From c266649550fe66757bb63b1387a65a24734f769c Mon Sep 17 00:00:00 2001 From: mhjensen Date: Tue, 15 Oct 2019 07:59:28 +0200 Subject: [PATCH] updating codes --- doc/pub/DimRed/html/._DimRed-bs000.html | 43 ++++--- doc/pub/DimRed/html/._DimRed-bs001.html | 43 ++++--- doc/pub/DimRed/html/._DimRed-bs002.html | 43 ++++--- doc/pub/DimRed/html/._DimRed-bs003.html | 43 ++++--- doc/pub/DimRed/html/._DimRed-bs004.html | 44 ++++--- doc/pub/DimRed/html/._DimRed-bs005.html | 115 +++++++++++-------- doc/pub/DimRed/html/._DimRed-bs006.html | 83 +++++++------ doc/pub/DimRed/html/._DimRed-bs007.html | 79 +++++++------ doc/pub/DimRed/html/._DimRed-bs008.html | 72 ++++++++---- doc/pub/DimRed/html/._DimRed-bs009.html | 60 +++++----- doc/pub/DimRed/html/DimRed-bs.html | 43 ++++--- doc/pub/DimRed/html/DimRed-reveal.html | 74 ++++++++++-- doc/pub/DimRed/html/DimRed-solarized.html | 95 ++++++++++++--- doc/pub/DimRed/html/DimRed.html | 95 ++++++++++++--- doc/pub/DimRed/ipynb/DimRed.ipynb | 80 +++++++++++-- doc/pub/DimRed/ipynb/ipynb-DimRed-src.tar.gz | Bin 191 -> 191 bytes doc/pub/DimRed/pdf/DimRed-minted.pdf | Bin 200716 -> 200949 bytes doc/src/DimRed/DimRed.do.txt | 55 ++++++++- doc/src/DimRed/cancer.py | 43 +++++++ doc/src/DimRed/cancerlogreg.py | 40 +++++++ 20 files changed, 813 insertions(+), 337 deletions(-) create mode 100644 doc/src/DimRed/cancer.py create mode 100644 doc/src/DimRed/cancerlogreg.py diff --git a/doc/pub/DimRed/html/._DimRed-bs000.html b/doc/pub/DimRed/html/._DimRed-bs000.html index 84af171c4..706f14ce8 100644 --- a/doc/pub/DimRed/html/._DimRed-bs000.html +++ b/doc/pub/DimRed/html/._DimRed-bs000.html @@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source None, '___sec2'), ('Simple preprocessing examples, breast cancer data and ' - 'classification', + 'classification, Support Vector Machines', 2, None, '___sec3'), - ('Principal Component Analysis', 2, None, '___sec4'), - ('PCA and scikit-learn', 2, None, '___sec5'), - ('More on the PCA', 2, None, '___sec6'), - ('Incremental PCA', 2, None, '___sec7'), - ('Randomized PCA', 2, None, '___sec8'), - ('Kernel PCA', 2, None, '___sec9'), - ('LLE', 2, None, '___sec10'), - ('Other techniques', 2, None, '___sec11')]} + ('More on Cancer Data, now with Logistic Regression', + 2, + None, + '___sec4'), + ('Principal Component Analysis', 2, None, '___sec5'), + ('PCA and scikit-learn', 2, None, '___sec6'), + ('More on the PCA', 2, None, '___sec7'), + ('Incremental PCA', 2, None, '___sec8'), + ('Randomized PCA', 2, None, '___sec9'), + ('Kernel PCA', 2, None, '___sec10'), + ('LLE', 2, None, '___sec11'), + ('Other techniques', 2, None, '___sec12')]} end of tocinfo --> @@ -103,15 +107,16 @@ MathJax.Hub.Config({
  • Reducing the number of degrees of freedom, overarching view
  • Preprocessing our data
  • Simple preprocessing examples, Franke function and regression
    • -
  • Simple preprocessing examples, breast cancer data and classification
    • -
  • Principal Component Analysis
    • -
  • PCA and scikit-learn
    • -
  • More on the PCA
    • -
  • Incremental PCA
    • -
  • Randomized PCA
    • -
  • Kernel PCA
    • -
  • LLE
    • -
  • Other techniques
    • +
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • +
  • More on Cancer Data, now with Logistic Regression
    • +
  • Principal Component Analysis
    • +
  • PCA and scikit-learn
    • +
  • More on the PCA
    • +
  • Incremental PCA
    • +
  • Randomized PCA
    • +
  • Kernel PCA
    • +
  • LLE
    • +
  • Other techniques
    • @@ -170,7 +175,7 @@ MathJax.Hub.Config({
  • 9
  • 10
  • ...
    • -
  • 13
    • +
  • 14
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs001.html b/doc/pub/DimRed/html/._DimRed-bs001.html index 8027c7441..c4c517c4b 100644 --- a/doc/pub/DimRed/html/._DimRed-bs001.html +++ b/doc/pub/DimRed/html/._DimRed-bs001.html @@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source None, '___sec2'), ('Simple preprocessing examples, breast cancer data and ' - 'classification', + 'classification, Support Vector Machines', 2, None, '___sec3'), - ('Principal Component Analysis', 2, None, '___sec4'), - ('PCA and scikit-learn', 2, None, '___sec5'), - ('More on the PCA', 2, None, '___sec6'), - ('Incremental PCA', 2, None, '___sec7'), - ('Randomized PCA', 2, None, '___sec8'), - ('Kernel PCA', 2, None, '___sec9'), - ('LLE', 2, None, '___sec10'), - ('Other techniques', 2, None, '___sec11')]} + ('More on Cancer Data, now with Logistic Regression', + 2, + None, + '___sec4'), + ('Principal Component Analysis', 2, None, '___sec5'), + ('PCA and scikit-learn', 2, None, '___sec6'), + ('More on the PCA', 2, None, '___sec7'), + ('Incremental PCA', 2, None, '___sec8'), + ('Randomized PCA', 2, None, '___sec9'), + ('Kernel PCA', 2, None, '___sec10'), + ('LLE', 2, None, '___sec11'), + ('Other techniques', 2, None, '___sec12')]} end of tocinfo --> @@ -103,15 +107,16 @@ MathJax.Hub.Config({
  • Reducing the number of degrees of freedom, overarching view
  • Preprocessing our data
  • Simple preprocessing examples, Franke function and regression
    • -
  • Simple preprocessing examples, breast cancer data and classification
    • -
  • Principal Component Analysis
    • -
  • PCA and scikit-learn
    • -
  • More on the PCA
    • -
  • Incremental PCA
    • -
  • Randomized PCA
    • -
  • Kernel PCA
    • -
  • LLE
    • -
  • Other techniques
    • +
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • +
  • More on Cancer Data, now with Logistic Regression
    • +
  • Principal Component Analysis
    • +
  • PCA and scikit-learn
    • +
  • More on the PCA
    • +
  • Incremental PCA
    • +
  • Randomized PCA
    • +
  • Kernel PCA
    • +
  • LLE
    • +
  • Other techniques
    • @@ -165,7 +170,7 @@ reduction techniques: the principal component analysis PCA, Kernel PCA, and Loca
  • 10
  • 11
  • ...
    • -
  • 13
    • +
  • 14
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs002.html b/doc/pub/DimRed/html/._DimRed-bs002.html index 9b0341c36..0b9829477 100644 --- a/doc/pub/DimRed/html/._DimRed-bs002.html +++ b/doc/pub/DimRed/html/._DimRed-bs002.html @@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source None, '___sec2'), ('Simple preprocessing examples, breast cancer data and ' - 'classification', + 'classification, Support Vector Machines', 2, None, '___sec3'), - ('Principal Component Analysis', 2, None, '___sec4'), - ('PCA and scikit-learn', 2, None, '___sec5'), - ('More on the PCA', 2, None, '___sec6'), - ('Incremental PCA', 2, None, '___sec7'), - ('Randomized PCA', 2, None, '___sec8'), - ('Kernel PCA', 2, None, '___sec9'), - ('LLE', 2, None, '___sec10'), - ('Other techniques', 2, None, '___sec11')]} + ('More on Cancer Data, now with Logistic Regression', + 2, + None, + '___sec4'), + ('Principal Component Analysis', 2, None, '___sec5'), + ('PCA and scikit-learn', 2, None, '___sec6'), + ('More on the PCA', 2, None, '___sec7'), + ('Incremental PCA', 2, None, '___sec8'), + ('Randomized PCA', 2, None, '___sec9'), + ('Kernel PCA', 2, None, '___sec10'), + ('LLE', 2, None, '___sec11'), + ('Other techniques', 2, None, '___sec12')]} end of tocinfo --> @@ -103,15 +107,16 @@ MathJax.Hub.Config({
  • Reducing the number of degrees of freedom, overarching view
  • Preprocessing our data
  • Simple preprocessing examples, Franke function and regression
    • -
  • Simple preprocessing examples, breast cancer data and classification
    • -
  • Principal Component Analysis
    • -
  • PCA and scikit-learn
    • -
  • More on the PCA
    • -
  • Incremental PCA
    • -
  • Randomized PCA
    • -
  • Kernel PCA
    • -
  • LLE
    • -
  • Other techniques
    • +
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • +
  • More on Cancer Data, now with Logistic Regression
    • +
  • Principal Component Analysis
    • +
  • PCA and scikit-learn
    • +
  • More on the PCA
    • +
  • Incremental PCA
    • +
  • Randomized PCA
    • +
  • Kernel PCA
    • +
  • LLE
    • +
  • Other techniques
    • @@ -167,7 +172,7 @@ This scaling has the drawback that it does not ensure that we have a particular
  • 11
  • 12
  • ...
    • -
  • 13
    • +
  • 14
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs003.html b/doc/pub/DimRed/html/._DimRed-bs003.html index b57689b38..6b99934de 100644 --- a/doc/pub/DimRed/html/._DimRed-bs003.html +++ b/doc/pub/DimRed/html/._DimRed-bs003.html @@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source None, '___sec2'), ('Simple preprocessing examples, breast cancer data and ' - 'classification', + 'classification, Support Vector Machines', 2, None, '___sec3'), - ('Principal Component Analysis', 2, None, '___sec4'), - ('PCA and scikit-learn', 2, None, '___sec5'), - ('More on the PCA', 2, None, '___sec6'), - ('Incremental PCA', 2, None, '___sec7'), - ('Randomized PCA', 2, None, '___sec8'), - ('Kernel PCA', 2, None, '___sec9'), - ('LLE', 2, None, '___sec10'), - ('Other techniques', 2, None, '___sec11')]} + ('More on Cancer Data, now with Logistic Regression', + 2, + None, + '___sec4'), + ('Principal Component Analysis', 2, None, '___sec5'), + ('PCA and scikit-learn', 2, None, '___sec6'), + ('More on the PCA', 2, None, '___sec7'), + ('Incremental PCA', 2, None, '___sec8'), + ('Randomized PCA', 2, None, '___sec9'), + ('Kernel PCA', 2, None, '___sec10'), + ('LLE', 2, None, '___sec11'), + ('Other techniques', 2, None, '___sec12')]} end of tocinfo --> @@ -103,15 +107,16 @@ MathJax.Hub.Config({
  • Reducing the number of degrees of freedom, overarching view
  • Preprocessing our data
  • Simple preprocessing examples, Franke function and regression
    • -
  • Simple preprocessing examples, breast cancer data and classification
    • -
  • Principal Component Analysis
    • -
  • PCA and scikit-learn
    • -
  • More on the PCA
    • -
  • Incremental PCA
    • -
  • Randomized PCA
    • -
  • Kernel PCA
    • -
  • LLE
    • -
  • Other techniques
    • +
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • +
  • More on Cancer Data, now with Logistic Regression
    • +
  • Principal Component Analysis
    • +
  • PCA and scikit-learn
    • +
  • More on the PCA
    • +
  • Incremental PCA
    • +
  • Randomized PCA
    • +
  • Kernel PCA
    • +
  • LLE
    • +
  • Other techniques
    • @@ -245,6 +250,8 @@ svm.fit(X_train_scaled, y_train)
  • 11
  • 12
  • 13
    • +
  • ...
    • +
  • 14
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs004.html b/doc/pub/DimRed/html/._DimRed-bs004.html index e7bf415b7..80787b95d 100644 --- a/doc/pub/DimRed/html/._DimRed-bs004.html +++ b/doc/pub/DimRed/html/._DimRed-bs004.html @@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source None, '___sec2'), ('Simple preprocessing examples, breast cancer data and ' - 'classification', + 'classification, Support Vector Machines', 2, None, '___sec3'), - ('Principal Component Analysis', 2, None, '___sec4'), - ('PCA and scikit-learn', 2, None, '___sec5'), - ('More on the PCA', 2, None, '___sec6'), - ('Incremental PCA', 2, None, '___sec7'), - ('Randomized PCA', 2, None, '___sec8'), - ('Kernel PCA', 2, None, '___sec9'), - ('LLE', 2, None, '___sec10'), - ('Other techniques', 2, None, '___sec11')]} + ('More on Cancer Data, now with Logistic Regression', + 2, + None, + '___sec4'), + ('Principal Component Analysis', 2, None, '___sec5'), + ('PCA and scikit-learn', 2, None, '___sec6'), + ('More on the PCA', 2, None, '___sec7'), + ('Incremental PCA', 2, None, '___sec8'), + ('Randomized PCA', 2, None, '___sec9'), + ('Kernel PCA', 2, None, '___sec10'), + ('LLE', 2, None, '___sec11'), + ('Other techniques', 2, None, '___sec12')]} end of tocinfo --> @@ -103,15 +107,16 @@ MathJax.Hub.Config({
  • Reducing the number of degrees of freedom, overarching view
  • Preprocessing our data
  • Simple preprocessing examples, Franke function and regression
    • -
  • Simple preprocessing examples, breast cancer data and classification
    • -
  • Principal Component Analysis
    • -
  • PCA and scikit-learn
    • -
  • More on the PCA
    • -
  • Incremental PCA
    • -
  • Randomized PCA
    • -
  • Kernel PCA
    • -
  • LLE
    • -
  • Other techniques
    • +
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • +
  • More on Cancer Data, now with Logistic Regression
    • +
  • Principal Component Analysis
    • +
  • PCA and scikit-learn
    • +
  • More on the PCA
    • +
  • Incremental PCA
    • +
  • Randomized PCA
    • +
  • Kernel PCA
    • +
  • LLE
    • +
  • Other techniques
    • @@ -127,7 +132,7 @@ MathJax.Hub.Config({ -

    Simple preprocessing examples, breast cancer data and classification

    +

    Simple preprocessing examples, breast cancer data and classification, Support Vector Machines

    We show here how we can use a simple regression case on the breast cancer data using support vector machine as algorithm for classification @@ -193,6 +198,7 @@ svm.fit(X_train_scaled, y_train)

  • 11
  • 12
  • 13
    • +
  • 14
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs005.html b/doc/pub/DimRed/html/._DimRed-bs005.html index 1ca4766d5..902a33e89 100644 --- a/doc/pub/DimRed/html/._DimRed-bs005.html +++ b/doc/pub/DimRed/html/._DimRed-bs005.html @@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source None, '___sec2'), ('Simple preprocessing examples, breast cancer data and ' - 'classification', + 'classification, Support Vector Machines', 2, None, '___sec3'), - ('Principal Component Analysis', 2, None, '___sec4'), - ('PCA and scikit-learn', 2, None, '___sec5'), - ('More on the PCA', 2, None, '___sec6'), - ('Incremental PCA', 2, None, '___sec7'), - ('Randomized PCA', 2, None, '___sec8'), - ('Kernel PCA', 2, None, '___sec9'), - ('LLE', 2, None, '___sec10'), - ('Other techniques', 2, None, '___sec11')]} + ('More on Cancer Data, now with Logistic Regression', + 2, + None, + '___sec4'), + ('Principal Component Analysis', 2, None, '___sec5'), + ('PCA and scikit-learn', 2, None, '___sec6'), + ('More on the PCA', 2, None, '___sec7'), + ('Incremental PCA', 2, None, '___sec8'), + ('Randomized PCA', 2, None, '___sec9'), + ('Kernel PCA', 2, None, '___sec10'), + ('LLE', 2, None, '___sec11'), + ('Other techniques', 2, None, '___sec12')]} end of tocinfo --> @@ -103,15 +107,16 @@ MathJax.Hub.Config({
  • Reducing the number of degrees of freedom, overarching view
  • Preprocessing our data
  • Simple preprocessing examples, Franke function and regression
    • -
  • Simple preprocessing examples, breast cancer data and classification
    • -
  • Principal Component Analysis
    • -
  • PCA and scikit-learn
    • -
  • More on the PCA
    • -
  • Incremental PCA
    • -
  • Randomized PCA
    • -
  • Kernel PCA
    • -
  • LLE
    • -
  • Other techniques
    • +
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • +
  • More on Cancer Data, now with Logistic Regression
    • +
  • Principal Component Analysis
    • +
  • PCA and scikit-learn
    • +
  • More on the PCA
    • +
  • Incremental PCA
    • +
  • Randomized PCA
    • +
  • Kernel PCA
    • +
  • LLE
    • +
  • Other techniques
    • @@ -127,38 +132,57 @@ MathJax.Hub.Config({ -

    Principal Component Analysis

    -
    -
    -

    -Principal Component Analysis (PCA) is by far the most popular dimensionality reduction algorithm. -First it identifies the hyperplane that lies closest to the data, and then it projects the data onto it. - -

    -The following Python code uses NumPy’s svd() function to obtain all the principal components of the -training set, then extracts the first two principal components +

    More on Cancer Data, now with Logistic Regression

    -

    X_centered = X - X.mean(axis=0)
-U, s, V = np.linalg.svd(X_centered)
-c1 = V.T[:, 0]
-c2 = V.T[:, 1]
-
    -

    -PCA assumes that the dataset is centered around the origin. Scikit-Learn’s PCA classes take care of centering -the data for you. However, if you implement PCA yourself (as in the preceding example), or if you use other libraries, don’t -forget to center the data first. +

    import matplotlib.pyplot as plt
+import numpy as np
+from sklearn.model_selection import  train_test_split 
+from sklearn.datasets import load_breast_cancer
+from sklearn.linear_model import LogisticRegression
+cancer = load_breast_cancer()
 
-
    
-Once you have identified all the principal components, you can reduce the dimensionality of the dataset
-down to \( d \) dimensions by projecting it onto the hyperplane defined by the first \( d \) principal components.
-Selecting this hyperplane ensures that the projection will preserve as much variance as possible. 
-
    
+fig, axes = plt.subplots(15,2,figsize=(10,20))
+male = cancer.data[cancer.target == 0]
+bene = cancer.data[cancer.target == 1]
+ax = axes.ravel()
 
-
-
    W2 = V.T[:, :2]
-X2D = X_centered.dot(W2)
+for i in range(30):
+    _, bins = np.histogram(cancer.data[:,i], bins =50)
+    ax[i].hist(male[:,i], bins = bins, alpha = 0.5)
+    ax[i].hist(bene[:,i], bins = bins, alpha = 0.5)
+    ax[i].set_title(cancer.feature_names[i])
+    ax[i].set_yticks(())
+ax[0].set_xlabel("Feature magnitude")
+ax[0].set_ylabel("Frequency")
+ax[0].legend(["Male", "Bene"], loc ="best")
+fig.tight_layout()
+plt.show()
+
+# Set up training data
+X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)
+# Perform Logistic Regression (replace with own code) 
+logreg = LogisticRegression()
+logreg.fit(X_train, y_train)
+print("Test set accuracy: {:.2f}".format(logreg.score(X_test,y_test)))
+
+# Scale data
+from sklearn.preprocessing import StandardScaler
+scaler = StandardScaler()
+scaler.fit(X_train)
+X_train_scaled = scaler.transform(X_train)
+X_test_scaled = scaler.transform(X_test)
+logreg.fit(X_train_scaled, y_train)
+#svm.fit(X_train_scaled, y_train)
+print("Test set accuracy scaled data: {:.2f}".format(logreg.score(X_test_scaled,y_test)))
+
+# Now add PCA
+from sklearn.decomposition import PCA
+pca = PCA(n_components = 2)
+pca.fit(X_train_scaled)
+
+X_pca = pca.transform(X_train_scaled)
 
    
 
    
 
    
@@ -178,6 +202,7 @@ X2D = X_centered11
   
  • 12
    • 
   
  • 13
    • 
+  
  • 14
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs006.html b/doc/pub/DimRed/html/._DimRed-bs006.html
index 76dff32c4..788009051 100644
--- a/doc/pub/DimRed/html/._DimRed-bs006.html
+++ b/doc/pub/DimRed/html/._DimRed-bs006.html
@@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source
                None,
                '___sec2'),
               ('Simple preprocessing examples, breast cancer data and '
-               'classification',
+               'classification, Support Vector Machines',
                2,
                None,
                '___sec3'),
-              ('Principal Component Analysis', 2, None, '___sec4'),
-              ('PCA and scikit-learn', 2, None, '___sec5'),
-              ('More on the PCA', 2, None, '___sec6'),
-              ('Incremental PCA', 2, None, '___sec7'),
-              ('Randomized PCA', 2, None, '___sec8'),
-              ('Kernel PCA', 2, None, '___sec9'),
-              ('LLE', 2, None, '___sec10'),
-              ('Other techniques', 2, None, '___sec11')]}
+              ('More on Cancer Data, now with Logistic Regression',
+               2,
+               None,
+               '___sec4'),
+              ('Principal Component Analysis', 2, None, '___sec5'),
+              ('PCA and scikit-learn', 2, None, '___sec6'),
+              ('More on the PCA', 2, None, '___sec7'),
+              ('Incremental PCA', 2, None, '___sec8'),
+              ('Randomized PCA', 2, None, '___sec9'),
+              ('Kernel PCA', 2, None, '___sec10'),
+              ('LLE', 2, None, '___sec11'),
+              ('Other techniques', 2, None, '___sec12')]}
 end of tocinfo -->
 
 
@@ -103,15 +107,16 @@ MathJax.Hub.Config({
       
  • Reducing the number of degrees of freedom, overarching view
    • 
       
  • Preprocessing our data
    • 
       
  • Simple preprocessing examples, Franke function and regression
    • 
-      
  • Simple preprocessing examples, breast cancer data and classification
    • 
-      
  • Principal Component Analysis
    • 
-      
  • PCA and scikit-learn
    • 
-      
  • More on the PCA
    • 
-      
  • Incremental PCA
    • 
-      
  • Randomized PCA
    • 
-      
  • Kernel PCA
    • 
-      
  • LLE
    • 
-      
  • Other techniques
    • 
+      
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • 
+      
  • More on Cancer Data, now with Logistic Regression
    • 
+      
  • Principal Component Analysis
    • 
+      
  • PCA and scikit-learn
    • 
+      
  • More on the PCA
    • 
+      
  • Incremental PCA
    • 
+      
  • Randomized PCA
    • 
+      
  • Kernel PCA
    • 
+      
  • LLE
    • 
+      
  • Other techniques
    • 
 
         
       
@@ -125,36 +130,41 @@ MathJax.Hub.Config({
 
     
     
     
     
 
 
-
+
 
-
    PCA and scikit-learn 
    
+
    Principal Component Analysis 
    
+
    
+
    
+
     
+Principal Component Analysis (PCA) is by far the most popular dimensionality reduction algorithm.
+First it identifies the hyperplane that lies closest to the data, and then it projects the data onto it.
 
 
    
-Scikit-Learn’s PCA class implements PCA using SVD decomposition just like we did before. The
-following code applies PCA to reduce the dimensionality of the dataset down to two dimensions (note
-that it automatically takes care of centering the data):
+The following Python code uses NumPy’s svd() function to obtain all the principal components of the
+training set, then extracts the first two principal components
 
    
 
 
-
    from sklearn.decomposition import PCA
-pca = PCA(n_components = 2)
-X2D = pca.fit_transform(X)
+
    X_centered = X - X.mean(axis=0)
+U, s, V = np.linalg.svd(X_centered)
+c1 = V.T[:, 0]
+c2 = V.T[:, 1]
 
    
 
    
-After fitting the PCA transformer to the dataset, you can access the principal components using the
-components variable (note that it contains the PCs as horizontal vectors, so, for example, the first
-principal component is equal to 
+PCA assumes that the dataset is centered around the origin. Scikit-Learn’s PCA classes take care of centering
+the data for you. However, if you implement PCA yourself (as in the preceding example), or if you use other libraries, don’t
+forget to center the data first.
+
+
    
+Once you have identified all the principal components, you can reduce the dimensionality of the dataset
+down to \( d \) dimensions by projecting it onto the hyperplane defined by the first \( d \) principal components.
+Selecting this hyperplane ensures that the projection will preserve as much variance as possible. 
 
    
 
 
-
    pca.components_.T[:, 0]).
+
    W2 = V.T[:, :2]
+X2D = X_centered.dot(W2)
 
    
-
    
-Another very useful piece of information is the explained variance ratio of each principal component,
-available via the \( explained\_variance\_ratio \) variable. It indicates the proportion of the dataset’s
-variance that lies along the axis of each principal component. 
-More material to come here.
-
 
    
 
    
 
@@ -173,6 +183,7 @@ More material to come here.
   
  • 11
    • 
   
  • 12
    • 
   
  • 13
    • 
+  
  • 14
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs007.html b/doc/pub/DimRed/html/._DimRed-bs007.html
index b93647590..8f57f5294 100644
--- a/doc/pub/DimRed/html/._DimRed-bs007.html
+++ b/doc/pub/DimRed/html/._DimRed-bs007.html
@@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source
                None,
                '___sec2'),
               ('Simple preprocessing examples, breast cancer data and '
-               'classification',
+               'classification, Support Vector Machines',
                2,
                None,
                '___sec3'),
-              ('Principal Component Analysis', 2, None, '___sec4'),
-              ('PCA and scikit-learn', 2, None, '___sec5'),
-              ('More on the PCA', 2, None, '___sec6'),
-              ('Incremental PCA', 2, None, '___sec7'),
-              ('Randomized PCA', 2, None, '___sec8'),
-              ('Kernel PCA', 2, None, '___sec9'),
-              ('LLE', 2, None, '___sec10'),
-              ('Other techniques', 2, None, '___sec11')]}
+              ('More on Cancer Data, now with Logistic Regression',
+               2,
+               None,
+               '___sec4'),
+              ('Principal Component Analysis', 2, None, '___sec5'),
+              ('PCA and scikit-learn', 2, None, '___sec6'),
+              ('More on the PCA', 2, None, '___sec7'),
+              ('Incremental PCA', 2, None, '___sec8'),
+              ('Randomized PCA', 2, None, '___sec9'),
+              ('Kernel PCA', 2, None, '___sec10'),
+              ('LLE', 2, None, '___sec11'),
+              ('Other techniques', 2, None, '___sec12')]}
 end of tocinfo -->
 
 
@@ -103,15 +107,16 @@ MathJax.Hub.Config({
       
  • Reducing the number of degrees of freedom, overarching view
    • 
       
  • Preprocessing our data
    • 
       
  • Simple preprocessing examples, Franke function and regression
    • 
-      
  • Simple preprocessing examples, breast cancer data and classification
    • 
-      
  • Principal Component Analysis
    • 
-      
  • PCA and scikit-learn
    • 
-      
  • More on the PCA
    • 
-      
  • Incremental PCA
    • 
-      
  • Randomized PCA
    • 
-      
  • Kernel PCA
    • 
-      
  • LLE
    • 
-      
  • Other techniques
    • 
+      
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • 
+      
  • More on Cancer Data, now with Logistic Regression
    • 
+      
  • Principal Component Analysis
    • 
+      
  • PCA and scikit-learn
    • 
+      
  • More on the PCA
    • 
+      
  • Incremental PCA
    • 
+      
  • Randomized PCA
    • 
+      
  • Kernel PCA
    • 
+      
  • LLE
    • 
+      
  • Other techniques
    • 
 
         
       
@@ -125,33 +130,36 @@ MathJax.Hub.Config({
 
     
     
     
     
 
 
-
+
 
-
    More on the PCA 
    
-Instead of arbitrarily choosing the number of dimensions to reduce down to, it is generally preferable to
-choose the number of dimensions that add up to a sufficiently large portion of the variance (e.g., 95%).
-Unless, of course, you are reducing dimensionality for data visualization — in that case you will
-generally want to reduce the dimensionality down to 2 or 3.
-The following code computes PCA without reducing dimensionality, then computes the minimum number
-of dimensions required to preserve 95% of the training set’s variance:
+
    PCA and scikit-learn 
    
+
+
    
+Scikit-Learn’s PCA class implements PCA using SVD decomposition just like we did before. The
+following code applies PCA to reduce the dimensionality of the dataset down to two dimensions (note
+that it automatically takes care of centering the data):
 
    
 
 
-
    pca = PCA()
-pca.fit(X)
-cumsum = np.cumsum(pca.explained_variance_ratio_)
-d = np.argmax(cumsum >= 0.95) + 1
+
    from sklearn.decomposition import PCA
+pca = PCA(n_components = 2)
+X2D = pca.fit_transform(X)
 
    
 
    
-You could then set \( n\_components=d \) and run PCA again. However, there is a much better option: instead
-of specifying the number of principal components you want to preserve, you can set \( n\_components \) to be
-a float between 0.0 and 1.0, indicating the ratio of variance you wish to preserve:
+After fitting the PCA transformer to the dataset, you can access the principal components using the
+components variable (note that it contains the PCs as horizontal vectors, so, for example, the first
+principal component is equal to 
 
    
 
 
-
    pca = PCA(n_components=0.95)
-X_reduced = pca.fit_transform(X)
+
    pca.components_.T[:, 0]).
 
    
+
    
+Another very useful piece of information is the explained variance ratio of each principal component,
+available via the \( explained\_variance\_ratio \) variable. It indicates the proportion of the dataset’s
+variance that lies along the axis of each principal component. 
+More material to come here.
+
 
    
 
    
 
@@ -170,6 +178,7 @@ X_reduced = pca
   
  • 11
    • 
   
  • 12
    • 
   
  • 13
    • 
+  
  • 14
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs008.html b/doc/pub/DimRed/html/._DimRed-bs008.html
index 9c9289160..2770b2efb 100644
--- a/doc/pub/DimRed/html/._DimRed-bs008.html
+++ b/doc/pub/DimRed/html/._DimRed-bs008.html
@@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source
                None,
                '___sec2'),
               ('Simple preprocessing examples, breast cancer data and '
-               'classification',
+               'classification, Support Vector Machines',
                2,
                None,
                '___sec3'),
-              ('Principal Component Analysis', 2, None, '___sec4'),
-              ('PCA and scikit-learn', 2, None, '___sec5'),
-              ('More on the PCA', 2, None, '___sec6'),
-              ('Incremental PCA', 2, None, '___sec7'),
-              ('Randomized PCA', 2, None, '___sec8'),
-              ('Kernel PCA', 2, None, '___sec9'),
-              ('LLE', 2, None, '___sec10'),
-              ('Other techniques', 2, None, '___sec11')]}
+              ('More on Cancer Data, now with Logistic Regression',
+               2,
+               None,
+               '___sec4'),
+              ('Principal Component Analysis', 2, None, '___sec5'),
+              ('PCA and scikit-learn', 2, None, '___sec6'),
+              ('More on the PCA', 2, None, '___sec7'),
+              ('Incremental PCA', 2, None, '___sec8'),
+              ('Randomized PCA', 2, None, '___sec9'),
+              ('Kernel PCA', 2, None, '___sec10'),
+              ('LLE', 2, None, '___sec11'),
+              ('Other techniques', 2, None, '___sec12')]}
 end of tocinfo -->
 
 
@@ -103,15 +107,16 @@ MathJax.Hub.Config({
       
  • Reducing the number of degrees of freedom, overarching view
    • 
       
  • Preprocessing our data
    • 
       
  • Simple preprocessing examples, Franke function and regression
    • 
-      
  • Simple preprocessing examples, breast cancer data and classification
    • 
-      
  • Principal Component Analysis
    • 
-      
  • PCA and scikit-learn
    • 
-      
  • More on the PCA
    • 
-      
  • Incremental PCA
    • 
-      
  • Randomized PCA
    • 
-      
  • Kernel PCA
    • 
-      
  • LLE
    • 
-      
  • Other techniques
    • 
+      
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • 
+      
  • More on Cancer Data, now with Logistic Regression
    • 
+      
  • Principal Component Analysis
    • 
+      
  • PCA and scikit-learn
    • 
+      
  • More on the PCA
    • 
+      
  • Incremental PCA
    • 
+      
  • Randomized PCA
    • 
+      
  • Kernel PCA
    • 
+      
  • LLE
    • 
+      
  • Other techniques
    • 
 
         
       
@@ -127,13 +132,31 @@ MathJax.Hub.Config({
 
 
 
-
    Incremental PCA 
    
-One problem with the preceding implementation of PCA is that it requires the whole training set to fit in
-memory in order for the SVD algorithm to run. Fortunately, Incremental PCA (IPCA) algorithms have
-been developed: you can split the training set into mini-batches and feed an IPCA algorithm one minibatch
-at a time. This is useful for large training sets, and also to apply PCA online (i.e., on the fly, as new
-instances arrive).
+
    More on the PCA 
    
+Instead of arbitrarily choosing the number of dimensions to reduce down to, it is generally preferable to
+choose the number of dimensions that add up to a sufficiently large portion of the variance (e.g., 95%).
+Unless, of course, you are reducing dimensionality for data visualization — in that case you will
+generally want to reduce the dimensionality down to 2 or 3.
+The following code computes PCA without reducing dimensionality, then computes the minimum number
+of dimensions required to preserve 95% of the training set’s variance:
+
    
 
+
+
    pca = PCA()
+pca.fit(X)
+cumsum = np.cumsum(pca.explained_variance_ratio_)
+d = np.argmax(cumsum >= 0.95) + 1
+
    
+
    
+You could then set \( n\_components=d \) and run PCA again. However, there is a much better option: instead
+of specifying the number of principal components you want to preserve, you can set \( n\_components \) to be
+a float between 0.0 and 1.0, indicating the ratio of variance you wish to preserve:
+
    
+
+
+
    pca = PCA(n_components=0.95)
+X_reduced = pca.fit_transform(X)
+
    
 
    
 
    
 
@@ -152,6 +175,7 @@ instances arrive).
   
  • 11
    • 
   
  • 12
    • 
   
  • 13
    • 
+  
  • 14
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs009.html b/doc/pub/DimRed/html/._DimRed-bs009.html
index dd7fef739..068bb5444 100644
--- a/doc/pub/DimRed/html/._DimRed-bs009.html
+++ b/doc/pub/DimRed/html/._DimRed-bs009.html
@@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source
                None,
                '___sec2'),
               ('Simple preprocessing examples, breast cancer data and '
-               'classification',
+               'classification, Support Vector Machines',
                2,
                None,
                '___sec3'),
-              ('Principal Component Analysis', 2, None, '___sec4'),
-              ('PCA and scikit-learn', 2, None, '___sec5'),
-              ('More on the PCA', 2, None, '___sec6'),
-              ('Incremental PCA', 2, None, '___sec7'),
-              ('Randomized PCA', 2, None, '___sec8'),
-              ('Kernel PCA', 2, None, '___sec9'),
-              ('LLE', 2, None, '___sec10'),
-              ('Other techniques', 2, None, '___sec11')]}
+              ('More on Cancer Data, now with Logistic Regression',
+               2,
+               None,
+               '___sec4'),
+              ('Principal Component Analysis', 2, None, '___sec5'),
+              ('PCA and scikit-learn', 2, None, '___sec6'),
+              ('More on the PCA', 2, None, '___sec7'),
+              ('Incremental PCA', 2, None, '___sec8'),
+              ('Randomized PCA', 2, None, '___sec9'),
+              ('Kernel PCA', 2, None, '___sec10'),
+              ('LLE', 2, None, '___sec11'),
+              ('Other techniques', 2, None, '___sec12')]}
 end of tocinfo -->
 
 
@@ -103,15 +107,16 @@ MathJax.Hub.Config({
       
  • Reducing the number of degrees of freedom, overarching view
    • 
       
  • Preprocessing our data
    • 
       
  • Simple preprocessing examples, Franke function and regression
    • 
-      
  • Simple preprocessing examples, breast cancer data and classification
    • 
-      
  • Principal Component Analysis
    • 
-      
  • PCA and scikit-learn
    • 
-      
  • More on the PCA
    • 
-      
  • Incremental PCA
    • 
-      
  • Randomized PCA
    • 
-      
  • Kernel PCA
    • 
-      
  • LLE
    • 
-      
  • Other techniques
    • 
+      
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • 
+      
  • More on Cancer Data, now with Logistic Regression
    • 
+      
  • Principal Component Analysis
    • 
+      
  • PCA and scikit-learn
    • 
+      
  • More on the PCA
    • 
+      
  • Incremental PCA
    • 
+      
  • Randomized PCA
    • 
+      
  • Kernel PCA
    • 
+      
  • LLE
    • 
+      
  • Other techniques
    • 
 
         
       
@@ -127,18 +132,12 @@ MathJax.Hub.Config({
 
 
 
-
    Randomized PCA 
    
-
-
    
-Scikit-Learn offers yet another option to perform PCA, called Randomized PCA. This is a stochastic
-algorithm that quickly finds an approximation of the first d principal components. Its computational
-complexity is \( O(m \times d^2)+O(d^3) \), instead of \( O(m \times n^2) + O(n^3) \), so it is dramatically faster than the
-previous algorithms when \( d \) is much smaller than \( n \).
-
-
    
-
    
-
    
-
+
    Incremental PCA 
    
+One problem with the preceding implementation of PCA is that it requires the whole training set to fit in
+memory in order for the SVD algorithm to run. Fortunately, Incremental PCA (IPCA) algorithms have
+been developed: you can split the training set into mini-batches and feed an IPCA algorithm one minibatch
+at a time. This is useful for large training sets, and also to apply PCA online (i.e., on the fly, as new
+instances arrive).
 
 
    
 
    
@@ -158,6 +157,7 @@ previous algorithms when \( d \) is much smaller than \( n \).
   
  • 11
    • 
   
  • 12
    • 
   
  • 13
    • 
+  
  • 14
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/DimRed-bs.html b/doc/pub/DimRed/html/DimRed-bs.html
index 84af171c4..706f14ce8 100644
--- a/doc/pub/DimRed/html/DimRed-bs.html
+++ b/doc/pub/DimRed/html/DimRed-bs.html
@@ -51,18 +51,22 @@ Automatically generated HTML file from DocOnce source
                None,
                '___sec2'),
               ('Simple preprocessing examples, breast cancer data and '
-               'classification',
+               'classification, Support Vector Machines',
                2,
                None,
                '___sec3'),
-              ('Principal Component Analysis', 2, None, '___sec4'),
-              ('PCA and scikit-learn', 2, None, '___sec5'),
-              ('More on the PCA', 2, None, '___sec6'),
-              ('Incremental PCA', 2, None, '___sec7'),
-              ('Randomized PCA', 2, None, '___sec8'),
-              ('Kernel PCA', 2, None, '___sec9'),
-              ('LLE', 2, None, '___sec10'),
-              ('Other techniques', 2, None, '___sec11')]}
+              ('More on Cancer Data, now with Logistic Regression',
+               2,
+               None,
+               '___sec4'),
+              ('Principal Component Analysis', 2, None, '___sec5'),
+              ('PCA and scikit-learn', 2, None, '___sec6'),
+              ('More on the PCA', 2, None, '___sec7'),
+              ('Incremental PCA', 2, None, '___sec8'),
+              ('Randomized PCA', 2, None, '___sec9'),
+              ('Kernel PCA', 2, None, '___sec10'),
+              ('LLE', 2, None, '___sec11'),
+              ('Other techniques', 2, None, '___sec12')]}
 end of tocinfo -->
 
 
@@ -103,15 +107,16 @@ MathJax.Hub.Config({
       
  • Reducing the number of degrees of freedom, overarching view
    • 
       
  • Preprocessing our data
    • 
       
  • Simple preprocessing examples, Franke function and regression
    • 
-      
  • Simple preprocessing examples, breast cancer data and classification
    • 
-      
  • Principal Component Analysis
    • 
-      
  • PCA and scikit-learn
    • 
-      
  • More on the PCA
    • 
-      
  • Incremental PCA
    • 
-      
  • Randomized PCA
    • 
-      
  • Kernel PCA
    • 
-      
  • LLE
    • 
-      
  • Other techniques
    • 
+      
  • Simple preprocessing examples, breast cancer data and classification, Support Vector Machines
    • 
+      
  • More on Cancer Data, now with Logistic Regression
    • 
+      
  • Principal Component Analysis
    • 
+      
  • PCA and scikit-learn
    • 
+      
  • More on the PCA
    • 
+      
  • Incremental PCA
    • 
+      
  • Randomized PCA
    • 
+      
  • Kernel PCA
    • 
+      
  • LLE
    • 
+      
  • Other techniques
    • 
 
         
       
@@ -170,7 +175,7 @@ MathJax.Hub.Config({
   
  • 9
    • 
   
  • 10
    • 
   
  • ...
    • 
-  
  • 13
    • 
+  
  • 14
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/DimRed-reveal.html b/doc/pub/DimRed/html/DimRed-reveal.html
index ba5ca09ac..bda964130 100644
--- a/doc/pub/DimRed/html/DimRed-reveal.html
+++ b/doc/pub/DimRed/html/DimRed-reveal.html
@@ -304,7 +304,7 @@ svm.fit(X_train_scaled, y_train)
 
 
 
    
-
    Simple preprocessing examples, breast cancer data and classification 
    
+
    Simple preprocessing examples, breast cancer data and classification, Support Vector Machines  
    
 
 
    
 We show here how we can use a simple regression case on the breast cancer data using support vector machine as algorithm for classification
@@ -356,7 +356,63 @@ svm.fit(X_train_scaled, y_train)
 
 
 
    
-
    Principal Component Analysis 
    
+
    More on Cancer Data, now with Logistic Regression  
    
+
    
+
+
+
    import matplotlib.pyplot as plt
+import numpy as np
+from sklearn.model_selection import  train_test_split 
+from sklearn.datasets import load_breast_cancer
+from sklearn.linear_model import LogisticRegression
+cancer = load_breast_cancer()
+
+fig, axes = plt.subplots(15,2,figsize=(10,20))
+male = cancer.data[cancer.target == 0]
+bene = cancer.data[cancer.target == 1]
+ax = axes.ravel()
+
+for i in range(30):
+    _, bins = np.histogram(cancer.data[:,i], bins =50)
+    ax[i].hist(male[:,i], bins = bins, alpha = 0.5)
+    ax[i].hist(bene[:,i], bins = bins, alpha = 0.5)
+    ax[i].set_title(cancer.feature_names[i])
+    ax[i].set_yticks(())
+ax[0].set_xlabel("Feature magnitude")
+ax[0].set_ylabel("Frequency")
+ax[0].legend(["Male", "Bene"], loc ="best")
+fig.tight_layout()
+plt.show()
+
+# Set up training data
+X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)
+# Perform Logistic Regression (replace with own code) 
+logreg = LogisticRegression()
+logreg.fit(X_train, y_train)
+print("Test set accuracy: {:.2f}".format(logreg.score(X_test,y_test)))
+
+# Scale data
+from sklearn.preprocessing import StandardScaler
+scaler = StandardScaler()
+scaler.fit(X_train)
+X_train_scaled = scaler.transform(X_train)
+X_test_scaled = scaler.transform(X_test)
+logreg.fit(X_train_scaled, y_train)
+#svm.fit(X_train_scaled, y_train)
+print("Test set accuracy scaled data: {:.2f}".format(logreg.score(X_test_scaled,y_test)))
+
+# Now add PCA
+from sklearn.decomposition import PCA
+pca = PCA(n_components = 2)
+pca.fit(X_train_scaled)
+
+X_pca = pca.transform(X_train_scaled)
+
    
+
    
+
+
+
    
+
    Principal Component Analysis 
    
 
    
 
 
    
@@ -393,7 +449,7 @@ X2D = X_centered.dot(W2)
 
 
 
    
-
    PCA and scikit-learn 
    
+
    PCA and scikit-learn 
    
 
 
    
 Scikit-Learn’s PCA class implements PCA using SVD decomposition just like we did before. The
@@ -424,7 +480,7 @@ More material to come here.
 
 
 
    
-
    More on the PCA 
    
+
    More on the PCA 
    
 Instead of arbitrarily choosing the number of dimensions to reduce down to, it is generally preferable to
 choose the number of dimensions that add up to a sufficiently large portion of the variance (e.g., 95%).
 Unless, of course, you are reducing dimensionality for data visualization — in that case you will
@@ -453,7 +509,7 @@ X_reduced = pca.fit_transform(X)
 
 
 
    
-
    Incremental PCA 
    
+
    Incremental PCA 
    
 One problem with the preceding implementation of PCA is that it requires the whole training set to fit in
 memory in order for the SVD algorithm to run. Fortunately, Incremental PCA (IPCA) algorithms have
 been developed: you can split the training set into mini-batches and feed an IPCA algorithm one minibatch
@@ -463,7 +519,7 @@ instances arrive).
 
 
 
    
-
    Randomized PCA 
    
+
    Randomized PCA 
    
 
 
    
 Scikit-Learn offers yet another option to perform PCA, called Randomized PCA. This is a stochastic
@@ -477,7 +533,7 @@ previous algorithms when \( d \) is much smaller than \( n \).
 
 
 
    
-
    Kernel PCA 
    
+
    Kernel PCA 
    
 
    
 
 
    
@@ -503,7 +559,7 @@ X_reduced = rbf_pca.fit_transform(X)
 
 
 
    
-
    LLE 
    
+
    LLE 
    
 
 
    
 Locally Linear Embedding (LLE) is another very powerful nonlinear dimensionality reduction
@@ -515,7 +571,7 @@ these local relationships are best preserved (more details shortly).
 
 
 
    
-
    Other techniques 
    
+
    Other techniques 
    
 
 
    
 There are many other dimensionality reduction techniques, several of which are available in Scikit-Learn.
diff --git a/doc/pub/DimRed/html/DimRed-solarized.html b/doc/pub/DimRed/html/DimRed-solarized.html
index 829228a99..f18a7b72b 100644
--- a/doc/pub/DimRed/html/DimRed-solarized.html
+++ b/doc/pub/DimRed/html/DimRed-solarized.html
@@ -71,18 +71,22 @@ div { text-align: justify; text-justify: inter-word; }
                None,
                '___sec2'),
               ('Simple preprocessing examples, breast cancer data and '
-               'classification',
+               'classification, Support Vector Machines',
                2,
                None,
                '___sec3'),
-              ('Principal Component Analysis', 2, None, '___sec4'),
-              ('PCA and scikit-learn', 2, None, '___sec5'),
-              ('More on the PCA', 2, None, '___sec6'),
-              ('Incremental PCA', 2, None, '___sec7'),
-              ('Randomized PCA', 2, None, '___sec8'),
-              ('Kernel PCA', 2, None, '___sec9'),
-              ('LLE', 2, None, '___sec10'),
-              ('Other techniques', 2, None, '___sec11')]}
+              ('More on Cancer Data, now with Logistic Regression',
+               2,
+               None,
+               '___sec4'),
+              ('Principal Component Analysis', 2, None, '___sec5'),
+              ('PCA and scikit-learn', 2, None, '___sec6'),
+              ('More on the PCA', 2, None, '___sec7'),
+              ('Incremental PCA', 2, None, '___sec8'),
+              ('Randomized PCA', 2, None, '___sec9'),
+              ('Kernel PCA', 2, None, '___sec10'),
+              ('LLE', 2, None, '___sec11'),
+              ('Other techniques', 2, None, '___sec12')]}
 end of tocinfo -->
 
 
@@ -279,7 +283,7 @@ svm.fit(X_train_scaled, y_train)
 
    
 









    
 
-
    Simple preprocessing examples, breast cancer data and classification 
    
+
    Simple preprocessing examples, breast cancer data and classification, Support Vector Machines  
    
 
 
    
 We show here how we can use a simple regression case on the breast cancer data using support vector machine as algorithm for classification
@@ -330,7 +334,62 @@ svm.fit(X_train_scaled, y_train)
 
    
 









    
 
-
    Principal Component Analysis 
    
+
    More on Cancer Data, now with Logistic Regression  
    
+
    
+
+
+
    import matplotlib.pyplot as plt
+import numpy as np
+from sklearn.model_selection import  train_test_split 
+from sklearn.datasets import load_breast_cancer
+from sklearn.linear_model import LogisticRegression
+cancer = load_breast_cancer()
+
+fig, axes = plt.subplots(15,2,figsize=(10,20))
+male = cancer.data[cancer.target == 0]
+bene = cancer.data[cancer.target == 1]
+ax = axes.ravel()
+
+for i in range(30):
+    _, bins = np.histogram(cancer.data[:,i], bins =50)
+    ax[i].hist(male[:,i], bins = bins, alpha = 0.5)
+    ax[i].hist(bene[:,i], bins = bins, alpha = 0.5)
+    ax[i].set_title(cancer.feature_names[i])
+    ax[i].set_yticks(())
+ax[0].set_xlabel("Feature magnitude")
+ax[0].set_ylabel("Frequency")
+ax[0].legend(["Male", "Bene"], loc ="best")
+fig.tight_layout()
+plt.show()
+
+# Set up training data
+X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)
+# Perform Logistic Regression (replace with own code) 
+logreg = LogisticRegression()
+logreg.fit(X_train, y_train)
+print("Test set accuracy: {:.2f}".format(logreg.score(X_test,y_test)))
+
+# Scale data
+from sklearn.preprocessing import StandardScaler
+scaler = StandardScaler()
+scaler.fit(X_train)
+X_train_scaled = scaler.transform(X_train)
+X_test_scaled = scaler.transform(X_test)
+logreg.fit(X_train_scaled, y_train)
+#svm.fit(X_train_scaled, y_train)
+print("Test set accuracy scaled data: {:.2f}".format(logreg.score(X_test_scaled,y_test)))
+
+# Now add PCA
+from sklearn.decomposition import PCA
+pca = PCA(n_components = 2)
+pca.fit(X_train_scaled)
+
+X_pca = pca.transform(X_train_scaled)
+
    
+
    
+









    
+
+
    Principal Component Analysis 
    
 
    
 
 
    
@@ -366,7 +425,7 @@ X2D = X_centered.dot(W2)
 
    
 
 
-
    PCA and scikit-learn 
    
+
    PCA and scikit-learn 
    
 
 
    
 Scikit-Learn’s PCA class implements PCA using SVD decomposition just like we did before. The
@@ -397,7 +456,7 @@ More material to come here.
 
    
 









    
 
-
    More on the PCA 
    
+
    More on the PCA 
    
 Instead of arbitrarily choosing the number of dimensions to reduce down to, it is generally preferable to
 choose the number of dimensions that add up to a sufficiently large portion of the variance (e.g., 95%).
 Unless, of course, you are reducing dimensionality for data visualization — in that case you will
@@ -425,7 +484,7 @@ X_reduced = pca.fit_transform(X)
 
    
 









    
 
-
    Incremental PCA 
    
+
    Incremental PCA 
    
 One problem with the preceding implementation of PCA is that it requires the whole training set to fit in
 memory in order for the SVD algorithm to run. Fortunately, Incremental PCA (IPCA) algorithms have
 been developed: you can split the training set into mini-batches and feed an IPCA algorithm one minibatch
@@ -435,7 +494,7 @@ instances arrive).
 
    
 









    
 
-
    Randomized PCA 
    
+
    Randomized PCA 
    
 
 
    
 Scikit-Learn offers yet another option to perform PCA, called Randomized PCA. This is a stochastic
@@ -450,7 +509,7 @@ previous algorithms when \( d \) is much smaller than \( n \).
 
    
 









    
 
-
    Kernel PCA 
    
+
    Kernel PCA 
    
 
    
 
 
    
@@ -479,7 +538,7 @@ X_reduced = rbf_pca.fit_transform(X)
 
    
 









    
 
-
    LLE 
    
+
    LLE 
    
 
 
    
 Locally Linear Embedding (LLE) is another very powerful nonlinear dimensionality reduction
@@ -491,7 +550,7 @@ these local relationships are best preserved (more details shortly).
 
    
 









    
 
-
    Other techniques 
    
+
    Other techniques 
    
 
 
    
 There are many other dimensionality reduction techniques, several of which are available in Scikit-Learn.
diff --git a/doc/pub/DimRed/html/DimRed.html b/doc/pub/DimRed/html/DimRed.html
index 421408fe5..b339bc54e 100644
--- a/doc/pub/DimRed/html/DimRed.html
+++ b/doc/pub/DimRed/html/DimRed.html
@@ -76,18 +76,22 @@ div { text-align: justify; text-justify: inter-word; }
                None,
                '___sec2'),
               ('Simple preprocessing examples, breast cancer data and '
-               'classification',
+               'classification, Support Vector Machines',
                2,
                None,
                '___sec3'),
-              ('Principal Component Analysis', 2, None, '___sec4'),
-              ('PCA and scikit-learn', 2, None, '___sec5'),
-              ('More on the PCA', 2, None, '___sec6'),
-              ('Incremental PCA', 2, None, '___sec7'),
-              ('Randomized PCA', 2, None, '___sec8'),
-              ('Kernel PCA', 2, None, '___sec9'),
-              ('LLE', 2, None, '___sec10'),
-              ('Other techniques', 2, None, '___sec11')]}
+              ('More on Cancer Data, now with Logistic Regression',
+               2,
+               None,
+               '___sec4'),
+              ('Principal Component Analysis', 2, None, '___sec5'),
+              ('PCA and scikit-learn', 2, None, '___sec6'),
+              ('More on the PCA', 2, None, '___sec7'),
+              ('Incremental PCA', 2, None, '___sec8'),
+              ('Randomized PCA', 2, None, '___sec9'),
+              ('Kernel PCA', 2, None, '___sec10'),
+              ('LLE', 2, None, '___sec11'),
+              ('Other techniques', 2, None, '___sec12')]}
 end of tocinfo -->
 
 
@@ -284,7 +288,7 @@ svm.fit(X_train_scaled, y_train)
 
    
 









    
 
-
    Simple preprocessing examples, breast cancer data and classification 
    
+
    Simple preprocessing examples, breast cancer data and classification, Support Vector Machines  
    
 
 
    
 We show here how we can use a simple regression case on the breast cancer data using support vector machine as algorithm for classification
@@ -335,7 +339,62 @@ svm.fit(X_train_scaled, y_train)
 
    
 









    
 
-
    Principal Component Analysis 
    
+
    More on Cancer Data, now with Logistic Regression  
    
+
    
+
+
+
    import matplotlib.pyplot as plt
+import numpy as np
+from sklearn.model_selection import  train_test_split 
+from sklearn.datasets import load_breast_cancer
+from sklearn.linear_model import LogisticRegression
+cancer = load_breast_cancer()
+
+fig, axes = plt.subplots(15,2,figsize=(10,20))
+male = cancer.data[cancer.target == 0]
+bene = cancer.data[cancer.target == 1]
+ax = axes.ravel()
+
+for i in range(30):
+    _, bins = np.histogram(cancer.data[:,i], bins =50)
+    ax[i].hist(male[:,i], bins = bins, alpha = 0.5)
+    ax[i].hist(bene[:,i], bins = bins, alpha = 0.5)
+    ax[i].set_title(cancer.feature_names[i])
+    ax[i].set_yticks(())
+ax[0].set_xlabel("Feature magnitude")
+ax[0].set_ylabel("Frequency")
+ax[0].legend(["Male", "Bene"], loc ="best")
+fig.tight_layout()
+plt.show()
+
+# Set up training data
+X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)
+# Perform Logistic Regression (replace with own code) 
+logreg = LogisticRegression()
+logreg.fit(X_train, y_train)
+print("Test set accuracy: {:.2f}".format(logreg.score(X_test,y_test)))
+
+# Scale data
+from sklearn.preprocessing import StandardScaler
+scaler = StandardScaler()
+scaler.fit(X_train)
+X_train_scaled = scaler.transform(X_train)
+X_test_scaled = scaler.transform(X_test)
+logreg.fit(X_train_scaled, y_train)
+#svm.fit(X_train_scaled, y_train)
+print("Test set accuracy scaled data: {:.2f}".format(logreg.score(X_test_scaled,y_test)))
+
+# Now add PCA
+from sklearn.decomposition import PCA
+pca = PCA(n_components = 2)
+pca.fit(X_train_scaled)
+
+X_pca = pca.transform(X_train_scaled)
+
    
+
    
+









    
+
+
    Principal Component Analysis 
    
 
    
 
 
    
@@ -371,7 +430,7 @@ X2D = X_centered
 
 
-
    PCA and scikit-learn 
    
+
    PCA and scikit-learn 
    
 
 
    
 Scikit-Learn’s PCA class implements PCA using SVD decomposition just like we did before. The
@@ -402,7 +461,7 @@ More material to come here.
 
    
 









    
 
-
    More on the PCA 
    
+
    More on the PCA 
    
 Instead of arbitrarily choosing the number of dimensions to reduce down to, it is generally preferable to
 choose the number of dimensions that add up to a sufficiently large portion of the variance (e.g., 95%).
 Unless, of course, you are reducing dimensionality for data visualization — in that case you will
@@ -430,7 +489,7 @@ X_reduced = pca
 
    
 









    
 
-
    Incremental PCA 
    
+
    Incremental PCA 
    
 One problem with the preceding implementation of PCA is that it requires the whole training set to fit in
 memory in order for the SVD algorithm to run. Fortunately, Incremental PCA (IPCA) algorithms have
 been developed: you can split the training set into mini-batches and feed an IPCA algorithm one minibatch
@@ -440,7 +499,7 @@ instances arrive).
 
    
 









    
 
-
    Randomized PCA 
    
+
    Randomized PCA 
    
 
 
    
 Scikit-Learn offers yet another option to perform PCA, called Randomized PCA. This is a stochastic
@@ -455,7 +514,7 @@ previous algorithms when \( d \) is much smaller than \( n \).
 
    
 









    
 
-
    Kernel PCA 
    
+
    Kernel PCA 
    
 
    
 
 
    
@@ -484,7 +543,7 @@ X_reduced = rbf_pcaLLE 
+
    LLE 
    
 
 
    
 Locally Linear Embedding (LLE) is another very powerful nonlinear dimensionality reduction
@@ -496,7 +555,7 @@ these local relationships are best preserved (more details shortly).
 
    
 









    
 
-
    Other techniques 
    
+
    Other techniques 
    
 
 
    
 There are many other dimensionality reduction techniques, several of which are available in Scikit-Learn.
diff --git a/doc/pub/DimRed/ipynb/DimRed.ipynb b/doc/pub/DimRed/ipynb/DimRed.ipynb
index 370f5479a..eec995584 100644
--- a/doc/pub/DimRed/ipynb/DimRed.ipynb
+++ b/doc/pub/DimRed/ipynb/DimRed.ipynb
@@ -158,7 +158,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "## Simple preprocessing examples, breast cancer data and classification\n",
+    "## Simple preprocessing examples, breast cancer data and classification, Support Vector Machines\n",
     "\n",
     "We show here how we can use a simple regression case on the breast cancer data using support vector machine as algorithm for classification"
    ]
@@ -212,6 +212,70 @@
     "print(\"Test set accuracy scaled data: {:.2f}\".format(svm.score(X_test_scaled,y_test)))"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## More on Cancer Data, now with Logistic Regression"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": [
+    "import matplotlib.pyplot as plt\n",
+    "import numpy as np\n",
+    "from sklearn.model_selection import  train_test_split \n",
+    "from sklearn.datasets import load_breast_cancer\n",
+    "from sklearn.linear_model import LogisticRegression\n",
+    "cancer = load_breast_cancer()\n",
+    "\n",
+    "fig, axes = plt.subplots(15,2,figsize=(10,20))\n",
+    "male = cancer.data[cancer.target == 0]\n",
+    "bene = cancer.data[cancer.target == 1]\n",
+    "ax = axes.ravel()\n",
+    "\n",
+    "for i in range(30):\n",
+    "    _, bins = np.histogram(cancer.data[:,i], bins =50)\n",
+    "    ax[i].hist(male[:,i], bins = bins, alpha = 0.5)\n",
+    "    ax[i].hist(bene[:,i], bins = bins, alpha = 0.5)\n",
+    "    ax[i].set_title(cancer.feature_names[i])\n",
+    "    ax[i].set_yticks(())\n",
+    "ax[0].set_xlabel(\"Feature magnitude\")\n",
+    "ax[0].set_ylabel(\"Frequency\")\n",
+    "ax[0].legend([\"Male\", \"Bene\"], loc =\"best\")\n",
+    "fig.tight_layout()\n",
+    "plt.show()\n",
+    "\n",
+    "# Set up training data\n",
+    "X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)\n",
+    "# Perform Logistic Regression (replace with own code) \n",
+    "logreg = LogisticRegression()\n",
+    "logreg.fit(X_train, y_train)\n",
+    "print(\"Test set accuracy: {:.2f}\".format(logreg.score(X_test,y_test)))\n",
+    "\n",
+    "# Scale data\n",
+    "from sklearn.preprocessing import StandardScaler\n",
+    "scaler = StandardScaler()\n",
+    "scaler.fit(X_train)\n",
+    "X_train_scaled = scaler.transform(X_train)\n",
+    "X_test_scaled = scaler.transform(X_test)\n",
+    "logreg.fit(X_train_scaled, y_train)\n",
+    "#svm.fit(X_train_scaled, y_train)\n",
+    "print(\"Test set accuracy scaled data: {:.2f}\".format(logreg.score(X_test_scaled,y_test)))\n",
+    "\n",
+    "# Now add PCA\n",
+    "from sklearn.decomposition import PCA\n",
+    "pca = PCA(n_components = 2)\n",
+    "pca.fit(X_train_scaled)\n",
+    "\n",
+    "X_pca = pca.transform(X_train_scaled)"
+   ]
+  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -226,7 +290,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 4,
    "metadata": {
     "collapsed": false
    },
@@ -253,7 +317,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
+   "execution_count": 5,
    "metadata": {
     "collapsed": false
    },
@@ -277,7 +341,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": 6,
    "metadata": {
     "collapsed": false
    },
@@ -299,7 +363,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
+   "execution_count": 7,
    "metadata": {
     "collapsed": false
    },
@@ -328,7 +392,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
+   "execution_count": 8,
    "metadata": {
     "collapsed": false
    },
@@ -351,7 +415,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 8,
+   "execution_count": 9,
    "metadata": {
     "collapsed": false
    },
@@ -397,7 +461,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 9,
+   "execution_count": 10,
    "metadata": {
     "collapsed": false
    },
diff --git a/doc/pub/DimRed/ipynb/ipynb-DimRed-src.tar.gz b/doc/pub/DimRed/ipynb/ipynb-DimRed-src.tar.gz
index 25c4809a5761d168d79ee562b22a722e51d0f7d9..9db24fa2541bd07464483ece4c3be8a7dba77f7d 100644
GIT binary patch
literal 191
zcmV;w06_mAiwFS!qoiE`1MSaC3WG2Z24L5oVoo558nZ6w!bPD-FAzzrX=`GW2=(^$
z0qv@#8!3f+o1bBZVWui}`M!wU-36OL2xVysbJC0~iQ=_hU`zlfgkT(FN+}?mB!mWl
zwma!%u#TDOl=du?73%HWG}bg9_N=eKGylY~l@`jw=vu8nX(wu}FObWw5Tso;0yN12dC&M}$g};8r^E}V{+5={0?sfnO005z@Su6km

literal 191
zcmV;w06_mAiwFRt2&7#A1MSaC3c@fD2H>uHia9|^nxw9UcHu&h;ssKY+Ne!xl7hXx
zeSoeMH${Yeo1bBZVWup$`MyZ}-AAiI2xXMQl$(sriPEK>Vaxy%#xO!Or34Trgpvr*
zdMCZ~&T~7S(wwDsQoo%W$LjjSp5+yI=ASq=%E9h1_{M0^#>>i>3*@RTrASma=n909
tZfb#*w_aHZTo1s7P+l3$uW@J1$@11E@vol=f*=UK_5cUX7@GhH004RbT%7;_

diff --git a/doc/pub/DimRed/pdf/DimRed-minted.pdf b/doc/pub/DimRed/pdf/DimRed-minted.pdf
index 6a752715d9c60220c27152b7dfb6deaeb406d200..0beed3110393fa6a6477dee9b4dce31088cf0c63 100644
GIT binary patch
delta 25554
zcmV($K;yrRpbYh)46s890yHz1!3zN?e~nq&a^$uVec!K`x8+nxG)eH9w~g)1rM%8o
zq;ekOREi=w8hAP6gyiVrug__8gEvu|9t!5yU%{I
z%iJsu;@$0G=ewSl;hVSfcfL|?5~sU3O58w2yW7+5-%ei*?*2LqJ?B%?4#%q7e_t!*
zJBLeC?0sj@ZNme)De6mEmDKal+yCeG@85k6g1Ib_6Qs(`vS2CarqJK?aQ#!2ce3_X
zJyn>xLAG43dln20-;k`5$c>|w=_*pzi=k{je{d_WYi=B>e3&NWUB!*+DKCclnLBs;Yly1`68g9g
zGN-Dnxz$YZnI>7C>ByN-?CyqHxL8XOZNizn=&xOfUl>tRfzK~|g^?zRV>L_uj9W!3BZhTErn==2j$yrVsyd0Y>&^^cdL)@7B;{03)p#U4=1
zce4nQOy}BN2Z@`eI7dl|q)+D>@`6M;XNn&mKP~p~oucY{-Q4pxgcMEeuh7jx=kp$a
zy1e{>&lO4o
z=dP>re&A0<-V}s9)GqTOr&cNo(dRl}sgqcBe)UvXg;DGy$HnVR#5UMxsM?0}xJ*DSJ=TUk9E_rBG^8j{-9khIDeax2)QAa+w7Xl{4kHgnb1@y__$1i?yYYlvVmYtx+fEP16L$}G0eP;6sxv?Rfe`=F8au*Q
z=Dhg*-kuBXA8;;le|FQ78y~sz2>IOlslC_z0Ms_Ue@@oih}^FoDw{s%wFqiY5K1wjXU%klt`u!*T_%x6>
z_Jf5IY9I2B8wP-A)~W4fJCSj|)OJ#-PpD2X0}yoJT}*Vie_pT(kdO+n=d=1dEQX)kOxKTuM<_LgmiF1QjQ1>^)dcVeWgGbbY2`Kei
zi{q}fAHu=Kt#_U
ze!iT(K7Z`rs+Nj!I$cQaA4&=%6GNN4XVhE8k((viV)axUlsz>C)ZLJ48DmmIUbfEI
z2qiyom1R=rS*|I-I%h&MWkSy%xUOc_2WWHIu^dyT`be
z0wGerzF{SL(wU?$`EC*tC`d&ANVPy-_{c%(@T@zT9#BP`@
z^h_AKe_2leO#aOlG5W;3731uwWTu>WDYhCVV*(|%EW|(_vfzcZj6CPBtuS2(;EDqi
zbXUAaiv(FBgC}xVU7aj~S{JOZIw}s`$PX7s3+-NAU)2X5p*-PQ7CO4INKtVEgr*>s
zN;z9-sSIV-l~6$`c>1sq!5yykG|OJFLg3q7f0uXK&>`xVZS!&13_swh8$~uN+wQ&+
z)M6w;ktO9An|Sw4^x{02fmXH-*U)ogPi*xM)LYK-=_JH!auV|}5}IU<^yLB?mPBQq
z{={ltrqb0YG7*t8#$*bS(5N>M2Y}?pp_pr<=9%(g;x_VZMe>ZDIOV9r7~RO6R&Y1n
zf5Kfa#w}$nbsKK2rpm{yBvY?w+OyOiT
z{yiFFu>HphOtieL$d3&<68CsgBQta(m7t=Agwh*k2~bezXbYL;+vEv%5I4%A<(gT{
zkaiL?nZyjxyeAMr65Ld^wGQiQ2oqXIe{+K6NjKBX3&Gv8^)=okQ$8%xDqMP`l^fin
zHc`)rK^3R!9>_Ns?*Fc{|L6&KisZVP(RD
zM$a{%e&yk1BT#Vna>Lkv`EK*kJX-`J#8ak5;G8}8%njyb{la51H5l(17`wree_wv!
zc3Dj@&}*?JrvG*^j1Mzv{ms0AaF5?Z#t>iQJ$SAgd9h8YlFNlWr~o@FDLJ>cI31Xg
zrIwK?VPPq=Fm&_tS-2=VV|8Tgoq-aKjTLVTbuezL?WC5c8zwSX4Uy#J>1+kg}B;U>Szd6XOaodrCyG>EXwdp$Ib8%I}SI$EIsPS7d-lD-Cmr
zF%JxZWkK>~s`SBFAk1!yvhQ!0jmNbpRQ>N_dHBoX?x!(8TH?u+JT-f_jzUKO7s2t&
z1K?hVOU*OSx-z+nCKCW6e|r>72!1+OBV#?!5pqCu&(KX++wdDBMqI{-jgijKcHMFh
z+rQ?G1Z5@>Z_Hj71GO|%xHeSAS03|Xgh6P4maZjeF#kd4ywCJTzR6eu-wmU9rO4Bn
zDy--58DEx(P=PVV*uYeo|8DG`;0h7g2@lYON}AE1t@LtFmAYx9fBx{EyqU4_JCDEg
zmb*G2n~QU7qQLRmIdS09!w{|bn>?xMrUWt$_A1UnVc2Xd)w!Fl$tJ9Vbebht^ifu?
z;wG?&Of9K0hCR3ZJTMcNKk0syvBh8f$a7N*Rr6tbZz-T3hu$Zi+dCl4$X
z{zEUF#*E{?o_30!DM;1sI!1V9!bXC{lpk+D{2%4Y+fxc??X3nyo%e^mqmOf9T|f7m1erDynp{AA@I0CujGejxD{j
zyYpYsf9eT|%5noZ8QB3$T&w`bk0BMaH4(A1vH4)_1ouZiQHu|qob4Pv82)Rjt!(Yw
zY`y;5YHDF?V){pQ6Bm01HCqdR2N$4}=>NoinBe}lnE{;vi~yhm0O)RP&hV$zUz763
z%=E|nA%TyVy`4S4)X>@q=wo3D{P=?Nax!!U0-POPfIeRTbo?iRW8wgqSQtBh4Ee`K
zf%~hvl&z^9fa_o84<`S1{m=4K{yovpP8zyxRt#~^R#{4orFl>dMB*njmB
zcd@pXH?#p#{->b-9c5@^VeRpsasL&f0sKRi^8ZM-a1yt02bw5aI2)V)OVxkbrJM~v
zhF-|l%o_NyB7d3G{_G>`kAeL-bS(aySpYgF4z~aDeaxb2&`e1GJH~Q#c{J)6p!)I(~{jnndwy?7PvDn!BozKMhLBr%9kO{y5{0ID?
zW%@T{{-|$i@elV0PqTl(kMib!H~SFA+{4}+X#0Zn0(y&{}Cc*
z=>85jM@4=jF7>3E36
z^vO1S9MRk@VxT}lK$Cw4d4Zj*B=;BJ?Fx?h7b*?f46m?V!%@j~KC0hCM=y)H4QoAy^~QQ1;!OC|aMdN#aQe7b`AO
z`+P2%ygSD61IR^vD=|&FyGt0(8$Rh{R@5qzYCy|>jc+8Y<}5C>X}b9FJT2FUXy?dU
z$#Yb~am4LsNb~l&VIb+P>(-#+2Nj*F
zFKkeX+2g-79;&|U&cOci0MoyCBfTWUSVuk&IRR89l6zl#HFV(=x!L}~D`{i79}6Cr
z75#>PKc+6GWhrTmY5h1S^-a;VxYCQbANOQivXhN!NRN(SB#4&uOnZ!NgT@=M$50q`Fi=_t|P8{j
zi^oETK2UN40sXtb?etYUDr-pk`Gib)5Xh3)6u4&Q@;ABsrgsqa3{|wK^KVW|zO|Kq
z*io@PcO6Funqe%9F8r$OaYJNKDjrI1}haNfgsU4?XoAMu)-R@}&7c)8A8
zrNpE{3|8d?lB{5aa!-?v6Pjk(te4N{SuJhVVXl=Dh|w+0K-mQxiEGj_%L36;q+uk%
z^7M#MU=QRS8Qu*CEc)SRsG1^lG@ogIxg7Jj@-@(M?Jf)N2CWVpwn>8arc%rWXt$CF
zkpz;((pbvqWQ(A5X*6YI$75GS=AkJ=*ycVn2*|DSBK>ARtry#+8b-}d^psa{tB2)`
z$$!oH4xL(`xB84NLqND{uYEzzS`$3GocgJ3DRVaFA!m`fj#nDX(&u{#XI}0X$^Tv0lB-SjiPenBwzb+hB3!
zB7q9J!zpB*2hU;T|Q-Rfh#H9PiO
z-u^43GSW9|cmaX4O1@qbD8%}JeP!W{G<}le+@VnINwtNVJHaW$rW^Q9CZ;exQv{_z``2M5YMEO+-C9tW
zJei8T{Jzz5Op2!K*M3pNdZv}E$_mCMN&FSq$aO3-ywLk9s8Tz}`-sX$PgCLvd`QZG
z_rN)F^x4Uc-tcl@*NJ6+LSfRn`@&e9kJAllb;mgdXjNaZBR=K-(P!s^ET~_VLLTE^
z06P}3acyN@7NtMEap15#Z4_|=V_7wHAv?q>xLH#DqkE;I(LKn?fF+%*D~1PQ-<%W~
zDeCY&FPF0_{nI9W%~m{*
zt94#W?~|Ht%hJ`zU1J{BOQHm%D(zQH;Wol$^5yOD@u|eE2;A|^7L%7_iN3{waBQKC
z?G<`QeXhHqejJP^8Ndb66x9qd>s_L(DkNz+TK767
zzJ5k~Ea!{^D||CTH?+L8~~e_#qyO?vbT^p7Kd0z?W^1)FEv_t0oKa|1Ep
zWT(q(tdGqa1+!|9#lgxg?EdjQsz-$SVds+%-zI$q!YnJmyUEa6Vq&kep
zmM2H-ANC82lVqZWVs~Lkr>tD*L24MI%^sZB8)#K>=G}t7#z{~x7kSV7i6_)gi>@)e
zI1S?zIwQ|)>ZbQGJjS@eI>7tYs{Nw{n^mSy)I$w_T)$(T`33};x{~;X_2WJ)UuXB~
z?|gQjzyfZ+PiDHm-%$RbXuDvFhLK9tB(}(Ax<%D#>H~!IlpqjAW%eIto1z^PKR_=R043#xaBVSo9#!T4~I?*t8&p)AB
z%l-87*JChnLp7i9m_S*fE4k`O3P~@tU#Bh%CPMH3^b*{?gNZdRMEet^9n6Sni#l`i
z&g8cZr*1bkglmo0Sb?qyf~El062P7kK~n>NF2y*gEXR8{(M4IsV%SnG;~WX{@>A&m
zHANJJ&Sx^m6PPs&(rIBtHBdJXg}Swlj%*>wvKfU-+;Y;O(OlM2?TcC&e{7Vxw8g_x
zCC=dVfH+iP?)rq!2S@#u%bL$PJ)T12io8|yzD0LVr~0gy-P7;}>6r9u6M5#2ZM;!`
z{?wS*i@#p;CB)({vo!dtm2Q>3A%Zj2u65?)E20P(GgBcc(_;QgR7R6WQ7Xpd%4o;4
z;ecps-F$z@R-ee^zk>`9f*t6cMtpunS=Y};7EE`o^e2M&(v_J!R4F5W&sHcTm!@y$
zfr4Z&*33xmhjf&20N7-XN2fkS#o6(H`pVpn_q%i~o`L2)D6h|6oBXBKFGg&HVX=(9
zG=>uov_)2)$=QMA3J%T@+w^g<$TjByUJ#yO`Q?Z4I3^{(#DjDfW}eH2+N)rH;cJ=G
z++Ho{(-j{f?}5*{-5nqvZCoh^;E$TvL~SKoyEmm4f*?9oFt>kernfcjjP`JI*E&J=
ztCYCS9S#kEx>Yrr0rNS++_wDDwa=OvV=#Xul1-<$(!6MG4VXeRy^rUz9|$_P(eyJ8
z;+i^0mvld+*u{LCoVGjLO988YZP%On&@4W*B~A6IfJ-lO6P;R4Ga9Wa>aSDuh$QXdBhKY(VMfVIrm?F
znH4snGys*deK8a_;F~8Yk~v(thxO4h6%SY;AhuwdyQ;tA+<-H;w~F8u!v^eo7Vhrb
ziR?2j&P+;7)}5H1f98lcPJ50zRNJJB2(LKC+dQ3!Z+DN5CC4@T$!R~A5RCa|D$oct
zt)i%l2C)hkwVMus2u?E7;NKJ#z-8WK0sHyrNN_UkD6+oV
zUAjAgrNaZx@txfTw;{gcn3&UXT-dmOT*%Za2pRLRbn{OdHW+Z(uNQVQPMh$W?
z&7jh}eGr?`FIyjfiktIXwpn#rh`s5cdJt?LIf`9liS!nyW;nA*G-WX9dWHPW`$kqOq~
zr!9g-nE8i|^)N2`8L6e#)Xj~p3EBhS1R`I2L9=0I=Hh+K=DA2TkGR?$c%on
zv29OsPs)4kPR^IpcE?oy)0xHb+n>gku=IVXO7dGY&p`tYh8X%?J1>|TESC_$EgVzE
z8HSbtwVOeIQvLcKGRa1?kHMGY_M(RhA>11V-6~(pcv19Emi
zbEd9;!M+eb9D_=$W4!D6uQ=5Wz)P^zWMmEa3cN)=pU{f~TWSDNL
zx^i|wmRCHB50))x-ptMB)~LzmK@44{wvy<7^1N8>sx5xSiOBOlDh$QZ-sF0*2?^D?
zkhlEWnWctfK{^Yf9yG$62=+wK01}s0_vFa&Yzezh;CX<|B!{vZSEWMm;zo=GHQMbo
z;%iFka^eL%f2jhsHP`X#kCG+*?~y*H#?F33C_gkc1A}OszF|HZs=KoAWrt)>X51>h~woz}JBV~}3?D>MIqzO8_?w3gOM+oK)VQ^j;ie)rQb4A@B+
z!s<-7TFR5#lCz%IWNO+)!2JRiA)pDKBevk{?$dR`W^_>UsGq>Vy;(82Lc6w1G{xq_XEw*E1!_qmzE^
zz3MRzdq2B_iK)_-#8UNShv}~F+4hJq3QMOyKv+J8hiNy?_Lg-~l`)Ay-27ld_?H#T!C;pL5j5O}~)+0}aF
zd5#qF_GX7$A~nY@*uR_F!6m0w^7Ib-%B6HI>W{!zSQ9KIge9ly+yx+id9#CMXs>J{
zdb#nPS_E4MQL}%NfJPq33W=<{;$lGCQjASbN@0a2V1=yP!8{ZmxXsLAi(i$c`ihOK
z;DiMgE7Q2-T1K0Rv6?28Tf<==s{n16jI>%ALH7Q7TWQ(zFc<8ES^|4gJt<7LB$33JTaW>%B`GY0zv6MhSc&Y2o6y!
zqD5rFUIDwk`w21e`@^6xB5O67+7~^rJTki#cP?W_0*tBQ!*MJRG=#o+@rkE2Z|h5w
z9(9>1MOsGTmk>mMYK^(&@9jKAZO?D*UszbS8Kzh4Cg!Mep6Y|+^!A%*TFO@u6IxZ2
zW^(JG2||Dz8u9mhz2C`?&&;aX7x{3?`fx4kdwDW()aV$ce2TRG1
zTi6voDLL*<{#Lk=F9!`4=fcXF)1Qj3oW$SJWpCqJ9B_Jn)P6=F(#ht0k!Iv)C}aSR
zBjNmD!+;e_|Eju&6TZ{dS|oKmbox^@%KZIEdP*HASUjd`Ks0e9f7Ba
zPz_aP=_8s$zT3-A=o$a>UNUXVlMmJGF>jrWx@?=jW%fed&xq1EulU
zpW(~>?k*31Iw#C@2As%y)BC{^K8(bo)ZCTG_LX~4T#d{cqh6!pYx%eJBMWCwCTExB+gzXVRY~-|lr;kQ3
zdB86p+7Pci2r-c2h+|gu*?qZmTt%MIF3ua4QF8!)6Uq{JIiq64a65f}D8QLUOq^{e
z1I>psEGUI~qlll-dHI@orOayjko>ux9`S3`mlX56p)@By(#@%Lta+de8%sTI4>0;y
zrt~8dFM0ah-kk~ao@c>h==UEzp+bSlqta;dZs2j*8`9`$jj)*_n{Kx=ww~J!`U?mQ
zpq3(kY{($2H4`Uh_4DWQR-TGa>lzK*WuPs%b3)P7eQk)o!|jAMuFoPgbBnE=+Qmn?#HBo#>Y
zAIDNBxTD4alfw?NksP+a;&ur3aNllBbAj)FoVdQR1lhxBNv|PgX}xpw-R$iX!ND4X
zOYw0JY)KzaOuM75;xCQh!9L|MuI&pm5$<%5lN(&K5LHgN8=1;|HyE<(|HSxyIzIM9
zS*orrvc0o7Hh->}~<%g5I=3%q>3dW!xdq<2C)xDXqFrLl?f|a!l
z@}byj@tv;Lg58eUJAZ~dF>CJNT+!3_|
zcX)L~Pm>!ha9yHKEwP8vO1Lqja26N=xM$cxO0B#*#fk&#o0G)1A~-+oG^a9u4~oGO
zP3T&!m-|f<%$ra!GBFO70vjy@ex>8uFB$mQvT9N{bxxCm!Z`!@@omJW%OAWqwZt7Vnt0m?XNg>ybGKL6Zv)m#ZTgp0dbwEaJgEL;#n%Cp;NP_E2G5>0ve)Sst^Gv562p*kY$}JzM-yXc~^YG@lL5bcbEUUX^|FNdXH;TmLN?ofvMNOEfca
zDVg+K4-1BO%%zHdvjw)uSS$^9-hD(KEo|~6%O=3{?5(kpzcvIX@itWedRo;HA&?V_
z4p0JHV>R__dVtW5o+O6dz^juM8A#A=Iv|*H+`R#(9_ezyW*O(
z^nxn)dp+st?k7s_MnTa>YY;xTc1p@jx6feZUdgOm(~3bHoNMQsg;q&Oz|XKyMM?J~
z5XgH;<&uwo+Io5Yyy~tvk9gdjqO~Ydn+PLM_(1^{1sa|i#M(`FtM!~9ES;7D2kBC=
znH4l%n?!YrsWFtG3Ij`c>2+OMGq99#qDexwe3vt=BrYT04?nFIZMqn}Q+jdVJy8Xx
zHy&5aU|FYK>VSI=_%?zTC9AMXtVc0R4~ENO`?X$wF8Pq45@v2F7T91fVN@BaRa<>C
z*C|;Q(V9-6IlS~wcYm&5_PhU8N5(4$!$~v0s1@n5Jjn)yG0!13<)nXiLQ|-G2;1bU
zV4w>wAb9EsJ<;$J6M~dJU5`@Pmwg+w@9MS?=l<~Mkp7PT99BsqtJ2u|r*SkyV5i#~
zxfY&(9r3RUo#X5^8(IXI?1H%8KaBm?mCR3KYtXpSFnu;P%`hEB`HJ5ad_
zuUW{yIdFeQBJWKghGBjN>E~gC0>hn+cDDBv*dPX16*dX`g|I!Fo-JZw$3M&^ODPoE
z>EeUKCjO;y;!J}#2yR5A*4P185&d4F%Qu97D6L!>C*DNCRuJ6yS;eP%$qdPBIw
zaa2tI2^ZiXj}Ve@?S1)sd+k-E_9x}@{)Es`@}1c=dJ%Tt_e#c=hoXy1rd}Em_7fkQiy?$$eQCC?r^Kgea8M`BW@|$^Z8*hi+5w*w#4T*ux;ag+Q}^3pYdyA-`!PNT8#X;n+es
z#!0n+MXLHGu1_QylH^d(veL~9$u>qUFqg<+DRl-A7r&5*-}A{XB_XV4C1
zw}k0~#N#c#7BVq~K;N?^tglvToNu`;&(xt)Ij15z=pg;B#b609L0#$<>a9_e&RZ_Z
z&H(BMs%U5isNPx&<2&K((yr=%V=3C5r7fqOE{AE6-cmTHF&i_{wnAp
z!zr9eW>){1`Tk8tq+qV1(-ha#g?8jhPB
zidE$Q@oam2IXyjMtsQ%SbyoA-Q2w$|@isI8HaBaK{;g;#y|ot2X|@l44&W|0zlt2h
zYi>e0xK1;>p(IhU_6=+|ziapAV$n{Peji&Vcgus0*mpIsd=)){677c`b$}?z*)sI+
zs6*WT{FM8&bK+I*%4k9=O(E`v82r|S9x?>Vu8#fMGapX>Av$81>a@T@&Azx#h^_M2sJV82)~83^|M9aRK8;XPi!_`5Mv
zeKYisa!R-*j_2X}=Vx{jd(slD$b(Czg0`Hh-62^XMPxy%Sd4veRlTDVwq#M`3X!76
zX9pUnZR#K}fv@>T&|6)?je+pOI9+|p-O=sx?E;`|0EEn7bq
zhym*B%`-z)QGCID@Vz+<;at`R1IS=P`a`(ocpSe(xmWI&nM<0kp=0QMnV@^s0#=vW
z>tBe(DDYz&M}mZZNsIG!!CXinz5o8*1f3RhR&aysnxqDLO;Z;nrfN2g?dFGn!(Nox
z5`b%ETgf|+A+AV88_Rf--OT2s!(TtYeIa&xWhCns^poCftg9f;Qv9T!B78vK1LSz8
z#2>$tbqXGfr0+|E2Lqy_QW`qvS0cVS2!pQMeoPG~By*B~Z;qqa%qC&0b#ql#fx$aJ
z%@;w~yqRRoQJMUs%qK=h5|Co$DcQR;8A~nukiyN!z+7oJXd5Y-8m=y?F~Eb0
z%vwJwzvqh~Nl%%|7GPpU#70}*9ZN|l>`}(UKFhes{~V$0z>MJYa7g69QEtEE=temy
zv9+<+M2W6{NwZXEsz9eGF4IT`%^B9oWtjO+sXii$yp@6MLChuz=URDY&lP&@y@m4v
z8mD3D1kDsP?(gQQUq}TG@1dPL9_@UN8zQ2x#O{P$85QJdl<2eeCPg
z;GLX*P6$dgb*{G4yr<03si1m=NhALT25_4l!RsGsT$VRrhrH&|UBRpHjj4`3dS
z8i3{sk0%*t<=TN+m?p5}uMy}m+qJA!_pR1DF>;jaWVtv+L2aA7((y^L!y5v8*Lx|f
zY7@0oj$_6?gRPk%!p~j7;-3h0VQ$D!oj~vgJ#}J0{4R~|
zX&Z$3?$N}MLy9I{CHQi)^cq;?172x=RIF9HH?O`^G9ZyE$M3w
zA^RAHvc!8VxRcfcI$NVO$9$DBHcWn)bg~S=6qXW^;VNzzZxuvOHZE0ayn;B&YqqD`
zf1kIdJowzDbTC7Z6pdNGBy{$@$ow7OOG%he513c{g5DUjqOhgew-~gVpm3{yvRgxI
z?+rQqQbT~0GA}_FrqBB{n_=8+Q;4E_f_<|FDsu6;#w2dU(*C_!!YOe-svnra>)eg_
zdr)vT+!9CNW;^-V(_&@Q^4?=v2hN7B23?*7G(3yy=gvsxv`^TBuO5Du>9<R)VlGRDo&CH9K&b9lu0&O}E_5J==~
zd=V|}ml8pDnfDO$(O7;n$>SWxU^j%5PM$k!4YelYa4xY@L`|O^?pGyil^{wCxQ-~x
z%P9VqGx1?1AD^YMcq24)CU*+QdgiRfjuc98Ck30mkK(!zJ+-{xEfCUw)&$vcL}j^N
z2*OIeFL(gkw>rdB?M;9|yr~>AIO7>RWq<1>DC-dYxu@NwJ2v}Olhg{s7&91+RJjbp
zqPll^--KqUGf+R%HFF;IT4W$`ZS5J323L){*8JS=ZG9}0_J|wD=BXP!9Fz+krYgIi
zKb!|SSZ2LX>Qqb)*Ef=X6HuiQ-rfF$UmzfsP#P7@@GjsD_w_j%S=!TnZrkV_LQ%td
zCu+k!3m0R~Qc1XWU-MWal1-jH+jEY!y;7sZb~>rtmdlGBy6FlwB4H4^<3&9#k}VOe
z{duII*9$YNy(QihY=c=D6|66+K6Z|`TQ10da~VXWM|YbCMmJ+
zo1W_EDxDPjMN4uT&nwt%3r8STs{iy&H_thep;5(eJanx@ytDD
zSdDC7yQGRk%rc+#o*N-Gtmi{19OT=UYBnsM)o4hih?cMI18YQgbKA-!3mif4g@WhE
zxqAxi@tBvAuRW-L&eW=z3!mS#;#+#uEDtlxmsm!HO9!QBjbatI!*oIBjg~eLf7ZyI
z*6L<0`X%<|RLC9>AE%bx?88b)fj();KYr_dE1)v{W^lI6kE)paF0y2SJnnb{D-o7G
zubXarTQtf!tZ8EWfa4d)Vqy@x*6Yk!s55z+xK278eRG(9YxRj}-E-?)Q5X4B#_FZB
zciy9YYHY&r!h>QTC=6o%Z1{U^o^cK5;KIpXMl@
zKwcTKcu&Ws_vXqBV3Vl9R+tjE!KG&PGONczy`-d1_n+9pDX4!3Gr2rT!+8lwiFb=-
zpVM4_bKID~=J+P5derjJGfjThBECYAG+Ua_*t@msE$Rl5!Nr39b#4NL*GI~cw@xvK
zDlQ!)!<1}HB|D{>_W$@-+t8w+d@aaJ4NLJLoKjV|tT`_Wv(;4n_ltlKX1V3WgS8H%
zUrU4p?|lQ8W4O2>I!~&_=o8WF#fsao=P&Pn&N9~TD9)Q0?y`K)0`D?yG-#~aTmVwh
zW4>dWe6CCD8_Y&0OuiJ%jwvz%`8{;w{%0bOFgM{dgVK55jNfpPyO^ZcKqeF!%P#Re
z5J(h-c8#{;(okRO)>;bkUELT$R%Bz1qM$SnRqJBpa}W
zRBWxf@3veY+H1XITVEI!^b{J&gZu%nl1FxXoS-fQ=jbbOso<5Zg0v_Kc@W|UPZocW
zk{GEes<^lTGd2mJtEn3wo{5V@Fs<`{kcHrGIL1k98dCHO){p-x@E*vUH>u9ja(=}q
zzAokuQs(T|kpSoy_zXHF=_kmh_a(E8>au
z+fS1~y?}jnRmp>#mYXz!apx^T8}8ho!e6M|1v@TH3j$U@=nMir?
z#6JR~QZr~^$8~vHWYlM*)1R+oOXSZ`PMN$Kiwkt`br`&{Zn@|xJSPQ6bRqY=FPe-}
zYS+R*2*EeYp1!#!>}8Q0zEVWZU73Xy?=DE#9BiR21+6)a@r(0VP|Ikwc)B9~F8Q>Y
z$3w-`%KXxUyZ*FTd|xz*4SoiH#R@7hY(e{J7V%S>tNgWcaQqomckXTnb~Yzu6LjuV
z4CjyIZ<~=*wWied+V&mFZTMd&Y+MN-jqK$Md>mzX4R$DWAf&AH7Sy+TWq5y(MOd1s9mj!k$)tl8PNjx8g@A
zI#G7c`>MY<9-p$3qC!51)|gG
z;+FB`PkhA=VmJQzg?rV8->#vRS`CH^**1o&
zWUZL!?BsM*iZ+T+^_d(2-m{fm(fQ*%`LwLUXzDm^!i^}=z!v=eMsaKl1+B$>zG&i<
z7*@2~D|eMEHkz`3IfQL7YW>bAtwfhqJx&t6cHU5yuuMeFZN88SLfD!VZLT
z7pGSh1RW7yirpPH71%D)H(B
z=W^td8?zq-ov#&decV)Y_1_D*ZKxTSQ_8(o)#0M$sT!ZkU9W~}^B(jj{9^le~(157TMEHO!$)1
z3(E1NBUL5`H$WQmsUAf`Y8wG_b0TF`2ozP8X$VBlx$eXZ`9eDxiW#|Q(=7Jr#MZF@
zp19nJ@4{iena1c6h)d(<9p!D@BG)gGe(9@wCt!RweVd4by$oD%+ZXCj&P1-Vlz5j3AC
z6Yq+Dep`**r3T?mWCN^_(Yzo(kC3}BhLol+B`m*bX^>Y_mPfvkh=qT{tmjlj_9Fgn
zIZ9sCG+|k2IMG&Xmg*5~*VfCyBuW%YLXE|Kc5evC5jfF#54d4cneG0^c_m3RXFtc-=TvVrjRy}q{M=L*gQ?b|E`B=Ke^LCU2g>0(V
zxaCJ$WcT)~VruU#p1g4^{oRpw~P^}4wtz#m6L5G?5b7}P+CD(%t8t&Q_6r&7uz|pTc6O^Z
zuFL$EmSk`~4I(?mDkcRt=F`cI=*5w-VkqRwV$i~TdL`kRgdRUPf@bdrQ(|e4UAaGcjd><^!0!^MSk03rxq|k_+(Sd))x^_
zsd7=+q|ckI4iwvm;j_HAKZFSjoZ0hfi-fb!qJd@t~(KD_IPV^v+`NK+cs)r-;|39
zB7L`*`=ZAWNg5z<@D$2_bFBmoo)hwIrXJe2{9X~2{941fi?z3foK;X9O!Vb(w_w2r
zcP9)QoZ#-R!QCAO2oPkj;O_1^xCghvU4jKCNO0KvzuJA++J}8ReXCEu^uw*LzUODE
zwZ}T5B3JbrgOytLGD_1cnETqyDD$har(w95c9t4FZxjX@WZ^N(tYOzSq_`
zb>PfRrAcJ?K}cuw<|pacQeDK#4Yz(dV-cAhy(s!`x0*u|>lkz@(Vda#sgS*0Z3sZi
zUvc|Pn>h<{TFf`i%*~<(>AERA%KLjo`<9O4q3M9sy_RtX1e%^m^M*M%6tA3>@>4z)
zn>ceJAL>kM!o6R!7oNnCYtz0FFRZ!C^hXoy*(9!BCnAoG;-@U4DilA(&QT|fdJbnR
zJP4edt8W5}agoSQcIyGL&0HnEsUFwc`tfmzsmU?)Z?D#DeiPU^oS%!B
z3o%rK;DNp^
z;p&w7kv_3jz8(|3KB&LU|5Z!bGk0-zKvmXfqX>?5Q;KnRdx+w5y&&CvVSxn*XYO2r
z_s2)~^AYc347KgbxJMg(*3PkYhBk*8T_u|(&fSVHu29z4=dXjYN)W@ieU8Kksct{j
z?-M*t9Q@d~95_6Oco+#=P-dN)W$IG#>Ol;6hy%hmT4GCC=cHhFSi=t@+-4k9gw;4h
zPgNf}gLUqKAxuLB61R869Yus7=C{GCbH|3Y^#?4*TPeMi|?=BUykh
zE%*FL1cJ$ju<%*sCgk)EYTM?UnWD8$MGdP=m|>3KKWoj-*`Jdj<4LTiCsIkQ(5z
z`RcRq=@%r^)j+KBo#Qg^%O8`ja=E>_K_|Z$r-rAgDa4d>0Y?bX9P=v>UKvCvvwNq8
zUNIsfM7!w=Pvq#_YrAV?h0~}ZLP=B`bkKRu&#kJgvd3xh`EK$cPN9q(GFQn7ILC~u
zqc8E2Tz+(~_Ny{(z8Tz`^7-^#T@5s7gqg~&>$=1Ex5J@qK$VYNhS-gOm@Nn4MUSxJ
z$P9os@W3!{FxK35TJAu;ZNi3%ck2v~yR6QjEgba>tN2uf@t30ULO!8L?AQ-e!&R#@5
z9Y1%qnDtcc;ow%=WHVNUc&_=Q+qNXvFQO{&@S)m9W|Vab)MR{?;x`?IV~hG(F5a!T
zo;~81*(qzdq8ty}
z@B@C^zacClZL=G+9Yl#HvRNdVY9DFp`f1sJx1~ioRJQ3B_Omh>izUWZzuf*@tM3cg
zQwy+WsYzDh2Dlr@8k31{T0M<$bqr^wl6>(ix
zS+{9+($^7#=8LBWi3L<1?J^aA1Fn$o#-(i|j~?m;iVOpRMk+m)z4x(Bm%0Jc@(m38
zaA}-igz~0dX=c^N=kcmD)h!mnePZC5r^$>xQTT7s1CBIRY+&#AfrF-mB6PUqfIO3N
z`;wS#1HH;F{{wK@RxSZ8pbbb)E1PY8mE_m|19lU;SCRsc3X%~u6U4qujM2HXa>Nhe
z={*Yr
zXB3yWH%K@Z1dWuW?E16_bp$_lSHfB(
zh#?1q=Zm8+eh5_jXN?J+16&CStx6^dIAj4C@tfV`HjFtYU-E}VGkOc_lHUX6X*Le#
zH7He=?4_D8DwNb&1NSS3f%yKggOjAyQEDGM;}B&(7ObtjDMj3-KAO_)S=ZPp316PMO*2z
zPZ~|z=ELA)vqgLr;&(?+k6MeBe(2pbUmdSR=14l%u}N1#MEN?}-m;K85G8Lz?UI#3
zEj(uZHqE>DLxgMBWeMe_dGebclIOIBslu903c1W!=4;8c=qF5Dt@|kDTpEk84Svx4
zeM*c*>~lHnuVogvd_R^wGlOyKKeh3+0|n3eQ3)%s_w3_d8k#937iKOP2oR(KN*6KRS%_snAc~Y@UU66jV321X;{8|T%`B3qGtaQ
zS)S3fEH)QdcRFXxEt_DiO`q$#ThH%&Hm(Z&LRX0Be>gnn#JSC(xye_!r%v}khH)rv
zuS+ddI+rv<{jtS)LE4Z_R{jVx(4H2z#%4|v1AR-Jkxwx#7x?T}q`l~(66Fp`;|a@S
zxc{J5KsVgfjI&ivI)UHGUYf}eYYckn%0jh^w#L8ziG;m6SWJmD^@miO;L5mwZ8W?@
zY#2n#^o|siG!@7&IYu;|nq>x3rjniAa)yw@wHDLU2jBjD62r)~DV-~5mXYoN6;(F=q&_77Wv+IC{)4dB`svCF;}#RHu59Z_$^Y
z0~?pmv_uX4pF>u#l~nc{{OYsVJl&a${1B`8nc7Bf{d0u*?y9a5rt(Qk;7Bob53C?Y
zk?UDV1yRA@Z+{Utw0GB`i1drF%zNi{=G+bEH#)rw*?W?lUc0=WNzi~E>T)D4HB7Wy
zbx1-zgDnwJnW7T8V{Tsu;U!wHi;%B68=b0wlmj@76q}k)V@g_O<0!orizxbH?p$^1
zqTa!fhx&~g1f$D9eORVhzm8??zm;X*fjn(0jZa#H7814;@Dw#xKTN#bi|p!Qoz=`r
zq^6Foub%&w$Kv3U45x#rLL&6;_61ISLg?lMA?qI%~nKKC*a5(#z)abiZCZBfa1K-j-!Q|`Lv~w^B_rT-EFvH(DWwj*U
zF_%w7&2WjTiXl0yWtU-M!6_GXjf@LBDtEXp6xZe&oAo?qO-2ODXY#ll%?UcHFA~B3
zfu~Yq%Y!H`Ii_)m1KSHXGJn2*gk^YJkxI2h5@=eP5(j|#7Oaqb!A}`eqs`xdz*&P;
zIr03UQYt?!f@Tj%HEorJ-z@$e%rzJ&66(K6z{AT~r6Of70*oOiHn0z3O3fMm=Mj7#
zP-`Yfl<3wGQ;V19SFMd{PA`D#3`pyt^SFx}sg%R3)poUN^eSf(_0(EpO_lBfjAz_1T&=qH_Qir1)pU
zX<@|f8IJ-{;D$8}tg5gy&y?UeKRaK%;OGh5c-ezA6u3GIaqgEW2x9T>^qF?r5Xd#$C<~j3qH%st?vgDayqnhl{39EOilXI6b%(?BF7wdLQZ&HOePVa-X>ta~
zS*QO=(Cch2k|puw-jIvjlTTRsVyw7U)3}T7lRa%v)4M{3+Vkhs0f%zz%QVKv%2<=L
zlhKPvH=NNTzk$5_y2U5}2)qzk=0
zM_e;kkDk{G%LlXOb&rJJa2ja1(!@~Gek;BFd|J3-a*vi7W7LP^l0R9GnZ*F>)c^yK
zlrWhq`C(hqnq+WgFt&8!{S0$)+0hSZhC3M|ULn@oCFJx%Vz@kTh(in~(
z39J|{ZkruQD?{=e24||pS2}Q@ohaY9ZHh5yxB~3lo+v&;R7DeGh;)`SYq1Nm8p@&3DR6SPBbmaK&O}+w(0fby|UuT(o`6oOQ
z`5R_bU5_0O0%Nhy)!_mk;^)m3P+XldG@)6qKBQ|eCu7r*&ad+;%1seZ=+-<3uGi`#
z7f-~)HfjSa>uList+@v+_Xx`qiXW;tt7u{(tE{0BAy09_pgGtEFy9&`OfA>sH54xG
zQooo#bi~l1Bth4lVe7&RNRQ%HFr{n9I{T3&x9b+1H-tYXSXkQ`eY%_hx?0*Sm^u$U
zZ+iA#ZT3;>TXWsUQLgYg>m_>wW}d@ZzoDQ_+x1dWF53#el6j+yM8u`ber_B|l8EMB
z!X2eulNH?A$yDP?#_igm>O!RLU{KyVApX7=nlozlnruQ$mineeX1DbO(qGlxfQs%MhdlT~lEtvgEFh9LA3m53KDF7M^&yH;V&jY-
z^Ew}zvrji@PG|J-KK4%R7%PrnI%rMFFYCa69XFhr;tMU!Xj6tHHAsDQOGLtsuAzv
zfCKo^C4lZ_9#D05OY*CI$`9!Rbdv(;=Pnt&+6fP865v^tGyv5}mL^%QJ`4RSa?G90
z7Ev4G0q3@^+zGb@3Jfe7p+UNfM>8ZtfG;WA#t)KXEC-sXD*0$gBt4L}H1@VA@6Y%p
z=G^jD3X+9_NOUV8`|xP!#cpq$0lkIi-&+7DGXY>EPCQc+Kl0BI48YEQjzmnmNMdMy
zRX-F;U{OEMuef-__vfCbT6Nr0K(}dA3y0vnQ@F11M2Q>fOf_=;LPK%#NXeIirr#vr^_no
z>WC-nVCZuFo$d6Ki02>q*1Dg{!B1Y^Y?Of!j3G3>c=#RX`z51zkk1!dD6_^5gJH92
zHHfOkLB}ACH)@EM$IsQjg?R^3(KeVLc9b}T2(zoxA
ze|eFZ+kB|^ibSmD+Q9-OO`It(yxgHeg;I~
zbI*jg%DczT`<*5nXIg);jF{{3-UmM{5pR0`j7$T$6Q(X(ID&RTRh23QOE_`1PZi1ovKH5-11Ee
z2yWpUM9MNp!H=SqF^OAXN@&sAn9(a{$keR-@L%p{mf_@zbSB)4hM_`rZ;C%WGbO3d
zuX~_%#OS%|r}t{b6pR~0k@zZYyeMp$H4BEmZ3?tI#gmJdykGtCEbTSaEPNB;9yol$
zIVU(N(h||3%Jlsu22qh}wx-C{sD29V;Vt%Y+o-aUgy!Epqjt(k>cZSF7_28WneQ#u
zrY6#$YHng8IxNqzOJ=DtnOlBeo^j{M9mGKP61_b08(lMJS+S$wxrW)S7|}}O;I!j7
z=ehmumayu8&f@p1khJ)-REpmdYX?xO-nC|qhDhzh3^!HLSnJ}Jwa=c4x{aau
zxf}shNh*^{)l8Ph8h;wy;VbjHH&T!e8NR2$gi1s$XqgLE+X8i>OH0JpScmipQ+qwDFLI&bHmi&EAM2ft;bsq
z_UO1FnB;xQPH9(>Cr_dVWq6};V(k1oB8Oz&N&>seZ>w9)VM`a1=6V>p4oPtbn|~Ew
zU5M^>y;^HcpC!B<&Tnqt@j@^~1bx&G9_mq=ip0>vijwNJIg4Cq8N6N!k_Gf_+q?@P
zso=6?uBhxRqcV9Jj!|i+z(BhbRc$}LI+c>%*W4u}#ehli0iJ`I07FaztS5CM*D|XP
z_#|iy9xV4iNEKd_-!0llSf2lpko14tQB+vIe;OaMgb!OC1}pGiJv9SXL-0SQ6&~bu
z9(G*zKPx_DgA9(t`adr|gfa~7i`9Q#Jc!Q;T-HA#)_-1xzi>9FX#gyEjI^wGxIiQ>
zer{ea?i6x(!2glr@D;>dXdKeEZtfnGyxg4s2bo35&CSK1ruG+(5!9`(=T6Fp8+26H
z_qEx89*IviMLxws@9$dbxonR`8ikD=E0;ndhJf4I%hM|fi~zD!lwhM4ZX_&8>_r_m
za1>f?n#$A4>NtHK{5@j`BY+)+hXe_xl**lFe=_@
zgi_j@a5%0Tl7-C2U{FaaVs&Z+A~Nnw7);1;ka}kB@Y0-LDIK1?Y$?^ZPpP)wW`bQ&
zinuDsM7qsLPHR3)ns&o;bqx59hTRzx2%oWZ!Q-J|_frPIvf&33SpjXoOA!i+!@+yv
zv&Eh0TQ!X1N9eJlm;MgX%z)#!Erp|PoCOB2z_pnIaQezd7(n|AvqJ=m$jHUC`>v@1
zQ{etCn#tr-SP=u$Ze#@-n0N&AP&sZ=hknZFB%A^Nz6v;WvE>j{C4z(rqQjKnYlQkM
z;k_$VcvwE?-*^fDswh{Lh$K^S*#QJ6zzjs3uYwu6%z!cnX
z9#3B544Rz
z9u672_5iMAy_x2rUt7cMhQ7=VUJEQGu6#)qSbh!^0zD#ZT5R{b)p7>5z3jUQy){Ou
zOUc_Cne;J0C=oRS5g#Cw;S6{1VjH2Fg7eE(EP5p&+AcZtOz^pT$1gs-M5!oIZHH6&
zmE(^HA-EKIFXg%9I^Qb4z(B>~Df$q=lU`za5*GerzdhQW$5h-z1$9m&Rr%E~H2lT_
zeac+gL6=2Ihdop>#Ueuo&WH26u6hkK39j$Ke;^0((KX3Eb8^L}eo;Vgcf9uQywNH$
zsj!}FKNkJHe!`G$f|`WD%hn<(&7X52M
z9vr#%d1_m?V=Z3`=1&~wqv7`HUqP_{-r14nv6EW2Y$?T&a4eb>cSHdn!;=}%7R-sT
zGo`z310@OAe%lhJ^@@)vPtHhERL!X?i8Alf}h3M#u-6-tc>`7
z`?GN}uv)CIn%%3B(0{?MiW!xQh&yqW#pb)7Hc0FFAp6Mu!evr(m1KsmakE_c~l9@a(K$^5lWn3&kRIWo2J=S(73)Z#Jrw0@P1SVh<|}G;qQyF5$#?
zFs9iwm*Z>@`uAxwW|2bYBz-|iOj^JiC=yM%)))cn6jC3_C#6F&f&HG3&1lob*X+LA
zD}^frHLz8K@=>J{!wDv|nS`2&T!5a&MV(JilHcgR-`
z29gv%#SN@+s-n3|Qk|+ZhOi^2Q${zh%{kIX`0XbR|5}?gT3xG-5{0TI8jLRfjflfC
zH)6^3w=H1G#2MlOxJxzVO*IzU5GyRXSZJj4dKSuB4P&N0(0v11CBaqXt+jE7x0Si5
zWx`S4;mO|H;^i#}YBGU37Jo&k&@0fy1+0iurz3tcCF4UNJz_n~cQ=btDArIO>SFz6
zCs!sm`NwUHp~n8g#$3I>V^SFUQyV{mb%U%i(HAdbLuD8YWJ~BH^5%i#`Ade#dm(-2
z5Ro=a89Y4`M5nEkamX_cG*a!;@I4~omJ3;kX9K4haK`x(MI(nsRK0UiG`8mxe^yh^+<*BXDeB1j1tXGG%E3tdv0zEyUY!?=
zKarS_ks0QBGH3=mHpGWUPdIXuyDkv2iy})BGuIhF=eoR#A-oN-sTD`K@(V?vBsN6>
zmP59Kr-`L6KQ&-ZSDfR@dg4ije^HB^B1s9CX?FwN^LX9Cj(W(*%(#X%M-jRF1P9DilPwaLim$4{U9Pd2@8WHa
ziDU6lA~?d8sK_&5jymncUVMH&Fz2V3syr52e6K_Lm@W`>eLZ$5^B&~u1MDVJ{PM@f
zSx^MA-^T?ML8?}`5tCLRvu5yDc5c~!-}hFSon5LkhRP^9LEjZZGMSrF)O7PAOFFkf
z^P%sz=%T3*$%XRtFdd8EOB*xC);ClukwYndpT4~S&a(BN907TvyE6}<2luxh6xTX0
z>DUiCBKu`s`_?bndGFadX?P;PKv(bAT)jJZA_v?cks^b&wioS^t!&*J0h6m}lk3YD
zb&#U&*1iNYib8bz_%isjRevt%0
z?wo-It?%al8*S@{#*H^k{4ovU36O{D)>S*!RpAk?fzR#0>XO+vMHMjs&T14!ZP}Y9
zli4ryQBuLA+1l1R6+Mo4BQw-(wE(@BS=bx5LYb40q3dI@elpHBw$@HXn@K7v*m=FS
z`*HR>^OgCxu$#@1qA1POg*hTGUdMwGv6bhb>uxPsAp^vpRa&}T2bVY`6c1w&SBxCW
z$T9B5Z@vOV8r8Ggkt4mADCA%zj-#yWSqmW3u7pN2Dehc#c9PqzKxXds>VsQI`tyyO
zev0R?K3Fa+TIcAdaevW7BW&+rzd4&DV%}FzhW@=XTgAZ21H!d8R&`Gy5@eVOHFgF`
z(O~E$27RB0Mjviz)2eKlN=e}OK|DIOhPV9v-2$%6D%LR@%DSmHY28{&^5%)TKTda_KXL?Z8`%AJo}FE(F2KUJ5Vo$1>K}>hKH|EF({^nY3}y`0-)$ln
zViHuRR({*|^aND}1h`$DO=W4d;5mU{6gW1;NjsQjLC*COu57x`v9z9Ggcc+{MgCK3
z{G2&agYpgL_+C-LwdV(ak&-x&aX?PoM?On;;uV@FsD8etbdT-B2j!YhfTkZezX`|4
zUgNKiX?u1+QixW!R_(g5@AKLGL~AAA5C*`Ye>SIYei?I@3b;weJqQipX9s2?&U9qbgyH`_9}{>Pu2YK4uxXCu?QQt72(0q2eOyl1d9?lcG0)O
z#cl;h_2vV>hcVFsg2T5H;z2mLCBMHzZ9RMP)2ykF)dl<}4#xR=ltGnB_YHBHjVvj*
zS;g1bl#4u;t}>jx^ywS4=>ahPylH5T?On#)MNZ>%hHPm*Fzvi~UnG&nQ$aXIMXqD?g*Swj5(+0{o^RG|mcfgnxLb
z*8zT1ZG&6qv+Mom;)M
zd{X?+An!jmv80p~&lgTfNl7jd%K!fsqqI&eczzJ?H#Pi--ParwoPqdAfkes(
z%Hc?Xi1d*b*6tX!O3l>az&042e9e=EW*_gH>U(_u%U|gkuSiWpocjeZrLr8-sF{|8
z&)i}N>-C5u_1H4?=sY&T2tWzEvp(2tF$P-8vBo;*5J{VG{ATgbL-ojjS!|<5kE>On
zha{3;J{G8Z7FmEO;Q&p93q>*^9Fr)-lPFw?=o@u)Kfhuht)e(;sTq=$eJ!=N_bYw1he{tUf*_G@?Opuo9WwSE
z+}a)e#Ld@tJF-=ZbW6a8|NA~j|AzqW?qTNU;p1j$h4zV)6Uf7bMn@;5B8~RH0Bj)T
AqyPW_

delta 25272
zcmV(+K;6Ifp$v?m46s8912Z-?m$456D1VJv$#Uel5xw_U)J?1ii9-TheQS@Wy@Y$*
z5v{Nf9)|-WQ7TNc$TGpwGGFgZWP+Qh>ba-@5I`n!dHDcEyT@I$`}2p0fB$&@;fGIg
zzAM8r6WQ+mu#3YeD&bqSi+8b7|3Nfdl+yWyz&{ePX5
zad5b_)jkde)3r2U+N!zKy2f4vz5Rdg|MtTtA?C7JjwobUmSQRAYh}Kf;qLER^}P07
zJ!MjaqFgRFmIQ-_S0u|k4YPD5`Lk(1tCz2pYT$6!B`2u}qby&pe?FROFN)xZi+532
z7HfMZX($SstXq5OH7XlrpsX6GrGJ`+8V9Y0X@crCHS~+BhUS$z5Bob1rw0`JyccDl
zwKun%DL&Ig3eX_Q7MesGXPqwrDoh
z3XzQG+TDpfEDAVBjfrGV=LY17Tn1;1pBR2R*dq=qZLMh^=^KO;PPDhsEtBBW9{%*I
z{kDgRZh119#4SMYyWw6xLVtwbQMm8!V#wx%giiPPq*C6m??(hW8U|BYgn3SG{@1=p(GpVR
zfjxGZXwTFG6y)*6@H2YReaKK_8oXGZs_27dOW+V0(h`QcMSgO}tAB^%KK43zgJ9bm
z1XnqOYz1v9vapCbXfI@Fj3j$GpB;pB{Xip6PV;K$dWy#rN404=a23a-Y971Z49637
z4}Jl9z=Eo8?0q#513kZMRg9)5kLT(pP
z2yUcP&m;i;u>?FhN`HXgfyPG(uJ=onMqhsX2tP3n?4_
z;OxPuyyc*2Yl;$zW(G&YvFES=)cqCW&#v*(sz55mFwA5u3KfbB_`
zcpnt#hI4a>Y-E}yVh`h#OkJq*h57;y%4wD-qGG)dvAtK0HCOE(FqNy(5;
z$zbUNm95gd0UA9BW%ww85n`;NZ=puRoR)_53rws09^+b!!QOIo;)gvUC5WW~)@qO8
zy&YCiNb-!$QoYGw>uJ{v2{;m>A}k_4JR~&AO99I4<$pxwJM}T|T25g|Rn%kvBp#@5
zcxuJ+0Q+m+sqZdry;zVlt$
z7{WODDxYb`e#*Q``Itqy$LkE>2TgcEHPa~gOUG0a1W+je8}zrd#;^ol!hq-rHGnZxs8Y>mOFUP=b0S07}8`0TVuIhAx65sD7v%0Tk`N31O58
zGvq2)@hUS64z|@lpukg}PbWdbp-{}z$Tadb+<(^#a9A>kIRGHtc$-VNqsW{^&Zvy5
zM$B)nV2(`6Fp_Mpjl$<@h>P1a@?|M$YEB4
zsvB<|($%phXdT@NmM7gzGq|M2Vtq|^xuk|gRwYY^bbLd4HZJH1FtFt4{s~>P;;<*Q
z{H4M#M^ArKylWdpIXBKt0xpoi(H|Q9dw*_U1aUxGiS$6jQVnR{dDz&96OlvRO61#j
zksJ3cAXrEwU5S8mTI#u*#mT#c#(1h2`9F+(g(d&|6SZqSp@21POPK!a#j!q^(U@=U
zje&af-Y`b^8tz4wku4~rYN{6eU;)snx#HaW+H@dNR(Mh-nTQoc#GtFrXXc~O8Gnl-
zLGKKcaBM7iTc88uzF1z#QIRHLDaFKL9VR&nPinTz`a2^11Kt17d@H6uyG*p*z;^@+
zWgIsHjVOFD7`hYOnJNJ7SDvTH=$5)v?{OI(0e_eE{#|z@P`Uhlmd#VRV`1g|1Nq|P
zfvZ;GaAg=qPmC+r>;*Ex)`A~Tj(;KWSjWF}z5!;trQum2F1VJUS1Y{xPYb!0M9KOD
z5^-B>5Lmg%1>J}7uA5ACcs7|XA&A*_%>UHo|3jmFt@d7<1JHD7sX7t}yx2~H-ixWY
zbgzcYf!wtsCWwLGBQVC_{B{kEP4{uC$bcdpH{3$6hd8Nq98B1&w&>7zribMis%wPjoXa2jf
ze;y{e@DrXu6D(>@f3nisJtY*HRQ~QgX)|Zz?>w*KYw7BMWG+gwi34YA=M*v&5mU0i
z@@toAYXHZ=U&tvc9G~sDK=;#C*~OJ8rdbR}H@tckIKqQ6wz$wZ{$bqFBf(6ZdNFpC
zvBhBGJa*6Q>&TMw`_ScXjA`(>>-G3P4B+k;_?Hh@$P@@4A)Ot^|Ly>BgNuaR-4!tu
z;h20T7E^w@|L}hx_^<;CWo~41baG{3Z3<;>WN%_>3Ntvf1QQ+$0XVZG9;62WIJ3wl
zV`e8XATS_rVrmLJJPI#NWo~D5XfYr%GBY&_FHB`_XLM*XATl*GGdY*Q9Rd~uHZwGr
z9fblYlaq`w1M?ewv&W200e{zETx?A~#_4Ph{8xcLO9GI!Fb3K>0skb4+x^RF^D)Yg
zqz}9E{}THcg!7-8*8d6zI01qG(Z<}+>91T_MMYVFjiH6DGtk!1*7zgP+0fa=31IM-
z?c)tJq4*a;AV9>$(eY0YxqrJH|F_M*sSDeET$!%5mye;_f3KLKt$&M?=RaoipOy=g*EUGd#68ZW?}o6O-@K&N?c4ul}`3Uc(!zMb|2HRrFV9B{ww-V
zJt0w9ZU84EJAjFc6~Oo*QZZW-5jz{357tg_f5s?KxY6W0O$Y!x*MA_{Au-4)njH3~el|J^nN9zd|&Cf2dOaAITO@;uh{e6GaPWWAlGa^gXH7vGP;|6vPY
zkd>5DQPiaQFMqlG6(nYBY-eI&YX)FuV+R;IIvRSwF@9)(nT-wL#q=R%6QKKFA^|Ya
z+uAvQqyX$)oP7YMc8+j=7LuI}z##O;^e@B#U=aBmaRL}b|3+K@2C@G|9E<=4@xKuZ
zfI;eS^ifmhZ^Zt=|8MlsLGEw#QB&c65$8uu#lI0VfPX>xZ}ibY<$n>^M-{`r(FaeX
zztKlcyI_~w}qAUkHyC3?|df4kH#kdfJ^`e;6LC85!1gR^GAJCi+{L3
zn3??pev~)=yV=Jm%suSQfwupM_^|ym|KnP${sBJ-TmJ)obhG&zGJP=pM=7Qcrgr~;
zA58840e?T3Is5~DFmwC|{Gj3V5BRZ6&i{ZPtX%#9KUlf`1Aee_`^QK=Sh@cLez5ZR
z2mD~=`4{|8c`N=o*Z(@U82{G$|J?BY!Ya;=c2+
z`^W3wzrOrO1G2x5%YS4G3){JS(XoE0l8%{^3xB}G_Mym+Npbl6N3QX|?md5J_PO6x=F;NoSHuRPA#YYko8p%<*8
zhTLYUR1d<*PU2Y3Fj{^+&ezKacO7w^bDI+EG||~bD`$xgsx=6&?Npe-ke1j%QTl9$
zCisP1B&4N~zH4j;3g=+N@o@U4H-Ec(FBKf~KwV348`pipwg!I~Vk&H`@AF>Yg7SAf
zS4L!0V>CddgwT?U$$hs$2V-i5A@-83aMkc23|Gxb80ebxMT3yJUOW~$^nsEa2wh>p&nfxx{D{}wwBkkv!^?H%DkUZrVz4SFkYoiT
zlzWuuT%QHCtC!i
zOQR_xJ080tG7n7|!Z!DrK|pSm7wI?qX}#Dk)i7#yqNluqTRkjiO#W-mcj(mmywzuH
z83Mvpd+iHy&T6K}Bj)N^tXi*Rh{5pg7sh*M^v4R&58xq#j`iww#!9X*!W5qe+Xjm(C*Si=V7O`^>SnmI
zc!?ZmdT#QHR>Ao!5v|ziZxRc6dGR?|bCg>3nwU8T3c@(?u(vC=u3IZvA*uum8$xTu
z?QxM?ljTvS(*@wnyMKEmso}@o7Y(_)`t0~1{pxRg=~f@>t=X~P^7daLm65(#!wU$U
zRr2+kKq1x#>?;drr0J6!=MIHxPpU1{+zC!0Hr>E?LNQ2~#-EmbhV0iOIPGp!R_c7v
zu=|wR-?dA`Ma_VyACePMU{n{zZN{ByOu2cnlt<$~;>NK?#(%M2ub_t|vJE1B;C&sP
zE>`N~;?-^44+#OzL^fR%@hNo}R#&}lo+2m(+P@AfQOn%o>DGd}!xUPEBPG-WXi
zL{xxz!a}DTJpV^g=yTl-_2XbX$p9{hrl@9!S?>~MRUt{s(Yn_uA3^BS3hHW&>k$MY`xvExe=D&N5#!~Nm4;3$TIKD
zn}2VI;?|f_nq>A9ZV9k81!hZ_+|yDx^W9U+T(g`v>hiv=Ce>j)wmdmv|FB4AVQ#ZYj
z;W5Sy)&btHR_z}x*sL;rq8@7C`W@@cH-8|=)Rn|9tRMGb`8vB-f9JFN1Qu}neKOPi
z{f6=fMcV~aG>lZDCb2~}(=Do=yJ{YTcru>Sl^FRwE<8>h`if>5eU&kuWchZIX_Vck
z!^(pCDI~qnex157m|
zV+Fb<2$}*|O8|RH1WgUN6yu<>9DncKL>FZhi(yN(jB_N&%TJ{P)D%$=I-kiLPhi$C
zNT-Dn)j-`m6zbMGIyhuY}3cZBG;S;ctLoA<(D7E3dVk?2$q
zB64n7v*Em%pk5WeO2T#p|DZE?V){Ik5f9Q`n0YQ6YOjKYuVqeidw;c{Pgi_|yazt(
zc6Weyv~i^vfIn(t6Sb9W?cS7H2!iNT!QB3>ncmj8Gup$=UF!tduTtVRcQ`Zv>Q>cg
z2F&LObKCMq*FI}zjKTbsNH(3~O7o(%HDC(O^gf=;ejw=FM$^wYh->N~UDExOVi)sm
za@y`}F9oc&U2{sj%YW@@fN)M*HEu9o`0A@{;~pCgXo2KxyC>Ab2{Opux34E^8$Ux|
zkL&JV1OyXDZPA7c0n??4$v4k(DdVgsCNk}l
zXrgyfUW0VN=M3VKDV-1MHZ1Vh$`$9qzFt2l<`F+cZ_c9T+<$-hWmedP(g0M-_Qg=#
zfN!3pNak?m9@a<4R6JmXfY^d*?yCNda|6!Y-YSAu3>&cTS-87zC$i7DI5R0RS$AT3
z{+T1*IPE#=P;HYgBD~@lZ}W5_zTG`KmK@jUC#U^fLNpc_PF`y3Oh+C)+F>8A4>BiP
zR}%HSq@QjEV1KAEH6ALsv}0$a#C%4%q8mevy?T9TabX_%kBIK~4izzXi^jVg$&eJb#ldsX!ypw2GoG8pJAG)NVQi
zA~?xRgMU+00GD}_1?=aele@EMPDCpdMBz1I3q9GlA;HPCqsaPdcj@kA8oM0iAeGao
zw>HI#jt3yPFARgVdv8_VXHyi9ezr0%Y`%r)n343sARP{D8a2qpG=oa>_Cah$zifRd
zZq9SrW`EUbA>!e?^A
zU#oyZFQR1x>EeK;pDY#l8uI$Pdpse6$J1z2|6-T_tI%Lf@#!+wk5zJ50g4I=Yu}D+
zRjiODJ@SI#qux}^wXPq`EO;P0mTD7Zmo~oAv5~P#rrJO>Rp7-Hym?Yv-Wuv|g}w{T1uXBb)r)NTez_3L}c
zB!3&xJ_cWo+lwA5gm7;dbgO(V<4Gk*I|Ev{JQ6#LIjJ4OdC!lU%5|l&BP)kOY
zAE0u;D60Q%ia$E+>Ky`+iMYBLtj+UipnsdHICn64#$fLRa_NGQ_iOXS4rIjIs6~0o
z@0Tsqk8w8?u;up2Fc=>e0ZjO-0Yb!x%?!(3?OrYK3e^P?#fP~!AL)k4D)|2Kr>@>8
zP`XhzvrNc(6&6&Ce`);sUQGZtN
zB?Nlz@}HBHM#1WQ{wwPT2X>$4v^aN$pjO0#_ce*4ovt&@
z6SHcgWrZUbtb7&eJKzQuO>GL=Q5(jf8`&uD-f2YZflb+TlVQ50>dM&#SzhrhK3KM(
zc{4YgTcakM2QhS++Df9!^J2BDwtx5)CnC@Ls4x^qdz0(MCL~nnLf-OgXO{Lt|J+?Lwjkbe^Yw3;HI
z7J#=zcUsqik3o9nugv@d`nCen(pp+4ZI5k)DHSbEGHwivP5^1|Kg(RE@eVdS>qfI+38
zNliF|KA)9Wt=xxxT9ldG9*vc)_S^V(;}gAKjNnt#8nTSx`dU~RSN
z^?x>=`r3z{0gBogwyP_ze8G8pjK%|M(grr&lFFV#UC)#xk52lv_o~M@?EUNxCZ@KO^?S
z`@{ynMk?_IWhPImf`3v#C6rj${sCPe?;B+X5uVTQNTntRkdK2$n3HrvEBAToBel-q_5ogqJr8Lf`>IWmoHw=Q&cy+nXJ3iPRjs
zVE=Av2bY{$$s6PTMJ&`f)f^0tW4vQYZ+}O
z#%h{WZViWhtOB%MGSX^g1ljxRZKY+;!(6ZvY6=`Tp!n9K1c2l$ec6~U2m`r#);A%|y
zr)Sifoy1S%y0?i^3@8i>Po0oP2A!Hcnn=eHbD$c+dIj&mN)N9V<`f~5f8
zn@;!<9s4uf(0QjEB4Yx0k2Ox6@Yvx^uc$PdpyZl|Q-2Yi4PV_Csn(q@J*(SW|&^Fo0y}@d8!YN)7x*NX(?YtOlVb6n#rw$CI|s?XvE+1
z^?oNmJ~OLkU*y9nkH2>);A=d^-S)nyrm%WW^alq+O`x${qCx4ssMVgVHp^yPMlAyDsn|2uGRk)?M
zsr)J?;5*YpH&?b+>YdF_Tjs*x#orG-viw6*fta{+=Bru@bOfFvLN!#GrH^P1`ED;e
zp=bQhd&#sdPd-$$lb`$1Z1Lb5y_wQBm}bNmou8k^_N6ld4V1=Xe}*sjySqH-oG{ZF
zaDO82P45Rc2yM0?d3*V7?GKVsQ6gjvud%Yg=VRA;$&=>bMSUs1*`sr4!rqa_Z*YmI
zwq-q!3002W6z^da8YZZr$t?3^X6ku%Hy`jUs+R
z=jChWl`^a8L-OZ(dc?0$UsBBPhSHq;NH?d}vF3p?Y%KM-J;3N+nbMC;yyWR~dv_+x
zd!7Z4q2GV>gbD>Fk4mG-yMf1LZ%Cu3HNs|!Y`Wdf*m`a^=r15JfLe;MA%n2iOn;o1
z;rj(5a2hl?Q>I|FD+bSIk3pFHOd7+hH!0;FX`ED8LT~;IuX2Nr%yf3QJHmdIrGy5S
zYE>WH`#QRsKPk&FQ2SkJM2ea+Gmagya{_jkWdc~IUa|<*kyIene;iAl;EoyxOb$E5
zMsnExirXRB!+pCk%>}-5;`+uCWPcB*CB24}rS;CyceA%o1P5yjF2%<^uqAywG3}1J
zioZ012m6%6xVA6MM7YyIPHu3`LR2~7Ze%L=-C)SB{}bc;>G;?aWvSAtGU|`{uSdl(
z2UJ%b6QcP>%du$w19p)}r<;VCin&N8S0#RN43zAcEgWWiPmdeZC)H0>Lx265`k>V*mRQG1a!gx9p2v*iI$cJL9#ea^iEaqoB
zc4vF~WPMdJ3NWyc`6bk?UemS-*Jc=b?_zMOqt8Mqb+Ts{0LCc*b}#seHK3kqdoNJf
zR&mO28b;wrB(&zkq>DwbD%Y{;Q@Ypx^2YUrw^t#k6a!6d9rc4YEiEX6-&e~eT2#`+fJsjD0yY-};X=E_~
zH1uWjo0qcbdd_b$JAanGoH=^7Y^ZDuU_cM~u6sP6Ej9S!
zxv6KRtd^KlB!yfz$`~Su&2o!uY$@wZt|^JOj1zx&R1|*Po1HH<_Q3mNdchvY0IiXy
zmtM+|4mWsQKGeaInBCq*V7bU0s_k+WiVJkejL_{`+W6YAUwE&Xv^Pu{Z
z6<-HXf`8Mt8$5g3$X=&kwDuGIN(@8Fv8fy;ACZh4Jl6w)lFoxr+p0q8yrGvv=i#2;
z01AduVH^|r>&+5|P5xmIh8m#vuq3l_Dv-&;Hn`|+8QuQ74yfdYGP*jxXIpx3c=P@?
zIYTx3CKN$N?tc@Ee&&Imlpz%|V^=xJ3)gg{OxI$%MtrgQactAoi-
zi|PGWX}6Y>&~rtVg|PW<(1uZN=C=!a-YaEdCck{U_t(L-rnO{2z0qNNA=5FwDN5`t
zH-NFC7DP_Zo02~pg!aDvTuHJ%?6K}!RVHGizJD^;2F6APyp0
z16$t#W+!xhoyaeo-AMzY5KtJn#N@}9`?lps2+N@OkTQph?uu*D(hI8G@AagoyPqhz
z8wEultwH$U+9@eB-9CerdnL1OO)Ca*aIT$i7Fs1C0YAe+6(!w|Kp^iWl}kQq>*e+H
zs(-uUJmPV8iq@h;Z6b_3;Rgj+6li#65NkKxt=4mbuyk4q9HdLdW>(O6Z4%Wfrp8c$
zDhw>)rPp<3&A?K|i6#lz@?FlflDLd~Km4>>wCQ5>PU*#c_e2$(-gsOwgJqp|sRQmg
z;M)jVl&r!ku^z=NJs2*B?bmv_4;7q@RZk3JiBP+S%SyV1pQ3RoEo#7sB>zdbWs#9se+wETvFrr;85`oA{T;i8Bq}
zAh;2cT4M)XMf7`xF5eKMv~p#fcz+WGTS0K+XBD67B{L+i>CjTy$5Ao;CtQGoJVHpq
zwfE)k?X_2t+Mkrq`x8P($#-Vg=tbCl-zyng9*Qn5nR;oc^crNupVaK`Dj;ToREKTp
zIIg4wUogF6_xh~?MqOpi%)=exWbBUg$#3StZM;P`dEv-)4ItT2Ix#e~F@N5BX>|Ux
z4pb*Rs{5-$U|ZvSV-JJy76Q%wE!+?-g#4zFAc2ljhGPrmkmD6h8Yk5P7OCo&xIU3+
zNRmTA%StydB-L^Je$OYnl!UOFp%~Z#I)Sz91hFJ;5+l!z?p_%^
zs~RQegi#?#$ZB57`rOItE`Q}QSC991Eqn(>6RtKA`GJH3VPnU8F#
z_({mvs@3N+YN84Rk4=f#opRQTD<*5GXp5>QC>`IXDdR^1>+k5udqT{IfS)YwWvqym
zS~R#GYm0R$=90L7jctzEZ78TZrtRr%{{|V4aTYSh{SDY2$X?!naeoI1<{bFOxfO+P
z=29%Pz+h5wR(dV)W6HRuH5dzK?`h2S@Z4p=QdF>2OHdZO20OF4-V&w{5|6j|TFAr{
z0)5Yxu)bQUalYlYJX42G<(!J>po8?g7K0_c1a+xbsJBK+MSL;1@>#oN#X*xal^`nRH~^wwH5r`bL@fV<%QDt~eiuek~3;5yChhLS|Z
z+BdM>{I1=bi$yzG`h9Gf+$|3}V&B!k@>TQ*O0*w()B&O-XUovPqYiQV^Hc8A&WTsK
zE29aiG=;buV(?oRddLtcyE^u3&wM!jhv
zO14#*vFI8$UVqg#s16{j*l(7>f&F5|WFXl0cT^GZg!gy>QNIG%^=
zpP$)D>`6llS8NrcQBRITrp>_0d@!jSTwFc7=F*RQ6vvjZGWQ{Yqp|Y(rabT}22Qix@oNMKoJy+FJ|$h6IL5P#PLf`)J^b)nW<_Qb)&kTI$t_v)0|_R}gy
z{tFB9N%H!y*;}*Endv$%6}&=jlkL^k>A%v9)%3MuaAddG
zX4z+anA%R$EGe+B#vPjK*rRKU#S8jY0~yhYTUO&9bV-65_pz@}gLiT|At=$*xql`f
z-$Vytl;oWQ5-u(|w)vZ9zoXxF#&Y4Idh7>cA#E%i_d@2{E4K$$fp8-oU(UbtwaZy|
zTphjSl?A%*CwAwb%zDXfj*Im85SRz1eUz`%-^cEuet)VrgxTre-e7fkRE1k7KY)2SY5zeb?PY}c|@-M3or#K=*uljY(R1+{JRO2;R~4sQtXUGJr^s!h~VIgT0o47O&5
z2tRiPi+>{2g}EU^bppW~^wfy~@w+s-lb_cz>ZkQ2?)1EFlR_j0B*9|QK7T)jz@N6;
zGllw9tqqBt;Hrmjnc<$T1?(wTvwBBE;RY3Zg9t5BDy4Evhchb38%m{4R*@2WtwhO4+?yj2iA*|=1x@e1N7ui2h%|9#$;^5Aoq(!mTt
zQZ#1$lF-@rBJ+2AFC}3@Jz!q(3wmSBio%v=-(t{eg2JuJZVj!yH-F^xOAP^1%De`GjzIymsrr&B6TcahWeqh}+Sr{fhxpi8}
z|88*#ga9>);!5hvRDTW5s(-QJ$rv|7m)I-9&*2g0I};r#K_HQ{@kO+>UrGesW!^)~
zM`QWTB#(0#gWV8LI(hD_HPo7n!@0yt5jA~wxL=j5Re~rn;5wo#FQfQd&cuh6e0-M1
z;*HSIncOKH>zT6_J5ngYofK^LK8ou?^wjc#w?If+6J*B`m4D@WAqXq=zTg3D-|7%k
zwKo9<@uqUf;EZSNl>M!jpsYjm=bm<#?%3>CO;RfiW6WSQQspuXi|XFxeG{6Y&OrT4
z*UWj;YmtG(wY6tB8eBE*TJv+exAn11+9Pfpo2PE{a8NFEn5yi4{%{`TV43wksZ%jI
zT;E7eK$S*#cYpg6eu02kLTOYq!@Gbt+}G!5WNA!h=f7tju-W~NVY_<_UDm;UN6k7_Lg{4
zunlHqRIt9N`q(+%ZrS|8*UAYKTk^s%)5R9($#$=ioqtikn54wMZ+fbuk8j7y*d#i#
zrElNFImV{qtFK6sbE)npDu1CSarr2)I}0q4`Ds~B_o+M&$20esVKuUS?UE`EG0S|`
zdv1i(u$~X4aFB0Xs@br3R-+-6B3iz-53CX0&21}_EN}$D7Yd#u=k6)6$75bfzV@Ix
zQ>$h!e1Cq^if`#rvpmc&Ut$>*E*+GjHHuZ-4$}phH(J_2{8=M+TC1D2=$F`+Qz3gm
ze4JW#vkxmJ1^T2V|M;!S
zIC8B~hIZWUa@Rgg{|4{qZEX*@6mlWfd&^G%t5%l)k&Z=QX5o()q8T{~4lNRB^v)P*
zp?_@YN)KqS4t!E#%{Qw_>-?-4w$%OV80}m15Zjl#VnZ!~=l&Se-cuTay2p8)-B!6A
z!6>tIziL>M=|$;VX*628mevPrAxsv$-o8{#xayT8y$l(ctGKJeh5$ssq4c&DN%(P=
z{wcQcH8J_*F!b2*nqq?=E(!nS7kH%?XMgPBg8^{I+}@_8eb9C*EI+FzGWr=b2F%;fSU4d*2!CEhKTeNJ=Dabp6TQT!>&oud2i}(sf(rjryWAE0kx2PLL1{VwZ*SQH0ULPq(-a5q`sKp;^R+BMpWOHYmU
zY5h=({8L3M^nZS<^ga?(7_RYp&epgXe2rhi|C@A*pG
z%nioPuS~YV;X4y5RY~ZEQK|RY&_zE|a#c#h_i6+4V6p2?kZiyfQn9t_zT0wrXs`8-
zZGB-_&{Jq65Ap}RN*>wmae}%KoTIPArGi(s3eut|)&zgyqDmR>UqhLoCcy2|9=RKO3k2w9oOY)
zkx`$KPJh0VEs;M%Ic4%{EH2Q!*J1F+y5*v)@SGGN(S_XezGyN^sa*>LAq3wnd-~>{
zu$M)0_(~BqcV!kmDp{z?OOCb6>=_&0E4di`
zAJ6NO{RU7~r+fNOweyOmU>upw~4X4;Lnt?zyOHj?t6!J?!76lsd~!J3`_;J5vy+&S40akkN{X4
z+fGaS((8>@48~uE@uWfDu`oP?GXe`JgVML93MF)PvqT4U8b58}@n?R@>SS=ag$1=5Hr%SY37ON#npG&?R4h$0u>>aXk?%NKzt+&?1Xi
zKi-{`&?FTduo&nMjx)o#@by$1({On9o$XdA^I5GS-0?U7?jf~n9CnYnrUzjr@Hpm#
zDv*Vh5{3}=hP*?+c=$IzQ#&(N|3e~fK;vVN`$aG>_8^y6L6toui`J#zaVp!2`uiRCx*l5b;
z5Vpmr^?y5~v=Uub^*BlN+Id4+!ZHyxxA{UU2w``^($Cdu6a{Q?zb;*}M(i&qWtCjT
zMs*JvttEa9Y>LLm-We61;dZmbYmWp7oCTrY8T&NkVFLAT!X^cOagXTd$qNml02%m;
z;d=&AgBLRcj4E3QSo$tin#OwewYlOfdm{MqoPW;GCgNRUz~0ChUYuT45OhR*DRy_*
zVA!EvctkEpk)S6)w0F(%zNsZCNm+!|nA2I2{Mbf0Y43!3tHi4loXe3*Zp?lVbiP)+
z^>I_p)qgMKwxMQRPAT_VRfmg~r)qpEcfA^_&3n+B@Q)Eta1zosLXQ2MZ2Wqt?c^*<
z_5(b#1l0-d|QJ{XG)JT4YZ*G2u&2FDS>8j#QZ(+yH6J
zr+O3(sci(z&54v%Ay8CVrXdhH=eiRwW*VbQ
zATEuYcbLl}=+OD`8;CnBe}NBTA@P;~?0++>1r0%(+iy%sc-l79fW1;3$U_G1!kNov
zCe{Ee0%J=OiaS#HcP%~A!&70B=C#^a;<0=CR=4r*tjWxFRTL`6jyNDKIHe-q@#iR)
znRH`k1gq5zS9^>GcwmEm2CrkfZ5(Ww!et=anSQq-{~$Q^Q;61KWPz_4!wZ
zn|Yc*o=Azk!OYHea#5XH_1GO9t$+OBO~qbs=VR#>&)ZQB6tbyanPrxrZ#eZHJitI$Y+~R8F>$u&b_3pL;7|wbxdU{6p!LoGdX1
zGI1>0KhZq}qiQi?flfivET{qx%bo42_ZjthVsy+3=Q?*3c{75i*X-ID=XW%q5_Jt_r%
zK;da?%cj>>v8~W&QElR3y_HgDYx@A5uf~1K+An0R!v+G|*{#;NE`Re|T9U!}G>GgJ
ztC$qrm`^7+q8CTTilLAzi%rXGz0F)~#v*x9`rLg8`@uxDKX*1$Cihvw?06j2q~r=%
zDjr%uNH&s`fSFWl>FWuXA>k>0JSNwB#vGr5;(X)HGvmYx=$t4rVxI*c3(-(xTKmoC
z*2LF1Q0+C={Q=cGgwx+Tx|{ZmFhE
zRz03aJw?e*mx^x^G%#Wz-<2Ob)7Jyy6!~qBom#*E;gd}*TVF&#rOHKNlRj^E6i~2b+UkTK@
zgp23ascWnRHE!p21=h%!3P(fsx$Z=q+2gIv&B|x_ZriAheN!$bi1giJ?u#BjBx!)a
z!BZ&DwGuRVPJhU^nR;m7@_R*8@@oy_F4o?HRgJS8H~rY9!!MJh@tzl+LucXjH8Bn_
z1|^`2>;Y&XF(q%k2l&x!%u+G#E}L5XpC7umy6HklH(v^6>fLTYe@-)(w1wP5a3x@~
z2H-fEWMccooY>BZHNnKTZJ*e7GO=wN6WbHpwv&IV?td3|b$8ueZ`HTyRo7RKdOvuc
zqeo;9w|-F89qWb9TEvEUceZvJLnm%p5W)Ijo{Dy?6j9u+Clv1|gcJZ;<@2*!8Q%uG|_h2J`VccSLmF}2Q}_R
z4^^vb5I_SLjG-*9DBtclR1z|-Jq&v|>yj8?1>QljUSYT0iPHC63
zUC(dCo~JPoRDw1r*1&cdi^$2b;mDoPBZ`Oha4Jm5usqSd%{S8{w2&sxB;+Ykq1~h#
znFdpb#QJ@`9`$>IaS}bcG>JWMzi!8
zna=kN#Ae(I{8fm1v2DoM$M;+tPj}?T>}^D7=X$L==}(HL%(RD&1>~J1(m#zToShPk
zwY*}HOSgc)=>Z$En(ubFl-=xpA>j8EeDB_Do-2$rxq7NZ$MrLL;?4~y$=U3ZVHu|I
z_%wVei!5Nn=Z*9RG07=#$UfyfuFW1+-)^;o`w)PNFRWNM)XmhDeIrNt{rV
zD0jt?d!)`rtFp^q+f}ZXSyOo#vOL$NEBc)qIk~zsl2#FQj;8BQt2}3?kYhRbMhT?F
zJ`mm5)!X$4{R{q{)n~zrY`7DcjHnnlI-^~iX%uFMx=_RI<;dDm%F7NYw4BUVol3Y~TNjlOhEkJc{LuDm)urH(vji={;Z4TOe_H1zCM8B~>uB7d
z-6uL6_r5<@MgdHCOQ@cqdJI_iY7bFB);j@09b~5^S9ePxFV!(3lC}8``IRtB7FRzC
z7nS-|spuEi$qPwLvo8-DO)oKa{74WQ$*~Gx9!zbVh|XN*>I_o)cVF5h^+*LzWT)jL
zAV%nUYS$$KVq&CkAg&jm><72%10j0aZ176*e~J!EBY9s9qT%VfKKZ`+Q$IU{Vv#he
zNdg8(wXcGgm+-!kPR_xeb2(HdF}uT3#mlT6-mosIxR|FZIuwq%EFjoX67x@)*5D5{
zVU`Ddzd1`^z0O(;m=;Lm_5HObfJjU?TJ6@7a+yo4KM7<#YGRH;S(LMg6QiT))KaoX
z2-}<(*l`49h^~E*pnVcn%;{`_;(05zo?LNg2D9T3wELORa;HA=%7jKQ>1yglJZjYA
z%w}qrefVn(F!H~WdbEN3Gw($3snft11~`RVA77f9WCmY5ld{{(WY%m?FV#5b?K$^2
zfWyD_I)#>jaTa_>@{tW%H>DJKYtuGkIu9_)*;`+FZ-%_sSWGqPVG05Tp%1ZtwWXe9
zmg!GHW87Bl@D6NLmWPp
z!`OrRhz~?2@OW+l9ivLRjJCbtQJh;5I68g^JD&v=;vvMPNpHr};&Ep_^q}IkG$s#Q
zy>B&k7iYQdr^p&NS1E#iqFjIFiW5gt0f6@`;xC#QP$|eQYGik3s6pGc_Ry_Xq$sJdod60vochbtMjI2&VMney
z3ZZ&uHg9%fyr741QLO_XGrG2tg%=rh-^pn^wz&D
zKU_{6vx^K}343)!?Tl1)Uef&d{CJIlC&I~-neC!(B(hG#$Hqj?#(g|^?L6#q-aoUF
zIt6HMq5Q7N8UW(E1-&n7B_#LIwqjM138&L!7_>3DahLc*DDXAn#N2F8{LrVgZPv>(;J^e#ksK{4dBIcB=7sm#s7BnA=e+Q)MajJW2^!EKyPju3HpN}w
zmiPZ*p{L?WKwI6%}ejlQY;M
zkN_8YX+;4vgkA5TUXxQ)Nof!5V@6rWdxk{`%jW?G6gWi52#&s>FnvFE7KW8R`tvOi
zKp6c}3vH%tKOk%8t_+hIYA@O9WVaW%tHLyScjT^QRy=CgQlddJ;d60JrHR;+8+YbZ
zB$AFl%LW>07?FD$BQE=G`i;+p*c4A6{tuhF|PLq}wh}2%_R>XH`0?=e}-dDv-
zHs(wHVy1V8&0A;E@tiv!kJeNf;HQ|HW8?D~BPRynunpU(tMqyWHv_O!AaK9p3*Ygd4_=%?p!7
z&L^0)yM^0C^H(m9>kBpgA{>FXN$}Qe`pLgHTuE&0@Pa5Ih~^^A{v1AblRiVeRmyI8
zT)Hmd2adIW+*K}GRE`fkG$2H8_qZY@%Yl3}R&*`kmcUam$qUKFr|9RNkdfh5#$t(j
zug2Qz=KSD9a=o2@oltSAWYLnnC#aB_gHH~2N`I__Z|)8_wxBt8Jy}2(3Z0$kYOPdc
z6U{Ab3yhwsX~cai^7-!Fm?7o6G8tXo2h%AQoQ%a@lGAs!R=@$2G{-Rtfx8{bF%GD<87
z1)xf|yW`p&Vz<4d%5l*5E1u~ha|69*4aB@ENYV;HH$Rt2xnV?C5^f?0gfwiI8nhU0
z`i-)V};4SFFnV$yFZ9>IQh}DP_Yhag(DE*Yn{UR?^agLYwq|bU-k&x|fBMB1WL)jbZSK^iKVOPW7tsy20Tkq!Wn@kgy9xJ}(C_&%kDOxtTT$G%lv!nB
z@rCL(EPn>`h#!&}Ls;-GYQvh5`HR~MeCbmdEwmZN7q_vk`>4dFczNkbS1In~F=|c2
zslY?7w?n#OO0X1?pPEnUEh>hQ#=jtheZZGP9*{XpclUI7|DE!<9DVap%l&&JIx%rQA<lN)y#bCafA+L^SLPq6OIHR>fd~Cq=mo9gy5uk!W_aO7r&8@
z@qsAuDq~;Mz0=PK3(MJOfXkA0h>?6TKS|iQ`jJ^>G%4}Wh6k&sC0^vDS`wU-=p7!M
zHTrgw$jZ+DD3=0K`>dM~lXk47zOMV##WODtgAL&xHWA5uKJz=&;3hY+oJ9I02(lZ*
z(Oax_7Vuyh%Bk_EXzs)KHDx7^HCoo3xRw=RyGlOBfesaQGlQ(9YLR>-7rl~_wS3G*E
z2e~J4D+vRCl}+2naX{-CsWdgYSXkE}+k69?nIFLF{%oynBZ!{)ER->Vq%m)K<`
zTLFFcSt*m5wk{TQcI#RJyQhY`(7hjHA?{6Yo>`m(C>)TQVgd>2qGGss;phJO`N!WQ
zS7+^7eSzVuv?0DDPk#S?Q^rN9ec_LQpP{P$beB%*yxC*c&?B<0JC}!#lLqGjp;}C~
zP0%GIPtVP2nm%M;=CjQuikBvbKzE{5IPph4S^H2difc3vDfoUnZ%?>d^w9caU$zwo
zEsali!JX~iazkqV%Z;*-Z}SQ6^k_}#Xn8a_)X6-hRblDN07HJsE;(`NcIOGcCxWJf
z2HXly(RYP!W{GZ=(2FQTYLH47WLj7Gt`&Y}rT-WFdWi6aa114E1XacrI8~Hk&1{4J
z?y#n)Rn{!j1@KZ6y+9h~7noLP=LD{zHBamID<24*KBib`r8QWE?Z+L?6^c@OCs`gO
z#ez9uI%V!eO??{sLIGiRRK2XQV6d^oEMs@6J*Ss3=zqqx=jwN^sdzbo*jH|2C#T;{
zbv(*h%n-93)_4-xZz%t+GL*}}HvDct)!$z!<`VQnNI)+bVi9SiW9-@CKMPBLabJ9w_kG-rf(I-0~A+~GU|lW`B%$RlsrGforsq=$)9F=
z&H5OUwP4F1E$qUx4M6}|co)XhZk0bIT$P8IqMvBwaxl+tDt$dDU%He=7qjG07?%tw
zQgz9};<(-VN?oWh?G~ZFzmtd&SXP1);@=Ti$>}DwYArT>_q3}6sST>Pu@h#xOf!^t
z!)70H8UCzp+V7%oYi4K`JKk?)44D6bP&XPD7)PzGGg3{I(_o(MGe0H`DOwRWAbgzWlc~cAElvAy)Nc*I@gR_LvV=y?k_0xXXihmR5
zKBT2WYslTIwrT-_u2f|z3;8>qk{Kj09;2`+3iBadqaNbd=L5alk(K!~S=((hld=?&
zR||nE!1NCKZXR!5=|hg#u)mKd$~;%
z`fPs#^Rgs*1?L7GVg!08tqCYvSqQjpA*7C10(o&k--_+2?r7;pV6<$aXrvY?2o;Ug
zeOvJ#Yp+NCD!3m)j#MR#uarP_>=wi-#=^28rqt}8ZCxO{HA!tycrcU{-!&CftF9?y
zej8kg*RU=Lx%ASa)8ubBD%wOfNK=lGz=^b`OSxbn{uWUv_tx<%XX34n#)vb6RACsTTxH>si82<0t#
zOKxa3-e?>{YJLHUo6+0d${Dkqy`%q}P0x#4mSw7Pr;)xKGl@lY3Z9d0LPlizHp>*&
zP+xCs;QX(aRQ;>r0telE!|ubO1jmPq`ylxuJC3}4n^EoM;o<2`C*!es^G8KEO{uIc
zPO5MQI662>1NQ?VHaJ_0+IMiiZ{Qp)NtochLEv01DjDFWJm5?%oeSU_lHe>Y8H5lb
zX8&@VK@jSu|8ksX5Y_+8sg}b>2(NG8|8Ne-Fc>CIR#q0~yxvCP@9c0>A^5GEU$`2qPre}@s`
zAX6-Yz@!O?Hn4ChTTFUT1Gu9YeOf$JNWUt|>Z~fbeCUs$^ENP5_5He{2X|U}Oh1HU
z>FeRRd7*V!`uG^|U)X@czN8=&m|qbbbwcXDvwJ&83Zd?>Jipc6yQltv0DKz{08{EE
z-UUbJlajAeuT4vG=YgYxL;xK3CFN7|r9%*Y6Om3UP#`+r&U?C>jv{9MwD#mPdH?oiyyF`hbroAaeV(evw6QlVXiBjM
z8`w}kOe{~O!d-?nfvyuvoDz9*;H}QE9qh2};=zM=F4xl_56#WuXD{q~Y4AdDCTUGN
zp85MjAp0F@!+1*rG_7s$YF|2f;QsD`UKEkE)GqLDj;~ZDst5l0Lvn}9C$}51%(=K`
zQng+hvMTd3j%AxWdn(p6&>%c6%nAucMP-T#F6sI8-s|DK6i=A~-641Q$pAv!bgkqb
zP~VlC5jWmVAvpRWG?n~{f)SJSmZauLyh)>dIWWARfZVAEq7*<$C7vi5D5y8g0qiCE
z48@LLVOAt}&q)<(J)l`!9#T4b3C62&N5J{Or`VWB!)lQiR+$H%d$kb$xRub;+fh}U
zCEa-p21)8!thMgZF^}&1uI2Dilz9Estulc#xia-SXm0jNWn0@sO|hiYglrx=t}<8g
zlehVYg4V}jYt2g6yy<>wY2@!gHwAa-`WY!zG|4)q0-zyrLC5+!5(jeXoQOu=u8%Gp
zf^Yo-i?`QQ2Q9O;eBRC2vDflbZ31l^$-Enz96wHDNE!OB%$OQR+^JJWw8gfRJi@7K
zfe#ccC6@(QccKQJ{q8et0k<`Fr99C5llhW&-NZaL>y#Xte+6jiD}@zlAfGKUxySP5
z4+`@fh)cy@K&F`&zZ6R?kTh6l^8(g@^zO{*d7p*nV&4t
za~ASlqYo`Kf1w1G#lmBs@{kFiC|LN1#84>zF$UtpfuxZSzjRcZ*2WYT59?
ztVY<<_fgCTIX4hZOJE
zxXhSgvk+voNp(7RNEstrHRothoI15{rOsA)Gt!i68Jy&u>jq7d>DP(s<2~?0*5!wy
z*a+i#ah4e&*dGa?IjVUltwU0W0De4ie4!&pNy)!14*
zAadm}e>?iJO-tqw1Rg?;I`)`PFcH`S;3~k4q($0iulZ}T!sz#NlU9L|?@uyCEJp8%
zzZgbD8mW$65e&f`7|vre#464aoyTL~N8stoqJe|3@y|G?Q|Jb#UOe?_D*G?l)P~0;
z(ZONh@kL=(bku@uJV&}?A@PRXXe_@HLA5`9xahMYNi`kXY=+2UR{Ia0^r=M-T-bk1
zT1j|Xp|6>zn=<5y^1rgL^d~hKZYovxSj#E%LD$B7Ji4F0l^W19>~`2HvN@Gs7Nx&6
zrW;u|ATQ{qBGo-ti~K}|yBD=GN~0yV5lxFD3FpDmhQRUUL8D<;PyK0*KT7H^0a6Xr
zQ!jr~+NwyS0n4~!$d>^D&yDW{hMFQlYlq3wAP$#=F+=6J;ez8$
zQWwtV<`1*0hM!rkveh77(K4#{&x$I#ov`Q=-V>4b*M7q<`afGD
zbDq&ZlLU`?Zn57nL6ck8WCGiI&v7}o2{|`TzMn>}x<8-8zWTm~hu;I_LC;Ck&;Q`A
zy<4kfPbb64Bkw*s-_f3%tM{v{!o^Rx1of@%i4CZwilZJPLhs7Hgk}E&IC^A8Him!$
zO?{6I@JQVj@CrEmzSVcC6X
zHZ_Q8tRLq|=0;CgI20{-O-zz-y#_oP<7?)hHyfX$zugEYHy$D)D_e5g_l56W9Sb&R
z@p)EH*36&opUi6X2pU<2n=&j^6mI=&Gd?E(##BV@qXTxSeAa(UL8}Z~6_C3gglKlB
zxm<@=&t9MYIo5Eox4X`@Tt^2vse#Oc@E=j`?=Kl~g%IDPCkv|DvsicHq^VLCRDF3i{(9*X);SBp2ldX$xXpwxz|U>rS1M>D
zEG4QY{JV2&2GUbzKCnjoJRv4jwoM-1A||@@dH<24Bq7MqEjt|fyT0xFxnixe!ngL1
z=6AS~r41z7F5FDU)IA4{o5*no_KfI`@=e<1i);eHuMZWqMSR1Mm@55aaiK2moDj$rHJF&kbmZr5xOo4M4C!ma;Xx=)MIo10TfI$AVLPqrQ
zTtYKx$@W%S9Q0TBhHAa5A*J&0Dwj)rj@_^nHvUAhA$MP6Pru|ugz~(%ZY+-b(LpM0
z_Z(o|(XN(yPcVW*7me3nJg4!`SI$N3VPu3i&(M8^urDfF(Q)Xkg-dsCjREP268Ee1
z!5CNfPf)pBOI@|{&q`^E56ZH2NJ$qtjZV@uEp%C1G+91B{ogXd)!MpM>544}Xw~r*
z?qF>kIb(E=>1PKC`SCIWe%hQHT56STA$XXeA14u-JVvTT^DU{DAsq)1_Et0L&Gz^8
zqCW-wtT|ON=t6XmgE$
zQe9qtm%W3_y@nJB*p$ITAIQrW4@gKXg0U0%h{*&APWCo1x~iLZjS*(5RJk>YZ1r0FPU)-ad0s|HU2K?{y9C<&Hb
zJd(Q^i^2^eTL;x6Yw-ml-q9njq21>f^`Z8Lm?R^VYAS)|jxImQe^ogD6$zi5Tg^hc
zFg4gf@_AW5xNtjWJ-$@`n`f8Ld2!^*Qz-f1I!Y{pCP3Z=TxBS&@-7(sZUEJ8KgfLT
z?q_7?xg>K_`_4d1hwliOT|Yaq7~T3Y){?r2gAExpwWh(n675x&aU(dfHmsj|g9M2V
z&Xo#{3<&^nbL+q}{<82i0dRA}GpYf!m;tN+=6^FKJ3D6pH^)D^R|%d`%GT5lz|HyJ
zMiijMC(I?z#>~pbA|@;%&cw_nD$FSgWMbt60)fKJ?4m$n0l@zs`S(g{F)}0vh$YYf
zBlPADL?#JD%lX@*4m>y9dh$6UjXGlrnTpzMGFRqF{#n6nS%5fg;d|FP$X((-IZ@t6
zzk+|52jTM&K?(}SsEz!H;`;ZL^`GX`IYTl1O_h`
zwZ0aS+dPy~P>9Kk!~X}P
CZY*a2

diff --git a/doc/src/DimRed/DimRed.do.txt b/doc/src/DimRed/DimRed.do.txt
index 052f91f72..6f5848873 100644
--- a/doc/src/DimRed/DimRed.do.txt
+++ b/doc/src/DimRed/DimRed.do.txt
@@ -139,7 +139,7 @@ print("R2 score for  scaled data: {:.2f}".format(svm.score(X_test_scaled,y_test)
 
 
 !split
-===== Simple preprocessing examples, breast cancer data and classification =====
+===== Simple preprocessing examples, breast cancer data and classification, Support Vector Machines  =====
 
 We show here how we can use a simple regression case on the breast cancer data using support vector machine as algorithm for classification
 
@@ -188,6 +188,59 @@ print("Test set accuracy scaled data: {:.2f}".format(svm.score(X_test_scaled,y_t
 !ec
 
 
+!split
+===== More on Cancer Data, now with Logistic Regression  =====
+!bc pycod
+import matplotlib.pyplot as plt
+import numpy as np
+from sklearn.model_selection import  train_test_split 
+from sklearn.datasets import load_breast_cancer
+from sklearn.linear_model import LogisticRegression
+cancer = load_breast_cancer()
+
+fig, axes = plt.subplots(15,2,figsize=(10,20))
+male = cancer.data[cancer.target == 0]
+bene = cancer.data[cancer.target == 1]
+ax = axes.ravel()
+
+for i in range(30):
+    _, bins = np.histogram(cancer.data[:,i], bins =50)
+    ax[i].hist(male[:,i], bins = bins, alpha = 0.5)
+    ax[i].hist(bene[:,i], bins = bins, alpha = 0.5)
+    ax[i].set_title(cancer.feature_names[i])
+    ax[i].set_yticks(())
+ax[0].set_xlabel("Feature magnitude")
+ax[0].set_ylabel("Frequency")
+ax[0].legend(["Male", "Bene"], loc ="best")
+fig.tight_layout()
+plt.show()
+
+# Set up training data
+X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)
+# Perform Logistic Regression (replace with own code) 
+logreg = LogisticRegression()
+logreg.fit(X_train, y_train)
+print("Test set accuracy: {:.2f}".format(logreg.score(X_test,y_test)))
+
+# Scale data
+from sklearn.preprocessing import StandardScaler
+scaler = StandardScaler()
+scaler.fit(X_train)
+X_train_scaled = scaler.transform(X_train)
+X_test_scaled = scaler.transform(X_test)
+logreg.fit(X_train_scaled, y_train)
+#svm.fit(X_train_scaled, y_train)
+print("Test set accuracy scaled data: {:.2f}".format(logreg.score(X_test_scaled,y_test)))
+
+# Now add PCA
+from sklearn.decomposition import PCA
+pca = PCA(n_components = 2)
+pca.fit(X_train_scaled)
+
+X_pca = pca.transform(X_train_scaled)
+
+
+!ec
 
 !split
 ===== Principal Component Analysis =====
diff --git a/doc/src/DimRed/cancer.py b/doc/src/DimRed/cancer.py
new file mode 100644
index 000000000..e1c06c60c
--- /dev/null
+++ b/doc/src/DimRed/cancer.py
@@ -0,0 +1,43 @@
+import matplotlib.pyplot as plt
+import numpy as np
+from sklearn.model_selection import  train_test_split 
+from sklearn.datasets import load_breast_cancer
+from sklearn.svm import SVC
+from sklearn.linear_model import LogisticRegression
+cancer = load_breast_cancer()
+
+X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)
+print(X_train.shape)
+print(X_test.shape)
+
+logreg = LogisticRegression()
+logreg.fit(X_train, y_train)
+
+#svm = SVC(C=100)
+#svm.fit(X_train, y_train)
+print("Test set accuracy: {:.2f}".format(logreg.score(X_test,y_test)))
+
+from sklearn.preprocessing import MinMaxScaler, StandardScaler
+
+scaler = StandardScaler()
+scaler.fit(X_train)
+X_train_scaled = scaler.transform(X_train)
+X_test_scaled = scaler.transform(X_test)
+
+print("Feature min values before scaling:\n {}".format(X_train.min(axis=0)))
+print("Feature max values before scaling:\n {}".format(X_train.max(axis=0)))
+
+print("Feature min values before scaling:\n {}".format(X_train_scaled.min(axis=0)))
+print("Feature max values before scaling:\n {}".format(X_train_scaled.max(axis=0)))
+
+logreg.fit(X_train_scaled, y_train)
+#svm.fit(X_train_scaled, y_train)
+print("Test set accuracy scaled data: {:.2f}".format(logreg.score(X_test_scaled,y_test)))
+
+scaler = StandardScaler()
+scaler.fit(X_train)
+X_train_scaled = scaler.transform(X_train)
+X_test_scaled = scaler.transform(X_test)
+logreg.fit(X_train_scaled, y_train)
+#svm.fit(X_train_scaled, y_train)
+print("Test set accuracy scaled data: {:.2f}".format(logreg.score(X_test_scaled,y_test)))
diff --git a/doc/src/DimRed/cancerlogreg.py b/doc/src/DimRed/cancerlogreg.py
new file mode 100644
index 000000000..45f03f36a
--- /dev/null
+++ b/doc/src/DimRed/cancerlogreg.py
@@ -0,0 +1,40 @@
+import matplotlib.pyplot as plt
+import numpy as np
+from sklearn.model_selection import  train_test_split 
+from sklearn.datasets import load_breast_cancer
+from sklearn.linear_model import LogisticRegression
+cancer = load_breast_cancer()
+
+fig, axes = plt.subplots(15,2,figsize=(10,20))
+malignant = cancer.data[cancer.target == 0]
+benign = cancer.data[cancer.target == 1]
+ax = axes.ravel()
+
+for i in range(30):
+    _, bins = np.histogram(cancer.data[:,i], bins =50)
+    ax[i].hist(malignant[:,i], bins = bins, alpha = 0.5)
+    ax[i].hist(benign[:,i], bins = bins, alpha = 0.5)
+    ax[i].set_title(cancer.feature_names[i])
+    ax[i].set_yticks(())
+ax[0].set_xlabel("Feature magnitude")
+ax[0].set_ylabel("Frequency")
+ax[0].legend(["Malignant", "Benign"], loc ="best")
+fig.tight_layout()
+plt.show()
+
+X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)
+print(X_train.shape)
+print(X_test.shape)
+
+logreg = LogisticRegression()
+logreg.fit(X_train, y_train)
+print("Test set accuracy from Logistic Regression: {:.2f}".format(logreg.score(X_test,y_test)))
+
+from sklearn.preprocessing import MinMaxScaler, StandardScaler
+scaler = StandardScaler()
+scaler.fit(X_train)
+X_train_scaled = scaler.transform(X_train)
+X_test_scaled = scaler.transform(X_test)
+
+logreg.fit(X_train_scaled, y_train)
+print("Test set accuracy scaled data: {:.2f}".format(logreg.score(X_test_scaled,y_test)))