3i@j{mI5D3N1?Wk#aR9`u@j` zSrgZBN+VN|#Gw9<7@q*+?9VhLqWBL7Db&FE_T?uvIDVjS(;- zF(ckUyjxf6*9#VR*_iEH_jTak?xWF#f-QG5PqnUPVCf8x?^coX(^-bEFptYBHXBaaz1rkM^$X$9fBM2jwl&;_ zilh%)cnoO6wjFTfzP`1w{{s+Qku|d^9u;D>1H}y~ZFgHAqL6(XQ*upYEU0wYrOVNZ z__~URRx(7Af7=BZ0eys0D=TT+?#URcZ^O)>FUyiI$d2r3$_$c(bh3hZ=2lkbV*^55 zvxi0D_iShSHOP|Jvu+`rCW{A_{Z($*ZD5m&s8q<6f$H_ zjr8Z2*OyTELo0jlHgEwn)7jir@sb7f1J?aMrS;_BEyno5HwNKp-hOqqiH~+5 zO_pDl!9Ijw2Qd7_L8v2WkGQK?iR=1MIpM=_iSJ1pxS;B zyB>c{?t}ZfZPUKpi3`dP2_9H;99q_Q74p0)5ehy8WtG8+dB==eep_PLVq!R{Waou1 zu+c8lP|8*lUB#;Lu&K|g&<+dXD5FToRsn}}4Fz(pQ-$R5as~|5T`s*j-M)3ch%yu_F_(i&LJXT%+Eb&E&9jb(R*gALLgciuZbwH6ymizO(~UBYdvngf1CunCA66m( zG&&D}B#&d-h0F?>AmM&VA9@#7>w_iYzvwuesav#^S3(;&lBA8>*e`8b#DBEmGPlgt z{qqfsMB)?S+SH@u(CO%sGrCH}1%I+tgkhzes3Q+eBC5kXxpP9)@ocRxO%x~>9~da{+d&1&f3_|rv8=4Fzh}{aZ2mmvV^>= z3vL@A*&LVo$m!uYK)@!NaVvD;h8Wrrf+^8w;%()C)sBqMw7F}>%74yrh@;{&HOkbx z#f2dv>hRzn`nZma`ZK-ytly-rcjvEpO(ATP2kOV{t3N^Z&&ezEzg<#HQIE_JH`p|j zdRMa&1ca$i4Z&Sa=<>&0ca@il-0_n&ecbg1FxPLrUTsm{sgB}J)Jlyu6b|?E_7>Z3 zxaE@%HJzP&fso>ZW$hM_O-~8piL5BCdnC|6$rktC%`nW8lQi)R&RmfVS;}$}go|E2 zSUN1HveaiE&(*ZdR8GYNo+iTvx}%?f7-DB`OWZ0guy`pW-~60s;3L)RPv#D4#=G!7 z=@Po^uJ^?h>qyOY_8o`q%S3)0IkO&xw`#{SUD99ccifZUYm!nu4@66km15gd$!it@ zzJRc8%>8(Jw88&+J0V6|EVaQp^?!6Fmj~x#Ve(Hk9!yu0nM1GtkSdRN(zF>d@k@XvjEVWwFQY~IOXVH)jP&e)>^(@zOR(^QrbPBYEk z*MI(fs_VnX7jINMQ4EG?ftODCD-s&RuTBGClILNG=0>O)<*bl~_j^!aKB+KN7sfNAt61Od@0|a_Yk93(JjAyPvZ>nukv`^{RJ0 zTRk=VysIkn{tM}wd1vh-6P^U9Kk~?3eoW72y29Wcng}OomDaT zgz0uP9Hvt`uJ~-CE+P9quhAqL@w_=D8pXav@aPEKepAcUa;+klR_okQc1(3!)STa0 z5u`eZhhR_Qhf2J%uWWAwcsm+;s3UUQD_eV}6 z0Ia`Ff;k=&OJ>WZ%9V>=63M*$U5Rkk2aL)*G6v6D%o^*N?)SyW19Kur?&rB~(4~`} zG^?ceIj;8UP-LMI5eL3#sbGpj7j7~~gXs24E5{|J5@5GiLC|f~JKj%GL~)D zh9yRMv)as)X36M2ImZK1&?l-A=j?YFY9W%c+#XS_CsE-m-4G9oY%na@&^__kLI&bv9)u$INh|)jW0g9xZ(B3=+$P>a50J{(6t_s!2N(M6RU{UJ|)C8#O4~U6%vw8e!x9!TxAQ+Bw#MLq8_N;FGo@mYia}`4@{8D!YZnkVV4e zK3$|48$lsDKSng=^Qgj`yxX8eFRUvmXsQTK_dqO{sa85$ni4aNL?xMTK?i)5#!PAucO;y#o8+pv|`2IquFqXR)4QLUn3Q;b`Ea!&@O;q`I+$C+&k9%nBOlfqTp8nQWGw1&b1K&?OBBB@fmL=ZYMQrv z(T2ZlTBo}-$(8OS&Mc4;n>DA*dSOl-5GgM>{{AewbCjPS5ncec;oeda>{{z9a z`SghPuEI!=VU9YUZRMPX>A^kQUw<(zhk%Q+JR|SO>;1m!-wLJ?Y#>7bro@qxOkN?1Tnhosc5Csu$QJw$No=&o?j<)jdkQj$lDFu ze7c&@71`chj${fR6k$cyAaIHCT=Sl#hvX_9C!n2m$neak$eSd*qDsywfk4H6I}-FX zbeYkI(^8+C<0$E;5A91vdUi83%dxC-s9mLOR3D>Xx=eUrUr}L1jSR#0&n3KP?`jPz z4H_H73RnrOcsK;}{?i|C#gxX}0L-;VZ#g+p&@nRq)YGRu!5^k;1mRZ7HJ^2y@8Z*P zs*7rt6VQE-w#|j$DNHsu?dZpq3NO|AKY-1&S}fle!ALYEHUiTK?E+|N#wX*ZUXwf@ z4pMbWJR4=1{P9eEkxaVaO3m3NCxhjp02k_2z>aF20YooVtBv)%t^9&Q&| z1juh=xdws1h4&=0+#O1@QS(eN?`kx>WNUtyO9{c++w>#|=4SjweV=f_YbE*zz-THe zlZIGjP|pr|nt~H~%&NW(i`L$s7FBl|Yk*7#C#ue9r|zg*`6fGQR*fWKzqF=S__&V+ z(_Mm!!J;3op?~{;fL&#HcHXDWkUDzQKia)(XKueI{bLZ7m$e|Kwitu)f{G`#x`b$P znwnddu2Tc$;YF_LWzgI*hUUM1&~H+zNZzb@4AT9%T$kxp(MTWUgkAVV$h>9We6}uJ zJQOTHIv9hvtJw~*D?SmFs7y$v{3Q)g(~_DB-q1@))_lyN&Vbw6`WwM5l9~!#Yag0C z%bAoX^snbg&d9r9xMpF9bB}-3M@1kfu#a=%GX|!eoywlU#=XwXH}*_su>CLm$-PWe zwE;`?fL6Ef=Ud`hdm@SUvlb8ks@9Htq9y)I2BcGNqO`Gk54ige1iHG5@f{K3aAb{307%`li#- z>nKN>3{RFzhfivfy>vsrp(KGjo-fv==^5BYDdHznTG%6BnXRpdme=XAtBqnm^WecO z9UYVTaeV08eDRsri+{iijOGJpNU-yjMgIAm0PDAbRih@@V(P;;A6Nj`9!IQqoA49x z(qF#%0}R!d{yb_D-43Io8_F)l5S-QXO+7mNrTg-r|@1l&Fr^P?mB!w#kC0U|gWb)wK31 zEkH$d*$xj=@eciQAu2Fl+1K@Wr{Hemy2>3&20@?6ErSSlfZud4i+OVgD~sUpco%@CRl_j5XzAX@RO zWrwx&(!4ZwGjDJ3>!j2*Y<_tx%My`_^~|sLH^p&O06xnz46w+bFvxgFdJ}^cW?LE+ zrfS8G|Kow*k0`wTNu_aE9x3|bSL#3ns5JR+W=_KT!c9n1m|pTO%yA;4EFl)CKUqBC$Phnd4k;=>IgA`5dXNhjO547u zFixA$*=NjR!)uDZ1xr*kRt3m?9Jp^|V;5;QxaG-EN-Od9-xIy5EtCsrOE}b=tqNvH z?N_#lfy0X7R4;7f()&Pb+?$$*|2Y(2mIG-E)}bE8PGYP$+TQb)8}OIz`K!K}uX0`~ z$|eS?dmQmx0gYnrsc1^`LrMFb_JCK1ru}d7=4Yh$8KG=Hei8P)D~n%ko`|+K>W6ectH6UyyuGWY-0T)_J^`y-Wi$MPU?LGi_zYqzkUHhNaTB!!yxhRm zbc`R6z{*Cx<^cuxXP&7i8Y8h$At%0-jW$qGhhfC@cL*`!kj4}y09%P=n{;4=^X?`U zzUbPAk+Zt+cUvgYs>f;fOuTRMG$MZDYJ-Yh0Q#Gm)O5(4KsCQeMu|tf()EDNc(Vp% zw`tpXNvaHYq-h&gzOL0x=Be?*xwAxqRxeEl<_WV5{j2JlMq5e1J;1;(_PojPg5Yagxj`R;o8HCQA$9~w=+L&j!TfC?;H2D zK}v*=y?%0=QOi)GpvS<$P&a!&#GhD(PFU;2bV!?)aEzjgPt|T>+?2+jyn{`dtO7-% zpFbFB=Vy$%&mIWV1#%~dpM(kML(Lp$zzrw6J)1=Bgz_k=fzUDdoun^qPp4>uVA7+I zqn*4Iu7%oUvF8Dkgfg!+O3x$AC)))p>4!~K?h}F}e-mP~J3PPQxMB|6 zFRw%!0pvU})(&akL(nQw1Soja+%Aj9?oP_(R7%UYdPz|CQq*A%=H+yJkc=|H1`_1D z-FjrdA>w9xhIJ_}dl8e1C2mVYWK-&-1*jq7Em;@8ah*|-CJk$r|E;^Pt>UxJq`%+w zbCQ-{09V=qKoJc8(%i?6dggX~Xc{*SO@zv=&^gc^q#w$ALz~51pnaT|s@!L5WeQ{t zYzBISzwG6;;5y8n37#IGU9htZ1q!R#47}AKpCKp5$f?#3$i0LN)BdR1ULTY5qp(6L zX{nQDj73o^RGW9f2*h65Pz-EmtGFdrJ2u>kG15y_)hnDm;HMqKPoeX_F?bXz8MjA2 zjC(TLQxjsGo>pUH*=j-0h!TZG4I4!7RSrgGbKR6>4-Lp22XY$vPj;T!$`(3;Gx=z? zp{$dVoN!@|p-v1+WHDLhv^tw9OUAIh;`A#2#c)C{%DcXV$tBn25;LE>$P5deGK{A>4@o?Pk%{HbBpC)= zf(!ZTVqT@uU=^W4d;tV514zMWI+MBcC>##BwExRdUdxMwN$^s=}0 z^(3A1uxYC?34t+`yux;F%M`lSb1PCe-3>jUWKct_vjU?Zf0jHB4zl~1)em(#h%dnK zFCvlQ<-_U1WIHm%_ z`0y=ibRBcpIX&uWwr)4mfU3KvHYhp7Yhv1tEQ0iGlKGQJjZKAxE^IbZ0WnAOp5L!# zR^xxi8T_wt*o=fsEKF?wH)p`a!Nkt;Kac-U(13}Pjr0Gsf&71a&70aB-L^De`L@UCs0LPNL4Js<@Ht2#l#>0en&O=sg?{YU}jHGqNOxnup=-rB;V(aQ3NLjYjF#kBa|D>XJb zK(ujktOE(;0!&&=d|5|^chHsqdBy9yz&`ZQ0y7gsQ#e3<-$U5eR2FQ$cob^mUIDat#OjEyoU8cF*bjssh zuxX$IIsl;8aqL(8_-U8rH=phRs6u9TWpZ@vdY9V7==R|1J$C2IcAX62WkzL2;Ix-G z19+<1YTBpK^{3lJ<;8dWC>XtnbVzym>{uYsRF3`@=pu#}bPn$=_4(Fqz`%hAh>i>a zh?yP751OUu;H9Jh!NrAeW#+DTliURS0(Gu`Jn&vu$=dLC&*b*mw!A&LyZrsOCbNTs zsbW*p+kr6=p@X~@0q{Qs#CI${kXogkx^Sf^NJ3+c{U|FeWWGeRD3;*4k9G{(B zJ^H)xJt?wy#jafO8`_xM7?j+ZUY>wHH}r5@^Fx~of$blnv$iv^0@p<38@}XwORPE9 z7l43Qxoe8ymy;fSsLM4+ld-mz$@!lo@W+K2^m$L$OECI(8l$hQn6QGRR^YSN{(Vwn zYiFo`X<`SYZ(s(P!NGy)!Os+A8VHt~yD#z&%S4IEjW0a3Uwr=#Fq8u@M+cBM&Q5U-EoOY@2S0Ei*tOE3#aY@R<1wvXaPcW4|;FX=5X4LB;n7m3Vg32U-k5V&q+3Y1?0P3 zd|65)(4^?ng=rU)Bie3uAy#fj(itu0LuzKD%~_|JDYe|rt)z&J8E z)j7Wk9jEf7_`g8|sJ_Js(gck;fA1lckF6|^tlkkp<;VD;*96`DQvfEvMX(R{?~q{5 z1JD0`oV`gf22wHdO@M?pb4dTz+o^Q^+WfRyKKW$Y`LpZ&jj008#krPdu8|(lQ~5P0 zvIBKybX+qD#>JE_=rDK5|9k!`2lBvQ`pw@ZDuTP!eYW1l0rW((bp>RCFh>-)F?#XH z-$UlBo9&CYn&)`T|MuDq&cN)*&g`$NW5=)$-vY3C(>kEeVONF`otd`Nh@s6RMTKn7 zVp|S*)w$Y42v67Z5lHmq6731hLW}LOHSqJ!$8cvzxxN&09C#haQ8Laa5bJs9aT_}r zC|t*eyYN4Mkp3ygBUut>T5sPFwPHS}wWkdNdY(5_xM`UQrAP^9Z>l6g>0U@_#-(Tt z;5-qNA=h+zC0#%sQJV6m*NHXsA69#0W7?AG52MD3BJuYLVtZ83p6)XiX~qQpXoWA&y6c{`&RWb27x zL_ViQMYtaC-?Rtc!sTPTCX;Nr)?Lnc2iP!b#)|4Yc*VZ_dw9E{{=-Q43PKh)UTCp_*hckt}AVAgGex+E%_d$1Vb zG+Im{$^%rJ^WFq&4l6loN1ER6sNJpJ&YAhXINheh2cMc&cW|HjObP+j5!`%p5P8-~ z)av^!t z16h3Bn0i{?`k`xo^s~PY%(akMg{o2}J>T5EX41qLZZ4N4X>MeU4tVo--V_DD_6kNm zbjW>{aR~`sL?`mFY=Zx1WkSY*)ip?wJTPr<5OruokB&Hv0C5cZ+q7Pap<+(m>;?2O>7wDtCxc;N6_&%K&W0B*KS;SFPA0-%qj$Cqd>^#7-5HWG^3G zASA?KGa_njals+~PsF`)^huyLQ>nmO8R`(xj9^MtN*XT}o*PZ#JeEdv?<$-nMGIv; zqbn>{p5CH~k+J7|dW4;%f%d8krK$-Mvv16z-p^I7^ckZY=8h}Y> zW&VpnHxvgm^9Foh$_YGBz#ddr$^*tF%Y%DGTJB$0bc8?5UHpPue5d+ZO`f1?e8*#y5lA8yBFasPm>(}y(6dUl2;H8IFGfu;pX*98L~gXMeI;oy9fWa_$I z*NuAd@G>Rwm(@g{4l$0ZWABKcmooRjMVQ>2@-xE9@FsCD9OYInkOg82`_L|tuJ6v<)rDgGw9BRvT_NC{`9ZY%A|s8-L)#Tpsz-|M`wYqItg_-6JIJm0<>Y9 z9dC(`26sXV`9kwsyjnnDz?H&Q0tpd6ClAhGJ^?Rd^YGCsxr;MA$$JGQO zzBX!m9?&nx0`bZ*JJU{6W*t45qgr2~9+A$y ztp~Y0*ejcz=)YOqNeQoQ%GdDw()@vt_0d7$4UrQIiK_VNp4f`u4+LvUGiu+4tda~s z%oIinnaj8W^@dJrpY6$8WKDNh2+iCq(+(se&JWBFgv|x@ll{K{@kyPoU2x0x>CMb_j~uX zUE09Ij#h(+nyZBgABRG4ZR{8pX_!+UosqYF|7nr6vn}k0bxYF&o*CC!=x+G?wmHFUs_iStgm%;(JKS5gk+L|{;2BP_j>?b?FK75~2snL*mIyT?;cRxO{%hbJ6aWN}Ry&5g`)3+;NsC{T{6;>ufE+ zx&|Sq#to`sC+ZDz_R{Ew2+Ri6v$jZczz` zvMkogpMrQf+aFF>hogo$c-dQ8;H#<7+60T}p_WLZ75AZt#Y9T9L3s~w7xmn^OERwN z-;>}%;)cejqhA^U=q*>Z^JIl{L`AC}uA`o7ccJU6(uXnzYuCHL{}FjLR_0h6_s=K) zI`vu5`SkYfAm-7YSUtLa9}1i!M!kNNS?ISO)pA3&=Q9Y$Tih8bgP1H051*CRwXmyh zP+JR-$6OUx7;7N7I%kltR=OG5j+vzE>8@sXli*jFYiOr zT5kfo?N?&gxHb{fcZ_@H;tgw!a5C2%ukx(SbozH*#mBc7G@#19lI~;XA<@82jhzy1 zA}oufs7<+87BVk*WboHtAw9%P1eN<+2z399&Y6Ovb<7RyWR>aQEs~#_hgLQA=5mfM zts%|;&+5ZZRxF*?mZEMCJ&)s%*$SeybKxm~RyG!pw`w!bznU@vl z6DIl#O8p%*kMZA>%hCW(=t0P5f5^6>YdgH?2cc5z6-v_l*=T?ae!Th+!5xb{RLD^{ffi0B0yj|LlQTQ1pQyuAvOD%EJ?wVmTrxd+Y<1a z;Q9GTds5^C!^|Z_ae*llrUPbahnz$G&~y?MZ+>;SjN zacNNX<&?g+A|glM<0aRKi6V%)cYe`fmPQ!rGrs}xF(pgi<1qI|)dd6l65fE$kSF7Z48*qI7FQ7V`b3gS5!uHlE80^ zz&-_$f-O9saV3N>93MEffyj#vg5Qs$U%jjJg9_kejX0564XPL0&iuSDIXXPh27hnU z*=_4t-Qf94NaYP(S)W_>?g!I&w}`Y3@)S0AY~49F4u1~MAt20iv$Q$8BIUkkK!&uD zQS-PwB>=7OwITbYQ%p1YY8o`m54Ouk{Sraq2=(hC#Ke22_$^t4vfzkAdn_gYjMu+9 zeObM2Pev3tSjUScfMc+Bkf+O<)e1iv(8Tr{y^D|Bx@HPBaLEki&%Gdb)E)lT>~?@nZ8oF`S3-o|^P zK_kZ7~!9Bue_8C$Lbb{wm`Jj^JaYuk`AR%Iy)(y^ShOnB89vvmvbJiA=*_Y*yXaMQBpIZz-&UI0KaHk>MH-SSD&Nn4=Kp2m7v3 zwfDKX`rRkQO&2PUWHNTNqLeOG%w>pR*%g{1vf@es7@KtdJQ0 z*8LWO=&WQubj4ZGbl)1?wq(n1DIiT%myJnxdjy%0MM@_KcrlU0OBd%WwY9fV?VidB6srIN>^##A$SdS`~`AH+We^>aD z)c%#v(!ebzDsC0x)zoj>k0wfHjJXp0Fjqa%$nn^XLZJnqHFR8$r2|eL$c26!M`r^q%B1?~WO#}w(HR#-WT(_N zz<;{ofYDW)xHP}NntzOCL-MahpRKfu;$1tw8YxF8airF8PMRwpyP&DBPu!-S_J{uM z@BV`{bFff$hhEH+Qdyjb#b`EE3eyaik_C4AsMl2w@O>NXqwCa*-q0y{MKfK91Ya8k zsZW9FRa#!LHqXh<;HX5HaowP=Hq7r}QJJIE?c3piHtO`&*a~Wm~+$|gV{=)L|munxm>`y6QereK7`j%``}1N|9yzPoML${16>v zCiGU49mGm)WdzlyJS24gZf34|hlhe8EJv*&`;SWGmk~Huid!gg=-`VY|A5|(WxHJR zLU*7Xc5DoXh##o{Ge~ZTkxc%1@Jy$PV{#dJx8b3wVc-QbP_8WeQCX=0n>-A_oc?NE zfwZ5Y98um`sg-s{7+0oyt5ZeqA>$#%7a=zzDcMxXB8KYA+Aip%A%jfJ-lVx}`X8?+ z7<*jgGC0Jgk&mqojjuEkS3l6b?vO8kWEGl{uzI)@HUeI?98^YHTMHc@KNnoXu>#9?*y_Ar!x_e6K1Gh^X1ik@xQ|Dq~gVyCnl#2~Q zkE4z*QK*O(1vx)Q%!{_YA^Ao(+@Zr&ueEtcNtc=&E|!*^owq+(+58WuTiPwv0fZlZ zn;5U=>}qjA{>2}U&DDrvfc_>0J%-JmiTwa@dA6g(`*89zCJmO_q*J{&D^r7dy{VWZ zg^kK(a8m}7Vp8A4=b$%?2%*tjB>0)+%vxb&^n2l2Ydf<`*CR}V#J@>rEn9^!&3?>8 zg$sjopZlD~Bfi;hqY->GO4r%~9O%Ew&qmlf zQ9P&~?zrw*CRMgL`1R#ap?7V2d(~l~ERU}JTt2%yZ(&Xiq(8Za6{vgo&JE?*FvH&l zxi}D4$=}0~ooNP9RCjM%d}1+0_!wd^1xN1`QyUG z8Gp3WV;+|K&C0*#{dF^Z2MF&1v-SWD);Ue2C3RtkviY!feOs+P1EbZG%B)7^KA_7< z-PO7AVNAHq5?XuT;`VvjKdHrpVMT8FjOt~_o3hi!ySn@mb)Q_X^7P-3ZzXM}N|mXf zK*%`h21MMtL{}@9tR`pfm%M5P#wn+sq|S$9Yx1w_aIKAaP&#FNgcD6R(?WdsC_S6| zY6%x2f6RwCF1M84GHv~WzoN>a%5M?4_)bV&5HiqIX)nPw+<60X$(*&!IlO|biSyp3 zgOM@{j;B0Z*OrU<%*gs@#z8os{`t{es*U*DePFYAZuvZbWp)8@ZiB zOYt7#(|cZU%H`6~Ha{h6vF3@WlPh8V2LM+xJDw^VJ=o+3PLfFK)Atghc(#Rs z>5>R4IX36~f{SI{xy>t5z}QYNd-1w^ixz6i!+7T=KD$tY@*5Zx4KuKS9o`}IpI-85 zZK_16sj;eHP23a@Q`|FXgI9glD01(9#=Gx%BuG9UKJP&7_}TUAp-4iV5E77cbJGR0 z$Wrozicjmyu)`olM_BCm6%5&LU;*T4Jo%xNr8F_MYAaQU%mW&oW-urlRX0`wjOIA!0~BEglCNf^(?4RF%8+Nc)+(SO`CJZ z8cq&~5N<+LIg4`^$3x|4qZ)+V~Gw(ZfvLt&=H=iOC2@#G@`@l~2ktlR25RnGE#yU&@st>)MA_ zwcgZ*$Pd0bQ=YCed^WjQKKM*rwtk11HQM{p2@W%zk?d@r6lb&nSEP~FmQ`yezWkzL zdw>4(j6%F=5tdG(NnY){Y|cZC?H_6D){U}qPcDrC7t~QLw$(0CBsR!#&@zxPa09TE;zSGmjAQE7K@wR9#ym-LY{oxP>WjxJa4Mv>l@^GE^UM zl{2?1MWp{S;!kC3lG$`+Vpu$m@o4G$O#`~Ul+;ePVQf34+f`x?T)j@f?Pkn^BmZ%Kdj zm*q{}gOA(=9_{uDg()8kXrC-rp_%bdgL2FRk8^uG*&Rf|;ywvP4g^MZTIac(@=Kt#^NiH^_L(yVCQ z!V^kfs$+}}V?2_Ly1W*0!H689+H_vk?Oa8|he|GLL1s&NAf&miS2)oio-n)6e!hw1 zTl5}}IiddazVKihd_GV%QOl0heDz(O!|4<2Tr$KLA`w^`oilZ=Im6jaB9CnGBe|`- zRrkY=Lz`2?bQevf3A7wr9EqjIf}aQ0R~T9h5=j)2_qONY-yvTyGR?REPXfoOffi57 zlOpHey{Zb~C?ctDqr*mOMi?Tb>14e4K;Njtdu!x--#=0^T=8!A6vr6(n)AxjHds^n z(3a5Hnp0_MDRj1|9&V%<#2R-wl}5jrc63q7tbfH4AzaR@WQ{wDX>>g~KDS=Ypg?Y= zqxCD}h`G$8SA+viu;0gZcs*0@N284=ddR{9Ug8T&L^#ij8y4CM%cvLUUuko!;k9B$ z^Ve-ujzp^zt3SdgrsH?w6^NzdmCPKUO_r3kSIlVNHgziFZjsU~-}eOS4n`V&cwhO; z&>#cFpz|jaB1%B$t=nu6@~Wr9`;+8TaD`TxU~X12df5?Qn4!!4S3@SuISeF!)xavJdYq9`;#pr&1v91U-D!^aP{D$! zCTVYRXMw?+Y&G^s{PJ_n zZNziHyR(!n2b<DWu`zkT;Yn0SRn{dGj!BBOex90l=p_g*8xb1O8 z)|iE_t9}ATK(~xd0>7C6gV?XV68;8A_cyFPs>y@H%uut?2@j;*;Z!Ood1=ORjL zD~#-_yw=qWBRVAD$DE=Vsf8zUjkz_WHnAf-@02!@TiMrn5R;)b^By;d3RIFX5XaI7br?d^<1BsP{Vb1D{qyx{Se+;q_>lz)ViT{ z?L$|OgH^smXG}|oQSBFG9c2%eHrfvpne0V2zW&d+N6!7qBSVK zuSGb-RgF*@JkoKOP4c2=4}M-9oShQ6(A`s-ygKwHE{Yd^q;8#C(IMaI%hkjgOfMRF z3M>1zd|=QB1uC~+m?*U#0%77zQ& zj@J}^IW<>;m6Yalai563^?trZqRSqvg%CfU@FygM`QDfIfnyfD$*n)dqIkZMg zBMk4|MV!}wL=Xa57{8%IMT}6G4SC9rN#jZ|j7R3UU8Ond(lh2s87O)BUU2+}A1;be zSePBL(TRK8(!293W3S3B?7V5BK_gZeg>i`0x79b@Puq}$(slZ5I0C&`&cLSBrDm#W z;rar2!H;Q7e5u9Au9IA)x?%#5y4k@0w*ksS2oX9A4EY zhBT;qVcqH=+_7Z0+8N&{^@%uO*0ZMLzhCvp7D$AJ%TPrr12{@PbNNIkS!Tr%OXrCs zE2<~b9(ebWIp4RwXn$7Gc~I2>xUUN+;TAnBv92C#F-8bs1T;5US8q3Nbx9rYU6x2> zFA`Pd?~TPEkDq|~G|2^u$8-8>ldsJ*D3D68 zTW(a)#`~Tp)e;*xFMkJIwoD+LxQ(LsRSvk2*e8VU8?R+wHk{4~9Uzl~ak( z8I4-0uAF(kGb#|eqHDtI4X`)1IKFEVlZU-N<*@zvsn2<$(yvhVGqY)FQkNpNbRY)Q zEu)CP)X&no>H4r4BieW6KYZ-piaPalrVD#TXDtqo@xgpG-y#z>Ik6b(1IrW6bWx9= zNd}c%Xl!CQpqLM(Auu`zwX@ee`EDC3DGoA!n!oY&eJpNN(Jy`Tax4&5c+c24?|F_B ziZ591ZM0%<$q`RRST#CIc-(NrjYDk>@$B&pmUD1(VD!5rEO}|TM-R6&#pVy=3rHnR z+ug`5$4p$zd23~nngcyr#)JsV&(}dk$5xg%Q!)&E=Q3EW1jvtc>^)kMvGT;SY@ITd z9{kOFG+)U_g!3KW@s#;*=Y3OqEc2sDj)x@?&bfYRX^ptY7|pElu9M04{5H2QMTX)q zDqL9P7nmhzTZb^36ATp}JINaF92sRVvkiVropF0(7CNB1mxXhgscK0{0+%2jang5S zz{r86yY|mHIrEK|k?%@!I?KK85u=QCCfrtKf=kRX4fcj;)Ttu5>!w zcnrFE4$+6D-GIq5q17^zeeh|Fp$`&-%UVwySxEUtSwYg8qxeUt*6HwQJQEc3kzRKd$l7QaS~VArshJufim;)%=ywK3$|S zEdephaxH8iYN26w1u?k|aFEtz3bd&oJ~C}049u@VgWKP;0v83y4;!hD6jFmPRb;SU z?1}6sdg;d$qdl>6}$q+IVcna zv@I!C_j%UjA&Iib%_d+JeV`v!jE%AQ^_`kQjB}{4%j&X8F1oqE;%g*Q474E1@_UPR zV(RRTg%0o48c{fuMifk2F~*cQmALb3cjTGrmx2zzeI;@FEFRXi0>S!x-G`z4?oxRO zG+xG~6AYC*)0)m*M)<@A@)n8Lp%s!7aLlFtS+khG;O~5O{8VvMMADI_?|}DM(k(0@&vl z<|$lK3R6caGO}ZyKEI)Vp8O^X{;Sph*eBX_3M2ODh_T_3IDxmMw^aN{1&pdqp^7W% znr@qRDE8IeoJ12|N$7)C`TyfVZm225xWYofjw*5{4^Hofu_?Bba`F?S=V9?wx@SOKsdVAYb699a+^8H$4GACIdDbAPxCK-bk! zzcr|?bGnk5pl!ohz_m9l4J#vcQhG5rb<%zY=C81+ejqj@gs4B)Q*0csjOFd`Qk*Jn z7gjzMo;+gbr%slcN)S4ahf&LpxvNs^rT_7c? zQ)lbQx3rZq(PPs>49uF=Ysa9`({M_DLlr#Y_!%w~QF>;$Fi0tBFanJ5m&v@$l?rOd zS-KKpykpi0#XbU5HPp$8C(w0?d8Ug?kXz7g8=3njLP>Tfik1gbLyCqrI#a!EVoXL9 zP41@Z-J$9rgkP)4DzIQN1=uK)8j24?Mwds#(^}GYoFp0m*%pR zX&ruKyoEQhTSo|PY`}YwD@cHxi3GJNCpNOLZ>Dug~K{n5JENJ!m;OF!g+jtR1N3_xCd*WA<|1Sa%i!5$92Va5`7=$Ng#C}1wd-caq9i>KB; z_g89mUvHZU&zxcziQtgqf3W0Z_@a%XVgO^|o}i7Q=J9h$)N6b6=~V-hr;7(=+W#xK zn4C|^fM~CBx64~GP7R;Y+@(T9N{Y&Y1nG?-f$8)Ec^ii8SD-~69iW_R`sZj?K5(49 z8`XE^q287Q=Qa1hljU*-^V3|&Ek@c|iW4b=^?|@D%dytE6hgM&tpPghLpK(cfYYg% z*Z28$tt!G898i1j7Ma7Em zo6f$SrLZU=J(y|r95FYT0SiqVpN$XrD%8L9Fh42lxKOUU{!qHIKW_S1R9*@6xbtG} za8V40P#-L|KywevPZb9uIblx{@C6lKv{v;mHd)4)~$1$(WW_L$!zlM@Qe#M z{|QCZHG1;#?Hb*=J(A>-%%TBR^I5gG4eH*whZeu?HHRLz$iivhCh3BnLC0=CvIB1* zTk|zP+>j@ebhg~nEyN7tzB^ zJyj|y-3zC^SCMTcj@4Jqs+KNmIt5X#;STL1jA{~I4Y}^#(U(ct8XYdC@yU;K*4m8A zwo^8Iu!-pip|CX#V-Rz;eS8K5PP4D*;ZJo#`kPL-cU6Q;HOnu*OfEaZdf>Gygbs?< zZr!fviFbNB<{4I{rb;|SS9Kt>%XhsS}o;_#gDN@tdkUx{h3`ig{rC!6+z4t*- z5WorU-SqKuRvUsDXsbXDhh>-!alVUsep~o--YnPU(u6gTuP%K`x*GD8M@>chlP)z6 zzi+Ul>Ah^&a&n8FmtV_UjLA||UO;1{Dq(H0kss>`vw-88#`Sr)Gjs_fZD)u^3Xgw= zH96y?!H1q*y8N=HJEsif>ZijljBf?s_3AW3ldwlre?`X_rP;|IIw)Gyv_b1=lrfe~ ze(Jao2U9o--B|zWB-hg06@1OPct7n|QgcW9Q8&2S(CMslMX1qv(AXodK>K2GD0B8_ zEtNE)nlNV(hIoKJrUlFW*|a(ZlxcXDmCEO z7|dB>@o8L@=Y@p9WdHL0Gh6B_TFpSYv!%k0fom zPnA2=$X~wkPz8Iwen}l;PRF58Ng{2RWZ?0FdKm(TIy3k0x!GZGvF`XNkLJR+dqtXL zT>g9@6a>JW#!aM=b(W9O(}9C7a*Wz%PZjxUuob+Q5xS|){y-mEE19rVs{L-LyA#E3=}-8wBag`5*@G23wnLeaF{0D6G?BPL$#VL#MUWn zA7i^zBz++0qTyi@O={Q!4MyeMY@c}{P%$vy{7w9U3dy593rY8~%dL~%*pB$!yt{nl zc3g@Z-Qr-ecSBFOpH(1lIUa(=P?$>eik4oGDnA^pLs$gH9$h?K>7&Mis*v~|NwY}t zt=M5GUpDZ)t%}bi9mRQJJp?;>_hLAZ*ttWP@006CE$a_r=3E%d3gCs7(W!>oMpDWU zoqY7ON&A5(Fto4uu)b>F)=U_+VdV^6RqWx=1HMsg%YBSd@aXdtQ{VkM5R@KLf=GH# z^H%Sd0vUu6)(cLj^fn<>Ic;)ZSC5-+gj`3Mpas~hqmx3NdG)$-ufbYPGYZoQTZ-xT zI?owI9Ud#e=Rx6LDNPr57L|H)2^bkh!sIh>_%0wL=-=~9YfYH1ODJL7WdJWWp=z-G ziRHrq>aHmkVrct1q~Ab1f%B(`5Sb5i$SqeyC+8YbF&0fZ?rzQ^$!6pqQ3`A{Bgzym z)#Jq&P}Gwq#$vb|L_L@VF&UdqNrQWH+&yf3mnG4eBuNW2`?@Z1(XcOvnkcn)-xEE# zBN4ZvY%5FAomKS^Fk5uSevkkYODRY@-biw~>D0kj#I^bvYckL1LY&%_ott;UXe?$a0T3qfYCgM98 zvgFUD z3dRsksy^~w%05aK8{CclWtNpQDrnYJ;m$eO`0*;s`wMqwtoFYoQL)Ggral+Xl;G1) zql6b)?rN+s-BaUm_wup6O25t-=76(^&7xMcgUL9a4G_di@eNJd0Gz5heL3Ehbm6#ZZ?{m zcPa9@rVK00q?{e5G>V6Iqg7TA@LhQchd^q}lqZ$L1>~m7cWcS;4i1;6Y7m0T+jv$= z^K*Q#@${8KZPWElfyWdH~r%`*P^k7>;vJEhQF=X8(!Xnb~ zsLf~uhh0m4GuvfK7!D(h*k_zUzh^m|Z>jTw-Srd+*9*DBjqW&Q8w3k1{*f~7=))wQ z{$r$Bx}0bq;ftfI(d`WEdyXMntd-${#DF;yaw{Hx+zNJ8xjC{E-3s!rIjuJfWNz-z zBK!JTJ20*WDXJ5zJa#(*9;rOV&XSg++2h;G zP9DINB%wVKS&ZcTX0wOg1o$B5>=F@G0jvp@GDDwqiEXseXlUS69c%aGIT(#`%gu$} z<(ueC3px?Gd}Hn)Ycyj15m%3KoKw+YQ+NS6xNt(nJMeDVbIztF>P99kN(tEs;(Zdc zZB2LM$%xtGucCyr#UKVgW03K&AHXek zsXsXG+Ijrq<*3UqF=*Ck!1VZ#AxH_mv0DZ^v{<%NFl4xVJhwQn8#(Yw5hf&rurU}} z3(pvz75Ayqqlo;lnGHuU!JYewV-8w2zX@`_>2!Xed9hR$$)-Vm$XZc@FkxBpLtrW* z-+}ot+G4e*Be!5etPZ@9<(97%2 zq%f9lL1l3PI!snbDc3d7l8x7j_a_phK2y7CMeWsu(U-z6v0vaKHYgA(6L}FlWWA3Y&b6RsWWaK6$|Y_1p2SUv zPk-WmI)0$q_ZrSUs<5XiCG(?op~Slu8(<)>wiB`*iAI(cVZs-q;S%-IVn>6nWFbN< zdOeWZRcTBfu_B?k?R6F*cZkOIahp?64LVNowJkVd7r)+@9geF2&~^*c|2OnGJ8mn4cFmdiR2S41yUf$1;2z+O7m?qhl0XIGxH>slk0n z!Fu1scx1v0NoX^)U2GOKe*E(FKb>-3GcfLe@k|n7w=Q+8RD-lXM%Hx{2XD)~O2{Z! z6zO+Y-d@KlCc*fEbM_uJ2e}w<)t7rpy~Rwh0_dGywGzbX$KKh8ccVXm`%lUc!#Y_ikZiL zp62icLnbU626T+gvn_j?nukD1Bg7Ak)}gM3KV#XGS$tT@0USfbju{u^YbsRdBJG$e z?ndP$=$i3~UPX5_y?p<%uU=8SS(mb5xaqUzAV4>?oCev6>A)S4eMpU#CNpscep2|6 zACWEB(ZEDKRhNew(J{D1;rz&b?8b~%-toiOUV>SUoJ@A^x@;OIZRtKE92X0^|B|Y~?>z*^EjJJu!96~agTG~F&=+8a z_HUw(-oVOr&cA3c6E!aG#&oLH9H?gvsGPLeR6ganXCGdGsBzy|%H&NVsp5L;08bg5 zge1EQXf2>3IjIdOYt?xsfZUhk;2CuIEN#p7&}W(wZ}(AfnR#!EUm8`zuknKJAfN|9 z@XONa^jmC81GbK@xwE-qS2J$3m)gJ8Gg6?i4e6*I1Rl zeI9LdZ5I9u3HigbB>sM`Xq(^J2`n#bJ;fHb_rS`abu(*M z)jay;p!(Z`9KKmow{dD4u|?=~Hk~)Iuxtli2`4j?P!A1eqlbTg>dro@Fo_keXYp>L z-p@uzN7i33Di4!cU*b~sPs(tTd1)Onh^s$5y3uZQR-KXv>fT4uD7Uw@OIc%(%SWl& z!H(>oDUG{{0tBb}cBW$HkT|~~OtSPM;;H!X2DJLO24(5a&ODAj910l)mf!Owp`ICx z9R_;YTX3k`PX~L!Civ?gSlzlZeA4|U2d4jwLfP%eh#kQaU9zo{Wbh%UvmXK0XYaAh z9t;|5My6i_ySb7683jaKu{0eraLA9CZm!NRl|SP2ZussJ z@xh1JSi1+S!0TvE7{j3`>U_&Gjgt}ZIpCPE8_s=z0M95rE*^~G#eOUD_&8;nS#~KvRK;KKEo(I5Hb~0fdf5$i%h8>r=^VE2_jW9 zvm7#aa!i&8!ZN*ynLjw8g-?-qiEI`^7i*J3u&l@(Pa#;(&nq)lP2Ei3;W=#hnf<&~ zS4^R}K7)wo^tlkdwkq6|cnR&cEU}m4D%N(QY+CSu?VM-*ljB7VxCJeq zumNyShm#3K3PgUT`rV}NO%i6z+$NgMPZ%P5vrNMsmh9HoLO2gwM{s6&!gfLO>LUb` zC~(`d;&KmOe7MgjyJ_K> zOVijhC4;xwaX<&*Vi6=wnz3c1uTLkT3@Ft8x+!#SOly)zBC-Mb2QudUX|wGQ-pi(Z zibX((HM@S3#4HOY*0k-pKErV;!e^Z^r!U(&v|vuOm+4k~!6ZeLT8>Er3zD~y|JXuB zYTW(NWFxC!nP(Q3F}BH+OI_n2*YByi$vo6Bp>=x)I1^4_bomIk1J00ze;QYB*Q4II z-eNeqv))@!yK{2g8 zxvjR~7h7m^js7&%cqa;_teZCd|Tfq1m`vM)B_qh^m zTdJIkw7TC|>WJf9XIm0;^!o>}e>T_rJC@bfTkEMweA16-i7|5|Cf+#DN)252E30|; zpe4A7-iS2e1Om#$61JL9#q!VYWx=TV@S+=;C1;bI6d!=o`X0)xmeQw>Tta8&@YV1y z_FU++(SFA0RX8LsY-NU()T`@IP+w4(`Jk)2rGr#asS4jYf?xe zZ-n1AyKj+SdaRCZqm{YH#&VFJAEP$!X;$_C`T&&haUdFMF!C_)$+**b9^yui5~t?X4s2tkQQ z$n2f4(Pe1Ni>|kj57tIU;vWMN=bOAf#52^@Mj*<5N~Szo5StEqDKKGd*Ox5hd;^G> z5BM(&fEvfGrM%VC6R6!5XJ_2f{_1^xuO)HYlJLfP)d}kle{`io6#=Sjn~$BMI9j~z zv*KlUm1=8cw9odCfK@sxqevG&Aes2P^(_ZeUJ#b4@p9uo$C%j+Vyaaz6#iBX^MQjo zg<5|(edfT@XY3%qTC>-dr~NI($`fix)Om>=3{hFNtKdt0%`-3~Ay zS-31dBcJPiO_uJjH2hYc*>m8m#<-?Lxq5)8Rk336@>$C3x^UF^PllcJ7fSiBLzZGuAlm(LhKkZO~M-tXaRqJ z5+DtU*30IKRPt%26RrE+d-Eb)!Z473H9A~D{6Dhy^6p4)&cu&?X})@eb^&< zN%_GjC$lvwB8cT7Y)EFyh)6-M2d!azKPXPbGH7z*MS;Pc!s*Cv1`Cli4=b)!r@j`a1Z<)&JUiEz$?n8*gGT~FivufH-@v#zm0&f!eg!mTWtdpUI*4G8 zuiU<=pbvJ7fi%G1@d#1m4GQPWZ|f?0VRAp{P#Brvh^{bFW%je~})zxEju+n^> zSpXQ@@i&=2@I}9Kk)ZX-b*RiaU2gq!N2jvh7g?sXhf;_nEB(}cK|WOzQ8Vf%Vpml7 zQ<#oz!k&aB-Ht~K~&Z5>aCpKAGfSxkYXc2)rKYato!iC@ysZ3`}BWuo8c1IFD$L+L}> z+{+w!q8a+pvqdUC$|7wePR9Sr7i1Q1?Je5gA-4tniCYDE9oYR?{Ape^&u!2C+%ce`w(2eT-ORD zFI|dYjm}!vqpvV=VNN@Vcbp>#Tl;Xp$?{zkWe?3!A**dpOhXx6dwxDrjMFUl*J6sw zoZ*XrBL#Sd-m^lV7DPn`~CN!pZi2YXMD#VRpj+7KOO(HGo<_ zhTNwIJUOqX+4G>{yt#PamGO1ilJ7<}LzBrTRk2(gvq4z3eR=e9s9!pRgz3l92t9xf zsfPx}D_)e5C7MkF3bT)s<)dB8Rkv}4crRmrX#TL=q!Bqsk-`*Og5D%&pg_aun}|@J zCqq*+jg{Q{k^Ob!bhQS1Fh1nhWzT=_lk7aZCocx2E2+Y z;DHUyUV>|Tffx3Xx_2-+PV>V>s4f8RCqGa0DLGr>&7tRnGVZ?6F^@oT>-by}od^s6 zxB6xFT^xxnDv>WQ)u&ZDl5Mr$*vHrHrfV4v+=Y4J33SRA#c*|-H;EnPCGypX1N}dq zSZquE1oDvFT4SKj+Z(OL&^NW9BkC)~3)WMx#_GMRshB{gx_ zioHm3bxKXut_h}e;8~67c z_!tKd$y+HoEPdD=)fW2(?@M>_@rlAu6}kN7R2g02^%zSN^Q9?CbaE6tB}yuk#pt*W730v2j-7x7o{^bHa<%@zh~%!-Q(Pg@`?(@v z?GOT9%BR4k`$0WhnIun=0a^3O!L(3}TgxO#i+hp$Hp=_*IqAR}XZxau34!XMX4i6LPqURoV*aB#z-r04K!QT; zU>l<;H`l@*JE3M2099i*j*)JK4r#e2EAj~cKmPIz$F6Xl572*w2x1DAg)uL zUV3KG))YBL=kh`%4ZAnJ3IEGvCInnK^}WAOXl(WnQ(U{%$x2p~c{8smVCLNGVy)js z$O0+{Bde%h3$dM*F|qiHEmme!&TB2c5p>4SNL}l~8$&64f*N6x2wFxBMWCqM2|-8v z`kVo=f0Ey!6?PGgufZBi>Y?~9X*VWu-rgHr1CWGIkHb*%HOiTl6EKG4u+-9{p{Ph` zyJ9QJz&J?i>yl8aBE0#TRPSnY(~wyb;$YB|1>E*|X+oEsndUCn4Dn!SK85mJ69%DH zRTCXVe1bqWg7H1uW(4Bk9Jx`_j75ePU7GNd+3!QS1HT>HCH{u!n~%%)hn)sZT=#&a zV3-wDnf$r3mq+xFG>(JiS${dn=iO4-oe9anp}ss2*-n2g8qa%%R#bsfF zUl;I2%`tp8!(S>2w&1GtmMh^1tqU~=gEk4!b(U zdS*30$?X&r4A!rvU=Qs-?tj)P93P84QtLVYRY(BXX=Qg@<6#R&E&&B3)!tdou_S5^ z7kl|L0-ab7-MGz$6jXQ!GQWOwnhL#;pN>-QhcL!knQHNoitjbtA`#X7s@Y9o-oArR zq{3o0NpfIYi0t7%y*_iFrQ!N6YU{?7c2^z|MtqPvhge@h?e}E<&n;+#^0wnND#t6A z*%btgm;sb?;1c>fWNPPG+(F;`Mms(=l6adbT_aG{R@?HUZVF>q*O&sAclL=Xww`E| zqnrLU!Fn&Q20WGGE(PaLMz#!AX6#(eB45_Brav@h zlG3`JWiEL=LWA@nUTu73YcgFN0OZ&ler|(3&1D_JnfakM6iXn*_vNelC6%baZ0u5q zQA!J20FeA3&zvqH=bMC`GUZth{b_GsCMIE~p6-5btgyC3qGT55of6#HbYc>^@P@t@ z4KrCwgJDjD(lX#ng4PL=*mQTr@fcmFdNy%I6;$~7^_jNCKbbz}_K4Vz!Mp3XtsOm_ zW8Gc1FmlX4|Jq@eORTW*kDR3Sa>}TbfBr#!CXc#rfOYi!Ln_A@zuJgDE1=>i&)VC` zz!Y7doYaGzk=^5RtSGLKV7D3v9ZPU&gA=osmwgu=s5V&{30R@scoP? zT1W*kt$TuNv~Ozs+xZ(?bSFWR;nxg*c<*kTmYg<6#ca^z!9b94?h;oC*y7UUv#?RW ztaMbydsZZOS7I_w4o_yl-y^-z6;qA^Nf_23+-Oip(iy$ zI3L@k?jP#0>w0klhdy+9TrYBSt+UePXIlVk-M6M3;1B$nXsaWT=mHYc)~oA$WB*iU zk5dcSNnSn4gG&kBOT%GJZbb3x_d-$eo1cb4l9}nv6yv^N)ZH_TuHKA97k(nXNQYVo z;brh+drV5>>0~*@BFJ&G5%RrsAuIMTMIbLzciX&$5!)eqc#pIcI=tA#>@Gip@=f9?`RAIE6$j&ff(v`b z$3SXI;LYe8Hgcc2QWuUlP-%QT%JHbmO<`l{-Z!*JcmCD&0n=m5SyuU^9wMBzRkEdr z($MqK*pHVxmetjh_<^Z9G(PWV3AdRbmQf`;%D@9cP1)OSs!7ws0N8AnA9^z&A&l;i za}uu977XRLEaEq3S1r%{e;vC-W}DCXAhgv;YI$<0qe;jVf7d4=HtoNc{#8}X*gs-% z9n_RC+4>^1MHN_kR96ZVollY&{HX8cGx`+}Ssu3|=3DW#jO^5v^WT_Z-ti&XW z6EYlAJz-IOgtgrHwXBb6Zt`JitI%OOM}GV2Fz7PYKMu*-YhYoaY}j@eT_Z#ZWjl=a zDpX9!WWV1>+ty{MJ`{VVqpGIk7|K`o3D8lixX@}PMK?5nr1x9#?*Q=JBdN?Ll z+Y?XED7YCLgcxr2__Uur`_C()2ESw*@(aS+8=xVKqA&$Yctl zz}F&s`iVC&Ez{RKU?asI<`oDFbaY=i9^fD^#eQKs=dn0SCdI(1K zQ+6kzEX1P%eLo%TkX6hgRwRJ4&OqOZ9-x2b@SSOmTpW>PM%Lp5>6KZMq#Jk{8M8;~ zlOwUmpCA75<{9ZmKX_B@ICL4($lI_NBqc$Q zA*jnd`&W4N$U|*Biaha(gTGP)O#-{{72vj~niA383pWLfUtDZ4o7omoJ^!moZxYsN2_D)s8v%@ht1+&8%#n$| z#mDfsM@YzSyQ`z8+;p~ER;w;^uwY{|=Z6gZOrwJ%s8mqtv_%)G|34s2k zQymUHyX~To-_aw7`o#iivO!X%p`LZP+m6~>7-zct6uQ#m%U%uQsY)EdyciGMgX zz$&R&v!`7$N1@%)2{v4S51D@0GWdY6vH8igqleDZYzP}+aId2kGz-@>^y2ndU=Y42 z0~n>``%|4tV@xXWBICIP+2ZNu^~N?3p4h;pJWTle*~6f{#Pv|1~^1W2(b zWV~$H%*4msjIgIIOZpOJ&B^{&w2r2`C+$4dYDmFb3KM5x^RSo2_M40o? zb>^l~iCd2Y%L4aNGi}3OjBEOuwb{a@S?Nhb)aoF57mBh|Jdb!3)b4tJ7E1iFDs1+E zE<8#_cnW2Zj{?+IJsrA*&iL>^0`GypCk6QH`AVy*A+>}dvJ?C2gBzX(UaQo9-2OIn z^YqT2Si!(XJD*574#-@9z$>Yu04!jjo`eJckHF2F!Jc8)!}p$IiXK}y#x@z!5T#Ma zjbRkKg!NN#90{#yN;w{I@vx<72&ePHO;<*b3z$uOf^9@>2FAk*5k_u%k`FuJTm9*h zbuJhWN9U6AZ2%+D9T%4i=x-c!mDpA~Hywd~HmvdjieJ#N{V6y}4!C_8U`TP6Aw2E2 zyJtD}mzv=t4B;8A?tpMAE*^xLTpyv^t7DHV!ZYp$|AT@5SR%pBkD40_cP(adn6S(n2y~ce3Xs)5$ zTVd8a#n3A^M+_#3mNpl+;>XEMB8pu%(b&wn1m^Fu zLX4Bzgim{L%|xm)HKO~&Wxp}#W-bDtH+~$JfELQ5BWLp~5Blp$u*!;m?)E&wf#yid zzM<}^Odg+=ssO`e@ul>DO_=BnYLG~ioAz6XuHMIoV9Z^h+!{d8-Zl`+>gF7?!r|CM z47PDfBcAO#Z-V*nhn!V<&1U5{HNv(&mN{u9!jzaj-~f+P;lBbyg{X_qDI1TCRWMOl zFz1N-lXQytR~^q8G5kMX066;ApBjteqZ@%85I=vm2CPM}_%M@xFeK-koz#%mfSpJK zM)DOehp1QLN-bY-?-8mS($zW4-xBA#>`@eAxTg;utE?UviwaT^8LRHRa!gH_NMnDE z7>iu6_=O*$&&`-egFX*#IEMS{keXOwmSA4`6cfl;sk|7i%v4P}totP34X_}4QJ+Pv zT&skfPXF5MIJx;x%b9wKn5O6YePV``R9@!h2i6L4xyzQ=ax-Oza{ocHpAf#g+jot! z@~iVtHd$bHq#(tH4k(iye@a}X5*@1p{y|YD>sY=1n=UZ(mhwR;pk%+?Xp94oJ+K@< z7e?stbL`H<7Cv4(W6ALXd?E+`wBDn;%TQMTwypJ=DqM9i*MB!|a>EH+$UgUjikaIX zdsK;^9Um2E`JUl=s^F{I89BK?K9ozAxj)0g=R(uZ^*X|9M>1SSv-&BZ9yHv|-Wx?( z+r}QTK*)?8se9DkT;{&^tX+8R+%U%U+qOz)3v7g>T6b(vzt_&s?SFwP>xO&+^YlyZ z@ruTRJAM5`NXdJyn^X`}T<$Z+e+D3of*{%U+ zr;DsWl5w_rQk?Atr;}hj`8pIb`oSe5+K4kXz;L+N^Y-vV&t*h%>G}mAF1-kRIl_&i zGWGt?mS?PISY7R3h8QiLi`O|348`jz8*Wr^Cy+>Go+?)N=7(4XATo zn>(wGCP`0%n|oIOOe9kf z%Uynww;zHVjw&i@iD<314OJR^&h9#Bcw5kjz+#)?)fmI=grMI#N2+XWVd4V3JxXA7 zKMCRd0e&|io=#b!5cEJ4^7VHbh9*E} z$_rY$lVuSComdHJhW#rb)HEKoV9vQHRhWT}^!%#D(U5HL6bPgx)55 zl@UcXspj*&v^=O`eL)N6bP&d$?g~jX?bmVyfS`KXgsY~snor33*VCiVv8OOjzh%>R zaR)^-;VUNj0us0s8JH%wcJRp#9xAg=UyhO}Fq=eHPbqy~%W-u3~; zrS63H3OUHN20F7I@THGd1cs)LbJ`6{J&tRSQ7NIYGU z#V^*yJGS|z@h9|W*>8J_#uDve7N<5@$@;8ri#bSY8&F{8QVULrb-ezoS1HM=wopz7 z*z%+;s~Miktqvw~zCFfF?X$SOId5<)F>R^$##(;uD~O%o@c&M8|Y{I|S_FksVO=dkNFb=n6*?gR5p zJLb>X{eBoSx5-yuCfw_ObbfE(gJThCm|vwwPUmomim-Uqr#h^AO3+r`^j2b>{NS>d zsuHlf3?&r1XZF2ud#_ee!S2DIwjIG(bh{`H%zTC)v?=20g1pI)qAE4B;1WuWDfQ^8 zqyekz5b0tHYk69L$yoYPhGGyu)!YE#l=4=XE+SP2^p*5Dx^b#^QizaS@hqncQOmOd z4-)5=-(_r@jGyM7|J-rj^Oo|OpwV-Yd}miJmVyN`6kbJ7lk=%FioV`MW-s3*v!^s# znp#{@Kg5ZRD5G{deLn5br&sNpbhV019Os}}2g8U3rh~WoPDEAo5_!D(M)gKpoUb)w zZ}5=eEQCMPtvcltY`a}Kn~me-c7g0||Cjxq!;q^C_La~1HDWk`IR9V7l4STz*<}BS zSa;xke`4cEl`K+0IBHZ-TlHB#K6aL459+u@@aG==$4pE>@S}$;ZVENw5JBc{=^Z$R zKiLfYMArGWS=ni1xq`J%EH?AmHDTvyn>>_NY?hkQhnJo>a^YbOT>J|kk>h=xy zELV=Nh@T{EDfNT1DV8vWg4qGcHbKyxmXlKIC0Z^HyR)kb!nav~r9dnO5PZyiO1XV6 zKqDWe$1PVC)|Dkc4A7yslodmXo;_(O_x@{&#-eESkeg@6Cr177Bu+?#36RgSMcKSf z3et~CgI5ipCR;)G`RuSgsbIrmY|tKFRnoSBvc4F0n<)bPD}j$pq|FPapdT1m8&=$X zOZ0*fbmsX{a|~TUHL1tv<)*|d7A5%mdh=r?^a!IdXT6*EmNjg@7{MQ9n{Nwf=~!qA z!pO$L+BLE=B1>%Y2a{%1LP!9GrP9S0=oO|!Yqx(Kv8Un5ZLwXuki4LXY)ywhT$8Sj zIePs8Y(@ZW395m)omY&qA~d43$b=dL3=7sr)@DnFnLHZ}ucS?(?Phu-?e?|%Bc2Vf zjTECER3xI36YO757j$OX!$DdI5j^m)>2G`^;gAX>+mz;1O`Ad|TVkvhCH2V^Mib?+ z71tf>cjq9_ov}daIby12dEb`Iyd$z5FY5X*>R%m61}b5HT}f9R$p5GeoF76_W@ z{tII-`~4F4%u-F#qZ0vekH?uwu_Jm`I)pg@IG11toicPLcpp)Mm@vKljsxrp?8}o@ z?YWs^UeiXTS(-*Z*6bZZ=S2p>zp=fm4V2SFwOvel^dJ;!C6iKq0A;=Un}>td?f8)p?0F~Q=}1t^gVzOwZrxltm%DB;sRWi z>6MH9x2+IRaF9``dN}*a+r^cvKJ#A!PRrO%<&3E5x+c#CiKDxXP-7s*- z383HJ4A&HOx0$r{X~tASGFbqy;eqikOx6DSrur#8{d`HU&&V}4_9ywzA@RkAARp}4 zxB8(3?~>QXq#!?cq&$tlvSs{L>3lCrY5P(lV%8c7ndsgU1?b7{<=Mp$Z^7OajEC?0 zzF)wTMZHJSz_UEQI6TQ}q}DJ~;$wxR1HvUA8@Ax8@7Irro_aDO5wxRh#4b@xVhhtr zro-gt1Ant&)FhA#EK=7@e7=-DD3@7Y^T2!_J&TgaS^!P9h?OmiXA0fjU+*4r0RMS2PU9^CLx1Aj8aLFqa$|Y6t z9-41YF7xIF(Dh5ibJnf`+~!gWuv^E%aNbB-TW`{KYbs9qt*C@p*3=QB4n0HBkVb2i zKE7o5MwMA@kJOU0bd>TIz8D%$5UZ$4*ak?yo~5*KM(8T0tRIwCr@{UzABgjgDFVeCSZ5eSfd6OHRsbm^5~T zi{x3`pq4=`xGn?_KbY4Yd18Mf`!ziS_a8Kd=RjX7jtacu9->HMSqf{WJI(-oR=96i zH-{I9z&l&njS zh%mx=jTmdsU9*6j05c)YxMV&(FF42DKQgiZRsbfAc&_xaoC6OniR`(a3C{2$ z%J}nhf5>(t1ApJJjz@-J2llslzq+DBBkI`3tvlTVo={Vfl%B@vT?%N#gpR+`6u{#$aM7oo0SABZz%%W1;>S z6RzP7H+#l>!09W7!@bMjo)qlP1>l7wgkEjHSI0MCfV1lLp&nro|2tRJq2eO)JR(I~ zc|NIw>+x41*IZ@vDe6fx9*FpsTRY5JtJC`9is@M4?*})K{7p&Z32ZBP267KsnDemb#awYeqEIv`N~v%|LBW#1ReM!h3lt~;f6Q$4tD zJmqr1~a3M4=W33(e%J)p@oJC+=@e>>!(4zGJX2piS6T1~pQFLW6n=Uf{GURB2 z&u)*^MtDKa_!F3+4iV1PLS2ZE)=jbANgS{qB0e8j0Nvu5S4_5>U>L7@hKrkyEK7T_ zi2%TR(g0^tW%(q(O!Nd-LA`P!5rHGf6$G0pyn09fYgd(1fHJxFYl6z1ZF$|T3*%Me zsCu)8bG~{ih$A2qoRA4!?ZQMF%cL$L$=pThMNI3{--Zq*rz@EhnvTQhFotOibX)Z~ zYuh-fv-U`g17r}`QW1P#;0`q%%^hB{P_k6eFnl;Lr%bNJ>n`VC zf@=yZ`P9C-U}@m!#D!>HvED#5PwOpn{&9S?7O5DFTua|t%m9znMSgIEOTdYdD#9c! zt&sG5XP406NcNJ29wImgPP@xzvBs|(prE7ghZaDD5D+o$cA9_3d;;;9Uj&0^5AJyd zS9fE@5;)aF^WyjXL*NpBm`U*NWZ z(0u#VuBXqrD81_SE89pR|po|-TLmR{W?DmRl)3uBwa>1i2jmA zjU2-*jDFI39Qmtroxz9_D${gEzbVzGA3SFV^>UbYE75XW&32$X=P?5d47(znPIkEJ=HV(q}S7#Jb zjNkRe^%R!$v@EYUA+A;2s}h#KaxDSZok1dKaJJ$eK`#k#ul2;}7T=z*sHLHa$QL8C zU;ZhxHlwM}sMkn-yVEcb)>3wW2l*i9dZ|pBwkR0SItDXV?YR8_KTu9DmEWdZ{dtHI zh3N+Th=^P1XM<%l^5U3)UYj`zoM9wG;@oqecf;s~;T^F14jjo@4KWqj*7)919*=Wa z8Sdt-LnFws)?i{BxuyLXBavF>`pc~;_;7pgJ$XN-wI;#}al=w_v=9CiI7d+}n<@)K z>yOh2Az^dCn$hXxg#+l<;m+Nf0wScCw~;xB@CB0dF!Y7t&2EMz0uM`V7}ws{(ga92|K7+{^WxUn}gI7(`8TrbbStn7pMXp)g14+*ky6 zy)Y4gp<_@1?n&8uWBt5!aA>YFAR0NVy|EKMc_kDu;^*Rp?N%F1`>K`Yv3HsSh3Hzcc9h;iwgf(%PQrKZ@DngDn zUwM=ZGSg4`ss4QmALK(NZSY_792E%59NeO^v!{w_mw^Z+IA4S3A#6lnoh5~_~5>avk>uBRA;O!;Xs$c zmLKbzH92?Wv2K*aplpB-PTGO6sO}kyeWQuvYq{bS;%l6zUgXjFRF2L&{F*Eo#wG59 zl=~rDP3l0MBbSV+pV{UPv8!7r_0jm6N8QJs=Q&k5Cct6Fz|e_ zT8QlhJ7UiyuPT~u8qI~t+-{YLRXE0kS;`jq2m!Z$5qe1qQy>00RU}aXys9qbd(e%< z)WXZHma_E9>23_Bds8xTf=sz3myZ9e$y&prjQfwr zMr==-lb9CQ=bzZ?3n8dyW(Xdw=#YD65|?$O*`>=9#hNq>!omd5NbiAe*{)dT$)7+` z?}*G-knqcFJ}ig-y=-}LwEqWS-xh@aM=b0Gw$3EoJpMz(%rzBSrxV+teQb+-EVTZI zZ$%Ujh>@dE{>vk9QyGt3nO>nQ4iYaX$>Bw~WNPkT+$;e%k%G!qkcnH5pO$Zj9fu92 z6^Ip)4}_w``Y6jRW&@8F%8(25wQ)l@t74#XaK>rYdw7zCiChYw>;OKoYT{Mu@RU+K zN{*Zw-ok1*gqB#+NT!Oz5{3}y^|ACjz5(qDD>Xm1=p)pP>g`2~4t{+n$e&pzVK}$= z={{(m%jOzyBb7OHuVjB6aeo2Q0$M12(DTf;E1XISDqA=grkxMtKQ6pGI6BAF4<-#F)3>_L^e~eH zv07g!TtIwov_IT8VOEtu4&ShlN44k08Mey_lAE{&f{~+fwXur5J4S=9l1QIImO>=d z+o|+#=GE z1k-h1zn6jaznbHnl;~GDeg@f6l67pTnsyRmL@htf4UYQHtx!qcqbD*;xX~GyjRK3G z;cJUQfBf35M}ur7tO}FbVz@nnAA-O2Yyg`^|FNq^ki$O!)nS~;{||JOg@J&9z|QC& z6c5k;xT61|ql_Gk|5GvjN073!vj0!{|3XLE7}%K!{`dO-fsT5bs$@^K(n7)w>|ncP zj_&O21hL=V!V$J}50HXD-`?H|1|s7GcJVaM^_br|?S6fM0cUpno2XW8T4nDPS5Bm= zV1mlp#1108zR|%Dl$quM;PgQC(kA9cfG!H}iJ|lliO9%+4M}z@)rcaD(dr3>}i57YDjq6 zVp^ZwUWAkRLL>Tjw|`+`X?yTU4o*#clqcg?D4GtyBu2LSdON$C3P8-Q><^Q~gDfzx z`!iz!ncvU`j1fc|Z9NU}Y6iesk0zk2sw0X@P*GT3TP!v8oT{<1GBB`v;nyBsSy>^? z1TY+ox;zYM01M_{+*!Gq*Z0q$i~aW$1^~3B?SMB0^Gvf$mr(m#OP?}0*3i(?D+|lt@sl~fI)C^ zar{&k{`E`C{?$G8tB6|fyOEh2-C7&}>64!xpIzCw!teQ*v&mZD;?(Bq;`pN`#xO9m z1pXb`{g}(p-25?3LRX7VMNn5YPWD31#VzXo(RRB|*WuCmef&+5*N}+??;n{41aN2q zOy3isvNfmrXJ_>_x`}xp|20gr_jzn|cMSPWud2PZ(Y59A*S0dfIJwoE* z8e3~%+3@$^!%X;X!wlX9kPc`d4zNBn4D?>MtS|Lp4EDp=*V4bX^`E|JU}kaz^V-4! z+}A_k&W+s>07xfSYoLGUNB2!G1mFN9L;GJ_iTt`MBK%I$*_fNb@BgNK#7z9r{meZq zeY;3n?%9@C^yI{l_oG;c0;trxc_0JN{N-`K>bH>S*jS@PTMw}OZOs3@GBvg|KYYvg z{W&gx{1r$&{Xw)kr`A6MWKn5$rfYmxRsA(ecVYBI=>6B0VBfApei_Su+r(My1HG`g zQ~b=({Sb#He|0VFw3!-P0XaEW^0ehbJ`rdn3e_^x$6y$VeB($=B`i>t{lH0Sx zyUUBSIR6hmDq`Tf^ndVC>}N7l<@zywB0zwQ>ucRzApp2qx?TWgFY;i1@DDW{ipQ~M z^g|Oc0McKvYJmLbz4IC6Tf1RMfXTm@c0hpW+Ng!S zaA*LQU+kU}mJfC#V=rMl^(#9&gePsxB|iZK0El03Z)&So_;g)fu8XJE-QSAC&U4Hc-Jq^z9;NC?i&+zYJ zQy+WrKkz_Z0e=}kG38TxaCR*^G}--rXaAjcjt*`A;{4lc+JAw2ZANbJ?_+Ym_To=y z^Un6R_q==h<^H5z^Mq7CnG61~1g8C{-~8}&{R~yn$-S)VOG_@cjKA>d9>wV7Qt>*c z?UsO_&HTPT_Wt;O#Q`V#!1wqIMMbW8yD>L*ohQZzM_>UO`Ye$^j82|^;xv-{ctw9R zb?nry{T}iV;Q)eo`^g}rp_;5i z#`^<90%QR*R6<@Im3Z)f5`k|bTPTFn*WKu#}W4@1nAmT?t#GP2umxGv2 z(19np#x3AY93NYx*Ugx;54FPqFN?wow{3FX7HE*`EYK-UJWXl&pYnI2P6cAp<=D>g zqRQWIK%$gy?sW9-7>$dfJ%iKE&0wE~>cveBa5}jaTO#NtjwU2isb-bFyyr5#xyT@Z zDu%zQ8f3b&XBcd<}Mjr4`0Z-ZXrLml)RmDB`79oP5=ULF$&WCd&@UY^Q z_gu~l!n^mfb9fbSlnv{Mf8e^pluH#qazZmYo*?yX(w!wWwroa)L{G)~5@M_&thFXK z4m5V;rOBx?QCB>ZU5#ZyLYBSJ4ACaGSxsB;FHTmJC|XdVBDB! zW4q=Rsc!wrc;JkAMs6_K_bJpNY`Tp9WQ|3;mqje_X_0DMOY{hu*pv<8#1<~&g^s#? zDF^zkYYJ&WjZ~ew#!JAN;n1h)i#3JKIUXR=PnVFyNzxL=!XEE1l$SU3&YAl_ll5P; zK_PWtlQj^#AJf zU&AhQQS~LgJwazZ_~y$PsUPR&(61lA=4}WrvMa5WQnzza%Gy!4l$rWnb49{2jp-oL zOPGY8MMfb?22tiaT`-Tx74U=|c7{ClLR@`%kDD8+IgmzO+^~5iASFSus-z%`{31Ug z&%umN zPWcXNZk9sm)`)}ImU-U-$v*3RJr!mVUipmKJnrvYpg;?T)*KU>446EytTKsp7-8^w z(rR`@tmhR%*RvR2KZytkeyH*t`Bz*Atd`#%Lj%vU4E*D=cU&NSv%#3pD1qy;Epu+TQRk zmTTfw{}`!?|5lS&#gS2u5cT#3$+qU4tof9s$qouR(0S+BLe);INwHQ?6%vKb>y_r% zFK8yC&Yv}bAkuJtlUN{(?W4ndaDLf%68>7#|6r`$UE8wX?>@fc6|N#E+aiF1%2j)O z%N!(F0l%#XHz~!8$3KDdTg*_>|Mn$jiCK0IS%(BpW{KjbBo4e?BjdZURZ+a{UTP+v zzP$Ssos7pkS$mJ!3qRWG&d>(Vi`X$8b5*DLUS+QePfJ+X;97$at*tD&7NgfU}##Z@

65+1mVVRG*fULx%XXf$`=rf<#Lz(CW^8HVoUlkyAaz? zImy|U)6U5V4@Yrc$V%E+sm@eQQh966UY=6h8x zBx%MHhDepbo`71Jd_9Y}7_Ncnj;KakDkGlgEg6{O7{vBCj>D}OmJZD2yMqrhRSki@*0>{&l}84+Sav?Z5akK(v?r zpxlhar#G<2($Y1%Cbs?{WJGGoD7-1!xhk^&ffjbSS1G%4ItKdy9&5HV20!`1)hgHk zQJD*l?HXj01qmrNJ%W5%0;b{|9K7e_!5(;__oe zrEQzOsR+Vu>-D%ZDK!R{8NC(2|IA5g&WYoOn+r1?W`u<1dsGRRoUF!+?MR4B)jVd-I%&uB5+dh1t=kK>V0(1g1^g5cc4LW>oReJ6i!MROl{CZ@ zg}ZVVEhxtL36?L|;nrA39)xizzqyMM)OOZXW-sNp?s)EW$lvbuibKE5@SR5*)&t|1 z_nu!Akzjm zt8iFl^hn+7z{hC7eU~f9GTm>al{MXF|L-UYDt~BD6U&`09wm-{ZWX0mtyfJ>JP>51 z+64`Pq5vwhex9g`!2p(|GM_r1D$T&by!B|yQK+TGws+TOA$=ODJ>VAyhJT$q==nXm z(^kQsDEUAQQvm9vftuQ*T}kaAwMRmOZ>E128FfRjJ0oKf|AWB^zkUKKPSDLiwCf0W za547p=*pcZ{+E@bZ%#1mvSNi_%p0X$W%QXf6#7p$DjR5gB;gwfZ~$`b+P7kz_6$xD zBEkKnV%r2LO054vW4gW-7-8>7!n+}#O z7wGU(SYW0wAkUr5U*ujhn4PrpedzD)qpR|d$9r8}wYqh^p1ndjYp!1Dd~mEwUGxqG z)45}otv}z+j^fK)vb2b`dn?mrto9|JHG>;P&ch79pYvhu8d%zdur>p3BtHst^XS)V zIq^5t*p{jyO&OoxyKw-D<`}D?&Y^(pb$R-y^YSN8nbP3k;Yj};UQ2N@?Vm`p8`J>p z%FhpJ^zD&aqAVM5|7eQr_}g%KHqt1#B1@u$HQGIxo)h#R-u1$frmVV*S8Gf2_~=iu zn!)tt@8aKypCe;J3hWl?I~k0SLFI*pKian^OEX`{vCxvG!C7Il{~PcK#SbmAmwUux zEn`$E`p~F*h_{bmpz>QEHZdLERh(j&?P0Pg_7bpb4NS7d$slb}Lp93@AdvYhtu>SJ zOaWKi6^&7ahMVQL1{Oh8PYf~>1bqvwDl**m=;Tdi)4|c-mdiUIj8$NR?+M=J+yha! zlw?4Am@TEYI54%l1FP)F?4n0m@GUyv#S=};R7m}CvO#zP#FC(XM|D+}%AU}%PD$5U z;H9ghqq@;8PrDYE7|DaP8Yj%=ILQ~T<2W^mDl-m2%w<0%UM}z7UHp~w!V&v^EZK_U z7~^*vH|sy>TI=7+Yja6@$9<*0=GGa41Z5NAXIrA8mxIPod^2qzx>U)fIYSaW&p_Qz zRf@SfWo)!sK*Saq+HOgma(N30cqHSvlu~qwxIKSv?rJ7W_g!+54%--*CuuPnZX#;} zu1VsdfLwm4t-@J+e|qh`khCHre_-;NY*x5sc`mk$0qZ2(yoCHSrXJh-^G6p9j{4#M7TpjM=gRF-j$m@%toIDsO9! zjeWJ1Y%~;#2!g^sTS&bG^m?#zT2UfrCz{@&so?9$L{Jt({PTrI=I{7^ZFXw zR5n&Pxmz*9Suwi`_i>5Ba-cEJ3#UT-(0ylgw*Y_t<}AhZbCS1*mxDou177;9ygrkdV4iXbg$ zc8q2UO0FsI<91*zU|OJU^~5=Aau6*cy0d8u9tXcxUBMu{MTfO%g#TmpV|HSr?&a;Xda*yf65%jy^ zHiq&H&{n6G;I+_F1+-bFa0&i*7(OtCofU2)(m8IXe0{bSFCb6r{^43|yuS9Ef^aDr zvtOv}@n`3l4Qq-Wrba}|^TWEpT{uuno&t1z3jNwQiPp?e-ckB5k((k$iTL4|T2-m+ z*47Ey+l$ys+pqiFsJD0(!~4(hL09+A$0KW42iEZ*^=qwRPvVJTeMFTp<%!gu%`$OY zO!QNw{VdAGC|2J9eA{R)U&@U&Qkb8^ptH)hoz>Ah@(<$1p#May_~H{7qM+yJ;iU6f zsyZ$lQxjHlB?28%5#lfchwRIClcSH0;5<`bR=vk^M)-tP{BRF8&mx z*39O;I!_>c(D`Nvj6nmqHMH=Cp+7Z_?rvCXrp>w@7HnQ_c=oBUUSgT3j3E5f|2Y_` zZgOy?+|(hie;9Ie1AQ`zj4Gw}j!C`<*a{BQrRx2XZ2NT%4~@Ibr6MxVWN3JynFuDe zV}%XQa6U_T3L;a=_FzkUx&U2~3y}(nAHa(N4%{Uf>prV;l}Jd~G&x|`KKc&)IR|%o z9BIMH!OgJB7vg!CZ+DsBQ5dKSyk&3?F(ll_k;d;5rWP+j!o82UDZ)wk<}bSwll~`W z+@*g!{112a5b1-P3zew-E!)(ERMob7el!MWU`7sb%MD8Zp*wzcWV!+>TZeYPR!NvE zkvv86p?5hXB3yFiq7LJuhU8FF?(SF%IkDEu=0|N&m~dK$*X?1V&rJfX(e$Y90deXS zH>9eqH+mm&0xX^h&%1e`PofE@iUC%2stmGKFs@4F+*&!LE4dh|Dxe%OyZ#cbI=FSJ zD&*47Ui7nuM%GC3CauYz?_8}yIV$BG6I@XC#VQZK(AYL}li*x%;YbDSr2oQS4TrE0N@7Gl;bBL! zc;lEQg*T>>iJO|DZ-(@{SGd(P-_20fYb!-EsM5rFp4}1M^5Np*O7}U3`DA_5={(Sf z5h>1k^+38k8jrxj;i&k$sI6^*N5>DfzyY=324jlT(AK5;jmg1@b*a06jy6hOqgxlcA2Xvu*2v~ziR{D?NOIorBzJE^E>kpzAR%DL_ zfQ~APhOSoFMDo}deu@G#nmre8t`HHwDQn2LZeArMYA`xE-9}pWS<= z*O@J;Ur|MJCqSa zEb4bZPA@FVy4Yy6Ylcmi_LrSRnxOu)(&a0m+MG;XW>g4mVY z+eQrLX^n|ltB_KQx<~XI7;;77yi(7%Ab}wwWp@;wprshK>gb&$*I}a?3qc5AA|6AB z*pd5c*}3WKP1yqxF4F4<8wrlEBtTRWWHWCje@5W#!VY_>#yUm@ic#uK+0jqd2)mx9 z&_G6GuK6^{*w}Sr6yAAPrEnH-@-!F<9w$=6JK!qm>7tG0`}xm4X?-(nFo^#Mk-3ay zPytBUK6+Q*ZHJ(gyMCx zsF%%V?#(?Ds~708FyHV+3n^ecNS&;q2t36$V`nFBb~!&~?*2xW=a-yi_)kAzP=)lT zm&_pjVr>z=3SAZnlw^vF%v43|EUP4f5}X?|;;NX~yDxu3Dx*y$WGs#7+4y}~?hz8O zKIuspp7#+LJF4Hu2{F-4P0Bwvy&xNnSc^StZoK6i+Yb9w6c*JT22j4ylNp|br#T)Y zdsh8f+Di60o6SJFnut1MEyq6)?pj0|skHh|GocDakcF}&qLIH|mC%2_w9s8aT*ekN z6+nZgwpmr3%zh#U|M|FEoGp@DTP50u%2SU4311p#*)L_*&)VM5twiE507UJ<2~j;! zdw^KgSn!dlu-(YqvuJ#4GpHiGE2L#`G*QiODcYLK)pywz2ANieW@XD-ZWfzUvq}q+ zFav!pPRn!t#U~oXM?X0I;iUF~ps$}bZHD-H2rCd|697*!7q%kURnMJ1XQzKz$$@Dh z?6OFlb2|QcrC96YS;YXiFu=&0W*`eMlY>MyD1yEyB~hegN6V|n&Q}!W;zAt=C#j~0 z(C(u;7kcSf4eXeIX3L}6BY&DB*Y0}{V{0_w8Wex7xnsbzRs zTS}-BE)cL9-7kh|)BCBT`arG)S=&9OP~o**H`{=ovBu{1tf`H$CSi!$k}JS9#f*91 zYYkHoBkmJHe(cheE76SI^$HkYh` zPeL{PBu!Wzb_}9vQ|PDB>Jo?xmUZ09A?U*A*>Hu-0I>kaKymLvIysk1w)GL#?6sGVLC{WoSFnk zz{EhlG(n$$Xk*gOo+Uiyn9+El;5|vFM2YYIiD9Y-s7}{)$uMVy7667%hoPcrVkE#VoEenrr@+ zfLseL=o*vqrE;A(<^rH#9i#J^k-?H3qkc324N0_+*j|-3ls>6MYJYG-3}eO=6%zkp zz|ydC$~o}NmDQK)B&l-Yu4y;RXQjhqKz#ZY(vYY(sq?#ZpsWl0_1`T0&@R0;zMust za|q71H0dJj#jqZG16t8o^fKYK;MY{A^uuSsea|&n(;q&c~!#Sc|T}n@! zwFAh>`7UeOyNg2M-a-Mb5_R8esbnZ^7j{d@P7~JR@X*COWs%I|v;Y=N_^Q^5&3BIZ zN5wI*WD$ZeCnyHkP)dsA>(scl$QNqr;jPnN?0#3#67{ORH1|io#c&*xcr5&lN`X4f z8GeIRSLatHnJlTcW4&=s4mgxHq%C9i zQCDfBqysY*Q^dshj%t9YwzqYV?h7MyOK>7y*#LnTtGIn)))t1a%ZrmZDiVTESThH7UXo`rw=E-3qgdg zobb?6uW7fZ@BK)N;$Skv59dnqJRk^p?2@8E9)GZVcFUHXOY)|D?!e`5g>zeWrb)k` zW=-GWBNW5!;0NU^uw&^8C~0;GL;`<0^;BK>$(RvqgL&m(e~TOi-4qHJ^TKq;i!`qP zo&{7nHz56kh1azOa!UTVgeO(sKH5t?MTKHqMZpBU!SOAu8{s5YwLaNFf3u|GsVi`AgHm&x9x}@nbEGzd(XP@1>-3j@ zDf7dQx-BW^4aD@bZ+2AJl%i<}sYTG5+oLpKLSej_l2miJ<@Q~YG=Lnh(dUqmT}{sp za{nio2JJTt%A%j34?)vRE;YW`H)lpgMV#9s#ck9h_osCQN3Oi=N#p0m3ajJq7as`A zj%}^Lz|#rK$J$0nAo#MI%zu%XCTzVg&uK6H28#%lETCfi?zC|yd-<$fCr z1a22RT^tkp#i_4^O0QSwFJC~M?QA5&`-t=gbFr95Du%U+R<~9VV&KjaaK$8zKWXcf zHImpus%WPz8{U{rCXD(G^fOLQ*P94mWBhqJeVrAh8bBd(`wv9l$3G*tZzdfPG%4F3 z0^||vEVxu!QyfXOhcO`S$25YOsh^1UNY(H6d{xx!e5A+-%Hz+>3-Fzo#yA3@F#*Ol zgJh!!uD`pi_I`fM9`BBT7tEg7xLqIRLNuyxKZ!R@~^cO zaOSb8_DwDbSCO#Ld45F4K4s4sQ#HL5IV>7<9bq{N4uDFDF$t)xg+ZV0d5nDB2V86{ zoSn1jqw*FregCZJkcO^pS%yU9n(c)?ao`1csF1St8?uL!^dh^+)VD5-I!ebZ9$6R1Pzn?9XaS3*VTsq1-^inamW8i{)=MWq{z zDo|ly^*^ohgF>YjJ#DnC|&9=clI~hj_r5pBJhAN(kj~o z%g8c%tq%y>I(&(7K(U4$rEhIFg z+$>VGh#`7M?z6D)uc?v;A<;_YBZYi=40M$t_G3ur0MN-2T+lXWDL0?O8fz;;fq+*V z*`;x=KTvf>_#eTR0$(>*2_X8Ki-!5}fzAuhfxyL_D;|AzHOELAYfW#}g=*U=!># zhv1G7fdL{p2shpQidV^}z>l9z_-M3M{8CTyzO1&((wh8~(rCHO=-wnkTXKOX%KyNZ z@~);R@!FwIi8xq1?_YmqTbnKmJ7uuLzg3O_VA`$gjUMYw!r9g3A?oCd8A$#AVgq5K zyAeI)+t3=YCcT|Rnk7miZqLik(d-brkfIhNmRHC73{|P*UcDBqaDXYCA8sUulcTKO zEYXPZq$K~vZLAm@FVY4+%~m)TytlK`tNKW+JVba=n1x%#N?iSy)>0)JE6Bi?D=2km z++l(Lwb~L{SJ4*wjAhc})s^K$fL)LZggHC?<8ws8Z=||A7n8$;&67K%g7QFlST^RW z-@Pg;Mdp4(`QIGq9ObNyE16OeD{$} z!;weYg^QmX=L$n+KHBSb-eZoHsc>VIO@?2RdqQN#KdXrrSy!NwWb)$m zCxfYaf}TWdK8dHri-7z$$i zyZyz#XV2ya!~&Gsggi;cUIi({S=HjwIz-SKmDF34jk?njuU77Ab^T!8Q13-HEld|Q=JX>GdTyxSImagI_q73 z?8?=XQfA}eHx$gwAt~BVBjGL$@z+5{3KV8z_VGMnM{D%4Pm9o+wod#CV`f*YzbadR zfVGioqhew9ixVI2~_q>b3hdT z@Mt`#I6N95h{&|IGV3HtZ3A&h7~1Zq7(2+T<^46_KGNmdTG;D_`+%*@R7GZg)!-hzX>z1Mdr*N;QTItNm{*HV4 zt#reV;Tt*d%N`RGIn!lq9U?N`=;aY~Xz!%+!kH|{yQPd?n5_qI_;&C)XaGfUxGhC8 zT6Vs)pJlkTsz>4DNc9jYKQ4k^^sDap&_zq?@c6(+-w(chxqS4b_caw>rCMb8`V{ zMe2`ijZAH@Byw>q&t)q_R_V?&lv4GUCY48s$68CC-hCl}kF^X01fLpF5OQs#{tuXq zG}NK>Qezke6@v?$DZcFfD<8!nQ~u4Z8lHyfD0{jCSecvyFq+a}?#%4m`Jj~>I`9le z94URp+(Y;o)qK=6&jjzObJ5Hv%vE!1)I|H;&7&tU$&NycBb5OKpYLoP*Cc>y1x;yK zX?EqH3+`H@cYsx-maj5-o+TGI+N|`m;QPLy%;Oa&+TlM((4DzTDVapZ{3^Kni#*DCr>uKe;NvpsBdDpT`9Z?z4%vs_m?E$R3`ocm8lnD6WYMFpCng%fjNRwG^ZR4gQWpamdK9^1!vhL$X-^}<8 zO_pk6=&Z98IP+@9v})WNIgN^|OeW2B<-D3l)Wa6WMt&W|C$P6_?$w{tX4hqLRk9x* zKImLH#kIHMM3Vwa|D@6=LOkVgUHC^%31RweOC zyeydt<403__`eBUC7v5mn&1zIz-bI>)V^zParfc2Q6`Bpbj|6vW!cNgdYJ>p)!ue^ z4%sg1QSt*Z0+slPYT%(+LY_%r-mYOc)~UyMt|s)m9YW}7hBq8bsojXF9xx$lo`_*V zw|XT}3dzKK!^QP?q1;RraTt1yBDB)xn99q_^Bped5 z$Cz$1a$QMn&T_{z1CVokZMvsV5XIV8FoW8&XWuT%x`qi48jgDay~*$)-|{_INqTo; z^UApvir{THx^WaGd}^ds6@cVsdFthXU1H>CLL`FAjo2Y0^)!)Ep0P2SO*`eeq32#v z6&43Zr#fHO?VIRzWSl7E3o_#6$c=0zHElZqja;Txn#OWP`oGTHq;}6VM{XOL1a8e)fNQN{klHhb3A#2PC4%(JSirk^ z!7rhBO0%8UGjN~|?&FXwQRa)`41B)NsMVP&bd0n$B@;VnT9%KBftvOFObZ1%zp*Gq z$#F|Y`*FA;dQfSIjyp)@UwrPZH`7jfD{DVqe`Z^V6(W_nhUxfcd;mh|GU*9^q&AU8 z+2P!^=exNnlNECBIR7ItU-VU0Bf=?IZF8TCH{xkGO z=lY_;Xum!eVCk4A*`S_*LDQ#e+F;P5Z)ESrsn$#6!{_Kd4m~(Asf2Z@UJ68A>OMIEF^He&@aRzq)esBXP`&3n*xB+x+*ByE7?*K?_WG|l!|5w$y zH)S?$rs!II1I5iGt`GRGr)s+BP?2&>ue6&M1L1D#rf$%$zpd?g17lk}ea-@!Lc1}; zom&`jPUW(^X?R@2TG;Q252yyN2}kpgVZir3&PRdk)xEOpcHsemAXprl=KEeeZ0#Jw z5H$SzP^F@7l!iWT=$UqWK+%o|dk(0SYRm?|hEE^S43@@lBWEmep20vIUin7}RQtJt^)ypn?pQB#r zwf)PA>1LC+OC%j8IJzlr=O_*W*goz7$?{O;6;rBb9+cp0$YK_n^KqeRQ z&%Go;{Emfva0OE4f&(QZEEu$2E3;{%tX&1hXJK3~Y155|;cfQdMl~w-U68kDrfe61 zDO>Y1DR;BI{0JTuZ%cMspOytstgv^%|CfjhO>eiv$V(EC8iyn?)y1WCxm`!re0m2U z1!cYIBIKLF=;|c)ru9H1bDb1Zr`Diiy1&rx1i4$eJ&PJkfI`zQf{tBE0b>;`RsP{^ zmA~~KBEO_EntdHFX5dV9d~xIv1G;8dbQw_ELZJ?~ANTu1iVbJzt^q@;`kT4Tw=P9l zs)S1Y+dV#;mR`M{Kar=&Ys=8?DxMN_$e_JG^v5ABFO_gy&=lw`CR;-(mx+z)lw5OB>4kU_Bdrw-Fd0OwgC&1F>&Dhe8?#( zq;U!zQ;1v(ZfEVYjh=b+5>nfqLQs8VBHwYiHcb_95f}&+ns=+jzRlq#vb1ZS5WPut>3ccT!}Ecdftr=L6gCGrFV(ZD z$@>nqq6Y2oqMJs5-74iSY5)aq*~!_*g?;H37`j;U7mOxv;`Vhaj8xA05gmx=J*P;^pzhe^z768a<}hgZjVsI4P^ zT`k}63q;aW*ia}&g%NpznbrVP>-c(dwAacgu4B#%D@H6}n={c6q3=~;59^&;51HK) zl5*!t+z}^kN9TJJoS4iG^y4r6_O^Ql^hq^P*|VyeODju=MabUcLCu=!zIC z_l9iR4bd7abvcLHB30bWYg~1AWGn0ywR#f^c7u~DIXsub0HpUgA?`IJ{s(l{Ox4UM z$MgxKnG#V80S?lAXpQ>~YSOV8|CaIiD_{wF3T3L>LJE#jw>#nl@;Iaq!8yw0qGwMp;DT}rnn^dab5t?q4^oZO}t zK3qoBD*SPtg8upa$9>no_{O0R#CM)CH1lUyob|G4p%&k~c*V3?wszy3R=y}6QIt{K z2C4IR+tbfAc0>2E0{OCuAqF&^H#*KeekNK9k24SPCmkL7H0U3fn944@i+X_;R@0>= zsot}fN5t`~k;~RrT`Pur4_@PZB0-rRYV^DfzJ}%Y8C5ld)3i`dMM|JDisETA8d3zj z$zXhvWJVPK9hFQ#iBJ&AQ(X2iOCeF4$b%g*60xgev$N<)=_T&s{3F+jCTrKGrg!t* zMxNLGt0V;Mpy>SI68vlB>hK2smG{Hp3^VhiD?U(nDxyp9?73ROOb6do9ej`XPYNAL z8m?WjM0|$vJLZeef`>ElnzPr$DOmA3d@0YE6kBvi*jAu;k_f)sTzfq3Tu3iP z#K8(p9nFs0fuBw1tQM`z%Z-U{?Qw-8Go4&caHP*>-z+dkN>mNChrDnwv@? zC-8wKsAPXHEBqWMi(%QLKY823j$nD)w z_G{Z4P#|S!ib%>EZ=9;O!O#x-LHgcv>dd5h9F>8H{Cb0J$vRd(OjJ>#w*#p_-ve~vcd)<=)Fp!7Aw>{%ptcqo@UH2)lkd` zqLBn1%s={a(;Vi*l-__z!MGEYN?9=EcD@7*9m2NGTqY%L&waSz*Nj`D7&`t79Z4Tv zR{faMBuAO;S3ROS<$q}Bi5CyL+arKTw;jWwtQOxmhEkZ?1DOyX`DWeJZqWxyd5g?= z;obQ;Qpt|UtN5tE{6EX{4~xtytymZxSob%nNmX~i zh_{}q-kMF4<$f@)qIQE_^ev21*YmWQaF^%8788!TEF>S9K!$}lAUH0MLB!gi(Xcxh zi=5`s$!`tQMZ*6E6kmH&ce0(=#yq@vG->lyL7xIIf1po^Aql*42B;;DD_?t^ixV=w zwwI?68AJBykAvk;+@#$rXB6&2UwX7))(eTi$3&LqfI`Y*W2LQ5$?b&9(``v9WW>%I z9RRx&juO9+6+#oek2h%H%A1B@+cM695jr_9cyrpbs3!wBPyryO_l+O!>KnBA{L6h_ z5bJonRT)YXJ-IbnKU19fd+&&wKT-rWhEA}5c_Uy#(2w$6I|+4ZQ$^X3x*RVlkp9!- zu{`d5xozyp7>;sqR(rXslZklVIz5gf#nx-w&e#HwVTG&%)LJa-3#Ps32TjK#xgpA3 zxP0=vS?}O6^5ViVDt|OuX4BXz1vUdudAJ<@DVbXkjSivC&14s))?Z>_n{AVTYW$Wo zHNv)lTm&%J6-1ElM1exE({}ex?7ks)-(z0H?z!&fEHeieRoo_tH2vD51-~w_>9Da& zlWjX)2fCZf;xugzV)G_Y!ztd#M{4Ts5_ zhjWLS)uxLCpX3+3g}nFm9f>*;5HZfov#In~vftLOT(X)FcGmpCrEbXEe>2^Ap7ji4 zUH6SV zzj}-nct9-o$OGPR|LjAtUEq0JximU`SGjs_?g5!u{9WG~7~(IA#8Vp*4iQf37x%C? z$#sisM2~;OXO%P{ZU@q|scc(Tm?-EOOAt|(!Bq%X!8N&=F30UNCmpf4(ngG452t4( z#Pwu?e(SqtAJ94XR&$U2jHZnq0Fe#gUVHP#vna?{H;P>(bDS{g@jqzF9DJ%HE3z{v zSzyU?KN#_snSdBC&}g#NBL+9SLAZR43 ?6*JM1{K+5 z1~b|T9zig`URfxH^+2O~EjKnpt$-b>r#zS)GoIFyK{;S{=D+)I+F1^p0*%+2>QFrn zPL{Rj&6b@BV+JS$vbZRU+vd3i}cyG$0Mob-aboEvQ953XA zZChE7REa-1+8`a^E)L|hzxpXQKdp85So4u#Xyyyd6{b=^5srr8s=r^=61ZnmmY}9} z6L3plyp|iGYv{|M=^1B092%337lPQ?6q({HVID4xrLjI7l@&UV{z^)Lbud6slZNW7 zPIrum|Kp8AP4D{krN=fEO+cVlg_eR=AMZVy^_D5H5-&nvb<(q)rw2_F%kQ-dlmo1+ zX0M?Mf*}YXqSsHD!ysu*yDQT|9S=^YgziF>{WzMeKA5?j6x9NvCV%? z|Cj>77LSWrEqLpOqH;QRW)A3pu@5?v#W&k!>F`x(wQ02+v8d+nMGfq z=p^y^`)?`5w8>}bv;MR#<*UP`N@x2iP>Z)>fHUFqJJTqnOBDGXMfmEtR6z-!S8Y1({qpk6g!}Os|zDaOatqWScln&@UU)In& zZ@U4*z;>GCQO{x0+T~?7Iw0>Ni8N1g`A%`!l2Kke{`pi|x^;>3d&|$dneN?FPtS+> zk7kNvLZ4*9@k2dwe$b!?EXw?#wm*LLR zZ+g*HDw)hV4Za(jCc7SSGKrbkH~M%_wE}fq=2b%(p?~2!_CI-X$h#ZE0E;X+zU&8w zb^F!io*Z36Az|idS^rgiNxC(sS1 z&5DCLzXzpBsU3>T1(}lIz_#tf3%Azei9!o;PPBLB{_M^6sF4b%JXlE=}d`x8!y>|9_47xQBtg% z-v;fY;h@}fa~F)|LsZf`j|1`OpQ5C!CA|pB&EZ&lq*wdkVG7+hYc=?6KmpxQotru_ zj;-Vr4#{!>zKp5CL6TXi)XA8{6m_XkeH^499_uEN8{cfJn$U?HENtRi; zhz6eMo!ivLtt!HU0xksEinte@Yu9+1_X_Haj9%6qrr0|Mw=05pzf25gvZo+dYwD`` z9E}p&}-8=P7CTE#SLyi zyz_Sif}~!{9d{^i&Mdh*O2>_$Smm$g7kP%A7rjItc4_5tc8lP_;mQSV0z=>PS$w~C zD}72pz3}(`KFFrA0tSlORCvo<6i=W|*zH;`6c!y@ho9XG5P3y}TfyCN-ftd&3`ye_ z83%W>u2WLodT6>8C_CGAK|T%jy@b?Z^FTWgTj_si38oujsc)H*!@vPo!yE(WX<30i zH7+E6#@Pi<_*o%WX`tB7y<;-7hlQO1un&!%c!jX836(D>-|p#fH1%a!yk3d%TacuxrgT|)ueQ9Nra z-xtF{w1Y)}LgVgpf7JZRTr&u%$c6~3o;@?82o!P;@aDonGz5sAc)Ybw8Y)n(D zLq!hYcl=ifA)Ds`z*=4a5Z-w>kTL4uuAhHkQM>0fLdn zfOj-hxKSIO$-dKfn2h|@VJ`orIsN@s>A|sHKbHu0KijTS)Tf}4I%`b`s&Qd zim(UrX^A|3Z>s4gTA6mc+_j`2fcX7+PIIT7vOt+Pm}dL8vpr@ZMAN?x}$VYM`gIq{@rng_>{`}zJGc~W`}Pc zpusS;#+q2xvL18(vnnSO>pU^k2K_Y}_r)oI^__D!uK)Vo2y?v(S*iX z@MRl^Y#z4VZy?JacgMOy!hzNree{5wj~+=-ODLk!Wozq1?ozDB+3}9+rG6xh8LF^(zj+6Ua zk^7>{^p_Ik7oJoP+-l)YkMa6lh}YhJ5~Ojc1QnHm!X^)Kd%qu40ZMDtir!#`Bo52t zX(Or>%|B5{v(kE&ZX;vmKC7jUuF48a7-co=L&Y9+ZFLe1_4lfCJw^eMRgzE1!GkJ+O9+etR%O2CZY4~cTOoY0Gyw~mGSet~$sB_IcQu)! zw?=*so$gi`kFLCMfW%aq5a{-LtM8rg#sun>tl?!AceCaO&vO=EZkLez19Q!l7`H1r zm(6c_wWQt52wubo9VM{9%En+VZX}hpWsZv^b9QJ2=rDb+C^B5@_>>Wu#y>h6sk7CvpvAp-GxZO~Vg~+w34TjAs|zX& zhHJQvvby78Vyc~LWdvE0lf^A;GWBpDii+A=58US@!BSVmf!MlUm$bh%;Fb<1Bhhmv z946PFmp|SlWkx^F3-CVyO+JuPV(5rlz1lA`paUkQXMQ4^rtc+sV1rwkrkSn-M(RB~ zD^bRjIkn@gI@do7wE$LQM|)D}c166d^4QRAG45?FVQeTj|pL=agKzw&6gllrpF9VNN(~zk=!6`6l6RkHgldLp-xq zlEn(uspFoU#F0A7IrAs%l$|nya{c{->Kl{fFQgB8McJ&QOH#gzt2u_u$>_Cdjz7D?MKjnVX>sP5q$U7;7a!3< z5ADa}NxJG-b*LiMXJJ^G=ssR=7aY@GKyk5~-q2JQ9jl827XdnSoKNB%sB`5>-_J6^ zk_1FTbPZI~mQ*V>LazmED??gOog{!OEG$?JdmVgOe!_f~T-}yc{()Ikkv}m05Zvxl zqmCs=B>6;a`>Bq_;lb`rWA&dmB-YLhgT1?Js7qumY-=R=?i=m74ZlD^9Z1)~fDrJP z=N&cDqXHNxIe-a=L_Y=S!H7O#R7`eLW5UGQ`fe$!Xr8!BD z=py?{B_NRnvaY&KNEz(Lgo}HCV88QuC;)3~y@_3)j!arWDS&^@wZy`X15^)i>zd|K%n3MOzpf8&1THx(_7J2YLsBvr3-P; zJ4nbGU%eSCssq^o?x#Hel?fQvi4})uD*KmvM5m?q+p<8>Ta; zqx9-Fl?VJjNAIBI0bV#!ewm_2qM8T6Uk>Op!~Sg7k82&zzNFrQcc$JTZnm&>E|n|D zfbYb@qeJj+ZglcEr1eRS1>u*_cRt_NB(tBI2H zYrm7IeAkySgIX)L035KP#IVwLugo;#%xd4|Rd<&RE%)ld8;-%ApV4CeT6le;_;y~_u8x-XX-aQu^Xzv9g{`K?X*x_@ zZHF3)vp)^`Ts`EPc6)ns1zyIHlljG<0J&7k^xHhA{ebnPSd1tOso&%#k}*fg@tg2WenVHgPlyWBNB9p)*n zhMX4UGc0hTOBB z&0wc6Po$Qjr9m(vp#zYpVLdcdV@WYmNMnwfD3oIX{9)}1nR?DH%-9WS(f*9vBQejLTgcK|?T`pi{R^dK5&U233I_Y2P36FbE4Vj6uB0Lq$6) zR_|4MX*jk~-GjS2k9s8FYbwB?RcOC>#g0^_Bl2DILRjP3_4-ikhdHMy;~BR7GHg6g zB=PNt#WuEJ)+l*vc1VFHP!9wDT^^$5af)jb83rRMnJvI?v)j=ytyw(jJGrWg#48c` zW<&%RZjq0ea+wAb{}PkOrLQ|URh)(r zOW8(Q3KTJLU`@`3m8Km&kzA__GAW9a*R%PB@`w7(nIp-V;nTGnqKI*b*I40r3-JDv zRamO=|0kD5cBarD14wG+eS^0S_0U5d_G0-pkzU^Stg^NIKDL2rqjNT+Q)j*N+JJkc2?#*Rxstm0RM@1rht&cn2K_yE2V+R{Tk9uz#=o(4C=l|uGa8B; zu_0^W=dn7jx1lQ$SNtpqhm{XWCW7=Sed&SBB7ywJ#uYw+6N)5$s+|Aq>_|fcWN=k; z$CW+2DQ1_CVtPzCcZtFJ@wgtC7QS7jf~s7PPPP0H6u2n(G%kQQu>l zSZ*v;lqsg{XAukA4>Ly8s@+5Gt@G5}K(2h@U&wL#-Vf}#Ilf;!J0yo@hR#~EZSS5q_?W3o5h8;w#L2# zBxXcS5XlvUfqesGxD%vEtA<2o$Q#~uX>L8a70&6?fCYW83k2P$cQuP6P?k8UBV9ce zD4@UVe&c`C^P1_cT*9^VR`m0W@@_h?pz!J+em|V#&-I^^D7`cW>lZ;D{IV6lp^;=tsZ9jl-&VH#xgbEHK zSS8%wql<`h*tDrC7Da$5X~b+mqH`35BD8QuzsE}H_VbTfn$`cA7WkEjTO*URuZ!g)R<}Jt zTIYD+qQ!cXSx+pfTYcSyEWoaK)Z>yCQ6(IG4r{gqxy1tVGpfI(3cE{ao4{3=UU`Ff^rC?30rbqpNj;lTPdDK zKRflfLn0~q99!?*a=s_&#y-5sQ>HOJ*vLO5g(#vdD$e4OCPu(y%t4wPc@d-Ib&lxM z#XpK^pRq=)e{R&p$oy?-55Y>=ts&r2o6p5l487b#U)-Dq0h^h{K!k1nA))#lsGFgp zo__+9gwTEcE8F4U4bc*#$QI7BTwi%uQ9~e*zeJM6hR=uNmhuP+y-4e_j7!1=q#m3u zP3NTQChaV!I>{7&HVvvxv=47kttNwsSWv}j4CPo@l&pACul!5uP?c5Wn18V4B|2hE_UIu!Y@ z&a~LPA|2SFz-^W`x~#4VoMI8mIT_h7gdq^9dZZ1P7-2q}$=Z?G(V)2GxgME|nk=+u zG46Pi>GZM+FO20*vo#)+jiRw=2Ao+MrtFa(`Y;LZH~IW6x}AtbOwm+>&;x-eHf%G> zB17n<-_bSOTVV&PHq{4~jr=Mx(OEAXC8H>mq#v>ZOWu85tj_Z4S}}NR8j0^q;D77- z%w$-b1B)5ke2#x^ut19+W|?zV|z>8JM`GAO*ce z(>80@{z>6~hU3I&0-Ur~P%cDVzi;(zH+X2CE}(gC6vwFEs0VD|tl<1yKnt4yEsD5S zC}grzXMTVhkUuu0Y3pflX_uoqMS|;DOLT%Bo`6N+*0ZbOmE`ju{HC_VE7rmj=_0Ct zep}>=ba`D6mHdu79VZ$efH38=S4d3xkvAlR3 za{*HYD~fXZH$&_~KFb8@{xc9&^I%0cp)4;K>t}Y%%or{GE>D{;8!v~V4U5DIg!@?5 zZ(7@bj6M2*GQ;`dEkU`{ zS`XeFx~7pjQLz)L{=BRfj3NA_pM&b+$vTbmUExt1Y6%L692sCgss7Hk38>PKt#Tm( zk+TOaF+yOSyj49ymeoI$kx8{JWvvncozk#`P1HY)jPm?KOB~T;{xMCdauX=Tk849w z-}LKKE2e~`D??dY_^-6cD=1cGpa9C}{E(WQdF_8P0Dw+@$-D*~)uoA7kbCtVs?8u{ zqH1kaZbzJTW_ai$I}UI*M;$UD%!bnqO3^xC`ep~v8fd4QZNjZTq}k)kF(I!O^pMe2 zs4oyfg~Ys+Sq|rWPoFZwYze4!PA#NgR0Cn^thtG=C(r4a=#lB#Z;iEL%|PVT`hR;; zksjDu$|n~aQ7cX5HvgRxe+<+$k>M=u{1`yHoPm%SB~v%!%8>le8UHsfTnN&Wkai%a z-O7{xCrcJ&k>S0&plwdV=Uab4Qr*?|3jp9tiS0HU1ArO>Q3(@*L8lUi0OF=@Qu6j? zY=zi#<^XZ1yx)U(<4j=(OiQ>h|FiLo8l@UVC5%T&#|x+($`U(N_3<+Gn+hvm1ovLl zS$0^iYnyP=*vcKpK2AU5l=J~N0qmCS;ka=A0TU0k9n^U4nr2KyeD6^AIdjz>IJijh zC+Up2K(wy=54ipQKWf1=u3Gjnj=Od3EoBv}(Xl`CuX%xpYxSdZCkRz|A?f}MKo~&= z2=)&?UPqGWXO45NSS0eV`Rczm7J*y2r07>Op7IBv8OCaIkc}xyF?m3PYea(m)@s*` zJTX&3Yri=Gw)1Kt6w#3LG71f%R-wK=R{q(nI_zkV_f0aph$s2=7u$)Jj~7gux7}4J zwsxVY?o|;db+-&fkq|3&#!3=Gvpt!HSoMUZ0*4Ad(}uVHQBj4_Yp6PX)X4@kBT_lH z$vG&>BkvF)h|9eh;9d1_c%C8J7iIdeh_yd;>?Pv*R-Y2)slv)=Z<@jaqCU?~WipJ~ zZ5L%1Q&_B@UwPY@;a7*!xG11Ix@QU3Kl42ZleW@EegAjuEF5OHKl!+1BIdHZgA*&9 zN<%wwTsbfPdy{%x9D#{7ZkNs6gUi#O(mq-7G(H%A5h!iO%8 z5#sPFBlZQ`5<*?D3KG4vl>vjj!`GE=8Gxl8%A4LQS_qR=Iq`V()LC&iI@B? z8~IcxW>~&1HX}oHf~8eH4X19AF&<4fr>w)sR>1LpT_$boEZRAL52)LWdf~U04?0vO zak@>DeL@r8ZK9HmZ8hb?+oH*3cHfI2z<3phvfrP5N$Xyp&udiQEvblA%@7(m9JZx3 zzKYy3Iuum+{+pvk-d@9MOZwp#deH+c74cEzqm$BN2t38$07E=JOG`(&a|e;kl$JVC zRuRcbW&nbZC-#&SAgZb97=Z2y|8M=?yeV$p;zu$bT^C%D^ssk_h4$Le*Kofx5F;Zz z95@XFkYx+TLN8}dQ{3zWjcj_DVny%wnQiQi(F+CM#;kr2?TFd?+T-ZwIgVnk7LD*W z+aMb`eUFcePN9?aG(8CajAl=`Q8EYM7{5HDl(P+sipRB#lflI@!R#rDnD7bIav)i- zDaFeMJuG8nmjTTFX?{21i-UIRVXcc%3Yk|oDVhM_5aU|SH+{GiNfcLLPcHFAy=8Ko zs6%=8Wv6}_l7aCI_}O6?01`vmW?j)Vx$~fWpPFBRc(O(+oy9YrIU{n2Xje>0kX<+v zzY$-Y1IbiiLR+xbaP`eu(<4gpg#$~k18SQIIjvXvg(};1^T%NGoJ!*ohX+8Q3J7A~ zCtdalNsA%!(g?MB{+{_hLETqLpBCSmCqHxzIDih4GLiK&;dYVA;1|A;($!9XFrP{< z6X#90s<<$dUst(*Uw(;H!R+M3@Yj*ERt12^q)$N%spJ^RaBBprU7Lk$gcspyFIfZd zEpIstiAlbyZo+v^Yh!__2ugx>twy8}`z3fClTPmisYk_lV>XIcxn!jdlmD;p}OjwO{Rv#!v`;3lXGKZIaFH*$pgrY>_2^C ze(c=F$&@t9=_Spd$+AJ;=*`z1FFrs|M`|j7(9@#Eb;I{iX>2q(B^Uiir^(0=Bs$;= z7T0)>WQ?*heqGvNg{amEKvUIRqaE5}NsjhUZhGSQQTq72Dld{TyqO;)7>ZV9cey~r z&~X()ay2T?&i0I@sU%?_s>7MKHQ*MV9!1MyY?wax?o1~Q5AxJ4qVsF!-lmym)Mp8E z#{P&5Ig1;gF*2wVl4m#kreiI=fFTbgAuHvqFs$P4&_H?tfbX4Nu4gEy8O_ z5M;-v-Mm=cqA0te<9C#QT%+X+TWDoITkLcWX7A9Ay-gZv&)?&5lD0-TMJu%? zP`aG@gPCr?uLqHPz~5rXwq?;#$p}bwr4nM*fRt%g9`GuBjr=^=Mw6Logm^`2MdIhL zMf>$DYvHkOui}2nSHuOgP#``~R3@vcaQSB+5SxL3RF}_1+^;mhE%yvt+tT--OG#!_ z5_j^c^RpiF-&vG207Ui)OZItHZsE3EcDg`{+$wI+8QKqPH|! zw~1!(NAlhkEHO5xXPXM~(j9`_A*HC|ouj&=LU%8MU$VXAj#4IQ~sf%<%^wA4rO$*b@#14_gdwA|!GvYs$ zcH(O#)OkJoj23I(TpG6xFFrq#pw+Ml%y3d$=uc;!{TN;mVeYkq&7eynDEid&O)!}! z408i#DQE&iI*E148Lfr5A~XsAPSW{jMu#t$ISv1Y1o!)$sUbwM|KA678u8#fH20Bw?Kb5B%_}~`2nzES9W9p zMqA)lx^p&8l}FoDC66$_gVI4Sxzvjy{|$<|DIGUIaN7qz!0cL{AlUv>HCsCsy4^Gw zr&)F#d}fvI7k39y*T0@BOM%w_1wbs8OJ$rc?Dw3BRe`<^^&^)_+?TO|&#xvHET zL*vxP)5=Li+>PZnr%44|kdVZhIt6cqR0GNN*~CT$R?56#%whSBcbt~?wd@@z=T$=K zR+u-dGa!80!;T@(TiVfhw5hEq2 z(Il?)M*<};P=B05i_3T;S=92mX8_g8Q>KmGdK%%vtGJnYal6{zSc>Z+wAVI-)IQU4 z$uJU+mxEHqgAc0BCS#{T>RBUNIk8R{EDntuOqQ)u;IXa2vVUCpe?=Ll|I3q6{;v=v zpqDeWRC2a~qL(FLWMKGTQjDXMGXV$le@|%>TQg^K0v2`##{bR8{C_wyH`-Tr*b@nV zwS9*EUx@_s&tER;Aoos2!>-=1y8AH(I24v7Or;#$j>ls^FK<pL43vC;Z_rwP}L!zwh|anL6i|GN5`o&yuq{pS=FztfG*Ic1B>@bUS)v>rj!|L z;*rcDBn$#UG!3~2iV3g|RpK?9k-n%Iw-ccBylZ(JEP*y`W^8o|7Uo{1cBVN@KfF#h%r{)6@nio@5o&W2aQP?7# zlSRMm3t4=eitXpIO;D7@1&Ai&vmbChabQLfLzNKPnbJ2vC1|H3eg@ovKDq;jlLu&a z5uMiYPmSbJ)QpDf-QK9BGp-i5!PZ5EUdXU-cr5#=^L z)~35uQXuYxyKPi&UL3cp^;x$cDmn9F@!IsN;E`~34Um(5-2)G#qJu(=o5h|yqBMB> z%`JGw7zjA*U|dI{K)_ZBkpGE+0%EZCY~aMyM6u8*|y|vK1bC zbM>N_HNL=jS98_5?ZggentZw;G$GCM#hI=WXMBwwpJmCj^UiJ_tN6J1^0dw!bM;AM zT%`q2H?0X?p|!p4aea-ng%m7-*$xWO%V1sE)HN^u26p>w(rlQ#UR|D9GFlA*TNf%d zH>0GpBqbX*)?~x~x?wUCF8%$g8Y)4)1wB_%alD8d2!|eYZn}#@^?WP$N#6X8by|Neg8}47Uowa0IA;ZFqsu z1kbC0ppUZe4*9@AEG(dS%rY#e+Yceu*cJ0`D7KMkD2@>kH+hXKHzyyx#kv9znqNFk zeuD(sh=Jy2l-k@TD1O!{SPIqf$@_N+35O|YTI@56($FuzQ3sFfg+r26Zl^@{CNIQb zss({|}s1v--d)oAk>_Dzw8& zORfl%0+DD08_{o<-(Fr7_%U1y3{;8NhaaG4e)o&LXe&Df*FR{-H!lb}szHq9IRQP> z++zL%e2j@XfmR1U1HzKUi!RsAQci5?Ax~r?IJ75rGlZssIhTaH)&K@X+{JljaO*J* zm~BQRhuLocdb`HPU^bWHgRk5J9qy@)Hk$wJQQIdcW??-%#SV-jXvNa1hPB8;>cVym znrnyp5lk&z#BS5}hxucmFhw7OigK{li+a*hMX8QNkC{+t=}ct^!d7T z6W2G7rmXm*gVP3Ky4^-pRG5TAKW&?v!pgkjon&+43;Rs7BA!QGeqkloJYb}l*!kgh zJD6VHuRa7rZAinq0-s+w>GqPgky#*N&;A0v(^1h{%@|T znCoOodwukh;^j{!lAQSV*}L;f)s3Cp4w1zg4uHFveBO5n#!zX#(I1cyh{rqC6|Rz9 zfUvPbtK736ffXdY@iBRBxmLgiI7+OsU4Hcj7b-hXAFD6kF2=G#d&9dVmI*K`4xWQuKabN{W>?&@E)pdC-uetl3xn&BDPj~v% zzb}eO?#7Ww%v$3-5{d`TA~&6!$_@|~iumPn$C~BodNU=J5$ddyvr2VTCE4aPwWHp< z1S5s-r}I)g{*(kha%lBbr7GBvf&dXsR zQW+bTz;4-=$gyJx2{+Xj^WH4Br>Dd(v+xEC%G+uXUU{^Ay2{saF)j$tMcqAF&VM!y z*9RoHnBraUuK^{R(ZGfr|9J1&W>+5S978Co9iEq3xN#LmXFP_fuyb|pR_n{~B zHatm?2;@7Aja)e_1h3R?91mB04~XyqxY!GjuqW%UGl)xz<9^&cXn9|59J@aVD9=Cl zwm6TQTEVYl%7b_IbWZWHPR>WlrnKlVc8fWc(eLzs|70J;jL~s+?C~LJfG5R+FSBYj zPvu8|H)P8f%6=mFROE-?cfGO>yn%aaj8Z(Ip5Z(=2IaxK`8!wHEzYa;oR^@>NZDbI zcH?>Oypo_>EyigZnZaQHk$IUwSWnsU#EPl%u`o+3-&2c>Uv(N(Kn)Towh#mc*F0{C zx*2&PlrR573*ULU7gR!r-^!0!)BSC1b>^xUPjUg%%RFYjk?IShOn?>+a0q)o$T)A)cYcS%ySe6lzIUz&UVnR?Q-o zkdk&!gLTG9n|8y7WkxTP<75Lt)&E4BkB0}BSwv-QF%xr4?~#k+FjAJUNKRhb}d-! z9?Zj-1up1JO0to-&<%yuAp~+<)?yIamZYJJ$QmDali$FzWyZe(cgj#2l`U%-J`{B*vUqAK{h+nU@ro($(=TN@nVjjw$>FkHZ0th5e3_V)B<5v`SAAt}}BU z1wR>75D~hOMjVMuw6Wa%ih-2rI#o+!xGof;^*uuIq!}c{2)?4BQoYEZP~C(A%yW%W zlU>7lljw!Pf@a-^-4iX}j`PV~_HO6oxR@!+6Uk<2^|h(e@i6sqgK>t?b8#=ZAOD@G zqDinh5reqo0}k`1^?(Ffb;oc~1JpGYhU)lX+e?J-2jPeKPFQYzMKJc)mrEN@u(8Zr zzD)2zYz5X~%Ms)6aRxNzDAEbz?J));hRUlJC>(@zHE7BPvWCJ19DX2jk_<-FOM62z zTZ>WLLF<72!8;<%*oCS$g~UbJB!dC#2$%<00ueq$yPZx8*1U48pyfA_EZ#}s$Ms5a%_!)aQ9#(NXZetWuJm2W=l z_EOtdtzcYf8)%Xzw9bQ{ zw%X`+ zG++m{jLRCuD1fASJ)6dIuuQSKjxbpBkllNG2!qKn0}BgppAF50Ir$$NK&0otGXG`b zv>NMOKjI`g`It`ns(mGOp4>lAWd_{@g_?R|#&K+2Dt#wdvyYkd5>QRO6sYzvot0}> zpUW!L=jq18eb3Rl4TS^q3glDVwT0@LiC>+~#O3rO&5q1)+|}nGZA~f+Ig_pjwu@8@ z4n#>sW^>Ra;p{vn0$EQk>>D^n!ewq);T%y4Ew*UCnzxF}a3?1^EvB>L=*Dd|3Ba^1 z`rFN4Y@Xx$EV5$cE*7hw^13cou(WWS=8iA(!ju(&jQ0&k?O1@sd0)8UKM8BWL8f49 zjeGo$=oV{e7VyfXg^wN{3yrIVBv^W`{Qdj;Bk7j}4bq)VRSm0yCRcxKbkJAJfD@V+ zlYoJOcCUwp@RGQw=OM7hAP>g%UI`D;`9#@^7P%?e90oK~FhN@2df8h&Wk)cU&k<&C z_0zv*pt^jjT4n7)-1v0Bg-`c&HR+lRB38|SfB0nRXMp_5|cms3fYvsdU7nm8IXG6nzZX=dKxpnFa@ygt(Bh%?{f(NcE zBTjW1Q#_Wsx$b`p{K$hDn$(81E-)DuRt!GReJ6>JYd9Anm$!TU(CAHsWm6s+z?SyjUi|69L0S~(eb?^s7jw)e)Ojo|E(Od)o9|{eKJK_oH6n?^ z`oobLe|kP3dz0&F184KtER1jsgeF*lH%n1=KvG&V1sn#p|auf-C289Pvn6q z`DUe&cs*xgjpuLkUyduvX+Juv|H2$It8zo1Oc42CPJhItJ7&JJG2~JYFDs)o9#xzd z*F7FT|3GywOZ}?!-LbR(hBur2-hUkH5kA%x(?iBfNP5Th9i+I>xBMA5aTh(xNh zOj>x+{bGFPp=%{+#%QQ}-&~&GfDa`0c{t%H(+* z#zjXDg+YKq;GKl+d1PVw-q5rFga^MH0C=7T;n<5|5Dp0(i>U8e133f$kpF9?fTDnF z2r(0+qawu$-bm<;$QL3w$jZ`i695S0w;3#eG&RiD*3JT;i-FB=E)jHILd+)=0z_aj z5~h#O=aohPsS5G#afhE^5C45-+bz@TKg)b~Fg!3%=v9pdg9M%DH-CJzS~O@{1_{?^ zBZgpD$C01XC8)u~&Zt+An8~yrS9Z%n-ta<*1uDD{a3_IEzehVIUgt;$CNmuY0PG&} zkqX6FVMvW?K+_+3Sg%vaMo6Py_&xpS-0n$Gob+-qtRnu>XnnM*{?c(1REm z{(|`G8bDY9f?>lI)WXNbpZfD6gMbV+WjalfXg9Z7GRHVns8k#kO9vdvLF5m}#P~78 z5ZDqv^K1IxyFRHpw$HKHmN&z1`Dq0ZGkFsJ^Bb?H-7UcAg=>~JTkPQ5zXqh`iZW(z zK>Sa6FEzf%K)`0hBcTic&NEG?@eC1<3VLDl=B^ze!}TFX%lP)swG@X5<_Q2`%!US( zG4wKsR}!AUo^5V=aete3p5$&?jO4=18-QU$0m4L(0mRIfMlgVx$6+8v2*3d10)Sz# zE{gc`;RPbHfb2mgaQH($G9CzFBLaTw^JyX@?EC*ANy8-?F(XXu%})pV4aNlg;KAZr z>IFlD+Wcvbq-X8`D4=_Sll{j8xb31^GqehCjQCN2Rn*n*?MG&8ZVD=IQujj1fKgSx^t`JL zp$WI2sO$$}a=lm~ZBigZ=}d35%uZa-N|Lq5$XEBjSylhgo2>hAJI9|ThPi(-kZ$JE z({b`3s%VE?CcSCMLA(RqO2&E1K2PXuDKtrV^AAr?mlPMYagH2(6#g|VZy>~uuIaAKr?NE}z1+k*`?)3%@ApD69_=EQ z-#UFjYq(tv@a&c`!-9;`uyXG7<=uy1bN4HxPEX_~2xF~?%tery+s*=IKW`FVWVoMX z)Lm+{#mxd2i2zvIYwf>8cxn-HBf43#%Pn7_GFFS<=-J&*X9+5=xtt{_g*c{v zVlIC`>^d*{9|0>KzriR@SMA!?$y{!2KaOoKxxHptUbNPMU5m$sb4;^+P0+!${-k7F zSIacp_&HDfC?KQc5Q^jSwzvm{<8YpiM4i=bfP$9fXX7rk^(`e8+5!x~QnjC!g@q zKc2w>mr7j|+qvqjV|+Q>-V^%kHM86L`C{%kG;{sg?(q6iZyc08vlc0nnZG=Re;Hr+cM;Usk_ph zU2L=t?&6j`?qwd(&e_^&I1r|G&|EY;Z7I*$K^gMi)0{P+MZn*c*7vP^{`73v~g+77Im-#$=cEk*PC+QvF+lnJtf<< zb4KSjlj85=Bo#Y4YDpMhi)md($|)U=_Nv%5ma2OssjY)oI#X%B7JS7l_uQX>ND}4T z>$Y}p3+EY1)~g?vzeDYVz&5<(M;Y0LO)VtK8}!~)v5;NUhY~Z1ZS$T}C7xP5svJMV zt~iKrFYH~=4_r<%-ReeiYiDIuCu1Q}GP^0&oflX^#@mK=^p}VK8bYZLSc7ryJNv1W z;Bh?*eN45D4uX9%$`#f0g6DA)%UTCB6x5_+9W{&XQy<}MqVeyul|<JuCn%JJTN+Sl*CZBma0BB;-f9GbcjiGN@cz0Y3z5H zn}2#)uLo2WC%S;4uG51bOhMQgPw*l5k$)L*l4v3R+rzgG8f;^j)wFKiL1locdCA0H zslL*#m>{hs-j9oVsFiXGs+;i0wmI(oyj&oso04q8j;w^C@3LD)Tu%_ewnpKlkB-@l z!)8|<@vl~XUAyd`Rb1Y^bdr0$G)AEZjpl-RuRGM**wq}BGAtnTTD(V`R+OmDulQ(} zId3N^^&*}zL*c~}-X?qanztq&K0%JUD&~ofeNv~%(T~(MhtB`JNJ@Ps(s#w4m1cBt zDcdNk*EunHigvZ!Pq*K~u!IHZCLB;>yVagIHmA_<=2T1{BS|8KX&q3Hv0KWfrl{!3 zF34{zYQKFvZVOK8KpstrBih@2tfUNQ%>KI6tT6TBHFm-LsHHj;*0bAHE(C^Mc>axw z_vn0(h?wOou1=Ro@Nfxu(YFEEk8{}}&6M+;Qz~?7=~@5om-7;KQyYw$u3%rvP-G$# zwySlf>T<(+ypRao?tezRRJt3dB!}o^zK|m~qqfv(f~8%wOZx-G#dm8;TCZf_@^fF$ z+aqs%27byt{?gFQP6lNscL~BKo^z{>d09eXWmcFKMK};Z?!udDrry?l?A`Zby*4$l zGNZN}Eq@~mf03+lqT5{brgHCm>3Ld}y2Vd>MHN*aIbR zR49xEQY0F$Oj0D=fn6*tq!Vubi`qtz36N|gq{g`IzB%5*7^k2i#Hhf!1n4ODi<=UoMS0A>gsq@9(8F!pP&0C3K$+72(D zw78-KB=B?#tc(}*MCbQRU{B+h`3YddTlL!j&vwV8BP}T>J1Zd`k^AZh3*hTd4H6Nk z)AEhZ4oAb>x4TqVE>ALWuLcZ|N{e+((m#25ast-mAebKjQ3N|TzK_n&tAzoz7b>s; zTqj2|1o&XZKxc1t#WCL!zY9v|2A&7m7}%!}KS^RTWLHY;nzTkzh5L=pcttRm%8Kh7 z1}u*-#=o%^sL<=k>f+1-7^{zg0~-*uuev6{f^i)Pr2f@l(2HO-vkO890E$lX&=9ub zr!&^O@s--a1Jucc3?7ZXcD>6D7TTGY^TU|wN4E(DDBdq`zH}Rj0Dy7=&~6;pvBiw> zl`11Gt0kiWSu|Q0v5c1`4jqsUhZF89{u|j-MP6D)35bA*4s>VKo64mUs6|bHgIiyv zP?UP86l`!BEJncV{u|bJy$`|8apLEZ&(OyD;n@>858&HP)8$&ExGjXVGe0Lq<5pvyo3@<}k9*H?o8fOQUZKk#+_ zMox5$4yry_HK#sr1P^y!dtqr$0Yvb#8Ex{Qx9WG)M+z{MLvDMO-z?UL%LA~2KtWl{ z1{8w|zWSTX{93C!K?eu7p)U@x`pdiKCmw`r0EhQ0{?&T`Ly+3l-h?P2UN_k5&5*=2 zg=<(1EUdNugZE(fd8_DXj|~ptAEpN}MovL=xB2+?(8Q^464RZbpPIT^taqfxS7lgz zwU2-TMKJU_3ED4vqIXrl=`fDlS6N8M;JZ!`4IwEVY$=Rqrd2(5fTE_qQ1y~+w7kng5L@K z$nKA-hH>QU!$ENT;tQ%%Kf#C78hwJJfZYxJ_!9x>cJdRXB!1%~bYg_@f9MF+6&63^ zC%wG`blve2w6rhoO82R{1riuQT=EmudB^(+(BLoa(t_0KqD-B{i9#Dt=Hv90@vrp7 z(f;eUgVLvL`vt5Qes}}@El1CuynI@{?t5+Z|9Q-L4-qsc8Nf3HXdLM=`Gx3@go1SP z8wlgw_)_P7dEE6Y`8DtRd98jDZ|UM0z#Sx}0N6uBIO*JiW%Ms&$X(@Us`k};>!m}j zn)!{Qx6=XuIe@Dt%El2^mjv4&Rps-#tWJ7n%#GKW)MC?+(C)K46)o_vO(_s5J}mhYG7`ko6VP zGSqVuB2`YRass`$9x^I%YN1p4icNW}rPq@xA9jbS*AeL{Ocawek}WFHc&yp?mho^O z5-e%N@LxCi%l?nG43a$HWCy}9?trfQ&>E0#%LWkmocW1C3|fP8IqgfZGZ&G&t64c0 zjSAbi2NM0uZHe04iAh+jMFLLod%B;Z!R!2r92`xclEq9W{GmC}3 z0o-7(*4+-gX(r;BYhw$wHgaq4JhKwfYrt7|m8p;9IfK0T4K1vemfsX-Te{a2^D+x* zy=nP{9%{@8c?QZh&M1>{qa;mpciqg2{0{v^hFPaI?RHx+2p_P}Y~{u1RSLO99XohP=W$kJ*YhSOvb}VNHq6s0YtX2R-_-;vG!j^_s7d-eOFZN&A{(I{Ui37pl z`AGo9OBhjBz`QY*Bz^U5iX#nS^3uZtKs%XgJH0+mtZwF4n3<8527hh~p)cukSxDRd zOndPoM0+(4XM!$>l{!}euXk>2pRBIJ0z*)iQ4I_rqw?sH&F&%ACRE9yxhsH>#u(il zUXbWQlO&h79gP9x(|V}nrZ%xCE;x(UI#(b+rOxZpZ1VmfiK0+(G|ORheh$e!?V zRWa+W>QmdL0blxhdxDBj$>*V-Bt1e)1}JJM0j5X8$>78-cM6WMj=l&Zu3 ziEmV;AmriWnJm{-iJg3jE$pZ-Ejkfgx`tA{ZEZW|Y0z;MO?V$P88`bYXj2mQ>GsH> z3@q8Qr(}>{|H0xqH6E<;SiY(E{sw;ZJe%tih7q28z0g}t8M&ii_xi1H>ub<8BZQ$i zL5avL|AfhLOM~SYIT}6+-4SF?SlcD1Zt@9C%)Z^)f+F*v+NTsuA@@2q&yn@owO=Ws zhg&giR}^jADIl%Nku!*sYu=#tw{SJfe(>vC=F|BGq7o>!8vkhLHQE@h^Fl)NY$&{g z1J{N;^G;6u)Y_Uko;Z-$k;Bk|DIpYm!7f~lRp1}axVW#5-B*l_ToxF~vp@OIQU`YvRKZNw|tY~OmA6pAL;RK-yB!IA@p699J``7<1c15Z zfJRV8yU)^-wJLX8tCd(p-H4%n$DJg<%`07}j&+*$x%1T7j)FYleYz5O>27Q`0w^oe z?3EQ+Oq{H@qfMp7wurS=Hc{!>*QrQ`Odq^w`})@NuM^VXO7ktfg&WX38> z;}lYnVT{!vPG%Vz)=z6hCZ;`*(*IbT%?<*6H?_yK5RPh+aps`wy(#(&nnLWGqZ0`) zV$E8Bt>w|!-8)6vtZv<|bwh3H)*D=Yig66UWDR~C_YRj1wIy2RK|OVtVhcvE0D!hQ zhBMN~W0qo?m(4ZAl-)99VF#Pz09VwH|EJjMFBjZHVNF7~^t%9Sk_=vj#6}k;OGO;} zWBsv-^(D{3@S+8Ow%fk(23%TELkm~qYst_<@V8k{j8_+`f~D`cVkS+9+rzfYC!7}G zVqAe&$ORe2sd@bss-69>QpaVZoH%u{-j9Ie25N$(y&@Qh44vk0BnB+WHtzi$o<&o^EI;{`$Z)>3~|otjF7 zV=alYc(&Q;Wc#WboDZpxQK9aFG&{7n(hbk`Cr*u;KDV(LlT;~89OQwgxN+gtDf86W zJ}X*0W#gWxBYMh=#OCw3f7$_ZWT>Rz;eY zH84TfO^9W`I$lPzw0q!V!GI18*U=r-cs)pZ{Rp|a*S34t`m>KZW$))c^1l$Io-@V6 zDKl2Cr_EI~Kc)JU?x`7j(tRAUDXx^JMrz~qc?Y~K|`WKroP>m3L<^viBr9QZ2%@wzDsDbE3>}7 zdkpBqLQSZiLqb7HIH%U4u@2d>m zN*f_iMpTO}W9(Ugr1B6WyerC;N&ZAVR*B&VpVI_6+EcER74wgLWHt5C4nWUXy()(#+HR!2^YuAO96*-$(6wb zTcu93G~=EPGZnwN$cehS;-_?DIphXVToYT|GZ6XM2`s44Rp=GzxuaMvpU1SRj9+%k zCL8o}yJ6_Z^$l?WQzKT^R;Uc=4=PzWD0^4YfhV`_CuK<^+e=fa>f^Ys+&puE zF$tG6%2{JIF=Tk98@UpWB41A$C|WUf{hZm9K`KvF1zJy(|zYJ*3ssS zIj*2?GwPZ?c9afJ$sDriJ(=Hj-du5&f5j5JZvuBnmbIgW?xPtes2Vxcb_LG4+FV?fg0_g3*+jO#^>~zmIV7!?H#pL+ z4iXo)?A+3TxNmZtiL~(nA^E0CNBS5GzW7tX1=5M?{>)^H;8Jn#nTr++M60%{8O63P z*N5LTl-n674xd{;F?gJ$Xwmp$W4F~{6EiwK%k5vigmDz!IJ^j051F5EC%{8eudB|1R5b!ERa%4w9rXYFq2Vkjb@LbAr zp5;X*0KVhvOkY>ylG&@+dAl=w6>}tW$Kphl%hIxJhK>Qu{OvV zqZRYATNjxPsMfblwNsbnGp#*#TB=^f{B@e!lP4_}IrDPt-0q(Xo6+FFm+i*czeskDNh^-}#Qt>G4wf z$AzZ;!JW}$-f@CT_)n#UbKMw1=tYm+6h*wLs7@bhaZq}EeDJAWNSr<06;8hht|Gah z!I$mbl_p7owqs&j84YEE>TBD`yi!Q?d2Zeu+KNlaUpz>y*iQTo{Ksda=Y2Rvm4}4HHrcYu50L^ zQ!1jA!(IQxvEAp1tPu8<`Z|2T-@Z^3f>^&JyY2ItK((8!3^N}z5>~NoKd*9!Mw{S~ zAorNi9ar?Q7XLngDm1C%+jZot*b4*SzG>kGxd4?@7rW(`zYWTXN;r>KkfX-p(|Bc!*2Mqq?f5G^ zo8Cq$!GK1PCH;Eifi2vI1=^#}x$5131kysGiFLZWxP#MV8ob&5mGV+USTiX`A{^;> zcA49?HAt%Vsya|Pe!a@M7pF>Ng8IG z@mFvhuXLnR_Pk)sXEcUM*XHf+`)x~H^V{}^YW>tp@)oaUUR31W-MW$Qix2@qi53c~ zHEfxi*-rya+u>sFYb>ML({=YX1Y>`k*Fb@?2BisB;Z~uAM_z2yqahF{a0_`k*Y{X8 z-c2HbXHKBKz@`NM?A8-hZlO5rFI9hBY=YKX^9F9oK=(%k%N*oP=pe;ksz%}~?5g;x->k>?j7m=69yLIUCn79e@=T(g`mG%H1;Gb>Jg zXlfJpkNb7uAG<1#!C5%}AaFSN^>@nHylBY6Ku|%?sK3G}`R*+l`;0(&6-~q|n#uIF z>!)8Yw4Q!_lM>u*bZ51SN-vG8i;J zAv**(XJ`5_aN#5nED~%eOqWr=zu#v867BdZ&QR1r{$dFOWnVaNH7}@@`@U#QVn)FD zN?Jb{GDO^($+;+Rd0`GF4LBfH1Xb?R-1eO9tZ93W@QKx6aJqD|%UoTulBy=ZB7cV% zw6^hLPmiBE5jW!acYTx3Q9YG$-@UX{E+b#y4A2$;n5J7RxuFuZBCZN-iXvHJCx%Ew z6(;T2ssI}%AaVgFEhhcrrXd)csn4(f0RtCTXSmk?EB*y2?Ebs=bB8QXsj=biHY^Xo zn~)KauZuV?8d}mfAedn+j7|QUvw{kBuI9w-B%Iq7xA{G*)LE1Sn5GEaWJu#>ZjTB3 zT`zjux@d~`(7e}0gB$dU_UE>BA%5J*!48B2uzdE*r?lvbdzKQbML2?XF=CrUAWaI< zn|JrzESoRLMOlnfmVuT+L->w3rZo$eY>X4J(xR6-sSB@^(Wx%GH10eMr@J#on z;08ucgEG6`Gt!gS>4#NN5~IiSy=IL$p#ke}3VzEIoelg)4CrjYhaBKlC7SFOtuy2R zm7o!3QH|pGwqOm1n#B>@y;v!b)L!J3_57JjMOpv#a+u6~EN9sKlcPIVlqb)%+UpV8 zX#vMnhBsNlAImkNcTtL-*=j-h-p~kf z1!y5-pPpkFBH};gE^VW}j0Uwv9Hi+7)OL#$rx2KN*LhL|%R?!3t*PK?{rFbwCC4B|2>+bfWlp=5I`9%VHH@&Gs!Kqss?w z_T+3p)5E9Ei#S^)u$Oc(rGw8p5s^&&6SUi^6?CHTCPZlBe4L)?C?10F@Rv1x6=c*@ zoVN)}W3k4tC;brYLJJCo3c}RDr%U3U=4UkJSF=75jAMioC4l0P!0xqAt;_xTyzmb88HXYqzOI0a?`vlepA)P7SD*>uiZxns zp<=f@fqsRNzP?uaPR4y!ny7tSS)N%r@vpu_6cSb(MzUuLubZ|+7(unHZdzB=TfGsmQ|bUOjUHwgDeM)!yZ>HXF|jWo ztHC})g;PxzAzLCm>CkXJJ^DOUDcMG6_K+?qMSbPBvD*t>tev~}pcqhugYVijsy1rH zHi%PSdp~l-1a)!)E_F)W-;zQ=e!mglt&N>-AGiK7L=RwX$`mV30^d{Q51Te-HC8ls z74CK^iAO~`-W^|8iW&T3kDkEqTj(0U0?IIQ?}O2 zYIir4ryCNJ?&7zs-3_X`BNY#_uX5A35mqB$j5oF{j|s~keS7-VBGKqzSPl{&iV+;k zo8Md)!o`6W7DN7%pJjB5_$vjYS8}g$-aNRwOov^xox)_7UQSA`rjNq!w9~l56**&T zoCu!=m1PZYrke5}1w)M~VB8TAEqk>=$+qq%<|&}bZV=O@$&k8;%u^a-H(1dwf?xcz z1$}h5cABJCP!t|)|Bg}3sr76J>d3w7!4cZmsNB58#|nu!6!rQ!`@4moJIK8()Qv{AJCP&OCsWz*trlnG2+@7{HA9zj-NS> zc1<{kM$%M7EAd$Tp1oPkukG!{@hT0dL{RvJE!<74wYS`y^;vd`!ZMA^=a0BPV@`7px&YbHx z3PJMDm(gr2Mly$5M?7V=bciFWN?>*8RDOP!eEXRKym{7b8=7KQe>jvDoCi)?o!B$7 z=Gdy!G~#|DoV2g+xaVsYoe@rTnjtH#$9?$M^KY)BSa_o<4IyiD*Tz1X%bIbwV(j$NlhI8jI@@>W2$>#OC=YaJt$#^Xy;l zHLXQMK!82(KBaoWiz5>BNZfQ$`&|e&-~;OtqaGl~<9j?z?$$fnF`@#rXL zn+LB`E!B=94x46WmP zkV|t-kqSCo1_&~em4`1A96s!Ok-yYa62EeazIDA-2IiC;&Oeu@7rvhK0*xcQfbj0C zbJ1we)Pk`!7#nHaRKTQR5x`|pyM1vD_|Vq${Z_Q~KU|GLyqVYL9vI&GeZWUsws2l} zTN469qi1DHWUVKN3Py6bK$JK^LMehr0GH&D*)a}ERr^m#vU7)WT9El^&K&N0+EPk_ zt!Sw>UqW9yB_aku!10_l&tp=#<@JurwVtY*k2BJz@Q+*XVXDBs5Ib2ALrpBN_9C^s%+Y9}6L>x1NLEjwG7N6XIGgLMn^_f>+RHQstxSb;R0aTQ@Qzv&RxZQX_+RepJ4#$M#+--=W1hicG%3cWu*(dE4jtC5P_>cCGH(Bll#Z6cC<@X>D zO1=+GKP`C<+se1q0;lCW4wk1-sTPTOj^T%If(lx2pi6hd6_QL9^c2t)Z)rB|h`qvL#i-XLEdDvX13ow<%u zUyewDL%K4Wzv$)pMS-%415NydK_5cS1fUFH72+tqJzx;0emwjb zP&^895|uuz5CL?=B4Dn)KZs*l&3-=tntUl8hAcy7;i0ezu>rz^kw6}pVqj3vpau-T zUu<4jPi$;nV%`Q4EJAsc07o?L7(f{UEnw>Jz{i z9f%ov-^O~v)lor!7*?l!P>{jO?*C!y9D+oFmL%G?ZR53VyI)!9sM*oh7+G2a*zWI10PwR}_hnoiP`<2YyL%_{XJcOrZMbLuNo6Yk_z(zq)?~m*!2$Cp* zxOe;k3_<8ed@`af0k%T;GLPa*?v}H>Ju-p^k;g;;4=O`|{d*+Gp~cFten;p_;QdI% zkI-kYPTdc)(28)G%dnfLIWRWv!bn_Lrr*1RfqKwD?6Y@rYPhHuHm&El7Ktm+m)Z{k zdbS?GWylZBn-CmWjoLq~Jc>7uS0ogG=t968I;uW~_SD$7nF$1@83j<)gGew(0a$Rj zmq!4?f=!wSKg;iS&{t(Q)qBZ>5HOfG_KWmn%pCe$hD# ze+1A0fC1#koxvwc?XpUq<7&7GS8%=0Y3@S*F344Z39mF4$ zs@q4LO6UitmxW8yeFdTMTT>LQ1tc7J4l@Qn$HWX&MvcaTQye)aG&_e7_D^P(26lmd z17I+NIL09nq)mncdFi7`*o|t!07O;xHS1v#pux&>KI6lytRnQ$`_M-NNcFiQS3>LM zfj*uhvqPqM3RHLO#y-7Pw(51-so4>n&Eyf={=id=F{s1rz-bFxs13z>^?aXxF9Iyz zYCZ@x7n4p!F7*R(dW1lLs~J-3S(RZwD-v4YKdaBN3x2U_)_nj#Jn&;xlp>d;4}%*i zk)Q}8J8VGxLom+SMi{lYBPa%WHf-38O&-j`Q23v;+bh?TFFELYVk_RB8QRae!!N%z zPt|^}E<|XOpY_`M%0!ciOiChB}$>u(G>D|2zJttj80KD&0 zWsOH?DuzC5oD}nwP+e1Z{Tx+^tLcEZIKW1>N33=$)QI+UR*Qx)@imjZBRsc=#F}s& zYXwXR6`0v=PcIXf*7R<%f{3-UcSn1(+X>e6m^;w~Rw2tY|)K zRkj&daV1yJTni0K`j#9CHfAsAqa4A!C_D^)_Z(heJWPG~vypAm{nFr_B-Rz3nv>$( zlw-rzRs|tejk7ec39dTh19_yMz;d5`GFe?k0|%M+KVic+ zDGRfF-0|L^+oHiUn)xom!_il`bNiWKZJGAHx+qpz6 zqm7xl@W-d!>&ySZltggs)nNyUk)djaFNQs6YxK5;qKBvSx4&b3X6D}) zWGaLEs8brzQMvv2ne};RRW&8U_}Y^Q*jpM0J=UA^on@Q+!fSSejbdP8cGRT!%Ap4J zt9&oXwOnuqPDyt1sY~;GZ?4;_@Zfm+L18w4dxl5xa-YKXot!Xv5Mj0L5`b*km)$jfh`}SQJ3BL zXdtG_(1M|Q5jeU4#NgMLWQ;(!cdBcxKYz7@{F!&quC2bZj8vNN-J1qn%Flcu#jQDylz3CPRp^*wxWcjIiF# za@G`6M5B4El-hnNS>ci2J$5$JdiJXCoqTEer6fCBwDH5xfFVT{T&|$4gno2B<`_ZYA7gH?N{Jd!zSe| zWTNk>&?09Sp1*W_Dx(HIM>JOn^T3y2tnZ!A=s7GQ~SyIMY&c4L`uv(Bw=K&F;e&E%%T;o}*@!L@lNV1>rwj)qT#t(%*ZUQCdvG7k49-mTRRfA% zLu3Jl0&R9zjSK2x>n^4u4=O5qwSP~hWk22pu0GyV*2{aXN3`}SYr%2C&Qk`;zc7;* z?MJLj%QsVC;nNEz=G6PC-|2WAWqg@6rgi0}2V?)-5cuYo$rFJEA9pc3eV0&L_a#6W zZw!tJsTc**%{W-zWQB_*(nh{FuWd4qa-2o?>UyQ{qG;sL8{Bs-Ozi?LCt|rcXI+F_WrXM>@~V+w#U? zUGBqnu>e{(J7#A=2%E^<@Xr*2Sn2o9RY9BQU$!Q=Tuxd}duR5LUAwLJ$fWz@j+be* z@VWG3DrUUHhly>L7V;j#k@bsFPnne-w;#@x5Ai3t0&myeM6K_b_(dYFAh4H<`A$5# z%>t1cx*jG(AG1wi9XU5bM^)?g)^l(J44Tm1lgVqX<44bI!73c*ZBql555eb6Rn($T z+opDon**Lkt)qqZpOe1@j}e&oS?OK!>M?S)XR5j!oh^#D4i&#=-g!Dt;l3YzP*m2P zFF6m=TwL|bG zro@!aG1uz94d3|09#F5w4b-H@s5n(sVvMOKhRAreS=Zf7dXl$a)S`6P`r?Q-L@0OGo zAPqCwIcbW~ULJB{WfsS1uXt!ZJ&zG2d;LZ#|Lzr|gk@s-n&Av0fVrevkJ)uOvSe&= z-8d0u-g{n|!Bi(`0qR1n(DTGiR97~(M0B}SlGqlNek$*Mem=4VNtl;wwgT2RKH0e< zQ+1oeGEZgVJSM*Pe!TfDjJdBPI%-KuRr9ij!tgz2xVSMIB=XDhn49DN(> z1-X+uaZsGLP7`hyMR7Uc zkzm9Px75nQw*)N-alTr&yfMLrzMYg*&G$??CNQjT;LiIL#ercWFK7f zpiX6y#VIYeI?QhIKy&K)sDAV_Cl$%fHlCSe_0Z!&;k%X^?|jk`S7?4OPm$Ya^qy}| zt&LsMq4Gc*>{%n&&9s~0sIz(fw854RP{=bcm+z?Uxi_>Izerj=J-|?G9FUnztW>SD zF~gTg;Pj9qZxyxAcoR|gq&+`1G${ZruEd?)8kT8WTP&wno*g6Ztk0ZzTtK1@TYRB`DGjOmQ{2;Q<1kk zAGt4WD&}zUM{EXI7I|unisA-7r3Z6o#Fo~r{zRqIw{5p1>>@x`bA6X7x~bn zymV+4>QYN4R+y^3B*&Z97ut$kO>E;BpszYmW zk_NMEfqtA1BU**nMr8)wqZ`t;A?A-SkroveF5-5{Tr(u_y`^i6ZbR<+HG5mZ8qHz? zQkzjP&b;jmQ)k}x+2vWKDe1Ig#iq=5W4sB-HPgKB7ybE*GSqL9H<7NEz=1Sea-XWi z_L{aQvjFz#yoDKYl7FtDF9BLqD_6OWuM1Q80|{kX?slt8;45$5M2=^48L#i8q7>P2 zUZ&DA zr;a)PA9c*gLH}Rsn2DL5>HiWL|DC~LVqs-x`(J(hpC(BwCu0W!I#DZqCu3n_Lt7(b zC|+JDM<)kkeQPMUjiygyrR3cY9ysKM0H@KC&dwsmj!w4e9Vb^;R|zPf1sdFiZ33Wx zfA+=uU7wj(9G=tHE^n~FG}RfF)#e*#CJed6$}*-%b-!u91-&>bCNAhq0Db{^*eg9K z15-m4V2$?mOio~SwY9ajxbkw+EdP=kR=v2Qaw<81zDo;OIab%;BSE6IwYT_kKt_Iq zIn0A5fHZeN6#tqMES(sB7l7<@K7P=vg9$5>3m|)70TosP0v-|(psTZED?{V!k4`Xh zrpH@}-o{~dfQOP37d?1%E^I%FF(883f-xXDAW!}%Rsc?V2tBYS*5+0aKN>(Req6tj zj-(7ueo<+4MUm_rPzAN^>3PJfFJCQ^B^GAmG=KyH2`mMGJTzo}QAO>;S5;U7h=JV* z$h;!J``;U2$gk|0KgEP4`ISV&GasI?04{!9bL+z=uA9I6qQLrKZ$hvM7UwW7Z>4~K zGaJjx@_xz7gM)#?6Vt=VpjTGI2WG$Ue6W_-=zid9>u>i7<9T^=R?Eq3g^#Z*9XidLF5Fp}h99+LQ`ERzVkG>mUnH}sH4UE8^ z_s-Y*&9_0k`?YlJOj5X!-~|>HI+FkkXYQub?m? zJAQw&ntsnouT5-gjV=I+K7El;l6VbtoLkNoFw#2!4*G$@7LSJH{F0} zl&rL-tA9zad;f`nlDKztVD)ui!p(=+zmKo+4WU!CNAHG@pZB18CZ`AQ`1nTD)7Sj` z=;#2ey$gN_gg>DS!r!}IfK?71w@#?27{9yq@8mL2MsO{4fa-wkZSDXiGczSRV^(qM zdwVBfZ*@Xon?PspvvmM!>il}D;!u6_)64LBkS?M>-TB;G04nvnCBB4s0F`g}L(qE3 zZ-H+BRSCWbe16Iod||1ueU#te^}uSzKLl_9l~?#fkm_^zL!f%ezkvi=JKq8bv{1eX zL6HaeAo~SR-~w%be!=er+RDBJ;%+Q_5sErL@IlUN?!g5*EPwuuH2yPs`OoO_-{|rO z{9cGi`jX(z@*QL%;z#>z?1m{hae#qzte+brG z53VItYQ2|6ORJX#4$fD+u2BG`{ta5t4B%JEj~eAoSKHm^GkZ(w`qywj2q9=HuDg6t zSD9cyZS&QVZY?T2Fgn+x~qs2*WXR+ZJ&1bfg$`mJqO+Gk_|IoALC}qEq<(1Jr(!yG0#rcOLhl#C-#*_l{)B$t zc8)fJ;eOhe6518tVt%xOC=WyJO5v^!3EHdYDboS|#xxW!( zGIEdGpkAwEycQ~xug+(M-x6+HDg1~?1*-iym`U|O)d7Wmn+W96;hg`(d_Rh6T0Ac$ z2UncAD*B4}5;0TM*8|a@P4D3hANR#-qdb8>fQoeBx0yI@ByRi#%rFc^`pWIRJzyu( zrKeq>fZ=VaMIUfanrUp+V%DFc-hFh``M8Mqhlw;5$E|(7HeHj4G~yU@t2_?UbPj8B zgDP}`C}9u>L~?|RJ-WO_eJ68B^J(9<;a$GH&y}SA>^4*BBXJv=a@YtsRJap9cuFID zVY)OiMd8XNX}v|jdM$Vzublk7aD0ioe&o+9W@`!D9AAKVNnNzU=oS|YCK0%*{Z3LO z52ABd?e;~jbl1Wcwbp%J);BxJSQSKE%Ujs-J7qrIh6WdEXcQLPO!Frh%AL5BEUXPk~zIjQk6ww z`CXQo3#)VL2j$QXGEFCs>SA!@M1qxBG6&C7C6dSW4@$qgq+u5|<>O34vb7TY--*1s zlFFEiU}QDis=AioC{=7a0-@0-U~JHFho3)gQur$NX4`W;(85{4kqWNLH)PF??56%KD%JJy?y`|>(f*DpOxMWzxpp$a=g`J{*vQv1(}yns(8r z1V~nivR+wT^YAC`#Pi$!u!nG%6iD1$EOsL^XCdklH6juehfN)Q#^gRa$}MLy86YjI zHx8?G7qg1Bt}aHRQ<_;OFMTh#wvagmpTSKT+e#vSnDC}YI*c)JDwI&pZ1HsO^t3ln z5^7VrJ7VnA`x)fx-q2+wsz-qgosVg#Cde!;CXa$B!@h4 z^NhdnK0k*=$*KvM#|9bJ6|Kv_=Lq$nMf0)um8ds?TqOUn@Xa$n!TI3eqPXOK2rn1;ya@J8Lls!F8aK&C_IQ$8cHM0b$gi+bnQ;alA*lEMe(J#JjCs`u^T(+8MOSOJX$Bko(7YIkw3-URL0me47t5UA=Vspa{?MWb}5Z@B5?W*E0 z%q5B~)9y|TyRKaIU_M|(9Hr}ytf=X}?DQJMZHD}I>{Q-fZndIBOb(3T?R#>*H||#d zX?bjAZZS9{v|@?zsReyzMs#4^KY%QSLLH4T4ZoOI*93JVCy_mx6HgAs5lE|}d$n{L zmKnNh3ebUfL}F-varFCwOV;=SH8o}|#A zwig{X+3=-TTPbW5yW?D6gLsuWjSO@fK5#?2Rf`-QS^{1scK`fpuYFf?RHwUhQe@X# z1d;A>XyjP2a(*+uOoWrdEjTYxs%M*DQafj-bj@N+FPm2TqHQF)VGj}IhvnI5dO++g zeYj*ecN?VWXR*;R?3h}L)1%J%;)USIAdTZD;2q)pp0vlH9lgOT!YG)uCV2)hboEHM z?N;IQxe^h1k2)&;aR`qtpi>};Ol$Lf-Q9hP<4|+U*XeOBVHy9Gh0LSHY$EJl5B3Ps z(PDV>mn|^s@&k_AjIxW{=%m&4wavY79zQ6@8pf-U=is7;(E>xaxp=4^r|*aw$b2M1 zPLIMo>OweVkxDCw6PoB>C!Ab*{PRgfT6j3Pw+#|yJuIX4-g#ySnURKfYm;zUL}zdT z-|@&$VkgDgp2Y!WsSUscKs>k-G6ZLVW7e3rtrgou&cPLVI-YWQW80$X2F z_Iz<+KZ$b*ji##FnETZ$ql`6(tI-E=1lKa&5SSrgj-N|-jYXx(CwDh{dMg{_w|>}q z&XdT~Am_CVeDIm(*TvRr6epf^7o+19^CMT2?9TP9hfEP03~m zv?wLV!*js|k(42xy$(&2ytYOEKr)bwgzP2IG3V;7kQNrp^V)Iw?VpB}C>FNjT2^08PsQK`jgzku!IATHjq5TeIsCsCOrD)`of*Iv)7Im(H-b+^j~qKnXx^16HLN(&r!#pNlBo71Pyl+POkwhiwK{$+qtt26fX5WVnN3WTY^3iXwW zfWbm^^Q$;zHuP807VzPxD9zqpZnd)8h7y@v=%z~Zcvn6Oh*1dgI^-o-D$!)QWEnfJ zskY4?Yn$ArfM1(d>Q@wuF$2vmkAbP0sj45ZpC*F70N1&=7!cLHLmr-}{Z41y1NJhy z2A!TJs8apjD3A1A9bsRBMHftnQmfgzh0)?A-7c*~N0T9@{h<{OMFR&?3 z(;N+F_rE9R3pEZQZ0wKe{^h;pl=)KW=b;+JsR#A`@pokys}S2sS_ENj#8E5J*zc3T zx$I3~8y_jq7mN%sfkDj#(-}Fq)9hjbUaQ-59Ly8EW=wBt6K_i-3?#?zuwvcyw!$WQ zMXeKu`1a3Xt49VSw0t}p@$#&UodQ&LfucHNFy8lC9R3_xk=Baz{SI*&{+x~a8`)mg z28kPdENdDCtCDc-ZC@M)_?WIT6q9w?Pe1l}Pl2br2mY0$z$xAVLwWf_1%GEnp)j^q zLc{p3f#zt6K$wsu3tvGO{hUHdEXCm-{HQ)P|5mE5+x)oU1VCVt}OlVVD1D3l%1E{PdEr}S>+g+F=UZ`W6>M+1HP`DuH zK-^4Az5E_YbC5^6>t%%dpN}R68v@A9o7I7yi7xs(iUJnydx9=-UKbayZI6~F4vQzK z0j@j(VmN~JQCKkQ_L`daYbr4`SErh(!eRT>d_7!w+>P_aRd%>wV-p=P!Z&?3)K@5J zfp4L=CJ%m;l1L_RkUBX7nQAZi&c7p{bZpB`{`f+f%jLI!OEAK-q-F1Wt}|zqrx(Rm zALXnD64M8c`?+)M@!Du(@40h%JB4S-3JRUfpk$O#p*RLs^Cl0Nts`GaHhl+IK1XjF zL+jF9>H&scL8%@8Ry{E3Af^F7s?U2A-x}`Hcf%uV(>dz{v0i=Af%y1i96=6VUd-afp7n?ih|(R`o>1~pi$yYeE0XXI4W z(+CUc_@=u6CKy{G7Bj!}+K#8Et_>T$X+3Z%57C9(bGr!XGcwzuZ7Igpu_B<`b|;W$ z6H%%to<5Fa>&7W8$Rs^zeGzlC`5f(eaJu1M%#Binu7_fM#+#bZuGHdVziCO|TOdLo zF4J#giW3{7*JW{T64@E)HC{t<8{K2f6Mze7)wq1Lz+4vq=^X0C)WDH`!*=oAf9u+ST&3G zyo+7ot&U61T;(~p#p|xiyl6Xv4_B0z$vxdV4pu|j78o1D`jJcUo2|j*5#U?2J`x936rb`GpmLFLbWR?0-TnocB+5L z-Orr0L6OIOxKdx3P#sEMvs>zYPOCK2ZU`+MTLaJ|$jHdCbW@O5>RSkE`=j;yll-iE zhCE%=wlejiE@FdF&7tZIf<=sIxenfpK@EyW1pCjKsFx=JOc~ojvz2;OB}R^Sg{vaC z`UzHU7X{O*5VJPC^|&__E*!<~zrqLXK&&DagBF{1o+(5CVF=M~+Zrddi2kYxpy!O* zDY0(SGJXj`n{oaFH_>)fcm8EBe7pqZee_e*%ln#8?i+CGueoS<%kYT;^dG+g9$BQ$ zK~is+ZateD=mtZH0|lB0ur#w;J58A+J)VA;E{|Ts&4&g?O`r7h(Ud#<+Ig;moOvpZjo$82^3vE-9k`mfBydZtJ`cz5JyYxd( zL)Z3lcwZ9l-QiMeaylCxD30ALlu{k7SLE8N&;0I)Av0aHlj+pm?dZtQ<(`idoQv1A zyFvZp$}$+b29$Q1U)0iB@xQNKmn%M2Eai&lF zQ;;T-Mxs3aWrq$YQ|nu(%3ObnjDBBm1~WEROa62pr1kxV*Fgi4r;hQbFf@=c5_(KZ zi7Kv<+?00>4yHli5i-Id8^%G%4eEnZm%1$pkmd(cyP|43SJl^#ktuw4(gjjVr_TvP zRO&|V^nntW-ez0|6SMQ=n{zBEtt0MzOPin5X=+(9@fBI-Mug$&qrq%eHRfIepgb#!5)K52#$WV%M^IFfpKH$Hgy*Z>m{}XlJ(#U zH3DQ)y-b8FwUv7{}5t*pWn@7et;B^k2u>O1N+vf|U}K3&YtZq^Xeh1S$Gz0uAz zuVS0H_$DV}v*Qbd*kyc)$e<^h*_AWH~r(bWydnd}3^f+U| zP<(;2waD(Ba-?I8Wo;?{0#)G-%|`fIH9c)T{^s3W$aGtb=!54++8wylx|Vt<5t$=L z&VOZeUjc;4%R6iK(Gm5IyUQpC?I|dL$Yy|#X;^w4ilbvmu!+6cKeC=K1<`3BDs{%c zEx+-5%iu*<)yOrP4IjAZSUzocr>U!>p5UUftkB2At$y=atZ3U_i*o4>AO3A^(_K;f zW~_VDi{a;5JPFn!9D_1-Gc}~Gf^3A$-R<=_b+w1~F4YV{iP#2vZ_$L1*x>Q2z=-W? zt}C#_L!W#fIhhLm(OkRuur=V~l!}-i4DbCyRUM5$HUi*Wcs8_V0j#ZKR%CSxYDi^m zCy)B_TbD3dajum&HM{UGbb@%7Rs~zioV&9;qG-pjfwiCy(AJntk87!QLK5G^9;IP* z7X^&rOD&OYuw{r`RJ(At10egz{;4x9qRwr^0yiJdegpb67)P<6*p_a+)cTbtGFnvg zz67F-GOdxk^t)(`$sxZ-$Sximu;O^L!pju*vf7xUaj`O?^@O>8OMl&iHLd@lkFCoO z*P2AAF?HKP$%I@nnMhr@jFTVbyu7ksHh{NqqR4kpQbk2mBP+0Bx|in+(G9&wAEm^^ zOOLga1x(edmB4e$_pzC2LzeRsPwa^#+)6wtPHOayve;vi6J4+C5A2x|aN;njgb8(p_v%i6b*nB2{ z(0|Ejo0p9NIM1V)Ik{?Lxj&V`p%#!6qp+!Uo{gC60Ma-d8TG-c;!}A7$$|7?d3Wtg zX!Z;n$^>D>x#(|aDp2puc`z||4um9BqjMbPO@^SiZ+CRe&-Yf(JBfn{fwZum!rSBF%n?1*J_P8JCiW!SIa zM#Jp);Wot-Vm(X(vkv%%cHU(Xh0U_8V;b(!Ff)2(QNDTXw5@+G=Z%V~#`$8r$I#C5 z-&Q;e4JR_|k8xNg{=@6Uw?6fqF4dF3xVlTnaGQSeWVh&&9KDg}2)X*{DkzFme7z3Oq zi3Uhd8h#oK%ZBykz&83HFz@F|^NT1T(E)BTxuezXEw~}yx1JN(yhes9%#J<9@7RsT^N{B z(TMk~4Zm8q5w241_k`^k)uG0Y+yLEySuZ8vK9Wjp4ZO?Cww)f*T5}a8Yk~LLP*&Q7 zMz2nU;K%1O$CZm^uOMQ1L$#vnwAaKypdp>}a1|Gbe|D8>{*E$UoArWIgN60KtwKn5 z54_S`u}>t=}NoH}77eMd+>g$HZa!{@QbTFkemTX#VPw19U}Psi_1FaN7+ngL42B%!<~@ z7Ot$XkFNgU#}3h-YB$Z~8h*gMDL7!b2w<=zV!}Bsgv^l^OQE(y_v5@K>EOYwO+(}6 z@wj2cy%{`8|F7{p1ZtLsOB zp-G025Sagy{yHVx{8!LbO?)~nw>R}hqE&?N{Au0I34HC&YkSQRiPXuKd*%twX6Uib zTm!}g6+i$94mk)Uxhpa+ndRJGKOc5VfFQJDb|YGbd&SE|tE)o+Nyt9Z2JoQH#2?=& z78z~nu&8lC-@6}@Y~pl}MZ*3aEnmBCTimHjCk#^RvarIL+osy~{j2Jor^a5OqV&@O z?G>Ytv|l^!330>xlfhD@+RuF2TJq9XW=)|^)2(_64bWep4C#B++oe7+5E-RdXFfcU zKwON1He)*$qyywvY2yV0zgslTIf#mZBTFV;%88~zvywm8cl`c6R9S4uJuO?r5H`_D(guq%>77aNw)(w~cI7 zLQml#v?Yqlwt5wR-#aj}6pj_LRwEyYH6D};=8&~8)Fy02PcOXhj?7uJcFo4$FQK&! zu&8}GwsmVoXAF2z%NY{zx)P2b6>%_XAr;?jQdcle;%90o1mY?fdwnWvjKO(A^CIt~ zDf*i6)ipfU*Y*D%*~b+myZA}9$BT_R<0K>G%6pk2Z`N3Hm${=ZUS)sQ-hqlvG8(gp zft>$kQ5~Ix8g_Togs zhBJ`vHUa@ZPhPE7u~xZsJbwD1XnChUt>?V;(Y7PvyXhsd^_|@BX(<0v04r4ecB$v_ z2EsA^=I&($2}Sh z8?t!{BR`iT?fXns%UBrsE2hWI|psGh;{ z<(4z{71ug@(xEcv9@AoSr&XYpL&8FC1N`IWmVD3w&v%Vmknrfj9~Nx zQ$3Y29{7bI<-_I4Vg8imo}!R?=DR~AQTUbpMk|ED!xkc$AwCjeN2_uqKu`TG$!v<~ zSS2#_1yFLHu7JswlQ;vZnW{^jZ7OiiIa=u}eW;_KAY*mMqRtRZF0#VIKxqcZykG_? z&z3B;Hr4WG32q}4fj*PxSyob2S8oeRW6$A4O2-q7aOCBspVhUf#E;@Mr1x( z+KIq*`=$+u?PtQ}Lg9G0b&P*$B;~d+fosOjQJWY@#QN>$cQUzm>99rz?skYp zdj#(jq7m?;u@+Y3J!u${3zTk4?#o-^z_~Mg6moI>%+g$^NVUn2EGv8}^$EE4o-;yd z>~^v9095?hI)5Y14lp8%v}-`hxXJe$b4o==QU&nZ&&z1+HkOHU+f>S6=ym356pwj| zD3&PSt>B@1t$$DsnAz^MG}N`JqhP8ke+SeAs=tWsQZKbBCyqmH71OHTk`K$24k5k; zZx*Ji2n<1Tr0O~qd?)~`dTrM%}Ft$T2h1uX|{*g;%6B|U$ytk<>J@`Z*a;pu_J?7FyFzGIf69$w_))9G*9xB{CT;XJ=eKPt;TL< zsr~xY2{uclch3!N{GpjN0AyTer66#9MUDV&kAJ zo_47lc?f2W!P|-`a(iFN4b3JB^UfE1DR{UEgw1bg#r};aMxOUejw}V<^%)BtS3W^j zFlzZGKc>Z5-1|Xa7U`fx@}Dls7J2G$wECB9F&I1}vVuJ!#7BEF@x#VSJi1ZaDXQe- zS=%ipPi+tk`hFym?oV5|C&fd=MpJ}dh`)nT-GlR_lQ@jv-ko6uD#vyJ6Dl0Kehu(( z^U7&Ow9tZ9v&UrS>V~kKZ_U1_Tuow2vkbSR=luH2Cx3_v3DPgSuB4l zvZJJDHI%X|B8H~c50Tlf@@FHZR2Fz+3E?Q{AR-j-POY z+Nt2fpgRH;cB##Kh-IJQW|qOhtZm?(IVq-@qU>rcJoaEnp-QL?TY>9POJ*LMLTg{2 ze>NkTNGGI}-PgHFVuK)gW5Jx;9ckElCRLd~=>`dHm659Ay*G%+to8GGa~YSd9kwT$ zAj!Q4YheWPXX~+%;tjVvxmKQCk$9^8{Z55hjr~qs+EBlFC$D|UKdG+uLO`^WO`tMf z$>7)ia>VA)V`OTZ#GHiwON@~e`;_X%DQ|qPbAXEUD&6_)qOWpzes@WT;&9{v_VD)* zcoT}!rHiEX%zuQEufSs!?g*^_A6U34y@)$ecv~aluj&PmAast? zf{$?+4=?zaMiA>#nd_Vg$y%8Qix9j}vwa>71zhCuwXA2xytt4+V7LdFQZ`Q5>?4vN znuHHpJkA4iFGZEVmoM!G=Xo2v3NBL2xUP4nH&yAY%2w@-)k5*r(!Vy6#aoH#Q(lH* z8asA0H|!U%j39PibnU))zODTWt&hx#2a;;8f=Y}o!kKrEn+r{+(=2T%_*Y&rlu%uPRW}w9-P${RL9Grirg8XKcIrzC znEm{{!_8W|s_|L@zvYevaEhqCl_DnC)75XJ8_vW-Ej?6@)xC5N;bavcVF|P|sy(&y zRfe1*5;pxTP05ohJVUYv zm>m>+2NA6?zGPv{?$|I?2|af~A6w4_)ZjFtZk-Ihy2(3vBo)U=vExyPQ>M-MjYTOs0$+ z30+nUNtMvPz>`Q7O2MKGyAZ?lh_Jr#yK?j*)y=rmUybSLnFao^m7{S1!AjuJ4e zb(xh=wLYN~Pu|@?9q$M&u@HczP%|j>W6_68#*omjRTJC3HgTS0>}ea2pR zfti8f`5bV#yv{VWIc_fgq*r92lAwv-%DkB7#2V^G_lhmNv0kwbX>yz=dU8wiq@^IT zd-tVb-8qd2tE+DyMirurve4J?L5kAr%iqVwFAKGXi;F2^y8|0)|k zA&aW~lIKs9kYAV&!8!rE5VsBJhvQU&S`M@N3s3sb;bmYWeNUV;V~vL7}wv5#?5hoxuE7{eN!xPU%X@#Z#vPRX0PKp@bivT5OUY$6oC z+B?b0%wt>B*I`o4{kGp8PTQ~&&!BaHrQUoM#Pe@0JI?O4?>*~Q*>PQX+l|_EV?|O- ze7Nx7!njcHyCPrpM>|S-Scm9jjG(LFxpKs#A41>RLJpGm7$D;0l(8dQ9cbFB`*X4h zOA7HA8@Z{!jZXRDc{f(QiR;b*n}b?g(Rax4K2af(+`$6K(1gdgt@QADD~XKY{RK+d ztA*jkNNrPOFFUDh(Vr;m+Gid`j?V8gXK-m&MKT^AU|zViI;kH7>mf8Tx?GVLj{N?S*$(XC){Ix>^;r@eotebI74@YFalywsya}rsPPiYDfO= zhoGuvZ9o|6iAPCHHf=nto=Sj4rtd04Qg*c^Am5R!$>J@IK0lb~>vHLh0)r&yp5Xpv zK}X)4)+{W^Ha$m5^2X3@l)(e6NoJzOqU^o4j9s?(atp-0m0tbk@f-z?E4%4zj%90w zG6+GT{xtsnDHn|EtTFkFWBx+NGn=a8?z<>UncC?6SiFL3e9SnWI@MfRS(DfKPLakW z{yVR_R^2Ch6IOS)8zOfn%(ORH*BAOO@!hT-rC7m-Dp{n;xmnQFT2RyEyqEt&BGrNl z&_89jdM$^;?v^VMdJ-jfg*($Y{00u(@wUq4DmJA_r(g~{O)w2C{j|~SZjILXC<(De z{a|~u{LloUT{!1%c){7ug0ORk?(NtOZpW(xB#mYVD2Ih5xgHHqpkr81zC9X>L6ag$ z`h;P6$Zh8{vYz~8M%h(v7*_9qFmW5&TRCXEYWE)4u zEe6~}DuqtI?`g~E=5A#@mgv%!Qs&UHQf#{>Wkb7V7}2msaPrY>pT!RKb*$W`apAP8 z)u#u-r?-vqInt9+0Hhk{ZRy`8F9$HM!GZJo36TKZhni6pDCq@0E8@!ZV@!xRq$AD1 z28pK&$0bglK^ubt}FWj*Ih0% zViKXGM~6h_JeL&rc3s@?@AgdN;@FlebTLi5kaibXDVx$!3D%kw;Ye>H=@A6^-8;Rz z6#CxxPRUn#_4vDVs?$Fx{m5~{PUPk_7fewi`Y<;>EBcd;Sy6Qc4dqap#p>(p-j_lu z-g=*&`5O8o=l1dzZ$@|_a}5@xf)qYjiMew4I@6?PxtjumGkY@56<~*Zp?JH30h`r1%x$n9`Sg5v0n9yd*7Mef@JWlt{hFl(3oCJV(iSLp zhYX|VV$KZmlnu^kZWa$qgoM6@y+z|d5bmc7qCB}w$6e(j-7r*ilF+Ns8F*|A-rgqz zvQy){q@U^H#!uKuLbtDp&H0EwVK^Dj%}C4p4n)(_b5M9G7w0uM>$e!4dX zR!gaw5S0cWv-o7UTNI!-f5XP5#Cu1?!PFJ}$=BVWB687lldmaX(gtgQgA{;*=fZKN z*mF_%)zLn*P@>mh(xOazP-4|xp1>4U-6D#0*6%W4oiWeCT1lmt;@M39%S=tnhm(S6 z1Le!rT8UWTZsCzf%>k`Kuc%9?_x4nyztN!l6a~_*9Ds)O#qCONP7M>!so3U*wb*zH zKZ?Wv$zd8bhf0H9h%IKoO10EF%Sx4DSV8<-{mOvzt92`ioD1E}D1FIa7_$0o=v~$RIwCo}>ve~MaUX^|Q6irblY~8*#R5)aVRJE;$A95|)Hz*i z9hG;yODBYBQ27A1@u$CW$Wbt+&-hIcIBW|3b z8myd0vLtfe0Y;4@9!8u_edB2*sC9ww<;HeOYhl=oM&%C{EBa`q}hIT*g5bujzLB`YMut zV9Pzz%wP~%xORf%Zdc1)hFdU4%g;@ybfdV%%tml4@2lgf*|P7F>E)KwQ!z1iul$D_ z;C0#dd`}i#&eGhE56{Pe4;b8AKCzjMA!&8DCldOb{sxb7U9Uc!tOQ*r;8bn(Ig;69 zUOs6`un{{q<{lG-eBr=g0?j9OC5(S&I#Og*C%%~7UBJGx=-afX-(XF0Tf%mmi-aR= zP4+_E3nS{L^*W?8=YVsr)P}YaFG!rDb^DJ$@8B`aW6mW&x(E11qm`PRi{o zNrtuoxsShy9o zrPpiQB)ZZ}`qtl}q=Lub+TXlC)^20iAXHz2uh+%zP~97R07HG{99Z|_=bob3HbA(dZ=FQ9hGUP_Tn^+8k7V|4B% z`s#KGr_R|)gj+iJPiNx(?I{wZt=b#*m=XdYc znmNTe2jkJKBWxW2vTN{+mPuJ73?!8WBNzI%DD>is2=IfnxKFyU!q#fE%4{1JAme^o#33SYgWhz{69@%4CvnkSTBJCb`yVF1y zSt<{3Eo&I$Bu?++t3A7(O?wirtHXXWC-N?mRjODWF~@!!6x{Yb-L(N*r`wK}gx+~@ z%kU%{ZnLo@(6ry@#^5q%viwRUPj2*hFzwQS-}{W-%4bJisc@?z>zT9t+S>%DRGH4J z1GZn~ee>Y)>Qa08;R>tvZUf!;w$V8!{nf{AH_2KZC#67>2FAi*GUOEE->BHPTx(b99iu zZmb9-#SN2?MdmMs7x<28`7O-fnOhtcx2L_Y`QS%mT8}V;^7*g#vtOr=bXni-F*OK2 zJkMr$

0_pfbP@C@(~kU|(;&-GvG<^jdhP4Z}UmeGB}^)> z!idBi=|G&8KGwJSy2t45wZ4wq4?k8V1aDqQlgUbp@$df)5RZBNW#_GxHSStJYJDBO zm+iMN2Rb`kqRc+rj~u;97Z<3h6N#u+Be%P@>d%xP3P_W)r{!H+9+5`(-tcfaY)?s} zPCZph)Z032JzvbZwvGPOui$o@T&$^st`OW%^0B=Lx{0TN4gOw*1K#}~D|7PJ|(96}uHQAhKthSS-qk^_F8*JK@#&D}XH>RcjPA#)V=?NBc8~h4d{bIS-D?u>e zlOtm!qT-!P@@7qbFWL&rwx1vJn{=x9$pSoHngz=lSZY{Q;|Hr7tK$M1b?m0Vr;vkC zzHGskDw)C<#?Q=gl+3OR#P*5<& zUs2CIYTr^e5Y9-_uNP8W@-)?x%fO|6FXSK@Cj$lB(R-;hiE&=dQC?5!!#d0_t`##0 z6LF?doy(xJaWRwVqWv>=zabACW2Kx1%X3Bf@_uz9a@;Xr@#(SH@;+Wd*agva{cADq z!j)pb4$S9@kygRt{;-R%R;e$lZYCvxcX#mtwDY%8f+)s2*H+|E8!<^yQ8vhJ@rEF@ zURRQQni~Fc4GdjQhG#9@P;^Yri*O5ipX$c6ZibKv@7a8uHoS(IiFfkt*Z7^Ulq8f2 zz0^#J17>P9j4ic9D}I$EUp(vIf=78I@?f}?4bVlsnCre!=(B6L=2cU{6HT{@`)#eb z*0K$-v=@dm4b3@qH-H&JP8QLX@f0^@=LB9HSOrrLQhrE-zB_g_Tv& zvg0T-==e;yvL7ADsPEut+^~@xl2ka%Xxqq58F4MSw&~O4U1FAwGE-3rX|WWCxE^ zQ03rX7%XX@%WB$;^ZXi!jK!b#4~Llepx3Gjc)pavOrgMW$%AUX?uwSopUySq>rBCq zH=)`Ss$F*nep3X{N{f`sWiA3Yb*+yY&7eNn@%vde$CuAL#OJ=KWhXw5o?x|Qsr^$)?VBNKZy$oQ%Z6d}|IVNY($ZkBVccg0dOR7$wXQtKO9fs5Q z96>7#3|nAOEJ$zDXz6$vuCV$WWvz6Ot+nX)eZpKj;T2AoJ{O0frAX32=3w68gRK(FIm6`&Z-*we{kmSd! zh)7?bBYmn9Yu(~n4_Qhe_=5Mq{C*cka^s?rK)i*~o>ONQYrIj5EYYqTsEtu#*P!=S zI!-HrHQc3b&D^2({{tRC;lFG?p8O^bRMZ<{uQCYEej*5TGz^(j9MgI zlDpXl;#MTCX|D4m1RZMg|eTovKnpbWQOPt8^`#D-4yG00*T7Wtx(bdsp@uWqv zQja~-X?}!k2hgu)8?2cXDG&BHw$pM4{B;Msvv)n600c*gnikQO!G?Tjj41OzF9Hv; z5zN*rTG+YhFzuK_M&N4-J|Di4__X8ScEdP5K@c7Q_Djj6rX=7jS*%@`&>7@2-#<2& z&FOURK+j!pF{jhjqWP7sa3oCOk`)RDq&JU$ZAEZoC{saNCfvcKm@dV15YB=h7B2XG zoA{4QbLx)JVA(Id8-UJ1mD|fBG2UevUoA1cAzgrOFW423qyVTKG?OB^#C@Ru7xu?G zQ;|ArOinQ;y4t#>>cun6%BB1O)e`=$!>vQp`*#He9^*z1Mlcls2(VW#*ZlTJVcc}O z=O~Yq1F2lFK_jOP5`& zQstt}7tz&XK9h@bw&gO#9*Ae{aN*+)gSHt3_34O)SNTU|=P{TtDs$Fn#)OO&;h~d8 zgu|2G+d5dFYy$4JPWZF;B@vsG)XiBu04ffW?5MEGxZJqO=m?T0@eH zybmpH5d2L#+o0w66ZxKoTc+mGiN!S8)m1{vtKisGjZ`6(F(Pi!W}OfI%V?D13uk%= zb8u)2@$ehp-*!f(u_O>KK6(A$%fum~+oAlDr30JoUm_+8?cDo&Jl!Sp@}HKYD9pDO9@N=J9g)ez4*K(`P{ z?m6=x?3r56A!T=`zIK3Tn^JTwu3l`_ROHUXO;|B$#P6nm-80vFdpc;I?YfWx3yjba z(}-|mf!R<(RqBQUjxHigY7Lr)!3Bsr(^ukylx#(PN|WfxIf05A#|yMeo& zo=8sVM5j0n^%il37xA3*89z4futiSO9DoXN!)(xlmZWK@0QXKn7Uk=o#V>Lt$u0f1h>)b!b9{}*5Cv^^y0R z!uWiGIkYpqSUYNOldAA8IW!S_B0@8~CB+PlO@^)+H zE5p7-6*!Lxa?w8dIa7uM-{KnZ%Dh8|zMSbj?ppcgvv~CBrGaDk)@njtV|~+UWtno5U+(6k5ioHc;K>1R(lAwRv+R0u?`rA2VgJfSHa?Nr zn5}wC^ZGbO&sXeRW9M(~esq!ysy;g5#87F#1Bz1n#A+&O38jXrF65uO7w`w6v(`jQ zAF~Ub-fak}tBmbkyeyoVz6sa6*KPY1i3KB%Q?*uCVP=YMqfD!>qdI<>6JMz3sMNSA z@p)B&As7`y>~R_o9p-{6Nnx zk`{IBzWu6ajU3yI>t>w|$2RAv7pCW#lCiGBUoK&0p_pB5JILA*fXajP=BAr=!g2j` z#qKLxm2|Jl)P&`mnx@QqMdj8p#mKg7@d6?+^D#n#{6md?2l}9-sI-Z2c?;K2HlZ5Dd(XRJEc+cY& znze!iYkMX^2@JdGU_e*QnE3xpcnEcp3>pl=Vk1#F>*%q5YltrX-N)xve-UO>@0bND z`e1Co1*A@TWzu->;JDGU5MX5#w7qy!#}~atj9ON@Zm`fU!#%IH47*uHy}Z*t z2h8KK6Pd?_cPGT?@@5v?&YSAXYKh~SVCnxa-s3G-J&x-#BfPI|;xPt&9{A+Rxt*dL zbY87zVN&!Kw=8kxL*f_f(`FG`I!oySMPRc=p&E|{^+4KJ@{jG9|0-k-s~g-l=w32# zNT~=T&YQ~l7xaZ&=#Bq$hh2kpyrVz?$au&md~Wfk9V?N{{~PTvx@WOiP;{tJBdV!$ z17x+hwM-T#);wCNRc&G6pPD|mUNJ)-R<1+v;M@8x(P9Aaj@v>)Cw?O%yWeU*ymJ4l zS_Yk_&v8YKMp!S=Fz@5#%XeX|&rqk=ob^$*%cYO47b;fIU7dqeoaYXVIY)Q49ej{z z+PlT~pxS9|{&Nn%?&K$<3zSv#EfxI}G5|OvBaUT6NA`}tNgl+vI3)JZ>&W^)Dl)d(#X+mcp zZkB>T*LZl90|N943)~`-?VG=E zluw)Z4>5?PStqpxAN;M~&Onpm@u(NYxSBoGiv7D_CSNw7y~q}??}8l>X#qvOcbU`d z6|*(R0&j-Juwz6@7ek4JToJSNmE0|8>Sgwdz|L>99zzX5;c)haW>8c54}IO^QqSV2 zKmHHwTTWrfIDyUdl)|?{yX^GboODN&_vr<1cqNQ<=Wd@peoI~*di+v$4f>9HH()h3 zp<2@ zqbB;4;LyF*nyZ(XYg~|#-Vx7?{f=xnjyR)aH@Corv3!tl#h&A*U~l}IBe=UtMnOGq zD(_#?x|PvejUD7dsq*snXMM_l9F>q=i{~hsPt`mY7X6>1U~I#{bPq#)_3^V|;;E_y{KvM0ccqou$y3tVs78oVYJdi<@Z{+T z)8gDASiBgRyH+wGVMav38zVlv*GwjYnxg?TbCOZ4U+>9;Dzh|OVklz>f{T9$>|SDA z3R$mXFB!6H%Uvaog5Q z5et03N6qf)X+V}F3fGF#Chdqz=o9Fx37*WAK+9RbaVoWAZJ^C@4#i_e5V|47sEpD0 z8_p5;U$<6=lN)Nj98WNa!qd--A2+@EmaYQjn!Y)@3*z!iri!6RT3!DJZGmZM$BTeA5-A5#Hj2DqR)f z=b|0kSR5UKPj`e=ejiQY8=~RE6Bz1<^%W_ECQ4B092Qy#@A!CO)xo`WzWyq2-*fw0 zSnRP7*Bq?`Dc^Yfw*2zMwGM-uU3SMiU;+ZG=vkc6uYCpjwA$%}=x+VG?_OIF3!11E z8*{mUq2D^+Ml>v_>n1F^)V3fRvTk_?kw$GrUBvq^&L~8JbT&jDy+c+fGi)fVnjaDB z?Lmd>f$$r&NB1oq9%nJrPR+U)UHDW(NbTxL$&549o?vZ{%|Zd83((OzQxhTN%=1E% z1=X;V&x*nK*Co9EEQ^(ycIn7g5A6UbdrM~I-8(hEnTa=v20Qn#9bXu0E&-!pSv!vY zsiYBd(b%s3TBq^k9M?Xv+Jx5G_SemEbFPe0lA|E_*i$q@*{)rMDo&=OL#);Gk>74k z>3iKPzcS*R_Ox6-9SUbpi5h2otXL(9&m=#r?X_|r`jXHUxZiQ2Ya_B;bPdryaHO}s zgil`2hc?-qTiC^V`kqb+Eu(yj?5dfOkYO#NT4HUzT(CY3)y*uA89w-{-l>iF2j+O^ zqsPkL1OLKR*Fv#E=g^Rr1Q0{lS6+0Ao4)drqMa{p)Ife*f2|MsTD1~CaQg_X%WzT> z$;0-?@*2rrPG}mNuWB`p$vSmvrZJmi^57gw_FF=2OYXOClp7S17+Q}og-5J){3r}q zYWzI~YcTa`90QZ98PUj2JX%9OD+jpWtV!0#^Z%Txv2P|`MGTWT(6o2OFzQTxUDFW4 z1I6KsFRo}OqE(1GG>cmb<0kZ`Mui)~O;o^3B5)TL&3ti+E~UlL1=~MKJSXq2(7F1b zq8)4bGN@C})`TlJO-!rs!kl=u?Fh&u$3e>c`ZAi1MG^P>QH8$SGbsSu7phC0{H(y* zUx}G8H~fM`*a5njtwT%;Njeij3aIDu3a)ilX>HmgFIJgT-xZ=s>JR8ETs;mBUo;Vl z3oy(d@1_}>GxNb0H8tUUh=6d^eN%J7le_?AZu$zAR|~5u)Wb%2`2FvnPPNrm}dZq|%T&{SX>G#z9{Kq7?1L7VPRsq-_m=ug0V5B-l zR@Gro=hrL1i>Hg;y3zHY$a0v>cJUs-vz=EiU>s*e{Zw9y!L$bR-L*}Dkj!7-=R?%b z8j*}&Sl=BsPAOpzQDF&}xW&`}`okw(=b_7Fwb8Zum;#-7X+l;@hnQ?I1gzY%tFRg@ zQh#UL8tV()*TF!QDURIwS^Jy>q(VG8%oFXu zM9ZNWeCgp{8*;q|_37n}q~G0bx{Wfhs$J?%R>?zaEZAySQ7(xA)QxLDcVmx2Z|(QunQGseY0D82 zUqyOlg_4``0)Qu~Am>I}ZOw*C8mU}j)#n#X@Uvq6MTUqsmrhXJs6(oTFmYctFVlEo zf#4uZxfPI419T~%+Y7Zc6({-^r-V#ek~ldV5pQsmNcCk~A`c93IyWgABj4w?8t}p1 z*(2R(8?iBzmcdwA&Bw*Upp+N)J`Xo$k9Ws?mqSv^N(^cFXB>Rk&9Ip(+?fk#SiwcK z_LLW8dlCLAtV!(BEudzg^I`{-+GKl3$SGMgfR zHAXND0^VIrvOr|PL7k}rSi%O$5I4co5Orrm3;j@r#quO+ceNTj7Rkh)HFcOx6wi8X zd3+NKQ;LK1lKwQe>yF%>CBFtML(#Y$dJEY3trPt+B#s zk_3C|*Zr>jH?J0UZBJW=!=U%=)gt>_DP(4~vh~TwnVb?XQ{Glqi_@ZW^_;IjoK~3X z(??*WLV5VT!MO_@p-%LwTyf+S;}Sul))fb7P#Dl)-t{xilD6sbx{Su@zJ9xJQIKFp zbY%`p24I}D$YbUnu0OhVMXOgjUS7XI!A60))|AZ_-kE@%720<>`n9I9!b9L%mYg-x zHTN|TG#&{_gS)a3bkAnrJ(cemR~ab^L9`ta&&D@5 z+dD<$FMe@};b*VapFE#?Ye73uSkmA{#uM+%kPGf~3rY}&s~(Wh(72Y2q~CITKb6yD z$T+kjMjB@m80D`)i)!659G>8^f~wZ*r=EpqY&$(^m_JL_P$U0*pQnNzTvaa=&b#SX zX$HgKq6Db;Y@*tMEwu7I*Jp0W@eR4uzg>Bt9q-jkk{Ui9?g1OEXvHAcC-tDE`>m@_ z%7xKvLi0Qf$TQbin*;=qU}(R+%3m&dJ{7D$@A}|X0C4aq_P~NQn+GSf3v;eU3H!~` zUPe<5lj>;dZrtPBUEh-&y=(j@I0U@b&t~Ox22Lvq?)h0@(G-~h>0xqSb(b=*swJFR zst$TlsukX5z&gE9=xhplI{W?v0Ri>%;_uJj5!Eh^axmC0@@WRuBv#neF%<0ZwBy#< z7YJ4RkVMSk>tDmmfE1nOOE2j~n)2N)EaZ1C;u$HWQveVAJ3oeIQGNN-8X0Z;m3ihx z+~{`Y1J}`X1RvGVp6U1>8ga=x!B*6%D$X4|I zq?}^fb+xn|NFm(=AzDNo1oy`DE(_Zz_oW#_94d0z`7xc`ZaVQ2DQr=TUwh5dMqeT8 z7a!xu6&KrqIpaPSHdoKlrmqYz&1l2@z% znkkd-mG#d57e6G4^!T282G&nEwMlL9a;?7+1*djV@fdz)?;{kj@L9bKojUc-O0nE7x zPE@B;rtQnNkX3LXhtRv%L)O}6GuXx0z`YDyfv8TE&(Ul$v@H1uH+;+}y@X_kc`p-a zTv?u@9G_171Rfz8tmy)`(#>#%Q$qDeX9(`(`NW^x7@UE<7Cx&`kh@p9vZpmJ(xhsBNUjpVseqB)TV!cMaX~mwsqmEOZ zlSKsZVPWED_s_S6Qlba9^`!s@*fQCUzUs6{Keu!o`M_P##K(B(NE!N5O)>;|6`S2P0_||dmxlF)yxWDk((SLX6Mvk-a{n7xE=^MWYNo<3_2zfsS^p^BoZlm*a zTOC!`^ob@WRDxo}P)y?4zmEb8TdE)^5v4i3%zMdb>#Z5JW0I;n{U8aH!t+i8gNPnp z(D(6n(depQOyEG!;7O1Xj#+A6Eg>r0=WzHgdNqMIjI=6{C1%fpQf>Y}IoVQ)TamO> z7^s`v*3E!BOrFxpP|oiETml@+P{C*I&w1TPsjsnbF$Z|{VXsYLDM%7m_kfvR*R?qy z(+!6YoTe)!Skub;O~sZ@!b4>G#K9jgh2XhkZ0ai)7Yxkjl-x@faT}Vp`~CW#SJ~N1 z-wNkn)a=*cMB{Hm_O)ZVj0Ejf)52q#ef|}!&G@Q;oNAQa4Nr{nl(TT8abUKr? znPU@cvn?BIAjUNpn4zXB3U7{r4$EKUtfr)s7!C>cmVE0OgObhus@(xPvqG5FpO zm7Ne~V33x0aH}@ZQr>q1IA{gi67;CLQQFpNXwFGI^0UHujmX=58aw0JB+u9c%NhkX zf!Dgn&%J{Y2ExIW(k>Cd`Yy=AtocUtv>D^Veeri3V&K>bS{jFH2QK1=2{QK5f$;gn zUzusBjs;~dp(%A%h{cLP>m1+xxdHh+1Wbps&6TfdI+e`oi>=T#R9=vjlhovr{As|a zHhX1$derk@aa;bOLEx2G=9>z`=GaEJ2*EB4;2WJ(Z#*)>)ByP>AF&4gEUjh$VA8a( z(~yTLNt8Kv*2YXk^iG7a15QK=@<@}HkO&+tM_}K+lAr9agbm@+qMr&CgM})f%H+lK z16peCLzFrMAHRvH&*8n6vK?-%okw^L|8|1%X`qL-#EG?4-&Q zOx46(?VxZ<7xPH8zrlJtf6Ezy^G!+SE%s$~LuM)-(=lfL0V$^9sw)w(YITYG4ybgJ z3$M7Ci3b-c2W@pwCkPy`v%FP2!!8gzo+iS^*#sC2ZVn!LvCAnp)4eG3t!3Al_0(CB zzG(zJ|MV)skW>t=>?(i3PnN2{`l%pnFNA?jgemmsq@9io#wn%-`q{Y69d@Y{NQ{UG zu=vp`L$f}9ASFvd5#ZBn3j<(mJv`!_;zv6etIPOO$+y6hBa==TwI?qL_CLdeDhl#$ z7}`@ivizbZf7yD?z(M&Vb9(zu5ZMeaYTWAt+99r>n1+bL@jMec6RFS(FERh$tj4G?^gic z3lVhrGrvh(ftoKMs7(;hUgmDWu%!bZR?s8LBk2nj9*7y^$;_4ol)ig0A;a`7aPmEd z-&8pVyqI838DFJ%5S|~?bGD@;A$JGFoZMsD-9jGV+aGU~+Vv1j6e|yRmb3pD5Lls5 zLSTvPaB&nrQF44YMAWhN`K^-8uB$Ds=_qI(_9vpQD1<~q1A~YjUeNdPgxji`IZX}- z6NtSc_MA3<4b;~*i5NGSq@rL@m1ipPxx{O!88`YIeOhitN2d(4Et;O9d1ZYiJlah( z>qx~=I_4RvHC;AIQJRoXvxGOs^4n&SLrcmlPF0*(&>oI$mQ2WXtk{&dJkcq9Z^WJY zGjJuE>@nqME%WVq@~IR1EaqZC#Skp6!W?7!D;|Mf!mO6c)HbwCYBn1 zl_TNZwvE-o8q6F}C))xDm*R~iZibh;!_bKkh5Ck%8+-_CimN=Pc(%<1OsvgqTbmsG zjAYi`eUOlq`A?Bry=GR>86 z-qbCs)QeWx3y{O}#%1IjyV<@XHYpP=opY)YO0tNuu`4~tcWAF0xF_)Q< zkX@iMKEl=_H45sY3#W_1os9WilH;a)vK?5E#w^+giR1`UGhpq2#-Kzc28&BZb>*02 zpYF19{Etp=up$qoiuOG_txgC)h}9ZNzpfwtU{N)P1Iih~<@$1i`FipxEAH}7 z1pz4K!Hjz_rg6SF!5L+ihkl6nfT`byavW*+frk7#guY%Z=Bmt?oJ!&ahx!@c0RgvGrxz_9AyHT;2l%MCtn;+I$d4++Erm!_i1F`LZsa7Pt zmTfa5w~1MoCdt@Wo;Il4?n@sGXpo&Q>31W}1m}i~gnl8Be}F^F`q`tp-TY|W&u8W5 zd9np(&p!##8i$%4o~e?v`+1ojW2eC{(=S-T=PB|e>;;`If@xS(-^z%xrG&i)@K>Ce zLH@%c4mHnHLl3KkjpKL(at%C^{2ZT4Q-R0CnP_DfF$<%GAZmrAt-BRRAHf@75n4G| zcd_K(tchKd;9Z`bxO;gU#ir;VCQtSt4fUV+DrrLxTcy3Y!u1SJi`D~q+@Vz(mgT|) zq)f!(c_m%zJvqqc(xQ|hTq@tMoKDp1WS8at!}Kqb9hgj+UPPKXE_}&EqlXg#|C?=S z*i1_hRJmF%&Brl)YK=?xDDf{3h2>RFo>k}(4bwJuLhi0qHzio`4|z#F$s{e4e<_;k z!|^#Ca@z>7$y%HWzXp;RD8bj`(NP`9jAyY;Qzs9WK)?L4A-n@pB8LkpQ7Xg8xwMVc zayYcXikf+s;^AyovJX0@znLLsQmnmyFH=eAR~LPhtB~~0g=$0>x%OuR-f5;?=34-! zDepgpw_GWjWsgVDNOgp<`RCsAnyuX-AO$oo9p^2RuhGcjV_IH9inBf>=o3PpZ2V_1 zBpN}YuCd-n&i=i@Q)IcT{OwOaOI`w%;WI}OmzjiHvyv{s4kt-K3r(7f>QF0F8kNv=r(`W( z%`c{K z5EEH9YE_5JV`S+MkNwB?aU`|RNHTxAWL%Vm&`vNKaSQqmqDx-ShY**mMS0PQNsu5> zM07sHSoXLzf8-pn!TQy!iQ~UOs$_uF3=!2>8hCt26$O#8P8Dr2ftZ4@CVo;!NtojS z3yj>^dxr z1o{1VynY^~RimTnIsN?QN%RvocmD>-bNPw8Ey(!N@V&Lwv6GB?-lG35q`hdMKphvQbPV&2-YH1#X*(h-zU- z$&0aX&}(TKDj`OB3umfU`rkfeX2VHupD&vzlLyz}!#m4xBOnjMA{lhvyj&N9VT%DD zhRI0>1OX1?d60|3iS)hr?6Fjm1-0yikH5p%3cRoeH8q+@=_qi8qN{OuI~Jthr+XfB ztM^*3ncR$9v;Rd)EHTS_r(_dBFK|ZAkpwR5yT=J=uKRyQ$WShgF9J)7h-Pa8``r57bUS)3ozEpH$rqL??8Fi0lU94%Lg3Y1%T+o%*wvV z9gSo!LQlw$!>GIINZgpC$pua~Eg~Z{_iNnIEp$}+YALL}sp{LUZE|OsTym_yWsYg3 zDr`iVB`Rd_ai^=dT5@k5z^Y~4!cv3YD*+fA>{stD0va=}lT=4~Gov~97oJ^QQop`w zR03Kl@l4{0JV=3sG&|p?$7_6)h@)^>*-=MAUZWcYv(|vL5=6aiE#NHhv#JLuWEaq zF7k@WE00BpOW+nHSVlP;%6PDhyxc5v^${;b||T+r67*J!HbTdv5s(K*K~t*-5vavOH3}#NPTksO0ywt_N*|YY8_&DDN|MA&n~{vpzRWW?c76$-Ll~h^ZGf3 z?Z{vQ?e#>z!}%2g{fzhM5Kg=7>kDHIW}&H0jXkIMj_7CgT!MLVjcrmdYD|~%U1K%+ z{}$d&5uH|^Pgr8ya6k$bM9_43)$ddLjLF#_jgZZ|Z|_xS)jwbC;yrSy_7)l7oRXHA z$ayvhe#{YtezZ`f>S4=UITf3!;fesNXe7Nx(l^zr0U~J=URZ5~rw!L_>!-rT8vY!A z)r17h_Th8YZ;d@X!hI=xIiLXoW`->DpSxM&HowJ-5$M7WI)YtCOMi5?Cm^7ki4W$x z1_}=!n3=1!c5{xgEuoP2w@^}>?gwNvauis<8SloIW+F|4Mjr(*&f~FrGIA?5!-$B~ zroJP(rJ)n-FnBA|$&R)V6;3u^%JHQ9LK`V}uS^Ki=STezB_3(IUqU^JnH#Qbu8ygi zwh`PaOLsMZZL0I-@$4>66I+aon$%)xQN91W=Ue4;(#fOY@sUey=Od1s09cv4JbfqH z%z@Leu)ES@*HK4b1^a`Qj2Zq%p?~tV+3tN@j=TN$Pn4Q2^2G7ZG(jM)$N&fCLjngE zM5XZM4e^W-3)2D(h)7LZe{K~sG3^b0Knb>#!(UrZbQCzi*dkm*rqD!Fv{?Y$MP;;A zLypHl-CT?QUSrm+c=natCh(`^$ervc>)`}KVR+Srr^&$ES@X51SSzgtlw1M!g$(PC zNn7d#Yg83=mH&TNVTexMvWa}y(0ia9SPF0mN-@g=me|H@NGQZ7&l@CK_{>n^i~v6( zm3qQA&22;Zh!EO5>p*dal~*^;Uro;p(anD zWx%q;&%yt)_9WzxPnX~JKnJzDsds0|lPIm3PBYocbZ+UoG@%s^VJB{QU( zW3UDI7mrE$Wv~Kzye|Rdkz+_{e%|8dIZ;fp&u$i&3*@x}y~rRZEod8cbSw4BBU#6v zBKZ5c7lY*UMtg#OY?O16*qk?mAULWTmZ}Tp4a9|AeHYMKm5s8g2f&0oCcGqr!#~SS zb#V6$E4oTnTB?;?Nm9r3<%0%%_B>-R^6gRraEX^I>jp&SErpyi_3@_R&RNpn;Gv>G zb*i5(vZ!(Oh}kibkr>sUw_wQd5ltB;IHJoLqi467JhBv65 z20N<5CMg?T!wu+I^kD0%Hq%jIv+Ha?HR5|7EY(wo5PrHYBmpf=ZwP>b)|KRe`*^r( zsw1aM{1d}%kuz%GFhNK$J15~}7G=%GFGCp_{`?$DEHV0jPl?kvAfCurx&Op?xA<37 zEn&IQid+4Aswdn2DT_WFqh=;A+fRHK&$W@Pkc=YkA#2Spw!$*WPcF*@UZ?JKHMyB!@m_Zw4v`gyqMOF8tgva7N9b-te- zLq^r(``{?m6v%pL3VKs74pXfTfhxgiBoCku&?0B*_J|H=6L#^U5!&LC`XxcJb3>Z7@{G@}Z;e$Mue5Lie_yW1n`6Zz`x30#s%1W1SS;WZf-CAIAv)RvUACqI^fO=VLFt(e#t`& zJTKjmw&$sFOdABTnsldL>;FmhwL*`HwjP~=7^=$y3%M0+ zmg#~*kK~d;ngT(oQ+tC@w42{O5Sdnab9IvyUg{r9ag<#@07Fzh#nq#OhPc$5GY7w2 z>tu3rY+Wb7zX)|FHc@^46FKC@&gY%tH@ysW0E#F4fc=h3NQk~Iht@6zqCx%s@IvXb zdi+XaJ!_)m#_21pDuz~x!C2`P# zzL=KF6oA7!TjND&8{2pV(;PC%xR8~2K}~1STX+P8Ur?#&tBf9}w1TZr1~PD92P_^w zPaQA~;BBj3?V*y@HbAeo>VjCdQ9XjYTx~^FPTXc2>pdQ}u^+%RZ?LkyVD$a}p9bk> zVvs4rEXy^a;#fzwt^kaD9a(5vl-j+S~TRvo$v6- zGH*0C>M{?qMm#6ums441VXSod4C%2LMTsHpD%RqmRB8{ZOtC_(Am@7id7_ID(y@N+ z3nvx`D-gZh1BX(tfM-A}ssOOFb{ec~~Q#VaB#4Zx{ z*bWB^{79S0beR*xR%=NfCH0Y2l{1R0W_}uBM+9nXTDC(@BijMP1`d2M!TsOrz+z+E z3`xfapID&L3NaT<;O|$|db`3C(ILPD%iMLpBEYWaDpp{#2vWw_rEc=y21u4IH|U{~ zNGm1i@igV;1GXo)3_%N8EarMgiLmDNMrILiL|01_r-*8#!g>#fwC!gGf?U_?!%O6k zuqq1}f*IhK43eU~=s}BX@_R0jn(dainUX0}w&W-+h5l-7;{I>Ydm>%mkac?-=n|vV zx7+D8URn=J^_nAcsbk6-mWULJ675BIm07EBM3e`xm-5^ca zR6@;5A`NIqmXv<>v`U(YzQX3ubcH*U0@Z%-qg5L$^0Y_5=*Z#cCXp$x1nC7@pv!aX z7n+@p55*zi{O+i9NDf-;fHgUofVSvMAEqp;Bn;3_05~Jpkx+|_^XbhrWFulh!3DnX zii8e>MS6a>c9MGY@&{@R0}LS0Au>O5C>bR5vrlL>JAn zS%uU6gPBj$*_U3KAevPXq7JzCj3`0%ROx-sd)C5n*s?5$;G$05;F!Z{-YrGIN?X5* z%UW@QU**YkpCHOQ@<|-eOZx>jKz+Hl2E|s(S40VLPluZdxCO*+lg zoP>xwujP>^|AniE8ImHRnmR!j-9q)JBa@Z52}Ut`_$2GZ1eEJ_1hXWW@SoaIN1$!# zc)v{zR_?O?7c`T-hjM?ilMt4x+*uSns8~5|xBB_a>Q3|QIb+X-v&vN68k_{(++Q=Y zBZYpc`87a4AUAEN4!SQ2h%JMegtDoAEdw!FDAJ3pm0{t&Ocb5YkwcteAU1Bq!R^#{ zPV3=W!f{C_;gA~%d4*CKkF;q^ZLYd$@2nTU&BMXQKQPH2b&oU-x{b`!kZjKA`l{Og z=Z+nXjAca`IzTXU%P9egOEsYJ?Ij^GI1%CEawaszm-k!6VuqNcv<))(sg>T34WC#597CUbI~- zf7_Hys3c=t!GlBeCho|?F9Ypr<6uf4c2r~M3AG#BJx^K|``Skp_7169bjmgTCHF7h zVSV7kQFC5tq1V5d>Ait;u{HJ715t=^zus1CBGZaOf6$soS`R0*L$W(NY!_NKtEov) zc8Yn7!gCPK23--v_nZ+=70m2r`R>#6<*Y$OQ*NOJK#`_LUH~}U5^nv+_hXVQUr^hF^+)Fr9X}BtVua;jqP|-JNTrjbZgH4vpV@9X`nSE zb%(XNvuk1>)68s8N?uONRQ+hZpLjH-2xYjOXP#k%DpMqSZbL`$#O(V~Q$|Co6sA;} zBR6GK9}+pev0>Md*;hlmH+!VAPpe?FUHOV9Q7%jCsawb^+4FURc_1u0eT?0JdpzI_ zz=hv*bf`6D&S9Z$@b71dyT1B-|8I$lwQl3frgDb`sIL<$P*=5Cz|H)2E4Yzf90>ZOg6G*p>$vT8Ljjkt1y)DYq zTD!P=a%13ITX-}RiWcekUS(d0+E=BD^*qO4Ul2-yl_qGOpTjqFM}e7MZDtU8)g&6< zn!HB)v$|{sVvW@QCFIga>ZQLM>{ED#!Hh%^0oKJ(Lkd(>C0?t4`r%=-8mhqwZx(t4 z<9%6C=VN2eey2v-L^>BBL2G?=U8z@^Y6kdb!)*(yZ8>7VAtDU3y>s@yBK`L`j|X|w z<(pZWIsrZj!S+?JE??Crqb_lczMqtnPQEwc)9iShoiQ}|u9V2S2Zmw7?9D)FVsn=3 ze?lb{(}<|>3Sq#|+z)oo?^Xbj6UgAQj4{{t{! zsjiLsuG&^}6py4Q(4ubId`ngf2~a^TT{$F0j+6UP&nutxBIiNGvYHU)7!;ZWvPNQ2 zps^T&D-dRL+Tp2&)n*In3!H4jqAMfPS41V$92GBMZ|MueuYN&|JE$MJgFrr32gCt_ zrwZSnch0f|mm_~>fv^*WNPoHcJ8GO6!M>(Llh@r3SVY)s;qfts)`9KImFDS*7{Z7o z>I{G`#mXfg=PYT#W;!qO8sc@h5pnQ-)UbXce+q=pm_87Mep3KbdW{}t$r5?hGI$iW zKB;a8aAIRQo3&M!d{pCsczSlYVp8#Qp@s)_Xe`L8|`>lii+l}!Z4XFT&BL%hYRq8vGq`R%8>PhqD(9x zIBa4!A*5M4(^nAh&rDo~opKqmb@I)2vhS~RZxZOPqSu8`g<3B`vKA?lVQEn=Oal{m zT!TcI6Bi>fgH%`ea?R>jdib?@mAHHv~&tglD$TV1H@ukkAk)rx~P*?S-=mm?=>4Kp%Q z;XL|HV!hoCR-;#mTTgfF49U~iXw0f~Q{)&Hxt1ISNZDxPye5Bi^Cul_bmrKbls? zA&I5(8p zjjLh_S@c&j>oyFamS$=TP`h8fNwW%3{$bvf*7=U=rY4#q`GDHaqRA1qpd+W~7~dIB zy*aDP5KtMd4<-B}KOhB6Qjb;8rLHx7VZs`5w$2CH z$$4kLlT&~nFo?+<1P57mNnX)&n`Dm9`BB%XG)9KV@Y?yvJf{Y@ofl{VtHD>z##b(t zZrxt6BPS@Q>IdC006@mA#(^HI&2h9hR^TOtqxZ1 zv#eleBTSi~IlmstT4^Vq4B$xKWLTGr{1&QBAMHCaRND?e+B1n;qnB6tZX8A~gmpNz zKTG;W=*)xH5H8{G8s3_={@P7b&$smG1bf?nSxpFYc%R$8XRsRy^hv;onQ(GKzDL6E z7J@Z?A(Q>Y0E?=?S`l-+bSzEiPj6!I0U{OBWv7@&n%T=(AVuy-Hg# z*f>a1a{m$@g~-{!-tlHkQNl;tHZJ3V@3TM|CAZh3%xubZFnUDRIfm?e%96T?%qhxcIxqli7@(xKt<} zQP^c5mpST?+ueA`Ne!Yzj=jIyS+xB(fR2EgREbkVMVB?!1csM)0)ER^H6l&4zkn}h z#OmStulPSvsz31P78iB_tH{K0?HcxtKjNXme`|^Qz(#Fj+fwwiNN!eZ*BUU*BOPB2 zzs*m_egZr6t3&0I@$jsV43)Qe+yHkV(AGX@U+<8fU1%MOq>E(Bk((!5x=6p;6$D%= z(2anl;KE|gys)~1O^&sw#ED&={t+tLG5ZE5sQwa-anN5R&Y>-l^ESgng||)P=AFf0yuI*iK*txtu*W@H zr-@o0nkGGurg|+cI;8t0M|g^nEIZ|Urbcn3yA1f}Vkek3i-zT3Ac8GW5V@#%X47ghPo$E)rpe{ z?l$d4BkYxvi_X4VQXfo!(qF%yHy(PLNVvo3W>`$poOu%I`~hnZU_akF7(r?Au*b3O zWdfGP{~Uo-;<~fP@{!~Ew{z*;CMEH7-G&d?jTSmbYqT}cDBjmz3T19&b98cLVQmU! zZe(v_Y6>wmF(5D?Z(?c+JUj|7Ol59obZ9XkF*Z0j3NK7$ZfA68G9WQGG&VL0FHB`_ zXLM*YATSCqOl59obZ8(kI5jphARr(hAPO%=X>4?5av(28Y+-a|L}g=dWMv9IJ_>Vm za%Ev{3V7PIx@A-x-MTf3yE_DF+}+(ZxVty*G%f*x1p)*Q?k>UICAe#FcX#KKcOTv7 z`+G-^uIe=(oAs=@)~FiQl%&dPjH2d_ra&o22M{AG6AK?eUP0B--o$~MQ59(EW@q98 zU}Iuo;Xt6I6n6odfUF%IBuqd+J^(k!3ZP^L`Uw5-U}51!pae(*9e^$$Mst9v7eE0B zGEw((0VO#T@tJGz1xO-)=s%s>ZAYX>0hM~Jwilb4INr4{IJ4t7SyznT6?{C|765giHb6s7QdJEgt*WG< ztVYZ9kyy>m$;r{>Ke&jiscT3x03<{e)g=KyO$LCphMM}{qdL&xgTEyMKvDg}{9)WXHl{+|K>8Y>XUiI17t z!^4Bg(#;jbSAd73i!I>e>;kj{{?iyY2lJ11f~Pb8$y|`w!Nx2!GorVg1o3kfVzi z^Z(Jct%IY7gU|oPEvy~PE&g`i+|7wu!@=6w4Ja$|AKo7j!hd6yKoEch0CWZbJ?UF=Ql{>^6XDrM~nG*`9; znOXg-vH!wlK_(yC7j>|-1Aa`(Kaj@XhZk8{v=M|D_9HmQ_-a)>fwfKbrZ+P13>4(cId>62QjE1u$`OG4VoR`49pd zCnvy%^+VIaBppRpRq#nBe1Wo`bkkN*!B z1rv~qwWl7-hm=@9@Q=@b{xbMq1t|Y}75tBAF)>F^A4U!yb^s$AFB^cBlk4Na%EsaM zzhce)wXOUkW##G=aZIPR#I_IvG)9iZdD}W4qu9pZ(XW!o^h;Y-a-cm5 zW?37tTaLym$*6gRg&1ls03=@a11h0liCZFl3c1+$5rbs+W)+j}6Xmt$X*_@yWK zK)@~l5zLqlHik0Sf8ZmW+GSS10FS%z2STJw;_GlFB*;;>l%M$$!s$TNJMNZ+hP z@T}kz{v<`kO8zYeFaU+!@huvBJ~5G91uqTG$UQTuVg{L7ZGVsMcRM?kj7c}G-oSX! zBX%Yyd5h+`?X?61YxRW0p9-bzU&yQD1PLyOqDPS9GV9KrWq-D6X&dzXQ`hHTrxktU z@(ESd16ny(f^ehOapW_mrEdWp03{B3VwBHtAcg@M17WJ=H36Hpv_d7_vI&FE+An)oy0_9?DJu!>+7qOGd*Y?4r64C52_?16cU%H4fDVQ(zJBXvIXwbiB0$ z_L!8P{0$Yz%Q)c`?GUS~u0$FHc=j&;z`Jbt>%YjhP_U5%J3lo!MLtc=)s$fa{-jT7!RKb*MD%s=>`h;y|#ZlPO z?67e`o9q2+(S3977=5i4@jh>qq=(px$hsy$v!m0~vL`n+^rXQIF`>(h(5!?wzPFv5CW9u77DZoi6vvA=b_6KS2}KE%ps4)c zAyIB(B3%0D3;3}<2az2xKJ6`xEF&>fMK*Q5(@DD;B0k#Qg(s7lTJ=xj-!r4DxDSlBy&O@?spY<+M7#}wF48SnhE$Mxhl=u zk0r^x(Pqmft)f0WMw#skdMl-+37|r66yP>&3Y>GKM)~eBR)R9{G%t06JEOUWE3dC% zP+@r1jE=BlH8ra3?aBZ=_H~W$(-I<=+~=&#@&V@!`pLr>1dmrwx5+-r>#@oqlVCfk zojqvhL%P22)(0cY@vMf(SrMyo4mk-b(_rVbDpWz3Zx1_cNA-^{bk@1u*D|5aIPsa+ z3kWw#=LE5*notU`adawv`qlwl^FgXx8mlDfdx*=uh|Am2gK@~*hyERyOqgD1v{vHZ z`2t>7wH^3Wl>OBkdFp5i5y4y7X&D3yKfwlWo+%Rn)~_y>UHR9d|^b?n_>&)4E1;{YIca-0vb|OW&u<*&v`@qyClnSsf|lPfg5%dBI(2jBiNK zWjn;AmW7q-b1R=V0}*yBLZycL#y4hh58VEM<)uAb9*E3Y8#t}VCx^=g$!JVFzZePx z!_Uwk(bKjQz#TRlB`MvpD*Al|7Z!tp6v=!FysJL|$Ko-tRoxn+*~k5yz2(J7=D!+L zwu%`+gHdkCV9X-onMAY)EIZBjA(eOil@e4Zec}9GJlyk2Kx-;cO=JDeFGo3*OFYy( z#7Hs|giV;|pAXyiJ?#g_u~))6K`D2t>FG13XEAdzC3+`b7>_qCXZ)o_FW(x6zD+Ye zX|oWS^;ep*Bt9EcqFZ2BbbsBNKt-$Cs7}0Wlhxu_Wx>Y>TuP(5TS0{D6r{BsoR z{rqj&q}0pQZfKBVLzY(W*--IVyCVyM60p5f7&~}dFDVN}2Cu_qQo_7X0tF-14 z`3q#bK#UfIQMNhv(aoxrZb+|iV5dx_L^qS$3v(0$7$Ftg*B#^5+Q78k=>a%Qx*I6jQyX^OdC**~1 z5-(<|S^Z?$H#d?iX5#katLx<`xNs|f>a~H_n8gbU0nPw96q@YM-p=q4u&{`kxmJly zp)TD*4R~VbnwK19lq(hVlM#EwgEntLF8_P6EmF?5;Oz&OrfP+zTv<~KPwQ0 z=r2~DV(s%;uT)Wk|C#isL#WGqg%B#n_=sC#|6ag02~5*^O!T=z>(|-=5Zf7mM0Aj! zRH(HL#Yz!BxEprxH7?{LrgzY*(q0ULf8$)K2}?UkCMY;-SGx`4?5}gmeA^^kP@_C63R{ zpW*C0r9}`9hP{fCd-JcnOX3tPa9%=@hVw*dpTyQTn1&{JYvFKMiAwo) zv+EPIVYF)07pawZ5kO<{5PT+JCi9Le+BtLRn9Thy5N-%3z{7~KFRJd2u6}Y{MyB{865g8T3@dpAZ5c?E z&C+q(hqjV4MdE?d#$=A3l#eRUn}kzUl%xJf%fI3DISbnheGl=4vlb`PcSJ2`27qR9 z^A-gm;_Oj24~Nkd4Y(%^)fMphq@9kampzEii<;4-!X0<4TYLNEy$bTK9prfQPIn<| z{r}i$smR@FY3ya@RWq?GhJ;M<*k6QxNF^QAFi-C5WO{pqfpXycm(`#S4MtPL&MOL$r= z^FTW4U7;Rgd31T6u%Lbmh42Mxpz8QiEhqAVmSNhkG8NY@R^_qpLjuPkh*DOd5RmT z9@&&^8GCMDOXK==G&U!JtOaK?b}XgXzwNbVUj#6944?Kpx1RUb0uaFs1C|-mPaPe6 zUz?>8XAMGvDmiptlfxzYpbEMOg@4tvET`J=E(3S1R{0`Uq2ZC~7A!=wHotTGOGGbM z1#wcfThyUK<6kIlUE4NGraOw()}85@X_4UV!+yd4VcKt<`U9ne_xrog)rd_q5pKU1 zUrgqDn7OCzciZ{WRY^T98EAt(E{pfV>3cpHO)?uxDJFj7Oe9!XW?>=6!SXo@TsECp zGt_iBW03LzrPK=q_qz?0h zuTYZP=%UxuOANY4&6&Y35q9(o-ruA#3N%4ai4tEzJW1ESsu~Y!riuH+ziBL_CyI&h zWHM2^s-tAACJMfS4<8qpT+o={sz1uMl#d(}KIx}WKs&OZtU1zh>1s)WGenOXMan1e zZ#bB{*g*6Tk%*3ca%Dkaf>f$90*3f^kL+ea`SWVBWDb|c<7xb83^0^;TWA(@6isDq z&lEaCx~Zp;!_iY^02-*)KIu02%zsa&2D5TsvAo~aJ;jR z@2^$3p!9vW?%U~HT-iMw!b@VN72bD+Hpnk)C-@VMJ#g46CXG_Q&peW!=yeHENF#3L zV}Er9v#WBZXYc!X;P6D0gwlWb(X%p`g;UojgiZCrIHLn)OXm=ADHZK$zpEyz+gUkuCu#cSpcUDao)AV) z_0h@OYhPFirEa(2I-%8l_rICU7}g)l20QXEZ*Sr%zTfYDQxOOZxweba^o?9nmapJ) zT|%vxIn_3S;2z#W8L+$z9zk0NBvPJ<<%W#e-2% z1Kq0#HSUmCam|pyy-Kl(rs3V?Ioak~T+bC>4|3sXl;a-#-UD%`?}vYo9SEHf0>I<4$%&g3;x;f-216%y#M z%ooKlz5s@b#6gevx%-$arZoy^&iU;n@vQG$6vbOWEP3|acuUM=JoNyCT>A4 zS3X@+=rRn36tEYPQ8c{V5pQ=3dnE@JEgR~$XDWlVJ&tMN<=B}KIel_;U>Z`E6Tcqv zai^uOzMuV2&bN*l)4Yw;LsUvO43RaVMP=z~gZY!!nu8#20(UXrAb1Z4k0xu&ek5lg4Sl2J-i#@E zSu%Aj9La8c@_B}}sJ?ZqsC$#C&GuW}CG})NwdKG`h1et{gsb|Bi+o%)NJz`*(JMC6 zF5&dYOibSpPieZ{80%(j|E!%=!OpW_2>i^uPml7f*KqecDEElSHH2;Nz3TSw-}Wl_ADX=K zDaU7&SpKm6T45TXn#cG;{$J0R-g_yNyKn~icI;R7;*g0lsfK?Ho3^*1mxW~osLv}G z2yS&&*vo>VGXfy^%Vn4&<0~{Sr@!>yL#|%j*!dPZk&i<&u_aa1kiC(roLaJ_-P~o; z^xo$;@3a~|v~`_2SMo{?;n5eTr=5?5Eo+n-B1}#9{>=4+1v$&#SlK?$TY0%BDZ{7Q zUF&mkJ*HAAj!*ZJOop*;lY^z zju)gZkY$`sCHg4T>C*`*7M~;@WaQYgVlX!xH&;W#XEE)I$#iTj%L?58Nrj z)FPAX-*;(~NvFWbqbQkZx7OR_W!dc;Dzv$z?mq-!(R;DA;hajtZq}!xpa(ov7A1J! zFWbCQj89s>B@rC5g^gDZxoP1eu>%!s!<4*dY%4@AdsEj*xVUm7y_E*{M{*6;7oW99 zRm?p|XYa3{xv$M#?yBe`!qys!7nAt+N_h}|YVEV-a_mBgJ1dCYz!WC=Bq@wu8Ib-^ z)`c|YWcbZ^bFv<=iZT7o*0tNbm$9`Se*9W+mN}NR)ztPIs*AKR#kKF6!y+_7ddlH= z&KLPdar=jsm`-e#@OaMV!z5-9PB~1n@b>E&9Ox*Q|jpV#^3Y=G&8*ub+3cnWY{?`MvqE)x(Bu zmGBnmpR^*5MFIhqMTpP3cIMsCK`rbuQA2*|+8#P6HZ^pLRs}O88|A|1_=)6v49S6r zV4^l%w{a6mJEXjT3-!KJx<7Cp zt)DN=xFr=m1V;~n?`!ZP<%O@{S`K;-z^Lao)h5`WXoyxOJNi@m_V9&xdH8NhOt&&E z;y)CkfmXBk1)*4#;XLQ<1q=0*Jwgng_!83GBtKYWZQpxeES*!8jo^h44-)C}7z%_Z z!;M33%>69o&iV+1WF6Ofx^fn+cazjp8 zG>oTfGQ;Q3g0pI7dNr{^6=%?UHweswsx$Ucn&BmNdSRbxR{ zuj|yI+Cq7!1b218jN8@pJG^(ao&I;OgO2d5lY2Ql9gXgzmV#OA5Pvdek|BAVH7~00 z89D+7m<(=HXi>8J4{vE5JV*fQh;0yJ%Y$WZYGCO8gef1^xs-1hA|{u$CT(Wn&+a1s zmG4!Q)kts!wHlo_Sci&l^I|CEN!p!HnO>~R1>jUv!Bn#;5w$i63lEaM_7x> zI|#^x&Ed{{A|`v`cQVv3#M&x2$_-jKg`!U>4WqkiGV~=Bazjwc|03OALCyGk>NfH>A4Cek&I}xF`osXtd zqaw6lg;a+Aqi(wZUM5bE;ht`tmcI5DpP8uJg=CMk3O3bSd6%KF6+_5bb>Z(RV$iDj zMl$sKdffI^FxW5u7O#nl(mrA354Hg_zF3FzXeBs(iu#tsHfwidPc~p^y9*%t&RH~)NU;uj3~R| z5UyWLj<#nzCtd3IOQj&9=!LFtbzCr5jmX3~iaycB1?{X`uW|yw4JFBmf^bTQZYZgJ zRMpaJH2ttkMfh!m zh-9USZW?PURzrb|!8MN-0u)`kNr;F#?I&xpVr~xqDS+|w6x#+akVU9|n6SgTI^(1r zewueglNvAIn_OgzV1KmRpSW7b*q$u84WE&fBD_4WD1!dXrLo&x?$4jFwR|-zSZd|r z$)U2|n>bE!-Ux^cd${qEpS&guCS~^WS~_P3fF^R(%{mcUv5TsomU`DvbV8bg;y=d{ z9yEU3aG4pjTdl5rOM0S!^wO=tJ_2ygub-Cw;e!=H>o>LkBMlvNz<};ET@e0r6=+`V zj1XS=_&NN$s@12;T|v7zVvp6!4Aq!XKSt+ID^h z*D&png(_lH1Uy!=ln8jz2x6qs?f1vPAU zp~ol*l|^~^S7v07;Dy$COctRH7L1KD0J`{|NZbvKf}bALd|!5ua)Q3fDh43`oM&c& z4OiSKz=+sdoGUg@Uweu|{kbPJ9z+XvzMWuK%w6@yl-!V=&Q*fo`z2LWzHJm3paYXM zsjY7lxsZ%P(5Mr$B*ptHqvk}jUrfhtq^eEVPBV~%nV@{O32g5B3mbO^fGO+1X-=qD zyqkm?Dlg5Mir-hDavtsW}44zY3`K)q)-N3>4`&? zpB3*p3q%!u>V1N3RW(*C@9%myg{esth*#cq*Q)ghrV@aoAAmfyd`NT@aPvr+CkVg= z?-jufdN`nJb29!_R7-h~)wXNsFgLr#tzN3@*FjLBH1cPe3F8|}gt3z`$8A{RI?L|w zP|L2>0=GJ)?*}86Z65Cs=?)0ul9cr`5Byg7`9@;jlDCRj^KbCAZY}D5U^?7y1!XyaYBCYDK;<{0x!S$*%5mcKFw3(1k~be2=|MI5yJev7DU%! z)btoW_zPrpO!I1p)c&|7NOBE0zD)6&gS8hxd|_J^G`E&)^qtpFa2>-Yb#59lmnJ9( z4mSnYbPdI|B1W)KAPc*q;w=tudI7#>od5ogM9*DiO@0ZOZ3~^J?Mq7OQ1j#n(%o;$ zm-3K_sOXXLe^r^yifP)Dd>3)Jhy`T-im2?z^Z6nm4sMI4T=FAz${*_PJC`9(xfnO#`1AmN@NA;z4#twAx z18L7$0V_P$9XYjfGuX*&*3FGc!yQnbXRC~q>n3JYmp3t~h+e29Mt4a@*IbTc4ArT= zqG)Ak8?Qtd;N)rD|0?oFq&S2XX?|`vos;P7=+!0hhYTM9lV*qg;_MvQ6j8B8Ytf%W zCup2k2$rnw_nPzo1WQ>_A31q-M6>2(7MX~*t@_Jjf{nWX9v<+=h0Bj{ha zQ)RO4fxn}Vz)q`{8a!@~k^-JL;+$|q)#e^~!5UU+u8}pT)>yvF#(8hZBjxGc(8v+t zHx=LK231)2O_eH8Tj?PH3I^?U1^V8j^2~qd4p%Grk`5cyp1+itvZ@1RZTI%y=? zSNmvKc8y~Eqq7n|8$I)DI@Rr_S<)#KZf-JPYAnx%9k5-{^73uuSkRRq>9c7U0iO4z z>vp&lVUh3NPQg_nS#w>J`}q0zNq&`dN2KT%W?a1qh5mI;28Jp5zsSEn6h2vgAzpuW z+V-aU-FvkOZY_`ibh2uFFI_vq@OXTF>HLg?iqXql={fC7efz1%$VFAQ|J8BIVvuul z+-t{&rq~|60ki>N=^kIy>)0R870rjx|82f+V3w@v1~ZJ58g4H7u5t1JlNr=H=!bk4 zyiovC)_<@FZs5Z+aPNoF=4bB;=m)!@w>iuc&vM=IqR5rSD?Zs+f0%if$&~2lMa)Y@ z-gUu%a3W7on0iWqO>=O55kRM-vMdvW1U2ajd|PQk5}dM^H!F$7I-h1`-+B%bCnI2r zs2-wcY_;i4$a);XV2uVe1Qeh+A*Mn%QsWUzD0dAdOh3gaX4iiDP37jHiUr$8BC(Zb;i*>PQ^t$(AIwPF!#NQUm@0x;F>l-3Y4 zk5G69E>tM~>+bcox2CJJP%`zjVW7lH`rM)!-vT_gAMKTLJ6artyUywp^See6QGOkb zRs|Yz%)LPu*`RMgRccOIAa)aFplJqD`pLCY`QU9BCjly(G-Qg@fO$J0JntM_6*B)l zrwlx40+IA01T8Sv+InB_f*cz)oWTW&`m3Zo>@;H@hHsRC z8Rpm2jghS@&EM@POgvdPvFM2FZIj=|LKU+VLpy_C7$o=$tF7aS7OilKmmGzg)#YBb z^45YAPVp|u%`5A(Y1SRmAlkRZXm#8NvhaygK`1$ooumu-4!3)kQB#39Rbbcbby5q_ zRSmN=a&96Y;gyfX`Vo!;q1PF^GJOVVCdh!BvG_#(mTI4Sq}zlha?3$kSC_SUNy z@H(2h1lwo-;q!NmPyu3V>bapB+Okq$p0Nm`t;mMcP3U(!)$E_DpXE7pY6WfpmPB@H z2f_f)1t0Ers*Uy^n=p6L1RNNG1%ELvjLqZ7{8q74o1mwc(_V~KJ*Qbxf?LGmhH62V%O^1Te~6Zt+l$Xc7nq@Tx` zANUi{1e-?pGBp;4*xQ1rI@YD35&=&U!%a7C_li4uFN7Sn@2OMF=#ze0bfMZMmhsG6 z1#s&=Sq-<&PLkSED2kM{wxGHVQL+~MgWEO6QY}FS>?;##)`aJTrYiXa$-Rr0HkfqQ zRdX}o1xNcHYaqCt>4$#j6$)5m+`gPy3Kg4x`Ru)4SwiIUk`r@)!Zx3CF8)NlfNGR$DZTD)Yem}$Xn`Ls zZyX{^?edVo$bqj46y<-TT?#Y?S||`*71+Ll-x6Er%?Z>Su1tuiTFVcp6tY8u;o+GK8TvvOKJDc3a>EwV8w=Qm*|H>m5JCOMs36%!h@T(y;%Ef0RJy0=XR~gjT1axE# z0TRZLR3366fIYnsrYNW$EnuV|heiU8InMKB;eg$Sd+iUH_#MI4r!iI^FJKP6=eh>g z(l=o~AtQE24P7G$rKx<``xaztPnb)_$t_*Pjf4 zAF;tGX5cgMEo_iS3291Tb*u8?kvN5{D?nTto>Bghq!^zcFV0^Rz<2Va`Dt!d`TMG| zDCxVw$O9s|a^%vO0qpQUA1vH)pZ#{PD&2{qntj%6*^^lZC{oKg1n*89E~o?fiJ$tC ztt_mAO;8Kk=cV0ixZFL~0osC#dDCB5mN`ampxN?v$Tw+jZv`fGkLk=-I2x9aH;Gcl z(?Q8p*~;GKznaQ8Hz=3^fGW~2(=d1@ryXgbSHMMu&}@xfrbVQ-Fuk6$OKr`f$ z?Q@5eXiPVU_x@N!o?Zzg0Kjt5Yii2^NCNaT2b##6418lT4!cjFoUNw`&@DO9y!k6e z)N0*<9ovJ@(&s=|ay?ZQN&MlkY1IhZ%gQPuZy~Mf%7hCUJGaOw$dW}wF;TD?K zcAx612S$9v6anb(!wc|CBqvj=4Bfr@YSZW|TiPM{%fytpbWxrnN(yNBw+iAmU~$WV z<7@LWywkASZPYZy*XVy&Ps<}>JjX-2!I0khV#z1)9e6B`-VNr2w_kAf{94Njk^vUc=@rHZ z*o}v@zrg7W{FPw!*aYrr!Cv+qU4E*1uW*Y~_vu6Q2BY88=J|cMRmfZYo4d3e21>eW z#?-%=IAr0*B6s_QT*wH2s)WP|A@DTaKLmHYE|n0xH2A7hKkARI?N zuH-7qfXFoDplG~d7>rL+%S@S}2&U1p}1~MGTH>o?Iwahe6 zvBGn{3#rb!W;SZ?y=+;Suy#8Rl@DlN<22Ed7DaFQ+sbN8M$Jnb`~Mlu#HgnnBX=9O zEput@F(xZ2VFY`aUH#Piv6`-n#;4j6BZ4~U;FR@|x_9)864|P%Lo%^h1xugCL^BE5 zcul*~MLSY*RkTw#)+%R&CtY{11ctTQK52DL6rkA6FrSb&FF3=fNic?Fx7Nfzd~Kv= zfBL;lL_)wVq*d!>s&&z>%o zQ)QS|j9e@()ZtlW|4_T7quQSr#mQM}v_eBx$7^-?moY2tX|vHj!E0^p!R{gUg5ZHQ zc5Nil^U%mHjgot71h#k-90_x_-&Ux=8B@d+ck!g>_c$c`jy9B3kzyB7`+cKU7}17h zsVjrmg?X6pb^6FnU1nQ#1I8egGfU!}LhQB7Npl)eIe)w5^P6HR`R^bP!HE{hX!U3DtgZ_dAuu} z>+{!==pa$zYawu0CbREX3HE!tu}~65Awe%qNJ7YYqu-J^8Un|19H6lfX{caLY#=6D2<)UT#bPAb=iUe1N>ZzDBk&q$Bb8&-(v_dkXDO9C)A2#aY$LIM!5{iRCu zk?1j|*fZIf_&>fZHsR9qq=>1QFv$7aH^M6Am?Q^_*@Mr;ZikxvTNkL>@Nd=J)%oZs ztl1(3;un~1cfH*mR;PXQ^t)Z(iSWxE>X=QUG=kf0f8Uq5#PFeCFQs)4AW+QC+HOT? zOT=!{Zg?WqyzqchRUh_T56$Y3SB=qf9CI-=y8dg8CMFcR8aFYnDaZH{BQREqf3_h9 zj34+wtjI(v7bt;nD$6z<{%_{(4T01a z0O2&CT@pTU8DkJOXphDOKDJ)n`Eu5hqFQl|li+u$WGC!wdM~PXb~&w7|3f3LYSSF` zj6ib!*tg5F^K=EOV}he>xh-GO_7`@Y-5K=GbLiyY{oS7K>X!$FvRyCq$W`dE6aU_O zu?wm)*c#1@*|AQ}cBkYQ*F$0?=J!*^-9%tD z2K1G#q17zGSQ82%B(Y*uk%~z&Nr8aVY*&urgdW6rDsaJ{Kq4RzLbrHqtUVMH(}Rop z^5E-Bf<7e8A=)Cma-Rlw7~6**dn$@McL_6b=qhIihr9E!#iA0crr?bn2*gOEZIf9N z@`bkI%8EZ(da$MjiHN|yCX#BJ|rUwMx4QU9~4qp!@f_}Aw5 z*HkM_`NLtP`(B*|!pAmhWY)$NdCFI>HRYpe*MTK5_R(VV^;O&33WSQcjy@4`$i7y1 z0}Zc!64%~l>$v={F#umsYdwFj#+Ja{#y@}9k8!g9yI9JD#JeP9)`iI~V^@7H1t$I5 znCbLp9A4UCIXSwckd~0L8x2)a(&Kb(-k~)oJ@~eYlGsN`XD^hw>qp*~gigdv*EU|g zKFjT{vO(qxOkA@r2e9xtH#lo4#kl{mF+-kWuYdP^x>?#E;DY=$DdtzUidnWVQ4*1m zy*Ri;sm+RJHgGV4$Sh6v!cIsubYVVoZ#o=%m6NV>toU=~ZKS;1N2XwJA6CA3QKW}S zgFcWm8+09v)GXg1ikJWzXr3U8P+}&p(278it3j^JF-_TP5j~ps)^tTiYl9W3l4|gB z)^Xu^srA6tqxV8ZQxe1aFt-qo3Zc==&?cL*enkNzoo*~e^O5GAw%*l|JZo~zt7I=2 zTtcJ8mfw&f%!hZpB@-^yJp2JSwF&)YOYWrBDFwjxxA`soBtTIx0?m=DlNk3SvfpYa z{`YuT*@$mN2TL_s;vrZVQZ%2ZAC91+!#RTS0c*_oI$gXVSM9k~RSUFRw>sib69qHzrdDx+wPJE}ao&K2o{f!!6qJf<-( z=#mq^?w70h&Je^ep_j2mp4GaWjOtmV#1)w26K%=1nL)zfd(hiVK>0x*kG5*{4e3%GN zH25cXF746sW!Kp$$@Szq$8-) zo6x!hPacNr9?98%9)ut@s!bS&SFFw3I3&FTMBTHOH(#Z07=!PWafFOgpMY90LS=rd zSdX)K_Y$^;Zd<|_*NsU22r~}q!UrBdM|=!2tebf_yro*FTro&J7LO@@Kh_WADTk;r zxN)+~gl!*Es8NlsmbA`}DgM;kx~Mgm#FDbKq&**T`?4pye_XAg- z%I+>xU3g6eftdxhB?Ua?5lhN>HKI=GZr9l$Z0Uy7?|!huodzbw0QG}%mkpQTiv5DV zE=g?KGyN^Ss_i-s8BN3;j)of&rXZpSkl;<0{<1ruG+fduO~PX+aOK);>-M|K+>vm`vpGD z=>D$9BW8w7n@4@T^H#Yswy?!h)%t6S#Cu&SKRbQIY1@HrW%$qGTSB94EhU50(s_oe zGNs6Ay-=n5pGRv}E5bAZK8JwvSjXRk5E<)ZKoU9R`;UkbT&+do@LuOhoB;@nwK)TJ z1}IW~TI8(ta(z)p>+T?Dk*6x=8raV|;Md`fVUu3pQV-(t4xjZ*6Zt90QY@z@;}YrP z1ZivmR6(d=ji;*vr{dm`D-`1qA_pbmh3SdIxOf~NVt4QGO;}bl86yx%XHq0LhN%FS zesZainXw~|9s7TyB|OP!SOq))pV?R}4&vAF;Eh0Oz@a`CX-D_DbWottnN zkccJZkzOb_Dx-r{KMXi-zlAh4ZX`HGw#Sw6m?Pwt1^seOIUR=B6FjU9-;u0FgB>ga z6UrIqm7pjSA0hZQ7CLo#3Lmt%xE}MC}UJtuczG)bgQ0Iu2bBI zMAT;O6_-pZ`xKw?rgM*mKSzXC-khmt`9X}JQh71)I-I8!p-}cp4yP8AF)Y=2n^I&J zoA9J(WHq2GSO&-QS-IRbF}?=)HF;Y>qWG=}r0^VpDIlgf+7%AQTAE#EOYB`eltSLx zH^fZxumoG9c`I3R%Uyt;5}TFUv=xC9dZXCYaB1=6FGPh~M?sGCau+>j_3zq|_q>p_=Fb;?Q z#t?5$b{!^FU@@16v^971cuL4 zfmq6fp>%Hjd;WlLJS0F*+^@;ol?A7Bfbnlv|i@YE`^G4;Nb* z_nGN(OY+F029DPZn%^`{N2f!m*<|c`Oyl}Jh$;-9SgpE9hoo@MSW{# z2Qz0BqLggia5n9MhqJj@^S3AyL&Ku2!kDS+Z1a4*&gu5!JJF(jlxB2qDQ;;MV^v#H z4@SeU#z@ha^H8p9P=`Md#SC^%M#R-wX)W$!A3Qc&ymY(4?U4Pm^<2n{UO_vDTtQN z*!z57kxW7`^BY0DF-~2__I3)_UGo(cWnFLxT2lRt8VPX5PO4vYC}{YWeBto9p8nMn zvPuk*d6#R^!M&8ty~E`v%BMFKz5Do!_j`-p?Q0GciS!v!jNOa;DA?lBC}jkE>u-V) zgps-kJsnE^V}#RQp}K$TXhgz&MQAe2|B^D!=GWL;b-iAOJdJwL|NcRGge*fBVIOj+ zGiPk9(dK=ZMBkX<`ka<1b}`xM+-eS)0jGv)?-xDy>^OEw^>Rv}-T`=2M=M>Klpl9J!apvx}oya_uq*&R(h>)cSSh_?;1DsB08L)%X*+4dOsT- zrm>=*rD=%M-*p(QnuqQw?nw6Dxn9}z4y8;E4N3)x#zX0P>1KIz%+h%p`6zi+5~4Xl z<%TmyspIsM*9`35Y0Cq{l@Rf$PGlTGW$z=dp!S3t%78mQk5IidCUNgRB~3++hg+!% z+`FurY&*I{KYc_PlSA>&wW8BaOvB@7L%ZV9+!{&hFwr3^E~%F^Srb50WEYvgXh?qm zhRo+5C?EL1d*h3rkLCnwl9#AH-M$2cY=!UueC+jlE0fo0^Z)Hi{4W_q)2Xc?FO_XSy>1tm87 zG7;0zhiaA&E^s#)X#@0%eXeEk7>$j zlMM9vQ4eWUeB6P`3@zkFo7X@>_qbqV+`!{rA>rp}Dl8#0-@c(OQ?aG`Z~ZpF1X-vA zKfFchujPGNze;LJZ%AZT!S}*CMv%yU`7Nz+Is@2}<6J1)*BIUg=x??E5D|Obc!_vs z$X4EabZ0M=E(iT}3$Lf$)mNM+(-R>pnNoJ00{%h)e_{zU>}~~=xemf3&sxt|BoXJ0 z3}Q4nFp4FIBRHHgmU{i*Q<;{YG;S1a0Xded13QSqZk!VipfR2)Wp#jRGt8vfVuG<- z$FJNTDk7>??i2{)Dba8pW)XZAA-0XMs|)q{2Npi|H3m_(H=;~_J*PJY{hIlAtL@+M zB*n;K558*EEzq)HrmG8!QtEf^UJ8NDpPNlgwF*;s6(#1urN%C}M~VQ}=hy4X=59!+`>!SPoGeC_5;^r8Ivyld? zBB5%FO-9_iF&>cqk$nl-oj$X!H_+=);`u$RBar-D8KFWhjd6`BvC4Hu zNwWa$E*Pmp7N(f;})38SQq+b+@V_6nm5Hq8murClqOEUK_% zCqX)AJisk5OK57xUhv%-Y*pFs(aW%DYI4daBGD68`RdVK-(vkItlwYt!pLW*}-m(G%3?M(T5JyE{M)R$~ZKJ-e8dw?CSAE>}o zyZ7=Bw+NQ)@O(g0OrZOoQvMNvldMNJtb1&<&aGgNh?Q}GW_L((kxAFSa%e2OBxrDY zQ>rP8yq2M$LD|-A0z*Ew@`X(|$_UG{dzfmE80#f~sl5X9v+J_MJ9+EjE{mGQHggB9 zMWiKFYJECnasEqM^7w6U>M~Jb0wo7|(@*`VVjiL(=c2>lG)@ISEhQa6&`rlkd)F;L ze&!+~xfpCD{79&s2HLbt1ubs{ZL>J-6fuIuOm^3sD+ z+?4JBl$600`j1AvI&W~%`uYm%XeBfA&FAGH*3xa6nbd+hYIe9(9}9wZN!86}Gt+Rg z)HS=^dIa~5(nTmuFgJ0HKt#EigIHM7g79C5CSIEtg+1i#*Oo1? zDEMn*E$5-Ok&$sn@ADQfAIOK%CDyOMD-#Qk>SCv_7Nje)R<`lfJ}e{N4U=Zdc3Zbp z%Uu(er=Tq{Q-WG<=A|h2-R?orya64Gh=>n8r)?c0t zo@?0By2t?WYJ*tVEw70w<21ueLJSWTz_7SCFgk(MGchqY5i2M}Z~?{U(B23biOmY=0i-v93M>qN?7Zv*WoCj0 z0uexh0OIbY{sSKkXRYQ8wV<2`-Jvsb#^jiU~}?P2Npa{BOsXRiZ0LX2`?_~ z2z|=Ouz`W`C;E9nGXxpO$kf>C;MCLxiV^q+fGOrF77&!X)isYKU<(d}0ZbcXGY#No z2B2C%CLk-SqX|k-QCjY1i}gU$051?BW(GlH|bvH-tp8I(StttAup5dp8=c`(Elf4 zYfUblTV=LDclESSQ1F9HHIO=BD2zs~)`$t*pu z&*W(cY6;WtHo`mslD)-4WC=C7)*M$pZUpze>By$K{U>!+f-BAPcDYkYA4l!Eq^fAx+i z#D8i!a2CJ}U>*>_g}uSxGx@g8?t|FugV>KgI6l26m$!daAkHCWyvjYfl zHjYjJzpjt^7n!i$0Z@ikHfIm}$5#X4H;N9HsR{W0TljnSIKT4G67aGQEcMh^O7=#8 z4G%!M023%VDEG-*SmEz4v*~Y=^uWlt*uVt3e-3pXj zT$?!iZ&8y=Qlk@CmS3AITjRU9{x3nA8@uPNmtlMx?A?#TKXUmuCV9qtulEjbioXiF zzkhJ_QNQyojimt$FgFLlX4}6a;Lh}&e()Y0OJ^H^ zER?^_wXZh+g(;U-yqcbJ)L7)g0WL>Oicte|PwT{>+K&PHxREYleRLKxytDUq|_O&%eZg zBYoiS_PazxZnt_c*Vi@xU}|`30MOjvU);uOcKLtztRG`y-_}5TU&s0Hb6h|Gfwuz4 zAtRjJ$h`(tDPi+>y@wyjrSm8}F?~}Fpsc2cg=kCUn-C5oPHp1>`A|Sv-^>)y*T*Fu zFFwgY0-{+c0~%~~)-#c)M{ED2gto+bAA3Q@4+n`kFJr6%F(09SOL0wD!kIWdwMMR$ zF=-!YhXGm=i52r&XTvd88`zttOqg~S(^Bo{?LwJpVl(2{&Tpa1-)}&mk#F{A?cp=@ zi(#TD;N?sttNl=(TTX)ACCFGV<7h>?O2KYU4O zSkpliNY9B3x=PvcV?AuH&>|3&=i z-EvAa{ElavQD0Q#AX@EyRxIcOEX2}Bs^K7{Kip{iJ5|i{W(ocsJ@Gp#Jr&qevj#bE zg}&|eTR@;=N?eEd8UbDN&L|4OZ*$zt@TRj!HTW=>WboX851awkXw5d(lFi6g*Ze-| zxdrV~F`)~USV?%No+DLmwx%wjj(B1F@4UafP-=-glvBKpnDxF~^UChZ6pTpR4!x?^ zDG8-AL=5tnD2^;vF-HH(B>LMPa(|sYLcWkm^ySDjDdcl>eHu1Y+^P6f%@|1oTZOJ)3CQedBqq0*CD(gvYn=(Nb+F{w>p-c?&E? zWxLZKY5?K|fU?VJBT32DE1Wn-_4f6XA_7Rx0xsh^75sJ$txIa6B!-S)@EIAGXk;I4 zTwhAO(o6?7b*I5uo<$dJ6_gqmq$B_H3{}BAb;&;X%W4xl=Tf$*6WitDRr{$aY+QmW1iLT%i&9@MqAii&wP(<~P-2cB)Fj}GKo!@^Ize`OPx08lq^U#sO(gLw@M!6%sBCBG>3&Wk9#mdXe zC&hrx#z|ESs+Mz)x9h;r6Rru9ZCWhx5Ugdxa$s~k}kbH+0f9ey?GQQBTrJNszjMGKhvRT5l@J@vL) zo0}~%qy<2QX&s$PlK<^0#>a~=M{6_e9sdU(z3jfLA8M4>44|l$@6w4Q8Ya5?=%$&~ z`fzj)*+SLK;eL|quYUd#6$65N_Bc2TaW6*^;Y*}aJ@`Ztonjv2kEO|h(<190?)*OA z%dQA+e~R@IHC=?N4|;iP04F6;eHe1vuUgipA2?9!XJV4TLw|3_X{+VLw|7{C84H7cVsPIB<02$RN-wsQ+@uUYH+l{Pu#-8m$MGJlWluzim4_XQ-P&}Vwsw&3^!uE^k12AL37fl2$`G^G8&qc(*x6v zIB9boGzmR~=>US33mJf<(r1U}&MH5$G>yNc^=WB05CxT2kf`N!QMA#^2u?;p5{l(N zP2{eboLW+3B=*KaDZtbp5S$8@Sg$)&BK})Q=&nUP4~Fq&3}lYT-&}j15e)5}o+Vw8 zfQ*r0oAiHa2*QT#U=CnZy=%H7Wp>N^+JEl(PGXT2m=uk1k(YVx!cwCsZmQqp7(E(8 zOkX3i(?|qXxtg8oteLV1VbE;PtMOAf*yq*s?`Fta&o~NeL3sFm{|`@b4?&=#f&B)d zx#Ee8+(1VhG8sldzh=@7&&8(z_+*>wzdqH9Ikm6hLl>FGsib8{Eh$^K8=*3nKsEqm zH%&E|fq8iiTduAMJMu!oU~nk`F^Tss>wd!;?fF-d{*c4)YGhL)Yf5nnfn4TgayBvAnD^4YZ#B2`@;?>#}bJd4~lY8U3Ra6yBM8e_0gT;xChi_A`iZe5vI7u zA_NjJl?3H9Rv}&rP9ctiHODd*R*GUbb%XVJGt~MH-oAy4QmFAqViM2oa4qY6zvm6(-H!Gj1Q@t|&-6#q? zdt#Spy+xjX#28?5y=TgEuuxyud8&+YBUhqWK?i) zmxKuZ#oG~lVo!aU3w5jml8R)81 zZ>FwqXR)ZXoDO#?^shS?E38qi3A}21+MYRBWnRSA7?;F;ko)xGU$oEy;z6FCl7a2X zO_CuO|1O`YfZF9T9c~g^8FDP=do1JKbZV6#Lms7WVudu1LwnTYn5~18hyE|nUw3O- zxlv8CzN$)@*uawTUy5R&3;V4Wis0)bEc^8J*6?tQw7gQ%5SSJO6OEdWb1=}WmYUn zKQ;rhcUDwlYLVG#Pm`%puP2EFxjyrerV_-&Wd-#X49FwJ%@GFlpy^wWF>sgOV{7q{QWB9)(B?IW~8AB zc+eewh_uT|ji)#6>`%JKuJL;>JEd#FG>3S_OvHY#!_Fu}QQZI9Eglz{!mXJ~tL$_S zf&iqhE8aUa-&jh)kGp6}Zg%;V3jNBz4ardoX2buAb$#WGrwc1xqrnd;IbijHm8xb3 z6F*K|-e%V_0Dv{w6zrlwt6}X%9syloiu_9ZujS%}QIw7=7SA1c`AnSEYl?#^aCEC_ zc4$d*%P*n(&_3=r8QZ&YLvDu+#aB*pN%iwO6m)F(b&4bVIo zq0=Ekru=G1_-qZ0yhxF`%3=^jNZtN=ziqW9Fs<4SB>7Rosz>>V_l%n@E)2=Mb}H&< z7y8-t%>9x$J>)E+U;&%ErxG2h6G8dP%EOnLlITfi33_X14S+}DXqftN7dk||Hh7r* z;7=}wrh~K0BHtu1ILQl}TQP@W0e`P>HPlsU(Kx-lSP?;G|HE$xb@9WU-ankpRH-qM zgF`Zj(PuDdTX8$5ST^QXA+y+Dw{njr&J82XH)#VHVU*0^$u%f5tjIzgw=d-9gqVGj z%y*9VWZZf6V5idP72P7|7V;d{zvtXmm{F*o@3G*UgYw7`=kJs)D2Lw5r@R#OST zod_c-jQ~qHUt@bH=T{VZZL>Oy`u<#09Yk(eom2bXPkJCQKMnLXR8t5%uKox{|L6W& zxg3FL+Hwxkg^UmA$zks|n%7`~>~|deRw92G)T!}ufhw({;2UZ*V}!5H&5TvCQ#P&* zMX=d^Xn50sAz21e=3{14LXTqGI~1l^H_|#<%Uz6*>Y+v(>=OV_F0?1;4OIr+0u<-a z96yp?ib@7@A3lrPy-u2)@C9WMJsQ~*!4usGtXrR3dc_<2M2(c?|MJv!I5!G96{(r# z@ht-uEC&aFzBDp3S{vM?g^>gnKdUb>yi~7!+)nym?PQh}V{Cu$^u`F6zW^tc9LxBR zne(I`6n8^G48_0!9uLJK;>T|S_lOk!$X-XU*Xp>fonMjF_Br)BnY=)$%57T#ZnjVZ zgRa&_jpP+Wq+Y%v5MP~??}$@%oW1w~?#q&`v(8^(AWmdx4t8OYw`n1--nBZ}+#*yt zpd|1_{+$=NJfZaVuM>`UO&H&8Vw_pi28?%5K!RX_tvJNkEJU zI=(ZQe~M^-ZhkFw3~($x>ZS*VSeH>FhFeCBvVfigROY630N>jwvdMPy3+jW%yw75g zKj1-2-o_N?k>MrmB#V#Uu#W_^$=qn9mD{Y7;~}lDP13E?u>tm%d_*L--h0E{$5yVD z^01@s%K6@^d#5t0f61~QY2e0`<63~O zZbStu(qo7!-0`4Kiyf++-0U0o98_YlwOZGZYX7XVS7T;LmS(ab zX@Mbs(X-MJAk=^f|t$%kDVfYzsk@v;1h z0fF)AK@Rn1J@Zu`VY$EB{x05e0TcU#so&!u&kVCPtFWb#ByY&dCE+Y60`y<9|$l_0; zT1}Edc$FEt=8Bkx!F>|&D#R$I)LIik^eb_<$Tpu5R3(LE)S;&!nwx5tpEpPkz*uV#!%FH+Eu_f{$$P`B`ndWk3 zU5x=-nN!!uG#+2?UBF#@C+l zyNwQfk5NVIHW1>2V01qg|G!siNal8$`z3RyZHA||S}2FX&w8HwtFWi0m|^7?>~{jQ z^>la+KPwk%Ojyo?o59`8?xdN+LNB5rhIeBji}_M`L|ohA?|-@t)s#@)ZAYcwtWzkM z<0@5{DC>o3YI#7uQ`i`SkVLH@5BK?v1Is9w@x_^Dd<#{Xpdl$jSp|?)6Axm~AtD1K z_I?c9&=adHRdS)+eCj8+-4e;Dv>7!Ngv)nNU!MhYqpudwH8Krrq)bxx(FNF2&g)_* z)QEP^Bu3#;3PkFM3O0VVlT>NAoEdXP^(P&-MJCsVx*p zw^5}$Mz@43M9NmtM+ML>;cMT@gYi@lON_qg`n+%vOzg)loQXO+dijXnlra4R?Pa7V z@r3tQqW^JLP3xn0Fjn(c;dBJNMSjJZg6uOVnU+vKTe*8-v!pCxRwAPYbiMJs41Ei* zF)`}8Gk0%5mVxK)5oX3ds-8x5RcXzkDEB7*4=yU-en7lqFRNCS@8jV`%TJFfMe6bV z@fe5dvEs=-3K&XwNtvVA{hb(g6V@}1pxl8vhhN7*VMYqt4mI0SI@FLVTAXy^@Ct*% zKGkOHe%#9Cx32=l7iHW0Cpv2Q1B#EH^U(|HS~516N0jl#AOb%4faxI$H-9o)*9WM) zg8$k*MZA+F;tP_m#fG_V_CW-d$1&E1WbmjOPw`T!-y73Zrs zpB=WKFt;?Y!jYC%#^#`8$64Ae>|!qt5;_iGwOkhz57#LS}<@1bW)FP*uYeBZylNyvc%{Fgmb&vky;Ix#V8^UrxMf{M z71-wz$6^7hA=m&F6#%|0g$6r;t?{$84%oG0Q* zS$0)m_lF}Z^NFXu69YHWG9{`x4wtg9Wyz-QDJJCk+i75XG5wE`JFILWpwO4cqcz!m zPB^?i1^V3Xi0CGZkdF&_2>D$%rDSJE~-qKM^_S{7z{!}9v!xfoX2A;5=W=0#Gq zi?%D1H_R{?rcv}p?qlN+y%9L2d~cD^8RH6qtIKRP0JxmMXQKUs0e<#LPLEhztE73b7VVf;?!d2d{;v>68=@SYG}ra*=I>XMLgNPej>bFjd#vEW)B@X zpQQxYy+ZCjy`YS<^beQ?P#*A;L!em zcawk6P%S|Qd@v#oMTPM1Z84tf{f3dt`8;(6beGtt0JvNu+0VNXDbI96seHo9D;y)I zKdm{5>6#@{MG9sU$daHF@zbpHABGP3#v5i?=t*`OOcEX2V>yP+u+N%S!rDG56;*Lk zyXyrhMVxLc1S*1~Xx$%|IsZ&`I^$a2`PpzF?W#^PG?IaQavwtgzTns;_vs@m6mw2* zzTj*}52XOnS6UWe`F=Ro66#lj5qs4bxpD-Bb9oBp#}TZm|ErsTX$Ql8I$#f}%Do6d zbY2$0g{;r;vCij6UQWVvj6uoR>Yp!>p{Q3Gu>D!!Q#BMCJj0c9gD)D@DjEg&{g^gp z1OM&xns;Ngsk39%N>KlYgaiB1!*|A4yC}yc3~GaV;q&R z4wg_sJIX-gx4fJO>C(im7MwvW{LYtUop(*C#qIf0T`qbb^=(V^l+;$2DKy<4 zAc-Mab08K`shvDJTr)MZ3^G8izl&`lgKDQu>~aW0E(Y}1(qTg=iRnb)&MWWG6u@^t zCeIHR15N$#_dfEjGBB)|v5KF);o_13PHfDayC!E8fmyYWJ(}Dyl9HB>2k|##p88$wEFo(;DKKyA{`ufk98q

^6t;@n{CJ6!U`VNx31`Hq5+R$h0 zw3a>R1oL(Di}r_VWNWeX!z#lhm8WtSo{w#+zex{?mUi}{`-ihbb+0nYD0#X5NS~Gl zFOi?^Kbvoj9B0~urSn3y>4V2(a`xT)Ikl;-v-m%Z-NUwMJ)kAvZQHhO+qP}nwsE#? z+qP}nwypYa7=7&Ikm2Eudd zNs&{wPtPrXLXTTAz_CO(;ArS8Mq9{K3ug*m7I*f2#WzmrOy$T1XY5`dV%HOZ6g?&( zzg9Y9K{|6$#`;Kocp!K6yY&m$m`_I~ZmH)5P73H{mf6I+;APybbH!Zt`CPt3^?p-- z3CFSEGkBb-LJgcmKwQ$NLbV}2A;+}dl%}2GT*}8$OhUqR4iWv<8L{f6;K4)N<_q01Q^x>GP z%BxsCcG^Z>aYhvsI6koeMO)i#%|Le0YT%zKN$j8z^f-d593JJ4^i3ZBpr@2geaa76 zux@p?qN-?R68TuR08%`0^@vShG5y za6FJ>=bFw!TPXQQSExZb6w;(ozd!o4f?V8~d86l9Lyss`mJ~z^-pyEg!q1H%e@@x( zT3qns%0$S#U};nb7~B`ejrAwc(;@0KCv7S~3#b~*NC{}h!@{m=sVI{DOL+1?#{>px zGf|DaPA6{-_|z|R>4fX1&x;bdQ&=RgILVt97~lmC23@Z}|CE>ZiS?>NYvl`S`HbK= zC@`a)mo{{BT{sA%OaXVOhn77USkp*geab~fU)bmaUVUl`nmJUbi_BaaWr|Yn6)r@% z#2%MI6(1eh+5Y{f*RQ@#ieEFFJGgnyQFrvPDVx6qQbR)2`Nck!*#$3Ez820|-yI6_)Sb;TXUl=$XCk zWRsB=-BpoT4!eX8&#XyQBU;WXj8a|Nm=g}*eDf(2U^rA+f^;9ym6 zh-jaYb;*IB9n3Lmq{l_Rq@KuH+IkKu$gpAgWjf~;J0qhR+3pXhgPUWp7yWYGoz&D; zr5z6_EK5QNQ7L=-#nxnCx7mrGkC+fY#ME^qN$q%Lqqz5w6EsF>dMvFr-=u70*Qam1{gf?f&|7v#tGZpf zrdM@_V7Gb_QY^X@vl}jOGpr%Bw0vddPB>?RnpE@QGol7!aiq03b<(>(zsg6FGbjMP zm1>RN9$~W&0M*0_$8{(aN=hf>*8y|rJkZo3T)Y#D4t*>8_J`7eZvKN243<8INBGjT zNqD)11<2vwsUzYTe=btCJ zW%v!B0>rThfhntDY-noBJ=vL**kNff&_X#|S&POckH~Y(6#ix|SVFL^fF_~G#Knza z64stcT8|%m*ptfthxBZZqrjf!^HOKB5kL*x({ZjHs@C#IOq;W(4EZSYhVA*ePef^# zY`5*Wq$%unS6v3`?D$k}FrIojOg~f|(K?8_4!@9vMdTR9j!lO3YU()`rfckPvrpL!cjp>D5QU5&guNP)5>0qiHa*zq;IvoGCf zRhZDDOC>H$%ckpF0y_J^k>Q`}S;P?b+jFVQJr z^`H%x;vH)Bwi_&Z!M{{b#sx!3tM$y-j}^j$)joA}&xdbk929h4?~e>K5{{o;=s4@> z49opss-yA!`F+)_-kejvmTZ`0Hz-@`rpFiz-$MrO-i87M6Dl67U~tV^84x-#h{ybo zsg_lmHZ^~#nwny4U(TBUSB5}p@ZZ#X(+Bv$Ke50Tuy+|6?tBv%)4B7t`nUo+n!#4K zM6^5)#buyEqk(b*B8WbJ>a{CTYBoD7n1g8t6!7FJ?ZA<070I4V<^hjAzE*%Vf6?<%E*nZ`5l*jfLdc;L+Zljrr26XYd{1H&=Ph8 z3yq!Vk)Qd52}m3*$dSo72`^jIN8c1whYOqQCj18aWLKK7uIk?CA^NjWSO%NCYK|>MH#1?#m39f_)|1Ku^uQo(0ggTO)0W#;jxN@6OimP4+Lo(;=at&KO7R5 zU#bu^Z@I(bgNQWw^r$=nUL&JUpL%lFBDXgLhg~ExI;u-|<}|A)Gk9uC13xJCoaq(c zDgcXSRtCO&Hp!q{x*Zr}ERxv1$5yN3*x-k7>@l)5>>{EiD@#5MpdV4CNTm78b>O=TE}gCXV0u0Q}y+&g%9DH>cz;=a38zZ@J}l@ziN@V`Q%pQzclXmAItI;jJ5%lSbD!)HO?$ z?cw3lJ&Bq&_tvQxeC!v2r*djY&s(Vbduo=c|CCkY~BeB^Kiy=rX9;}J;OD6fEV0W#FjyEi4{08A!f{% z+&@+SazRN#|AXeU@J}D2yW86PBDL+Yy_w$^mQjg3eC~Al^AJ_Dk7_5UQ)T01x)5SS zN6B4Stqq&liA)l>dWyMsg1Mxl2-rUa&Qo%|JNmY8PZ?HRwMn7TM|=IZQy4>&W_Lv7 zy;WiETz%K!jv(+K&G4VwNnkDhLPiAkPr48(gVYYFvhb2LqePlxt*q9)9%WeZm@k%X$E+*wncX2RO{Zpt`vXI$XvJv`)A~P>QB=zg&l49Gce8<8+dcZl zY+gzp%x!O1N?FR0BLA6TOkQujcY;-A%aJ932cd*d$=(*$Omj4b1d! zibA`X6!z)sv+M++Q)TcZB2RAGHVK}yxcP-&ezz8!M>~vLRsq13?vqbuv zZiFnO>Th|!>M!&|^^8{i5cUHK{@6r!H2&Q9MYHqpfANSdpFHL0E(V+!7kNG#Gf4&m zYf&9}d9SG|QQLyICp~bEB{okd7qjixdE=&n4&K$tuK}I@E+pSD$B}T8dK&Z9@y^DM z599Uok?DZ)BcVn&pBs62QA)-!xLcXXd}Qq+h&z(AQouMVB^y9vy~;9hq@9sml%j0v z51%$V%=l*A#O>oYof5%0Ti2Wc8q^Z63o@JT!nDXZx(p{70n>vx!4P>7hxr2}(mrf$ z)k(zy4do8Iv_H(!8TXivY~R`5^782&xyY$8An9V=+3k>L1blT*u;jg>3wKB8m0hqJH(> zz%rWb>9LW$&hupcIk!Z6rwVL;KAtq&!;9jP;ZacSNaaIPr9gh9%BM0t=H$Q3^w|O} zoV7I&{rk6h4`z==FvQKphO>RC&1aZjO5W&=VxeSi%$y({vcx#2gX zDb}OCjZ&@ielv{!Dc11%X(F3#9~f_#v%z>kBJ6cGplEoRrb;yYT9WS9mW*W2U;?QI zOGLs)`4Zso9k;YC$)5T4b0uPnZ2;wX^0{`?9cHNVlns{)_|h}>;L5_5q(B{qB0JGI znY_3*hCG^_SU;<-TW#9%ZQE+u5$`@+6N#zY36?g;V@C(CLSoz-Y;n|Ie=8Bae%u$A zsN|=s)aoi~>#tU-`mcB1c!R#ONh5m;0@wBhjG+m2_WQ8C9p^Z07!K6Rlr)T-Kk_wk zy(tJ>kZ$}!RHC`Nm439v^0p$Gs59L@9?+{-i@MxpkPEZMRA=_i4HZh_j+vuOs(}&- za?}ecp#X*0AD^=>wt-f(scKU&VJ_p`p0E3}H2p#4xj7;fZJF0$bLH#^=Z!m1NXm1kp@)@oKsFmRMSeep*}O zO<2n01|~O5?bWm+AuoK@LPLr@Z4N3K`;a7(;UHc6pJR~ex4LaVKpqrrIku+;-Zv@< zk@Xv^s+=H7Jm^{MAGI;=98JOpX=tDHU-z~K1YSW%?+Uc0gsMT!Y{7_+x;CrX!&md6 zte^c$d2Z|oX4~=ncWM!ldcebs|ENu|-&l4h(QUvD)d_KVzC74ZOpQXW(tNjyFv2p9 zTu208RRwF5^R>u&Klyu2s^0Bvch;4XmBPnt`G}+1evC!38zALuLDIx%bsKE5YmICw z`AMrX0qz^i8@8l6Dws2Cc5>yQ&hI~H-Xh3e=ZWGw@Cccs*Y!M64lawLVwShmkhzMy z$|<L(M+8!K52`|M3Tvw_`j~Bzumr*v^mI1nx5Bbu#AIk+WnCk;e zHM*#l%VQye@zZ{>VY%Isg01h4IjjsB{xdKTbP+30KVI?M@Q;DtQH>%&f;OX1H@-*$Aw#*fUA(nxHH zZlj)`#7%!N^#-pjtT&MTj>XtNrm84Rlj&Vv#SsY!#JOSa-oGkcq3+-vPq@Pgw8K$& zbTF1FLf+gPT91k}ZyeoOaM#<0qEkJg4q8H}c{@@=$_X;UkKxc+qy7G43Ks7Zl}_dT!Q_Tl#3;H7zL>`k#`~@%dZVj& zdPqpsi%@S)yp8>(eb_|l%9O>)wI^K^iGB-Y!052CHHg@HMg!ELwJ}s0CAfzVW2Mgu zl>=FI7coF17{ni-Lr*inYTj0%?Hk)nzF~1MTsBg@-No#)64Y`@$*I={8jjcipdQ&7uZrC|bPKTqw=;{yoZEUuJpFT)p; z1_NG$gj-nA@-+$|mzw*iGXir)n3l~V1?PNX4z^#0y#5)GmQ*~`Oy&@uK`R{+EN>U0 zo+{-^^}(Q)%V3KUcu8NhY=V*Jk}W$B*(^o42MZl6oYMB9^J{fF^*b{l1lxNwyLmER z&)&A}4Y$=U=w-K^S;8iFiwU4JTjOPS&B&KI|G4IdDtbwa=T~q-&-KuADBn=Rn{pwp zMSzm}p-KI+V>EGWa40wQxNaw(`eWj><^y-`yulDhLL-g{gp;#q&7lzrMFY@=zvqoI z$Uq_WjZ(64Ab2sLK2n2KDi3SQdOk`3AIb2X*VBc{Loy$ib15^fZU|BQP39knZA@0vE8E9fDi+(BZ``ICrA^|UM* zn@qJ-8%zdX!Z40S$<6K`cVK2^Br6Q=m`Z&a=!7$3e{)QGo)~Q0Sikc6VKE;qqPq+! zfvc{vew~iz)_-3*#$LOjL&H&qJFwP0t3M7cN+3$(xDbzn`QL@6m<6?)p#};Y{8$|@ zpx!d8zOxU+nSJ8WY`f92%DKN>WL@^MzP%?}v}gKg zdnRKl_aUV1%m8rG1`UR3o;nXp3v~OhD~4h_hWfhtISqaXPjEWS`{{2eX8S?GcWALQ zc`rK)1y$bLS$(7DB7yFo_Z7?`C91l0R} zwywT#x%hzo`pD_4X$fO+w6QeH#NKWqR-3M=dj|qT$y}w3;cP;(+ZM5%y}T(%$wQyoF&)Fo z42WN!ZBG_$EEL1(taR!!}Lv#1sGWdRW8V*HX)B{wc zuP2RRKOJm+Aei2NrJYG6Q#?xCjLK*zYesAEAmGitc0%$yG`-`&33@QfAG*nY-+GK{b1`uikSI-PELbnJkwkYsG#q*Tngl7R=!69}%0LD%5?Z&#+o?UaOgyHbc`C@^aGO3}K-NjLcBXPN!C0Ki zsy3lbhU2#}$VcVa$>_7G5@K|Xo4&ynw{ z#}VI>u1{|XwjIrMF&fT4JhDegv1*tpSypri+U)r}48vkY?K6Oh%s1%1#8=KHWtQ46pB>d?U?18uHg=_>d~2NoDPYSt znB@Z+6w;%06w5QpXo2)R3|obRJDk`S@bU{y15#ZP6@(0ugf4P(2-hJ0{Q?p4&YJB< zjEpm?`OZw}UdJ;(KcL$MRt-YJ{n5IV*RJMEX$DqRo?`kOAdj2HymaC9)H>0!Y(N;S zqlA%4yMyz%th-pY8VO_P;`NPf-n`f4cuOPYn?~i6h+0FTCgj%^5)7J)E{dU_%odM2 z5%u}%=N}yxtz9-rV?rH(4(FXlbG&7^IR+c#us5J!F>Xq4CC`hHz0#f4Zk2{j zD^9x4g2ANkD+Xrh$k7pBSdM`3jDV-*^xHK+1Z>Cy!WV@j_2f=(m#sa8NP>aWM)q#BX#Pm1+e=fVkA-Q zp{*!mhAq`iZ#@B@7eJJRq>c=ujTsybb(2Nu!ZdDaE%eGQY~Y9A`}aVnVTfd|vC46| zW~XB`{GDYbu3Tob9cQ=zk3PkN`_oD^g-(}*2z0>jP^eJ>6u)>G6vObY61}QESUyZ{ z{56eBdM5I=PYs?o(61n*i&KM^n;(quz$SY8Y8>Tu2#g(TH+=Afid0;QU|}a6yP$ z4SyC~lbIr^+{5W^@$ec*3S`YeFkO1gzF`|on9k_!K)eek4I0SZ!%n}EVmpp%pUBss zzZPPtW7%(4ER3ynfFq=}hM*judBJv(mD|guqJHYMeDG-sjTEVMP2zwzs;pfZ*;y+5 zQ(I<8e)dnU(YKK29hR8y*fhpGHC#>l^y>KVUFBHjG23_;_9WduHce5CdL|%{Nx^9! z4+-iyOh&eh)S4JPz({5So ziJxlb7m(qS0y4Ga*-4Z(bD85%_pukXnndR+eG3=aF7THiA^1q>w(ZzBJW_SrbwAl- zpsSMy<_2Xo(%RAg=~|)e{g*bkB4#JCsy9KZ{QHWY?3kR7ZaRB=4m^(!UydY$@Rz0& zd8bKNIR1d~eyAhElv{`1n)!wU^6rYJq+wGu+CWk+_j5sDB22)r&_hg}WG?aMJ^mKF zEbS#rx6sk5H{=qO>nhc7pAXRsIsMNCgDhyp1!-djYMUMoz2}c1ZjF2ZO)6Xp^KEK| zM9oR-E$jA7@%%H-s3|YEw5`kIDTt{3&_1Nca~{2!5FhGeksI}XZrYOR^y|&m6G_n0 z)Hw~-$9jCFeHTRIM2h3H{y!d}bM8bo+1yp+J=aZ(I>A0EfU{jeg#hU`ytxD{SUs4+ z&mB+$39U^|RQyp6@L(hkCESVYzG_WKgjlkQ%(2SbtCb1p7~TCCEysl;fqH91*|}VU zC#XD=?Qz|1wL~0uE3Z2vKqXzayqUMT2;rhSW*H@ z=i{TWvQ}Rb2CarU&6uvd@l(2lLT|~bzhBEf`eRf{`-O=%-zdrtOC<$D_l%szjn4df zRA!>7OTl4OHbbHyg);KCm{0m>ApVngD8)s;B!p@25`8w~9mB*KlxKP0pZ>ovyZTu$ zM(x|=<(*8;YG9Vdx>q3CB3=1=E#Kls^G?SDBg+CQ*^&F!ykma>)rt$LR@Wml>~&Kd zdN6ugqEPjg2QZT4A5eFWzphzt`Zj2+k$J|b*T;Me-(6C<*xqHpt8RX5SoC$>leb*t z(}%uCQakFv8D*6omM5Sj%oVR})yTP7AsiFw+KnrIl~+xZRVVVz z6;vUdko^Vz1D#w;+S*DWEbkaU+yy}lEr%6vN+hwm@ zc`Ebj`xxe*;!LS+#Zi5wyK^>MA@oi>?J-s1rYmUrifMzJ5{O^@VBAE$))}}-6VrNm z@11Be4}3O881~?NMh`|-J4#beig@S~GHg1L(QpOR!>wmkm9FkrBnFjpmF0CD!CGSDY#%8%h+>DO8=&E;Rz<@4Jj zjrMT({>6BI=RM0()9Q^3|DxKzF5epNlKz3Xvv2N~ZUtN~rUy&zxZd2uSMjP<`)2&1 zPhzm1v3ozkd(;{|974LHCsZTLLP|h%4c8V&ptHq!_y;IwZ`+N^G?@gZ>GYalmgG|7 z>nQSodxcQ4rt-btF)#FO-nEn{vt|X8*^uTChU^ z%lO(EPBnE^A&-BO2!&A;pL{v=z=`X$1yMhtl?~MgmWV+yu*#u8oee6bwr=*fpfW~f zs!Kzd;E}wk)_Y)m*?4E={$Sl@7yq;BR!FG-fO7^R+d^v!Kv_qG=A~P4A2-cTz9>l5 zRD=%D#S+RS`w^JQ-Cw!Ez(t3Uv(R7Vt5P zsCFg&asq!{TlSjn z;`(AA&=f1n@DyK4*Lh7ecHufKL60y7ZfAG3R)?`+7@l0+0C~hd!A1Vam&YyKH&5S*i=3J<{+0Gk^*oHpkjS=qsTigMESb0C zR|Eb0{GeK6`euyHX>2W}?k%jdQgGn#5m9pH1u(lvUbgCXn6VdPi72K@p9%hX*L6=^ z@DsZy27_5TKeky8?q`3I_udObh=$?H$>6o0rUy@%QPRLPNshtq&P>=U+L)B&d;{OL z0Vk1dvv#F~Rp{2>v#J@9ZEk+F1RE6XlI6cBXZ4P8>sA@6SS)2KydIh0+b{2pVI|L? zk2zTRWDRF`xoL$e@lpg=H9Z!%&MHW_|MH7>A00UnKExuN{8c=gM+_9^i$LC(aR^n} zOm(M{bi#5K2pKTw)z-G__BNG_yj5+pT%gh&p2G&uU;G;lEN5>WO_EmtdUT%ZhZL&` z28kCJg}e|oWv^DeW3q;1?cM(vt}RiaBneXX2Yn`s;9y(%fgW)CXQPrk^u%zr%z7yO zjc^hx{LgQbaDC9FSA|cx7(EQ|4oy)8P8}kxe+Y2yhf{RbD`su{Y z#LsmEUNDa@U#;s!l9K6B#`DxQ{upQwkoB9c`II~Ig{C%ksWD&P^U(8Q8X!~C3`y-$ ztHqgm;it0x;#uH(uG$Xz|6bMOP37AMd1PU#5^=fj*r9$`ZJp~pvSj%>Tk_PEwI?VC zoMu0%_A_*m0w(w-qA2lb(>i0ht3u`+EUwTNxbq!V7TzFfo#ms>k8i#JJpfdv`BnDYtQu}&g`gYjwMEy&@Ya+?$eRbnQ0tcGE+tKrOnW!X4f1duwWHX zv#qTKt|pbkL*yzN4)6mwrI(RP{V%eKEiYDRu)t+#|AUq+*0@Ep7-aj*pn#KqVL|01 z?pm^`R`kP~!?=Rz$m2J_#rdvLIQdh@Z5noBE4cMVvfpa!1$YSKZSnk9dYJ1wR(*Nx z6&{#UjFuX)lZWP(b^tM4?6v@NBS6eKQGO5xeRSc+G0|;N8{fRlWp$UhxfbCuV71S+ zvkZjmuA7I93-%NO()9>`%}qRcydm;wwaFFmq6Wp6WK)Zi%Qm4NhC=1~EubwhZ0#vw zCZUyxE~3Bt&x{}foi=|ji7233m2L*)aERM2{}r7ez0xVzmRBV6=S6eY$QPyOUJdO# zCE93`m$#OV$00WXXyRMO;(O0hGL#zauv2s44Cx2W z)Lj)XnWeFX;BJB+|09{4+S43|d4w@g0dCf;&619qRc{e82?e|Gnb30?m17@~sQTOK zsnI*cToa)wlpY#2jR}r!3;mkw7*muoTHc}Kk3BAQ0g9={LPwvpKX?ZS!6_W1Q{D>j ze2-`kW`e8S@`~fep8BTQaM%&IK;iFH+y}iHVt`p>P!KMG>B)z3@jfNX>;Z6;!-p3Mw$?;gz{>{2&9QF zXiH%4AtQQ7oG+TDZsZ)SegBEi2s?I-7zm*z%@RG66>U~(MdYA-@V2Q7gRL21ehmPA8tWVsoV76?nZ}%NOSW|f z#z7LP+f;e1PUg5;KyaBJh&%B}%{upcU?Po2>r-~Mp;Kk^-@oS()BZnqQ!CC+>S*oZ zaVA+b^o^vHMS;o~@#eJMT}smPZ_pp>ELWEKSGNU%PsRhY!sMxXT@XIR{Gr30vM)b_ zet?-8tw__XIw@@=@E#yv z%#MkZ1^x&>M>6WL-e*>lRvIMqif2s7VZUo4xK+8u{=c+X=xYHz{~c)wR5E^&U-449 zV0iwrb!u>0M8XygDCFo^O=k(_W_M}3RL5i~K`M?CnnD}j(wJ%hG`Pzb*3(IMWikZ_ zj=^^PK++1?A`jw%hd8cC_~zoy%XRhWqZ?g_&0pD5t?T>Z(XLTrx!1cMXGZ_#cHrUTm$#s>$%yR5xpJeOvQQWMGvn4nJP?!$NHdR>GN1 zdL1G2U16riYS{Faw0ymng%(g- zO*%Lc=3aHsLrI~13IB&Wo5)Tgli-CcTe#FjiEl5x>>z9#?RFIS4spr!grg}$n3)c^ zvM$6W9c&GlnhH}ArRbpOI3^;ym+Jy9%(qMrJ3N~hNu=!BQ+|50VxUaJMNtX;=Ykw9WYX&B^1Mcz+3^j z7^?(P?kGJO`ipqpibpFZ0>uHro!trzaE1YDcb9_;I7_ZN+$J3S11a4_L>T*1tQjeI zwiWBWj0D?4K`VWJc_-4x1`y^)Kyj;BLiYgAV+7_)X-0j=2fT_a3(Nt#rAB^`lkQa*+USBc_8J;WqHj_qy@L0$CV_F>A z_3T?Hbcf*u^W~GnKmhWb_lYhMEl_Rz!=hY~x^BZENvglBw2PRN(cZvE;tC_!oxOv74 zsXW->#VnS-ykGxFG0sc}B_$@@=5hBj(+y-&agfgl>+% z%QfrvPsdvF%pHE7cad$~_#aUOza=ZUA#0ph+Fy3yQKJHEtjxD?Q&UbX)hzWue(y84%ifq<6+@JMA5ng95 zuPnsWge}@mwHt4ICphMH$(-l)~(eHTK@hMNN;+S9T=P z6bHFp)e!cQc=-ebu1%X~IK)S6vII`%7&qe)*M8Pt zNPXbYyMZzNM*Im=!ywF3(AB?d${=PI3J842XR@9u2WY5PBKeWCV|E|ABiz|3h=2n+ z@EZkIVrfSVrOa4!9;MLE5@Xiw>TWI_&q1&65@4L;ZenWm=Z0hEIDnuKEKVfWxX z^q7GLJ!X7TV3s!!ZpI}rL~cBAa$ooFJwG|(tkTypa%%S@8?_0!=6AdZi78#>HLF_J zph~XeocrWPKbKVN@H!B8wQ-2Tm$vh$BGpH4PSGIC44>Tv9SahEzgh&c0jvEg{*GNe zZExoJN5jrWqynU+@kIFJzp@LJiTNF!bx$aqA7jm{&-nj7oqa0k{+LD9AbE66Es|YG z0m|L#E&pxi1&A&H!5-q^ohkp6?$JP6H$wb_)vASE`_bX_tB< z9){YI1ZTi5g3D&(Apgni+V) z0UhHytfN0;gSX5*t@*goR5*a5%M`87P+V^O5PtI*GJ?<%Sy1XirT+ClIwCt&^#MI2 zSV5(=W8y+2Mz-f?B0!dCW$>+(^$XS;I$=gLedK5yVncZ_5RmZH5ED4|&(8loN z{y@jq2uBgUcolfz$?-E4HyqO6J&E`5Z;pz_7oC<(1p&9`I|Re!4|VIsemtf81b?2U zeP{we=J+L}7c(#jiC}U^>zv*i&u%GV4omMvau4?6s2A9L+ng3+}#QrimM0!ZXJq-rPh5@u4={ppu8p*qtR7FWP zB?^G#iKZ0B?E^_NVKN3fd7^G65;=4wI3b5PB;8vCJg!~h8fevX*sdhU9ZoUAvzpSsZtc_rx`a>uj%-_2Bqrvz`*AwN8Z)h}ixFTd zTe~g<{Ss3)yUVF6(E)D*zZ%SPkR51N9Ix3SU5kzibiZGz4`-mN6l6w;j9cz3Zc zrVtF9(M~Ht1g+;@gZ*C}JV~h-1ZWS*8~}_on<PSY~vln|K zU65GQmY9W!HfQPiior+fz@Ep}bTv(m8n=}Pv;9SW&0w1?W@~=&@48UB=A74(7s3jK zCP}|+O=3qQYo0n6dw@RcV~UErESk9~r|m;ygZB|O17-uf=X}udR9*0Qc#&QScv2rl zoQ!xJ3>Jh&(kLj{-%?fhu=MYDynA+;6JQI%&v};Dp6VLaq;5J{2Z-PWNms>j&PCa1 z6;?&uVVBa>72@XCw$eLgPF_ zEB<9xK0BU}Q8bQMT%ln39ht+!E0hVVr8$3b*dki`MX}u^nrIIr&L(oPGt}*M;=M5E zqMNKPI(dSSArJAZ=-X^NT9Rj#0EiE8`#y*=Y)}WrHBDA^Q`%n>D^Bt5fk%svC8T zM8)=A5Zn~+&TekP5abc|v76p_e!7qSptJj=z$P(_djb3kZ%0&7&|xI)Y8QT!%Hkg* zoZtU6X|LdCY88~O0HyXDcQWleyylK5cyR3si*9ocgKyHHkzt@p(XxFoaRDnH(`A(U z%IJ4zpKh8rfNwke7~uHzt)pG@vrPdej=|r(!X7*Y>Duu-pnS&O#sFFe>p$6hujnv~r82h<>U;lY@*{gtU%#u9PW`bZNPqqgi0934saOW>5$DDd}IS9%9bYpQ3yr~ub zyNGZokMs0uU*M4A#y0FXSgqL-e6svX9R4{PfZBW%IN?-& z*{OI>vg5jz6#zKUm9?bT!(b+M0q{sp3VtpquWc$NjSU42t6045*l-9Xjt0O^QM=b& z2_+56@0-}mv@bP~?)a`CBt%o}1;qJLLmiI4D{i#Q8@%XEB_KL^Ck|8#k)ZgEIF32f z?U%JeTKKY%VJiFtiN*cDY--m3r%lbo^nYz?MkaP9hX0-ZM_K>xkbseygM;z^)243e zG*!vEYGspgq=p-Tyw%#?+1UwBA{f}|5qLO6ai>Lb4?n!69Vzixh^uvd-LSrU`u!0T zdpyqc;(f|>Cc`9NKCZ-?HcB;c&S${@F*GwbJOZbnstCrN5xBOQc0ynV*M9zt7y){`yad`eFg_A41sewt6i8DqA?@FW&IbuexAfu! z+yHbj^9NL|rw-ZJo_`8=YX1&U{Hst5OO z-`oP47Dxr~L?cBDU|K&?wSdkC4hA?Cz-9)g23Uv6hg=4=u#~8<1~6GxZB}7D*VF^H zZs+6-)cp;Q4hbtwg+UfjlF?L#0$4T!8n>{t`1M=?3DWz}W(G`V)qnAS?8f?aR8L$) zSz1<3F*5&gh6iv9;GQ2@d6nP%I~04Ggt@o3s;gC_GZ^bb0ywuJKp^*1Q-gQ|CPPkL*DMdL%C$+B*eMC%ETdTu<-#J*g_Mtb}l2X#y1tAYhIksUjSP zIXs18_>q471kqajl|_JvcW`$6ND{u#r%L;ieWJfS++X8ncDa9GetQ2;rv~om@cJwI zADtS_Aik=GdidooLLeZe2KXM{dyU`R4Ej^Ll%$k~oVKK3qWXQ)gTRdR-r7}w20?yY zf7SmmQj+Qs*khv|@Y*LEpmmKb1xQmW02fzZvy14DGOzf?I)}gjoqqLiY3-Gu9(?@l zZ?eWlFjdW;%KqYXxDvF{)eU%3@;C0K3gTmr$(jJ99}IvCI3O2|t=fO_XNA^(_}YKC z&h-At5r`YGdKyP3(63dke+{}Zt}25Y0l?7_`19ji`5_M?BM02T(i+gy_Ib@v@Rzz1 zXB-M<|4aLO@>u_WZ0gScZ0cWEDcS)7aQXn!d8iCkOc+mFh|WL%)9XGb$u3R~iY!i` zi{Gx|KU?!ti>v*YywpF_Lf~)VR5{G+LyOaUeJ*WvOKti5l1kd)!PC3RU)n$JG>q|0 zKnMe~GZRDaxtWJHv8y=^&hFQ{E4ap=rorFyjz8(Q%}lgcg8VA#fTSCG)cY2kKhf_O znBRG@dZ`QQDCu*C-@M(ADH$jez}99^O<<1pc0e;aIy1d-n>yHmff3MqL;o9F>HIT& zTwnuJz(B$}IskI^2>gDi>!9y56;DrqYA$~fe?mS!fU1!{A~!ZRfT|OI0`36RcldEY zgO*S5W1#xYU+l)d&!AsGecyq`;_D@{(WldIDg@PYYKY*3Dh*{ zmw6)(YclmeT|s|MihqMx2)y;zrCw+{G+wUyGA< z^2xN9!P~7YhyEJNUMET%!|24%wE0qi;h=c2$(~^F@GW($okMqeGm`~_+62CEW+ABA z+I057L9C^&F6idHXXEPw2lzKS7?&99oQ+t0Batv@-$+Zc11HtN9L1TKeCP8OE1Ec4 zkt#>LMhAP^TY>Q(l3o})+Yp2^Gh{H3Cv$IKHwS-S?O^;a&$xF8X-%cmxfhLRW#3tI zsf%EE{}1+Eq~62byOb)|MZRUQ>7Edl7B7qlh6Z>UPI;$#A-G6DsK{{7WH`n95o7t zlo=U`-Pje@i#Rv5fn!dYoBfNirA7!OyW`rqSnnqHQVyX|SE6ywhk}YH&hZDhTFw9t zO{y~>b4PTPz4O zv*au2zZ+^@zJXW7YnT^%l=()(K{-A9MiBEQ(HT zLD8T_)u3~F%9E~Grj#igd*Q=wM|783QKOT{(Da8CX)b==o?#lYy~K&=g1Glv+gtwYt%Uk+OfUNERa zB7E0VY75vPHgGXS``-!dxT-79s)e9hl}{0=5-Gf2k|eNI-aU%z4tJ#C_^I{TMe|e* zA1E~iwFcR=)o26Dzh+Lp2O3&Cv<+9)B_4Qj#y*SY#pggw1$dS=q-K?iO%O?;0H09Z zART$HfWi}Rq09gL;2cAKJEC5a69*DG8(iQ8kxOM5`&^WzFNI3T$G?H)*H6tH+W#+T zX+^)~fbn&UIH{KAg$yxN9*3${?{;CYv>jUgr0mf9 z$vS$iu8I6*nzX4>T7}T*4|unnNWIL{F$A?%2w$((AbOb^#b>?2UAoYB$ zynWP}iRM+`iMnw4AaBza@*1oEd?tx)al2z>DmW?CSc1Q`S-Yr255W*#n@Kj%gaF zv1_6g60&jjOgt<+6R5IG!p~F5yqJsZI5Q0Zj0Z22MVMB3+@E9D?J2gB9Y|e{C@7L} zmwjNBG=wg^kXMP{w<^1`Ip6TDo8`z-zlyFEY8z^*jPh`dYaTPfJIBc4C(mrBxNP5S zE@OVdd(WopHFJ+3G$%OJ8iQ7%pK^~&ElEJGi<7x%)C^9dOkR%oN3&Yd^@M;qTKhvgFC%W7MjxYze-{Ru*itn-oCGo zubI2eUXN%Td*_1AmG+?}N&8Q-sRU$-HDeBWL0;?qyiByYVzu!xOiM@dB)!<}4z4EIK8kY!trk#zK8q4bPu$s~$^PUJ)z z=*(ML>3*z?Q*_^+{WHx<-1-P-hcT%Gnj^gZ2dzJZnYQG|>?3jgo+1JMtm9F7ZYPb+ zYG+Fq)S7m!dYa28T~u)4Vozi`W|obF!Kj9pyQ_=v7pXh)^UG*XO(3(7$UH})G>akz zX6Cpkr4d6;rQ##_4MI_fWJWMV9l?2V7 zBqJRu@FwiBHcT4f4KUf_fB{>zzGkyXB>7aIX*=@@;zXC~k3=llhDNVnV&ZV6e4fsP zT_v3cNtu#g3*=qNXpo$sCZ0ds)a?_srE)C_2Y#F;h08O^+b6RMV(WE$#wKmgE^@gr zB6~76`9)u2sq9Q!v73{5CIrLHY?Msa=foS6wxoyis1W-rDyf`@_3W^{`$Y&JkLrmFIg$Ay@;trZ3x z?2pO2VttMD3Km&QZ^hfM=Zkj~PW7G67f*n1E!(?e*y)Wu^PsI##IvKI`b&#GG zM)CiEdwS4u6l+r<5gM4~@1F(WV)s{(dPb%%8_lCK9zyq_-!!Qt2UkS z7}MnJHzVb&7RLNC$gb1ipaN&HC3e-mmh&n{VRA)Do~=q7siZm{R%|y5SwERdH-7o3 zcuXoK4Y7*^F^UMmbZlML#LO#SIirzcvTQ+4(qUK4X?VUp4qK@nrij{edwB%dj!U6< z5B`!oJX%@C#>{0y1$4hqDJ_HYtScL6uSRON4gEi}WUJCDmtkD804ru@lTSf;(x>UQ ze)(lt5J>bM1wRJ|-?DHy7kN+$S(6~;BX(eHXd^Nqyb=W;n69~7G7gQl#MRN4-=tUI zIH#it9dpU-mB1zgnx!DN#Z{1o@cGp*oewKbNc1} z*wcT(-m-p4sQLijB>Q6pr{ip(-GYUln>Wqx%fCr&tHaW{5QkW7^W&m10(n{Nza=bC z&3If7?d?bH=9bjSh6`#KN0IxiY1H6ua>l#O(DG^s94k$x`ftp;vcKcGA#z?@#gQ(} zXos6ZCZvRMa-Alxi97>mjts$n`w5P)`G zZaomA3GhqgLnNEa>;IPjo$mkrsGJOOD}uSQAXk%;kb6r=R-Nc{Seoyj8~PThxJae<7t>6 zyY5l5BqeXpyu{yOUgpP&AEw^Jz$Wdt$azVfZcAF#MYan=l{i~lOvdR%5eKy>f1RIw zMaVVYnP2h)%Wf}1@jGO&6%|&J8t&4@O9?BGKIp7Nby8x7|4LSPp*)jNv25GKCc!#Q ze@b*$7ZUp@f-*%?W`ggu&V;MLLkrmyY*=8B)nY}fP+l5 z!+T%_et=wYutaeLVJv|gpzPlMc!qB8y7P_k)?~#Sc1Rd=uAQg24g5Ws4&->enDVgA zuD61W!nOaooH--*MWhz~dA6S8;oC4!TCiMtXFK|^iXmXrERl6RFrVVAP$Pwcw_6In zzB^|jUx8$6zr1sx@n8=@3H%^jhhk=wjNjN|dTB8JG^yWtF zUe^d@iZP)$C>xUlC4iXzq~cC-Yfjpa&Y9(%kn}2d9Ub5~$R6uXkkwuk+kHAVZ_Qm) zwWi&?L)*7;C_rl$u9eumq3vER^B7W5OA33d*H}A0`{}83X@yjpt3%o;s+T^>15YKd zbvvds*}tM%|B6N@Z7T)o`vL?t{qf zue)eIw9rWLQ$+sGYJD9?SpBbbp@`3f9=$5n*{tA=Fe#zR)K3@?U@M{-n>UX8!$C7x?rQ0}?p4(dXbEx}VB zn**D`q;;E@Lb)Lf9nI5%_}PQQP%Dy&x2bQgF4UGMbGn|P+v^$(t2Qbv1!D}tlEN`W z)+Xb6*34hRX^iF2CM@rURNK`R-Z&d2pM6w60FP?Zgv;wPcT$xQj6(u}K1*tO8x{#u z^jx_nq0>`WJ>|HDFUxBqYZZjZ-BnxkUOYDLd0>xDzbK_1gl^d>T2UB#*r7E+YK}qs zI3zalwl;wmC!^7SQzkxYBQx1-BA+gesIJ8o-G8d`O~aQIH7 z-gCRc5Qehx`Xd)T4bhRb7tOJ)Sjt7)Zl2zmOi!Y04J9mkQ^IWCbe^1f=Kpqc5E2hc zS&-Wi3MN5Y_IteFXaSF;&eo%wUl#qwE$V^`yq&Ur#px*3s;nR}YF%8dj)+uahQGH&=$R|&ZQlr4n_GC|lHarc zEteY`EYv&8fC8e`i~wKJSluq8QkNB3;48;3kboN&Kjr>CD3L~gmI$R<@DoDnOdq&{ z(Y@~zdPr*1N1xwliCU5=Hw~_75F$zzeV@~-@E+WS7M!doih23rzZOHtQeCcteGBy3 zb_u3BN&3t^84WopY$Ew3!b^X0#&B32BNb)tYZuiO5#yHL6iWvtX)ppa=&p+pQmnLr zXI|9#%;OxUCB8z8p0RcjL;MScrBA`m-%J|uK){3HxJ}D&k2!&Jo38P%o8!p3Zf3k@ z*iAEG9zxv|t9JhjmWOeVT#ZL;%0ooou+a^r_l zT#*GuY_GC-p~25$#7-8pH4IPvZlsMev(i{2guGZ(S~XgjcKI9cYq9S2(0k>Mq`l%bBC_pbLtGaA%E#kN zUisk$E_8DX8M{QNOs8G6=SB4qR2;?J(T_cOBRA~9-j{MQ)}lMcxeu_3qtx@DJ7p5! zL7fl7^`px^9`#Ep0FZsJ zKsTZekGfHy>?Hz|pkdyA@)K-Qw_= zr?txE=m$TF%quNMBee7wf>0cO&ukJkc3^s%;b!~kXs4^s=6x@x-1Lzy2o!ZgJ{A0m zAR|*D9tqsWg26povf3l-13w<0OrPF7H?x4ubJm-_eeMvHv-I;`+TND>+!d!9xQ7=` zaMs24VqaB-8>phhMCc~-020VsLiPz01@dE-4>39VZKK$LaZ|)zlt?=a8n|WJ^gX;H z>l?da4FVgXmY&$X_Og@zYm*U=6EYpUACt$4@ z_o6NrBfB9{Qta}|e|vYtZkgo>=T+2tx5+$4?y&P}X2KYDH`@?h@DgiWCG(#d(#zKb z>f)!`Y*JVnPhLIiaH>BnT^r%A_EdpuPqD_UuX1kvi#e5WWmqZ{3-Fe8b}8l zOP#1#WODl1NgWIupGewOU~(f@_Oi*)p_2!;hn{DJucqj@Pe#h7ct?xX^zu|@j(;&^ zBFTGy5LR$HGe4e!ap2c;P*um;$x`|oK3UH|V|NT-fDQz4V$eqAKB2g9_+E+o)1K+~q{pW4 zLJ5~*rtXj5t22NHD~4%i%~~8l*5=t48pd9IvK8S14pq40(amkV+*5qgd;y|$k}za0 zOO2eN;k#*wz){Vsr|o6^J71MY|9B3z`lHn5{_fOZ}@rs+^&pUbB7+xFkM)UC(6#=y*D6In-d=x1B|@IYg5UGGuK z@$R8v{nSdad465iJv;h0A0{p-uAr}ML=Wtl3u5yglUmkpSU*#X5Z6#DThZbIU!8iE z+>QsPj!Vg`OvYtxjOh}?lUeicP7Pmw7(n5Yf86=j2F5`Pfq*~3LhHE;qLz=0j?vU~ z&Bd043W!5{U9=Ml5S3!T8&GV;D1VY3Xn1-_Kb?C&WJ;_|oyXy!{?4|rGS?49rF<(4#V0+x89qfwwX-YYIm$)QvB=J*i_Y8A?DDO_{y{0d~ghQ``6k`?Qr-R}YA%;w59f?r9{D3X`+J8HPse!o~!PLW6rD9?k((k}g zNOF~90cdt%QsUlmF_rUsVgR<(i6sPDzPc|v2Z=f>gmPTI=JX}5V-t4M4dE0b9 zSd+usye{?qN$nJ!bU~xH7I(_0FL-ok;Sk>;BmcL~)jXV>y9GY@&B~ymH+|%8mv~>+ z3vR_gwy$8J{ebqRF%5op>5vnK>7-=`q@0Hy0F@+Ob5~KZ?l)|0O!D=!>tUk5;;UQcW}4iKXFXJoCj+c4DcN0wW#%@uBxo{xml% z_kPu32W}FTAIf@Kt?$NrVJ5lJ()WWtUFw=o@#6U_wZ6ah|Mq4t*Hgo(qWPlWDfQHa zbcq10&vL0E>^C`vaGPi*?VOreO;7b{yxOMr!m+R7Ngf&A6;~2<9rc_kVqho66W;Cl z50kThTom7T4)S@M(>(d8i^cZ7AuqDs)^zL=yo30?XUnVQ8+MSX(wsZ(8s zLvar`XS&KF_lPf2xETJ8$j0r*aP;;6t{0rACSMFOZEeTPRjIi-OpzhYni}iSK2p2U z)i)WRPouq8VVG366s%Q<|0vff4XaD1`;ibE)purZ%{T@I4+kH~nt8xSs+X;m3fQiq!!tqi?|3O; zK*`9pN-%mnHO>Q0Z=;dJyHEvzisNw46@oW~ z8?Yo8RMIv54&2S!W{0{Xf#LYD+z0?y5O$q;!fr-YJ4Kzrc3RsKbYQKLa7M-LJD$(p z3=|;D{rZYbbndaA67tgbCaSG!?F(h3Kgi$n7#!d}Z|z#g_fSVl-ohY1OL#8_jqZ2u zcptASlp!v7v`ImH`V0j^q7NR?hud*kI;R`wWO^nVbPQJBv^WI3NVN;0(kY;!siqK& z2Yk{l7Cf9XfF)abj(U&*TgV~Z#Rp=obhUx`{aWBSTV9hkK=MU<&FF|YRD|JR%JHVbY7mo@@e+y0fAs3$e6>;B?k4I48bY;*K!z0eoDiKPyO z_r|ZON82K3)l0ac<0~yG7>c_=xUOtJRSYW#Mb*F3abWdv=O z*PN2&C(%}+W^mY(PRDkGpnZ|SfX1$`FJC?%8Dk+XV9b}fid?R%tr8oT3Vq#m3bgPS ze7U8!r8ex=hAw>~k`p+3cu0|&Ygf~hwu=&|A!N-XQtFH_Jze$;a<#*H6_ieVTm;WN z8qF(Y($-9Fk!+KQo^fMdC!Qu6}S7s@`p$AZ%><&pPw^jH@s;sa>XjT`E-PRa~ zDMMx--c_j@T0iCh!-VxIBhO=S%r(L?g%=JlwMvWgY}Nqa{i3@=ov=K4SX*Skp&i$6 zpM3BliiWpN;%D$*G9neWd`qx67gdL=1_6$Nbk846T^7OF(QO>CSVY~jFG5 zeK|(oKDAwMYX0=kb|nAEZPe>_(x^i+7nE(Sqn z+Qq|c?!+l#p85Rs*MmlJwg`Rm6Wqzw-8VBgWFXS2D-#TN(n3O;)(fyn0<$^d$@zM= zXI9)&jzl%!O}phWntbe>y-7Kg<@gfkI)Ap7R`NF|FyZO(#+fK!{$&LPl+LU1PF~!FD7{Kdu9g=Fp#jc z@f%Gzpi+KI>8Jg3QEuJHtaqO+)J%v|d484q7!IM|ZF(>BzAPP3^BxgB`@`INxk2cO zYExzVYWBLA2Ka>~&t5N@r&eHuktvWH<)yV;afog^77{AQz!8?4PtZzur~@kUO^cf? zhYG|mV9qC*MVEjuyRar+%^c{PAR9DsSn}AY*08xyEe_YEM=Gog;ByoKp34Ij^TMaX zR&TMDwvEn(-yyXh>K?J@ zVsWK-0+)~RJ?QZWRcxGI8e%dU?`SXj)#UnbSOT5R$xp%XrStR>>WOJ@v1fpzdA}}e z&1yt!S{HPo^f?q2O=9?+`q*salX;LheM~97Cj$6aX_qCh(SCUQf;s<&AvH}=!Y@l^t(xcMPAR6_lXN(Z zEO^z#H#G`2*K|)r$Di|EXuk7`lo8 zM_xjfX|}Bi!{kE8yP)|+H6sLbd$b#;jkdso0SuFfTeu%A6SIc!z0RQ}5%C z{GsePM6}FF8_St}HUzblY>2LLd<^^z^y{Bf+y4 z;pOI}6$r>PpZ(Xhb&Xef2txp{XpU+2er3CHQ1`d@z~A1Q}WO_VA=UO zc2|gZ#d*rsfIFE5$m9m?ZD_wB^5S0w?HbRTAr(*Mc5BG~C-BzZ!)bLbbr_(#yA!GG zYz2Buuk~C$ij?I`Zv~QKuwqRPRn7P>l)8kwRCgkU?D*f0ow6m-{Dk7-B)o>O%aNyk z*>`ZcRoN|$uo|aNb*5|VY;F;t#V|v>boY3{6fe~WAotmAWE0sd-IM#S6Kz;b7nmMdZ%G{U5AEED+ALv zh=e|8%CMNHfU9|8b?*$&#l{s8^N5mgS4(#f9eVn>=s|kR3pCQ!dzwf`yTcX zjCzj9q?OWwvG+Z3T6aUV+1#FP7C^)}?k&^(EK|+g&!O)e%YSh!MwdHu+kiDgxY0LD z1&44grrntD{@+3!W0eaHBntAFuM$~(v3&*`1wltz@4YStv5y^gz}f$~D4a@c*Dl9j zKk@Yb%mwpEKpzrz--ZxUKaLb|K)tH9S5cv6BLGyI0lJec-HS*|nimLRdBrD5jjDfA z_nSfI^QI*QIl^QJ$&EL8t=G(cq$B>faA1Sy7w81lxDK?1X^m-S8g_;v)Y1^ zImk!3UUUpbM*rvOx2emmfHE@TcmtL90u%mxm!Gt&iyh=hd9gVE@NWmGuYNsShhyzX z+K6F|%0E$AwO<13n`?r4zdHSfi~o&yrO0C0HDxKkRoBkKmCP=6EovBFk(`bvpq23N z2{&j$fb=e6)u!)*JMHJ$9qIeAqXUHQ6p#FUD+z%Owq{!OteHs%jrH#dI~mbI%tqa- zNFPH*osVczXyyT5HjMVI&?`XDI2CRy-on|$3*X78R|Oq>^y3kRT`hj`W^{7v zA6`pQ01G!YJSC5%s6%)L5>#pa`*lOGN;$FSP>ZmSg=Fu(hvA;ci_h%|l3nfcIC^S$ zdzON>8{V^ik21y$hW=TSM9}JDMwFC5K@Ex5;H|A@&Q<1@^U$@TTw*UI?w)`FJCR*n zz3eOsAv0)Y!b!?CNjaO;TDD(P?x36}6z^c(vT9#e=)rJa|;b+2TFW^ij(lI z<(bLxf&)*ujKPjpxYNJnjcp3!lH6=5?i9|gs&H(hw4`<})lNLC5(CsPNy|VF)j;gK z1yHB;woo5%bAMy2{TL%HGQp&+{P}D=UP*@#uyZ_$lDo>r*mix7Xtrhni7anZ<G5EBrhNX^mS3o1fcAjnUjYeb_k`D~+3CjWpD<|I9Rw zc+Gf_^ZoBN*fEo2hLrWU4;;i_cg`HL0c&0JZ~bMgr+mxWp*1QE`h4rl?GBiGm2gJX@0j^>fQ^g?Hg-Jzv@6IlBk9nWXTc z@aWBv*}0&nm(ftv*4vw56wSfad$vlY#m$p>ufF#35d%(BJVU#_>aMt4?AEWbwa{e2 zw^+{AFsGD8GDyR79ASrE`7e*L<$IuWp$)A@+$-n#q%7z}$elQO@*AQHCF7Q+{dPN* zOye`*@PNb9BfIdiAPxwiOH5GsKZf_z2o3OP#I9#rysyW6w8xZV%=ADrxt{D4d+~oY zE*wX`!156sYqeqHU)zGC^F9#yr|&0l-!_5xO8w`M0x2>3AkadhM*)y6#kjds=u6PR7||^asSX^2!wMss!$M!A4Hh?@Vgt z6&-zF%{6g!p&p&uv|4JwSJ(Oj+#Aswaxdc`eH^=s-fMKmVnv8$W+pQB+Mk{2<}Mz! z^rxHhg3A`Q?o%>iFVGWuDx*gD_q{6k)p-0tYzVlZho7+9 zI}lmt zsZ`~%QhuC*0{eW^!(>v(0R9PrC1@`nEVr2|MEoANppbd}^L?}HEtfPe2@lC!z7sa% zi!Ox644Y0hhft539C~B8t4>LZ?mQ|BeRt?(U)P_D6aZ;^Bfs#JdsiL6Mnihkv^EKpFsPNs(iJ`$#$A~nxVG(Gx&ZIB!g8L zct5UT%ygJ$YAnZi$m`xO@eZ?6PjxmuOgK2ZS9#NjZL?M{nU|Tgl7W*4uxTDqUVH3f zJA>CuayOzpVp4$}FI}Ia1u`9uDG7C*AM>~!)LxU2Nexeg%eqq|^53F&*RV&$m4(Y# zg&#Z6-OdD==O|N2bd_d2IeIv`KZN;%nt&WDO65!RUc<|B`IngKM_`n|+UT1(v-O0O zCxh>=uHsApZ%^nIcrAIg`|2!5$t36}CUJbAE{k49%2Qo7kT<-~m-zwhUc2oPA)HmG zyrzTn-9bM+H5?JIWPTAbBqp1nLsWQHrh4dpg4f!(f>bS%^XdjHkX@;==F=e*cw|5a z4BNstbO7^#TqnKj=o1Y#4kRS*b2Z+g=Zrh}m=;!0dWI2oNt5XF&AoETlkKGFQ6ETw zdDely)C9{Z_FfV}OzY`%E+nA&16$l6XpvhvB{>si(R7v2p+4dK&iPTzXZR83lp#*6 zTEEO8h145G4Ti${NXgRqtqu;CZkfEx9gfv`>Ykqlp;&wGFCbIew3I=*a;zkP?bz|rW#{8Vq~ITxZhMlw07SgJ zXc`CJEPzdj4MlRBXVqu4=;)^1Ye|Id?CW&yz} z&`Nh2mB|^_?f`&~yr`W>`HLuuDPwDeJvEXK`eN-M@{T7K#;2~=1y$6R(?53~MRKRi z6ur18AjvU_)>JvRZ6I~QRFss2Qj=@CD}%{Wbm(1^@IEPa5C!(*=N(FYBVp&UEfPKCbeA2 z{%m{MBjZEKcS%h{!U#9y{|~Q507WlmY2#w*L_jZQW9VWkVrpz}VhY8_2j%SI zWNK&&<+0hK=Eyz(kHv7%Dck}dR$IPJ%g?cw0f>}U4 zE&4zbl!?24b0jQITuB56Y(`AN+|7@HI92uph%~T0Edv+sd>ZV664XKGg3=5uv1&mg z0P0CGZW!PLpxlr;1ZYQqer)VS){g+lM<7)m=|KR=P$Z8_!4Pza=>mcrybTcnjb%(W z%OF1-PDG>_2@lPqcPMFeBLcv4aj;Mi5E%x*l@95FoE=V38!BB_aAD zo&nUyGD8OmTsbAQ!JG=f041`B zhnq-|00Du>2r&iVi5PCQqAwxPK zLJFoEL=X)G?38+O64u~$<&5~^f8o@bQbXRxJ!2zZ{TgoVASw}ZmeSvxogQ0&tF37O zklxatDbuE0LpX4wSc%CcbD)EdA%i}M59+vhGohf{A^K682HmnhXZqu*NdIQN1i7iJXAh6vj3-nGET)Kp6zm zap%y*8r~Q`{$;o28g&get@OFTSU*Hf{MAdG2-?qmFn&rppi&vwAA>${xj)G;LOO(m z<0S$X9>*?O<&N+vaWkRB>_Ms~=5&~ij=zmj3DwjM0# zA_)D9Hc}u)9UqZ=Y6y^?0p}9MO1Z(Zek=0Inmz^Q8bKon%ct#MQVw*h)U}=jr>dFw zy(qela9G=gC-*jb^>r{Tkcw<~S(02BwKThFP5D%fn=1b0w=qc0 z@Mgr(OXB4-7ckOLDO8R-axa^$^0K{d7KzW}#SC9S;=`DZUb^m=Sa&6*v$<#7<2Lf6 zsatAfb-!9|I=kSZg3T1^M4fx4mloOHFJCa-{+Cv%oy}eayjfJ$SN#j zu=Q=4Vb4 zkoH`*;)a0<6jh;<`w4i}ImvuGcG!KHiar!N@BeN==eX?Lj_qSI4AC3^QL?YGsyw~l zfFZkj)Qixn_SI;7ObUnIbTdP}vZSNt`EyuWy3V0Y?mf`MYdIe2N&mepe1Y?Vj67_% zZms3}xArTD9~D=2&Zy@x*iZ^r&CVi!H5y z)OIz#3pp_lf`?S#CDdG zW`)7$K{Vt4*F|yQ4AgJD@NU zm{kvPUhL(Zr{=yV_?%N5nqzr-ew#?(+wmcH6xfL?o%2rv!8{I688_=`^N*Er!O(P0 zLB4Zuibh#}*zQiPWasOyU1m`r!ghSo{h_DDdMV~!}#@VHrCA=oyLlmSh?qo58i2~ zpZJH!V%ef{V%a9-qpFg@9QoNQH@;A7w~aC(UIQL&N7wJMRDt_na};b z&y0(~mW4r{*{)~sWu|R3u-=8_N1H@f8~D`3Xyxn`MvJuJnJccu z@6z$VTqtObZbwaqjZLdIz#cpQ#apq|O|jgz=I>-!jxO?Q`BGb6Ou>I;*K$yECTkum ze8S}E|2-Z44Tf^(rcGo27;sf{dAy!x-1ZqQ<&HJ^)kg+;*@900055vg0L7j{QCT~q0WYV{B!+16#q_oyT|Dc&^4gxZPeR&6e7NVSj=dS1+&b z)*aecxe@H>IiHTou5N^iH<^s_$0G7~{PR(6pj#I49{OOm_Vre`m}9Lex{K4)T__RR z>Cy$ow5=YC?b1_=JpZuuboKg?dg_Qh94(1w^OtY-6lB}DQgWJub{=R0DL$(#XWvcJ zq^tI`SXr%jQNQ@l-NHk~26IQ8vh}`7G`_mk5mzGiH3z%3D-N&S>Eog5)+w4h#)XQ* ztVdnn%*(w@9Gg6k_MYqa zeNNdrJEd7X*8H;@>@o8Wq(N>`?~IUwEe_w2*K|KA=Qn>S#D91n>Yb1;=}FP<05YE% ztUvKMzi~*e-0}wLMbj8r+xP-9bF&6afgJusY_EWMp93pdUk#hEnT0Q#)2QdJH%D) zqCZ2o?e$`dq#x|biCR&Kj^N!(VbaQA|MGTCclN60wH7Dr;F=9c7p}Y5_@v{#nHLil z{1^rZ_G>WX+K)SB#TdB*l6kONHruTZV3#!A@EXxft5FMjyQDZXwf=OLVfC6d_vp7eO2K`3Z@o3Vx?E>G$7!4P{eXA00%8AevXb+Em6c2^|05)s2pBn;+5Ttx|Hw)P29Ezb z2Ic>ctW@#lQMte_CuJ3aq)gwXJ3?YM)+wfq7Bpd*mT8ckVsv(xp5|PdUT9BAQIa4m zRqBkC*r1#2nNPLLMj}!%aouT+-YnhARMh^s{{8sMnag3md-A2*d-Lp@lg=A3Qr-s9 zz_<=AZ|xK-JQehAIbArk=zfeN5cl?46G$YCJr2ND*Bld=9`D3}48oj(8Lj<8N4yB8*@# z`b#Lv45W6164M5Z3FD_mij+uU^sP?KM-YmD5~cugj2z1zEH=qswjlbYAMg$#fZ;AI zRUC;mMHl~+%cxh@X*4wBUhuXY@BI`ZVBSQ1ley#FYr`8lfq+SiJq{6MFAPGpHi=>u znkR|9IK4YMnKYgMhPWX_Asu}NCpm0nU))NRi#<3Cx;E;zMH&;$ZdDk~&7}y+L zWR4F^_z`ThpuXV;p&f4M$o$hgXAgoJOeHITK$GvFU@XFXHTZ@j^jBQZ29X}l8lVfP zkKs!^qo7HV1BywG(Af$u(m!GMvGEJ>{ z!Wcf%AlmQF+WbyD2IBBf)Gv8&aa~|*2p=h`P=Nvl(W9WDJ|J}ed%_hFSqMFVQF0Oxb7hpl2n2$N z5w#$DX-P$##-u-a?-bG#sM%uzLE#ArM2z`=%id8P8i1J@1f%m7o_GvW4RcZ+n%+Bt zX3?`9r$>taBIrg!gh%hTA#^4ZF-%%R*vvz6-Fv^cesuyJ)f(N)G<9;JAwkG%7CO!lqm#Jfb*NOyx|Z-# zF{mRPWtOkBe5y7l5E}gsrq@~hY&;BUH+oOmp|oXS~qlJFXrxY9U8inQ8vqbl&RQ=YL;sO-xBD_1R$E+1K#C5^leYluFdP zO&y`Cx#GyD;r9sfRg$5ek2J|^`NVK*g5{t zcu|NGdSX-?x{Oyj#nnfLspcd0H*HFL)6_tL0WB)Fta{*L>3xwrtf<=YQi%z2izM3B z<;nKrWa#xjPu`gG6AxSoVjqDeK}##Ae87T5EauhNyPNw9D>_=Yd_I$>VmzHNU_`+4V3XL9G~Db zzf=>YOQY%lS?!g6jNJK*OD&Unv$KLwtEk^x^ebm}rj&cr*HIHGhSa5JZ~D6-^$>GO zisIrt%<@~|N-Jje@ZCqX#1z`uU;ys zt=nDt6NPh?YJ8E~aO08E`=idpGn)s2kO1g;OMmrtD%mB1n0|=O>vR4BTO#dMj<0Z= zKNMx22Ie{LOEL*+mfS0Dt~LGmBR606NrA58cYls>FXY#Z+p~Ohv(GfX_}s=8r~#YX zB#i6Y81|ALu}@b3Ha)tvM!B!W&N5qU#`L-KwqT)D?0s|z*!v4GU_OnwWg*oTt@B@` zlcmRjer8U2zG35WdzV)Ri&;t9=ixWi$ddexskd__R~N@4&)L?@xCpEb?w_+J$->Ez zV}kE)x%Eyz4YRXw7@Gz2>we*31z(8tJPF}YG3V;7PkK@uY%=hXb|pB$4pf4$FLx3uLY-ppZKY(>s3I!UUco%0=g zKC_vnDL8k##PrLqRoxG_14SgT5nhOVJ(^pXWR#4gTQ69MKUIFpm6~jD&axuk_z$Jl zlw9e{&RX51Ec+IUx^kY4mJR}7g=*f@igw%nw>V6pnqMrsiy;G{TBE5YRu!OKx|n1* z%sYy@#1!iSJM){i9D`2=n#&A1BayIfiaWo<&oxK>G~3r!?p`!@(*(_jG zX#3A6>1>h>WN19UBBkNWwvCdsi6n2gGv2m78OhIv`3~2kj}%UnuXZjD9N4H^W{(Ae zh229_@+P!}#~$;!JTshW*}uFfJ09Q4%7vp7V&aYUa(c5CF&OQZVK>ZiOA-odX!!VN zUZu5|2xg__$2`Ohd}(cUx|J5I@hPkCo3ke+$;aK*PgdeU5Y2M?DlqkTqhs@X@MMeY0Y|7 z#q@4~sNa|lOE{a*at{Ie#J!8P_rR*r_%3 zN&3)3x80#3$Cc7lzHF#qHao{J#zEDMy$mbsCIq`I@ufz&mQ}Mb0($P)0=migfZWGV z0PbQ0V>xtF{kr-ocw)4=)o{p8^TO#1U)!_E3=&irn>w zibJYzsf=vMuiy~-4%>*|i@s`RL)rx$ zF;uik>;A)L62K<5N94A#sB`L%xpn(0sYEw4+zp900ERu^@YtqxvkH>d>$XE!M~aktkw7#8PJXQ+kk#0Bv` z+uOK_KuAOcUn6e0jV`y3-=8|#^jmT}m%6i>(B|aIEb2ov*;4_GvIO+d;8dU3{soDN z$YT*f*&?{+|HgJsp!SSS%uYq~^AnswGXiX~`4vs|_pkx**TC`&0Yi%fiPqLZ<;?*a z_z~t%51K&!dI2Z<>odvNnGt~d?_A;MT^xa#GO#%XwFMAZX(7Y|CK8<=U0qvR*#Q6O z1Vx1$yA|)QBUJ%_GC_Nh$+3B9`D2bD<@HqzLCL#51Y_6$KFJdJbIgrQ_5V8v=p4ue zP!n^}G7CV%1*K*M7ED$A0M^aSjBL&CpopT7l2j@t0ZC021uTFi6R>eJGqd0C5=bDy zS9LnzL#R86X?<&9^fIEQBwzXgDH=S)0f~h~>5<$6Yjcn{M zGr-)YfB%f#+8Q?xk4E+Oc3+ij%}Uh_-0$6DQ@snYJ7@a_FhOn&KlmQsKwA(sp<|$p z|1ag8ml8l^3mShs_MKZ|^kQ#TECL0g1OnIF?ht{Xzu&6!n;Lr`EW}qFE32vP?-6B1 z1tk#9w6%af!5EA&AS6g;!1gvk$?qD$F27WVA0h}4@wSeR->dwGUfN~v<`&Jy834N5J-v3L->ZgA2fq>FtZxthK=JTJliJu}KjN zvjz%b5V)a&V_+GuA*fftZ^Vy|oVFw)W2E;NfUK-60GF|Nu|R5~Y-}vTiYf!)Pm3!57_?jmm zjfzb_8tZS_8(ky!8Vuvd0f7}bfw9~VgT*(m3|xGDh&E44uN^G_G&tR^-#Y=ag#kQJ zM?0W9=#K*+A>|jkUiee*I?x&l$??(=OZw03*cUP>Y=eEn|B4em+yPB%YD{;?ui`oO zj`jiH9Qj2zkj|cG^8nV=1!VQbq59Twui)z(9Yx-|)4R9P=5qy+S9r)_P7=Ed~IryvoSuH#fb8UHL(jV7x*NOiszV`Qf`iu{2?#rX%XCgJ; z{^7x7KU92p08}9^F6!{LegFA4PC3et=lhqNB_8|U@BY>e9RTFt)s$C7%cceyyy9gR zQw9-7mr)y7gs96=*^fA*B<`;^mUTlw7C*jK1QLk~(h6!YkIF_l-oXGZ0t=FKu^7P&2pIkIuh8kDYgsZqJFlJXBqgtXm%$c^7vm^sHkI5E? za(4cWKFsp&|H%=xP+#xS1-R8f}ezqGTIY}=%24iR!sm{a>}M^gRk@XAfjpYQIaivzoQbG=ynG*#w$SBZd&(D3EolziiXNj29EkD|>8726-d`3<5EarD@p)8V-5v4l7jU&h6s{1U=4#c$VLD;aR*JaHs96~k3yT>`woZsV zU(T*25(^~Rznx(-@pXVlwcC|FP>!9VN;n%<#aM3LJqylTtjL72Zy@R}|01fVNoe~u zwG?OeW35<-5skb(H9j$U5DXc`>xNo_LtYx$DDykP@x!wYnS0876QQ~F^b)=$NOmoP z-lGHkg48s+=1sL?wSuRxTIRp39fNky{Yy4fuT_De@!~)<{+5-Cz>+7iAeHX@)ZLVd z0Tc4Fyf6%|=KkD>VKPiK;96ra_c_T+=wI-pbrvpO_mKS$lPhG&l!#Ze+u<6lYL$V3 zV4m<{TWD+(evo@P1J^_>qvbF(Mde5>*<~$EaBF{yu$06qvwo9Cg}`!3p`8_0n47+w zVKpDx`_xcYL?g8yn9{C=9Y!tC{xnd)(O5oYru?BiR#W=L$hBO}<`#MN$qHYK%~{)1 z;3eef$YZ5_5h?#&HmEPj>Ay-~OLp-z)9rq)tC|GAl~LYNMOJB)JH(VcKqpK2h$cSp zf@jxRN_q&`-B$OOA}vVr47Wb=aN>jL2?EvvrC3)1iTHd-*MbdZ(_8&aRCr#Oa)134 z-4q?zI9SsUUxL$%@=Kj)e9(im(*@2@uaUQbI#md*7~JJWMn5MabpAs@>xDv|lLO{7 zE$+?NR2hx#l_iL^j#>hCpn=!z!kdjhLJH6Q@RaN!$m(}IDgqG#HuctHWA0hz3y4;X z(b*K;*~+Gea>d|^V(m~fy#NmGTv!5YweIb5RKu!xAdk5k4)w2W6dD6kK)?PNlN-S+ z?TUW{u$}V=i{djDy(lC{;Y8eV8iiLBW>NGF56}5VvgrLRAyEFSz*EOYN7PKzSBe_* zM(9%v>g9R8S`YG7)=1$k#02StaGO$2eIR;_Ks&kuwC`UC)B0u#q`Ld?+K3`4hne?~ zI95gH-4`l6$1Rvdyv|`Fl<`M&cOHEft@-6|)6-OXoj>aoJu=;Sq~J~~)daTHLQ;or zpyC}j?ZxN%EXme+%rRes&x8+RnVnCMe$IBOaX(^+JdbNp+Z1(24%9&Nx%6Jt5rk5P zifKd#!&-@KU&$-B)${Q3@({_Cr{?BH`<2iuoes)2$wvRIM&uvGo$0?PRBq8r>-CY8 zz}^#_HC?mLL1dl4oFMC!1#*xVjA0P$(K2G_4LeYdWaG~J&G&*j^JPu=S7!Gz%!xRE zQV_cP<-N#7I%8+x&Vw!6s&>OTf6nJ>ar&s4Q8fa8uaa$WEPmUSq?44s2PbkauXMY9 z?A`Iv|IzzxcF;sSq(Vu$8nUOFv}P52a=j<$%9w+fKTW|4hSZdC{0>>U%{*Cu4(z?w zr1XS0vK`B9N?S#gWNeE6;^Iu8l%UqPdc6psbj8&4V*JK9XLYx`I=ab6ncf0d1`AoG zD}2M~7|9S=jlRAV``2D`mKisui&@b@1HI!9Xi)g)cCl-lm!7GIs9`cwK@ zqi!*H7SK;L80+KE%>uV^zpc#!e!vq)*_`K;EvHW#SQ*85sD|hekb=t+9o)?oc0kro z#9OA?;MUlg--17{&A%vSRO>JV9K02?D zji~q%<_Ohfv*;}o^d1~qBQ3ntYEF=M!2)|80`If^fThzhLviZ4)7n&(Ok86 z-5|E0X0HMwK#J`+l&K4?O6TDtZWE`L{2t6$h7X@z+PoSDcGs`S5bN9p!ZfPqZFhg$ zAkx@yx9#uwMO|^rR%t>fN^=c_Bq1t`2I3gD_50 zj!P$BdY+a599F5$fJy>xZJ}U_l}+>_c?ZZu+Grnys)te%`*B3UOd|b79)8FbrJq8+!Vvx(9#%$F#AYF-SCIFlK*@3aT%JM1p zD7_F9!Bd{{i5a+aSvsXR`T!U&TMH1gA(Nj9cz`ioaAc$_mHO`1tM=CLc5c8wG}_@Z zznhkBh2%&v!Ma2=aHR4bQbOkc6>vSAV>D-I zjj#ZL>qJ(?{H#OGJ^k~UkCyLqqtE~;cf%is(L8iYB%m1uDusGI_-!V`&#~=n?Kv+) z6b-m{XyA*@XlAJX#p7yJJbx)|8sw(*zP@RMcZKM}Gye)o3MIM9J{j83AP^MkwZ!L^ zA2ciLH=lZbapXQ=dL`%)pIKPon~97pZl=RtmzY*cfA`!t?f*T_;xtn%j5+(yTi_d- z`z<*X=r4yO^b*8Uc~zlzIGQ3_Rl}t*KcZ$jKI>WbnD7SN86=?(IM^jAsbT_(YCddd z%aM7D{rdJ~=(VUsq_k@mk4uDTB-`UlN@*Lo#{wr9p(|15KEP@4 z1g&6PlGl~kbKIeUw|PG{b+d(2&v=GMs^ew-8(sP?zffu_ZZW(B=wHl;0~o@%Kr{2* zwV`|4eHo*$3neYGd0w@eh^_PDS}NWQ6gQ#J|9^*7kPju%b32(rgZ!HvSlKV}v!OTM4aJ zDv+cS`Z&tgxNA~(M}?LD%~VZ4W#y^1Da*UR4)u1nQpLI~$28)S+w_KHQZ1v97v}{r zA#QPP_l&IRG?My2iP>B;JE_p|LIKesyL61KRHel$qS)SYBlZ`5Z$IRrHr}Xi5^B!L zDV&|r-)wDYJ~!^l)jo@p32AJFuz0!P6ih9OH6vmxZ8x=sBWPd2Eme@YCg#V*Od+p>kG(cB=H0+mr?2|Q_FD8gyQj4?D|Cw#!pm!T)!x7#(V}jv^tJij%PMXncJC1y=}gWWlJp^Y3!ZUjBZr>@=HSrx zr1L<(CG?4Q3MKvtS%GQVRk{U>ho^T{rlDUb#MRHzX6BdxTn<5zDecNGsEmK zwTL5WD5uWt4NG=p3rYC;ox294l;Qgl)5#Tu;EdiwMY-#2AcISK6$wQ1Yy5Wf~&S~?#ESwSSPf3tFVEVG8WDnjPE#3 zC`8Wjf1uFfCv|?2lvcpcH3x@Bw}#^C8BYEQSE!^{U{BiH6(8$c-$G^oW5xbQJ-Lzz z;wFCP40~&l4+8B5wRm7`qk0+Q)Z0AS2UW0Yv)BL^u4I3{0%uB_ z-oSepuQ$6eBK?$ z*+-+UYya3<6+h`I<|Yha+`MNh&x6|yL&M^N;9^&)LpV8wLCd-0oWB`*l~|ZZMam;( zzB9d(Ek%OdV=&H9y`@ec|HuN|q}@=eXc|=1H;n>zYO?y*NOkVrMmb!uh|ZWKbI?~| zjv`P$qKoH&7sz}2E>15|QT|!^_IU$8;O9#818P-`Zc}_9qHG*qynw=uT6x%RAU!1? zsJBemnmD^dPA-ZvMd&+p#q1oNx>~s9B&35K6lapRrCJ|P%`7Jigh@{0Je{zT)V4G_ zsmCCJT^}+;S*vVhV7sSdk}IYt$E*WmU{MT4=YoI5TwtA!5h=ds#`uFH4f4>JePM$8uy{PpSdOqDiaSK*!D6UjFZKBil1{(F zo&O^BGM1@@a%Sn4e$DLmhtW4#PPdTCBhI&A3NMl7Z8`ZMRdBTUmVzdy=clq#oUGwh z`iVPYDkZBV&i^fNH;dmyJl6O%l4hCbRp1{B5aH&-ZWQpnIsstWfI#qvhYKhdH z?v=^IylL*hRe4`BOH&R5Z?K%5A+659E~@9~x+M?m zfhyNpV5CKp@3KBy4yn{{r$lgBUfLSo_lc~)HPqQsQKEmLI0g_7 zdH5>}WV>y$`-jVg=Ay!S8?|LWhY^F1=;J>N{8^A0zdZd-qfpaj^ zl#J*nvJyh=Pgn^D$Rp`9OW552c`2V^ogIwQ>25oC-O|wn_7$vqvW@dhWgmS*5EA zW%M<^mtf|?3{B)}!DmIEl*o@XZ&hK_+D(KzM4L1o?|u!#>YVsihyYUeGG~YKZE~eh&U?xDdt(2rKlnz7*g8GB`%y4tT){nTZWs@tN6$7My<#onm*p0Ua|7gPcj9^e(X#tE% zIAf4JWzA03X23tj&)GB4{vpAHfo25*t8>#m-mqP!_c8tx@WAgaCB+bjI$Al&-~%Pa zeWMUY4W1Z9JN#YJ^|aI&6GZVir^&->q=UQ8906^G5ycoEV3y`D(bO#(9cl1rsXISu z#*sLMu3hbor{o+D+{Z{b-8h#nf8p`z@10)>sTrif^PO&^oAm(jxtCzW)ACrue7~%` zk`-tuH3*rgwpzA8*~4GYJQlNAn|qke3N9Mvp>#Y$kgg-`GnvkmYdrfGoORC~;E(ZUW()eB-w zq1YgkP#dnfe7NGHhp;)r{y;zh&2l7kq*9n#C}UCcT>io8W4VP~TDimx8;y>HVI#W(s-P^j&|) zG|>jN2OU*wPgu}{3ba=nbJxq+32C|;2KDGmGapq(uZl-t%qrYNKOaz*>dC6l!-c&~ zN`{G1Rz0f=pMYvLe)*VTv~PKvU!Q#M<3WP%&t7CcH@FPz2wzd{NJvA)$J^xmSW1+r zr#Ma*V`G4-Re=!cL>^=bNV!4JFdV*6ZElqARRU&L`;BSrwA>r40=xsb1b)NRE4A)7C2RSy#NRnZr$mEs0Zhq#V$_goF^wP0J&- z%RA)mIQCai`LaR+>Fl|LcBAIjo|MDpH<$K1%oPk))aB}tl>?D9?jG!5Q=TOS8s}o$&`LN`oTZt|Sbsr7%zL%ek`TQu$sbuE; zS*8FaTW-_uNZwpNYS6mto=CS{WQhv*gC@uLG7U4i8j;hZhbozwRYd3< zW+QNXs_&mU{dzd8mWM(|WaMuQVE0e$)h=WE2iYG^hAW8J8g9HvHKQAnbtm-|*~%m( z4G#J$WyhYXEyri!qXr{81UHjCNaI#JhCizl7=wJUMundb7lkwB_Nl8pZ`;U5N>ja} zSJsDL4szmOp5BOr+5>xyz=%VWWkcL*J&>WcWSA38MM4NfZX*`>U9gdR}On}Awfs-KuK8BcbKiSikG=a)K7+KZvq zK>r5M?O3(J8UA2=1FrJ>B3KXwRs}$-j)Ocag1VK6xX0Z8%6lHKl1utt<5wsqfPEVf zfSKBnqd}Hho3|*QAFXy?t15e@7C_!Gk!37Ce%HQHzv4L?p^DDUP1u-&rK^Ao1aL1* z*g$dw*g}(45~{+S1!$5OwIp`gGHr$wIQZ={91EvCFi0kV7IHP|9&BpD7x*Z;QZCRW>bX@h6-PZ* z^p{R_7}n60U1KXb>pUF{+IZZnjA%=8ctfoZ!PT)&JGu@)oG=Ej6^5}s=f2507YBpM4w1;Sp8kQ&C!RIqeuT3=>`rdk3LmW(Mxv%;cUid zlQ2QPdgtqqy0wn%?~m%MoIUF!<})|JEC!`(MF}}3(XWi_78Gv&&zgB@92LPX9?99^JdjGWP>iy@)`({5FgdioJjp zRnBTDwQ9J_LPI(9=|S3h$wLE!`}%|SrS9B?Qem)a{EqRUmvv=?c-SzQEP*!z5)nav zbyS9m{br6F$6W?0w0cT;RJi%N-9EV*M4|`wchs2v>)o~nKLgkfyt6!Sx%{-G190Ok zXR2-1DLPi&;shxN3IpHJP}H{IQbdNl@k`l|bXc>qk8OUZmwm=1dpw9K63ti;W*oR` ze-CbmM`FdoF_wQsBxxV@pxKPUEj*c4J97PT5)Ez#oA@oV3HxnfwFQdj3Zy`>-*Vm{ zH7OIoWT)kQG4MjZ2t>_*PuJsXvvNVK*O%#46rK`7cUocB$@~+=6P(7Ba-9> z1ozKV-XgeSHZ-fd?q14CLrB)!v>S%mC+$!t*Jy$SQ%|p@c7W8p*V)NZ z!OW+#{yL$lnANqSAw&s>q3S8A56q|T=B;=?O$sdbI%py(Jb}5)j{Yh~2~U+m>-jZc z6mtC0CQ~lEJW)l}#8mkhsxV1Va$(OKObH&9S_qT7WBasq0f_6RZFOCYDDv#`ezpN) zU3ygpruR#f4e^k4KWt;y1Eu=u_N4;8onV)&ZUo__n8n`q0C1P zfvT~C@i<^s_avG;!xdd|k!8K3Zz+*byhAk7FViNkCm+vcSyET-4w(lXybrwbjdd?# ze-tH{Mh@QnV`hpI*jY1&)I(RMCJ}Rk&g1fozlj6o$U9M!_uoUSgYJgJUK{W(--Bhd zjvM!ZC0X$mzY_KKi>I*kBEFQ>?9_41`fUWn)`4wt+}ZVFx$t0AF3I!{*a2IEX`}I5HrZgkNgY}QNbQyIWarY=gxaBcaD~J zW(DB+32OsuU28+H4C5E?W_}0s5B6+=ZB_|BC7U1ZS{9TOKvy#H5i)mT@RiXhlZJb? z6R^|UI-@LGVZa}o++&3ea(p;q5(@iUHazW+rDYbN&L8qxWETaYnT;VIlUz@GDr`_( zUm0DhSJZUHEf-aC$7077iA}@vhGd7la_mq+9C~5LHX%tI$*|wx`)+(V{t{n>6&6rr zy)Acou!ZH&=MMud#mhB>H!UCC+iQ-i+wpD^OZlXA z5;-TWy|kjLJAQkx@a7q*XyLY>yaFBZ(595Pt_b;FDqC6;@m+v%7;(?M(* z#_I~JovHe=bja^zG{pK{!k!t+&#*(ng7f~Z$HJ6jqIkH4B zMbrLz(m=BRDNf;;QMiEoS2mDVdOfq*q4_Z#V7e2QP-nplkakmf^&iD9rIitZ*KZsG zuD{Mu$~BYFkUqU>y6Df(mh38NAPxI%`DkS?ud|xPvVE`{iQTXq@qJ7rF^!l-+>>=O zQuv9XMlv{Z?q(rFm`N_Q*LKEgK&F^Bn3?qDM&H9Yo}` zNHxXL#1w2dGRzy9Kb+PaRyQiE8+~n?V*uPwq^4Yw?M*PoFGaG1ctK`6*QIa%qM~+9 z-01TK7zb5ws|%}Hi)Ltcb?Pts(pe6JuZ!)4uKMM&~51iP>Pq7Ota2|*23X#19v^p&*(h0Q=F7l<> zr2ZYwhn7Paqj9s-?hL+0SfEY-)Rmd1>=+E_vvE=MKRJ|^;z>s~27iEa#c#@|nA48K zMB3I$G>UMWV7<$cxk}%o7ME8bO-7*Q@XYTW;Op7+q$wf-R0h9 zy0&BOr1CAbcPBn+F9?>|PeDsy)DR08sK|vtnjB0KUCs$k1n;r{&8Di{HH_HlM@bFN zm*b{Z{e1Z2PN$*Vz&pKzj@FXr>`y1b8Z2jDe4j{jZoycEBYD&?ZJo6naqZOrUBGZL ztsm2hq!d;|R=5oLt+Miw-G+f|LU1fA8rr{7oFshNJ{$yv)6q{ck(JDB0&@f((2nPR z2fXfUec&F0hRxLx6OLNSZq+Wp0&tG%d~%J{FG8U{1R0|QD$<5rEuPWL+dUtRNkFPP zfj?V3@daw>8}KHlDVCtl_&l0C+<^7oQa3RVX!pl*Xp8M|W<|8ON}`}OsM!9gP^RI& zPj=xkN6W~&4;ubXj|u>M5r@8YrXM!X-{ujJh@=nO@lEp{7ZJfUnwf~rCQVYK!~f;z zWG+kYfWKDnEwcC_1Y;FVt)h`agmq$Dbx_rR;d(_nAN3#{S1fE&^!kuM{)OIYTQaO9 zD6k(DmMg#b{$d3IV^=O=g`WrrI1Uwp^vzZeTvV3ZXYj13^A9|0yv6ue{xy$a>s6b> zD_0DLHHWB2zkei}cq5s(cSEluWVQ;_{rF0)uv8e>N!B}ud_aN;Ftb3hY9hAox^GfV zrM1akNlZtVIWPJZ9Ex2?=7%9r)(D`>`t+Y*O=&MAX?(hDKOBPmGu);`?*T4zvnAIs z_aFdvjC@0~AhAniLl3vq;rJIh<3))xwGJjj*~E~HT`7fngBKHyceF!XI=0s z^c>P^(H-%K%q4eM?=#7MiEq#{WJ)B~UQEysi;37?D|f~eR2uTdCfmFjjuZTfw=}-% zlP>z;OmsF7qhr+6B)3d)&6$6gM2>6kV(?LGf0~}r^oR|0+&8-HbAB{`(wpIim5f;W z$VWEhCUSgH#UFU)iXWoKl7TPG@-e<`PHSurwi?uHQk8YQgSFm(S3phE(sgj&fY8tD zOe{~X2Q~5hfkcZ&)PXyfC+?!&1ucyJ2j*Sa`I_fEZR5#V!y*^?5l;mK9(PvZv9G&> zP8~RbVY^dyPd0k8%_KLP{=H~9k4bQ{WerriznfC@lxZu9cnpnQU=r=n2!^ll|FJq< zL9T>n>PSMCQ9z`rH=R3`lF`;njQOY@YfwNGV%23ovu2bJYjH<8*4Don?3p!#IzC4t z{9FJ#YsX%BJgLzbMxMT9F~06?`Q-i+t7>0F3hx-FU4znO8X6L_3Fg;r!c>Dg;Zq?? zgQ4t!bEGf58J%#0@=eaQi~FM}>B7Fs?!rXKNg8H;$3SFWKCT*Xl)pju=@^Zjs-pq5)*!5%2MT~DT7js&u}k@RHQ(eGqMSlVXQ{p5`@%qy_jlQ(r{m4h+!!AcM6n3M=S`8Y~sZU;D8dyLnC~(%8`Xs7En4oq?oyO>_lJ9qvF)2^XMx&sD0F4eaq8sM~jtIEeaNaAl6* z$_nOs_#?Cz^L0hxx@AmVNdFIx?d*@$&+S~@Ux*W0Ts?~=#+xbGlBRTRa|;jV(Dhx; z_<7y2w)%XO2*DP)*u3KoF$!@B|K&a;Tht86#4cL$_3{RYN{T%g7l#6|!xyZd677lG z`}9U0>y??$ZD}XK$V3+Ur88-c-38&6Db7(fG#;{ZFFUezXX+%F*u`c{Z&zLqCOqtF!>XHvO}>C|2VPR*?^pLOZ9`;MNkU>(7|e(7lY0 zC&bvOsA!v=n?X_+LC~j{=>LL@$`*-D5z*djZ^S*z6-gt(o|Ahi;0`!rD%+txmO71{ zZaqL-BiFp{zAj+K?sv6>F?{&)_ch2#88O=M3SGI_;ZtkeGLF-GR!tMhAo`v2JqZD3 zYYrx2*G%f0!?`G2UDjaRNr@Q2g364Xt}?h9I}HSb#-Z-e_2463@F^_7G;+Z$pXTG!f-MdgxHuhD0x~d3>qlTt`(Y z$d%IzCV=&{-?hOugixTxuCjMuFeZ&NdW!6?t~w1xdz2Sn$^KlJ4vPMxde|9z{e{zK zL1(atW}iVU9HZ`M<9@__ePNcEkRlcBoU&F*A)75Wb(bzZ;lVz=Y-jM05dDU?g#7?N zHwn?a|B1=g3f&`eGG(U)I~wB-)b}Ae8V5NXugZ;ldpx>$vLjg79ntWdbfuM<@_0?6 zMWsg|oBNDimpVwr_4{{JqYD|ymh#YO@aQ>=1oLDU2D^wmyTC|>i{a5T+w?LOi&2}N zW{CsO&UqVX3NMcO{<1;e3gUMzDKaW`ik3TDR;aH8KpK1%A{EQtdsIufq&_xuaeyX4` z#9MnxIOi$5{?LLYMhjIt1>1rKkFAEAUOIV-x)d26Nc0&ajD@;%m*W#^qmSAYb0{Gf z3|D>2rsi_=8&a6h3AdHH^Qw6&PFIU)x;xkY)e5u0arAm| zEzS|MAzM#d4!%bf!wkf+-(WEedk*TuUb9wydbcrm!J$*db&0d_v4G?4SpTx@dC>l4 zFDF>SH&0Lc;r2yoSogKds@ZJe^d^PElr5}VEy%}(X#Z}Cm2A=>$^6bp!fHmhk9?7} zbNkA4g`Hr1Y))aPIHNqfBRQ6tUsfl>5+X|3l(U~lUGU~Z%}6~asyy-#U+4&%SK9b;Ib1a+i@vaa zF?rbDW*}Y7(7{=WH(s(0?R9Xi^N~cq)u|#*%l2j73m0`~qE%^9a7c4*q4`9@Jw2A; zu#%Ds(hpY@o)1~P4A|09sc*L5c>2oRSLw-Ji9{g~IxI{`BbcN_%-uik)uKk*OarP4 zZ0(v{4}Td{(HJJIOc2g6Qd(u|Ap^@$aVwj`%(7Yr7<@tFY>D{mIa-C8nQAO zW?urqR5z{mbjiK6sB}x7TJS^h)Dd!X_H+*J`a-(-LdmPaVv4HtFlpTovo~|7EAnD# zcTD(iB}FGNBeU@n5zGh&Cno|ln#W{O9-3(AF*`z?fFNfkj~YaM)zUN1`de5k0ha9< zz<`Hc?d~?>WWVvd7;WSasX{y1(9j+c^yg0ieueeu@WdtJqhHF5w=Vti=|`0GlWok~ zO?%GHpQ#a%KA%<(D+EAMMT2%4g}Bkavbu~Yt=dfHT+v8udkkCs$ckvhbbe3j-zgMx zyLY_QNAEf&-zN)q(BQ8rP~xCN)3dKu-$0>rCpXXLe_>AD9!oAXDRog+zhYKr5J##0 zwtxIE6I_WhfXv}sI%u_PiJKW!{2IU-=WNGcP@kU`-sEK%G1MohrypOuC6DMj$hxoI z2erJo>;I;e%iNa#)>eRCxX71)_e~vi1Ryw${^4%+G`psW6ozETRc9h0?p#B~i5$rO!Z>qNq_! z7!^ko<`1AXD&da>NPSfO59&8HjLG$bsn`{(r_L)Cm@ zHdz&Norv0rnAI|_TM$$_%>d+kGIAv((1)~@JG@T+T1XP|ZT$otf%x)XSFdlkoqd}r z^2`go1cm-y->Jv#{YQM27K(`7%3O6eg)C4>Y|LYhXzJY?07=->^SrRAB3;lS{g+C+ zS7$rycQql6WU*zkGzK}+toQ0C=U5+6yW2lk-)DzH@mh-2z44!-U(wI?&Hweer-hy^ zIk&+<2fH+AK*b}rGVVhR9FVMv=$LTINuc}HYlSr8z=;{OrFq(N)Sd)O|3tWJw6MD1 z{(Dmkwl%QwfO!zod(uc5nSS~qfjqN~M@K-+!C>DjRK?dC`bT@E`hsr#_o*6WG!q?W z`m`=o+J26`+#V6+md!`1A_1>}DW6m4e6jtdb3Kiq?f^aLI@7Z({lJn&xE8G-*Pr+Z zg6EHV^J<`d4~+32!HE*rF6Ynq<@#rj z*;P!DSq2y6r5+k;oZz*^2NvMfsbEz?r&r8shpfNCDC{AsUCX-}I9zskaUj6uIOJ90 z{1P(@X}V*f?TfIPwC7g$=i+ZT$|M*lTDx`cWSQ*P&G1(!RO)$$?WMtGhr3IgwbFiR z{tDJ56nWXb)|xanP^32Y`Sxtewyod{)Lr1n?I*@w*13MJe+Y{`k~zVjI%Ss?!m#B3 zJ*hcd3`4SP_M|RXZvZ7BjA0Wu;>5i_22e~KHkrVuKM1tCibn1{Pg+l#+pmhnZW9cK z$qDGhE3PC2++>sJ&oMW@ZCjh$i79#Q3LF)x79yN8LN+s|)I@}jw`3CKXP9!|detu$ zB!c!4-ITkI##tEIleo-ar|S=(!1&g^#_i0+Jw{~^+y1m|09cS{IUK+M4W_jVr?ClB?b=KK zstt zknF5c?&;r7IJk;e5Jx`#z5shu!_)XY*=)tUwElD2+sG!^ZNAmM3%XubmNyqRupz}X zjFR8fkUWWlF2FA-Bjm$dGu5a!cdmKT$2W}T@1yxchO+dTZ6)h+9P_wpkaAxej%~T} zFb9_ro=c#!yh8d?q`G*cc)qHk5W!rQ8aF)AU|9LqycSk&_s}$_h+K>FWPr3J;4zr- z2ZAWII`z|!#Y2Spo?)@g&4%*(B9ns{mQOU~wZ2LuYt;JkUTnBZE4OKhsCmcneh9}F zaxv)~QPlx>AKt=^`mG_P)WwvDoaF7$&0|=MjqM_n+MJ?WQ6}SosIUv8ZF8gqBWSj4 zPVodHXqQ!Rex+w1YY`E6N-Hc;YZ=@&<{vQ8*RauhS`Eb_B}iEaudt)(bbMS+RfiqP z-zpT@DP~T&Ei{gUVwbm?^5K4D5P>q(USXd5Oebnm3((S{N-4 z{!&e#AmqRm%qR872zJU-wv^={kz%BF>yw&0Nr(FB?W6KRy?eTuDU9JqVEqXfvkysm z9o!JdJ2>xdqB}iQ2OO)rYP(-aU#wi?cQy}H_^+Ck!FW+V?KNpdSzZy{@SPJs#-uh1 zi5taVQS1hk>}kOmC}w@V&MoQw!fA;_xSQRLvG((&WrR->BLG$5P5noc4c(u&LpO0G znEoBs6a;7;#iXmFWDH$?52XQ@N`$FF8s6{U|ClkoRv{+hp%53>xMR6pteLB(GisvD zISL!GZ=C=5}&kMl@$fhD%xI7;y))^8jA1CdYrE=1bs)EynvU zdDXtiD&~etbsydx9bPIC=lx+yc2-Efb!VpI5XXD_CO9 z_bsObN0rP$P8%*4)TUf{+e1RL#!PW;Yz}>w2e#j9dtGo@@~iqio6eC~>?dDB+KT%8 zMe?th!MxfRwt;+n-_uP}P;656Hf!H{HR(q7EatX09eIyNO@( zT^-Yo`Z0Gd7@{Kcxqm^k7W*{TyMlOlt!u@G2^8|f5k1fNVqggPvGWQZ=~ReMO%LxB zNCO;f?1_G>w~;g1d6w}M8HA{iNKNqn0n)E%VvK}#WKRCuc=r;Zr`cS}9CEY3)DmI=giAVikURn+Q;16|SPfi|uR%w!QOB5ixvR@g#g|WJ%-B+#Y zP(U4X;o$%J-rirdZmb4kLjWvwtJenqVsh+@5`R%F*-K2wp=XHQhh|&xL+~_(;UE4V zXZvjyt5s?8Z1y&Ak-c%(>JVx4jhKBtcDxrk2;T5>``8&V9(`Ul^}m`>p!}mD=I+i-0i-0ucyf& zNF`JY0kg^kgvx!W!1qQ-(xFa$UL(+)<%DF@fn(S*(jEJK-0Me4EB(35i;*sSxj zj45}zNy~gX2KeeWAPa_8%(HVvcE~va8^&|4Un&Y>&F4!wDwhgcN;=Boq&}jn4IB;% z5=Rlr=KX00w$<{Gge#7F)&Ho9?VVmMDTUQaka-M*@ihNzDltDSQ`#}mhDON3UT|Zq zEP8k&#ovxw4c5{7)rC??b*4qX+1R}{M+mCC*!(<*!kA(21*x64;4$Im%LXQM1|^nO zY!-i%LU#}%!!jxrJ@X!N2DCE+lqEQAOg8Nr#>^V^@~s%c3#LV3FmQ4E>u!y&qmk;( zT+4~NJah6HY_7|Y@6>9QSzThWT+8JUK^`c*3Qz9_adeY-o29s?{NJ4YU0dzp5WtC}T`eifG7c4U z3tz~Xzok%wbAc`N$dZ<7+?vA%Kc|Y3f)R5SMjQ|K8XbP3{4j9M%@;s_M8S1c`M!KR zLC1ux(-e#&p?h>j%+sY*6D3?I!#fFrCtTd)k8i+e4mv3RMo*HfoV^!JrQ^$cCIC3n zf&3hGWBoS;BU+o8eTR}Koz^i-4nh7%R{TJ#CCrP>OlM-f#gGcu5=}$Y}_%LF}>uB=*qjA z35b+0%Zn$-!P8;k=O9ElS5@(%1ySwmJKd;pN`|MM4~Lw}jxZ1Q3IAj-=;jj-2}bJs z7%hu23EOg~73++3Fdl_;wJ8>2RY09shY+!|*r_;9f76-FN4p}+_Q;7Xq1olr1T7L* zH0x8S2@d6BG_N69HveHU)XUh24+H>6a*Yn*5i8ExurzXY_6gm)mNohX1^g0r8Ejer z+PWP9x=g2QEF9>RBAOCCzll=bMywj}`>`Y{##rnT9pb_$A|@cjcQZFPo4BbVTB!6! zFXAh>?4codz-CY>Lqc|n8@=l|L5VvE4qwlLjtmW!B*19s`w&!xlgC}rf-q#+(w2SP=D_8>@h(QUg|=+)m0TmLoqrOU|ehdF*|S~MVa z4|}T*;-X#cLM{o^CA(JGZzAnJti*414?th9mL`sXpfaNR=!qb}O9R!Rh8;+%$H1iP z_bt#OPAFGqz;38$`4!+%6Q0lFD#^G;Y(KR2L%-ixjrluiB@hheNwgxm-@M_>6r+dvP)c+ai?7^tiD7IH& zUddi?2C4`@*K;6lRpxHb3#!gu* zb`jvrd6klXoZ@z3s9-9WlL#V5Ti9#Zaj)JSzJ8rGmjV+959t$OGNfoXep36PK@5>O zd$NRr?X_I6+w9hXPU51!nFV)pCSrh{^?>o~3HLvFE|wh8ePFLCpL$^l#KQUv@4(L( zVXxztj2VeON*i)20vakZA-Fg_`(3A{sdNYP-tN*Ur2lmK$gL8snroHb zR-yH<(x^r$Xd8Y0sl~7gW`F8qaIrb=AL{OI;8Feyk0uFMTwk1j>2uQz<3NiW&$Y+t z=MRrv4-i%f>No`y_Cx8b4OfLU-7G^-JS##iJhakAR~F>Y9tMkM%fz4_649E#AVVAYaqXAQh*~R8DAY+IDl_%~@=b zpdspt9Co;%>Qpc6e=DQq)?-kt554zk zS6juFA|%w~hdnLVTd~Fxf#*_~(qZhmCTKP4?t?C>eOqg_+0xQuu*VYqy>=Y#zYFua z!?ZzRczUk~Zy!_62`Gf=b$Soek;e1VAG@m1^CG>Pph;Qcqk<_&WRk^SR`i<_1mA zB<|hzgTo9u(P;Ade!@-nWpL!O6KbYe7u9EGyjhdtE6T8rUnTMckJ_tIOA}QgtcY8V zOYW+~HRm_WAlpHQ_bUVZxOPz>U2O3;#(upcCn{!l@(y*VTfL61Yb>f0R97un$&c~F zIugzS+6oXsgB1}T<+&D<+fy}TP%_Beui-eMfggaiRDdG?55z3%b?+$;)=)7;Ss3O< zS||^GduCV8;UMm%kh!f8Z9n_11!!zu(3_Dw{FK{yWq)9=2CtL-}!BB3O89@-4;Og%3h zP}K4)0*ncMet1MAYGj@hpI?SNnqy^}y*ezKFc-e7vv|bj`9iuF^-ZI*CPae`f?O7E z&jiymP-sFO`j8{-4$bmb1S0p6z~^c z4!64;0k&op%oyPeuepV{i)48NU8l_Cp&`44q!+y{asjv5DPsyckVFf9k6;X_Am58iujLiI$#lHJjspW2PNoqPGxZS1JkcdH z)Aqlh`xCS)oH&9l_d#+_jTn4VEs2OBv`mok6_rcd4C;qyfW#4kRH2ba(Kwl7#2ovg zAfzgkj)=3fBjDYBTOyi=Q_if!VFGvx%)p^sE$&w<+7#Hm-_IvQpRI)dmX3<( zvhOZAQeEgr<)Uow(sS$Z#qaR5k$0`Onpxd9xFBam^af4g+Y`aOZ(OU+98^S@#TB-h zHs)D7YPNu`8MU(^U=ELk69>j1N_Q{;E(b*9NjQ(Uf+VO#DRLBsjP((rwZK%*2)im3 znE>)dad zpv31Vx=lhQ`*)wf4m_cbDGRaspB^C-HkeD*9GA!Xfj?T@4DS8r!MjpAv86#M{utl` z>Uv1I8i!OVE@~l*W}bB56mjX8Jh>(AAbwzToPT2#JiDe+mN6GMEry9Kj}Os=X{G6A z%a;U5oi)=5o369ty-;w+T7`7WHi9X{B_REBvrl`M;91uNpx8Wh+s9zua6s+fgGC5I?Xu15XeT-J^CqN)t;mhq4m+@#S)z6(pqugw9-a4G|K zxb9wsyf9>96X>m3+~fbp{FXC90h-DL$cvd?>ls1}?v86Z0V?ng_^0T3rE7IZt=E0a zI7}H+o5QfMDgM$72RVyn$gll~9{(ucTBvF5NidCW+ISzF!CIWRH9vkpnq5{QKlLoe!@D{{!#R3dr3+PR@INy1S2q0ALh$ z%~0b1gA_DA^Ap-`{8iJh5Vl7LjHYh5j_Qka->XznIWZ3L^sMR{)=8M}a!v24{2&M5 zNIN>6iQ-aGaIQlkTC@8;K>-x#kU`8Kka0z9sk(CRRj zAZj|VNOS|YVJ{}gqph77Cr`bR7eNu1swqvG6i)RcWV`PGmYkTx_=!_5^djH=J943D z4MAXFc9SJmrzSg1#noSSMwj-2aYWa0U*1D2JrP#k^5KJbufa&`uJWn2uo^g?qIO>9 z?btxG3EO2O>z3c{6y4xjR`PdI!%6D(Y>G?sJiH=<2WQENg4;z65Uqs#C7@NeMq zZOzA2_!=4^UvR}o@-GfNCd;VYw6(#~ioBnWhh4XFS~kQd#SAl)J8vP^?tV8CXkH{7 zStkQAUTmN}Io#ylAt2BxB<*GKN7nK&v|+D8BUKO@#FkjH=0($roEQ;JDN@i{gcu9a z(ms7hL*zWsL?7byqqI>e3(8$o6Cl{$hG7GTbqv>zQ9)cA)@e^5J;&^9<1He3l`zi` z1QLZ$OTlVPqP`4-zeXfdFA;8wA<$GJYXjuo`*0B}=ky^jxZpMBIP6E%4Hd`IfIR7 zW0YZLvzA<42R(RjX@}0C*VDfp1coTjv=+#G%@}B=MeGDl%8xnt1N8Rcr5aVNWnHgE zpRsGOavyp~C6{^Rh?VlD&mM$O3SaLxqs7VxeSkv2S28G1sK92mLv2*Pc>PdkVkmB3 zTA;T(;vA9Oi5j~Pb>>uF=wRA6N2&6C{{leoHbBXsO>I+K4a|zr)byVM#W?$k4X7ng zGf?!#aE^r?@~_=zoFMr|1~e@nHgxZuQg-Z`LtKnEdFTjV{YY51Evd7d2xChu@uH*88 ztn_hST_=G3JKe=yiH1b9K|o-b4*DE~)ij4dxTJ_$z8}Y)ft7x17H(V3!#@B0IVmLN ze#=@ViYX9YQ~J{eP1b187R^V<(1xVRx+X{|%eXcdr&mp}+k&_xF5Z0rlyLwf7L*JA zZ_GKNlD?7n)Q_)#v_q2GL?&tIQcQbeurc-#P%88%Ig6f@%PxYw37ht^$rK;I^>t@T zI~>y0KKY6??eu>P!)MCHU4a;;*MLXpP=5H_!oXQJNp{j&)&{w2#;-^g?%$T6Q zQ{O*#`z%*&HcrjC>IpY3XG^QUV*>ox(7lVZdUo-5a#^rH8N6$c+LoF--N$U}2~*rR z*hSHB%$U%mjKc{@w26|enWmq+Kw|;m+A7PQ&ml7)OW_6i{%EN;c3w!FIpoH$(y+k{ znwpZYR?4S>v$f6c0bDJA*hp+ZS|+D`-(&IZd}Ya&aZ6Z&JT7t`4Y_k@zMK1C1*{4n zteaFMF(*~Tw?l*d2}`>p-NseAUQ-fJf5v8j92hG3ol1&5>gCTFm}$6p z(JoD?;hl0hhS#Vd&_DaQiXc4L7T%|_9RDEmH=nmLriFsfFmkA~hL~3=CLx|30fR>$ zxE-b00o1Bf)&GM9P6b{JEo0J(-dpGKopKdmhO! zXA1`47BFWG#Et7iQHvxT{~!W1{OawWPi6@3yS~pd_=d>XSXY#ka;h^FtOGc@ee&WE zWD5LF!d|%ENcBbx^ecwGW~YeK35`%0jkdCS4_X@9vC_FD1o(l!(wl9Ph_Uj*s$WQb z_C(Bb8PV}NGyX~b)L@|kvL*jf!^+hssJlV>b<-7)cUr<0Xh~KNjY+l)q@r;USPcgh z_V=(7Z!R9DhSzC3vAwZ};Y;Uq;Nn4f=NZ;#O}n)cx|ie#TqAQ2kmP@(2F6x`dpk*` z^mixeD(8hjan84A40LK1VWC{x|9Ylygi7f^SBR#r1dSMvAb_63t6AadwwbNdD@KUI zS%PZ}U23rdC&Fz?Dc5(TD_Pf3;VcRp)0$`>s}dzJG^s6=7$k!2wK!Ks3Br3}z|lBx z{?V-JTXeI*M#jRr;eTaV4_WvzR^OtHK>XI(vmu3B+F*_&y*5$^)Qzlw@QdPj?r8JQ zbC*Wrd@a8ZlG&2=ONRO=e^ddFB+Gm`>H6dbW@h?LZf7uiBdaLoR5vFnA1z0)F30&5 zas2p~ibg|}>VkH?bZT22O7$R>>)0YGp-!zRxtn${vVnXkpr4~twUSU~>lg(N@%4QlTop8PT1yA&i4}sO%q~akxscSV znDnFnt6dinA)?a%iLbwYldOfh(6N+{ssWI9qI|`dqOFQFRa^e)1ODiW_ z+?=j^P<#p};~c4=+*+RdaWL+3>oXHk4%;2{jgCY8_)u37n!=Se`m<&Jwi%gc93B1q z>iLCk+D5l2=PT!VikpXFi9C7nx*{Y%8v}8vcJQxZCGRRZtb)``?yCel^D|_vW8hAX zy25PZ{*B&`Y%roWE5hbKIjRk@cf)l?{}29xh>{L9qqt3jWbzy-Zt~L6;E)9?kDe*C z{N{@`o!U`auXp&vjHpJAVP|pX$q7AvGsXIjPAp)MiE@%|a#Xl+;6MCE3U*V7RQ=M;Z16>S(d!alup5 ze{3!Ts>uU8t~4$8VMK??FlrHJl>n7DG9x2yfSSIS%@1NgNJ=oacC|RG^mqi*-7vJNZl|yIc#<|tq-Qt6rW!ksSj}SrK~>uWOO7&hNfcsF zs!Q^<{G&vTHVYWlxcP+QI`E~7loNAbq;Yg(49=pl{`K0bwucaM){(23cX;ehR)`QC z!%3u|mFYfyU336Ls6MDd#^F=T3{E$6B^{x=yx5VnE)r&qAW*sed;k|SS&fVoUZ{jq z>>)$army>cVSJL70=jl^sAlNC(_W%1HszkRy7JH|K6<2wBn^ubx;Rv(3O07mUU|w( z4fm)E!ny(awq8)4>7gO5q?;~21$g2g&}y#d#)kqWMK@EVZCsCww*bH28K02fTQh}O zdH{9;!ixx|1qrCAxqws|5erdkj1%P^S@kiGKoQLa%$EM4sRx9se%c7+S;fZPM_aFy zUUb)XT(W?W6>s23TB9C^iSwXa8z@{&X$-?T%GI*yJOrI}qs1d)G8Q>Jtv*xBB9;i$ z#ZhSu4U*oIT~<6cOa)87e>L#7C*8&VDyc}WbBpq!Ne{EkFor-EGxZN+YD{z{SV$7y zpe@8l-jd_vDbK{>v%DHh*AdUPLd}o_MG(yU%iT{BL%WxcIeK=**hvTGt93b^WkMoa zTG+(<>r+8U>6$yJBEI8*kkEEzeJe|GAZqU>u6^bt0Bq)tOeJayV-eK;i5roPz+m#X zR;_h1_>aJWL8znK5p_I`m~k>Sol$gO&)eiP?Qo$H@Z?xrt^x@!nZcBj`>$m--VEm( za_>^lo{>HfZfV{~3&J0n0~ubc(`F~oM<~>vnnf|VdIMk*Ptj0JVB3#yM0^#tannNk zMN#B_56(fJB7}iv4L#pMUZ^un!t!kLGb8UX5S7O1p84K4Lllc`VBqImWM%1%3}nSyZ+F+E^}+~xy^@zHxQ2NCHA5)+-k6Jj;REN+7k)3ZvdDG zv*<^tbI6f)=ST;NiwiPK*imr_fNj#tX9R26l(t}OmUR1f=@Bbr=g-K(u5oTFihZLc zdB96rpuq7V1yC6`h2->s5b0$vU;0`{PSC4G2$heO^7T@K+!9zNRBpH9kqW`4TVTx4gOVSdb>}L_kQ2=gGDE z8##WI7TW4nFAC#G1g@KUUq#j)0)0QVfFAaTY?39$rnyJ4J0C{aScoL5!Hd-x| zJLE0=>GWoTKrPs`xuuznYx7wdA6wjz&@89~XMt*06pIxXdK0c;PCNDw#k;Fl78Ydz z<_mw&b4n|Q_g*b;?YZ>YtIiOGm+v>6lvwgr(bX(tPwS-iQlj75t_ww6G>V0ety)hr zj=_?uW>O--OuNEWcrrociBClfPzB!QLJ{eF)jL9al>ztt@H)T|%N4I!hjz@ej05s`?}OHsAG5cUSy0iqBY8nH1J^eHgwPbmh(7%C{t%il*od#jk~FwywFU7E z>5Z8%ih>7mIxT`DP3A6d?Ld9HA|m!og`t6HEjxz}7x^ZPgWI*_dcf+DAG2^EOSVn+ zv!Ex6Kl_z(Onmdd{r6SYWt66C2Y*o*h*;8~?{xXaqFo$)#3XSnv zO#_IzX#{JeLz!T*<+S1E zB7YFttmqK*_+jG0O4H|8gV!fI(8V9wQwUHC*hj;4=d)V}VV8`vQ~id;?uXUAUHbIl z(KJ2R3%pCYIkY0^&S{huw9qfvnB)bihs2H%z0D>QR@zvuGM75A6+1Vw7s88O(fCPv z=n9EGdIUxo1;f`n>M}k+GxFUg8mIgEf7E7c#k*M)z~AVSUMbb`=E0N;G8Z^}YdUFn zR*wb(>#$N=yCi%aETlpd{iDzSYeki!YRK&k5<$wi5yA=rEa}(0KY|(R!<=81v;dCd z4z|wDWVz&$14qDlGc6|>Yva#ARNTn|nI=@8?wrSwXf zPJe$-9Ws56#N>MF!e=gsr{C_b8fVM;tXOr+0yV5I;NPBSk_*EYOn+HD-4+mYj~ddt z<K+S(<^?At4+Rh^$$+OtH7i$! z2u87Rr#(mN=PLLd$+^QKF<5{Gj5o>r)Sw3_mk!H%WUN#hSgyCb<19G{8(~BMtGr1W$F}cg2k2j5 z_N_y7x^y4Lepc0#xZI4S+g6+mep)*uz_+mXevG<*y{0fdYji!CWDjymjaF&(;$pXD z;xc0hYnK>EQ-eQWytbcYH}v_D`_bnjf1H^NrEw?d7?UF_Vc_Aom-Q=>inOO@laGK? zW~6|K*-|=@0N;tRkrc%N6emI+J3kg8hV&-%BxB#&2KM<397Z#~%qenuO9w4bb6kc~ z<@6oiB!%Yl;2Ylu9^`&)__$BFXZaKFO;enzYKbjFE)Ss%0frnl|Zh&+Tu?avGUU% zXC~N!2nMb6X1+}ZzWs2lf#b^##?F*FAa9nl)nm7x(FUU*e(m(|DJ;vYqB!jR;Ft|d zp0x_Z?7f-FWT@OXSIB^O zsK{KEiLj#Efjp(2+PT!ADUbf3nb4W9Zw}#b;pSkoBmu5B$Q0axR^u}pl{JpgTK zTP#9G5bptn^xokYXvNU26Rc-;SyRtOI7+1H41qapn`cuJSyk!Z8DK)&9iAcdA?c?W zULaWS{6VL=R=LTCVS&XHyU0+hBEULA%4ByxpEpr2bg|NWC1^}L3m6$7G1G_juq?eg zgrc>HlkpP> zP5?T|Zxe#j`x(5Z@6*djt~F$&5LeHz9!P(usA%d$`VnY^c9x&&ANgT2S$6=71=C80 zJrZx&UtCEPo^3};e*2~l9AL^C4F#ut^bs|}Ljok6#8_FCr*htojCEZQ@Xmm6VptXWNU%A76KqvAL6?+A)Np zWkpse>sWq&4tsz$-Pnti`DuT#x7RvOcsI-K=6zm$xANm{x}y(nv}8n3uzD?nB&G-#ucI9I>StJ&w<^ zRmy1k%=wLh_-=npflc>W$^HS@-86cb_>+B6=J1;cq~rY^1X8JlL~CI_KIwEceIr@F zvWPm~e;lUFhu$8Jb1@2`XN?yQ>9aOj!WK zS-R z(&tB(y@FUEH@!%Ide$&KES|Kpm^XTc$<)lBx?*CG*OWC3oQ@5`fyjaIYOzA`-)?Ea zzsft9cz&3Uqi2qwoBsYsTwEen3jyObVjm4Dl=!$JvozQHzVA{6e6pA)un<{&B;R>p;sqLUc$!-r7E|I2c4 z`md%8=2~|c$1<<=wp$=tpgM0f;t3J$Sh`i`h>;3mT<>1n({>ortrIo$;zwA?bGW=S z^o>InceB$+tf9-NYIWs)Uv1A_ zQkh+8uG|9ay%;>&d$MRk8>^1o{HmfO$6U6yC2z)=?xY9B^YinImR-66>(=6@)*k;pAAOW39#U*yPT&mKB%XzBvK-(f7g7qH0MuQq%rwq9tgF> z%DF1C!8rO4C(xu+eoh2$)yDtQd+gnjXr>94ZBBA?MQ6yCRHbveYXpMYpy}1usuTUu zGBw72G`0E!fH|pK+=$u$De377t#vire(rz_ijKx30qYd~7CL9F{#vNH2_;UB>@*s`yB?SABcEkrb zf_H(KzAQyOZI*YNu+^!u{Veom>9_x zZeh|BsFl5P?)Y2uW%}8C>|MpL$=<>B;s1pYn^3TK-_YF|ol(Hc=S~y!Edu1%najwC zOwO(dJ|*YFa*ES;r%$hlK?0FBKPgQ#EjV>n2VvkpVo^8$OIq>&4sDd*B%2D zBnet$|5&aHb2lBbwMrMP3slYfRfnkThxyn(OFkm$XZK^Z-$sk-=07olA^jtj8Oyc% z{KXV4C3{LndC3$?PPhbc%X+Ccg1_MRtcdYaJRxc|DY8naER zP%!-%f$_27YGHan2Smpk{=kNiJycFuJY4#x_y-N8A5;aSKLBdOT(ZJ?uxnuOvmlfH zRY>tCU>IIGKfRFF90P~Ll5jRC$I@jd?R6JpxiVtKpcMUDII#gl%C&u&uE1Na^ww*@ zmv&wJ`ZX?WTblL55ai6DiDrQ4&nq{Kx^=B3`yGVBlSw%`_Br6H=lvz{qLXXqsbzX~ z1)FXc=(-8zUePLcK@u&@d_rL<}B0th~D*g++)h*EesoOwQ8#U_^}56zflrRC~gGB zahNmhee5`9&K7cjQgc>-e{f~$8p;bI^uIBrw-IZf#I&dL%!0q0G_|*b48+#zLPVJ> zDbSixEOL|AqnXG0EUWepx>gRUoc!DQB^CX6aMk$UsbwEfbgtcy@!q@k%0nbLc~CPU zD^5(J(gS006E(>7j8;RcocR;ZOh>EQ^spz^fsaYuBGvk^yXKp*ViwsCL6q#xe;HS# ziLmi+x@Kqs0AglVNvll#l;Mwm&4mlnMCA@B0$NGWN%_>3Nbb_ATS_rVrmLJJPI#N zWo~D5XfYr$GcYy^FHB`_XLM*XATcvGG72wDWo~D5Xfq%%3NK7$ZfA68ATlyAGaw)! zARr(LFGgu>bY*fNFGg%(bY(r!GnldC4GLp=sOb9WCCa?+?TofC}v&%ylMNvQm3xcR9%3@a{ zqGCf33(dv~Zvt5Gy>s3>IcM_!_kQ2~Z~3mpOirMks{o6H*%*r3(a2OM;2-GbPon}V zl|iOb=_Csa9)gSEFLsi}GFYxaFx2Tggu5JuaKg42!u3E51wKkKpfLc=!HG8CiAn`@ zD%J4|5R*HB#ZU|)0RCj)gQ2j3WZ{lU)p8_Cggb$+M_|Rb1~f;<`L_LZ;3|RT2p>X$ zKZJ{53GpHy5`#dDkHENkNC+#I2*;&P6pBivB0~}d8Iwo3Sla>>f{OqbR>1NYSOD}T zg8)bZ50*wISpc30Q4BZ&F(Iykq>144UHtI)G&Wj^&Fe`UiMw zDctX*>CHq9T8$Kw0wGZbtVM({@ki1qpcok7awV+QeD4@~BhhF;fbek;2}dC)>05Nd z3=0SD#Qfz*EC{6%@}mK&{`d71PAE)(pjngT-T6g9zOF1OaAI~bHieR#*WDZ zcJ_2arwn@ra3l`ep>a46GMLAATyIo}f%yaJ5`FrTa?GFxtOggv8VrpUfDy`tfz>ym zhf$eSKJlUb-*WG#{5O^Bqx?^y|CkY5DHiwptp))9;fExMSUrI7P>NyskAUUyz{2{& z0z@hK(d&&v#16QkQR1)dLlkTz78Y<2oG%)XNxvN?)IXH#f$}kdeog312LQ?CkeWm# zv_K~YrlAr0?7anaTidp#8M9-KnHgip3^FrQOffSvGc&}@?3iO_W{#PenVGqD&ONW* z?Y^n;?itn9T@7PcmbA3Dq%G}l()u+QsgM4%AvC1{YR^ZQ=+Cxako4?gy+{!veYUdM zVuAv8sdw%4IG-p)TOW7VI>P92VXwuTIWgDST^;4F4iQB)M-`pMj}GocX7NllgNC!* zq#{hUQg_Cl76u=b-P+7|%k~cm2&7%a$B(Hav~&d`v>t6v%N;0CR;#8-0%K$?=C#r) zFQz2}vtr*?8ght9{d{3^q}JkX>57`sS8FpZsLj2dI9vMSSQ<(XcGf zk1g)mQmwA()c(_s{x+f!(n^i1>?(OGPB&r3k%N8mv%DNmbWuYu+E4Jnc&$24OoYB=i)bW)idTC%J%$i zp6X8{tsAPfnb(w|3GT>Ku>pkU@~><3I9J~^nFy7H-4Br?i`Wz$6A4T>T;-hG$){T< zVl~~#*svPjP9I|d74C@ZU!v$ctBMi`y07lStONCxV)#SfA~m0JkCyBVl1yybZ#~6N zs8&i}`e}dKY==q~bewanjh(R1$Q!8$M8c*_W;~U;pJufVpM9K-XtEwnk(F3wHaVt9 zKxr}#yq2-DSzV~%{y48$7_9R=JTR`ihHLao&k)V?VksH#+WjO_Aq|W1Q+<>w?a&dt zHO-QGj#RZ#ZN$C-4#1nE>(Mpw{!L+DSyVAftYr;TH}7;~sgjN6Rx2*pIaSy|bImhg zohvdn^}Y4H2b^#>lcijuX{0&k>S!b`cR2jan(@a@fve2o%%~Z()$5>G^k!XK&Bo=m zDaE)Ah$litr`>Sx^JYQh( zx>k)_M|qlg!iJh}q7v=qh3YoHII4xVeS5NSK=qj03!UJiaXA|E5P8TkuZyjgwv^5I zj+)xIFMdot^yFaej1%WxKepuQtBfN(xwo$JQe+w{d%o^`2B9>aL~vt(YKY6rJjr%o zsLEFR7W6$<&GGY1h?_j~`RAfe3;wa@Myxc0(Du2xlQ&)oR81Carg!+8&MF5ty79B?yQ`?IA z^d+tB?f^T8s%c~t8|F-Ol=BY0Ml9);2W{C!B!l?Gl@{E2#saICqm+c2`!>DDzC@6( zG=s8%LL)d?#%J^yMeWKKk3#7e*wb=5_}Y9k>Ei?4Re1Y;a`%lnZ;IL-&(ahg>&4&% z{<@{|EBf5AxNYIcToJ2PNdEgXDgy5 zNbE)r&!~r0v9htQt-BN6NYwndco#caw^l1uN6Yi!m#1<E2cn@E9NW2DLMw8xy>W==#VOr0SUhyyWa_U6+mB2wM!t{IylcO32@L zj~)BYCxO4HinZtv)v{LNSsP&Ofz9=MotM@7I3DKd9(&Za!tjQr?|?UYOrz8~yE(nd zL|~|db=e|m88v5^PA6sm#P)Met>||l5zD`XM1*Y24F3pq{*742Qc2o+iypOgU)cpQ zVF|3yNemQCp!%A~cEM7qcvg#uVO1>p2U6Z?hjx^!_kIU-^*M9tw)rO=|A*ItX zscz9Rt${VgQ0Ue?V{Po^X`VPV zNfu9L6yu)vo6J1m)BEL&nyalf&+4X5X7=nE2#}1Jh~KEcrL>)zyuL5s!>4&2UJkiF z1F*~q8@F;|Qp5f$D&w`4Qf z-vk{7U`r0_>xes2ZV>6hbCINjiPnNaMSg7I=^3;7;vhk!Da-h2;6W$-=@N8%BuXLt z+WUc|u(8aSA|2|}77aMpLY$NjnQgW#08wMyCj%;lBl4P%$pp;-9wwLuxtMx`z047J zKf*Xt&^fy2yUm&q_`9D6TBK~I{J>x+iSJ zXEbTIKj&{-c1yZ>-VI64FJeq`H=JenpeVELJrQRup?qtKavvIaM+ndSzCp(~Qmpbj zYFPi38di4Jf1et~VXG}Zq}FF@BgB|kFkSqGAZVh|1(mHzanzhytuFlP#L9?p_MLO* zVkjo$A&?nbWoKn)e%=?ibs2}*(^DaC`#at7)qJ{B(&si#VMIh%Wi9Mwgo?Ru_1No~ z+6(UM3F@4c0~T@COKvW9JJRa-bP0H>ym8^L#mjvB1a}Xn?=nLLMfZNPcS%oxnbQFO zS5-vOfv1#C+@Z6tBBtIJ-sLTKfuY7@O<;4D36AjO?4r{p*sOTBct`a|6|oEWpd^1l z({mfzKHOghKDc~-t&=A0evq>pE6uwyd_CbEH&kfyG^bo&u}pX8QaIYStR+*J8;H=K zI~%5Sw;|Dt*LsFMT)-RUO5*92MH6>J(^_3!a`IS9ZP{VoVNv@JXu>SiSKH!N3QrJ8 z*c>%@29l;Ohd>}NQ$LX45`LBL#TWZPDExpL2d4620^9)igIUN57!2R09u-$K2OqH} zEi?<(hdfEa53^rQN;)?*K$3~L3urwcbr}{Intze{MuLX}+k4S7Hy;RN%IMRQblZr4 zBnNN8O+|d{VRuCM)nfz6!iXOiKKX8Kby;3t*7;9TsyEoA$Zk=TK9|A39U;hJ)oeH2E>JU#`_Y>c9J|7-aK zTY;8@-d*Db!$kl<;Dc;Y($|S-dAgz>Uq-p1S|o}P#(l>toIvMSj=M7JpuJ47UZy@j zg$=|{h{Ys<<-Y@g?O#Cvu(1F8ApGe*klOykeaQS$4vbC@ z5W{s+PcW-f7L&sGh>Xg&^!9w4mQvfIgy_3}nRc0mG0^^aSDU#|zuqkFP%&G>c|V_R z8+Y}*hq|(&=xNX0e0J5-G*`EM^2pB(?^)x7;iz6-tklr4;(AqVjxawLQCcA$=)USq zfLfYsrM*%SYEB-!eHS(Ph6o0KH~eOIQCjhFNOZRFXHIRF72y2&j6cP_tPYW;rT!%9 zU&@Kl#xwpznASsN5Dt7u_?cEh8L8*t$?w5f&4>H_Q4 z$6C{s%0n$p`gynPJ;yd^jd`zBrcZ?y&LqxjS3$kGV6WQjVLD>kgfk3$l2cQoh zByXeHbT5Dwrl)}#<5GByHL+Cg0@CLoDkRfmw;^o$v~mkV0T{MD*f8Bxw#;pWcPxGs zLC2yKI*nej8(obrnA2(HJpGEskBHtoGw%bPu$!6ySW~)61Td={tlTU`Cory|U*_9o zw+r+^^Qylvm#RY;j*OZ5C3UWrqA^svF>h<)8EhHP2(TG1{pQH%M&Q~FAoNYfB>FEX zOE48YkJft^P*g`TIqc^M^jFhw2R8Ogv+ll*J?S3|s(K4FmGbhRvmeWbjz6Jn)r_I0 z+RvqI;q8|)%#s|P*2lOqbxi7uO7qFLDF9IGxnF1aK8kbmY z7+Xx2ea2hH9*jtdHPCISTfA;mz1}B10xyNBFJ&slgX`k_z;?zqpS7ri5YYyCP z;N70yX>x{;)&+KU1{ANdVPH*K+utM-y1qKKH*Oa%-*$PF7mRydd0?%H?P}6{;;ONo zv?x11i>GC&Ys(h;`y=H0_X*Tl?Mu2LdRauuNOxFWXWU#l*?rEd(lwM`Z@xC2Zq%J_ zYt|{h6&O8s`_x$e%4|$5vv?+QHf8U4>Mek``(4K>+Cl&(8QB-pHvlXKF#r-1q>TH{ zcVxS9k8cm`BPNy7+(%4yOxH*s@h-|Li9QxCggrD&v_eH4nxc==zGRpzVRPy7xh4_u z%{w3Dk6=K`HyM@)o=1%J*GJDs3DI3`DNEru^&rFh9aYFXKW)O0A7I(iBu+96P;UfO zLCCPPL5u-pUefK&b%t#LWw2UF0I~ttjj%QmX%t)9tsoZE0EI0RH~tfIa9Ozds2^Q6 zucTf6wvP-2Y?_|oMPpY4Z@-G(qkJ*CTWrv$v~=(wSA>{Ec#BRTUE*Wr`_+%Dbm1y$ zq8VxpKk3hm8-*uz?$i;{RR;jUrPALrSm9+h*n%jN)Jn$k7=-DcNKXjgl>5d|@HEpA zT18Qv%HXh9%G=+}ejEi{|D5+U8wl-0T+Lg6ovOFrn*CfBa@H+HuW<$yGZ520&M{SE zp{751H9hFj%lhJ01*EZd%N~w~2SwDQU+tSme|s&Ez`8A_7vQyRuknr5Utj|9Z38mf zu6haVVmm%z4IHNFH7->T8Q%cH`4IR3#~t_})dYM#x?bb0@8iAW4+exZ#-IM+^hQ49 zeIYpTL6+rFy-G2zBhK$AWaOLI2*-f|_-tErij;GA-%?eT56I?Ad&9rQ@1?OXInXQLuulbHB z$L{i*Yt3(QevUJ@-=DedM=iWDBkrtq;IAZ|*W9NXu7{vz=YQnnXqakr)gF1(b&03c=c3!w?AyT28eHoO^ z9NXqJZT+~1sDIpF?yWGdo&Q+0KYZEf9G&e@sa7<1>LI-T{(84D^RkT33!pm}Efm1U znVW?xJM{xI~^6vL5x0S{`fg`iSWe0_%u0XnZCc5r)A5#2Q`T zZG@41j2wF|5b5;womBe;omc~*S@u@&7VHRxj$Oydg*z@F8ddW@O^L^ZS1N=2Y$R}{ z2|_$0HbOU_J~3mKXOX(l)AA17DrJ40_;)bkz zM6p~gTkG&~`Y$+Ou>{r88AM(!x?JF;v~aG7B{6fo5*f*5`d{hZBz~BKWXLch$cyG~F$aU1Cs&l3XQ1;>DoYLfj87zio>h~Yr$?6h z4|qk+zj*|CQS%5EqGe6Cpk-~Qp=A}dKft{LhsOjw`kR4-dJ*yXYlnNkP092(#!_ zk>7Y#Hu~R@`ZrQRUl7yegQUJie_RHf00 z=4spIM%(3%CY_>Ni7&@)XpQZ!)?-5HSIudS=Hs(<-hmuL%3V{0ivg(gWS|V7AEAf| zpyBC3)!a|OQU+{CeMg{>>E(+{QRwaHuW&)){N(f!!pvQ8X{hNx%M^90D4?VQ8?*x< zWzQFx#HMl_1j(C01J&r!ujRc>p(|)l<6V}cx!U@wg45jxiWzZSDdj`bjV3h#yQM;t zs~@S{`B5J>-OdUo-8TJ!eSfYqaoa`>qm|ZHA`+t$%{?NOd`qJj-kcJ;uy^vtcVQkn z6SJIv-+%K-+z)E|%3eyl`3YAc1b7N*?&nIB`+s3+@gULP9rR^=b;`KXF z8GZ*U;D3%wT`h)+{WDb!uB!0j;u*K#Y;9I`D4C7pgiornt-0F2!_?4J60~P;eDgTh zI9@Wnb11|g$aZxycGR>gmR`@W=5n28dzL&VpS#cNZgYI4fH?knjB)S2m%8{y?O9za8J62qdg5G3O)Pb35>2p0U!od#e2ryHXeR9%Vk0cgv`!iE@ z8dVj!S;lQ812#A3%~;i)J`d7qQoN?%b>)aU&$}W|<_#~yX?c^@^q$ty=#P{kAJa z39fUs$Y~~}nN6P#;K;5yjBZXfhu(4**;qfH{b%%hVuLH1`&Xjsa#4EzM^PyWSX9#f zv#9hJs(%%g{zCQ7qLSNRiE8d&MWvS)FZ@;8X<$+5AE@F7JIlMGj1UCP!9Sn<2daM+ zmHrde-2Xr|EB#MYryI^ZmRPM8oc;+_^Z%<+{g0y3|BvebN=2nVQT^wl(sSc~D=Mu5 zi%LMR+SR)r|2tIwRZ>OB%tZfhdsWBSQL7Sqq`-Txu%_=~hkkm{4bs#B@zI2alLRbs zvtOzxsN^eSK#E+S-7LiNe`Gi`@eExd^8Jew{U}>WL~%0?^%|w^5m{vMy09eIwI&7WD2vHCS9Mw&Y8tVsJC5Clq}P{M>V+MVw?*yG?3_(gYXUzJwarTG+#T38v$y8wqfHZVw3Xb=_rpjV z_pMcPG-$>d)k~Me-6qPH_cDrq%PhMJ<}?KC*x=UsP7h&ljVx)i6N9fIeflg)X`sM^ zVvhtZhl|`#h)y^sB!moZ_cQq;7_qGiH94rLNx(5gl`c5wCI_jX9cDbsS20t*5u`H( zG7Yf&S#0`<&Zqp&UB&VIpj9+(HGcCU0l#{Olu!k zbn|38{s5|RKZZVeP6>svfCuntrMx5}U3@#Z(k9x=)Z_jGIr~*iSLS)?kp1QuNCeC{ zNcK8{sc_Q+FYva_B9jL;BK`1HAugS6hUyyA$;{ItLc^zY+Z_N=N1J{ZUYJLnI&Ts^S&u4`uii-P+hehv=@_@rWmQ$81Hlc2(@D!TefJ1M>|ZEe-- zaN$f-+}fs!425|UkM0_C+knDdGZg{e)nY!sY^RQAZry}BZSA~xdzzrp?lL{j+F&V{ z(fiPq*0QXbx@WN|C}v%U+{E1+Zv{zUYT9?>n&v$Dwxp7MSh}G4ZNJ`pYNX!Ov3gB@ zC_#mMaHvM<&88487&p?N)kf_xC|LEGS~Q>&hge`2I?!UsSPub|C`_;$vlp~CHAKvh z4*`uymyZ;|yiaHYI>M%nkTsl-P$O9AD}sMDBn%aN2v)aeF|jfIp5SS2SmyYqK7w#3 zOz`B}L}3cNr!lXoJ5CHUY;qj*v|0{ZQHnPhYq;PB);2u-xE{O$Z3z9YDt&>9nW*Vd z3R0F%bJz_d>4;#YFUvfc7Vs41HY>*=iP~%ac&WiQ)d`|MMOb(zn92aZ8uf#X3l09A z+#tLvEd)vmV8jEmMlb6&luPw4byfOxzEDjh*O^xr&PxV6zY|JP63YepXCmtbdT(i0 zDd-T%_QqZR0sLmgccqpJVJB1d28P8B!lMbm0OwQd%OvWSsHLTFJ|f2lfW3RY8pVDqu7}##-F4k_ zHL|es;yx!gU2J^}ZA#kea~{qA@?!An?j31mRb9}S?y<`2ru}5+vCJK^Ur*)Kj?r*z zSzlK2MjO)LXxkd^rTD5{dv&h#O2&_G`+W2+6IpmxZ19khe)lTsoFVp0yW|bifR|F} zN1wCWL^p5yX>oxU(vg+~R*1A_c^V4qwD5FSbTZJfDrwgHQ(rl|1x=K13HF0z*&*1K_u!0*gfdi29M5Sx*JE`?D6PK6+YB{Ib|XWu*=Z zxe?)4L{6wpRzh4Z_5lb1^xz}})4CZ(Br2(b+T|HT#oMaZZ&ACnJa?xPSY>FfV!~s} z^q*Na!|CU>K1GI(f3%04TWGFrMfW0gSNIu2iorPSBL|;hF(H^zqiKfV9+EAmf%6?; zCbCfKyI;x=QoB>2Q=-Nof?I~AUuJ~J;Aq#+P4b$~aH!CqH}3MK^$>oWot{m$D=oWq zN~cUWOE(;}pf$DGu<@H6gw%9=smV9we$-$yL%2iNDC9M`bGg{eLdw21c*N>nU3xC> zFqga14c@&}OcUfe^KgZ3C~QRwXiH z>EHBOVoo>x!W!NLpXl9&?9O;}Z9_|{hi43A<7 z!(r0FhNK|3Lr^K{R7c#SQzx&MvLuIK?LakMK)lk$`P9vOX)c&tji&eJi^s{+odhjG z&Y`6GQ++XS-z%c^=JJ^x4*|CaG}f1-oSR@I#+eOQ?<+kw_q6q#`{PAe>GQA*H0i^Z z>-LM&_P3i!fuVerokzwi$H(oo#MYf{dNX5w>iNy=TP{@**nF^WK_Mv+pZvu_DjA?* zyM4u`afM*?`Xrg5dlj(g#Q81wc7y;(smO%Y-|WHskX6WWp~FP`6(Ixp;d&V4A3~Hm zKiWxR+0=^t5L3zi)RTks?GhQYW+6R)MWj185F(A90nFEJfgQR5WOHMXoffGY;Nc)v zlAq`+rpuVWT0omy9JO2LM&o>0XB|9dQnn%d*{vU+9zT>fs| z1<7KXKm=)1v?^q{*@1yS{!sdxZMOubNFXCeN_70*_kgrb+ObssRahY zHL>&s07Ytjv2pkK3nQ1P*c&cksyi!}Xfg!<%}0ga8e@oehWTE;xJoVkbYvbeY& zH{%P7CT#?sGJQDf#A^;<(=*&&;5zIGdAqGxoA8GyQ}$(gCw=ht5Qvdx3E1gUY%?&| z+l}9_@ij)S!&U9LlpSDJ!;kr%(8e*{c78?GdAW-;_%~t^JAl@i)q|>%y%f^C(23hoX{o z*$9GM9z?shveN>j6OuFSm#oFJ%2g`%<; ztGQpb^W#A%TgcCyDYubE$K`uHQ16$VMs4Ou>c3)8eMzn7$uz#vJ{RDPetEG4Jgq9+ zFNu~UB)`La58l4u$0o>mjLNTO=N7k_%}SgN=wd(5m3#RnGh zksg^3RHy3`nGdQD5n*l!T<2973DnmQl_?*6x@p6lsXk?QW{31NLwvCLNd_kQ0o_L* z0aW4xnvWp_0=)caUJAsA{M7iR6_8()PoF{amKZ)nVshw}=L-pi$0y6rAoKabqyLIy z49`^x+minboK^k+Eh(NnmObOsLmw;{01-T^;7{{p9#=GoUTn$1vM3NYT0pLTq3Yh} ziF3I>LLAc^p(R74ls$Z3XgQT$O??1=ev zjPOq~iE3z|3`X-R4jg;-RS^9r&l5}ZJkGH;bKm4?XUpum@uH}B_8Dxa+QS|togixm zPLM-;#8jdJ>f_6B^8y%56;4>}0;YaR!@E}cVacCLwZ(+%x*dvnUPEw>3J?@&O)bUhqtwfhSrJ$Pw(X%{pNe)bk@TsZ3uLRs$T9H&1rX<2-mf8xD<8 zT4Q6XD&=>NjuzYPx32ZPEa{gXg8O~f3%1dYG(iW)*|Xs5G1Tj`8Qs zv8=7*z52F{`a)S>4^H^;{Yh;Vf;hAH+Ke^I1&a3e`}ft6mbCrV9y;@GUKPyB^M=Cx z&ZlIgV3QYGY{?~AUV?__LrDu>9N=oIY5hhWf)oQL*=AdP!t2{k*v1K{ynH$aaV{$e z6qy ztdnkdQEv?KaQGgpCmyX2jibs_uftsfp`~9E;GZW65YxKG(+&qksrBbBi_hm{vihCJ zE&Ahw2VfhZH59gFkE$JOC8m7ha6wRPpf_~Yn7#(_EkUI-AlcLoS#}1-@pVoXclPA) z6Y0z43Pcox_MrcqV9-vmI(~toU8u<-Ro}<7)>@a$&5RNcko46Aj zB>02b4LsP-$p+|3lTP}BtpJuiz0r$zqi-0}YWyAi!Qy&$8v(Ypy7F7tF5)X7#>}XU z(lRuTpY;X^eEA}17fK+d$Uw@>RcV|2a=QymZ6YGT z4okGXfW1M#!j8B{chf8OTPt;4s zH|CF+9_pUpxa*#1nvR&=M1(>-RpBktx!%jdd&45~r-gQ*?s z7%;Wdvy7BI%9GLO%ys!pA3%YQ)Ja2Vh07#BjhDkvMGh zDD&uuNR-X{wN<sSeYv7P8R?*)iv47=pGvJb-o+K|zBa92^|m+aE5;%zzH^W2 z`3}d4KY>`gH+RPCo0S`f(6M9d3mSgr*ma4?3-TB5kzC*_+Iu|9sku0Kv*T-X8T67b zS!_-628Lz7pD(vXC1u?`^t-xnKLKLgB{h3=Zwj}aF0P&2?k1O#rCZOQrFRN1ql{Vg zhqk8^)WOTCY{fO}L+8Y?ZDL7uKoLN=q|70Qki?#Sm>@wT=bGRfZIGb>Y8Wmm%#;Q+ zW{N-MlynwAS+K<{7{xyVz<=cKR)dHhqo36OVMUq+B?jUVsPN>U;z9CmsKHf8Hjwxt z+I2s$JfVZ2!(&2fW>ofTlwrLsj~(TRQy;J$(2GkUq|-;TN=aydRfKSEga-)}42I^* z2-7st(^Q4dc5|bE8n28WdS@MZ7tzpm?nd zjS0#k7JHA|*M+>W-(8HD3b=SQn}jz~k&|W&(@Zk*Kq%Nj$v2QXduok-!1>l90`v(t zOfr^zU@3fkavgnxh5ZHkb3_8Ml^}w?e9iUkC69~u5&XcRMvG%pYt?1N`4Jqqj5E(c z3K%>P<;!C6NpuMKumJ7N%mEJM*IesQI+H%S6SE)9M`lwuJ5Eb!Q)~Ua{1M`McRD|A zoAl)MvRs5Uf{mF_8l_2R_yLJRyO}DSJHJ*EIYqP%&V7h#B ztE39jn-s*PO0WLdG+}(G1K(466aDK0@L%nOnQ3uPtz}mu7;Or5dRPrI0-Ip4R_Toq z(~Y$wdbhD`gbjxsvd$gRF?r%L!c$zpa}fk>$fWa7DnfcZi-L{3o1&m>f%~USgB^nn z1H9+Jpi*P)pD8$3+1VyoP#uIF!Wn)XhKXdJtP|R5dJn4Py+5@JdVhUU?p>hHf3W!S zLSwI2q>2PvxPzNrwqs7pow!w_nSCG2Q?X&ch@>BBC`ofWYdq3+5}I9;^9FbPZ6?cC zTdw0IF6@cv0B9qUG>*n6-GyHRFN%0EcpT9uG~|qZoeOft6zz(b5;Tr#4@%^-Cci6W zP39_OT~6Bl87q|T&KRG76as~hg+SrspCN_Fxr}F!S0PaN2;8(q0VsS_$jWbjf_nuH zkEwa~Hv{7auQH(U5g1Y^l#Iug0ELfgp6KmQG#>3yZ91U>UKPdB3v{zfo?m z{jK@H|GgU05}95J_2V_sZl&d{Vo6)UhC>W;h!5`5`Mh-ybBURIryKs-2p6yS%8Sd| z*i^H-ea(ed7zUfee!E@G$^IGe_dq+Ejy%my9uDPBx4Y*TgqhFfOU-gCMrj_?tlJqB z)=Nh=<*R|W-m#A7((d^`q8dFNJB+t~7zOKB^LtzN`xkolAo0+&2SJd%*}YCU zoce#gmzN-*T;e#6R2e}Uq2Sl^ifOc7YR*y&Wix?`M|V_+m=5nhI?&|5y3@x53OCML zbn1|=G2ZPlpAn~AF-EeUmTsk=fwk+fYu<+a>B9{|pf=6U@q`Wk+Ug=}rm9&ZjM~-{i~tifQJ3&5!Bc^~PM5WoM9P=r8@_@u8?|E*5BB zf8>i0&f*tRF>u}acI%e_zCQRs^aN>u!F|VOE${)Vn2EmH5RzzOK*+hsDZVb=#yL@u z)FvLO3FL&%sDewiIb?#+sf2)N*>OQdVitBX?DuyDZTIdw7LwTvU4k#zX>S?eHgvKt z=6qE!?WgCs;5kkc*!e^Tnk%ru!RszM)~GdY+%D5)b(x5Yi4I!P3f)|9V+uoJX# z_?gW#!j5s<-h35{ zNuz1RU?wXi>U$&0Qzjsk?pwT}a@dL+)^SkoiHUG~s&+~N{x3B|{&RJ`<`$x5mW}t>Y z*ZAzJS?^8}(i?ZDS~H42%>FRlYFLGSkyCvQWX$@jE)0 zSldxj(J7ibSQrAY6&&@<4D}s=w*-MLWgLKyN$CQ2A#^r%Fd;N?wK24_Gc=;2lQeX7 zwze~{rvyI7pBtK5TL}SM(hyP#aWK*|GSRcsGchvIGq5mI(=$@g(^CMqleRYauQpM% z)3vcNGywLu(Y3HQgrWPYD6B#&;%H%^r)y>P#{npsnA#Hp|Ne0xG=$#_?d*XsO~^#c zz|6qH&dAEb^o5a@^*@dfxC8LLOPM+u68`b7s0ht%bRA6W4Rr~f{@Rh1mWdYlK>&N` z7+M)v>zTnYF#ofzKoNMvgmkicW(p3LFmy76%)ovdNCN^bK~YU4Tc#1o8M@R5Wb_MRpO4o`2E`3H#{yRm zI|1y~7QFkDx%Lc5dnAaT$fM6;9~8wpp*Qj6W4kB<84wY0;Wm#yCFi3C)4|9IbRpyC zc=2t(3M)S334#l@Q;-qI6K18-;XeBHCxLkbQk4Nqxm72oHM%LCON4dO*a>vno7D10f4^zze!DBJ;LB zd@7gA?9U;U4>%lzp&}qefp9jk$@~Bg@nI9zm(KU#<;fgsy={azV7$5(5OFohm5H|4 zs(&(U17kckeh#HlhRj3P+ z8M*C|7(jRIOYBA)Sz2Sv;6obj7j4ba@mi0OHQy6=-=j5H=;(xz!UMs`uiOhZ>C-u< z5$4lm0JF>Kfb2Pyf1O;3maRpd&Al=p=n-mt@de>hP%*$I-uE2PJxLRVqT=xql4TSA z-o0+)7iaTCk6;^S?1DbJZ4tf?7;xqCmAoPy$+Gdyx|B>{^nx~IQ(mPze?lADaq1Tn z$BfzGC6=1{)_o@7#Y|*6EA1MRhLN@IiI31Znr4B0wxSD{(^nPwxpL7FF4ykVa2dh9 z?~bEO^J_1a$rjvuzCOctH1tO<|LCAcW%^3ob?a!D2a zM7w1GH_*Y<+WFl9cl270HR=NKrD2Zq4eA|bw5Ax}&;Q1T<>mFwttc2T%s3i{4}nAX z2%MkNUZ}#a_`u&XN4rR%zNEW6o?NwM{UR<-@TIl6x%FW|{R!*jBJRuUOen;Q;``G0 z4{mg`2=C704-e3>l6;0KpG0)$nz0ufy+e*)>HMwE<7#%lbfV8^UsQ3f%`f6gq|G;< z^|H{aWxRvqIBIZMYSQu@!UrS5>{2*4G_;KyQg(Qz@EC}<*CawuR8f&R9Fpg6vjp7D z3)|1N?|`9H=O|r#1_ObU+N^iME`-fkEnwIvL$>`|$=mYC#nn=ZTt!`Ey8%*}Uug?9ncIj<;LPU74+ba)QE`j{Ee#E3^lPE;^3m{+J(O*O zP|N0xDK{edz7k4z)rViB6qaVCCiD6){VHhgQkp%Y<@z)-^80GX zgzmCIU;K)N6OGgdNW6?9mnO6pD4Q?J!-?u1KX8+6eXkCET!PC$YC%T~WVj_HO`#3I z@`T(ReoHEu9&TvUyLkjM-48$*|Spvl}1L~sf}$I0a$XE( zhK%QcuAk9aM!H=G&faB$=4=W(laQ(g3z;a#>lo%mCRBYZedAG?-Bf94Rr%SqtEE4g z0EV&E4mQ6GWn}|cx$~-wOvK}(T?8w;(W;74Q*Y!@lM^jbGnBlKDI=|jwlv4m5~Mc4 zD;uXaDe`AsfcnCXdy0RadHg(D(Q}78GZaII20t;{`T4bn|Jn-BWfj0P|Pmur)l#ocL@^Q zJC!QL3{ET3j!kfie4AEa*mO>N%7}TJ@-71<%g@hpMH2W@^UIc^2g`fJJk1B=1Jc(U7Fkv#%Aeoeyy6oV`&X)I;pBaK1x6|7W&RF5BpB+9DG zn07L3Z{X>Vm34BqXiMN&5o+K}NNg>@SYUX3u!!M^-;GOD!2Wt=lBezcGX9|D){GDK zj#h!#J>oM5<(up?3(2CYWhsTye@v-~bT;Euo7&1jPZ&MR{Q=q46Ca7{=&P@kTozQo z=L)pXoW)X_{B<}&@9Tk!N@Si-9keHfG`zE{c-GAf0iTmaXM$d)V6%~i$?;k==a*Nf z3f|LtlAlhHKvCYJMU}r3hbpu+R2a#*>EQ=|+q`{OE!MmLUA+6xTHt>KZ7TAHMlf{Z zRtAPHgzEH!4D8IBFmwv0Ziau{grQR=RA(S$B4hv#%3E7I00;l5FUkM0hmke#-X8(< zANPa^)wzV&goFiISp`{G1eh6wM1<+-1^L++0E}z^5n&cq76Bf@|9TVP>-?j&BOwa| zq$zSJluafVOg2R384CHd|T88XdU`CF+{fA|UL(aIRr&n>MI+3V;P zJ_#e`5R%dp-2r$qRLWmXUClXF_D^UPot(r=s_|15mx3)b@CH>VuS{(k)|X?_;$uvqm1e1OtL#P#!>7}RQ+ zkOM%7DvD58o-+UtG;KheH%Zfd@ecxPCZJyF;R=-`roWkefZW7|v$e`}$TXmH0io?C z>vs`xPEizX233CC*<6JV?`G+*&O5dNv+DkW>A!3Pbyh6pQ3aczh|gQ+AiJxyIMrx& zEhu@fl*LU1F-f|Pq3-Io!lUH4mrGX2@w^(VtejdE|I__-O+MktU_`#|soYJH>J|(o z(QGD4N>!rd{f2k8`29s>W}H@N51m9?TWD{4^KCinSNHX=sI_0?>oLNKFB#!9j-rX$ zB10)+hM(6KJHF@bk)(tr6sCc~V=Vc+!Pd%F-er%1enI149yHvp@2`FV6WTBq`qvZP d-a*&S!3CIt!7#D{*w|TN$jF3cMgF&O0RWiZr!W8j literal 0 HcmV?d00001 diff --git a/doc/Projects/2023/Project1/pdf/Project1.tex b/doc/Projects/2023/Project1/pdf/Project1.tex new file mode 100644 index 000000000..e602f817e --- /dev/null +++ b/doc/Projects/2023/Project1/pdf/Project1.tex @@ -0,0 +1,646 @@ +%% +%% Automatically generated file from DocOnce source +%% (https://github.com/doconce/doconce/) +%% doconce format latex Project1.do.txt --print_latex_style=trac --latex_admon=paragraph +%% + + +%-------------------- begin preamble ---------------------- + +\documentclass[% +oneside, % oneside: electronic viewing, twoside: printing +final, % draft: marks overfull hboxes, figures with paths +10pt]{article} + +\listfiles % print all files needed to compile this document + +\usepackage{relsize,makeidx,color,setspace,amsmath,amsfonts,amssymb} +\usepackage[table]{xcolor} +\usepackage{bm,ltablex,microtype} + +\usepackage[pdftex]{graphicx} + +\usepackage{fancyvrb} % packages needed for verbatim environments + +\usepackage[T1]{fontenc} +%\usepackage[latin1]{inputenc} +\usepackage{ucs} +\usepackage[utf8x]{inputenc} + +\usepackage{lmodern} % Latin Modern fonts derived from Computer Modern + +% Hyperlinks in PDF: +\definecolor{linkcolor}{rgb}{0,0,0.4} +\usepackage{hyperref} +\hypersetup{ + breaklinks=true, + colorlinks=true, + linkcolor=linkcolor, + urlcolor=linkcolor, + citecolor=black, + filecolor=black, + %filecolor=blue, + pdfmenubar=true, + pdftoolbar=true, + bookmarksdepth=3 % Uncomment (and tweak) for PDF bookmarks with more levels than the TOC + } +%\hyperbaseurl{} % hyperlinks are relative to this root + +\setcounter{tocdepth}{2} % levels in table of contents + +% prevent orhpans and widows +\clubpenalty = 10000 +\widowpenalty = 10000 + +% --- end of standard preamble for documents --- + + +% insert custom LaTeX commands... + +\raggedbottom +\makeindex +\usepackage[totoc]{idxlayout} % for index in the toc +\usepackage[nottoc]{tocbibind} % for references/bibliography in the toc + +%-------------------- end preamble ---------------------- + +\begin{document} + +% matching end for #ifdef PREAMBLE + +\newcommand{\exercisesection}[1]{\subsection*{#1}} + + +% ------------------- main content ---------------------- + + + +% ----------------- title ------------------------- + +\thispagestyle{empty} + +\begin{center} +{\LARGE\bf +\begin{spacing}{1.25} +Project 1 on Machine Learning, deadline October 7, 2023 +\end{spacing} +} +\end{center} + +% ----------------- author(s) ------------------------- + +\begin{center} +{\bf \href{{http://www.uio.no/studier/emner/matnat/fys/FYS3155/index-eng.html}}{Data Analysis and Machine Learning FYS-STK3155/FYS4155}} +\end{center} + + \begin{center} +% List of all institutions: +\centerline{{\small University of Oslo, Norway}} +\end{center} + +% ----------------- end author(s) ------------------------- + +% --- begin date --- +\begin{center} +September 3 +\end{center} +% --- end date --- + +\vspace{1cm} + + +\subsection*{Regression analysis and resampling methods} + +The main aim of this project is to study in more detail various +regression methods, including the Ordinary Least Squares (OLS) method. +In addition to the scientific part, in this course we want also to +give you an experience in writing scientific reports. The format for +the delivery of your answers is namely that of a scientific report. At +for example +\href{{https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/EvaluationGrading/EvaluationForm.md}}{\nolinkurl{https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/EvaluationGrading/EvaluationForm.md}} +we detail how to write a report. Furthermore, at +\href{{https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/ReportExample/}}{\nolinkurl{https://github.com/CompPhysics/MachineLearning/blob/master/doc/Projects/ReportExample/}} +you can find examples of previous reports. How to write reports will +also be discussed during lectures and at the various lab sessions. + +\textbf{A small recommendation when developing the codes here}. Instead of +jumping on to the two-dimensional function described below, we +recommend to do the code development and testing with a simpler +one-dimensional function, similar to those discussed in the exercises +of week 35. A simple test, as discussed during the lectures the first +two weeks is to set the design matrix equal to the identity +matrix. Then your model should give a mean square error which is exactly equal to zero. +When you are sure that your codes function well, you can then replace +the one-dimensional test function with the two-dimensional \textbf{Franke} function +discussed here. + +The Franke function serves as a stepping stone towards the analysis of +real topographic data. The latter is the last part of this project. + +\paragraph{Description of two-dimensional function.} +We will first study how to fit polynomials to a specific +two-dimensional function called \href{{http://www.dtic.mil/dtic/tr/fulltext/u2/a081688.pdf}}{Franke's +function}. This +is a function which has been widely used when testing various +interpolation and fitting algorithms. Furthermore, after having +established the model and the method, we will employ resamling +techniques such as cross-validation and/or bootstrap in order to perform a +proper assessment of our models. We will also study in detail the +so-called Bias-Variance trade off. + +The Franke function, which is a weighted sum of four exponentials reads as follows +\begin{align*} +f(x,y) &= \frac{3}{4}\exp{\left(-\frac{(9x-2)^2}{4} - \frac{(9y-2)^2}{4}\right)}+\frac{3}{4}\exp{\left(-\frac{(9x+1)^2}{49}- \frac{(9y+1)}{10}\right)} \\ +&+\frac{1}{2}\exp{\left(-\frac{(9x-7)^2}{4} - \frac{(9y-3)^2}{4}\right)} -\frac{1}{5}\exp{\left(-(9x-4)^2 - (9y-7)^2\right) }. +\end{align*} + +The function will be defined for $x,y\in [0,1]$. Our first step will +be to perform an OLS regression analysis of this function, trying out +a polynomial fit with an $x$ and $y$ dependence of the form $[x, y, +x^2, y^2, xy, \dots]$. We will also include bootstrap first as a +resampling technique. After that we will include the cross-validation +technique. As discussed in the lectures for weeks 35 and 36,, we can +use a uniform distribution to set up the arrays of values for $x$ and +$y$, or as in the example below just a set of fixed values for $x$ and +$y$ with a given step size. We will fit a function (for example a +polynomial) of $x$ and $y$. Thereafter we will repeat much of the +same procedure using the Ridge and Lasso regression methods, +introducing thus a dependence on the bias (penalty) $\lambda$. + +Finally we are going to use (real) digital terrain data and try to +reproduce these data using the same methods. We will also try to go +beyond the second-order polynomials metioned above and explore +which polynomial fits the data best. + +The Python code for the Franke function is included here (it performs also a three-dimensional plot of it) + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +\begin{verbatim} +from mpl_toolkits.mplot3d import Axes3D +import matplotlib.pyplot as plt +from matplotlib import cm +from matplotlib.ticker import LinearLocator, FormatStrFormatter +import numpy as np +from random import random, seed + +fig = plt.figure() +ax = fig.gca(projection='3d') + +# Make data. +x = np.arange(0, 1, 0.05) +y = np.arange(0, 1, 0.05) +x, y = np.meshgrid(x,y) + + +def FrankeFunction(x,y): + term1 = 0.75*np.exp(-(0.25*(9*x-2)**2) - 0.25*((9*y-2)**2)) + term2 = 0.75*np.exp(-((9*x+1)**2)/49.0 - 0.1*(9*y+1)) + term3 = 0.5*np.exp(-(9*x-7)**2/4.0 - 0.25*((9*y-3)**2)) + term4 = -0.2*np.exp(-(9*x-4)**2 - (9*y-7)**2) + return term1 + term2 + term3 + term4 + + +z = FrankeFunction(x, y) + +# Plot the surface. +surf = ax.plot_surface(x, y, z, cmap=cm.coolwarm, + linewidth=0, antialiased=False) + +# Customize the z axis. +ax.set_zlim(-0.10, 1.40) +ax.zaxis.set_major_locator(LinearLocator(10)) +ax.zaxis.set_major_formatter(FormatStrFormatter('%.02f')) + +# Add a color bar which maps values to colors. +fig.colorbar(surf, shrink=0.5, aspect=5) + +plt.show() + + +\end{verbatim} + + +\paragraph{Part a) : Ordinary Least Square (OLS) on the Franke function.} +We will generate our own dataset for a function +$\mathrm{FrankeFunction}(x,y)$ with $x,y \in [0,1]$. The function +$f(x,y)$ is the Franke function. You should explore also the addition +of an added stochastic noise to this function using the normal +distribution $N(0,1)$. + +\emph{Write your own code} (using either a matrix inversion or a singular +value decomposition from e.g., \textbf{numpy} ) and perform a standard \textbf{ordinary least square regression} +analysis using polynomials in $x$ and $y$ up to fifth order. + +Evaluate the mean Squared error (MSE) + +\[ MSE(\bm{y},\tilde{\bm{y}}) = \frac{1}{n} +\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +\] + +and the $R^2$ score function. If $\tilde{\bm{y}}_i$ is the predicted +value of the $i-th$ sample and $y_i$ is the corresponding true value, +then the score $R^2$ is defined as + +\[ +R^2(\bm{y}, \tilde{\bm{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +\] + +where we have defined the mean value of $\bm{y}$ as + +\[ +\bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +\] + +Plot the resulting scores (MSE and R$^2$) as functions of the polynomial degree (here up to polymial degree five). +Plot also the parameters $\beta$ as you increase the order of the polynomial. Comment your results. + +Your code has to include a scaling/centering of the data (for example by +subtracting the mean value), and +a split of the data in training and test data. For this exercise you can +either write your own code or use for example the function for +splitting training data provided by the library \textbf{Scikit-Learn} (make +sure you have installed it). This function is called +$train\_test\_split$. \textbf{You should present a critical discussion of why and how you have scaled or not scaled the data}. + +It is normal in essentially all Machine Learning studies to split the +data in a training set and a test set (eventually also an additional +validation set). There +is no explicit recipe for how much data should be included as training +data and say test data. An accepted rule of thumb is to use +approximately $2/3$ to $4/5$ of the data as training data. + +You can easily reuse the solutions to your exercises from week 35 and week 36. +See also the lecture slides from week 35 and week 36. + +\paragraph{Part b): Adding Ridge and Lasso Regression on the Franke function.} +Write your own code for the Ridge method, either using matrix +inversion or the singular value decomposition as done in the previous +exercise. + +Perform the same analysis as you did in the previous exercise but now for different values of $\lambda$. Compare and +analyze your results with those obtained in parts b-d). Study the +dependence on $\lambda$. + +This exercise is essentially a repeat of the previous two ones, but now +with Lasso regression. Write either your own code (difficult and optional) or, in this case, +you can also use the functionalities of \textbf{Scikit-Learn} (recommended). +Give a +critical discussion of the three methods and a judgement of which +model fits the data best. Perform here as well an analysis of the bias-variance trade-off using the \textbf{bootstrap} resampling technique and an analysis of the mean squared error using cross-validation. + +\paragraph{Part a): Paper and pencil part.} +This exercise deals with various mean values and variances in linear regression method (here it may be useful to look up chapter 3, equation (3.8) of \href{{https://www.springer.com/gp/book/9780387848570}}{Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer}). + +The assumption we have made is +that there exists a continuous function $f(\bm{x})$ and a normal distributed error $\bm{\varepsilon}\sim N(0, \sigma^2)$ +which describes our data +\[ +\bm{y} = f(\bm{x})+\bm{\varepsilon} +\] + +We then approximate this function $f(\bm{x})$ with our model $\bm{\tilde{y}}$ from the solution of the linear regression equations (ordinary least squares OLS), that is our +function $f$ is approximated by $\bm{\tilde{y}}$ where we minimized $(\bm{y}-\bm{\tilde{y}})^2$, with +\[ +\bm{\tilde{y}} = \bm{X}\bm{\beta}. +\] +The matrix $\bm{X}$ is the so-called design or feature matrix. + +Show that the expectation value of $\bm{y}$ for a given element $i$ +\[ +\mathbb{E}(y_i) =\sum_{j}x_{ij} \beta_j=\mathbf{X}_{i, \ast} \, \bm{\beta}, +\] +and that +its variance is +\[ +\mbox{Var}(y_i) = \sigma^2. +\] +Hence, $y_i \sim N( \mathbf{X}_{i, \ast} \, \bm{\beta}, \sigma^2)$, that is $\bm{y}$ follows a normal distribution with +mean value $\bm{X}\bm{\beta}$ and variance $\sigma^2$. + +With the OLS expressions for the optimal parameters $\bm{\hat{\beta}}$ show that +\[ +\mathbb{E}(\bm{\hat{\beta}}) = \bm{\beta}. +\] +Show finally that the variance of $\bm{\beta}$ is +\[ +\mbox{Var}(\bm{\hat{\beta}}) = \sigma^2 \, (\mathbf{X}^{T} \mathbf{X})^{-1}. +\] + +We can use the last expression when we define a so-called confidence interval for the parameters $\beta$. . +A given parameter $\beta_j$ is given by the diagonal matrix element of the above matrix. + +\paragraph{Part c): Bias-variance trade-off and resampling techniques.} +Our aim here is to study the bias-variance trade-off by implementing the \textbf{bootstrap} resampling technique. + +With a code which does OLS and includes resampling techniques, +we will now discuss the bias-variance trade-off in the context of +continuous predictions such as regression. However, many of the +intuitions and ideas discussed here also carry over to classification +tasks and basically all Machine Learning algorithms. + +Before you perform an analysis of the bias-variance trade-off on your test data, make +first a figure similar to Fig.~2.11 of Hastie, Tibshirani, and +Friedman. Figure 2.11 of this reference displays only the test and training MSEs. The test MSE can be used to +indicate possible regions of low/high bias and variance. You will most likely not get an +equally smooth curve! + +With this result we move on to the bias-variance trade-off analysis. + +Consider a +dataset $\mathcal{L}$ consisting of the data +$\mathbf{X}_\mathcal{L}=\{(y_j, \boldsymbol{x}_j), j=0\ldots n-1\}$. + +As in part a), we assume that the true data is generated from a noisy model + +\[ +\bm{y}=f(\boldsymbol{x}) + \bm{\epsilon}. +\] + +Here $\epsilon$ is normally distributed with mean zero and standard +deviation $\sigma^2$. + +In our derivation of the ordinary least squares method we defined then +an approximation to the function $f$ in terms of the parameters +$\bm{\beta}$ and the design matrix $\bm{X}$ which embody our model, +that is $\bm{\tilde{y}}=\bm{X}\bm{\beta}$. + +The parameters $\bm{\beta}$ are in turn found by optimizing the mean +squared error via the so-called cost function + +\[ +C(\bm{X},\bm{\beta}) =\frac{1}{n}\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2=\mathbb{E}\left[(\bm{y}-\bm{\tilde{y}})^2\right]. +\] +Here the expected value $\mathbb{E}$ is the sample value. + +Show that you can rewrite this in terms of a term which contains the variance of the model itself (the so-called variance term), a +term which measures the deviation from the true data and the mean value of the model (the bias term) and finally the variance of the noise. +That is, show that +\[ +\mathbb{E}\left[(\bm{y}-\bm{\tilde{y}})^2\right]=(\mathrm{Bias}[\tilde{y}])^2+\mathrm{var}[\tilde{f}]+\sigma^2, +\] +with +\[ +(\mathrm{Bias}[\tilde{y}])^2=\left(\bm{y}-\mathbb{E}\left[\bm{\tilde{y}}\right]\right)^2, +\] +and +\[ +\mathrm{var}[\tilde{f}]=\frac{1}{n}\sum_i(\tilde{y}_i-\mathbb{E}\left[\bm{\tilde{y}}\right])^2. +\] +The answer to this exercise should be included in the theory part of the report. +Explain what the terms mean and discuss their interpretations. + +Perform then a bias-variance analysis of the Franke function by +studying the MSE value as function of the complexity of your model. + +Discuss the bias and variance trade-off as function +of your model complexity (the degree of the polynomial) and the number +of data points, and possibly also your training and test data using the \textbf{bootstrap} resampling method. +You can follow the code example in the jupyter-book at \href{{https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter3.html#the-bias-variance-tradeoff}}{\nolinkurl{https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter3.html\#the-bias-variance-tradeoff}}. + +\paragraph{Part d): Cross-validation as resampling techniques, adding more complexity.} +The aim here is to write your own code for another widely popular +resampling technique, the so-called cross-validation method. Again, +before you start with cross-validation approach, you should scale your +data if you think this is needed. + +Implement the $k$-fold cross-validation algorithm (write your own +code) and evaluate again the MSE function resulting +from the test folds. You can compare your own code with that from +\textbf{Scikit-Learn} if needed. + +Compare the MSE you get from your cross-validation code with the one +you got from your \textbf{bootstrap} code. Comment your results. Try $5-10$ +folds. You can also compare your own cross-validation code with the +one provided by \textbf{Scikit-Learn}. + +\paragraph{Part g): Analysis of real data.} +With our codes functioning and having been tested properly on a +simpler function we are now ready to look at real data. We will +essentially repeat in this exercise what was done in exercises 1-5. However, we +need first to download the data and prepare properly the inputs to our +codes. We are going to download digital terrain data from the website +\href{{https://earthexplorer.usgs.gov/}}{\nolinkurl{https://earthexplorer.usgs.gov/}}, + +Or, if you prefer, we have placed selected datafiles at \href{{https://github.com/CompPhysics/MachineLearning/tree/master/doc/Projects/2022/Project1/DataFiles}}{\nolinkurl{https://github.com/CompPhysics/MachineLearning/tree/master/doc/Projects/2022/Project1/DataFiles}} + +In order to obtain data for a specific region, you need to register as +a user (free) at this website and then decide upon which area you want +to fetch the digital terrain data from. In order to be able to read +the data properly, you need to specify that the format should be \textbf{SRTM +Arc-Second Global} and download the data as a \textbf{GeoTIF} file. The +files are then stored in \emph{tif} format which can be imported into a +Python program using + + + +\begin{verbatim} +scipy.misc.imread + +\end{verbatim} + + +Here is a simple part of a Python code which reads and plots the data +from such files + + + + + + + + + + + + + + + + + +\begin{verbatim} +import numpy as np +from imageio import imread +import matplotlib.pyplot as plt +from mpl_toolkits.mplot3d import Axes3D +from matplotlib import cm + +# Load the terrain +terrain1 = imread('SRTM_data_Norway_1.tif') +# Show the terrain +plt.figure() +plt.title('Terrain over Norway 1') +plt.imshow(terrain1, cmap='gray') +plt.xlabel('X') +plt.ylabel('Y') +plt.show() + +\end{verbatim} + + +If you should have problems in downloading the digital terrain data, +we provide two examples under the data folder of project 1. One is +from a region close to Stavanger in Norway and the other Møsvatn +Austfjell, again in Norway. +Feel free to produce your own terrain data. + +Alternatively, if you would like to use another data set, feel free to do so. This could be data close to your reseach area or simply a data set you found interesting. See for example \href{{https://www.kaggle.com/datasets}}{kaggle.com} for examples. + +Our final part deals with the parameterization of your digital terrain +data (or your own data). We will apply all three methods for linear regression, the same type (or higher order) of polynomial +approximation and cross-validation as resampling technique to evaluate which +model fits the data best. + +At the end, you should present a critical evaluation of your results +and discuss the applicability of these regression methods to the type +of data presented here (either the terrain data we propose or other data sets). + +\subsection*{Background literature} + +\begin{enumerate} +\item For a discussion and derivation of the variances and mean squared errors using linear regression, see the \href{{https://arxiv.org/abs/1509.09169}}{Lecture notes on ridge regression by Wessel N. van Wieringen} + +\item The textbook of \href{{https://www.springer.com/gp/book/9780387848570}}{Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer}, chapters 3 and 7 are the most relevant ones for the analysis here. +\end{enumerate} + +\noindent +\subsection*{Introduction to numerical projects} + +Here follows a brief recipe and recommendation on how to answer the various questions when preparing your answers. + +\begin{itemize} + \item Give a short description of the nature of the problem and the eventual numerical methods you have used. + + \item Describe the algorithm you have used and/or developed. Here you may find it convenient to use pseudocoding. In many cases you can describe the algorithm in the program itself. + + \item Include the source code of your program. Comment your program properly. You should have the code at your GitHub/GitLab link. You can also place the code in an appendix of your report. + + \item If possible, try to find analytic solutions, or known limits in order to test your program when developing the code. + + \item Include your results either in figure form or in a table. Remember to label your results. All tables and figures should have relevant captions and labels on the axes. + + \item Try to evaluate the reliabilty and numerical stability/precision of your results. If possible, include a qualitative and/or quantitative discussion of the numerical stability, eventual loss of precision etc. + + \item Try to give an interpretation of you results in your answers to the problems. + + \item Critique: if possible include your comments and reflections about the exercise, whether you felt you learnt something, ideas for improvements and other thoughts you've made when solving the exercise. We wish to keep this course at the interactive level and your comments can help us improve it. + + \item Try to establish a practice where you log your work at the computerlab. You may find such a logbook very handy at later stages in your work, especially when you don't properly remember what a previous test version of your program did. Here you could also record the time spent on solving the exercise, various algorithms you may have tested or other topics which you feel worthy of mentioning. +\end{itemize} + +\noindent +\subsection*{Format for electronic delivery of report and programs} + +The preferred format for the report is a PDF file. You can also use DOC or postscript formats or as an ipython notebook file. As programming language we prefer that you choose between C/C++, Fortran2008, Julia or Python. The following prescription should be followed when preparing the report: + +\begin{itemize} + \item Use Canvas to hand in your projects, log in at \href{{https://www.uio.no/english/services/it/education/canvas/}}{\nolinkurl{https://www.uio.no/english/services/it/education/canvas/}} with your normal UiO username and password. + + \item Upload \textbf{only} the report file or the link to your GitHub/GitLab or similar typo of repos! For the source code file(s) you have developed please provide us with your link to your GitHub/GitLab or similar domain. The report file should include all of your discussions and a list of the codes you have developed. Do not include library files which are available at the course homepage, unless you have made specific changes to them. + + \item In your GitHub/GitLab or similar repository, please include a folder which contains selected results. These can be in the form of output from your code for a selected set of runs and input parameters. +\end{itemize} + +\noindent +Finally, +we encourage you to collaborate. Optimal working groups consist of +2-3 students. You can then hand in a common report. + +\subsection*{Software and needed installations} + +If you have Python installed (we recommend Python3) and you feel pretty familiar with installing different packages, +we recommend that you install the following Python packages via \textbf{pip} as +\begin{enumerate} +\item pip install numpy scipy matplotlib ipython scikit-learn tensorflow sympy pandas pillow +\end{enumerate} + +\noindent +For Python3, replace \textbf{pip} with \textbf{pip3}. + +See below for a discussion of \textbf{tensorflow} and \textbf{scikit-learn}. + +For OSX users we recommend also, after having installed Xcode, to install \textbf{brew}. Brew allows +for a seamless installation of additional software via for example +\begin{enumerate} +\item brew install python3 +\end{enumerate} + +\noindent +For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution +you can use \textbf{pip} as well and simply install Python as +\begin{enumerate} +\item sudo apt-get install python3 (or python for python2.7) +\end{enumerate} + +\noindent +etc etc. + +If you don't want to install various Python packages with their dependencies separately, we recommend two widely used distrubutions which set up all relevant dependencies for Python, namely +\begin{enumerate} +\item \href{{https://docs.anaconda.com/}}{Anaconda} Anaconda is an open source distribution of the Python and R programming languages for large-scale data processing, predictive analytics, and scientific computing, that aims to simplify package management and deployment. Package versions are managed by the package management system \textbf{conda} + +\item \href{{https://www.enthought.com/product/canopy/}}{Enthought canopy} is a Python distribution for scientific and analytic computing distribution and analysis environment, available for free and under a commercial license. +\end{enumerate} + +\noindent +Popular software packages written in Python for ML are + +\begin{itemize} +\item \href{{http://scikit-learn.org/stable/}}{Scikit-learn}, + +\item \href{{https://www.tensorflow.org/}}{Tensorflow}, + +\item \href{{http://pytorch.org/}}{PyTorch} and + +\item \href{{https://keras.io/}}{Keras}. +\end{itemize} + +\noindent +These are all freely available at their respective GitHub sites. They +encompass communities of developers in the thousands or more. And the number +of code developers and contributors keeps increasing. + + +% ------------------- end of main content --------------- + +\end{document} + diff --git a/doc/src/Projects/2023/Project1/Project1.do.txt b/doc/src/Projects/2023/Project1/Project1.do.txt index fcd0759da..ba711ea43 100644 --- a/doc/src/Projects/2023/Project1/Project1.do.txt +++ b/doc/src/Projects/2023/Project1/Project1.do.txt @@ -121,64 +121,7 @@ plt.show() !ec -=== Part a): Paper and pencil part (also as weekly exercise for week 36) === - - -This exercise deals with various mean values and variances in linear regression method (here it may be useful to look up chapter 3, equation (3.8) of "Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer":"https://www.springer.com/gp/book/9780387848570"). - -The assumption we have made is -that there exists a continuous function $f(\bm{x})$ and a normal distributed error $\bm{\varepsilon}\sim N(0, \sigma^2)$ -which describes our data -!bt -\[ -\bm{y} = f(\bm{x})+\bm{\varepsilon} -\] -!et - -We then approximate this function $f(\bm{x})$ with our model $\bm{\tilde{y}}$ from the solution of the linear regression equations (ordinary least squares OLS), that is our -function $f$ is approximated by $\bm{\tilde{y}}$ where we minimized $(\bm{y}-\bm{\tilde{y}})^2$, with -!bt -\[ -\bm{\tilde{y}} = \bm{X}\bm{\beta}. -\] -!et -The matrix $\bm{X}$ is the so-called design or feature matrix. - - -Show that the expectation value of $\bm{y}$ for a given element $i$ -!bt -\[ -\mathbb{E}(y_i) =\sum_{j}x_{ij} \beta_j=\mathbf{X}_{i, \ast} \, \bm{\beta}, -\] -!et -and that -its variance is -!bt -\[ -\mbox{Var}(y_i) = \sigma^2. -\] -!et -Hence, $y_i \sim N( \mathbf{X}_{i, \ast} \, \bm{\beta}, \sigma^2)$, that is $\bm{y}$ follows a normal distribution with -mean value $\bm{X}\bm{\beta}$ and variance $\sigma^2$. - -With the OLS expressions for the optimal parameters $\bm{\hat{\beta}}$ show that -!bt -\[ -\mathbb{E}(\bm{\hat{\beta}}) = \bm{\beta}. -\] -!et -Show finally that the variance of $\bm{\beta}$ is -!bt -\[ -\mbox{Var}(\bm{\hat{\beta}}) = \sigma^2 \, (\mathbf{X}^{T} \mathbf{X})^{-1}. -\] -!et - - -We can use the last expression when we define a so-called confidence interval for the parameters $\beta$. . -A given parameter $\beta_j$ is given by the diagonal matrix element of the above matrix. - -=== Part b) : Ordinary Least Square (OLS) on the Franke function === +=== Part a) : Ordinary Least Square (OLS) on the Franke function === We will generate our own dataset for a function $\mathrm{FrankeFunction}(x,y)$ with $x,y \in [0,1]$. The function @@ -241,20 +184,16 @@ You can easily reuse the solutions to your exercises from week 35 and week 36. See also the lecture slides from week 35 and week 36. -=== Part c): Ridge Regression on the Franke function === +=== Part b): Adding Ridge and Lasso Regression on the Franke function === Write your own code for the Ridge method, either using matrix inversion or the singular value decomposition as done in the previous -exercise. Perform the same bootstrap analysis as in the -part c) (for the same polynomials) and the cross-validation in part d) but now for different values of $\lambda$. Compare and +exercise. + +Perform the same analysis as you did in the previous exercise but now for different values of $\lambda$. Compare and analyze your results with those obtained in parts b-d). Study the dependence on $\lambda$. -Study also the bias-variance trade-off as function of various values of -the parameter $\lambda$. For the bias-variance trade-off, use the _bootstrap_ resampling method. Comment your results. - -=== Part f): Lasso Regression on the Franke function with resampling === - This exercise is essentially a repeat of the previous two ones, but now with Lasso regression. Write either your own code (difficult and optional) or, in this case, you can also use the functionalities of _Scikit-Learn_ (recommended). @@ -263,6 +202,65 @@ critical discussion of the three methods and a judgement of which model fits the data best. Perform here as well an analysis of the bias-variance trade-off using the _bootstrap_ resampling technique and an analysis of the mean squared error using cross-validation. +=== Part a): Paper and pencil part === + + +This exercise deals with various mean values and variances in linear regression method (here it may be useful to look up chapter 3, equation (3.8) of "Trevor Hastie, Robert Tibshirani, Jerome H. Friedman, The Elements of Statistical Learning, Springer":"https://www.springer.com/gp/book/9780387848570"). + +The assumption we have made is +that there exists a continuous function $f(\bm{x})$ and a normal distributed error $\bm{\varepsilon}\sim N(0, \sigma^2)$ +which describes our data +!bt +\[ +\bm{y} = f(\bm{x})+\bm{\varepsilon} +\] +!et + +We then approximate this function $f(\bm{x})$ with our model $\bm{\tilde{y}}$ from the solution of the linear regression equations (ordinary least squares OLS), that is our +function $f$ is approximated by $\bm{\tilde{y}}$ where we minimized $(\bm{y}-\bm{\tilde{y}})^2$, with +!bt +\[ +\bm{\tilde{y}} = \bm{X}\bm{\beta}. +\] +!et +The matrix $\bm{X}$ is the so-called design or feature matrix. + + +Show that the expectation value of $\bm{y}$ for a given element $i$ +!bt +\[ +\mathbb{E}(y_i) =\sum_{j}x_{ij} \beta_j=\mathbf{X}_{i, \ast} \, \bm{\beta}, +\] +!et +and that +its variance is +!bt +\[ +\mbox{Var}(y_i) = \sigma^2. +\] +!et +Hence, $y_i \sim N( \mathbf{X}_{i, \ast} \, \bm{\beta}, \sigma^2)$, that is $\bm{y}$ follows a normal distribution with +mean value $\bm{X}\bm{\beta}$ and variance $\sigma^2$. + +With the OLS expressions for the optimal parameters $\bm{\hat{\beta}}$ show that +!bt +\[ +\mathbb{E}(\bm{\hat{\beta}}) = \bm{\beta}. +\] +!et +Show finally that the variance of $\bm{\beta}$ is +!bt +\[ +\mbox{Var}(\bm{\hat{\beta}}) = \sigma^2 \, (\mathbf{X}^{T} \mathbf{X})^{-1}. +\] +!et + + +We can use the last expression when we define a so-called confidence interval for the parameters $\beta$. . +A given parameter $\beta_j$ is given by the diagonal matrix element of the above matrix. + + + === Part c): Bias-variance trade-off and resampling techniques === diff --git a/doc/src/Projects/2023/Project1/make.sh b/doc/src/Projects/2023/Project1/make.sh index 19fc2f11f..0426adfca 100755 --- a/doc/src/Projects/2023/Project1/make.sh +++ b/doc/src/Projects/2023/Project1/make.sh @@ -51,7 +51,7 @@ mv -f $name.pdf ${name}.pdf cp $name.tex ${name}.tex # Publish -dest=../../../../Projects/2022 +dest=../../../../Projects/2023 if [ ! -d $dest/$name ]; then mkdir $dest/$name mkdir $dest/$name/pdf