From 54fc7da531b31f241e6af39bb596fe91539f6d07 Mon Sep 17 00:00:00 2001 From: mhjensen Date: Thu, 26 Dec 2019 11:06:16 +0100 Subject: [PATCH] added pca code --- doc/pub/DimRed/html/._DimRed-bs000.html | 105 +++++----- doc/pub/DimRed/html/._DimRed-bs001.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs002.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs003.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs004.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs005.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs006.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs007.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs008.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs009.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs010.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs011.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs012.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs013.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs014.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs015.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs016.html | 103 ++++++---- doc/pub/DimRed/html/._DimRed-bs017.html | 106 +++++----- doc/pub/DimRed/html/._DimRed-bs018.html | 202 ++++++++++++++----- doc/pub/DimRed/html/._DimRed-bs019.html | 127 ++++++------ doc/pub/DimRed/html/._DimRed-bs020.html | 134 ++++++------ doc/pub/DimRed/html/._DimRed-bs021.html | 168 +++++++-------- doc/pub/DimRed/html/._DimRed-bs022.html | 189 +++++++++-------- doc/pub/DimRed/html/._DimRed-bs023.html | 162 +++++++++------ doc/pub/DimRed/html/._DimRed-bs024.html | 155 +++++++------- doc/pub/DimRed/html/._DimRed-bs025.html | 157 ++++++++------ doc/pub/DimRed/html/._DimRed-bs026.html | 132 +++++++----- doc/pub/DimRed/html/._DimRed-bs027.html | 118 ++++++----- doc/pub/DimRed/html/._DimRed-bs028.html | 126 ++++++------ doc/pub/DimRed/html/._DimRed-bs029.html | 132 +++++++----- doc/pub/DimRed/html/._DimRed-bs030.html | 121 ++++++----- doc/pub/DimRed/html/DimRed-bs.html | 105 +++++----- doc/pub/DimRed/html/DimRed-reveal.html | 134 ++++++++++-- doc/pub/DimRed/html/DimRed-solarized.html | 162 ++++++++++++--- doc/pub/DimRed/html/DimRed.html | 162 ++++++++++++--- doc/pub/DimRed/ipynb/DimRed.ipynb | 168 ++++++++++++++- doc/pub/DimRed/ipynb/ipynb-DimRed-src.tar.gz | Bin 190 -> 190 bytes doc/pub/DimRed/pdf/DimRed-minted.pdf | Bin 253434 -> 259876 bytes doc/src/DimRed/DimRed.do.txt | 88 ++++++++ 39 files changed, 2866 insertions(+), 1735 deletions(-) diff --git a/doc/pub/DimRed/html/._DimRed-bs000.html b/doc/pub/DimRed/html/._DimRed-bs000.html index 57753922e..8da1ff958 100644 --- a/doc/pub/DimRed/html/._DimRed-bs000.html +++ b/doc/pub/DimRed/html/._DimRed-bs000.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -203,7 +218,7 @@ MathJax.Hub.Config({
    [2] Department of Physics and Astronomy and National Superconducting Cyclotron Laboratory, Michigan State University

    -

    Oct 25, 2019

    +

    Dec 26, 2019


    @@ -227,7 +242,7 @@ MathJax.Hub.Config({

  • 9
  • 10
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs001.html b/doc/pub/DimRed/html/._DimRed-bs001.html index 5eacb11b3..026609651 100644 --- a/doc/pub/DimRed/html/._DimRed-bs001.html +++ b/doc/pub/DimRed/html/._DimRed-bs001.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -227,7 +242,7 @@ data.
  • 10
  • 11
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs002.html b/doc/pub/DimRed/html/._DimRed-bs002.html index 1a1f68938..f281284d9 100644 --- a/doc/pub/DimRed/html/._DimRed-bs002.html +++ b/doc/pub/DimRed/html/._DimRed-bs002.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -226,7 +241,7 @@ ensures that all features are exactly between \( 0 \) and \( 1 \). The
  • 11
  • 12
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs003.html b/doc/pub/DimRed/html/._DimRed-bs003.html index 2d3bfbf9a..d2bfcf82f 100644 --- a/doc/pub/DimRed/html/._DimRed-bs003.html +++ b/doc/pub/DimRed/html/._DimRed-bs003.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -229,7 +244,7 @@ techniques.
  • 12
  • 13
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs004.html b/doc/pub/DimRed/html/._DimRed-bs004.html index 32a4b3df2..b9946be9a 100644 --- a/doc/pub/DimRed/html/._DimRed-bs004.html +++ b/doc/pub/DimRed/html/._DimRed-bs004.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -304,7 +319,7 @@ svm.fit(X_train_scaled, y_train)
  • 13
  • 14
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs005.html b/doc/pub/DimRed/html/._DimRed-bs005.html index 0f389a6c6..c1f31f395 100644 --- a/doc/pub/DimRed/html/._DimRed-bs005.html +++ b/doc/pub/DimRed/html/._DimRed-bs005.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -254,7 +269,7 @@ svm.fit(X_train_scaled, y_train)
  • 14
  • 15
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs006.html b/doc/pub/DimRed/html/._DimRed-bs006.html index d5e29d62d..2f361bebf 100644 --- a/doc/pub/DimRed/html/._DimRed-bs006.html +++ b/doc/pub/DimRed/html/._DimRed-bs006.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -233,7 +248,7 @@ logreg.fit(X_train_scaled, y_train)
  • 15
  • 16
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs007.html b/doc/pub/DimRed/html/._DimRed-bs007.html index 9a71aec08..1c3b2e6e7 100644 --- a/doc/pub/DimRed/html/._DimRed-bs007.html +++ b/doc/pub/DimRed/html/._DimRed-bs007.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -288,7 +303,7 @@ applications.
  • 16
  • 17
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs008.html b/doc/pub/DimRed/html/._DimRed-bs008.html index c2a86cade..ec4603204 100644 --- a/doc/pub/DimRed/html/._DimRed-bs008.html +++ b/doc/pub/DimRed/html/._DimRed-bs008.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -218,7 +233,7 @@ We have a data set defined by a design/feature matrix \( \boldsymbol{X} \) (see
  • 17
  • 18
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs009.html b/doc/pub/DimRed/html/._DimRed-bs009.html index 710178b2e..05ce56b2b 100644 --- a/doc/pub/DimRed/html/._DimRed-bs009.html +++ b/doc/pub/DimRed/html/._DimRed-bs009.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -267,7 +282,7 @@ In the above example this is the function we constructed using pandas.
  • 18
  • 19
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs010.html b/doc/pub/DimRed/html/._DimRed-bs010.html index be8c93dfd..9fcb9607f 100644 --- a/doc/pub/DimRed/html/._DimRed-bs010.html +++ b/doc/pub/DimRed/html/._DimRed-bs010.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -262,7 +277,7 @@ $$
  • 19
  • 20
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs011.html b/doc/pub/DimRed/html/._DimRed-bs011.html index 0b8212479..77d5296ca 100644 --- a/doc/pub/DimRed/html/._DimRed-bs011.html +++ b/doc/pub/DimRed/html/._DimRed-bs011.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -250,7 +265,7 @@ C = np.c
  • 20
  • 21
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs012.html b/doc/pub/DimRed/html/._DimRed-bs012.html index b8c4a9bf3..7a8763b59 100644 --- a/doc/pub/DimRed/html/._DimRed-bs012.html +++ b/doc/pub/DimRed/html/._DimRed-bs012.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -252,7 +267,7 @@ The above procedure with numpy can be made more compact if we use pand
  • 21
  • 22
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs013.html b/doc/pub/DimRed/html/._DimRed-bs013.html index 3e0982c2d..2ab2ddff5 100644 --- a/doc/pub/DimRed/html/._DimRed-bs013.html +++ b/doc/pub/DimRed/html/._DimRed-bs013.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -234,7 +249,7 @@ We expand this model to the Franke function discussed above.
  • 22
  • 23
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs014.html b/doc/pub/DimRed/html/._DimRed-bs014.html index 77882c670..d7c6c6753 100644 --- a/doc/pub/DimRed/html/._DimRed-bs014.html +++ b/doc/pub/DimRed/html/._DimRed-bs014.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -270,7 +285,7 @@ matrix.
  • 23
  • 24
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs015.html b/doc/pub/DimRed/html/._DimRed-bs015.html index 865870b95..86ccf8390 100644 --- a/doc/pub/DimRed/html/._DimRed-bs015.html +++ b/doc/pub/DimRed/html/._DimRed-bs015.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -250,7 +265,7 @@ It is easy to generalize this to a matrix \( \boldsymbol{X}\in {\mathbb{R}}^{n\t
  • 24
  • 25
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs016.html b/doc/pub/DimRed/html/._DimRed-bs016.html index c1988b0b0..6f742428c 100644 --- a/doc/pub/DimRed/html/._DimRed-bs016.html +++ b/doc/pub/DimRed/html/._DimRed-bs016.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -258,7 +273,7 @@ features/predictors.
  • 25
  • 26
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs017.html b/doc/pub/DimRed/html/._DimRed-bs017.html index 73aecf9a2..3d8d8fdb9 100644 --- a/doc/pub/DimRed/html/._DimRed-bs017.html +++ b/doc/pub/DimRed/html/._DimRed-bs017.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -213,9 +228,6 @@ $$
  • Keep only those \( l \) eigenvalues larger than a selected threshold value, discarding thus \( p-l \) features since we expect small variations in the data here.
    • -After this we ask ourselves how do we prove the link between the maximum variance and the feature reduction. - -

    diff --git a/doc/pub/DimRed/html/._DimRed-bs018.html b/doc/pub/DimRed/html/._DimRed-bs018.html index e0bd197b8..0455b1a78 100644 --- a/doc/pub/DimRed/html/._DimRed-bs018.html +++ b/doc/pub/DimRed/html/._DimRed-bs018.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -184,23 +199,102 @@ MathJax.Hub.Config({ -

    Classical PCA Theorem

    +

    Writing our own PCA code

    -We assume now that we have a design matrix \( \boldsymbol{X} \) which has been centered as discussed above. For the sake of simplicity we skip the overline symbol. The matrix is defined in terms of the various column vectors \( [\boldsymbol{x}_0,\boldsymbol{x}_1,\dots, \boldsymbol{x}_{p-1}] \) -each with dimension \( \boldsymbol{x}\in {\mathbb{R}}^{n} \). - -

    -We assume also that we have an orthogonal transformation \( \boldsymbol{W}\in {\mathbb{R}}^{p\times p} \). We define the reconstruction error (which is similar to the mean squared error we have seen before) as +We will use a simple example first with two-dimensional data +drawn from a multivariate normal distribution with the following mean and covariance matrix: $$ -J(\boldsymbol{W},\boldsymbol{Z}) = \frac{1}{n}\sum_i (\boldsymbol{x}_i - \overline{\boldsymbol{x}}_i)^2, +\mu = (-1,2) \qquad \Sigma = \begin{bmatrix} 4 & 2 \\ +2 & 2 +\end{bmatrix} $$ -with \( \overline{\boldsymbol{x}}_i = \boldsymbol{W}\boldsymbol{z}_i \), where \( \boldsymbol{z}_i \) is a row vector with dimension \( {\mathbb{R}}^{n} \) of the matrix -\( \boldsymbol{Z}\in{\mathbb{R}}^{p\times n} \). When doing PCA we want to reduce this dimensionality. +

    +We will generate \( N = 1000 \) points \( X = \{ x_1, \ldots, x_N \} \) from +this distribution, and store them in the \( 1000 \times 2 \) matrix \( \boldsymbol{X} \).

    -The PCA theorem states that minimizing the above reconstruction error corresponds to setting \( \boldsymbol{W}=\boldsymbol{S} \), the orthogonal matrix which diagonalizes the empirical covariance(correlation) matrix. The optimal low-dimensional encoding of the data is then given by a set of vectors \( \boldsymbol{z}_i \) with at most \( l \) vectors, with \( l < < p \), defined by the orthogonal projection of the data onto the columns spanned by the eigenvectors of the covariance(correlations matrix). +The following Python code aids in setting up the data + +

    + + +

    N = 1000
+mean = (-1, 2)
+cov = [[4, 2], [2, 2]]
+X = np.random.multivariate_normal(mean, cov, N)
+
    +

    +Make a small Python code which plots the data. + +

    +Now we are going to implement the PCA algorithm. We will break it down into sub-steps and across multiple cells. + +

    Compute the sample mean and center the data

    + +

    +The first step of PCA is to compute the sample mean of the data and use it to center the data. Recall the sample mean is +$$ +\mu_N = \frac{1}{N} \sum_{i=1}^N x_i +$$ + +and the mean-centered data \( \bar{X} = \{ \bar{x}_1, \ldots, \bar{x}_N \} \) takes the form +$$ +\bar{x}_i = x_i - \mu_N +$$ + +When you are done with these steps, print out \( \mu_N \) to verify it is +close to \( \mu \) and plot your mean centered data to verify it is +centered at the origin! Compare your code with the functionality from Scikit-Learn discussed above. + +

    Compute the sample covariance

    + +

    +Now we are going to use the mean centered data to compute the sample covariance of the data. Recall it is given by: +$$ +\begin{equation*} +\Sigma_N = \frac{1}{N-1} \sum_{i=1}^N \bar{x}_i^T \bar{x}_i = \frac{1}{N-1} \sum_{i=1}^N (x_i - \mu_N)^T (x_i - \mu_N) +\end{equation*} +$$ + +where the data points \( x_i \in \mathbb{R}^p \) (here in this example \( p = 2 \)) are column vectors and \( x^T \) is the transpose of \( x \). +Compare the computed covariance with the answer given above. + +

    Diagonalize the sample covariance matrix to obtain the principal components

    + +

    +Now we are ready to solve for the principal components! To do so we +diagonalize the sample covariance matrix \( \Sigma_N \). We can use the +function np.linalg.eig to do so. It will return the eigenvalues and +eigenvectors of \( \Sigma_N \). Once you have these, carry out the +following tasks: + +

    + +$$ +\begin{equation*} +x_i \approx \tilde{x}_i := \mu_N + \langle x_i, v_0 \rangle v_0 +\end{equation*} +$$ + +where \( v_0 \) is the first principal component. What do you observe? + +

    +Finally, collect all these steps and write your own PCA function and +compare this with the functionality included in Scikit-Learn. +Have the input be the data and have the output be the principal components and their associated eigenvalues, sorted in descending order. Can you think of a way to make it more efficient than the algorithm outlined above? + +

    +Finally, try out your own PCA function with other data sets. + +

    +After this we ask ourselves how do we prove the link between the maximum variance and the feature reduction.

    @@ -228,7 +322,7 @@ The PCA theorem states that minimizing the above reconstruction error correspond

  • 27
  • 28
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs019.html b/doc/pub/DimRed/html/._DimRed-bs019.html index a6e981422..cce5ff3ab 100644 --- a/doc/pub/DimRed/html/._DimRed-bs019.html +++ b/doc/pub/DimRed/html/._DimRed-bs019.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -184,25 +199,23 @@ MathJax.Hub.Config({ -

    Proof of the PCA Theorem

    +

    Classical PCA Theorem

    -To show the PCA theorem let us start with the assumption that there is one vector \( \boldsymbol{w}_0 \) which corresponds to a solution which minimized the reconstruction error \( J \). This is an orthogonal vector. It means that we now approximate the reconstruction error in terms of \( \boldsymbol{w}_0 \) and \( \boldsymbol{z}_0 \) as +We assume now that we have a design matrix \( \boldsymbol{X} \) which has been centered as discussed above. For the sake of simplicity we skip the overline symbol. The matrix is defined in terms of the various column vectors \( [\boldsymbol{x}_0,\boldsymbol{x}_1,\dots, \boldsymbol{x}_{p-1}] \) +each with dimension \( \boldsymbol{x}\in {\mathbb{R}}^{n} \). + +

    +We assume also that we have an orthogonal transformation \( \boldsymbol{W}\in {\mathbb{R}}^{p\times p} \). We define the reconstruction error (which is similar to the mean squared error we have seen before) as $$ -J(\boldsymbol{w}_0,\boldsymbol{z}_0)= \frac{1}{n}\sum_i (\boldsymbol{x}_i - z_{i0}\boldsymbol{w}_0)^2=\frac{1}{n}\sum_i (\boldsymbol{x}_i^T\boldsymbol{x}_i - 2z_{i0}\boldsymbol{w}_0^T\boldsymbol{x}_i+z_{i0}^2\boldsymbol{w}_0^T\boldsymbol{w}_0), +J(\boldsymbol{W},\boldsymbol{Z}) = \frac{1}{n}\sum_i (\boldsymbol{x}_i - \overline{\boldsymbol{x}}_i)^2, $$ -which we can rewrite due to the orthogonality of \( \boldsymbol{w}_i \) as -$$ -J(\boldsymbol{w}_0,\boldsymbol{z}_0)=\frac{1}{n}\sum_i (\boldsymbol{x}_i^T\boldsymbol{x}_i - 2z_{i0}\boldsymbol{w}_0^T\boldsymbol{x}_i+z_{i0}^2). -$$ +with \( \overline{\boldsymbol{x}}_i = \boldsymbol{W}\boldsymbol{z}_i \), where \( \boldsymbol{z}_i \) is a row vector with dimension \( {\mathbb{R}}^{n} \) of the matrix +\( \boldsymbol{Z}\in{\mathbb{R}}^{p\times n} \). When doing PCA we want to reduce this dimensionality. -Minimizing \( J \) with respect to the unknown parameters \( z_{0i} \) we obtain that -$$ -z_{i0}=\boldsymbol{w}_0^T\boldsymbol{x}_i, -$$ - -where the vectors on the rhs are known. +

    +The PCA theorem states that minimizing the above reconstruction error corresponds to setting \( \boldsymbol{W}=\boldsymbol{S} \), the orthogonal matrix which diagonalizes the empirical covariance(correlation) matrix. The optimal low-dimensional encoding of the data is then given by a set of vectors \( \boldsymbol{z}_i \) with at most \( l \) vectors, with \( l < < p \), defined by the orthogonal projection of the data onto the columns spanned by the eigenvectors of the covariance(correlations matrix).

    @@ -230,7 +243,7 @@ where the vectors on the rhs are known.

  • 28
  • 29
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs020.html b/doc/pub/DimRed/html/._DimRed-bs020.html index d1e36909d..536339c85 100644 --- a/doc/pub/DimRed/html/._DimRed-bs020.html +++ b/doc/pub/DimRed/html/._DimRed-bs020.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -184,40 +199,25 @@ MathJax.Hub.Config({ -

    PCA Proof continued

    +

    Proof of the PCA Theorem

    -We have now found the unknown parameters \( z_{i0} \). These correspond to the projected coordinates and we can write +To show the PCA theorem let us start with the assumption that there is one vector \( \boldsymbol{w}_0 \) which corresponds to a solution which minimized the reconstruction error \( J \). This is an orthogonal vector. It means that we now approximate the reconstruction error in terms of \( \boldsymbol{w}_0 \) and \( \boldsymbol{z}_0 \) as $$ -J(\boldsymbol{w}_0)= \frac{1}{p}\sum_i (\boldsymbol{x}_i^T\boldsymbol{x}_i - z_{i0}^2)=\mathrm{const}-\frac{1}{n}\sum_i z_{i0}^2. +J(\boldsymbol{w}_0,\boldsymbol{z}_0)= \frac{1}{n}\sum_i (\boldsymbol{x}_i - z_{i0}\boldsymbol{w}_0)^2=\frac{1}{n}\sum_i (\boldsymbol{x}_i^T\boldsymbol{x}_i - 2z_{i0}\boldsymbol{w}_0^T\boldsymbol{x}_i+z_{i0}^2\boldsymbol{w}_0^T\boldsymbol{w}_0), $$ -

    -We can show that the variance of the projected coordinates defined by \( \boldsymbol{w}_0^T\boldsymbol{x}_i \) are given by +which we can rewrite due to the orthogonality of \( \boldsymbol{w}_i \) as $$ -\mathrm{var}[\boldsymbol{w}_0^T\boldsymbol{x}_i] = \frac{1}{n}\sum_i z_{i0}^2, +J(\boldsymbol{w}_0,\boldsymbol{z}_0)=\frac{1}{n}\sum_i (\boldsymbol{x}_i^T\boldsymbol{x}_i - 2z_{i0}\boldsymbol{w}_0^T\boldsymbol{x}_i+z_{i0}^2). $$ -since the expectation value of +Minimizing \( J \) with respect to the unknown parameters \( z_{0i} \) we obtain that $$ -\mathbb{E}[\boldsymbol{w}_0^T\boldsymbol{x}_i] = \mathbb{E}[z_{i0}]= \boldsymbol{w}_0^T\mathbb{E}[\boldsymbol{x}_i]=0, +z_{i0}=\boldsymbol{w}_0^T\boldsymbol{x}_i, $$ -where we have used the fact that our data are centered. - -

    -Recalling our definition of the covariance as -$$ -\boldsymbol{C}[\boldsymbol{x}] = \frac{1}{n}\boldsymbol{X}\boldsymbol{X}^T=\mathbb{E}[\boldsymbol{X}\boldsymbol{X}^T], -$$ - -we have thus that -$$ -\mathrm{var}[\boldsymbol{w}_0^T\boldsymbol{x}_i] = \frac{1}{n}\sum_i z_{i0}^2=\boldsymbol{w}_0^T\boldsymbol{C}[\boldsymbol{x}]\boldsymbol{w}_0. -$$ - -

    -We are almost there, we have obtained a relation between minimizing the reconstruction error and the variance and the covariance matrix. Minimizing the error is equivalent to maximizing the variance of the projected data. +where the vectors on the rhs are known.

    @@ -245,7 +245,7 @@ We are almost there, we have obtained a relation between minimizing the reconstr

  • 29
  • 30
  • ...
    • -
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs021.html b/doc/pub/DimRed/html/._DimRed-bs021.html index ee37e1a02..2ecf2204b 100644 --- a/doc/pub/DimRed/html/._DimRed-bs021.html +++ b/doc/pub/DimRed/html/._DimRed-bs021.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -184,53 +199,40 @@ MathJax.Hub.Config({ -

    The final step

    +

    PCA Proof continued

    -We could trivially maximize the variance of the projection (and -thereby minimize the error in the reconstruction function) by letting -the norm-2 of \( \boldsymbol{w}_0 \) go to infinity. However, this norm since we -want the matrix \( \boldsymbol{W} \) to be an orthogonal matrix, is constrained by -\( \vert\vert \boldsymbol{w}_0 \vert\vert_2^2=1 \). Imposing this condition via a -Lagrange multiplier we can then in turn maximize - +We have now found the unknown parameters \( z_{i0} \). These correspond to the projected coordinates and we can write $$ -J(\boldsymbol{w}_0)= \boldsymbol{w}_0^T\boldsymbol{C}[\boldsymbol{x}]\boldsymbol{w}_0+\lambda_0(1-\boldsymbol{w}_0^T\boldsymbol{w}_0). -$$ - -Taking the derivative with respect to \( \boldsymbol{w}_0 \) we obtain - -$$ -\frac{\partial J(\boldsymbol{w}_0)}{\partial \boldsymbol{w}_0}= 2\boldsymbol{C}[\boldsymbol{x}]\boldsymbol{w}_0-2\lambda_0\boldsymbol{w}_0=0, -$$ - -meaning that -$$ -\boldsymbol{C}[\boldsymbol{x}]\boldsymbol{w}_0=\lambda_0\boldsymbol{w}_0. -$$ - -The direction that maximizes the variance (or minimizes the construction error) is an eigenvector of the covariance matrix! If we left multiply with \( \boldsymbol{w}_0^T \) we have the variance of the projected data is -$$ -\boldsymbol{w}_0^T\boldsymbol{C}[\boldsymbol{x}]\boldsymbol{w}_0=\lambda_0. +J(\boldsymbol{w}_0)= \frac{1}{p}\sum_i (\boldsymbol{x}_i^T\boldsymbol{x}_i - z_{i0}^2)=\mathrm{const}-\frac{1}{n}\sum_i z_{i0}^2. $$

    -If we want to maximize the variance (minimize the construction error) -we simply pick the eigenvector of the covariance matrix with the -largest eigenvalue. This establishes the link between the minimization -of the reconstruction function \( J \) in terms of an orthogonal matrix -and the maximization of the variance and thereby the covariance of our -observations encoded in the design/feature matrix \( \boldsymbol{X} \). +We can show that the variance of the projected coordinates defined by \( \boldsymbol{w}_0^T\boldsymbol{x}_i \) are given by +$$ +\mathrm{var}[\boldsymbol{w}_0^T\boldsymbol{x}_i] = \frac{1}{n}\sum_i z_{i0}^2, +$$ + +since the expectation value of +$$ +\mathbb{E}[\boldsymbol{w}_0^T\boldsymbol{x}_i] = \mathbb{E}[z_{i0}]= \boldsymbol{w}_0^T\mathbb{E}[\boldsymbol{x}_i]=0, +$$ + +where we have used the fact that our data are centered.

    -The proof -for the other eigenvectors \( \boldsymbol{w}_1,\boldsymbol{w}_2,\dots \) can be -established by applying the above arguments and using the fact that -our basis of eigenvectors is orthogonal, see Murphy chapter -12.2. The -discussion in chapter 12.2 of Murphy's text has also a nice link with -the Singular Value Decomposition theorem. For categorical data, see -chapter 12.4 and discussion therein. +Recalling our definition of the covariance as +$$ +\boldsymbol{C}[\boldsymbol{x}] = \frac{1}{n}\boldsymbol{X}\boldsymbol{X}^T=\mathbb{E}[\boldsymbol{X}\boldsymbol{X}^T], +$$ + +we have thus that +$$ +\mathrm{var}[\boldsymbol{w}_0^T\boldsymbol{x}_i] = \frac{1}{n}\sum_i z_{i0}^2=\boldsymbol{w}_0^T\boldsymbol{C}[\boldsymbol{x}]\boldsymbol{w}_0. +$$ + +

    +We are almost there, we have obtained a relation between minimizing the reconstruction error and the variance and the covariance matrix. Minimizing the error is equivalent to maximizing the variance of the projected data.

    @@ -257,6 +259,8 @@ chapter 12.4 and discussion therein.

  • 29
  • 30
  • 31
    • +
  • ...
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs022.html b/doc/pub/DimRed/html/._DimRed-bs022.html index b4a7a67e2..285b05f8e 100644 --- a/doc/pub/DimRed/html/._DimRed-bs022.html +++ b/doc/pub/DimRed/html/._DimRed-bs022.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -184,59 +199,54 @@ 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 final step

    -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. First we center the data using either pandas or our own code -

    +We could trivially maximize the variance of the projection (and +thereby minimize the error in the reconstruction function) by letting +the norm-2 of \( \boldsymbol{w}_0 \) go to infinity. However, this norm since we +want the matrix \( \boldsymbol{W} \) to be an orthogonal matrix, is constrained by +\( \vert\vert \boldsymbol{w}_0 \vert\vert_2^2=1 \). Imposing this condition via a +Lagrange multiplier we can then in turn maximize - -

    import numpy as np
-import pandas as pd
-from IPython.display import display
-np.random.seed(100)
-# setting up a 10 x 5 vanilla matrix 
-rows = 10
-cols = 5
-X = np.random.randn(rows,cols)
-df = pd.DataFrame(X)
-# Pandas does the centering for us
-df = df -df.mean()
-display(df)
+$$
+J(\boldsymbol{w}_0)= \boldsymbol{w}_0^T\boldsymbol{C}[\boldsymbol{x}]\boldsymbol{w}_0+\lambda_0(1-\boldsymbol{w}_0^T\boldsymbol{w}_0).
+$$
 
-# we center it ourselves
-X_centered = X - X.mean(axis=0)
-# Then check the difference between pandas and our own set up
-print(X_centered-df)
-#Now we do an SVD
-U, s, V = np.linalg.svd(X_centered)
-c1 = V.T[:, 0]
-c2 = V.T[:, 1]
-W2 = V.T[:, :2]
-X2D = X_centered.dot(W2)
-print(X2D)
-
    -

    -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. +Taking the derivative with respect to \( \boldsymbol{w}_0 \) we obtain + +$$ +\frac{\partial J(\boldsymbol{w}_0)}{\partial \boldsymbol{w}_0}= 2\boldsymbol{C}[\boldsymbol{x}]\boldsymbol{w}_0-2\lambda_0\boldsymbol{w}_0=0, +$$ + +meaning that +$$ +\boldsymbol{C}[\boldsymbol{x}]\boldsymbol{w}_0=\lambda_0\boldsymbol{w}_0. +$$ + +The direction that maximizes the variance (or minimizes the construction error) is an eigenvector of the covariance matrix! If we left multiply with \( \boldsymbol{w}_0^T \) we have the variance of the projected data is +$$ +\boldsymbol{w}_0^T\boldsymbol{C}[\boldsymbol{x}]\boldsymbol{w}_0=\lambda_0. +$$

    -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. -

    +If we want to maximize the variance (minimize the construction error) +we simply pick the eigenvector of the covariance matrix with the +largest eigenvalue. This establishes the link between the minimization +of the reconstruction function \( J \) in terms of an orthogonal matrix +and the maximization of the variance and thereby the covariance of our +observations encoded in the design/feature matrix \( \boldsymbol{X} \). + +

    +The proof +for the other eigenvectors \( \boldsymbol{w}_1,\boldsymbol{w}_2,\dots \) can be +established by applying the above arguments and using the fact that +our basis of eigenvectors is orthogonal, see Murphy chapter +12.2. The +discussion in chapter 12.2 of Murphy's text has also a nice link with +the Singular Value Decomposition theorem. For categorical data, see +chapter 12.4 and discussion therein. - -

    W2 = V.T[:, :2]
-X2D = X_centered.dot(W2)
-

    @@ -261,6 +271,7 @@ X2D = X_centered29

  • 30
  • 31
    • +
  • 32
  • »
    • diff --git a/doc/pub/DimRed/html/._DimRed-bs023.html b/doc/pub/DimRed/html/._DimRed-bs023.html index dbef737d6..67b68fa28 100644 --- a/doc/pub/DimRed/html/._DimRed-bs023.html +++ b/doc/pub/DimRed/html/._DimRed-bs023.html @@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source '___sec14'), ('Towards the PCA theorem', 2, None, '___sec15'), ('The Algorithm before the Theorem', 2, None, '___sec16'), - ('Classical PCA Theorem', 2, None, '___sec17'), - ('Proof of the PCA Theorem', 2, None, '___sec18'), - ('PCA Proof continued', 2, None, '___sec19'), - ('The final step', 2, None, '___sec20'), - ('Principal Component Analysis', 2, None, '___sec21'), - ('PCA and scikit-learn', 2, None, '___sec22'), - ('Back to the Cancer Data', 2, None, '___sec23'), - ('More on the PCA', 2, None, '___sec24'), - ('Incremental PCA', 2, None, '___sec25'), - ('Randomized PCA', 2, None, '___sec26'), - ('Kernel PCA', 2, None, '___sec27'), - ('LLE', 2, None, '___sec28'), - ('Other techniques', 2, None, '___sec29')]} + ('Writing our own PCA code', 2, None, '___sec17'), + ('Compute the sample mean and center the data', + 3, + None, + '___sec18'), + ('Compute the sample covariance', 3, None, '___sec19'), + ('Diagonalize the sample covariance matrix to obtain the ' + 'principal components', + 3, + None, + '___sec20'), + ('Classical PCA Theorem', 2, None, '___sec21'), + ('Proof of the PCA Theorem', 2, None, '___sec22'), + ('PCA Proof continued', 2, None, '___sec23'), + ('The final step', 2, None, '___sec24'), + ('Principal Component Analysis', 2, None, '___sec25'), + ('PCA and scikit-learn', 2, None, '___sec26'), + ('Back to the Cancer Data', 2, None, '___sec27'), + ('More on the PCA', 2, None, '___sec28'), + ('Incremental PCA', 2, None, '___sec29'), + ('Randomized PCA', 2, None, '___sec30'), + ('Kernel PCA', 2, None, '___sec31'), + ('LLE', 2, None, '___sec32'), + ('Other techniques', 2, None, '___sec33')]} end of tocinfo --> @@ -139,36 +150,40 @@ MathJax.Hub.Config({
  • Contents
    • @@ -182,37 +197,61 @@ 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. First we center the data using either pandas or our own code

    -

    #thereafter we do a PCA with Scikit-learn
-from sklearn.decomposition import PCA
-pca = PCA(n_components = 2)
-X2D = pca.fit_transform(X)
+
    import numpy as np
+import pandas as pd
+from IPython.display import display
+np.random.seed(100)
+# setting up a 10 x 5 vanilla matrix 
+rows = 10
+cols = 5
+X = np.random.randn(rows,cols)
+df = pd.DataFrame(X)
+# Pandas does the centering for us
+df = df -df.mean()
+display(df)
+
+# we center it ourselves
+X_centered = X - X.mean(axis=0)
+# Then check the difference between pandas and our own set up
+print(X_centered-df)
+#Now we do an SVD
+U, s, V = np.linalg.svd(X_centered)
+c1 = V.T[:, 0]
+c2 = V.T[:, 1]
+W2 = V.T[:, :2]
+X2D = X_centered.dot(W2)
 print(X2D)
 
    
 
    
-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.
-
 
    
 
    
 
@@ -236,6 +275,7 @@ variance that lies along the axis of each principal component.
   
  • 29
    • 
   
  • 30
    • 
   
  • 31
    • 
+  
  • 32
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs024.html b/doc/pub/DimRed/html/._DimRed-bs024.html
index 534db1781..2e151571d 100644
--- a/doc/pub/DimRed/html/._DimRed-bs024.html
+++ b/doc/pub/DimRed/html/._DimRed-bs024.html
@@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source
                '___sec14'),
               ('Towards the PCA theorem', 2, None, '___sec15'),
               ('The Algorithm before the Theorem', 2, None, '___sec16'),
-              ('Classical PCA Theorem', 2, None, '___sec17'),
-              ('Proof of the PCA Theorem', 2, None, '___sec18'),
-              ('PCA Proof continued', 2, None, '___sec19'),
-              ('The final step', 2, None, '___sec20'),
-              ('Principal Component Analysis', 2, None, '___sec21'),
-              ('PCA and scikit-learn', 2, None, '___sec22'),
-              ('Back to the Cancer Data', 2, None, '___sec23'),
-              ('More on the PCA', 2, None, '___sec24'),
-              ('Incremental PCA', 2, None, '___sec25'),
-              ('Randomized PCA', 2, None, '___sec26'),
-              ('Kernel PCA', 2, None, '___sec27'),
-              ('LLE', 2, None, '___sec28'),
-              ('Other techniques', 2, None, '___sec29')]}
+              ('Writing our own PCA code', 2, None, '___sec17'),
+              ('Compute the sample mean and center the data',
+               3,
+               None,
+               '___sec18'),
+              ('Compute the sample covariance', 3, None, '___sec19'),
+              ('Diagonalize the sample covariance matrix to obtain the '
+               'principal components',
+               3,
+               None,
+               '___sec20'),
+              ('Classical PCA Theorem', 2, None, '___sec21'),
+              ('Proof of the PCA Theorem', 2, None, '___sec22'),
+              ('PCA Proof continued', 2, None, '___sec23'),
+              ('The final step', 2, None, '___sec24'),
+              ('Principal Component Analysis', 2, None, '___sec25'),
+              ('PCA and scikit-learn', 2, None, '___sec26'),
+              ('Back to the Cancer Data', 2, None, '___sec27'),
+              ('More on the PCA', 2, None, '___sec28'),
+              ('Incremental PCA', 2, None, '___sec29'),
+              ('Randomized PCA', 2, None, '___sec30'),
+              ('Kernel PCA', 2, None, '___sec31'),
+              ('LLE', 2, None, '___sec32'),
+              ('Other techniques', 2, None, '___sec33')]}
 end of tocinfo -->
 
 
@@ -139,36 +150,40 @@ MathJax.Hub.Config({
       
  • 
         Contents 
         
       
    • 
@@ -182,45 +197,36 @@ MathJax.Hub.Config({
 
     
     
     
     
 
 
-
+
 
-
    Back to the Cancer Data 
    
-We can now repeat the above but applied to real data, in this case our breast cancer data.
-Here we compute performance scores on the training data using logistic regression.
+
    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):
 
    
 
 
-
    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()
-
-X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)
-
-logreg = LogisticRegression()
-logreg.fit(X_train, y_train)
-print("Train set accuracy from Logistic Regression: {:.2f}".format(logreg.score(X_train,y_train)))
-# We scale the 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)
-# Then perform again a log reg fit
-logreg.fit(X_train_scaled, y_train)
-print("Train set accuracy scaled data: {:.2f}".format(logreg.score(X_train_scaled,y_train)))
-#thereafter we do a PCA with Scikit-learn
+
    #thereafter we do a PCA with Scikit-learn
 from sklearn.decomposition import PCA
 pca = PCA(n_components = 2)
-X2D_train = pca.fit_transform(X_train_scaled)
-# and finally compute the log reg fit and the score on the training data	
-logreg.fit(X2D_train,y_train)
-print("Train set accuracy scaled and PCA data: {:.2f}".format(logreg.score(X2D_train,y_train)))
+X2D = pca.fit_transform(X)
+print(X2D)
 
    
 
    
-We see that our training data after the PCA decomposition has a performance similar to the non-scaled data.
+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.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.
 
 
    
 
    
@@ -244,6 +250,7 @@ We see that our training data after the PCA decomposition has a performance simi
   
  • 29
    • 
   
  • 30
    • 
   
  • 31
    • 
+  
  • 32
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs025.html b/doc/pub/DimRed/html/._DimRed-bs025.html
index 213172eef..a1f2ee2be 100644
--- a/doc/pub/DimRed/html/._DimRed-bs025.html
+++ b/doc/pub/DimRed/html/._DimRed-bs025.html
@@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source
                '___sec14'),
               ('Towards the PCA theorem', 2, None, '___sec15'),
               ('The Algorithm before the Theorem', 2, None, '___sec16'),
-              ('Classical PCA Theorem', 2, None, '___sec17'),
-              ('Proof of the PCA Theorem', 2, None, '___sec18'),
-              ('PCA Proof continued', 2, None, '___sec19'),
-              ('The final step', 2, None, '___sec20'),
-              ('Principal Component Analysis', 2, None, '___sec21'),
-              ('PCA and scikit-learn', 2, None, '___sec22'),
-              ('Back to the Cancer Data', 2, None, '___sec23'),
-              ('More on the PCA', 2, None, '___sec24'),
-              ('Incremental PCA', 2, None, '___sec25'),
-              ('Randomized PCA', 2, None, '___sec26'),
-              ('Kernel PCA', 2, None, '___sec27'),
-              ('LLE', 2, None, '___sec28'),
-              ('Other techniques', 2, None, '___sec29')]}
+              ('Writing our own PCA code', 2, None, '___sec17'),
+              ('Compute the sample mean and center the data',
+               3,
+               None,
+               '___sec18'),
+              ('Compute the sample covariance', 3, None, '___sec19'),
+              ('Diagonalize the sample covariance matrix to obtain the '
+               'principal components',
+               3,
+               None,
+               '___sec20'),
+              ('Classical PCA Theorem', 2, None, '___sec21'),
+              ('Proof of the PCA Theorem', 2, None, '___sec22'),
+              ('PCA Proof continued', 2, None, '___sec23'),
+              ('The final step', 2, None, '___sec24'),
+              ('Principal Component Analysis', 2, None, '___sec25'),
+              ('PCA and scikit-learn', 2, None, '___sec26'),
+              ('Back to the Cancer Data', 2, None, '___sec27'),
+              ('More on the PCA', 2, None, '___sec28'),
+              ('Incremental PCA', 2, None, '___sec29'),
+              ('Randomized PCA', 2, None, '___sec30'),
+              ('Kernel PCA', 2, None, '___sec31'),
+              ('LLE', 2, None, '___sec32'),
+              ('Other techniques', 2, None, '___sec33')]}
 end of tocinfo -->
 
 
@@ -139,36 +150,40 @@ MathJax.Hub.Config({
       
  • 
         Contents 
         
       
    • 
@@ -184,33 +199,44 @@ 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:
+
    Back to the Cancer Data 
    
+We can now repeat the above but applied to real data, in this case our breast cancer data.
+Here we compute performance scores on the training data using logistic regression.
 
    
 
 
-
    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:
-
    
+
    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()
 
-
-
    pca = PCA(n_components=0.95)
-X_reduced = pca.fit_transform(X)
+X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)
+
+logreg = LogisticRegression()
+logreg.fit(X_train, y_train)
+print("Train set accuracy from Logistic Regression: {:.2f}".format(logreg.score(X_train,y_train)))
+# We scale the 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)
+# Then perform again a log reg fit
+logreg.fit(X_train_scaled, y_train)
+print("Train set accuracy scaled data: {:.2f}".format(logreg.score(X_train_scaled,y_train)))
+#thereafter we do a PCA with Scikit-learn
+from sklearn.decomposition import PCA
+pca = PCA(n_components = 2)
+X2D_train = pca.fit_transform(X_train_scaled)
+# and finally compute the log reg fit and the score on the training data	
+logreg.fit(X2D_train,y_train)
+print("Train set accuracy scaled and PCA data: {:.2f}".format(logreg.score(X2D_train,y_train)))
 
    
+
    
+We see that our training data after the PCA decomposition has a performance similar to the non-scaled data.
+
 
    
 
    
 
@@ -232,6 +258,7 @@ X_reduced = pca
   
  • 29
    • 
   
  • 30
    • 
   
  • 31
    • 
+  
  • 32
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs026.html b/doc/pub/DimRed/html/._DimRed-bs026.html
index e3a14c74a..99b82253e 100644
--- a/doc/pub/DimRed/html/._DimRed-bs026.html
+++ b/doc/pub/DimRed/html/._DimRed-bs026.html
@@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source
                '___sec14'),
               ('Towards the PCA theorem', 2, None, '___sec15'),
               ('The Algorithm before the Theorem', 2, None, '___sec16'),
-              ('Classical PCA Theorem', 2, None, '___sec17'),
-              ('Proof of the PCA Theorem', 2, None, '___sec18'),
-              ('PCA Proof continued', 2, None, '___sec19'),
-              ('The final step', 2, None, '___sec20'),
-              ('Principal Component Analysis', 2, None, '___sec21'),
-              ('PCA and scikit-learn', 2, None, '___sec22'),
-              ('Back to the Cancer Data', 2, None, '___sec23'),
-              ('More on the PCA', 2, None, '___sec24'),
-              ('Incremental PCA', 2, None, '___sec25'),
-              ('Randomized PCA', 2, None, '___sec26'),
-              ('Kernel PCA', 2, None, '___sec27'),
-              ('LLE', 2, None, '___sec28'),
-              ('Other techniques', 2, None, '___sec29')]}
+              ('Writing our own PCA code', 2, None, '___sec17'),
+              ('Compute the sample mean and center the data',
+               3,
+               None,
+               '___sec18'),
+              ('Compute the sample covariance', 3, None, '___sec19'),
+              ('Diagonalize the sample covariance matrix to obtain the '
+               'principal components',
+               3,
+               None,
+               '___sec20'),
+              ('Classical PCA Theorem', 2, None, '___sec21'),
+              ('Proof of the PCA Theorem', 2, None, '___sec22'),
+              ('PCA Proof continued', 2, None, '___sec23'),
+              ('The final step', 2, None, '___sec24'),
+              ('Principal Component Analysis', 2, None, '___sec25'),
+              ('PCA and scikit-learn', 2, None, '___sec26'),
+              ('Back to the Cancer Data', 2, None, '___sec27'),
+              ('More on the PCA', 2, None, '___sec28'),
+              ('Incremental PCA', 2, None, '___sec29'),
+              ('Randomized PCA', 2, None, '___sec30'),
+              ('Kernel PCA', 2, None, '___sec31'),
+              ('LLE', 2, None, '___sec32'),
+              ('Other techniques', 2, None, '___sec33')]}
 end of tocinfo -->
 
 
@@ -139,36 +150,40 @@ MathJax.Hub.Config({
       
  • 
         Contents 
         
       
    • 
@@ -184,15 +199,33 @@ MathJax.Hub.Config({
 
 
 
-
    Incremental PCA 
    
+
    More on the 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).
+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)
+
    
 
    
 
    
 
@@ -213,6 +246,7 @@ instances arrive).
   
  • 29
    • 
   
  • 30
    • 
   
  • 31
    • 
+  
  • 32
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs027.html b/doc/pub/DimRed/html/._DimRed-bs027.html
index e5767b817..95a52b810 100644
--- a/doc/pub/DimRed/html/._DimRed-bs027.html
+++ b/doc/pub/DimRed/html/._DimRed-bs027.html
@@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source
                '___sec14'),
               ('Towards the PCA theorem', 2, None, '___sec15'),
               ('The Algorithm before the Theorem', 2, None, '___sec16'),
-              ('Classical PCA Theorem', 2, None, '___sec17'),
-              ('Proof of the PCA Theorem', 2, None, '___sec18'),
-              ('PCA Proof continued', 2, None, '___sec19'),
-              ('The final step', 2, None, '___sec20'),
-              ('Principal Component Analysis', 2, None, '___sec21'),
-              ('PCA and scikit-learn', 2, None, '___sec22'),
-              ('Back to the Cancer Data', 2, None, '___sec23'),
-              ('More on the PCA', 2, None, '___sec24'),
-              ('Incremental PCA', 2, None, '___sec25'),
-              ('Randomized PCA', 2, None, '___sec26'),
-              ('Kernel PCA', 2, None, '___sec27'),
-              ('LLE', 2, None, '___sec28'),
-              ('Other techniques', 2, None, '___sec29')]}
+              ('Writing our own PCA code', 2, None, '___sec17'),
+              ('Compute the sample mean and center the data',
+               3,
+               None,
+               '___sec18'),
+              ('Compute the sample covariance', 3, None, '___sec19'),
+              ('Diagonalize the sample covariance matrix to obtain the '
+               'principal components',
+               3,
+               None,
+               '___sec20'),
+              ('Classical PCA Theorem', 2, None, '___sec21'),
+              ('Proof of the PCA Theorem', 2, None, '___sec22'),
+              ('PCA Proof continued', 2, None, '___sec23'),
+              ('The final step', 2, None, '___sec24'),
+              ('Principal Component Analysis', 2, None, '___sec25'),
+              ('PCA and scikit-learn', 2, None, '___sec26'),
+              ('Back to the Cancer Data', 2, None, '___sec27'),
+              ('More on the PCA', 2, None, '___sec28'),
+              ('Incremental PCA', 2, None, '___sec29'),
+              ('Randomized PCA', 2, None, '___sec30'),
+              ('Kernel PCA', 2, None, '___sec31'),
+              ('LLE', 2, None, '___sec32'),
+              ('Other techniques', 2, None, '___sec33')]}
 end of tocinfo -->
 
 
@@ -139,36 +150,40 @@ MathJax.Hub.Config({
       
  • 
         Contents 
         
       
    • 
@@ -184,18 +199,14 @@ MathJax.Hub.Config({
 
 
 
-
    Randomized PCA 
    
+
    Incremental 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 \( n \).
-
-
    
-
    
-
    
-
+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).
 
 
    
 
    
@@ -216,6 +227,7 @@ previous algorithms when \( d \) is much smaller than \( n \).
   
  • 29
    • 
   
  • 30
    • 
   
  • 31
    • 
+  
  • 32
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs028.html b/doc/pub/DimRed/html/._DimRed-bs028.html
index 9cbb7aa9f..0c9c7f308 100644
--- a/doc/pub/DimRed/html/._DimRed-bs028.html
+++ b/doc/pub/DimRed/html/._DimRed-bs028.html
@@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source
                '___sec14'),
               ('Towards the PCA theorem', 2, None, '___sec15'),
               ('The Algorithm before the Theorem', 2, None, '___sec16'),
-              ('Classical PCA Theorem', 2, None, '___sec17'),
-              ('Proof of the PCA Theorem', 2, None, '___sec18'),
-              ('PCA Proof continued', 2, None, '___sec19'),
-              ('The final step', 2, None, '___sec20'),
-              ('Principal Component Analysis', 2, None, '___sec21'),
-              ('PCA and scikit-learn', 2, None, '___sec22'),
-              ('Back to the Cancer Data', 2, None, '___sec23'),
-              ('More on the PCA', 2, None, '___sec24'),
-              ('Incremental PCA', 2, None, '___sec25'),
-              ('Randomized PCA', 2, None, '___sec26'),
-              ('Kernel PCA', 2, None, '___sec27'),
-              ('LLE', 2, None, '___sec28'),
-              ('Other techniques', 2, None, '___sec29')]}
+              ('Writing our own PCA code', 2, None, '___sec17'),
+              ('Compute the sample mean and center the data',
+               3,
+               None,
+               '___sec18'),
+              ('Compute the sample covariance', 3, None, '___sec19'),
+              ('Diagonalize the sample covariance matrix to obtain the '
+               'principal components',
+               3,
+               None,
+               '___sec20'),
+              ('Classical PCA Theorem', 2, None, '___sec21'),
+              ('Proof of the PCA Theorem', 2, None, '___sec22'),
+              ('PCA Proof continued', 2, None, '___sec23'),
+              ('The final step', 2, None, '___sec24'),
+              ('Principal Component Analysis', 2, None, '___sec25'),
+              ('PCA and scikit-learn', 2, None, '___sec26'),
+              ('Back to the Cancer Data', 2, None, '___sec27'),
+              ('More on the PCA', 2, None, '___sec28'),
+              ('Incremental PCA', 2, None, '___sec29'),
+              ('Randomized PCA', 2, None, '___sec30'),
+              ('Kernel PCA', 2, None, '___sec31'),
+              ('LLE', 2, None, '___sec32'),
+              ('Other techniques', 2, None, '___sec33')]}
 end of tocinfo -->
 
 
@@ -139,36 +150,40 @@ MathJax.Hub.Config({
       
  • 
         Contents 
         
       
    • 
@@ -184,28 +199,14 @@ MathJax.Hub.Config({
 
 
 
-
    Kernel PCA 
    
-
    
-
    
-
     
+
    Randomized PCA 
    
 
 
    
-The kernel trick is a mathematical technique that implicitly maps instances into a
-very high-dimensional space (called the feature space), enabling nonlinear classification and regression
-with Support Vector Machines. Recall that a linear decision boundary in the high-dimensional feature
-space corresponds to a complex nonlinear decision boundary in the original space.
-It turns out that the same trick can be applied to PCA, making it possible to perform complex nonlinear
-projections for dimensionality reduction. This is called Kernel PCA (kPCA). It is often good at
-preserving clusters of instances after projection, or sometimes even unrolling datasets that lie close to a
-twisted manifold.
-For example, the following code uses Scikit-Learn’s KernelPCA class to perform kPCA with an
-
    
+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 \( n \).
 
-
-
    from sklearn.decomposition import KernelPCA
-rbf_pca = KernelPCA(n_components = 2, kernel="rbf", gamma=0.04)
-X_reduced = rbf_pca.fit_transform(X)
-
    
 
    
 
    
 
    
@@ -229,6 +230,7 @@ X_reduced = rbf_pca29
   
  • 30
    • 
   
  • 31
    • 
+  
  • 32
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs029.html b/doc/pub/DimRed/html/._DimRed-bs029.html
index bf5d8c9b0..e605c052c 100644
--- a/doc/pub/DimRed/html/._DimRed-bs029.html
+++ b/doc/pub/DimRed/html/._DimRed-bs029.html
@@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source
                '___sec14'),
               ('Towards the PCA theorem', 2, None, '___sec15'),
               ('The Algorithm before the Theorem', 2, None, '___sec16'),
-              ('Classical PCA Theorem', 2, None, '___sec17'),
-              ('Proof of the PCA Theorem', 2, None, '___sec18'),
-              ('PCA Proof continued', 2, None, '___sec19'),
-              ('The final step', 2, None, '___sec20'),
-              ('Principal Component Analysis', 2, None, '___sec21'),
-              ('PCA and scikit-learn', 2, None, '___sec22'),
-              ('Back to the Cancer Data', 2, None, '___sec23'),
-              ('More on the PCA', 2, None, '___sec24'),
-              ('Incremental PCA', 2, None, '___sec25'),
-              ('Randomized PCA', 2, None, '___sec26'),
-              ('Kernel PCA', 2, None, '___sec27'),
-              ('LLE', 2, None, '___sec28'),
-              ('Other techniques', 2, None, '___sec29')]}
+              ('Writing our own PCA code', 2, None, '___sec17'),
+              ('Compute the sample mean and center the data',
+               3,
+               None,
+               '___sec18'),
+              ('Compute the sample covariance', 3, None, '___sec19'),
+              ('Diagonalize the sample covariance matrix to obtain the '
+               'principal components',
+               3,
+               None,
+               '___sec20'),
+              ('Classical PCA Theorem', 2, None, '___sec21'),
+              ('Proof of the PCA Theorem', 2, None, '___sec22'),
+              ('PCA Proof continued', 2, None, '___sec23'),
+              ('The final step', 2, None, '___sec24'),
+              ('Principal Component Analysis', 2, None, '___sec25'),
+              ('PCA and scikit-learn', 2, None, '___sec26'),
+              ('Back to the Cancer Data', 2, None, '___sec27'),
+              ('More on the PCA', 2, None, '___sec28'),
+              ('Incremental PCA', 2, None, '___sec29'),
+              ('Randomized PCA', 2, None, '___sec30'),
+              ('Kernel PCA', 2, None, '___sec31'),
+              ('LLE', 2, None, '___sec32'),
+              ('Other techniques', 2, None, '___sec33')]}
 end of tocinfo -->
 
 
@@ -139,36 +150,40 @@ MathJax.Hub.Config({
       
  • 
         Contents 
         
       
    • 
@@ -184,14 +199,32 @@ MathJax.Hub.Config({
 
 
 
-
    LLE 
    
+
    Kernel PCA 
    
+
    
+
    
+
     
 
 
    
-Locally Linear Embedding (LLE) is another very powerful nonlinear dimensionality reduction
-(NLDR) technique. It is a Manifold Learning technique that does not rely on projections like the previous
-algorithms. In a nutshell, LLE works by first measuring how each training instance linearly relates to its
-closest neighbors (c.n.), and then looking for a low-dimensional representation of the training set where
-these local relationships are best preserved (more details shortly).
+The kernel trick is a mathematical technique that implicitly maps instances into a
+very high-dimensional space (called the feature space), enabling nonlinear classification and regression
+with Support Vector Machines. Recall that a linear decision boundary in the high-dimensional feature
+space corresponds to a complex nonlinear decision boundary in the original space.
+It turns out that the same trick can be applied to PCA, making it possible to perform complex nonlinear
+projections for dimensionality reduction. This is called Kernel PCA (kPCA). It is often good at
+preserving clusters of instances after projection, or sometimes even unrolling datasets that lie close to a
+twisted manifold.
+For example, the following code uses Scikit-Learn’s KernelPCA class to perform kPCA with an
+
    
+
+
+
    from sklearn.decomposition import KernelPCA
+rbf_pca = KernelPCA(n_components = 2, kernel="rbf", gamma=0.04)
+X_reduced = rbf_pca.fit_transform(X)
+
    
+
    
+
    
+
    
+
 
 
    
 
    
@@ -210,6 +243,7 @@ these local relationships are best preserved (more details shortly).
   
  • 29
    • 
   
  • 30
    • 
   
  • 31
    • 
+  
  • 32
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/._DimRed-bs030.html b/doc/pub/DimRed/html/._DimRed-bs030.html
index aa8394e93..283c55794 100644
--- a/doc/pub/DimRed/html/._DimRed-bs030.html
+++ b/doc/pub/DimRed/html/._DimRed-bs030.html
@@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source
                '___sec14'),
               ('Towards the PCA theorem', 2, None, '___sec15'),
               ('The Algorithm before the Theorem', 2, None, '___sec16'),
-              ('Classical PCA Theorem', 2, None, '___sec17'),
-              ('Proof of the PCA Theorem', 2, None, '___sec18'),
-              ('PCA Proof continued', 2, None, '___sec19'),
-              ('The final step', 2, None, '___sec20'),
-              ('Principal Component Analysis', 2, None, '___sec21'),
-              ('PCA and scikit-learn', 2, None, '___sec22'),
-              ('Back to the Cancer Data', 2, None, '___sec23'),
-              ('More on the PCA', 2, None, '___sec24'),
-              ('Incremental PCA', 2, None, '___sec25'),
-              ('Randomized PCA', 2, None, '___sec26'),
-              ('Kernel PCA', 2, None, '___sec27'),
-              ('LLE', 2, None, '___sec28'),
-              ('Other techniques', 2, None, '___sec29')]}
+              ('Writing our own PCA code', 2, None, '___sec17'),
+              ('Compute the sample mean and center the data',
+               3,
+               None,
+               '___sec18'),
+              ('Compute the sample covariance', 3, None, '___sec19'),
+              ('Diagonalize the sample covariance matrix to obtain the '
+               'principal components',
+               3,
+               None,
+               '___sec20'),
+              ('Classical PCA Theorem', 2, None, '___sec21'),
+              ('Proof of the PCA Theorem', 2, None, '___sec22'),
+              ('PCA Proof continued', 2, None, '___sec23'),
+              ('The final step', 2, None, '___sec24'),
+              ('Principal Component Analysis', 2, None, '___sec25'),
+              ('PCA and scikit-learn', 2, None, '___sec26'),
+              ('Back to the Cancer Data', 2, None, '___sec27'),
+              ('More on the PCA', 2, None, '___sec28'),
+              ('Incremental PCA', 2, None, '___sec29'),
+              ('Randomized PCA', 2, None, '___sec30'),
+              ('Kernel PCA', 2, None, '___sec31'),
+              ('LLE', 2, None, '___sec32'),
+              ('Other techniques', 2, None, '___sec33')]}
 end of tocinfo -->
 
 
@@ -139,36 +150,40 @@ MathJax.Hub.Config({
       
  • 
         Contents 
         
       
    • 
@@ -184,22 +199,16 @@ MathJax.Hub.Config({
 
 
 
-
    Other techniques 
    
+
    LLE 
    
 
 
    
-There are many other dimensionality reduction techniques, several of which are available in Scikit-Learn.
+Locally Linear Embedding (LLE) is another very powerful nonlinear dimensionality reduction
+(NLDR) technique. It is a Manifold Learning technique that does not rely on projections like the previous
+algorithms. In a nutshell, LLE works by first measuring how each training instance linearly relates to its
+closest neighbors (c.n.), and then looking for a low-dimensional representation of the training set where
+these local relationships are best preserved (more details shortly).
 
 
    
-Here are some of the most popular:
-
-
      
-
    •  Multidimensional Scaling (MDS) reduces dimensionality while trying to preserve the distances between the instances.
      • 
-
    •  Isomap creates a graph by connecting each instance to its nearest neighbors, then reduces dimensionality while trying to preserve the geodesic distances between the instances.
      • 
-
    •  t-Distributed Stochastic Neighbor Embedding (t-SNE) reduces dimensionality while trying to keep similar instances close and dissimilar instances apart. It is mostly used for visualization, in particular to visualize clusters of instances in high-dimensional space (e.g., to visualize the MNIST images in 2D).
      • 
-
    •  Linear Discriminant Analysis (LDA) is actually a classification algorithm, but during training it learns the most discriminative axes between the classes, and these axes can then be used to define a hyperplane onto which to project the data. The benefit is that the projection will keep classes as far apart as possible, so LDA is a good technique to reduce dimensionality before running another classification algorithm such as a Support Vector Machine (SVM) classifier discussed in the SVM lectures.
      • 
-
    
-
-
 
    
 
 
      
@@ -215,6 +224,8 @@ Here are some of the most popular:
   
    • 29
      • 
   
    • 30
      • 
   
    • 31
      • 
+  
    • 32
      • 
+  
    • »
      • 
 
    
 
 
diff --git a/doc/pub/DimRed/html/DimRed-bs.html b/doc/pub/DimRed/html/DimRed-bs.html
index 57753922e..8da1ff958 100644
--- a/doc/pub/DimRed/html/DimRed-bs.html
+++ b/doc/pub/DimRed/html/DimRed-bs.html
@@ -89,19 +89,30 @@ Automatically generated HTML file from DocOnce source
                '___sec14'),
               ('Towards the PCA theorem', 2, None, '___sec15'),
               ('The Algorithm before the Theorem', 2, None, '___sec16'),
-              ('Classical PCA Theorem', 2, None, '___sec17'),
-              ('Proof of the PCA Theorem', 2, None, '___sec18'),
-              ('PCA Proof continued', 2, None, '___sec19'),
-              ('The final step', 2, None, '___sec20'),
-              ('Principal Component Analysis', 2, None, '___sec21'),
-              ('PCA and scikit-learn', 2, None, '___sec22'),
-              ('Back to the Cancer Data', 2, None, '___sec23'),
-              ('More on the PCA', 2, None, '___sec24'),
-              ('Incremental PCA', 2, None, '___sec25'),
-              ('Randomized PCA', 2, None, '___sec26'),
-              ('Kernel PCA', 2, None, '___sec27'),
-              ('LLE', 2, None, '___sec28'),
-              ('Other techniques', 2, None, '___sec29')]}
+              ('Writing our own PCA code', 2, None, '___sec17'),
+              ('Compute the sample mean and center the data',
+               3,
+               None,
+               '___sec18'),
+              ('Compute the sample covariance', 3, None, '___sec19'),
+              ('Diagonalize the sample covariance matrix to obtain the '
+               'principal components',
+               3,
+               None,
+               '___sec20'),
+              ('Classical PCA Theorem', 2, None, '___sec21'),
+              ('Proof of the PCA Theorem', 2, None, '___sec22'),
+              ('PCA Proof continued', 2, None, '___sec23'),
+              ('The final step', 2, None, '___sec24'),
+              ('Principal Component Analysis', 2, None, '___sec25'),
+              ('PCA and scikit-learn', 2, None, '___sec26'),
+              ('Back to the Cancer Data', 2, None, '___sec27'),
+              ('More on the PCA', 2, None, '___sec28'),
+              ('Incremental PCA', 2, None, '___sec29'),
+              ('Randomized PCA', 2, None, '___sec30'),
+              ('Kernel PCA', 2, None, '___sec31'),
+              ('LLE', 2, None, '___sec32'),
+              ('Other techniques', 2, None, '___sec33')]}
 end of tocinfo -->
 
 
@@ -139,36 +150,40 @@ MathJax.Hub.Config({
       
  • 
         Contents 
         
       
    • 
@@ -203,7 +218,7 @@ MathJax.Hub.Config({
 
    [2] Department of Physics and Astronomy and National Superconducting Cyclotron Laboratory, Michigan State University
    
 
    
 
    
-
    Oct 25, 2019
     
+
    Dec 26, 2019
     
 
    
 
    
 
@@ -227,7 +242,7 @@ MathJax.Hub.Config({
   
  • 9
    • 
   
  • 10
    • 
   
  • ...
    • 
-  
  • 31
    • 
+  
  • 32
    • 
   
  • »
    • 
 
 
diff --git a/doc/pub/DimRed/html/DimRed-reveal.html b/doc/pub/DimRed/html/DimRed-reveal.html
index 291a8c15e..d75134efb 100644
--- a/doc/pub/DimRed/html/DimRed-reveal.html
+++ b/doc/pub/DimRed/html/DimRed-reveal.html
@@ -148,7 +148,7 @@ MathJax.Hub.Config({
 
    [2] Department of Physics and Astronomy and National Superconducting Cyclotron Laboratory, Michigan State University
    
 
    
 
     
    
-
    Oct 25, 2019
     
+
    Dec 26, 2019
     
 
    
 
    
 
@@ -982,14 +982,120 @@ $$
 
  •  Order the eigenvalue (and the eigenvectors accordingly) in order of decreasing eigenvalues.
    • 
 
  •  Keep only those \( l \) eigenvalues larger than a selected threshold value, discarding thus \( p-l \) features since we expect small variations in the data here.
    • 
 
+
+
+
+
    
+
    Writing our own PCA code 
    
+
+
    
+We will use a simple example first with two-dimensional data
+drawn from a multivariate normal distribution with the following mean and covariance matrix:
+
     
    
+$$
+\mu = (-1,2) \qquad \Sigma = \begin{bmatrix} 4 & 2 \\
+2 & 2
+\end{bmatrix}
+$$
+
     
    
+
+
    
+We will generate \( N = 1000 \) points \( X = \{ x_1, \ldots, x_N \} \) from
+this distribution, and store them in the \( 1000 \times 2 \) matrix \( \boldsymbol{X} \).
+
+
    
+The following Python code aids in setting up the data
+
 
    
 
+
+
    N = 1000
+mean = (-1, 2)
+cov = [[4, 2], [2, 2]]
+X = np.random.multivariate_normal(mean, cov, N)
+
    
+
    
+Make a small Python code which plots the data.
+
+
    
+Now we are going to implement the PCA algorithm. We will break it down into sub-steps and across multiple cells.
+
+
    Compute the sample mean and center the data 
    
+
+
    
+The first step of PCA is to compute the sample mean of the data and use it to center the data. Recall the sample mean is
+
     
    
+$$
+\mu_N = \frac{1}{N} \sum_{i=1}^N x_i
+$$
+
     
    
+
+and the mean-centered data \( \bar{X} = \{ \bar{x}_1, \ldots, \bar{x}_N \} \) takes the form
+
     
    
+$$
+\bar{x}_i = x_i - \mu_N
+$$
+
     
    
+
+When you are done with these steps, print out \( \mu_N \) to verify it is
+close to \( \mu \) and plot your mean centered data to verify it is
+centered at the origin! Compare your code with the functionality from Scikit-Learn discussed above.
+
+
    Compute the sample covariance 
    
+
+
    
+Now we are going to use the mean centered data to compute the sample covariance of the data. Recall it is given by:
+
     
    
+$$
+\begin{equation*}
+\Sigma_N = \frac{1}{N-1} \sum_{i=1}^N \bar{x}_i^T \bar{x}_i = \frac{1}{N-1} \sum_{i=1}^N (x_i - \mu_N)^T (x_i - \mu_N)
+\end{equation*}
+$$
+
     
    
+
+where the data points \( x_i \in \mathbb{R}^p \) (here in this example \( p = 2 \)) are column vectors and \( x^T \) is the transpose of \( x \).
+Compare the computed covariance with the answer given above.
+
+
    Diagonalize the sample covariance matrix to obtain the principal components 
    
+
+
    
+Now we are ready to solve for the principal components! To do so we
+diagonalize the sample covariance matrix \( \Sigma_N \). We can use the
+function np.linalg.eig to do so. It will return the eigenvalues and
+eigenvectors of \( \Sigma_N \). Once you have these, carry out the
+following tasks:
+
+
      
+
    •  Compute the percentage of the total variance captured by the first principal component
      • 
+
    •  Plot the mean centered data and lines along the first and second principal components
      • 
+
    •  Project the mean centered data onto the first and second principal components, and plot the projected data. What do you observe?
      • 
+
    •  Approximate the data as
      • 
+
    
+
     
    
+$$
+\begin{equation*}
+x_i \approx \tilde{x}_i := \mu_N + \langle x_i, v_0 \rangle v_0
+\end{equation*}
+$$
+
     
    
+
+where \( v_0 \) is the first principal component. What do you observe?
+
+
    
+Finally, collect all these steps and write your own PCA function and
+compare this with the functionality included in Scikit-Learn.
+Have the input be the data and have the output be the principal components and their associated eigenvalues, sorted in descending order. Can you think of a way to make it more efficient than the algorithm outlined above?
+
+
    
+Finally, try out your own PCA function with other data sets.
+
+
    
 After this we ask ourselves how do we prove the link between the maximum variance and the feature reduction.
 
    
 
 
 
    
-
    Classical PCA Theorem   
    
+
    Classical PCA Theorem   
    
 
 
    
 We assume now that we have a design matrix \( \boldsymbol{X} \) which has been centered as discussed above. For the sake of simplicity we skip the overline symbol. The matrix is defined in terms of the various column vectors \( [\boldsymbol{x}_0,\boldsymbol{x}_1,\dots, \boldsymbol{x}_{p-1}] \)
@@ -1012,7 +1118,7 @@ The PCA theorem states that minimizing the above reconstruction error correspond
 
 
 
    
-
    Proof of the PCA Theorem  
    
+
    Proof of the PCA Theorem  
    
 
 
    
 To show the PCA theorem let us start with the assumption that there is one vector \( \boldsymbol{w}_0 \) which corresponds to a solution which minimized the reconstruction error \( J \). This is an orthogonal vector. It means that we now approximate the reconstruction error in terms of \( \boldsymbol{w}_0 \) and \( \boldsymbol{z}_0 \) as
@@ -1041,7 +1147,7 @@ where the vectors on the rhs are known.
 
 
 
    
-
    PCA Proof continued 
    
+
    PCA Proof continued 
    
 
 
    
 We have now found the unknown parameters \( z_{i0} \). These correspond to the projected coordinates and we can write
@@ -1089,7 +1195,7 @@ We are almost there, we have obtained a relation between minimizing the reconstr
 
 
 
    
-
    The final step 
    
+
    The final step 
    
 
 
    
 We could trivially maximize the variance of the projection (and
@@ -1148,7 +1254,7 @@ chapter 12.4 and discussion therein.
 
 
 
    
-
    Principal Component Analysis 
    
+
    Principal Component Analysis 
    
 
    
 
 
    
@@ -1205,7 +1311,7 @@ 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
@@ -1237,7 +1343,7 @@ variance that lies along the axis of each principal component.
 
 
 
    
-
    Back to the Cancer Data 
    
+
    Back to the Cancer Data 
    
 We can now repeat the above but applied to real data, in this case our breast cancer data.
 Here we compute performance scores on the training data using logistic regression.
 
    
@@ -1278,7 +1384,7 @@ We see that our training data after the PCA decomposition has a performance simi
 
 
 
    
-
    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
@@ -1309,7 +1415,7 @@ 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
@@ -1321,7 +1427,7 @@ instances arrive).
 
 
 
    
-
    Randomized PCA 
    
+
    Randomized PCA 
    
 
 
    
 Scikit-Learn offers yet another option to perform PCA, called Randomized PCA. This is a stochastic
@@ -1335,7 +1441,7 @@ previous algorithms when \( d \) is much smaller than \( n \).
 
 
 
    
-
    Kernel PCA 
    
+
    Kernel PCA 
    
 
    
 
 
    
@@ -1361,7 +1467,7 @@ X_reduced = rbf_pca.fit_transform(X)
 
 
 
    
-
    LLE 
    
+
    LLE 
    
 
 
    
 Locally Linear Embedding (LLE) is another very powerful nonlinear dimensionality reduction
@@ -1373,7 +1479,7 @@ 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 1708ad00f..686a4c32d 100644
--- a/doc/pub/DimRed/html/DimRed-solarized.html
+++ b/doc/pub/DimRed/html/DimRed-solarized.html
@@ -109,19 +109,30 @@ div { text-align: justify; text-justify: inter-word; }
                '___sec14'),
               ('Towards the PCA theorem', 2, None, '___sec15'),
               ('The Algorithm before the Theorem', 2, None, '___sec16'),
-              ('Classical PCA Theorem', 2, None, '___sec17'),
-              ('Proof of the PCA Theorem', 2, None, '___sec18'),
-              ('PCA Proof continued', 2, None, '___sec19'),
-              ('The final step', 2, None, '___sec20'),
-              ('Principal Component Analysis', 2, None, '___sec21'),
-              ('PCA and scikit-learn', 2, None, '___sec22'),
-              ('Back to the Cancer Data', 2, None, '___sec23'),
-              ('More on the PCA', 2, None, '___sec24'),
-              ('Incremental PCA', 2, None, '___sec25'),
-              ('Randomized PCA', 2, None, '___sec26'),
-              ('Kernel PCA', 2, None, '___sec27'),
-              ('LLE', 2, None, '___sec28'),
-              ('Other techniques', 2, None, '___sec29')]}
+              ('Writing our own PCA code', 2, None, '___sec17'),
+              ('Compute the sample mean and center the data',
+               3,
+               None,
+               '___sec18'),
+              ('Compute the sample covariance', 3, None, '___sec19'),
+              ('Diagonalize the sample covariance matrix to obtain the '
+               'principal components',
+               3,
+               None,
+               '___sec20'),
+              ('Classical PCA Theorem', 2, None, '___sec21'),
+              ('Proof of the PCA Theorem', 2, None, '___sec22'),
+              ('PCA Proof continued', 2, None, '___sec23'),
+              ('The final step', 2, None, '___sec24'),
+              ('Principal Component Analysis', 2, None, '___sec25'),
+              ('PCA and scikit-learn', 2, None, '___sec26'),
+              ('Back to the Cancer Data', 2, None, '___sec27'),
+              ('More on the PCA', 2, None, '___sec28'),
+              ('Incremental PCA', 2, None, '___sec29'),
+              ('Randomized PCA', 2, None, '___sec30'),
+              ('Kernel PCA', 2, None, '___sec31'),
+              ('LLE', 2, None, '___sec32'),
+              ('Other techniques', 2, None, '___sec33')]}
 end of tocinfo -->
 
 
@@ -163,7 +174,7 @@ MathJax.Hub.Config({
 
    [2] Department of Physics and Astronomy and National Superconducting Cyclotron Laboratory, Michigan State University
    
 
    
 
    
-
    Oct 25, 2019
     
+
    Dec 26, 2019
     
 
    
 
    
 









    
@@ -950,12 +961,109 @@ $$
 
  •  Keep only those \( l \) eigenvalues larger than a selected threshold value, discarding thus \( p-l \) features since we expect small variations in the data here.
    • 
 
 
+









    
+
+
    Writing our own PCA code 
    
+
+
    
+We will use a simple example first with two-dimensional data
+drawn from a multivariate normal distribution with the following mean and covariance matrix:
+$$
+\mu = (-1,2) \qquad \Sigma = \begin{bmatrix} 4 & 2 \\
+2 & 2
+\end{bmatrix}
+$$
+
+
    
+We will generate \( N = 1000 \) points \( X = \{ x_1, \ldots, x_N \} \) from
+this distribution, and store them in the \( 1000 \times 2 \) matrix \( \boldsymbol{X} \).
+
+
    
+The following Python code aids in setting up the data
+
+
    
+
+
+
    N = 1000
+mean = (-1, 2)
+cov = [[4, 2], [2, 2]]
+X = np.random.multivariate_normal(mean, cov, N)
+
    
+
    
+Make a small Python code which plots the data.
+
+
    
+Now we are going to implement the PCA algorithm. We will break it down into sub-steps and across multiple cells.
+
+
    Compute the sample mean and center the data 
    
+
+
    
+The first step of PCA is to compute the sample mean of the data and use it to center the data. Recall the sample mean is
+$$
+\mu_N = \frac{1}{N} \sum_{i=1}^N x_i
+$$
+
+and the mean-centered data \( \bar{X} = \{ \bar{x}_1, \ldots, \bar{x}_N \} \) takes the form
+$$
+\bar{x}_i = x_i - \mu_N
+$$
+
+When you are done with these steps, print out \( \mu_N \) to verify it is
+close to \( \mu \) and plot your mean centered data to verify it is
+centered at the origin! Compare your code with the functionality from Scikit-Learn discussed above.
+
+
    Compute the sample covariance 
    
+
+
    
+Now we are going to use the mean centered data to compute the sample covariance of the data. Recall it is given by:
+$$
+\begin{equation*}
+\Sigma_N = \frac{1}{N-1} \sum_{i=1}^N \bar{x}_i^T \bar{x}_i = \frac{1}{N-1} \sum_{i=1}^N (x_i - \mu_N)^T (x_i - \mu_N)
+\end{equation*}
+$$
+
+where the data points \( x_i \in \mathbb{R}^p \) (here in this example \( p = 2 \)) are column vectors and \( x^T \) is the transpose of \( x \).
+Compare the computed covariance with the answer given above.
+
+
    Diagonalize the sample covariance matrix to obtain the principal components 
    
+
+
    
+Now we are ready to solve for the principal components! To do so we
+diagonalize the sample covariance matrix \( \Sigma_N \). We can use the
+function np.linalg.eig to do so. It will return the eigenvalues and
+eigenvectors of \( \Sigma_N \). Once you have these, carry out the
+following tasks:
+
+
      
+
    •  Compute the percentage of the total variance captured by the first principal component
      • 
+
    •  Plot the mean centered data and lines along the first and second principal components
      • 
+
    •  Project the mean centered data onto the first and second principal components, and plot the projected data. What do you observe?
      • 
+
    •  Approximate the data as
      • 
+
    
+
+$$
+\begin{equation*}
+x_i \approx \tilde{x}_i := \mu_N + \langle x_i, v_0 \rangle v_0
+\end{equation*}
+$$
+
+where \( v_0 \) is the first principal component. What do you observe?
+
+
    
+Finally, collect all these steps and write your own PCA function and
+compare this with the functionality included in Scikit-Learn.
+Have the input be the data and have the output be the principal components and their associated eigenvalues, sorted in descending order. Can you think of a way to make it more efficient than the algorithm outlined above?
+
+
    
+Finally, try out your own PCA function with other data sets.
+
+
    
 After this we ask ourselves how do we prove the link between the maximum variance and the feature reduction.
 
 
    
 









    
 
-
    Classical PCA Theorem   
    
+
    Classical PCA Theorem   
    
 
 
    
 We assume now that we have a design matrix \( \boldsymbol{X} \) which has been centered as discussed above. For the sake of simplicity we skip the overline symbol. The matrix is defined in terms of the various column vectors \( [\boldsymbol{x}_0,\boldsymbol{x}_1,\dots, \boldsymbol{x}_{p-1}] \)
@@ -976,7 +1084,7 @@ The PCA theorem states that minimizing the above reconstruction error correspond
 
    
 









    
 
-
    Proof of the PCA Theorem  
    
+
    Proof of the PCA Theorem  
    
 
 
    
 To show the PCA theorem let us start with the assumption that there is one vector \( \boldsymbol{w}_0 \) which corresponds to a solution which minimized the reconstruction error \( J \). This is an orthogonal vector. It means that we now approximate the reconstruction error in terms of \( \boldsymbol{w}_0 \) and \( \boldsymbol{z}_0 \) as
@@ -999,7 +1107,7 @@ where the vectors on the rhs are known.
 
    
 









    
 
-
    PCA Proof continued 
    
+
    PCA Proof continued 
    
 
 
    
 We have now found the unknown parameters \( z_{i0} \). These correspond to the projected coordinates and we can write
@@ -1037,7 +1145,7 @@ We are almost there, we have obtained a relation between minimizing the reconstr
 
    
 









    
 
-
    The final step 
    
+
    The final step 
    
 
 
    
 We could trivially maximize the variance of the projection (and
@@ -1088,7 +1196,7 @@ chapter 12.4 and discussion therein.
 
    
 









    
 
-
    Principal Component Analysis 
    
+
    Principal Component Analysis 
    
 
    
 
 
    
@@ -1144,7 +1252,7 @@ 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
@@ -1176,7 +1284,7 @@ variance that lies along the axis of each principal component.
 
    
 









    
 
-
    Back to the Cancer Data 
    
+
    Back to the Cancer Data 
    
 We can now repeat the above but applied to real data, in this case our breast cancer data.
 Here we compute performance scores on the training data using logistic regression.
 
    
@@ -1217,7 +1325,7 @@ We see that our training data after the PCA decomposition has a performance simi
 
    
 









    
 
-
    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
@@ -1247,7 +1355,7 @@ 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
@@ -1259,7 +1367,7 @@ instances arrive).
 
    
 









    
 
-
    Randomized PCA 
    
+
    Randomized PCA 
    
 
 
    
 Scikit-Learn offers yet another option to perform PCA, called Randomized PCA. This is a stochastic
@@ -1274,7 +1382,7 @@ previous algorithms when \( d \) is much smaller than \( n \).
 
    
 









    
 
-
    Kernel PCA 
    
+
    Kernel PCA 
    
 
    
 
 
    
@@ -1303,7 +1411,7 @@ X_reduced = rbf_pca.fit_transform(X)
 
    
 









    
 
-
    LLE 
    
+
    LLE 
    
 
 
    
 Locally Linear Embedding (LLE) is another very powerful nonlinear dimensionality reduction
@@ -1315,7 +1423,7 @@ 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 08bc92a30..7b1ba18c7 100644
--- a/doc/pub/DimRed/html/DimRed.html
+++ b/doc/pub/DimRed/html/DimRed.html
@@ -114,19 +114,30 @@ div { text-align: justify; text-justify: inter-word; }
                '___sec14'),
               ('Towards the PCA theorem', 2, None, '___sec15'),
               ('The Algorithm before the Theorem', 2, None, '___sec16'),
-              ('Classical PCA Theorem', 2, None, '___sec17'),
-              ('Proof of the PCA Theorem', 2, None, '___sec18'),
-              ('PCA Proof continued', 2, None, '___sec19'),
-              ('The final step', 2, None, '___sec20'),
-              ('Principal Component Analysis', 2, None, '___sec21'),
-              ('PCA and scikit-learn', 2, None, '___sec22'),
-              ('Back to the Cancer Data', 2, None, '___sec23'),
-              ('More on the PCA', 2, None, '___sec24'),
-              ('Incremental PCA', 2, None, '___sec25'),
-              ('Randomized PCA', 2, None, '___sec26'),
-              ('Kernel PCA', 2, None, '___sec27'),
-              ('LLE', 2, None, '___sec28'),
-              ('Other techniques', 2, None, '___sec29')]}
+              ('Writing our own PCA code', 2, None, '___sec17'),
+              ('Compute the sample mean and center the data',
+               3,
+               None,
+               '___sec18'),
+              ('Compute the sample covariance', 3, None, '___sec19'),
+              ('Diagonalize the sample covariance matrix to obtain the '
+               'principal components',
+               3,
+               None,
+               '___sec20'),
+              ('Classical PCA Theorem', 2, None, '___sec21'),
+              ('Proof of the PCA Theorem', 2, None, '___sec22'),
+              ('PCA Proof continued', 2, None, '___sec23'),
+              ('The final step', 2, None, '___sec24'),
+              ('Principal Component Analysis', 2, None, '___sec25'),
+              ('PCA and scikit-learn', 2, None, '___sec26'),
+              ('Back to the Cancer Data', 2, None, '___sec27'),
+              ('More on the PCA', 2, None, '___sec28'),
+              ('Incremental PCA', 2, None, '___sec29'),
+              ('Randomized PCA', 2, None, '___sec30'),
+              ('Kernel PCA', 2, None, '___sec31'),
+              ('LLE', 2, None, '___sec32'),
+              ('Other techniques', 2, None, '___sec33')]}
 end of tocinfo -->
 
 
@@ -168,7 +179,7 @@ MathJax.Hub.Config({
 
    [2] Department of Physics and Astronomy and National Superconducting Cyclotron Laboratory, Michigan State University
    
 
    
 
    
-
    Oct 25, 2019
     
+
    Dec 26, 2019
     
 
    
 
    
 









    
@@ -955,12 +966,109 @@ $$
 
  •  Keep only those \( l \) eigenvalues larger than a selected threshold value, discarding thus \( p-l \) features since we expect small variations in the data here.
    • 
 
 
+









    
+
+
    Writing our own PCA code 
    
+
+
    
+We will use a simple example first with two-dimensional data
+drawn from a multivariate normal distribution with the following mean and covariance matrix:
+$$
+\mu = (-1,2) \qquad \Sigma = \begin{bmatrix} 4 & 2 \\
+2 & 2
+\end{bmatrix}
+$$
+
+
    
+We will generate \( N = 1000 \) points \( X = \{ x_1, \ldots, x_N \} \) from
+this distribution, and store them in the \( 1000 \times 2 \) matrix \( \boldsymbol{X} \).
+
+
    
+The following Python code aids in setting up the data
+
+
    
+
+
+
    N = 1000
+mean = (-1, 2)
+cov = [[4, 2], [2, 2]]
+X = np.random.multivariate_normal(mean, cov, N)
+
    
+
    
+Make a small Python code which plots the data.
+
+
    
+Now we are going to implement the PCA algorithm. We will break it down into sub-steps and across multiple cells.
+
+
    Compute the sample mean and center the data 
    
+
+
    
+The first step of PCA is to compute the sample mean of the data and use it to center the data. Recall the sample mean is
+$$
+\mu_N = \frac{1}{N} \sum_{i=1}^N x_i
+$$
+
+and the mean-centered data \( \bar{X} = \{ \bar{x}_1, \ldots, \bar{x}_N \} \) takes the form
+$$
+\bar{x}_i = x_i - \mu_N
+$$
+
+When you are done with these steps, print out \( \mu_N \) to verify it is
+close to \( \mu \) and plot your mean centered data to verify it is
+centered at the origin! Compare your code with the functionality from Scikit-Learn discussed above.
+
+
    Compute the sample covariance 
    
+
+
    
+Now we are going to use the mean centered data to compute the sample covariance of the data. Recall it is given by:
+$$
+\begin{equation*}
+\Sigma_N = \frac{1}{N-1} \sum_{i=1}^N \bar{x}_i^T \bar{x}_i = \frac{1}{N-1} \sum_{i=1}^N (x_i - \mu_N)^T (x_i - \mu_N)
+\end{equation*}
+$$
+
+where the data points \( x_i \in \mathbb{R}^p \) (here in this example \( p = 2 \)) are column vectors and \( x^T \) is the transpose of \( x \).
+Compare the computed covariance with the answer given above.
+
+
    Diagonalize the sample covariance matrix to obtain the principal components 
    
+
+
    
+Now we are ready to solve for the principal components! To do so we
+diagonalize the sample covariance matrix \( \Sigma_N \). We can use the
+function np.linalg.eig to do so. It will return the eigenvalues and
+eigenvectors of \( \Sigma_N \). Once you have these, carry out the
+following tasks:
+
+
      
+
    •  Compute the percentage of the total variance captured by the first principal component
      • 
+
    •  Plot the mean centered data and lines along the first and second principal components
      • 
+
    •  Project the mean centered data onto the first and second principal components, and plot the projected data. What do you observe?
      • 
+
    •  Approximate the data as
      • 
+
    
+
+$$
+\begin{equation*}
+x_i \approx \tilde{x}_i := \mu_N + \langle x_i, v_0 \rangle v_0
+\end{equation*}
+$$
+
+where \( v_0 \) is the first principal component. What do you observe?
+
+
    
+Finally, collect all these steps and write your own PCA function and
+compare this with the functionality included in Scikit-Learn.
+Have the input be the data and have the output be the principal components and their associated eigenvalues, sorted in descending order. Can you think of a way to make it more efficient than the algorithm outlined above?
+
+
    
+Finally, try out your own PCA function with other data sets.
+
+
    
 After this we ask ourselves how do we prove the link between the maximum variance and the feature reduction.
 
 
    
 









    
 
-
    Classical PCA Theorem   
    
+
    Classical PCA Theorem   
    
 
 
    
 We assume now that we have a design matrix \( \boldsymbol{X} \) which has been centered as discussed above. For the sake of simplicity we skip the overline symbol. The matrix is defined in terms of the various column vectors \( [\boldsymbol{x}_0,\boldsymbol{x}_1,\dots, \boldsymbol{x}_{p-1}] \)
@@ -981,7 +1089,7 @@ The PCA theorem states that minimizing the above reconstruction error correspond
 
    
 









    
 
-
    Proof of the PCA Theorem  
    
+
    Proof of the PCA Theorem  
    
 
 
    
 To show the PCA theorem let us start with the assumption that there is one vector \( \boldsymbol{w}_0 \) which corresponds to a solution which minimized the reconstruction error \( J \). This is an orthogonal vector. It means that we now approximate the reconstruction error in terms of \( \boldsymbol{w}_0 \) and \( \boldsymbol{z}_0 \) as
@@ -1004,7 +1112,7 @@ where the vectors on the rhs are known.
 
    
 









    
 
-
    PCA Proof continued 
    
+
    PCA Proof continued 
    
 
 
    
 We have now found the unknown parameters \( z_{i0} \). These correspond to the projected coordinates and we can write
@@ -1042,7 +1150,7 @@ We are almost there, we have obtained a relation between minimizing the reconstr
 
    
 









    
 
-
    The final step 
    
+
    The final step 
    
 
 
    
 We could trivially maximize the variance of the projection (and
@@ -1093,7 +1201,7 @@ chapter 12.4 and discussion therein.
 
    
 









    
 
-
    Principal Component Analysis 
    
+
    Principal Component Analysis 
    
 
    
 
 
    
@@ -1149,7 +1257,7 @@ 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
@@ -1181,7 +1289,7 @@ variance that lies along the axis of each principal component.
 
    
 









    
 
-
    Back to the Cancer Data 
    
+
    Back to the Cancer Data 
    
 We can now repeat the above but applied to real data, in this case our breast cancer data.
 Here we compute performance scores on the training data using logistic regression.
 
    
@@ -1222,7 +1330,7 @@ We see that our training data after the PCA decomposition has a performance simi
 
    
 









    
 
-
    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
@@ -1252,7 +1360,7 @@ 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
@@ -1264,7 +1372,7 @@ instances arrive).
 
    
 









    
 
-
    Randomized PCA 
    
+
    Randomized PCA 
    
 
 
    
 Scikit-Learn offers yet another option to perform PCA, called Randomized PCA. This is a stochastic
@@ -1279,7 +1387,7 @@ previous algorithms when \( d \) is much smaller than \( n \).
 
    
 









    
 
-
    Kernel PCA 
    
+
    Kernel PCA 
    
 
    
 
 
    
@@ -1308,7 +1416,7 @@ X_reduced = rbf_pcaLLE 
+
    LLE 
    
 
 
    
 Locally Linear Embedding (LLE) is another very powerful nonlinear dimensionality reduction
@@ -1320,7 +1428,7 @@ 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 700af1f72..82b60b04e 100644
--- a/doc/pub/DimRed/ipynb/DimRed.ipynb
+++ b/doc/pub/DimRed/ipynb/DimRed.ipynb
@@ -10,7 +10,7 @@
     "  \n",
     "**Morten Hjorth-Jensen**, Department of Physics, University of Oslo and Department of Physics and Astronomy and National Superconducting Cyclotron Laboratory, Michigan State University\n",
     "\n",
-    "Date: **Oct 25, 2019**\n",
+    "Date: **Dec 26, 2019**\n",
     "\n",
     "Copyright 1999-2019, Morten Hjorth-Jensen. Released under CC Attribution-NonCommercial 4.0 license\n",
     "\n",
@@ -1049,6 +1049,156 @@
     "\n",
     "* Keep only those $l$ eigenvalues larger than a selected threshold value, discarding thus $p-l$ features since we expect small variations in the data here.\n",
     "\n",
+    "## Writing our own PCA code\n",
+    "\n",
+    "We will use a simple example first with two-dimensional data\n",
+    "drawn from a multivariate normal distribution with the following mean and covariance matrix:"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "$$\n",
+    "\\mu = (-1,2) \\qquad \\Sigma = \\begin{bmatrix} 4 & 2 \\\\\n",
+    "2 & 2\n",
+    "\\end{bmatrix}\n",
+    "$$"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We will generate $N = 1000$ points $X = \\{ x_1, \\ldots, x_N \\}$ from\n",
+    "this distribution, and store them in the $1000 \\times 2$ matrix $\\boldsymbol{X}$.\n",
+    "\n",
+    "The following Python code aids in setting up the data"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 11,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": [
+    "N = 1000\n",
+    "mean = (-1, 2)\n",
+    "cov = [[4, 2], [2, 2]]\n",
+    "X = np.random.multivariate_normal(mean, cov, N)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Make a small Python code which plots the data.\n",
+    "\n",
+    "Now we are going to implement the PCA algorithm. We will break it down into sub-steps and across multiple cells.\n",
+    "\n",
+    "### Compute the sample mean and center the data\n",
+    "\n",
+    "The first step of PCA is to compute the sample mean of the data and use it to center the data. Recall the sample mean is"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "$$\n",
+    "\\mu_N = \\frac{1}{N} \\sum_{i=1}^N x_i\n",
+    "$$"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "and the mean-centered data $\\bar{X} = \\{ \\bar{x}_1, \\ldots, \\bar{x}_N \\}$ takes the form"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "$$\n",
+    "\\bar{x}_i = x_i - \\mu_N\n",
+    "$$"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "When you are done with these steps, print out $\\mu_N$ to verify it is\n",
+    "close to $\\mu$ and plot your mean centered data to verify it is\n",
+    "centered at the origin! Compare your code with the functionality from **Scikit-Learn** discussed above.\n",
+    "\n",
+    "\n",
+    "### Compute the sample covariance\n",
+    "\n",
+    "Now we are going to use the mean centered data to compute the sample covariance of the data. Recall it is given by:"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "$$\n",
+    "\\Sigma_N = \\frac{1}{N-1} \\sum_{i=1}^N \\bar{x}_i^T \\bar{x}_i = \\frac{1}{N-1} \\sum_{i=1}^N (x_i - \\mu_N)^T (x_i - \\mu_N)\n",
+    "$$"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "where the data points $x_i \\in \\mathbb{R}^p$ (here in this example $p = 2$) are column vectors and $x^T$ is the transpose of $x$.\n",
+    "Compare the computed covariance with the answer given above.\n",
+    "\n",
+    "\n",
+    "### Diagonalize the sample covariance matrix to obtain the principal components\n",
+    "\n",
+    "Now we are ready to solve for the principal components! To do so we\n",
+    "diagonalize the sample covariance matrix $\\Sigma_N$. We can use the\n",
+    "function **np.linalg.eig** to do so. It will return the eigenvalues and\n",
+    "eigenvectors of $\\Sigma_N$. Once you have these, carry out the\n",
+    "following tasks:\n",
+    "\n",
+    "* Compute the percentage of the total variance captured by the first principal component\n",
+    "\n",
+    "* Plot the mean centered data and lines along the first and second principal components\n",
+    "\n",
+    "* Project the mean centered data onto the first and second principal components, and plot the projected data. What do you observe?\n",
+    "\n",
+    "* Approximate the data as"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "$$\n",
+    "x_i \\approx \\tilde{x}_i := \\mu_N + \\langle x_i, v_0 \\rangle v_0\n",
+    "$$"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "where $v_0$ is the first principal component. What do you observe?\n",
+    "\n",
+    "Finally, collect all these steps and write your own PCA function and\n",
+    "compare this with the functionality included in **Scikit-Learn**.\n",
+    "Have the input be the data and have the output be the principal components and their associated eigenvalues, sorted in descending order. Can you think of a way to make it more efficient than the algorithm outlined above?\n",
+    "\n",
+    "Finally, try out your own PCA function with other data sets.\n",
+    "\n",
+    "\n",
+    "\n",
     "After this we ask ourselves how do we prove the link between the maximum variance and the feature reduction.\n",
     "\n",
     "## Classical PCA Theorem\n",
@@ -1320,7 +1470,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 11,
+   "execution_count": 12,
    "metadata": {
     "collapsed": false
    },
@@ -1367,7 +1517,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 13,
    "metadata": {
     "collapsed": false
    },
@@ -1391,7 +1541,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": 14,
    "metadata": {
     "collapsed": false
    },
@@ -1415,7 +1565,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": 15,
    "metadata": {
     "collapsed": false
    },
@@ -1439,7 +1589,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 15,
+   "execution_count": 16,
    "metadata": {
     "collapsed": false
    },
@@ -1493,7 +1643,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 16,
+   "execution_count": 17,
    "metadata": {
     "collapsed": false
    },
@@ -1516,7 +1666,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 17,
+   "execution_count": 18,
    "metadata": {
     "collapsed": false
    },
@@ -1563,7 +1713,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 18,
+   "execution_count": 19,
    "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 cb7d0f3ef45648511fd83d24751f006e56622ffb..55bb01bf7c59b9a7601fc78711ecc4c1db61e57c 100644
GIT binary patch
literal 190
zcmV;v073sBiwFP+h6G*!1MSaC3c@fD2H>uHia9}h*^@rlzEoUiSbg;2uXp^Y)qs`5RgzB=KiWZ;$FoKObYQDqm8n5*rPmfqs46m}GUoN>+b+m;B)S%aNEKe>3{Ep6m6-6j}6ssJrK>on%xG
zTOmHx&DyAF*A7+~*S2;;JKX=Y{@2a#E0x76*Q<>#VpDmUUz~k#BUY=ZvD1XLFJ%7O
z_I8-s5hu1YNq-huNfPTS_reMh_J`dr&5X6QY+se+Qggn&`KGW3b!&j^iy$gn)dN$5mlUwvp0t3IQh0!*=DDRY%JVD1{5~f}hXG
z(Xq+%ut3X#EMH|g%*PtMAcRSBJPc3T>&OszI1E|bdp%s$j^h6&fY
zie@O5nrOU}@0T4h;m2mW5Y9h`_K|qjO)xf1f1d96$JC8yFPN)XmGB0cG(Xkk|8hTJ
zNKm9YdOyu>9L9dFOSz&Bb0+#+&m1kn{<=J@H>^=J4jdgEVVJpkZn}}g$hPbH77NR0
zboD&C=XIGy7qakE+t4|?_P!px(7B%NS&Nkto}fFxD5+T?kxPO^d3P0J#hcb14U8YqdDerHYu*65csLj^2gRb$;D69
zPGZ17f23VNNZsO~1ZNA8cU
z*q|6e4$pHkE)C7>WM+q})NV`hMZL6j43N
zj;G*DqTwTANjc?dGB!B_}e>Xr(1aIk=iURsop+(V9&Bi;a03e;cv(8yxiRk`Z
zns^~a5~N$9o5f9s5d#=jZjWjgW@m~4Ad8A{*9@PJ>k@u;^D+QW;3ka7IY_&9e}lT)
zJkh?tg
z0eIm5EpKJ;NJM2?uul*kHIzUqY7qnVZO5u#h!QU~&{+(hi3Si5(t|MHhbteA7;olU
z+$xyp)jcdK#Zk0XR`@-gPm^&#sJt{}TE>M!DFqu+EF;_%^=9nXD+fL0g
zO){9q-Y;NF;43x*(T~$Cf0VcO!_w6*$nX(DsX`47M+i^3<7Vcp*Ao?+EZEEn!VK+y
z77Jn1Pb9^A&S-+ic4*oMu8SBpWAAS>Sh1||>qBRU(U+_D6p)zyI|+4z?PM4xa$k2r
zT#qcTsk`oZk1Q|YQI}}8(6;fNuYu$>BV-wUVEB*{Y`0tQ{t{r3e8Q9xEV3waxQ1<;*`n{WGLYQIX{$W+@uAY{%3E8qz2m-c}lDOg#~Hw1=XAVC_6S%W#=|@cLLrO2M-BjKKI@N04x1
zT$VrRvd6`Fq@tU>e=g8e;ffa(S{qacYUfD%Uo2`IUnV(nK@y2c5Dk{m6P%sbjxtH=
zvAcM^FSsr_WEXFSwMN?dBxA7(j-jADw!dF5dZ7}DflydgbVH)wNF}YRSW`<%It}}i
zP>y|DnF=D~3rHKR{uXU{pPW+wIA`_fBnE@A+rx!qsOuVJ(A
zJcDACG+iNdccFayB{3_
zviGtM9#4gqD4IQtj51>ppQG5hOrwA&$8TkoqL)w6f73I<2&Q~(+HdW=;U@eI=t#XK
zG|Fm1zb~L*mENXi@6yfwDaBm|fA^kZ5GRiBiRM|}RAJ|Ix(~CxXUX%*P^1BC
zoS+6Vf9eJ;CB=5bcK{GHBaQ78pSyiZ_>8Jlc^V1v9a#>z%TKr9Q9EqWu74%_
zx$_jDZu7LK=2!STdhdhw=bY89aqTl(+m7Cpu+J6e$vGYm66lZc+%gi5-DWxnA
zC+76xpW6H3@{Zz*+jn92_3Ya3zV*to(!Z@}>QDe6M^YJFFR+NVuI3tgD>nZY&f_+V
ze{L|J{7Q29vDB*~t_py}ahRsSiy~X0XUy@?7Cj^~&{!VUZ#4J8nA;g{b4(cy1JDW3vhhBLHi-Yt1J1wUsIif
z-ER5uy7@c~S7YB>*R+gpSdzwxxC{*y6#C&)RPEY^&O9;Chnsi*15U`ATeCU|1wjNd
zH!(7kVOJ`DOOM;g5x&o_U~Uv(wfjxdh7ZXi4v@!Yvpc{pu-*eLaYn>S;)J9;_P@{9
z-PIJSAvIpd2o^c9MRr%$tE;}M7EkUIp8Vy8Up)VGd-L+G5=p@dD{XRnmk6D+OvuEV
zj7epZ+imi5`i_fi^~>#FhjlOCifpuqvm(cPo`^($n4H4%lEX}LeGL`t9Dal^~wc78#Q1z1&W`lju|mvXad+HHNj-+#hV5nk5E
z@NCX|H+SK=tu}3i?aFTLqypC$ylcZ`9#{v})mkYv-rg9;Z`H92olF$Dc0-k{1)lOV
z^(YvB7v_=~X_zvk$iG$91K5H5&2j%JjQSl!X}Zc0e)$%BkLVFCsTAW%O=<1Jf<*?T
zDwvVkbkSG~I%vUhk*E8zy?1&J9F@l~D#N3z_OMsA4O9DFTXnl;zYQM;eZC04w{Ycbq8L+o2-7E)%b3mzEFhC{K1Y%>DRi8)
zpvuU-)RL)uyzO08_NUhC4%~P3aZ`oiqth*otBpysI;UyY+xk!)JLs__`d(>%+7@Kn
z(oOSH;muv^PTL{;2&x@NKgf{TJTcIJ7pxwHIJmt|`@V+JvS2jFoU-Y$X%Bu=jN$fu
zTfaX!gN8T6Ek-^Dy1`wu-#0G&JyKk<>7goVz7Lsh0ELGW8g>iL<))U{;p3}oCj^nt
z!mka_SP{5?$EBqeYP14bB8%?-#<^+k!}iQ^5}K((%$z4RIdhpF&YYOx5Pq0{4@6$i
zP|!@vJWi052e6sj(FO+J5O;dKtZ+^T9=XmUhf>ebhe1uAtf3d0r&nv?=*!>)i1hy4
z5Ch|n7h6(Bm?j2k=1&LL(X}Ws>zwcCe*&2
z5qr#nRMy9i$s%w3EER)Y7bTE?8D~bT>2eq1^0cXaQt7WnZ^f}ZzhL>NXINGOp|`l?
zc*aTNd?{I>E;hM)&L2P01e;j62SUB~7h7mJ1=Dz3acgahWHjkD9LHwqEBCPBk;TXB
z@G|aJjCPaAY)VbG`*t?98Iz5-yMM-JDi;W7m-=XZ0EEx=ySfWLN8t;9!O=&s7t^Fe
zVNu&0=Zpdig)5bIKypxE>URyP
z#^qIf+|ul_-gaRWb=OsYeLvvW(?g)pN9qGKxCmYUD1VJ01k+0*MbHDoD=a+j>Wy2w
z3u_;A+3`A~ciZk_*@jn5-b9ebXEsP99d|QP~)D5eQ*S
zI3bT)=Z^Ssg`?Yh+&*467-^=?%ZA9HeMC?N@;gG0mzrER_pr-hCy%`F+ez+5nJ>cJ
z8)u9K^rUhG?TjFQpbs#b2UThTmy43Rmpd8^l44AfVy@q7JdKbYJ_RM^2f^sW`N
z)q
z*Cs#$=gwPyw&To5ZlmEUu3d3ld!I!0;(oTio>u1?@6y^bM-5UyA|sx9Pd?!S#a-?m
z!^~M+C0NeY8T8MfC?GU3WWl>x6m5Nj;VL_vS*8Bx*y<^>01#Owzlhu7Dz{6}%svOr
zCSateMBT+SM`b_8cPK1p1X^2|7=hKDg5o)+aB}H?-zHZ*vnN7RZpL!GnQy0z9OpM}
zHc_HsMveDc=-x~~-}51LYPHLZI7wf4D>D$0f(o=}U
zUm0wX~?_kU~O13eXyQXJoN9{Gx1)(Z-!K
zio+oY$3`1QD(Jzt=v?oP^k7L3mh@l)`kK-xd!n$T?j1nteSQ4HYHhUj5YPF|CDR;=A{r9SD
zk298ma%C}eW>61S%a~VAFfYmz;M)%?t79XK
zE!4Zk0COx93eQG97Q#F!M0jW#R@B}O$C#fB5yGkX7I3ACxKZ(##KR9*f4%6BKt6(h
z+%dXGkq>Q2Dl3X&goWM99SsHzVC-6f8MWAlRv_McWE^i-B5{CPYj^~;{~l^N85YB|+^14t=up6Q5Aa+H{Z$2b!F_W_S5wopvq#TXy|
zm&i%uB$767W5WWcG_%)0N8zF(vVi(};LThIIWvXT+5c3r?0^L{1`B-^4~(=v%LVoC
z&jr=zxnO?X1v$$vJfBFyc>1_sK&4
zQfNfF2!p8ZSddtt!qLt*$4i!dVPv4&h5l?L42-UFJ0<~o}XbV6fk{_-N`
z=O1rgeDYeURW9;GC99jemDF0KNw!LCDwMXXo89V%=ucH~-_*sSdS9+LDpygz=SEk5
z98U*sZ=1DN(J$-7L`7Q__15*v=vee^^^ONN^lWbXqN-<;o!YA2R;S|N=R0oJKizx<
zoNlBPxn*un>kX{a)Mbt9^_@VXPhQJ(x(yLa0|VogT*=Hrzg{KA3YD#Hjxfj9&03|=
z1J0De1AaO(_=B3C5dnRv0&%cZS|M6
zwd1j1KOc*eozkJ~JO|pu!m5yiALBUTQTZwnd8#uvY8aA=%)%QjMW!W-d-;fehGndU
zjoNS>vEHOMj^3;@={LMBYU2IeaS!iUMR(_V+gDA^oCPX95vNmaC529Dk@_SK6^QU&
zl-2!pCS?vAQ)w{VX@=2sh6c=`0j^DEqQCX@;-NYm=v7PbYOSM3++B0VKPK82yjQYG?ZnlUFX`-xE<^bZ@&)k)OfueexPKbFvpBa${B-6$@p^$^010bT%Y^*nTYX%x}(S
zA^}l>2!!+^dj+T3sX9_sUkg{)j=Lew=bR&U1!^XW+
zh0bkp0(~lX^zRmDdkm(qumx?}uBSgI7(RHX;AlMeDL8yZ%wIaLYtq}~&b@9xW0d^}
z@u@in5OVS)Ar`(lG|q&7Vt=%GbS#USy1e*GCjji(6@5WeY&LNOcI`|F`wop$hG8~n
zkaR#Qyv#7JE4Phn!@$#Zk(mU-%u^!k{vR^CU_m7!wO(dz<5>7N_z2>kWtVx`LD!bq
zW7)yGQFaZzYN&q@{GdpqeBt$loKQh2_Te&vJT}X+1Dngd@CK28y(A;%HNaj-3#JBs
znrxDGpxqA8@fs7o*%v*Hbv`+CF5oe5yRs$U@`vC0d}v5a#9y9}QCR0)1%b8-~N~Ou^~x7x>W-
z56!RJd^nAk0W+ViH!?v4K(xtZMpV8um@!`9+z1HTt7n(RXwKX(tp(-?yIEK((D98@
zM#v0B4v&!;z&0=(ghXZ-!82LF+!+{!Vc71um
z9mR@0AM`!6RnM=PpO*d+xT)#SH?O{=#sGh55kCv=x}c+(Cl(0Dj0W9F*$MZ1C)Okf
z=+y6_mHKFZ30;1ENH%(8;E~>gln>{fbHhMXT|?hG&V+@vf;GS0R&T3*^Y^l7>t|8~
zz@;|M!{Km*7O>J^ixC|YC&ag7@Z=yO^xeASpAbAH6s5rX@@bNzHnAa@npiu0nhvXB
z4<#Y7gQIRa?7VeRYED6l33}-QCH7n@W=CUqL4$*TtCst^j-ubTDlq!Z71D1(LI6ov
zNR#xzrl5b2V4z~E1hSQuLa5SscJ8?=J3t6K0?+WjX?JDoOeKq66-+c*h{C;9^;;&s
zqh~=6msb}IO&zIAbq%DRZ?U)JrF-seyr#~!qjSIvF`q~Yww`P&k4xDP5V|oVOn{WG
z0TbGPAK8r0J>o&e3k4lcGbS2j+ddC))rN$8&0xu%6U3pcnEV>I4?M
z@<2Q8=-r;UWhB^8=QtQ^8f!U~RDcarMA9x5tj?YZ(Ov;Z`Oug);%@#3t}
zzICDmgju&SPJB4)LAfw6oaTFFfq4-90Z4mxLgtc?k}E14DZl|D?_da
zd#mH=P;JL31@9gPI(YxKIt3G%VL{k`Rf|r7deQzKkNAq&Iyi0;)Q!O`^=x}8S0Xbk
zL0|XO+A?CnvqY%Ej>wYNg(rl1Krcroa(mI%EBR0XsG2iV0Hin{YyT2Cr>f8;^qq34
z(_AFhhFtz|O_?$!jMd?QcMC8sfpBA&U|awPF>;5<_i|~VQP~`2UtdZtFBM{cVn*4Y
z;v=C)>K{vd00@vezXtb_ktAT`#Nb>709VH8n8d{K;+B*eXyh}8Xe^$|Ox1o`Fz|3t
zhIQOvOj&GEbM_Z*7zk7ri!5U^9b?t5I+k@eR9ORo0X|Ffd2%?*u=JQcDkcchUSZ-v
z~IQ9}I?
zHCBrTF;U~w6B=m-jRgDAX}Cj&Gjo3s974u>9PeZ;uIXb4J5;(u8E#^KS}aa=eS7wp
z%avWx$x_>iOQ9b&H*wrC{-prvw=L~ThiCkL65_`g~#yO>OmXK3RE1q{jrZMDXQ7DwEHXw9~QZ+&w
zeNCzq`2w+qjQgpSi#i|Ae;}`mf|=8q%P30X0!IZaPrcy+F4#<{K6@VIV?c}*xT$;+
z5HAEMZ7OsIK0AiklBoYhfa6rW>W5>H>iUhKOeLs87VhVZE0IWM;JD)1fH#Pn;^_U-
zJ!0l;a)G6%B&=e8aEV}D!3r<&RyUm%78kiwknLDCJD&H8rin)IFh{y0Y(ok&`R-t*uCu2`7DX-Nh+O7g@oWYc5VDO2l~*4*2#N_kiVSJ!1^(gT|M;
z@bio*{D=QTK>$AtnLtncfMooXryIW!d7`{4!8b|CUow@)Q`K?_3$MnTvD8bIGej5e
z91bq)U@U3;cUh?H*0Msu1kI_hjx?K$_*06FQ&rG^8`4&KSJqpqMeq0w7g;qLJb@k^
zj!L=mf(IZ1D}?vn%C(#G%1w_
zUjqz@9@Wp{F=g|G^*H##As(mYdKxG={qIgiJ!fPoos;hS0PQc(xC}l2U<*Gop>MV{
z7b>R}?7?E~rN!a}Chm*JlEtD0i{L|qmDUgxBS2XdC_OC*>RglR9B0
z1Tr)=HMHeWLwgV)M@dEq(n8Se{zFC>
zLrNBz6bex$vio)RHvhhB7fVL+_5(bW7@qw@vZMJ>4TiH*&&Bi9Pm?9i$<`=x`x?n9@q<^d>3M9$oqc>-OB?Th!7ih8M%;3_l
zDV2G6mW_M1UH{O2Y@71Y{k>mG`>D+T-n$<*6?{T+|GCEvygeJNR_?uN&DYPqho`5`
z0{NxhlzkPx^N^a_22-R9zjPCGS`#~a+1J_^^;Ng=>j)m103OFk0F3?igr)<`#?lfYGhTeWn^d^vCpWoRwfs!Rr94Y92Nx>Z@aaBD$C>zz3h!-pj>L)C4
zV&n`JYOa}Lmy8C*5tnZ0zS38TzMv9(G|NbnuGp!rU}nVlc9DPW@CC6I+L2%+SAm$~
zyKnFBe!dIXqnINqMJ5V}|Ab_#&E4B~IDcEik8d)Jm@cwsTXh42(q_<)*}w1p>C?t|
z=0i)00^+3Wv2o*OX=?v5kLSgN}S2ox1CdmMpa3Z)nXpMMA4
z^LRRJ&3Ot(g!pfj(8B?QXIj4o@oa+lA@b8?Xf(;~j7#uYE>WCu36=5)9eIQ}
zwCkj!B?cp^$kaG#Y5_s^Qd37Jw0~iLROe4~152DcFl69_d+K;@oaq#dE(t%Q$saQ|
zc`a+dPP{9QdFfXPwqi&flG;QDaDC~+OhlX6o+85gs%!#he(vhtm#%mIB@ps~TJQO3
z-J2$V@x>{^M@}2*auo;Oa
zm*hl5L-Nqno16wtv^W&wC9CgGKWaSSv->4W|?C)U;i^cyYa%!M
zS-1Pyr$kT=TXq6#XMfu?;;D&R{{ehB!5PT2D{8wn;H7TLzVe_kie#YL#e{0}+!5_V
zb4s9;bRK)9GuW%RjJtA~
zY-s~CFOY(@WL#LP7BZf22VTO5SnAJg{n6KOs!GGbE4+MK2niXyE}NAvu}OGb`A)H7}+<@Lh96&=j^C|iPxR)-HOCw+GeBu)8hy$K?QMk(ZA_Rt<6~i!!c8
z*c?=iH-_IP`Jd!NK1;cWnX>w0Z02Tr$HZ2EN4($i&$Uv_CC65J#SQqLLno&-4Q!)n
z-G1pJ8-E@w-7zzgb6_iG9%f&(`1X-%g6-cCB?;&3%6Vj7jC)q>N1z%8o3(h|0=4ZLLF(DLaOGb
z?7RBK+%DF_ANj!y!>QkaC{7?FQ{ipB-}Q{H1N#78{k*GPWPbD{Xne<;LBB71<0{0b
zFwqbP)-$S-65|c~vJ`9he}aseF+=iMQb_(m79lvQ1VQtExkWKw!>i4A{=w5juoKDxd*T>g6Xf9Az2nFN{0Vin^Kk;qcmJqo1T3_4?-iqyd1
z9`^eAGPNOWr_wZBH^*v&%_QvG^0v&Y>W)VC`R8(9{Z4d07`5>bS<6-x0N1e=b~4nN(;b
zb-ee74+{*j!fFF+oJ!oMIEEC*P84Tyun{KCg3%;@*kms}%m!M-l4mK$lqgumB1+Ry
zA1u_%da063SQ}5T_8oFR4&}ottuiZ&nQq!!lrYt5xLB&NCMcDAVhl9
z*aOIgjX2Ed`=@-@=Jk#Ne`O|Nn;{XFC5-xdtorg$l|_rIHED7|?X^=LXn-l9`SNPJ+w%Zr>e&kS0+cD)WZzN
zN#u75qFZ>SiI6s(ADc~%wJ~6nkdimNTA*ze36&(MTo!)D#7$0ve_^Ww1te;k^m5?_
zV9!qPqyKNl%3BNAxki^Mv=XES%5bJ2t9~V>vTO-*kyZBmDslH0I
zkzB$icmyzWI5S$3#e}Z=1pg$t$6qe@FS46+CM0L-ea`jCYx-sZtX9f`7QmVnJb_GW
zf5%jZ`9}sO&1Mc7f13h)s7nhd)R(s-4Ea=oQ{q}}(H+P#t$SYFEIqi`ipZp2rF@eM
zBqEt8!(gugv6^dyP%vW)oLQydkb-SksYn>b_x{u63huT9T~Y}RfciLg*{?Ef_b9lX
z3islNtBe0!3?^4AVY4K#$_Oii_5R}huTijpF)(mKgNOXdf6d$nFfmp5(5v9x#Xs03
zA2UPgOe9hVR$3ub>zT1+yi=neyW!*73&&UxKDZ2uZSZ%5&1y+iz=mdnru^=paso+b
zP-CE=30@Ak9*{;&nQnH
z*om!>QWH@}R%DT!Ss{wIqi|EUr${>@se@u=gG<`^0bX!>8JlqoA34OvINGhcT*(l<
z>w0mq52#lZDhT7~8fg+c~3#7_z&B}Zc_e=P8j9XQom0!dn-O++Ga4ki`w
zmm^3C$Ks4B<=8)tEbmK#XoxcuJUU}j7d<+^{|^}X+VHi@Uf97fV!Oh`!nUW6QG9TB
z7EKuNc@4x$mZMK+=ohqHi8xb_`_H3D>|y_T2+o#le&)Oy@S_qvH~Kof$O**;$QH4@;;{tAmarm65}2OD$o*{J|&3IPYWXA|C=CU
z!FK$1fHM!m($wsI^Kd9HO$siO3JiqufL-n}yxyKi)3Y2A`M>V>ly96XrI{)vV%3l@
zoK9YYdQS3v*f~a_6Ey^0SsKrB1DI1%G3N%le{zqn9@eGdpu63E%^WSj>cPnjoQ
ze{pbo#N-=wEGY0?7;!ALWnsLzCmT3O5d0y8gvp|kGk?9y74JEfN!BGj%L}@^x*CbZ
zv1Ju}GBg*539wLtCp!tY`c8{Q#nRa9_Lh;6X>vlpWG9GzvG?|ukU!Ody>#NcIW$S
z^C7e19AWIEX911tyeoMVdNOJP;BZGYAx>8))7vd!tGiHy4hS>qgdfil=Hjp`oQ1i-
zHNeoI-37k+&mS*>_bxmBw_|&_e;JBkHgxvoCO`D3!`LCgmhv`KE#&mLj=l45!TImj
z$F5`VKXwm#VY8Svb3;_H8_v(dyJy`{KbqeY{ALj=7F=I^c9oqNcJ8Q=S6#y`{IL)<
z5Z=XXPq4*i%~#a0@eU-1>2ok|)bTGYN1-{R&%c22IF$fA)Vd9tw%9Wpe_r7tfPp2I
zQz>}e@XM;)7rUk{*AC*@WqnRf6qa*dCp^Ou;F)19B$YGa<`M&$f#vLev_)A9ew&kG
zZUkO$B2w?JzirEUT^@Y>_NBX`MyG|spq~1_2Ti)`yf1xkZ5Iq)qA}A22G5rSM<*x~
z^9lH_>cV?n*gu@$zR;z5fBW)#E&>>;bm72IqlNYsC4~(#@GxuNbi5s(aX40f(x!w*
zi#*CJUv9jo-lh~Fm@nV8#b)r`ELQ1IX#EY!+eT^ls%-H{ZIz2qeRIQiE3v$Zehsc=
zt02Q0hfkzSMSH07y6`g`KtQW1{i=19{$)`stEjkU(O=H~u@7p^-D)*F<
za82fP!wZg`;{o*i{kVV2m@vyk4CWHc+<+C6qr2M;D$*&}e&(trBWsfZcuto4x;)W9
zbzIP?YnP#ReR)-#e_XM9v`M;R8{3)DYjX~`2b6kjPqM*#8hmb7+-1{}0Y0Frp=ga7
zs(|1Ve}-vmzsIl^Rj?iT>pfT2;aKWmszx>F`4y6w@&FFF&K_`w*k{b}7UuOzp5`&^
zOts9(Y(j8UI+y8O(h7og8x86D1E_Sqghx6RQ8MIW8$Of`f36{O{)vHda6Y|J9YGnX
zvW&snFcbhFQ7t#`I`9BGUj!zjZvDk#2gz`^uITA|s1gsH@OApUZI1P3vK-7|Tka6O
zfvH*eZe4yX`_)fH-qz1f#(!=VC6@6NzMc
zGvl{vyQ11skNZDQ$H3<6;X|4TLO-Nv!&fu#SaJxZ3&e)!tQo8!<#uv}*HB-hVvio0
zRe9Y)&{K50!KSI7^~(=Pfi`9Avj7Ko#U$4Nh8~F{IDEbLnP`5=1iU8~OyeOY3syP<
z|2^XvKu#w8aCPxNMH4=}3T19&b98cLVQmU!Ze(v_Y6>whF_V!U69YCfFq2_dD1WV4
zNsrvd6~6md^d&GGn4wt7yauh2_5CfVd2;v~%uxqAivB?e>jmEz|Z!0#*VUKjk
zrCH0Xug0elG_qVutBU|)0(~+qh31mgXqn$g9e5nMsb^Cwc{R#Qdo(*HdG|6^}
zjYFL=7l-QPn!akAqOJx!;u)65Drg_@8__n@I21$CyFofFp8wPcO^YXe_kRxdoXJlP
zHBl1Ac6BJzujTFUX%vv=X8WE!pLxG0DNy^|wIHQ5_|=s|)jaZGC0A`5q=YL@?4VSl
zFwfRffv?e`+#`vCh3~p^T{I5cIyg8~VuM=LnZax1cnxfmsIY+)T|c}c6rIFjV)Nyh
zTT+?bQ>P}3^LV*tW-xd=F@Km`3?>JI3HP_w*x-Y!y>$4lRnM&>aj@i+>}D+AS3x%`|&puUT@iYxjemopIdG#dtB=)i
zTf3rbUYEZYM4ZN9lxvp62VVYXT>dTID$;UDWP-9Tkk6ib)#+TjlUECm38@aoo(&hq
zKERT{|H$owyM%zg5?KQIuVX(jMO}Tw4ITHMxa}=GUtc{jq4G*RR44+If(?`IAKK2L
zv4EG{u_Dvpp{?r{6@T)?*Mp(5CGvw%6z6kYu_>g&juCe=vw(zSe$&m61LG8Ogz$=w
z#<2=jCbFL*lC}v&q>LgG_V9ddVJno!6NP6Xn}reV{}RPf(+*C^$&_)t8hE4_K^sGR
zDu$|ri2W>d1q;3BZdr7Wmmv%IEhSvL4&`=L9VQR87IlbtGJj$VuBESBdeGy^!L4Nxt0J3;pT%~=!u)v|Pbe_1GrB(iG`z`1m)
zU&JDG(v9TB6W-Rrf#p&QVvGzNv-`>aWAcA+avr&8Z61Iu$9&q24=k#tzZ`VUrXbrG
zl`s!Kzr_Vwaes(`@3HNw-~4tpk8yG@hqmK%3{j+&#sl2GMagR&xX%T;yv$tEyq*#S
z?Y7AIYC~*1Ude^re@7s_lG)=3azmoTqA2H?;0=iCNH!GPKte8<$1@L%y?YpIraV`U
zycL?9|3Rb;+7p;RGSBQv*j$A+
z5_PzuL=j*j&}JzqyOtD0RRKo@u7MjstE$Cnfc{fe@IGP;*8mG7IvXdei+ny~ij4y1
z@nxd02Jm768&ID4&&K{~XjAP8KEX{esnPveN-y^`S(wGupyWFz+Veh_@y93VIjlmxU7O9WLm61+>avN
z0(}^hxfCeG02LKQeNoko*ZHh?@p3Ogk)naIligu8BjK%ZWg|#{iDBkSFCWZ-6f6pa
zBS{|oZsetss~`eM$@@RI1I5x_#(DR{b({sHL;nL86z(sfS`BM5Cj4(5+Myu_r%_T
zE{UyzfdQ|Q<%9%?Yy&vr_!09Ot>zC`3p2TGG7Xi7$%W~@IzdqFB=xqdJl?d;Z4XF^
zjL7hHU-g}wedjN2CvHM|n)?m%wCO5QynntY?5G5cyl?u!6?6+N$WlXGxvxN+T~$Bx
z0LOsV%ZZIadvK0i4szfS0a;{7T
zkFIeYLs0BO4fOzhPHvgvgW=1CQ!9m$pCo>=Yy4zV?q|Xe)-1wOrhWVcqT+Di#(#KT
z9$HKQ@r6nNkZ8#l{Rp@xDJu!7UI$=Ik0L2CSf4Ita-KA@$`3{p7L>v2%3V&E(2GrxJmX2p;;bYF+nk+sWK#a|EzBa|rg2583
zx1vJ$c8CeZX05jv8*Q^@qq$K}yF`KJF0l<=L-k||^`u~5m|K^WKsGNat@44r-;6~X
zlDM(fqxCobZDBaJzu_3S
z9Td!jJ6?J;*iC4%eB&t#DuWInebsT=O(g9m=-}vwvqusb1morg!zdpmaoHuCVLjpj
zRkCmxEGfZe{<%kwADI-Kp*~!_H$X*e>{Owmq_P~he+57*NEdmuTD2kprUMO{9!^Er(;Ws4^PdAdBd|?hF#Up43
zoFReY?6lOemh7Ya-1>X
zfC0n`(VI_>i6C;I=mkTS)x$FzQsQAcP*Fwx!sutq=MXwcuNU-5Hh+a&A8C_S#8P<}
zYqN@2TnHO+5U^r=MI`S43EphPNO{aizF@6@g2D`hxj6OzZmsKDY~mT>MyNn*i{m|u
z@-oM46h2#+xGh8$LO_9;v9XgB5`~d0BddfFFqxa}kQ$|7oCv&=Vr~kbQyA=|csEw^
z=*x#$J68Y(<^H&YiGOE&1aOy_rh@OkqON-1lpQ^sLUF}!Ft4G)wOgnJ(sr)=e>QMY
ztLZNTfKDgwVD{h^am(}g5&{{K1c4)$=UaULNaU;dfslt7N~qxEPJb;u^-N=MG6
zT~7NzZ_hAQ)Q@ckk~uBUfZ!Hh3qi_2au!2+hqWp|xv?n*2k-=X=%rBeO!_2Al0N-Q
zc*<{k@1chvjz^}8^xR+l!u6V-Pk
zN#I!onPvTNC;JST9M8UU0T+8`0*AjZ3tfO;7v2nb5C+Y#())k$Ch!07CMO5i|L|s>
z&ZOr-JL=YpmNoYI@F)t%fV8JfIsaK4e=7S`ssa_fZV2V*f95{guDtqLM)hQ0?ZyqA
zJ;aqR$KX4A&R8qpABx<2#Rjsl|H4|0vZU?8+SMn6res2yce&NNZhF~229IpEm1#%r
zCP4CM)vxOb6IO1?=6s`#zmi8JjNE{D%bjgt9tB^gftH|B!d2bNOIOaDjQftKfKKMt
z#N=MxFmw*)&FSZyQBk#mI+xr(*{!y2zGk+jt(Do$&#{7Up6{2d9gz=M_-HjME2Q=E
zI=r{smihYMvGhmFn8+Iux|`kkE46<4CtN9r7PAfq1?_5_pUPbD%nuf;8i{~gs$^i@
znQLuSF0l}9QnAKfE-YKV!B6=kUKrhwLZg1Yu3ujl(L?s8`wk=R#mX-B*|qht=S6sR
zPX24C*izDrm2JJ`3vto;CsfGhhyqMMvFtqstlOHEA-`A=QY5?bj24TKa@KG23a?^T
zu`CRrV_5x$L01~&eiNzNBSis8F#5Zp^tS+=VEteDgn=XD1?4G$n_WE>Ah$MNP
z`qj=d1#x$oV!fd;UA&&biHv<^iLY&qzcYsΝ>vP82}&_L;GT@`ncy(0F%R7NO4f
zK`J)EpFplL;c(WoE@StTpPzN@CnEamW-@M{vj2+UGf2N9V+yafIA+D=-+w*?UI=|2
zzt#-Io4H-D;QWknSyc5bZBoa~p0kP`by_42SM@Vxww5!cQK=~6AmteCPYJkBubwOC`7hlqb?+gL|@LO#bBrJS9BObKk
z7fU-PE~B@U_i1BjN&y^0&L10YU6-;kXK9)FKF9xQJaeAMWZInH(xO5YQCGy(o~s3t
z2uK}zC;M1xE6E(+&Ck=gCWEO+mzF*eBl;cQ>&oK}_;Ans6vi{20&7k*eVzg^!+jry
z$sn0|S!eSn1ZF@)Lm*bP{)K>wi)8}nxKgj3|0fTd$eH$j
zI;CplQsj~a11d5m>cKRW1MxwsA|kCY5|l2FKb%*+W(5FUt?H*IG?vWv^RdVH@IXB#
z0pkBINlDN~?LGdw-AtOSj@LN{ZSKX4_0rzhM)dt%hp3E61X)yQuBaG3$zqhajP1O9
zkC~e^4ddoQzl>8n(JF=Ad0AXXpUUww^3ogibKt(65%#$49D9`;Zy>(S?w8*W
z`uG7gOnn33qz{T9A&Ops?RL8s{V$t5LQy|mE-+L_6h3`sf)VjG3c_iAc|4j
z!D~rEC_kN;7K{qkY!Ra1t$uJ&pU!(>7UigJtKZHF(KEpcC^<9asK7~PmorFO7+nia
zH43m0-|RQ0@G^iB02fgtzWo`5n(1wF{-DlyitUbH8<%mPN>7;e{JD#NUDy8CBrro@
z%tJv`rs`#jJQnXjvZth!%V6+ICGFthCHHL*Q#xf(QN@HF|IPE@m+cMny1zb6YRs-6
zcKd+&z4fZQ10yN?kOgm=ddq_~#{!s3{{yg8?C5t!G^&la(4vIdu6I9NZeR7@YM7GK
z$n(ZA%b6O-`CUt~NGgR!sK2X^)+SkQ&k!=E;USlCOR=Q;mpE3x>eFD>!yhf6tg^Mf
z0-#CfQk{=LLjXmO+prod-$sLP}Q=
zs#g4BW{xOAxwyL(_^ibGhAL~;Q)662-`0Q?m
z_mv%5Yc9+s>^2MxJb}WEiXAZOM5;fsc;u-G^;CTPYe<1e6RMBpjgOOcL{;QLjs~Ek
zvI7VSa%IfIfjzf$&Hzegjf_UjrnRZBIffTf^njcIf9x9yTEQQLDqXVkIl;xT#O4Fn
zbh4+pgH);b5tfdCGYo2Bn2hE5_rB6)lLDkMZ??RwFPzF%iNI+w^Ny}>@#_K;u%qV?
zHHE$3f0CUSvlU@?qD8RtPDtcACC8_$=-V!SSK;=If(mmg&;TINu=)uleLTVZFwaG*
zHWr>q8??`Ir$wO4uFNE-l{nJMiWozp@Sp{oYz2d!mPj?Ap_yMZiRaoOO;C0sS~bD%
zjBR>Msi6x^)|>4aGZXB_MsGw+z)BpNFC1EM@2V6U8vbV@uNl=J^NtC0_AXB*0VoIy
zJHO_$gR97IQ09XIH2nVuC9?AV-wrxnUN(~dVMY%PIk#LMl>g2sox|u7qCr4Fw^ZE9%91jU5MtqUohEz5w-iyWIlq=1hW7eu?PNLm1T)lw=xCrf
zGLN2fLeF_jLyVZFD?m6?oiT=;0MoTPW9|MO;!juWs1+&7@sX#>D*Xw83iX8$eHGad
zt<$RTqB}Q{CW3BdyJ$6fLKEQS=oDLo3!SbTFN
zmYn?w;lJSn^M8wl?bcuMn$AUx{M+fgqbPk9lhQj{W@D`KZkqsu?lMM%vg>(OH0Z>c
z?I!+DN<0eYy^Gr5%*tNy0M(n3w8^=cl0U}bHT*=I5Hl%rpgg^wK#txvFYFdNg1@vM
zJ^bbH(Gf5lUc$0_8OOWJUv(d5Yrw~MV~9{v@}^1r@oc^E#Mn!ZC=peLhB`gPX%G_*
zJnty?VW}%xmf&x>X!%9-sn>JYZ7|
zj_ihE35sM>Uu`SR)5y_l%Ah)wR^kZUbMSWfdz}Y*JrR3Y$dY6t1c31&v+({O11%db
zEARg-|7l%4C;Se~z}5O)v-bw}Zlf6EVDrUN_+}B53K6{_}b_pMX0We(6|$e;b5n#Pgei34$>ctVHI0tjVhH-bwZKQLc
zgx2dIQPv(hI5^twxc($!kJHdQz
zCb2q;b@nQ2DoDGjav@PHNob52LpX9_CV0{&9Lj9XJcNvXWLo?7mWe0m{7<9IK)!&=
zK{TF`VPA*lftv^39#8%9>%OZ;dG9yTV(i`Yw^N3#zsFZ#2M*BQtIJrg=|z|7$l7bT
zZ2)xlXm1Yy=DWm`UYRGeb;7^Y6tFosK>b?#@AYazli_p@y0>nOzQ2%1a8;+^q7Xz*
z&krJzts-gilt;6mz}2!u&+~AB+!nVV2GQcdq%$Nf^N0Owr%ujkKi`{o_06+8zI*tY&nAI+)&e$fkFw8Bg6
zaKypX@Jt(e*Q~NUF2Ud|f$ngb36WO)qH$`rWGn{zDO=*wE9u!sR>lIkz;BGw*v4zG
zl`xotLGdIn{=0#ZgCs8-u6wnG2!YSusS?S4qFm-3`MbOj!TW={M$u5UOK@n=?RP<@
zc!b@dgbiZYoE^~V|9YSRIcEnPf-Z|xw9W}qqT(-6l;J-?8uX)=?ZvSjaMm{HKfkSg
zU_=IXplSXdfWSF&iKs0cYm6gMb?G#3Z;)oN#jZ#R7slmE#`5|6_`|Y8Q#1pY4UJ&A
zztwZ+KmQ|vBhLB;UusEu#mq8L)bf~mJ)pU@KdSk&cIb)}ggOYY7kD!Cz6Qzleug?U
z1Kot}aWOnqI2H-{_XXWHHv7;lBmgVc@=I439#Dw^W9rA9`pqG=wBTT|iuj@Qs
z0@&jZKdXd?i_-3w?X$?SS|8+ec6912{-fb
z)o=Zb;&=+EJZ@#Z`mw>`OU<7~-ULppJ{$f-|WjW|n@Bef4~%76DPtw0)IAFqa+Jz9a_eoDOk5(;}!
z!1U)r(2rL={dJ`B5^(Is$gYmH7DTsZhb*7gyC_KDTRkt`CMdF^U;a{F{Psj&i)%>xtm4*dB6
zKST0?5BzCghtK4g0baX}`j`aWHHW6-HBUXWqrxll*gD;xAs^hIKH47|%MjgG(P33wJ
z-5u(+>cVZftXm-Y6FZm5*&e8U;~U^#RlJu0)-SYL8#v0OWJGfe9(W&b~)mTbG*wB?ki|=8C@_U>@*H%v*?a!
zjB)vhHnAMBD(5cG23hR}^EjrVH8@_?vjs}1PE=9&4u+%8L|8ZCA23~LZ7FUY5a!$f
zu2z)AHoe*<5!8kVCp?V7^RSZTVjHm=cBP3`3b7aj^ko=h{&Xg?G4tzOZA+W`KWW0&
z6PE!Lb^XW(G+C(kWVK8b1FAHsum~XyAaNL`vw4W(1T$LdE;S+{)?p6D!k*VN5QV84Y(VCctBK+|2w%X1*t_%umm=BzNlT?xY~=%rKZ%|
z3XT*fmk{syCM%5^GXpgN_k8#|OfI6zB_DCIlD_Ko+Td9FyF
zM2qodt;Du`#&S}j0AAwWBrXTPH0LOYt_tO6W|h1V%f1kOKriN`&im+x1$!1C+XT4x
z>}t)-Y5Tu2J6{We{{
z(4bOMdv3P7dQp*(kB(TTH?IJ87JMt_ge)}N+%@tAlQUUc%8D34@KYJr;a*YNE4O>?
zViA^GNnNg<)O%Qp*io`C61?oIP6KLZvlRs8u{%R3N|IHVC{Vy5cXcKXZ~=X
z4pW3F5vI1p`_;}#?8SeVgEjrU8d^7Lh1-bfPGt?(F~V31jLeTyWN88Ph&q^=xtG!h
zWnfQZZmWabo2FAb?@jH{1Du0c?~iIp3(fX88SZtEmjV$pk!|P;)-CZVHyk$#vr+AY
z(@8KvLp{WDUup@;DE#B+TF_WIWT~_HUi)o9h~jGoyh@X1vlQqQ4j2e@g&*>M;vHWw
zIT+a_4^Nqs$Y#&emplR0=`hn08l0G`eDxRnuyle7YMz9NzCY|JEn*h#m8Fer1STlL
z2-HeFE({w;;HniJ^qaR^YZxGLU@7p`lRjsG7xrywR&O0?;2sPSYmEjla-viE|E`)!`qo(d3$x|lSej&=Rc#KO
z{b}Yvd(pl6!#M}(CPuycXEvS4l^pzSXBGHPXC%(;opU{L+5K$MkAE{F$%q>2xtTmU
zkT4xGM|_6U3=<^)gdrH5=!BxWmQ1s{Y4VB`LI){wUP*g$1|FY}%XW7M1WsP}?MW@g
zL_qjrlAvx!i@l!NVYKA?ZF)3nkM(@0N%&X@iwd>>s;)m*%R1xabYTX943QCkG(_DB
zKT1}gHhBsh%p;!RZ*mL%I**ktJ2zRYSR>F4J%#=MU#VwSih)r!c?bWZ%0!9>!(?D%
zV`uwcNM|EqOcYp2KS<>M;=2|7Z{Lu0k8y87Micq$2vj%E=q=Pxek$fbRr
z4el9EI|y5&tS47QYaBwenCA)=2o)a`6qdHCs)=-AQB!zcZB!9^5`6!)JyH_4=NF7S
z1PkzA#qwYZ*l=*G$XKE%dQdX72-c~6<-Q5BIfOFK7bKjg;!Lwtgd*{fh#ot<9?6k(
zR6|}*WJ!2-$3p>cmW|E#)FZWS6cUMrwY|m3vAG2(OEV|vkN?c6LUC9iaQu2f
zsB@ch>lnsXaLS;zAnNIo>1m+*viDVR6KQ}uF*<$M-;E94FTL1?#+WJ|L=g#PMM(@0
z^%BsiYP#x=f7M{kpL0%9;7Mx#aK2MdhVGQwYbs+3W6DEl$Uh4>L2y9pM5L)>g>Am2
z@z)~R`gazD#1sgx?7w6|3R%27TThBByxiPM*xg)Q%bVOdN}Aff9BjSRL|49_;NP4B1&pA7v~j>4@`8Yn;ma>h>j#yI#6sk*2Oul}WANW>
z?#j0-G!JO-clzk&+*!RgE+iywP<&B%E&JTc+{)|F{My{w2(0mK=G6hVwb&PPut-ot
zY{@|ZLE|^~;cEfMH?_f^YcmVJ6+=KM!tuMyETWZ3G5@=A8erb|JF_xAIKFmf#0~~p
zjVO~+E6~~6k=a|BJru3OJ}do`U=TV#v_Cl&{T4QGu&_I^ym;!0g_NJUw`t4gRLmmhudW4*&fioeB_O{Oz;MZ02kMcmA6EC_4O(
z`0Dw@00>nl647lMH!cu&_2W
zxw_j4{5~i~`XEo=|3bm#i8_7&A*x7!^{_^?p@bW5la
z#(O?k`2Ff;1!|3}^Y_c@vIIN6);?HK5u_Ce2Nkxx74{W|`<7-50t`hZ4P`as-hlNZ
zW@;5ZGk$b$wAdh4KI(3lYqww$yL}BLZ&n(c#6aXDQ$nf0B6nt*-LxC9A
z{JT66!S$OIDI&22`1~)C_Vr(40$7xWO8-VV8Ox0T7RPd>fW=7LgKre_A!)c&fP_4V
z=oR_nhs_)E#~+*L|03HjVsn#E5@bKO!Yg3`A$zRv{={^>FK}UV-A`~~cinGr;js0O
z_&8#*Z&vn$NaZeIH&Xe%ugw2-9Y1h0VMRpkq^~v(_-_5W=>8o0Ae|sOwKjQ4WMc`e
znYtr>{f$ro>`I3{2;Z9FGW|ETMf+F0JhATPcQ7%Tz*jJFr{MSC_q9W~aORtl7uL|u
z^w&+rmQl^+rL`5LdK=3>?}hIn$(6aa`9Xu9dk2Fi9}U3A#}3f>ogVaf({)cgFDB-B
zZP*Mj5ZHt0QYmVJkZ+73vHrJzzJF-d-1QL&00y4K75Cq-v$e<|U|YeX4RQ~jqnn}`
zGKi!jy~Q1=lqjXS?$j1I{&=dtcjbPiI`9}tG`2R
zik_}GT}tEkfNezdHumyF6^E2$+qwf)pf-)Mfs}+2X_2bgygA_vi#CkE8DBw`
z=gCcGT4%;+%8hPlw4x>WC>g8QP#pg?#l;s8)~ak_C;#4X!J^&}&D$XsmX^kQhG9~f
z#yJc#j*}H#_wpjMQ=vYpJYCh8BF++#()2vo|4XCxcd)wiU2D6jf-xjrZolcn`4Sxv
zDv4UfgP4y{UXfgaX&G{0Jkw2s2b~nS$pFaYB85%dIHKwEXb$ui$I^#Aq
zZz6#b)FJ$<3YkiaYqY}^{3R#*LK`y=0wfPPMZ9FNJtM){Je%K7J#u_%5q&g_`FGDZ
z#cLPJ+_XeGG6w-}f;>#u5hkXcun#{cEL`Sid9XWJ?H!2t8e`-+Wy*R)o#1N`I
z>ez8~0yEovY5j*-2ESjU2+4*1_uxCrNv3w8Ab6Dsmj7hSzzp2vu=MD|A5obWA~I@-
z&}Fr8WeLaSR(2ITrU$xBrR+s`;~|_O-Ak+vIT(DVtLj*#iSbgmg!=?dGb22}uwB_H
z^eH#W!6gzj^QrvL*;SPSaMWKPTPQMK3u~3CXb=~~^Skuf8^s<|Vcdl=XTZm5(4b4a
zsQ2`28xkq}#Gs5a$%A8ZEpS49ho3pP{s`LUWHRj8HS#kJh0
z8kYK-UvA^?e4|+6`Bvo>sz6GBB<(;EQiSPHTz<=-x$PNAd#?1#Y8)%9mD2QYj|@qz
zN2ahni)6xAofOs1lDaV3W!GtED#qqy2*$Bz!M#c=H1sMiwXc2a9kl#qKQ~sl6f^3`
zw6i(fk@*BZ)-)EBIedssqvA}xvX-^=ueiuy)xUid4t*I1G$BD3?mS$;0wjk0+T2Uh
zud5n*ln3_~#A7+DoDK;5)U`20--t9}cKq8_tPZ4gx-&O?1qO+dh9Ho%pDHFlW9X*(48NIw_01$Y;b5JuQe
zTqN*s`--_C-P7B#MQvAA_V{fyy-?|YUZ8b{RBl|Aq`NjDl|Z{p4(TO)ln*Vb%T!C2
z*3;EBglqZcYB~pc?R?E(b75Q|#CWvi`haK%LfD_{D7p?vb15UGMawx^c~)+m^!3CYij9uMvlKtJ&r;Mi#qI7-
zhigXL+-obBwwZ*L=$$4?@E-FZQ)vEMM{;l(^l4F|>~mOt$h`J^egQ8hP$^^h;@*z!
zn^`(jBRt*fW~JY4&4NymD@nB|-iZ3(@ZR82f#4}qZ;_b@>^F}xq$HLh)sAxwpP;LE
z4gV~_l`4ffkKOT(ofr2|^^&zg@;v)NY)j_96!mWvk4iu1g4nqg`+2-6(kL+&3gZJc
z`b5ViKw)({*l>fX_IfL1qI3EBKFr*pP_RbK^xvApf>lq=ztA1CKy=S;$$JykZCo-^
zlQ9y|K5(~UKsqSN{h9lDQz`S=_O_wLrha&*kW+7IKut;(f?eZ^&Mo|bDKL};OQGyi
zw%k#49+Ed#P5{IxAKBqOx0`su@-vvf6CCNO^eyb5ohbjy@Ix8CL{PbkP*RadYmr9F
zf(q?iK87R;DcgMC)UiggAr23t!C7wba2IBWxTe&L$o7a_Uafk)jpj;~zCZif-z
z$8GQe)OKkQwqlWfoL~~>4z$+~5E!E-yAmCzvhxj${^Shfu?9rPYl?~1W1(tT>s#0sVxp|k>Q2~BraS+KAKmnd=Bhpq`!8#rqp{YoE`
ztu2q%GNk-3my6S2l%gRWE5RM^Bl0k@7OfvRU$Gk_1c=PY=djRv5svwPgbPQbtPHS~jbW_#jj`rT$y@>)hwuSxZ>a
zP~quTMlsD<*+WVfjYdry$``B%gpnD^&C%Hmr{kPT&2kW2F(gQS8PtlQrt~^{EnTy@
zleU9?kMnpmDO`+$z8w}jA%KF|RX4z%nk!cdvTcvUUq8nyE3NI&9<$2`R|P;;G+%ID
z0EQ`jL#1C<1e^V@UHp6+sIbxrk;v2QD-P`zJ}}l<*7fswf}Ks3V`-wm>(hH{433f3
zVaYpE7uia<3AZFeh57XunB#b&YyUj}x4jF$t
zaCHI>*oi}bTBLqzS?U^sl_E8ZZdIN}9aYC2zFne=pk1vAm$82NV<12+_RJT5PF-qZ
zJD`n!Lq>O*Ush(3HBWO7gKK7e(!D_U6OYJ;waHV@v(S9R!u6frW>{IDeDT%NuU?t;
z-#a*eEkW<~B<@&40Afq+K5G}=V*d%r?HTbdQ2{i16_nc^TQ9@!i}Kd*V|Zj9WA?$80cVT07nPU4j^jK>in
zy*WfK1DD4fRx^{KoY|ik0c5nuu8`!6r!dcEW^Ea(H8g`^QV`e&e|P`n;1oxGT$8F%
zYcJ*^x2O;H)^vs3hYliu!+hu51UM4)Qm3pByQ+FMmJ*E@XJt8aj4KmL#y^n9l-nHE
zm7LKvHrTD#AiM!?(LX7B(3#q6vEg?M_i%GN97KQknkScTIg&Fvc%|z^
zGN*Efnk^1s$orIsjDvi4bYa~}B{{EaIZz>c!ZJxUOjFX34$vB}CkKk*9!xyl%rlUyS~;UVClwREn>a
z_JhAQg<1Zg6e8GrAQVH(2fV~}?Px{D&RiUZ?Su)y|Hsd@7H?_70w7o44Su^p7VBOk
z-1`r<5oFwWBiv^ytH?w!G1>dfhaPp&jTeG;$_VV#plFGJw(5)=y!Q(VyPQpxR`O&v
z{dv7vQ)`IYA!R-oUL)dDvrMQhO)6-rWgJ
zq_q6c=PvzuB(acc$1a)_sQevM`j{Td)m;mWc$afz(4Mqg6IJ!n9#+i@Z0Ki{k_CO-
zw{T_()zlhRRd$NjMkXk3pm0cy&>?*uz*Iqb-Cspv7zVG9_z?IToKI*YwdHv$qVhfq
z>Jr;Sm%;CvS`zhL*jf#L4wQ?CB*?_Jy~sG9Z?)kIrI$?XMRg08y(b~~7)6PuV5tMm
zF-P~)&a)z{JL;UbBkHdfdz!PXU%@YSdP+p*@;bVLZASgUR)F22t01(0MbSZv7owEf
zD-Vo!xt2cnOnJ(n$I|rIM_UM{)*a$8##{01LujI%p7
zN}@p13V-9ZO2nG@?%s;Ax+uKo3m#B~tD8|mi!^g3GdJp|^{_-JN`DQZ!QtGCG^JK=Q{7eqLoB6D_V
z8B~yBKbj1~;xR6JSa&p!ScP>+qAIXLFxH)$pANpf0u)WyTR2#`mtqyqUjiUEobiHB
z7+Q2}rZd2C-^|~mmQhG-%pWi&><3^&>D73I(c9<1K4TzW@kUlKwsn>Y_C#o)t9Yi}
z<8QT;{-{}ln?uS^23xZYdNpIj?d)_^I3E@Utk=S@EZkWbo6-?(W`%R&bQmYlYI9Wf(Ycp~lLtPqkN!(I;AH)^p
zvg=ea_lQ_#P2QSbJ?P(={pCjB4JS9YmX|X%+!Z3wzi-eQ4PlQslj4oX%M&f*?+kT$
zdg`94A~+9XrB5#YvW4w&1+A~@?_V>sXXYGpO3Z)mBbdiujPVw;Qvjks?P~kL0P*a5
zalfvzBml6}WMYYrr><1kmn&!R9i1aXR2pifV!jvR*wm+I>`!7C$ef{i4rqHKROQ6Z
z`w>EseuY{+5$zV+vZ2ed$x=}JE@pPj4H<~f@
zd646*Z+~kAMp#p(sQ4!Cg8hMDq%osmQiBmO_BkL`+;v1RlZ5xOueCF0UHYcs6Cwne
zXV^qTX2&|uNiDP~5{X6*9ge|hPv1VGECU?Lj%--Q*xQBl0Wr5z1XqzH=D!i92lfR!
zxe+X0K5T*N&Uxd&uuwYzGQyrg+X&+WmA9t;aIj{j^|O40<{lf3VJdUBmZz5kiJhXu
zmqp?@3KNTt@SRE}NX(nx7dO=`hx7vzqjwp}CueyFq=Li;pJZ5?4x?AcmPINAQR=qm
zWhs?9-~$0+K!tE#kl%~*`;&|G;Wr?g1
zxFa5HI0OlovXbd32&ivgC?Mm$B)6z;5O+6nP*0h4wPvZ~M}|MZFsnoy2|}RZTf}l&
zmKH{}VI;$9@4CjSUv-j#ryRcff4RfsdwC>KWhrDJ05?Pjq?rDWJqmJ%+5LvL5!h!r
zk?9lSJ~qimCGojW@tjg%ko@Z
z_-Sx^BLe>u{o%*IqI`AaVp)}5^T8TX1vk;i5I(B)j{~Er9`!>!TTR@qZZ(blbJPG62*rDL%x3Fr>THcJ
z_RGJDuqhaJIzOzL((tf*bx0rh=A+poMya95-}N%-*UEa}p+O@oV%j*^MxAK<;9e$j
z8#`r5&^7_)%J}SyW}w%7KJuUAAz#05Qln%6~J<8VBN#K(bZl3mUq=~Yi
zb6fPEe~zgl9o{rk>#?`DtQ{&vXG7`}<~kXB!dv@Rl+v0u%59xBK7M2n*U*E#$s2Px
zem1>2i~cyFV#$vwLB+iDlNPWvqU_E6$$bEoWqL*l;
z=iFGV(7}7e*FY#>Khyns(XEAi)C{}R)LrqFkC^|nw-yh3*^KDn^sg8Zdcn*eLBA@q
zbT_?V(aeOke-|J8aG+F|0Ob!Qx_&;c5nv=_D;-K{*D`Cel5fd^Av#)gvr~QyzSRwB
z^kOQeK?VgsVL(q-$6kl7uSP^b+C+n~y+)YBNY_d?rAxKxq1eyROAwsGKQM-vZ)v@!=GjyB;PCLFuwaTEoq&hKI^R
zVRnm!dc2ik{IJ4bQ@TUpI62SBi|+oC<6Of3eTgWrdh2377DhE;u6cX@Q~k5{wtnTY4Ud4WBm-@pl#^ki_io1-@H
z)m?YoP{IPypwxpj{{XNb{;@ZWgNDYBG1^BqMe)Zd(g3t~6&*mePSe}3Y%EZ9wfU3a
z?<$S{Gq4fr>a|3<1`XUju+O#lOnziEa`i3{L
zo)E6tjr0EuiFiRxjFP4b2F`6y2nJ#T1P7SGAT7;$y^Wn*nI-pWJ+_@4O5M?M&oUX#}WGQ_1lSH5$
z%B`tz1(n_|2^?M>Tfl_jjSr1vjz~PgD5(qmjPSF-k!0X@QO5p=l~wirckMQgE9NZk
zn9wot%hbrlle;2**n^l2gc}Jp2*S6$U
z1`xw0anbGZ`%}sIbZ@qMtGObE_YTkDWc(#rpWcMP6E$1V@j4@Wd_H
z6j280hG(wS09tMyf~EBxka?iQ1tWAW1E)kJB?PtyPrTjY5uBfb60L{sWJ7#tFZ)cc
zP<*)S@isY@vLmt?C}9slHG)Ol8+0Nx2t;Mtx@yC5He%F`C^xPk6{BSOTM#!92xJlz
z>-k4Jk)%XS!zak6vG<~{qQrg*o9#|!{vt(&TCi#E|2f^EMMcx`OY9doRZ;%X&jM#X
zf>c+eO=p9hu{JG2?))|zQfyJjvqJEhfndY>uwv!IhtmoTXiZaiMSSg?9NiGl9UwrS
z5@*>Tt3d=Q}W>9yghe6M3$xm1kXfKEQqB`=Bv;%nY^?j&mC{>WMN0+YgU7daIPLiS(gzn}
z$()hrtxns0FNNk(_r;?s0>s3kk4GU53h924Z!1S}9bcnz)
zjW?-OmygRFBZm?vA1L?h1yic=@n@bYO#R^;)1ewim`_y7;jBQddx7_*#W$CaVsrZ0
zmj!PF=?8Cq0Tvo<22Y?uqks;<1m99jM_Zm4>Ay+_o||Is;4Q4O3tUR_>6~E+ITk_^
zqOV0ewUI34zmV8n_&9ALBdO0ig&&{bYW|L=%YC9+q5j6oK9HM??q2n#ezj07V|E_P
zDz!J(KG<9dMCBCbRtpp>VG%THgFs=aORAotvS~~0Zy|&$=?YPnR9G2JxiYcbeyoR0
zOP*xs;i5q*WmpDt1yILy{8?G$+%2lw$KPDTMfsF|7B%if1+Cp^YG}o@W9>WmH1YQ=
zt$L>a`~B6*>H|&tGyX_kPkqo`uM#;Q8w>@iQ8OYEhcn;IP^84FxQjf|4xuQZu2C!kA^58Wzlxb%;sa9TX0{tm+4Ztp$c8?6anh{&}GDzXAadjwx;k5OD4{_WN$
zr0qR&?rVfuUm&g^jRe+#`M=^bzakv(_Ba(GZbp={6N(fLGAN!MQWdBi`cgMo;_TtY
z(SWl0rCZQ=m^c*@Qr>l9}!cZW1j!*{)QcZh7E5!wU9Pwo}u1d;F9t^0ig;@PRo&
z1_R#soSPd>u+7gLn5oJFxLFd0D8@Fd|BR-CZPWP&$l-;7@cOl*T`eUmLAM(C)C;j%
z-wVv2sJJ)RUM@&#Af(S%sfa4o2x4d|56Ey{n6`yKCJFQmH+AvV1g;aA*Gsh?kSwxH
zd@bR1wf7a|j6V+yUc7vFnl1p8h*eB6|qElNO3n{~rKIK)1hR
zn6n>VkI-|zaECGKTEclXF&aU5$%G4!sxtvH)hb8>HU{2e48!Hs8RriH%ldFu%?AB1!;h`*dxIPmOd*eQvj*9MGp$rzx
zY37<(e_l1Q#?)Mvaq;73M2V@;Y0>%2z+GFHR6`PhJg
zEt=7Z2&#%AGygg^5_)G{zDiJ+>zIcOzC^6RpT+27G1CAuty4`@6`}Uz!5Ilcq1WZo
z-&ikv87u(p%Vbfw_(pHoow2oZ384ip@ME8Lf55joj`_PennEm~>c=?5rFsavbsB9d
z0+`6&u91%+ca91|?uE$@68nXt-*%=n2O*KT_c8iLAy1kt_~r!=(wEqfDPf;S$jB+I
z5)541aZsijzVp$ex>pG3Gr+8~p)FybkE4S~_NOkQhHlP%U9XfzXiL}3*Q{x8>HA&+
ze}7?(B$gehD-O1Gq)wrE6)6RxBD7#uf=;bh6EOIYT&WQ&K)p*P1BYfn$4Q&G;wNcB
zoOY)jgn?O}xU}$)F)!(|YVpg_izj7Opr5+vYSHTt4RQ1ZLcSm7pU2agr%FK
zrf1DG%VeLcw(W%AWDXrTHHZpT>zMSne`7r8`qOhn*h4FY{A_*c5?QylT-rTH;l5A;
zm)eNc&Pz)w0r#;i)_MAzn1KlFJwjzDr$zI6EPkoXPYO0AoMF~wvT^f!0eo2AFiT&^
z>Yy0IYZ_*b%VQu{OI@Ia!Bip3r6~nG>YouPjDB`_7R=U^{aQ1hUUJb<-(oQWe-b(fkL`$wt@nT*dsfMohuvRP?U6onc+eWrQX$-GSmWjfB<6(>=Y
zN?d)hG>RFU-NwOcfqgeQ97=2HlsqfSOAQSQ36lc^&a(+5dL#=4j0w^rq2Uh4ExUal
z6w4K=wq`!UC{Bq#JhSLgVa*;Tf2w%CNs7?ouhvy3RouIq8nqRE?9L_0h~jM2r){4I
zn(O4+^Ux-)R4Wg6U>yJ@|8+qU?ynWk-ZRXSJSv$}7zc5i#XJMp2y5$p%!b%$e4=lDD<{hj3jZcGp+gq_83G#Xu+i&*bxX5s+@H-Fbq9y2BuAgQ=D^H
z&d0TF5ffBHO?4iE?X<7Xwqk@UP~t$d-9Dw~xllk_zLzr&p?V=9r}5V+wSCu6twyo;Cw
zt6~-Sx2h2JT|6poQhDWxgAt}bC7Q*Tk5ZMtjCe%TsduT_=Jqp!!^FigTLfiEYHPhd
zeRgMkgPZ8r(_kY!e>DqBkFo`@vwtldTPz|E%NwcC_TkQ#kQwL!-3Yz7gcwW?n8!d*
z3eZl2v*8MP2FKu=F^RpB_2xB}ChZwtd~bi?G;~|!w%iie7EMGJZjbbUJ~)>?u$+gQ
zel&7S6BXu+Coz_<3sMMftv7WncpF!yYLTtDFWFD5Qbk&5e;cnBgS>CTk<8G>WDg<^
zpoqICXpV~3ifz4L*p2#Wc;EG~C6F&LUv#JBGvC&i)Z2POt|rAN>m|+xSw!AJ%g=M*
zEwz4;6gCw4^>GTT%jhJ%5pSgqOnn_rf=DFp>MFZxihu2X4>eu90-T}-hJ5a8am^jI
zo_0{Rf#YfAf3iTM)1Z-W;Wk4l;i0Q=|2tFS0GB$oOX(?^fl;FsxwCcp5P9`xxXWX5
zT^}EMk(|%L`s%yvJdzY@lDXq;nsH~IDE=pdw=3jNM#6F$$bE5VCtuSJBfqOng!IQY
zXq{cvMP9w148%{{s7%vz3H=Ug>yc&_conuObre_Re<|x8HAH9D*pU}k$#rK9EL2gf
zR(j>NAbUwQaHhB?STNkp^uCM4tm8^=tJoahC_um!+?{5^%diXy2-;uKV^OE~
zvzI&$(w;l(Xb?qbYTnnAOchDVr}9-v-B}~ZuD|1pFQ%M%QSGKn3HHJLjmG?*(Y?IG
zkUa$_e6g2mL~|D`cyKj!L&z`j^O`jauQ6cD<+O8i$YVHZzCy9-f6S8XTB8X}@Fy}xs1AJ(eRtJVTDK$^
z3OWp9#f+q9OQA5>%x$GWFo^gXK8Sgu$YJsP7BNGoI{kxhX*2MAD50vw-?DdoEyOwI
z(tL(PZ~CdSExs#iZsnax)pEQ6-p=N3JY}zk6+Peyu(cPgR^ow)KeFHadspZVG|m_h
zf4Ou2kj6a+z3os_7!e+Zhi{j<=3^yxDy9tcJg%!SrJ<9I>&%c~UB^~*k5SVSOO39n
z(#4yU7oTrF;CXVH4uSOxqs_BJ!KGJ(-fXsC(E3%xijS}j7PDnfBf5K{wtmC5LO~@bRl#x6A(J4yyL2d9axZV
zCVQ||@T$gHP?OU&b}JZuqf@wA7-%xe*MgUrb1SUL+K4
ze@;xF@lBNOm~^RLuJ^C%rT&%0e{v=8n`lEPp&?=$-veStVkf4YqIxW~294Ka{(@nN
z05{RM-uD|y62}f;lW)N%nPfQ!F6gN^XKfavwOw_3v}3$xY_}shxYrCJpr!NF7cu77
zc<)mK|1jx_x!USA|IU(f3ElD~{z@*v{N{3Qho=H{=nM4PPD*`?
z*JiG7?R=)R`7cj3gjXH-&yIHj(%1!HX(QMld9MQHRaR$MuSlam4GO6Crf1EY@?;5@ye80p}
zrrqsCCcvPmK>%=
z4*9N#vE)L2^9OX2m<2R{_F0kO?x}3+AQR&pgLWsYnVeQKzez}YJE=mEZ5w$M$uB9+`4ogPPOib5OQ{iRd4fs)3nqpJm^vcBpTJNCPX`
zWw~;{NANX(*~B1vz1Nu|UwiU4Zi8eb>Y9Zs;PQ%Op9BZb=wK8Jil)#E;s6tz+3CbG
zRM1e^oJRgo5`6de#X%BwP2VCIrE3HP=Wy1l7E)Olf7E$eY5NJ_NIeNwO4~Mxe&Xa;
za)ii2ukxFRUm^roBLR=ZTZ;}59=Jh0M;7!ORmqKnV0q#E49icDQ{haQy?kujzzVxL
zLZbzXDw8X9_2mjCmSU<`O~+2ZXEAw0t1X{=7;UgES!W#yv*$~^XAv;k2E47GU)oQ``+;fXhkNN(`|use?6%VlxBQM*ko$~uA_Haz`!|qs+e$i
z=mPn=1VG<}?W1df8Gz=%
zW`#vFBC=wxVcEZ62xCxO+2y|R3(It;?Njeq9|GTX%@#D$=OB@>D1`}vWVs|dqsT8o
zvjd}FJSGA|@Vr_Byry&4^x8pq#<8bue~3F?8ex+7|;Zg^~K`Y=k+^G!o#ypG_VZD0^+{54l*J_mkbGmloTP#y5?YU1NW5|5n!v_e9r{
z$G{^m`D0Rx@Q$J}<_qPT|0mfoOP}7AXE6zp$N7RzLET&bK#@kVygMqCB$ef3f0?j>
zyXg^dD^+`tip^UzQ?zFf{XX-VpH0_07LQYcFFBYAJHfc56>NHid7{c)S@hz*h8C=P
zK6TJBVL6z_#a?~_w}S7q7yIS-DGUnr1}Xta4kbF-GpCRCo=EhvpWa4xqsVu#nrPvI
z`;5}Z$J|Bn`V@rBNN4mou8q&uf9afPpR6|Ox+fPIgH>n?2!kP+pE)->w>K=9A2`z%K-e~)W}zlEFV2QM|4
zpl>>;4UiP7254Oo^+({dlwk5Fqvgbn_k77amKO+M?OGf=-Ed(WO;5M7Vu-`|yV>`7
z4E)yDt4w!=znNlUQ1mcnrlSj|6prbrMDK;>wRl{^!
z!>y%M9K6r?bQS&Ne5E3{1MxP$T41q6c*VQ5gLT}hwwNZWK8Qs#PMA+h#<3&wyIiY|
zUT<`GhJH3+P)a$}C<829M?2mWPet9;&MCIJqJ95O7MPNK$GgrkGtYXf~f$}H2vr}I{0?JzWj`Q}O$sJ^s_>5^S_M<@zd=i17
z@ly58UZHGccD)vh#@wZi%X`A+a~f>4-1(DwzXb|Yfd<;$Bw`u3tU!4n$+<
zR{jbCNC^)s*)lfQB^L=oUap(xikahPTtO79sT!L69$18P1D`xn4nBw8HrC3$FJ1DiEbonf7=6or6UQ
zzT8I{r){1}(I4g3g8HS0RL3Bbhte=symv?J)vL|UzGBswd6^LF*%}MuQ!(A*42ijl
zu1%*ze;NANR&F$ND2ts+I1bGxNd$Pvrtxckh;H4jR|s1-8^@fSJ3=s$4Ss};fTT5s
z176q|WB+M3AMC`_gQe
zjx}Y1#r!~ltvJnZ)Uo9-%b+P&qkx!pGCN0oGoKK(H|5R2k!i3Z+nA
zdc%%0B_CSt>Di5JX3Ye9U&-NUuuo+g7(p{5rEN2BPT};hKU470wlZ2gKvp`l%Ik)0
zf1L(P8x^+mpN0@hLM4dq1L6V9f	t)`u?Y_!Me=adJ878;K9hN|!;zMEG9ggRO>-
zuSNO`v0ub4r4=*sZ7RP{MMG_Utxvs^@x*7EPwlV*
zFr)`w#jY|taLaqZ#SHy8R0K}1c5g`5Lr())0p7fzj2K|u$!n#&kfx*u=$F7je+@@M
z;tNvOPc(&9zxJ;+xllbQ#YAfK#bN(^9m9?#+>9z(W(gXPu|zGaB$5zHxn6;DjCiNo
z=P0Pkn8hFfCoI@&fQV@pMS`;}tY(M8RpDxL2X!puFg
zX+<+}_&Mv|l27@Vkmp))m~YVP>6kH$HXED-S^`zztJ94(dj*0<4uAf-dDEa%7BX{9Ze^AhV`xAv{_VRl!ZglTsbt2bckke%2jtX4fGa7fC!E54}
zM4P?um&Y!CI^4+1nH0`OQxIK5lZKlIM&_e;{gVhdB;7S=2NgOk23wJyg+eyhDq!a9
zIY~RFo=SE;2dbqbb?@aW?Z=NjV#}A*6+7%u++9wM7|i;UCu3cHf0F#?F;VnbLBCBf
zQ&MD)XQR7aj(fIA8w!oqu-v5Cy>-anIG&%iE-Uz;QkonT>GbA@g4wRyhf7gGFarYR
zvwR$s@o6;92N4Q~5pKJMDe;Cg!ZHYwQJKJlV`-iE5)&|cwbzx&5CXJ2HL-CmH+Ki!
zK8IybI`c>8duZR~e@Gn)P1zpEj<0nxGyP8QL^*aeANm@T7Qrqj>2k=MIYavddmxrF
z75FUCy^8!x7RQrj_tuUrwT4xdL1`26p
z0FdMQhv~@shE`?lEkdf@K1qhXmM^S_6~~Xf=wU&cUaSP38GcVQ1_G8RdD2f@8$;Vuid&jB@nMs?vcGxwCf6|*Et;blU665$Dor!VwfP$1J
zPnI4)klbG}}#4QXGJ`nFlf>H;GorP7POUKC=!x
z?-r^rCHh@6`xjkSIbN{DrmmZi8yUe(@k^g~zE2t;2)m@JbzaTvV{YVPdv0z;?@X#C
zrZ#nnf0*9YFmXuk7CH9INcluVe^}o7;e_ktGBwJsGUAU*X4V`^d0yjJ
zqlA_3UC^|WH?$yNj`LWazHYBmtG{z
zA#{k@(5z2{J?ZS1I%#T1u_XY9ZG?;o`&wv?DYl>t2b7d6WRTIMt0q-OH>X#MeBbdp
zCnUC*ak-RvN4$y$cge$+ZAIvwFKv$cfBy^6C@X#!
zFo+w0oV-&n)dnAdH#6LnRCRR&8W~ZdnCMq{gGSmOsy)<7(skD|Lr!kRa;p>iScLj4($#16%DivD-<2f
z8)k*6A3Q;6fs*a6s0hLQ?y#W&#V>jbNw2+$=z}d_=@*vC7J4o7`ME+*dk~XpC=H@w
zD_)l3&maI65t!S%p;RZ$u5sRuFXIv}#D1W9qjl`RxOuWRGDFu*f0wkde?mC&+>Ws9
zn5hxJq3ya@rRn|z@Fq)0
z?+)!K_F`|MnaV*a)>x|n@<|tquYDYM*ok?1V;YL5KD$h*;B%maK|87OO>4DGwOGyB
za4sUN{XZ>53Hbtj4xj6Ce~R-f#B;=yFR=u%5ZbT{!oGZjF`*?kv}BKVkp4!tGnJN}
zoA~V%VBAe5;0Jv%$+w0lk?pXEd6~@rJu-~MiDBq&>#r-YJCdbs4=uz~W7jp{I#Z1S
zZo}5r>DjVI!ydHfZo!yQXlSlhTgoV6?-4n7ld&O5FOZg|SAv2%~HY8T~UoRxEa(>yQDM`TF
zC?8jU>p!K~T(QXYx77hVV4tjy!J1gMK^c}(Q)|Hxf`bRc*8yShazIt*h?(8EpZb9E|L$
z9fmMR?MF^W)LvM10~okr1>onnLyI{Dxa>e+WpAhPrlLR!XSip3ikd=5p0f_d0LKBR
z$=(YGAi_^7rc~Lab68MEpfa;3%^(LDO0q)r9mV(a8QX%D1f%JmkOGxsTrZ(@Kc3TS
z;Z#nD%XfV}f9(Lda7r%FX0&Pl5(QDuhK9OTLWtKCvySGVBeyK7q&wEX+tvuOnQ!b^
zcg?asa4%P&=dzFj(rgv5%Mdpz-f=@w!8clLUGBXNeM1LJjal>m0GOI@9%&XQKwzVp+o%Qc{_FC20G7C_!PHzfC540ZZHMVm3c!z}?vd`u3{<
zs489~2?eum)ee)_euK^V1<|->(r;FAa$Fw5s2*Xt@n
z$g_T%e*;iF@h8w&dBEIv@&;Ff=u-RIWjtFbjPCs2yj2Xm7BndX%RM2R{xNnem}%iJ
zmwhLMM5Pvc`^*1#+AoY;dP`}S<7m6K(4qQj%pCQI55{4Z<&QLOaNH8!kaG>K;-UDa
z`>)_Vp^z5Y6w{E*kch%)(j6T9F^1f5U?WjQf69@4(2iEQhXyOPayPrP1!^|wemRP0>{FFMQEiPP)`s3tWyLGoT
z9|qXta)rLCnztIOea|c=?xwW^H6_8;$#UXqXqlHWsDykZ+uJ^|
zf4}7ePl!zp4mv1Sm*e0_XP8N-*<~yeS)uYJ#$+50kzI?#L)s{_@WB@GbcOqt_uJT0
z+;x1joNK`n&LgI28dV#0bV_PTxcO5tEw
z584;I+#rS6e#^EM+iU{tJM|-&;k-({mLxOpF2_FD1}aV;U*+L6MTN9EOolBwf5+Ej
z9N?dv55rVJnK;OH_HhcMNLLctOqgpid6?;oz76@v5;=wZ*5(GWz{o(}-m|FqqhIBV
zEwUvX1NC{rrnU|G)R5y{jR(M?(dY`1v63ql+G8*#SM#{x*&`mY@3TNGDn*0CHhdW}p%Bn`e_>Q?*PBT9T)pu*93p2}dn%&!ML4AA&Y@W8o;T!o
zJ60TA((Vv}Ov1ol3vqwK0QLi&1V6QPo1fpz?+TKPIqn#Y9(x(sKXZ>wFkI+gbQYJK
z(qv51)EP?{7p2s=-P11nMYU(%2pW2xuayCB^JIVR3jZH7`xAyULD479f2bM=AzHR+oe_2o%Kn
znXSTC;lMub$lEUu+HFyne^P4KbIecWWPYs6eqlC8sXwmARy8$+s
zQ3$!61F9HEO0mB^@9I@{JRv+0rYPzB5QJB;3l`L)v}9h)UrRJVemzjD+f3%ZOJA2F
z*?f7i5!?8?kD4`xR!=fc&UNCf3=lRmOVA3&XoLc+5{b(U??QIF>e%6LzX}ZR+mhbz
zlw`Ml^itti*LZdBe^yz*kF$Qe@K1bB^AAho2|4q+ilwW?LU%Ga=67*8FEObMoj|v!PGpJHIlFPqs{q2k1PHozy
zhyVP~ngFSc`aNTt=X2h3MQWDUaJO5C7*AeSxoT%r!HU?{e`_n>tusox*!=IdU+lf4
zyNT`k;?|`sF~A!-*Dtk$S8R}_qkLM`ewynlhZpu!5IxHN**K3TEjwL1ldasdX0z~a
zet&J70sfdN5(I-6^kR>7aMRo4TR&hq9TUQB)OiD2;NQSfGSXRZ@S@C#z)1ebIQHPW
z&px}@p#dkCf6J(r7I0e~5bPoyA+|~Kl>?Gb?og
zkGT3Ia$Th=Qj1?LMpE?3cItGkMoiE_x68PRFD<%*F&Mv{-5mRm1R8ztWPCw&srx;C
zA2VIi7FLZW+_oh~TGBTS<#Ekxkp_qA%HwiMz^BXpe>DtgJ0nMIJa_QY6b5x2E$a}7
zTAYLIBnArN*KBDfc+2&1oYd&@GQ*#l_IqS0NdA@sM5?SGc4OT%R1D{BEwo9D4%(^m
zljkN%A{rsVTHtg!DF_4wTW`)VCA06_f%a8j38GxKj=MzOy|I)>wp
zu#P;^e>y7B@w>^!ow_qk!}-xd1zRPHstPvNZ0u369%mD|hj#g0sh-aapVZWk*2~Zb
zUGBhr5Vun%*WAflX=nDc1|TY_TWNWS%LzV3w}3J;Q1CjvV>38o4I|)v9ar9Xy&+Z<
zCAM?5bNqPA&#VUp4plQ`6TX^3difH{XX~mw}
zX*wuR)>`E|$d~MUrbK2|@~)xjtq$e$Z*0|fZC$aqO+Uze-=tA3&US}C7yp=9;~UFm
zbRpnO2^7RkfD;s7ZEipC`Z(HBroF(Y
ze{4T@f8Q$Fm9@Pw{Ylg@y5t3FZr;mDC7R*?;I~`AR8~HKEpf
zW{;AsGFsu?339{*lq{c=i`J>!v;3Hxe`Jcc2aRBt+EcyvrJ;s$ov}tuD>>;R7r$-pqb8^zhH!%l|32J
z*f2aftQv@&dnq)WmZbifMma+21rr7iLB9ZIC7;#MOR+xe*}{l)Kv1i)FxO_2f5_#4
zEib(^t-K9L&V4l7-mn6+Taz_3z_i63+)%zZyKX|b{u0gLI+-Awe2k8W0cxS69D-PS
z4tR<7ZzJZUs|V79aBSBda?>yu8rpd-5LM2Oebr!gT*99^fiSjOLv@8LQWWkupNc~K
zO*brv^4WUlYyH+_!@@A{xJj|ke;jo~!UA#PBR_mEf1TZV8t9OWb%qR%Fhtea&cm`|
z#)SkwA8dWPExT3yG;TXNbu;mA#t+B)iE>UG5$01S$fXHL6x$`c{;!x@SU#%)FZ`j4bhym3hGGzOx2|KO+rYAqZus4yqT85x##kaT*{*
zGTI?^cG-s-i0x%$2oSw|e@1U#Lq5e-=sZSiX^&ZUHKujcODzl+CrhfTML}rsTsOD!
zLJr|kFKbiKuSQ>U7UgK;1?s_L2@f^EM=iJsNo@FA!g_6STa|6B@_wkxG`UibYYWaoqJ{cAv%Mc6?0fABdLz
z}eY^f!>G9Vd2PCFg7<
zS!UzL?iHO>7d)7Hf5w-J7P3r?pjNUDSts){p|_&UJ7;CKy(J%d1*mo1tfT*}x4wr&
zZ>?oW(DL++9U6i`b#Z=L*5Si}q{xD^$a@VQ58U8b9P#~Z$|_Jw5q+QZZVRVr9{%JO
z0DoyL&+Otze}X1(cbUb1C3PRxbjVLn{qNv4-L92>3JW4(e*+?<;#sDlVXnJeDVLs`
zMqob__=J{5$8W|Hwu^D`XW1Gk|)_pP_@B7!#{eoxe;aG`x7UFDDj5u
zQ!cMl$RiLR8un9M4rkR;-G6y>`97_6{3ftzeKlokU4Wtq*hC~MAmB|EIvy27H-Z<4
ziiEM~&^-KdBcp@nHLCVp$JYF-YnL}WE7N-ky?f$of0tbQUl(FTOi^a8Ae3*wle^^0
zD0&0Wl#_FYByPk*mJDON&sow^*@>b_=F6Vc_oUc>e4GHTAlj6%LyJo5a@{t7&HUs5
zihaOo2*|Je2LNBTD$~QR!xE`(gDsNFW-t_Xty&!r(Y%WZ~JJM{rV3=N6R;Xy0d4ydFGmshVjZ)RqJFlE?>6ck0D9?^uYH+
zu2a6`%@3|bG^z=!SXtmEK@E@YFIzYDFX0;af1q4}rBJcDw)QYRL{-$}a_FW4a4T
zQdC78H7fOjpu%9gk_**ke*k3CP2LU#U=RmFz{!!c?lC~;Sb4rq@H?98zO%I&r_$okVHeep-=1tt?iG!L(a(mDr^qc6?Awn_O6r1sZRI(+FRUUQ8r41RrX|bK`y$RAe%T^~CqaZpNE7=2E?O7O!+QPa%j|*pCDQbhP894mn
z-<{AkUK@58+G5%jyg`a6sEq71WjwUtF!VlR_VLqu%W*V{^c|UGW1OpMf0Jk*+k=ZP
z`-Z3MJ)9bkmnKFB_Ws$WK#Ln-m?OIKB#PBUB;gD2dx&E!gPh?q6`!NG6(KxnG}zTe
zbYx`;e;$sOH`bxuAPq70x`!&u>#0v4<&p{*qT&NLEAqK=oq`&b;S++7$UCaWLgyQB
zncDB~e`=AT0**%4Vu|Y;f5BZfm(02XxbTJidRSL)Bh4Ak_&(NdtEkTtPRQpe^IC1k
z(T1K*1e6a&(U%F0x(R&RwLW~}P!~#O9Cu4qY=4D4^KGf#Yx`Z5B@7Vmab~@ZKCt;J
zK|+NIR9E@=kD?^woDoq
zM)VQs!sg
z9X7n*`e&FKlA@uac(Um@@2OLd_e|HqBhLV~cPu~8d>R}Cv
z9b`|xC4P*OU--oRV&<^
zu;y+pu@g=+e*uY|DYWFvoYH-W_jiD0LRU%TW%{l9OZ8AyC*QjtJqzh^fhu}~
zMGe4bTt&iv8$Bk2XEb5JkV%TC2C#o_#^`ax3pYhde{v>n%@6s3*~Cwaqv@R;VI-rp
z-vJA@3(k5;bwfQG&)H@3#C;;Nxa5m+`g)44&(ZfQ{}QVJ_79nW%R9q>{i}HcTC4)J
z1(2eq?j=IFh27Ox}X_=ZQjkn6`?7E-f
zl%A9ke=PjTjKSBq>V>C=aizXNHA3+>=;1*yes;k=({@#XZe6FZ_!Lw`qR;{>tRupP
zX+`8|f%!mdZckwr31+OGo^#1-7@RO%XyQh|&)OBe`!3&oZqE7%vd<4BBf^kPF*)!I
ztkxuNrU!?R_p*li!GL(l@8uVgZ_Bez66r^^e-l6aJ#>TeO1uq~vBs&}o{uU>ca$hd
zK3bI=@qC5|&^5Us#>!B{?NHB+b_+T5rz`
zeoO=#v#ZQ#&aUVS8_;SX1BH3c{kM=MOUbutS|FSNrVlA3TjA?^uj
zKNhA^nQ_a@1{Pp2?)ii7GbEQ$5GZ7ixu!>?=2@XRgO>a*YiMt}@@9W|*VSQEXOl|*+U3$9UY>oeiaZE-G4y95Y2jtTSC?pfb~2%f&qSalAdkCn6LBf^
zgYzlGGr3*Hk_UyOCRr^9V|~d8+g~cU3jl~W?T>9cLCkkJmN|9vmpvpH=C{nnxvxfL(~z&8ml(5X4)bsav(?uZj6W1+E8
zWH=BayJSWAXA?6tErC~pEkg4&f4F%+6Z>bE#pNWK<7nKke}Tc)s??}S;t$^EpCon2
zkH{U_!+H4&-k8f3w7Mi>P8496r_3s3JPM-MZIry};_z=truJO_8c&O$Tl>2Y6%V9J
z#>U;!!6-oIPi00MUlgx$Ji0~rEmHX*VvXnNi%{QhCDFT2JzGym&Y(N1e_~kSpP^eT
z@%A=ux)CXGq~tMKW6Q&kX>sGTjc;Ps1ZPzdqVnYL-T@;gb%lQ3y#RWt;lAVcU|U79
zv6`)4{&9KI32NkzIDRD~!=Bpp&3_GZ=hUC{^kKI4WlT8Yt06Y@kIY&*p4m7yWa%||
zVl&9ch{GCO)Dfli%LmToe{rI^2;xGZT>)B}GJu`c9r-fd?QlEjr6LNDZPKCoG+5KZ
zs05)K1QLFJl>8khZ`FH!u^aVvaAvbD_d9E(&DX$DN0xV538`7#j&w#
zvpf)!axw#^O{~@Z2TL6K=P4>v2Fo+&`f2=dz_zR(T^F7;-2J)Nf0WJd5THJPWWpSW
zXWy%SR(C;PRsS;NFpumXYQR8d%o&M^7cF==u$_A4%;wgq
zXy6|QIjiXKtrrHW4+K*&?+kXig3I_jO$bbu3avkzeAv)?pd45VaQGVUaOj0`tbKtW
zi|2%os!f=~4$RLpgj(V_Ji;7@thS51MfyVky0nycRsA;Hjd+~5B
zO%1v=z7VZ3?dO3FY4SpVMa^KIBhEK@6JnJQyP5vqS{Pif76?*~bU;Y_yhI@+7PetL
z2&@B$VxAhS127BC6%;ryW$Z^3zy4j`N(ngBd38d(e-P*S&o2uLSwDLQyq|8D!?+?g
z)Wh=W|G{%nc8O)~gbhlRh%*hFw#WLD2yGr;M<&_GIxNN_n-JzvVf&zXo%NA<6QgkRFMhaevc1=DC$e
zBU!$5e+UIBmIGl}SBelj5}8svlDfyfi|U(avabGgS|gX+P2>~B&yS12b&$L5w`eU7
zw+Vj5W`uFPDW}W_Xq8~jc3fiQ(rV1GVG|(0Pr=!&O;7zsoA*)PJYMKzCDamOCa4Wf
zji3G&6?l=1&!J5tZGIi6#$R!Uf}+0E1CK*2e@Pjcc>^UdB@{(%Zo^Ebtg0si(!dk&
z*~Ezhb#-CfROR-PhZ}f{q;DX7|z5WkNsu7lPlb}?H7Ym
z(7Fa^N<-PlxxVLFtEK3UMp^bSJs(UMuRd4YE7|K>ez|U)?>?@^
z)d$$ogdI3%AYfmnG>FusNCrNHuj-ycuP-6{l)~6OxO6)Y+t9tO#ORZekTntSGE3sH
z8VTco21$FYVl@u!@KI@kAdM>?0YVfB)!d;h%G4QJlJzW0AeoLHw#Z}2dAVU@f9W!7
z5|%4~c}5|nk{;L^Oh2;EX}2YiDlYI`q`G31^g1){I-Vpec30mo3>J_NCnxIeIBNzb
z2tXk~*LW}b^rFw|`XR1tIsI46l&-MrYjpWovE475RaZwOyOs+f-d2&>*$i3
zaimjawRY{dc#5C6@MQh)l+(BNR2}twmrO~>gIB44ct@SYs=XKuv9TGVf51Zejx7f#
zfqO5xZlB(aD1=Y3gB@Fxl0S5wKGpvIB=}KC3oD}2_QvJl7W_vrHjT{ax>5)`Fqhk+f
z`19AdN?z!xQ0Mh5xg&wdzlRL#_S+xWKLuxW%*?H*Zwe?+dZpN_e^jmYQ`0o`<*CG|#L7ebzMR+MVZ)9nWL
zu0!7G%?4$SOnc60!2c8UY*5rt{3EwASkPc!o-#RqXr%UBQIM_{`3i-2ElV=a!mM;#j
zSIgnIUpQ@{wNQ-|b?JA)`7$amgD0RJG=p{Nz?=++bNy_-QzO2j#Kk&wjrMH4ME}b&
z(q@z8vQJfr#$UN=Ljv#kvI7+*$=brF6hjT2>u*`KwOi8QHc%{*-+FZKY@?}+J~yuv
zyJ(2}(}TE+ozhRUx!{I*jKgs3a_Rxwys-x)`8r;{N_{BLQAUemb_`leiyHY{cozEhVvCA2+8tkt$(K0x6K?eW_KtV*hpTg
z8IW6G-C6>)%RAmGFokPuedrP=ihjqJobW`T##c_78sJ&hHyemxDoEIG;u{LIWW$DM
z$U*j>&dLT9Y|*$nBO2{Pe{rpslp}Dy>lSgr2$r>5MY~N37
z6Y=&$f`MBHlILcY$qSu7Tm@dQiR&|Y6uoGyKX)VkExeuU0Ofoi8^_oM2TTu|KA1$=
z#_pKs42*b`1%F4>nVZ9mugiYpSl`W?nl$0QBDcfa
zUGt{(1Q#5GQ_Mg``DY!RH8F|st$fOhC#wjym!sT%>9Z%JKHhbGuXPB95edetzZSPh
zR~J}THD(n~F*DK?cTM>Abn?S8{=UA}B=7J13+es9e(8K@YP$SsyN3uuVZFy1aBES!
zeYi35=NO&G6h64|Hf0l57NabJuPs8Y)Ff*7F?+s)vZbyX8|XdVEFU414NhA!=lL;W
zq<)DjkJ@;;bIa*_KYPmjZSad=cm29BrBq#q=`pjh
z>Qh6x)pA|mZHFRYp#F~PPZ-NxeCMBrCQ6;xn(2otly#qLUd9`ak}=L{0#n7c=}kgD
zOo&VdmbER8N}3iRo3~o-+}QH7eSebFyaTmxT}gSxxzyT{S%Sk^O9I}8^c0o&r4CTz
zUVH2}r<>8##R{QkD3?mvVG*_#g8eBn9`1y!Vy{{E>+)FI3I9>*FgS17$nx7DXE7ef
zCV#)`X04X_a6gny)3-CXh#YRqW}c0Jv>g1opWkxULT?~dHPuEUEn5Aj{
zN+IqA!Fulm4sQvI$T~vz#&jyA-4X~*_Uyh-l$jV{)Gbgku#y(0-XNJPT!4s{HaLgW
zZe%9d9}UVP+HFNQp(IxRgBzPY87vYEvq=pdOxG5iohMls9Soys$QO(U8H^*@zXc4F
zmWB0y6*xBvLN0C=cFzCu={Bf)@L_y{nxt1^|LRI`delm;cstZ0dt~fI
zwbf~a!>*Vnb`&U@#%FeDm|~-8k?wh%zP-#J-97t)-ud|^Jv$HYKD-3CuDGJoTcO{b
zAEk7{Ske7=VBi3hU}@UghJXLzL7yYmL4dtHkRo4B8e#w_NFX8OoE>R$NDL7Xf#Hgf
zCM0oy7@9#eLx9tOkghE$4S-WDAORu=Tv$j^vx~ob5kCj2#}BDPkCX&}{Y*VAp2zau
zXvn^ZB+T87ifV8B2Yhw`AIinCUw6KFVhX5qLZ+h+$U$xlBYobs9whh1!K!l2c
z=RWgjux@XZ3g)ZQS16(+m7I+P(2c_RTb7@bQ&d#s7s!i8?kC0r+x!Cv+P64{6@}o7
zrQrs=MFwz-`XUb!8buRQ52D?cbq`zu*p1v#A@Q7J^Z&jdxb{S$rw28dd{TuFLLTN7
z-c>mQ@W5Xr@g={%T2vNRR#J6CA&MH3z<%#V#t#ECS)T!5d>2EbeGCptusrl$0P3Slt_o#P2l(s&
z)EJ(;7mI*6j{8p>icy_wAS^+YkQbQ`C>a}GbXkr-N~M(SZgU`SbvIcL~vVBSqN9hJcr
zjk_ZxWF#~!1hAE4dE`BU-<_AK83Hm9lU@Em?iVN&nBogfPkJ8n_&o?hNJ
znJ-i>5+o)D=p-q>z97{Xq=ZDp5F+_HBk?Qa9V7;b=SeX)YRTb{6#E*z2KR#ntaimh
z3Mn2rP((eXenwLC0|K_UVDYV-&h(wtP`hVNvTr5CMRIm@ZOx-vDInE3ij_
zXF@U{IBL)bSj|QzC`2R&0oJ{y0kk2CvDW)yT-oJPU9S5%+~RlHwLxMFzKare5cCdV
zEZy??bMW$C4n0n=zklY1k>c@HgZ?rp4bE-!f`9cH$b^#QT*kK!eGnP>0#SnZ6V0!}
z{{_N$>F@?0-pI=9>tXC2AV1AC0l^(Xj(niE)4*pti(i-cLw3^**2qHiS$1ia8kXtZ;D+DUT
zoZ+5evSDkHdbkj`cVxTMW0#cBS>eV!17aOU(d5p2I5pz0jSZ*t72M$naxlB4Dk+xY
zjTj6u5(+syyP9JYz2*7I*UE){aaV9QK0MqA?HfWc9ljrb>9{DKGuJVu!sH*xTZz6{oFu9=v-RQjHfl88F{|#W
zU$IeGGe=kXBc2AWH$~PyoqtiUHmB6m53W?=sS0{}Hq3n~tCx%MTh?w9DR8|$I;Dd|
zW>I3Em?~i~k@4>s|2(;8*9+cz#4>QjsNZ`zD;KbEfLs0)W4s%jMtF
zZ+-TKO3#{L1WY2E`{qAMKV6wd(_4HVQhYlLMfBHa(nFixYM16?`ANNtPPSK2xJFc2
zi-!}#hmY+Gd{U)$D_vxdd;1gg8k|U|2Xvx$?(m7We&SHwBqUo&^ur$;;bwD6!Kx~}
zvX4_I28mGbWM(kdzF;!KoCDF(f2Qjx3?=X)uasy8YW7af)&_aP%`Z9UC&$YSQg!yF
zo=OU(#xngghEl4RiIY8`RDAOfX-+c*vw*+DZfb|!pWwQuJW*4jc)1x@cQbAfyF_%V
zNH<)J&2=dfec(BC_PC`g-%s8YLYn=qVW7VnYtFpJL-|Aw1
z$wNoI%bAK~i*N1V`^$Ml870RD3di_n8sVV-wLL5|Cy}cFi!ZR%^2R(Md+61d&kpA!
z1Sx&o8Qf+*Mot5co#Ki!U6)@UuDcWGXo&1R;Vy~YpB3pS+W51&loXSBmH&0`YfR!_9tsbh6)D>v*q||7nRRv4l}^SF-y_>-$T*^X)TV&xs2ZG++w{a
zN$R5%*<0Uq%P4RUTY5Hn1Wd-NhI9VoDt(>_?(dMsclA>tJ4}NFcztSLHC3rZLB7v0
zdGQBv+vv@fkKULKs
zH+h2DU1OH{VaX%}>@xn|T{;hq#wxpB-;<|uJ$TpXy_`0$bOv@M(;DgIdw)913&b=I
z@hGZkZ^X7^xGp_R#KfA5N=veUw*@X4
z{%P4Mq^`UwlGwFo^*&uTYBcdG#)`ok&Bvd@?8(3>l2IbBf6VN#qetRxvhw-0kC1Ya
z^yBpX4YnN~K1FO!!DpvJAN#6s2%ev3>zOIcPb4ql8cF4o!%(-_3QE}~YwZESp2_(p
z+Ort(2?8A_r|N5sisa#`M!&YP!T*i(ep4`6E_|7FnY_SIZJN{Ghz5*a=L=Iui;C4#
zs+j@tiHlaFj@>c{5<8{#GE++YOJ!<~EL1c`rB}tH0h*BBT^4R?!dI7gkX_W~e>k$$
zJoS$?b5&zqrsW#zJCE*&J~A=W(BoHa?{ekmDGp@TsteN}%J}&DlC&5D^>M$Hn+r#*
z#E8gEHupkQL3$6z>(z|Me_&yaw{eA!MeTtdQ?j^P+Zxq9%`b9FWxbiuH+$TPOALmg
zk@-JDC&K6w^D2r_tRILDJUQN7%1vawN|l&EcuZ!22`{~2Ci**M-dP`&i%tahk30rn{zK70jBq^gamk*hE%}*
z)dzJR3AzDHlI)T~7{sD`8!ScmD6eAt+Z7=OcfWSIVg;K*Tc_w@Lt$_#O-_Tw^QI>r
ze3;9oh))p}Of((Q<2`(;-Uxj3f81}sdl$voBhebQ*<7fxp77qlKulQ%fRRiy|8aVI*ZE}qmNnL`S
z)AsGp5(ke$qHHDSY?ZK+wZ_nbL2e^Hv5E|bb*bx$$q<_^wDM?F7~h{@SU$k^Z(B32
zea|}g=$6q%*LRSzFo|+^sw$GoVE@Pia;446-OueL28HBYv;DW{eEaD3ye%}22PaJx
ziuaQ1nXmdc*slRubR2XRIoA-Fy
zAK3npj?W+F&NhO#R7<)OdQZ+fbu
zegs1bbo(>%z1$L18pL$n)~9a5xPv-MGCA@dZPdBq{t&!TXN#?v!A_kq1x4;9oT3Ct
z7rrU#?pbPQb{qo9u-lL~Q|pBFscl-<-(CwQhEa@+`wzf^KOgpk#66-J$(rNV5sLSx
zGtGkZ%)fcw`PM11j3VocLNoo?M7gs_BK4GW(U0P%e`zf`<|BG
z9A{Q~A&?lh60CwcO_Jc?hQ>=<48CvMTLuvEotaWduM`1SMz&U(8MLxN4EsKPG{08g
zhsM`BU4?g3>BrmT8h_*-cX#|LHx9)I_t^TT%|HBu)ppu9#b#&It+FC+KJ_hCj2wS{
z=Y+(IAU2D0asMcWFEfgr^qJ`UTA)NwYYk6h$Jx2n2T|AZ^u*98s=Ql&y#%45oSKKL
z!p){9>HG~aOnJ41$%f&<@q+FkF6-a-$gtJ^6O0@v%Pv(}DY56#u6t?JK9!Ao6pIYq
zw;oJB6nMLDY3|Fi+TQy6k&~L3zB9g*G7I6NX8V>=|@#*
zXWa5Elh^+GFw%X>W%sJ~-gjxyilu?xd*s+lNgA#|9_PdNGlYmLWk$9uHo`BY;`T_s
z#s#wrY?Ws2qtm@s8k0^<&DXKY<|PzH-&cD$cTR$1FNnYDX8{g~`^EQ@AwM7|!#9iR
zPlvBB*b=Xb5z+CZJKyVD4?`9iE~XRUr=eW_eeNNplK}hQU2mmTn766^TptmeI)&iJ
zUSJF`T&L~TztGop;9!S0ncZdFozIN!`B(B}aPMxMd$?frWI(lJV%uxF!0{t{g=Ws^
zvafoc-tbs+;}K$b68|X%OX|W#G{)c`yrO`J5_8R}#eUx+IqoK;8bi2%(ZKh%%s&%Y
z`m4isKSzUT#iBFj0&JQn$KNiA+7hgzUgCU!o%Zpk3&t>N(PpW)b-;pYS1!Z5ne9uT
zBgYpRe=?@L`z~dg6q51nG~4MqHJ!kpzVHj1?h>>PU-&N~cn?``9vyzux$K`4x8P$2
zO09z9=8dRL+ev3pLvUomPC;>weViliPV1$rB79^`wd1W-drXsShj`}`UpguKO-6-?
zcZCOUSO=qJK;F)Yy`C(Nnr~KzUAu!8|HUcHACmlw4e8>U#$#o%A{}^BhMxM&X?>Rc
zAk-E2HW%5e5)dHv@Ur=0p=|Wx5OlPGGH!~ilQ>M{o7+cMd(Qpsqntx8MFqN@-i7INTUerI7&Fgcz|OL=
z;4iMy(TPn!F`w2|eP`tDV%J}fe`DXE^u
zhS+IK7bdD=wdLHo9>!kA_LyCbH&GoE6e6Sicgj{Sw6lGpk{b^~=vu(!B5w2&kNr`?
zUs%KmN27<{{AXQQU>Z2GoxktotK|pu`EwrsBsb;fPH3awLOI5Fo+?_tXwH-1%Mpp$
zdmL4Z7~2Tzw-!&G7qCFkjHYala)3uT@0fRmm!ESFAtQbBuzw%*)phy{3=y&3S;Qu*
zz^L@_nj^FuQLeN8D?4s>akDYf=}@|AoFPJK_{xPg6se!iDRCdXMT9jTZneKwjI+}2
z$h2CNQ>M&f%CC)dc?TRtbonW?T*E^ZI%l|>;1Sc&q%;@x
zUs??M$i0-m7bxnKolvH~aTT&(s%KOYn8{f^(9S7)y68p2{_V)_HnL(=<>Gy+HHpv|
z5GZtCDHEo|Jg;ag7yVcGRBc`ATfN<0@6WTwls4oams>fac%hpO6nO1e&}1I^ZTZxv
z0m1WY$wnC`XNRmRjEwImH(RYfOJJO~ee&Ny4?J3&!RT+-k*4*hyIMyvkqA|=$ub99
z>o=LSlpX?f0u3CxZ^k1NCCa-^Jp^lfEspXj`^t?sWo$z?$e?Q6v1<$7Gqi(Me;rZW
zW94X1bWgGCO^E1C;2SG>l|zdKn1y-a1NOoalEd#dy+gud7%te5^S|7Ohj~e_mxrYx
z@s$GXY=K95T=%@k_vzh#;qGegZn?op3f}K010@W~+Fx^sAvbRCEg<6WYW0^0kLP`B
zZoK4iyLOGVk_AuWAE0N+$08$$eUJ4WfH)R}0&<_`Gn%#@!1RRPeMh;o1#mmxO~mx7
zS}g35iFKsqf2OQ$bYu>jOOZ372N*nBPm>L8Pj;yC4&Zqu+L}}^f>?YIc3p*oB_?%J
zLThCN41T>uj=#?a2b>gBd80RuxExLnkfH_RC}i%;v2moJ<-+wJm#zi=S`O-{7F}i<
zUBV?5GvB|G0Nqk$I-Q^YM2!@QPg;@v+&UEsxL4IGp;{mbyF|5YqI9>aepBik5*294
z3~57at!ka^gwP<7`4MLVx71(N8tI#&rhh-px4eAO1zs_S9
zdkjlQhd@{Tbq#cuouZCyUz`Po!K5NL`y(3XinJ{;C=Y_27_^8MxvqNkxR`Z(%2uY$
zI)$pSIZ836-o*H3g;08nvj-@TxS9fPoEfM`T9K`9kH-$TNb;Dd5(#Hfer_VFk(s{R
z@p5nBoxx*#WvE1}yIdvVcORVhouwSc{ebMOC}0`pMtL~(ZyQ-9ke&6JazR+ul{3OR
z;MV#l{KG5NSHt10XY|&}L}e=3)zj+;1ntlB*vTY=IpM>6BIpXa8c2$2ek9YXvcU@<
z<_~I2RM~{MtTP^tDeva=A+|v(GRn=lKiz#tq8@?*vS+cS*
zHo%yyf9k&P6b)l3*Gd(ob@p>7TRwS6NA>
zz{pXEOxC9^Kf~YnO{>rzQ!u*fL(eS{D}0QwaHq|5l=iKkJDceWwH@%aRIO;+YEAt8
zHI2f1-chL^iShEBKZn%d_w6V{;Hf0hss#=PWbW?Id1Vza=y;&GfqS_Y8`DJPq*L0%
zEX?(O=3qHezSO`N5rWf#0$m=qhz9gXhSKGV<5DU2oFUbDBpmIePRS$j&SsaZS>UOG
zY+gLtH(tX2eAIj5q`N~4@4c>Q_TGj6OZHrev(_g@)5`6Fsc@?6hlNo>l8y?qI~&l;Qa1_P
z*JM98z3z2T;?$-wCk-y-xPc@Uv!M7%_2!qadg4xNR;~eEp%Ir|Gz|h_7c3T0w!{22
zqm@1_3O{~3Oqc3)c`n<&*@|5gQ^aapC(Dn<%Duh>NkqhG&g1Zqn7@I7wJChX#pipF
z+w`XG{YIw#+!gEoj>vULwCO0%6Vq0COS1;eI5=wlL^$O
zROUA*1X7muVVL_(#<)14ZTt&c3maNGW^0+mS`L({8WUm*f3gO`t1y6t^oZ&4=jx%<
z$K~A^C^7=B^Qu|`o3_S2X|Fj9e9M1V5{DN-1Dddh2)#JbtWFy9h!uyR0hc`mrhPh$sP@{8^zmvgrKHnB-a4Yv2
z7yaFDLdqK_oSdA4{|k2bz60^1rVII8`l+$@WblJh5H3@BvZ@nw=*f6@vwM2>MEk}c
zEBiHl-aS`opc&NbZaDj&!YEc*2inHHHfxVh1m4XrX=T63Mr4rAR=RmO4A3v=Ny+Jb
zQ#mQ&JGPI8ud1x^G~6VKkr$rz_Vj2o6-=c68ybJ~Z?UmO4a^MqtW6}q+>wx@jcA4(
zJFplDBmGXgWP0##>~L1kF02f{qjoocsEM=mca=CzfUH$!wzojhdE`#6A!7E2w(gAo
zn>lb@Wovv1V)KcgG_~LTY-a
zW&5g(+7RF{oQ@PsLKzUosoYSKESgax71LTR*%NarEB2A^W
zrZi!Fn+dx+UqdYS6XcIhny0#*FQX?loovl8+bX`bJ(NheKy@W$(JI>Gu4^gP5
zSZy(G`VCjh?6`(duN!9?y;pSTSXLiL!TS6fDt{sclsw`8mv5rXs?YR;+}e(+lOB45
zE$m3|yf!HKVXatfojr7RFi+*81Z9-_0ERUl>eg6c38L#sV6(rgz829jffs!v+Uw_I
zER*hvZA+m;SBQu?r%g`#plkr-`Wwp~@=|
zQy+~I2N5E>_{`qJtaqO7BYIP=c0RlkS^GuNJaeni5qL^D!Fa3we0Y|dXSIB^lg^QgBo>|V!-on+Am4uatmHU6L{%^2=
z{W~kme+LWHJo!}LwiHIvq=Ka~%fC%fwK=l<4WhT9OlENjF&f9M(?qAGQ^F3WSg8&D
zlOOhVOfi8+s%)A(n#3`Md#OHk^>XJqVRJZc0R#$WKOK4RqiC*>AW6m0TY}GMN-@Cy
zLrT%nMAf{M)Pkf7F7~g7-dI)zhv?2z{f_g6s{qNNr3{rw@S?iG8^Fqcj1g9YoiI#~
zVVVM}Ba;01{9vp|AR6NV8o+LFQ#9L;6A~;T)I})zuo00~^ptAc{1zq4ZDHE5GM?
z0~su-u{u))#(h6X2WJg$Ijf#^FptLqQ0<2@Y^cm_2J9^akVAa>Lx;xU#w>Y(0d^*h
zXBuWK4Gatt@Y6>`gb}$gGFzjj7BhPxfj*=m)Br(T5=>nO#B_;(wGn3+E_BBbh!mAD
zl1Mn`0J$8bf7t6X+^~yvpU}`;0Rtn_8;&q3B+sI_9^fwgi|eDK1lMGWmJBfxIcP?b
zSpEa%;7n&fo)O|OxBvluXz4xM^VQE|{m*UV2GdBS&!$BsVE&L>+0+s_HK`V#z=d?;
zX!#&ZIr#X2^L1-ZB`lGqke;4B#i`&Kk2^Gu*_x<*#2;HP7t(<~R$!peSh?KT;SAiE
zW!HMb$lVF(=-p35e>HH>@wyNe&<{+=0dP;=`j01g$+J(%gvYv>)E~?>3^@i)l?q3S
zH1JxK#83Jf3=As}2^#u+?fM^u-vU4igAAdNubmV8Y$bYj7&WIMHzfAm>@^+{78B#k
z2D6h$Da>9fCD1Pyz~!LLD=hr@XS(sAkU=53`Wnhf#?e
z8{c;5-8OJV=04wFj>=KfZ3D9R_X`{!A0o0V!0xpyyrQAw7HBcsTE!9x8%x8^B_u9S
zNb(&A5(tJUhl0t$K>|Pkx@1AVR@}h|{j)?xKn6RuRVNV&!2!>}wXytq^(bq-Syia%
zMxbnIwOEj+!n01`Krxu*s2~{@M9x<+lNx~ix8C~=pL&Lx!4cD8HpHhVOyw9S1N1)J
z4@IeK{A~lK>s(w6K}ke8K8Es8LEwkWYp&nBAn#q3rKspIGawuY6+>t3C|QlCDI|*Y
z?OBw>X^%ift6!s&Rh>O@@6ppoXEv`*XAAuv4y3nJSA+3!!
z_aHR^lF3X6dR{>gMDg4x^25}=^a2CJJ1fakM8K8n;jOKRJ%Aw~zY!mw
zooemhIn#RMG5jR3B8G?CJSoGSRkq>6-OBYupu5s(UFvM*Zor04-iEhu&-CX}(qp0h
z!^E#h#+8AbKd)}#Qc1FI4GAtai&!4w`jqY8kJYWe43ZR!mV=^IKAkSyC9bLt0C?-#
zG0@K
z7;glfM2kU}q}7H-ulHI)FE?EF1@WF3$NrI(1=)6Tjlr7OO1Yz(ySsgq^6b+wMCb1S
zM3}NNY(
ze<;H(l`xwE1Gt)HnAOdMhIg8#$mEy9o~9>^c{H#sxcYehw4+k#?504Y@H;W!y7roD%gszcWj8@H&<^N<
zd!zIs!WlTLx11~#)cy|d>-7CC`iofDt0Jbm6IhrBy1^A=!A03`cY!Pq4>~;wSEgbj
zO7eYgT=$#vEc6-NR7DZGP(wShVO7RGJmyB2rBmIb@_$2V
zJf_e+$NcvjC3^?uQs{KM~G1|bm|j;Wt|Dpo#TZ383!q-Wsa4eC%+-0YgXFG
z1KhWVcgQ|`WXFhno+5wzSyk*yZ!hG=jjsIWkOE(zf2eUb-v4SGB4Of=aPe$xms5|`
zd!f=t+!`iG(B`eec7uk?=uu!2=6ommn|g+Iwm`CG$E{MU)BdS_S-^N@u?5oCA13$}
zn59g8h#6hTjn`sOT7DcZttY~=nTa5QkJW{}&0)&)b#?iB1XDX`s$-MY_ukcUQ&mcQjto`ueN#w(&)
zYKITuoE(HrRoJ9A`R}?xbtmZ@m-4Xy=!q6_H&G@9PEJ5MO)N(30*yaUh1SPRg%@h;!plW_nR?>o!3{Hovc9uxoBd#&a
zV;OU|RC~plI9TAr$Ov@M*KZUFJCF~m_p9r$&}tJ)GrznDYZsQeoLT$Em}i#;TwQ#1
zyWBJB`dkSN?}wj>?l=fuA{Ca`mJT!!-Vxf2VsLud#HA}+#skc
z>kT)^KHprAcM4UYqOkDbod>nhRYjf^1{F;OWkyR5v%7Y1Dy5TtltF%0b4~^?b~{0r=1LYjxKd<=xu!x
zu;D-J_Gi7#KD6Uy+PkJA3O#S41t)3`A6?u3dA^s&nG|YgoN@M$>nVBYOgzucjXG#O
z$zA3{A{lT(h2k0N!^Tx
zguF0oJpCz%3OQfEs_{(C+jZfLc$d@^i@juS@_f}*uo0|V&)^ZTMA}Bs78r-}*^M7w
zO-#TCqA>$;jr0v`SAnf}tGe2iIqn=@{W{B9`NnBYH%uDbbhG5V+p%N$z$N(^<$Yh5`r5
zX$*AD?skzlIB4Om2Wv*y4uVa?yoJ7kHfn{wGgOWgC$iabs|mWmC`x^J
z`l&)$M^`^KmTI$r*#Zk`S_{f%3rZvD9(5xd2`?|Vcm|3~i~ZV1m&M)vgG|`Am9^%h
zS1w!is_J^ZQ8>!k49i)Wqt}ZM=?llWyUto(PHtEJiBKCTX6T8loSPf&Rjb$Qx2f5E
z{gVuWboU}QUtQPy^LZ+n+wQfS8scf4x=Xc&yEo}gd%||WjU2^y$*UD?kEMH0q^uHH
zzVTPirOY3e(-y8;hd!0WH6-uDE-Fq7oV}+qNIPeekXp8Z8sNlFnvn+&8NtC<>bza
zWVOA%^eUzWX8kq(;`O
zh(Tjwg&lmxj=_N896%TuDYPv3tN~S&)@s&!orfYMvvUTA{Cv}nG|@|CyI^tB7s>4U
z5cm|{64o}$*mXJzNFq>@;&_MS;W?SSj%18rr7vFr7LFZ&Rgs-ylUXPK@_LPP?9qbg
zNH(A!;!G`E+}6nQtKzd2%tSt+MQP`7f@Xe!uHQ2pqHOkFz}tTp_k~V>m6aXn0#DbcHr8Fl?_~e%
zUh9;g%Orbix#MMqypH5X6{+m0
z4lr;xh)7LZ_98=Y>X7m$;+w`67HV)<~
z;!$0XC#j!G_)fH>EJ{j&iS=T?2lUS8JAlOMn-!S2htUg5MFeQ@e=1+_s5L3IXn?!|
zRYToiJ3_Z;T)xGPqPK#xRRs1g?#5}vWu1aL)I?t$0L
zuiGPf6TQQuB|&_t@h&e3sY^ZRIJB`<*paMB8I>N-I2cP39cbfNME2815#stx7<&jv
z_SysN(<@$oWIW17pCSC#4jc5>1%7MSiC}UX%zbN0JQX)+Pn^3v4wq^oREB<2OOxT)!h7KXAq1>PG<&JpH(b>!K+)12M-dajh_H
z+JE{r!0}%ILUK|I7)r81DHskU2OA4la!(5w4Y0*c6~lT-K#of)(C<6R#>NKWY^EqO
zo2yK+tB0%<9mxjShEynf-0bX4&*{QP{~lAT?dNyvqZ7glq19f<~Z+ocYw
z7^u}M{5u!x{2bhp9l%3^a}2Jg4aU$8MkmDO!stLv0^_&&&YXknbo
zdkokq8I;qldre)9+Z$nEMDT$}{?=w(IY8V+7ACByERe99z!pl;RLuygScETEgcraM
ziC7I-O$#gsL$&e4QbftApuLm$M)D*f57g0%_s7K3oZA?I+r1TRk+QP38e@UTr&Mh)
z1qczE`(-p1E?#wM3w>t?;+y-GVyS}CVXU9m^u#4(MPv>1LIN+I@Bjn=l$?mN4Tn|G
zXe3q|?zV!pqJ|B2*q#9-z)OdVi&~hIQ$ayN9p6KPOFb+CsqEWv)qB0|g)UeY5EUB+
za6bnH3;wBw<57)SpJCtF-39S}2`{uVi4N-y_30iL{9fT2l!Sf?nb`J0&0M?Qy!Ob=wt4Y5D8zpMn*b(u~Zvvte=E&|I4=<8-pB=Wcu*(aR3lk4lVM2Zlgs{)N
z-b-8jbr2v*l~_f!n8cZRW2F#eq7-o`@CxLVg^95X=CiFEiu^<>yD0v`Z9uPoPbv1k**s=*9q_adra_m;6G6{zern=}VxQ^qRv
zRDubKcnhrwRxz`zN(!`}j?DpqEQ&s`lgoKw^Q|vo^Fe|f9}c1}lDsibUGktYG3;kv
z9WS-8h<`lCU9RqxweI>t^lrRdvTI$kZ%w*A69hj%=!W>HsH;g`Vyj(E7_dLurtgL>
zpKhQGzZl97W#uBSWRF|(Xo*FDk)aed&2+=w}SR%HZMT@yfRuZZmz-A4n#Y4EBx<0W;+Qgpj0n(O45idSi5U=Ysi=?1hp?{ees3
zG6WP0Nt%5{*$*nTwtET~12u)r+(CZR11lg15&WHN6j(;0V`9{_OF^tYF!@UbQ=(#eoYtq$+|VS*TWVZj
zG26a@^|5irXK`*%#(3`1Tvyuo#EsxAuqg#$^Z?TJ<*fHTn)b?om3KH>!dWz84>CYM
z1jwC>ek92M7P;r3mS
z2Ws4}3YZ*V6e0{v!U;;xRjzZL8yN7@4Pi8&)>eKAYajUH=bG$k0R*E=>dDu
z)y1D1;C4}-gE9t8DNYe>DoZBiy*J=-hz)y%dhjMk7pkJA0UZ8`eechnYvKBm6
z!tbH@kD{sq3z`v&O>Jg^W~*U5JdHZRJVA=Q4x&;LHv6Lj5_t0IOc1)?SG}WhDK%3C
zdwXL<(;HGz_t^XqJI
z6B51(*(0iaS0>}HKx$FoS$k`KN3<=@$*TWP=sP;UwDMjFWT9kRV35Rd>VsM^4d&+t
zB$lmljZKmFw0E#HZEJf>U@{W73W5i?F1EwNKtodCoLN+o_v0*KF%%o&fzPJt7a1AO
z;TfLe4PU@I3P7(#K*2n_RDB+*O$D}Wl?yIj;Bucv=kVL*t+O+v<%$)=1lFOnY3@k;
z=_QU{!I&|Hocj&|3^k+_EUSpuuUj!p@7^m$?W)C=^5fd(FZ&U@fxt*^4kIN&Crx%R
z*~unX6`QVp<{YclAz`)bzfw?4_faqfGPBIYenf05fLa%c5-;RRSMEEOM#G~)&~n;K
zjuXfBp=#WDs8gn1egD3HomY{;%{AAI9$WjG@LHKUmSn{PB#7V8
z14OjVPibEEOq6?EIu#+ck3v3@C5V{7na7U%XrBf0N+Os*duc-bpy=C&Qya`ev>~>p
zmIyUY$L*XaoNiaNQ`9euT-5sOmGZMySNYfIl&4gW_m~UzAzdmeS1$M+dYEZ}M8sXW
zwMx90{JvZpu%l9NONB{EsCD~ORi2oM31qESH2?Bh1|9x18II#mQ*jc|oDt{7_Ers6
z`d4YEXj1#sg2~XW`t}wYPf0!_hk*YM#m80kEVXYNP4rjdC#H!abqYf~YR6;kxSqMU
z{p)ac$JxY$n|cubj+%SAG)pa2;cJhIcJlA177MLdU?mL8mHQ!&*~!xf9sBd?NW56o
zk7^8R>flw|CFMP;-xAre_m)aJ@w2m%$TUbClg;~7@^4nOmH6nxR~sRO)}uR7s6`H
zKS%VffIf98Hk>IAPR>Iq8N3=p+Lie)blx@&5(Z%x!^(8p0qblJGs+n%{v#)De@5po
z$OjJ}3glT{d~PBFuN$>o>HAoLvFCruRJoa!EIgpdXz8umv$9)gZU;Pc8N3)~rm*8x
z^3K-69VOFU=AkHumcEC*j#4qlsQIRL+h{mG06ns&K6hVIY{&4-rCtf^tL%2rSF;jR
zJRdJ(TXNY%x7+;oqWV=)%lFnxqDbs;M$Q+1yk_t;=cIi17q8x-*|-uIGrhMQJ2z@=
z6pH+gBO-(Ej^3VyG2t*X@$H_a`gyi7(Cb_JtJA3hfPloG-@sT!_*0BH1Nks6rU*vB+ck&o^)8FsLHQ03p
z!Nb}(^H>R1Nb#6Tu4PT%uD9^miIH2P2H=M;4XsZv+x40c!*xUXS8Nwuy>))X`i)3W
zpc+Ogxhh%2Hhtk6qxz<+(V&pKvZr$MQXmdxo77B-OG~TDOT6mCHJ7&;g@r0}{j|sF
z(~5}W7iKzkFFt81n>aRW~t6wAG%&djnDh@XG<{{T!t
zv%fz%(;YY(#{VjC)yVMIq&bLX2>axlkAJ}YRs+>KsL7c*iH;&D`tc#^1~sSs``Gh_
z$wq#En>TipzFnk_RyUd{d&v##vP?P`A2FYxP<&c6zIcCP?VYp7`gjr@W(1=laI-{Q
ziYa|8!0y_n)aRmgojJL9F5d@*kagi6q_0W}!tOj@(F%T>#T$a;h0ftE@>MU22+IIa
zhyK>z*KOA4H796clZ(pZf+}RKiFbJfDvOGlpeqt7SF0?%oV@h=|Z
zJtwO%TPI!(ktTd0aRlP__>!+aR*Q<-a5Eur4y8T(a24Vsy)fhEG2GnUeafcf-a6@j
z{NhkjgX-s!;OVM0CC-P}mn+!08GT2-9^_Knvu{OP?(Hczv$?a&!xx2pFZ^U@^ft>g
zw)Q)dmg!|naxCjUT=pE0cH})e{A*g1TCu^ja#!Jw{D}AW1Jo82Uvk7OBqW2{8-$DP
zgUn}ndnRE?n3t61=SegOH_#0y+?jcQbJm4EyP5lZgM+b*z24H#x2U1F4|xjr+v8f|
zha^|CQIi^#H5i`9b`ZLt*Y+}JrFZISG0U9l%GVYaavk|SkJ6YR9-9Av;Yk63+2O1p
zW3aAKRV~$T+9&76b9aX(KlHPJ$e+bOtk7+DQ;CZbSLYNk>3KHr$xU<6&TP|vJ^Hjh
z890+)GxpM%em~$*=#V`0#YWHM5`mG>j*L>B={4Ip4-sGE
zt8ia7KU83-w1{B&^nTtugV5#IJYDF^NztSRa*J{C9%HW;X3N1x)S3g11zRUnb@|s>3f!QKHJ6H`QL3Rh6t=C~$mq8e2|=lr^0h+S<}K9IF_b;mkZmk|xG>a|^D=S$;zz!4H7t;g)Oz!2cU~mjz+32zglANu{II?JK>CGp(#Xtz9z2I%58L$6
z0tLn6=yxJCkR6(PA@DvIBWcybZMTJ+R8XtTT${j=fTBWf&R~`Y&SMP0Ym~_6Nq5SS
zjM_h-Uuf((Lebq>ZJPRV!CFFf_VhdRv#eu9Vn{B0-26zVU2>b7_6VP{Ec?9#->j(z
z%Db0=jNM18vI$}BUrSSeRI+|JeDCKvLF)pNP8*=QIt&jN(5;^`^v(0Jo-5qJ*?bYl
z%i}B>WFrYC)$7!>33H?2(XQyvQ8QhloZ4&^6ZTVZ5&o>6&ybJ>lWk2irkR&AWr)}A
zHr5G#uuiHy!e!5eR_L^@Nw+X&~%~GQ7R>-e4#{(#k
z$gh5bcfNCgD}hOT`lP8`Lcg;R_ZVg!DW+3LW{PxywRsv=dU_{-
zbdH1XTPQmpW6{XgSyUhsa?vXjvosM<*#8`L#fj+-$Ie|TT$=#`flJyv`|T5(P%`0zgs1W4`MUv94d1a~`mVfBx+oO7
z%eV8YZ`p~gh%M8Eiz@4enJAAJ&*g@JOl+Ag}zu)MzherW0b7
zT7bjBvwSzsxJ>Gr&xq@`yRRZD(;cUpnoY2UhHD+#Y%vhjmb}}x+Y-k
zRm9K-_r(KY7aK*LmjaJ$QW(_+!k6vpdnSOfO
z14Vyeo!K9^mLX96@)?~Y(#8|=tpqh>WZfQ(aygTKr6Y(0RmQg)^DV9IhBX&KJ^R!k
zCX2vLT*QpFc(?~LdIc*^H6gHXU*D%J}|0(&gXTK
zYK<{W3xzEj3rcsS6tC*P6Eic}8dStCZ_vH}HW>c~K%?JFOyr
z3ljC%DZouV*MDV7{_OZw?YS@NyW>4K;)NAzy{B;I0rh(liTCTdVj-q?4ptvdauty#
zmugL_-1dB%wqw-OACLS!Qt+sE_r!pb12cI@FG!QW4t-rwHbi6-lYFViB9oFGBj}C%
zf#!2lkfrgQ`Xxdu{@4BK*<_ClU7@>wex+`V?q4c5aLV*oB#bA$k6&FC`L{Fdh`zel
z-jv>dS{Eryz)g!qp-veYwBE)Bu6Ds_95X8&zW#|fWprhrHO|}1L{;T+v}+4ijXlh|
zH-YpVRoETWo!EPxIkX-7k+R#;@u21qsTGKdD`UI?UUh%H3sHN`nUxmDZ@IjG&Ob|d
zH^JnL7&DgcF}Nqkza!$%ULKRqqgjSC_#Fi%o1Q>nPFM{!Y2x)2(G2EfdtapwGiKY3
z|G*p`wnrq{=+Lh%*+h$~>fp9=9TP
zzlA4Uy{F6SV*w9wID_;Pi$Ujq+hvF;ZA#M`LSJ4KK)hO!&16w=&^X<1^+<)kF^h1d
zF^%fd&l!eWWDyVXR*IjmsoGCh6u2RGv7zntivxtQh5r!6MVmal$y#sgI&yU9HH_pNq5@GA!`jcOx=r!8xMV#({frL`fa5)
zx(H|Htued)$yJob;#WUDNuC3m1E>-OrR0j=yw-my!=&9{$Q?f8T(~gAg_t-y|7+`Q
zxa?$zyHtDMxBPb>_ceNCSg0z}{5Zc}_@s=Gvq)Yi)~bY@t)=RJT05w8!Xn3bi@%;R
zu9r_R-wo6w@@;s~c+2)kZb!DvdljhawRG@sehq3&V^9BXEk0a&;rvV9ZOQnOp5zRp
zMjO9bD_173^|(jp4AZ*ZYgrFB@tAi&JR|h<&D65wcaYAvbQ7$z3)A-|*^joClxZ9n
z60vQ0BFjIGY3=fVdb2}g*|HqGepH#i(feUJ^iBKzsKEEfUD7SXvt8N9l#bCwvps@#
z3+^jY?;5xHNSXZb7wN4Qt?hfD3tXpnPl797_9zK{)VI=1i5b5%ymw32pvxd^5BT;o
zsjU=fyIbEsy`JnjKjvaA2Ro{IN&OhL<;<`4TDf(f
zsL?WfZgou!JHEMx=^wB)b}4(KYvXcZ@n<^Y9q9`38C@IaPYBxQNfmIw6J7`7R9BFc
zrz*!t6uzf$-IgRS&6g-K6%J0DVy#{PxKJp}>ZS01BdD5|t%=zbbwI#uTa&`LY!O_#
zJrE+g5}Gq7$-J?`aAbU|W8~x1FDnx&f;hsgKC!2jIi%9?Law|I;h609ofO0eRt%YfO*{2h%m$yCE=qnBno_c)K=h*T+(?AD?}3wJ-G
zgRPK%(P!O4W5U{-F6PzNTPusuIQG8@J=oDr4N{T|vcv3>oDQa2+P%UJH2>oCd10A^
z0Gvy00l+OMia@LP)ps}Qm;ChZ`?Os6Xow~k@2j2&Q{Hk@v
zTrr%yoIK6r6O>Psk-Rg~9iyLF^qF7|2&tKWnftp&G>a8E_G~i1R{_s(&-%oaiYf2>
zzTCoNrNPb>&R0MVGu)g;jvWX`uJ>x&8eGi7xg372b#}o%dSRi4V52n=+$sCFIY&!^
z01fm}lic36fpHidC6SXZs5mtPhxC_DFNF(5ka;rv6gLS(Jzgdn{UE^S3lr#9G%II+
z`4qdP=8cNo3TB#pp;Yzsc1i8YYUNs=g0QQPykI#yz;ubR`S`r7>(vym!~{SPGjTqG
z**lgQ3qOf(=me`J1lLIKOa2j2@Hj7uTa6
zT7LkS69wEUJJGMS%8Zg}rjbr}pfJD{CW%pF(3*9Kr*6reJ!|AAVrj?o+o;rketKV_
zRu?nt_>mY`529S(`8L9*H5^1nYts_Ja8egoX8aR{4b*th!YM>K8yIV8zv7i4RZc{~HRaX=t~5LB#{S4cb4kEJdvy`Vcld#C
z=p(w5YeLURcaL++XM{5yM9~j_#4a=(bNyB~qaHBA0$9q{jzKziMZegWakMT>q`0<*
z3fqQpYi~ik6ybt(b)x|e4=a*e!+t`%r%DQE3r=x{6rP?l*-G--L2s#&kX5u^2dgDT
zAw>62&qbtAPpdKSrOXlMl_HT?qV!|ly~W+VwhiC|nE@RBrw8
z)xoSlhum_3pC}N_a%GZ2$Pp1uOtiI$(sc*9B22zJgRkf|Qra@H_{bWB{p5h|Ug(z6
z+rf5$PDB^0scG9aj3~9X_8pGf-$C%BRk^~>;M-w=j}%1hMRkpTly7p{KPg)v21ZfU
zp(10YV4h~ZafHvK$0q=lNh{$lu5}M?%d_
z_t092$|A!>q0j7pLwv-vH>QE?B;#n;0jxZ|Xf>+^#hiMy*(~L6c}-@6r#LEqFK1d|%{6_LShXP^TYZ12
zEusKlX#Jcx4WL+&6YPw~*TlB(;H^t!XUVx%GKb~orf~>aT{kXqDuKX+VB5$PlGxYq
z%w_iu#A=*=*hpNep32Qc6$+A8iq4(=z%~b4-$P0H$a$JAADW(Pju*V>uz{=UKd%|i
zv;P(m{a{vqJk4KVVoDYQ4((Zwy7A+I37m)?V!@>Su(||U&-eS9pU?8c-kMnJ0$`ON
zJVS_vOb3IFPYCz5RF_VC88dK`0`E?W)ECF!%jYp0k{nAEf}Is4l3}6kR{C3I*B6X@8L_$W!|aAXO~i_S!}n00*=7h=;%?i(RdiK85~P){
z%i0N!aUA!`kji6mUV9I&z8*06#oktRMm9cfVC+Zfc5vY;?Uf}>>IHgoE9O6PKETCB
zTvl31kBZ)20P8mpC7-N+I*2DTsL$%|fw|`|-Le3flbEUbXNu>~y{H|8qg!m`dNk6s
zrzOpQ=`S&JeYQvAf1h?~g?sAtEe6>Bs?0g6N!8=oO2I7<`(1RU+$Hi}2qg(oU%i4E
zth^Jp4+7oWXs1^}U8ow5{XP6ZA3sw1$}+nibXUeonn+DZQN3tVDEhYyCpz;%+fn$t
zy@EkHofc`p!5YYLq1C*F1vN+f`I9+w^ZVg{adhiI#(vjcxE+BF9{~($K*Ybr-E6jRn#}(nX68mM3aP&*JtofNz}HCQ8)n^
zr*c&n^5PV#A4_~AwOjrAf-QC!*+&xur%$^Lcoi|b60WwGzrBmN-sD>sPX!l~Y+S8>
z>H_mRsH!!kwd5MG5-R~
zYUJ&bG^kkd{U>QYva*(98O#sh6NQip1Eervc1fxyZadRiITv!C}_5zRmJtibq?+v+)T4K(^?X%>>L4HpW#%0uQ9VO
zgNHlvqEHN`J{UV=O@gJ_nMNl*tX0?u|I}IBeEPL0m`)mR$40w9?;SeSy&=K=b`6yhWrn?ZJ|hvD8Z9%Pk{I#(}ufkm>T=uG=Pb74)_K
zo*LCGr*|nUDwl}ukbz}gz_Ay9kwqk}Xq@-73FgoSVLN(yyi~c(S>V5A~MS4wkJr#k6xpKiZ{Mx2QjkI)spv(I-@HnCR4ojVuW(dINaS$ps&1<=hnsn
zc*38JSJDI|y1*-p-0-^&i(hKlY0uE+K+IbDWMR|3($0%LO)RZLdKA{0QT>__imibG
zH^Sm~j3Le5HY5=)uaUohNAIZtFQuAd2Hq?rz-FyTx~Ox03xy^14FDn*lFKSwGFul=
znk}VI0}>J!l1Y2WAZbbOYKP(lgYsEMWVmpVh5ZYue#LzdXTADR)5D3c6>R4Y3t(sD
zqk}JWU+O*(uKJj<kpqL5exznDr%
z36@x!$p<1gzjq(r8mXf$(Ni$|^lFeRUf_ujffad5g-1hACD)42ieae0k9&15j;3-1^aopns0&Uw+!?FU8HJGz8L;NHFzpcdytJ-4L
zeloLq&tI$bj{Px
z08RUj=VFFtKpEq4mh1e=wI_CL2hPt#_t~MutWpgKmS>;8jxcW|Vd62zpcgTx;%M}a
zg68|s!FQ7ta;(N`$-b)qjM)qPK<(OL9df3kcT3!}y{qUrd$xBQTkMcn+`}s@(JzJvU(7R(I|B2~q7RTp-_vJ(
zCBgD??NWVzLa!z}3-@y3^8JySPRNE7&;e$QGPd1^(pffl{)%HfkPwFfap#HfNu5z*
zhuN4d^6iOuN3*NVx3=pfb@j6sq;msQr(B{LIMz46mg0P+
zmT@*dYCVwLETZ!u&Q62QS%0~<(9)WWmBDX6IJD1yrEmp~b2NU=ZSgm`o);ZSTMc%i
zFZLBUayHU1Pe!MXv(^2YPIBjUc#Ux{Kx1;@ci&_NyTvhzemD|iL0&DuFwCXT$x2hQ
z|NMr`<6GZy;`BQWIaya)gG)_jE;9onNGb+gX!LT0*7FORLXm2z*KGdV3*z;>ly2Ib
zwvOk2cLnC!oCiBO!No1c3Z};!JgH#~{>BV)b<=qAXyQf|`=y+ix81Ev;bfP?goi1D
zrv{F`@-auADdc-)eRemg0aO@kXRQX$I*AwxExJUQPt;?BPd~jJdneb~NKPuEe3{bH
z$$3^u_EtjEGEqZfW?$Cc@{hgKs$l?hzPnL>-1gr7fSMsq&*1XC5|;!4Q{?TCm(x?7
zD@AeJ;z80D^q9L<5lI;uBo5Y+NNp#}O8uhky5czV3DzBnH<>Z$mR_tEf7t*zCX01C
z_jW2;bkVR$>TPJn%8#7p-VK)gy!4&r3VLU2lWb;;cbuq0VCFYWH#_$-99&p%LuH45
zV<5qK5(>`-&d9u}MQD>^!>BzYQ$OPKxDcpy428fbq(;U)4~0!4f0VYNVYk4dJuc3U
zo}6BgA`CsKZ9~k|aqI_Q-EZDi_yY~5C6gj@@8L~uTHVb~swMOEU=tHIG&O^=q+s6P
zg?&k%lorbAO!S3aP)JVAy!KAmNY;~o)WM&@B_0EWF_G(}gbU%F<|Fbmz;o`~Duc5{
zXE!}Ox#$pXXuXzK8q`g~XmMRr+
zQuWBFa8>!%yR31#PPZ3Z(W1buUMc-8!%wGpIlk0vrm%FrqkQ(5lQLRkfl(6dbB}z6=x*I2eOEiy~QR=$GxArU*N7KFz&%wlO<}=
zD@rsiGBp$5cgo=
z^-fnit%&Eq?1Qs=(-ls>sz`+E6pR=L9oE1NuKeSAJ^dl#<2Q%nH36xADGH(CC*IPW
z)}WUT;E%!s*|O>k%<=3;d9AzJn~kc|gbN#=T=~0~wcU)xWhm;WB=k77M+_I(;W|J&
zJi)FBgYZ2mlSfS2w?s8kea+=$hG(nm69@wqS4n{m#;
z*o4(Py8qodg&O{0r+lw$@MOGhuacEi0@kb&WqHHqWIdq}Kl@PRnmAf}M!2|b`)seQ
zyK_J~rC_sWGl>W5v@_ir75uGmu8`mQ9?66X$2Xo>HnUxcxqd!;MzS9pSj}JVnAXZ8
zQ`~?nMsv@fYaA(meGYCZV)m^=yPBdWc{V?85T6#DRkC!LAC1#P{0PHGRS`JxNvO(L
zj5%^1HFdw25c&&ayfn^17_`vp?r^ujz6jWKvTLD=3okokPoMjMQu{_~fKGlz)_GeWiKv2Jv=VMQXW!u^DiU)YT*juN6k;HIV-V
z5C4f2#k}8tIB(DV>J>}^=hL>n+g?#sg%%IE?$(i=^(Il+p%Hca_8Rz@jztmN1I*&T
zx#HANFt&Ruin&M{E-m9|`I;Vd7?CZsoD?K6y(5|A=Rv{0IF@G_-)3d1qI$*J1zMZ(
z&{ivlXmjFr+u3Nv!Bqa@p5ez*YKOP56mZPXL`mv@ncp`g$g-_Xds`mM7^9Ov)-ux=
zs|48YaoLX8pPVOULYh~|PNczuYhS{7kWM`CX4Y0-_N+NGNqF}irFB|E
zJlwQ|VG#&>XnSm8UkV=@1}IRzda!xAFVIt@NTu^F-%t@T{(LVZ=wA%nnyDR=QZjN6$~hzxIpE
z_ET(P3n8*vdsL(xYR@j2#CBt7NENH?gr)C)?~p-$sasRnQ(bheZ61>hNZ{?B5hE#m
z^Y{od4@4iq`h3M@^@BKT0QkU+>47q%%$n9C>sr3imv89Q!JO@9ad&7=%oETQe6Do;
zQ%i9}Ue-Bqg(~?jh$VtB;e??8@%1bChF3x+i!qLH+~gyo!KijG$5)?D-)*7Iu_CK~
z8-6HPljg#d8cTA@C>NcK9upd_=81EiL#YEfIk?dSM@Pe9A1A}Uy#C=
z#7S;0yrpbC{pq!k&vhn^AoAjvx9tK2nU)%KqWp_mAE##puD}2{+^{1HH*R59FlCT2
z?F~8%rwI9p%Aon}-sE~(d#v$~kXIUi^-kfoLLu)Gw;HkfUgA+LdxN4~Y)^KM9Wc7R
zNPgm}cm8}IQAN?3nt)kfsOG`{{Se&B7EEs@ocW5O;Sd~S8GHef7A%qX!8taCpIO@0t4Rs?DQy@Aj>_PV-_e&CG=F;gyW3q8SfIFn9lEot
zWf=vBSEW9xn?8>xlbYceTq42}4*(iDvn>S_9|^eq20r1*{VKrw72OARv|PzNcZts&
zl9wTx@6#LnX?gPP%H6h=lh*pGYcUXkxiqBqb%CzA>Lh)tI90}q$^lWH{<{~sE)p)~
z@O|A|5z60H7b#_h(Qx_mhTdy`YoC}vB1_AX+^18#==;lU4CC7jCYlTa3qg2Mb2lF=
zf3U=nzf*BbCh8mBb`2sKNlGF?ELp1Wbsg0K77=uj5(;O3b}9Fa>F-RwD_%@!Bx_B~
zEV~QLxmCQQzw0K%a$0e*e`wd)O%p<8(AYhXlW1pfeGDd9N+tucBX!w-tWPG!XL3^_
z)bCJ`aHLUUVc=QTfH9Qov-s2)F$jhC!);6U!$0@Je*8+Z9|P(#ZoWe4mM0OMUSgDc
zi0IpzARd>b`^nH!Ho>qgF3G7o&F$(m-e*NxU#)nVj`!m86-*hf4`P}7_o|}j>ZZ;J
zX-G9te)9axsaqSR07VLaF;l{;8gwy^7$subbH=kA(wnrGKovqr$O!l(5>uFdh+UU(
z!NHyA>o{=rag&8OX@W(QPmct#&Q~o}wL+>4=Pi2f+*)u$Z$I`swh^&LKkBNVi$PcM
z1XUY7P81;Ai$`rKb~2ZaUc<(04`jLwi|v<0ODH#fJhX3$x@qly*AZ@EtyBb~kQfUE
zvubLLifU#dKih?kIK~NS7cNv|LT_LvhNvnmAwq(Di?zsYm%^*Rz7>-itxlECx;aPZ
zEhA`3W4{Jp6pq6`s}cs!r-F6iO#Fn_caT3SVWscs<}cN4@$!VhS#vULbt|=({p|WY
znql{=^LZrPBNhXHsN&qyy6q7rv=P2q!;m$vrMpdD+0A1thqrx@$fG8R8~n`XKhk$s9>
zRH~6};RY{OZnJCzBKsj!2gb&UXLhQo=OGFRvjcr#%whK2PXxm+Xcx!q{shoh`7TIH
zv{i9J^q`iGx~A%uYI^w|%q%C_sE-y))+;wgbA6t5%?HP^c!E71{dJ_*F2d3iYkATL
z51v2tnRtkQ_&OC5Sey+3WQyLxLR3&gH>I%W^(2u}oir1zxIq
zIO}poy7W5u%hUq5#z)cQSi`!nq(v@K?l>LiF>n%_
zJ81^RNHqhl|I%#(h?nT;{jLw^**ARWnXq4Y8Aud=-=bzw6Qag4RfcP?D&qQBSIuiA
zL=oGGX45KVjoY0`*uKD8!+dgUT{vB5IuFmvN*9R30Q@>78qRHM{oGn#wV2Oi)u~J%
zIE1fpM%6JpKs^%pniJ%wP9av$xe#8+dc{Tg=5z~_f3xONLnyXnDEn7oy+!9O_5=rw
zmSh)yLbWj@X=ZgJc)0TZXs88cG)2`-LpdLem!XP9l{h;msqZ#gG%W;ho4G?i
zm!J9Qq!Snne1oNKg@oz^S|+i9HS7^6E(Dgn+U9;pWUXb0Nv~9En-~Xg1no*b1&uel
zJ(mslJd!Y22-nivJqsz)OuZ)`YK`td5jP!q6!
ztO^QnY68`VOT|+AA&V~CAj?uG`twDYH^5T*qr8D9@eAPXG_&Rj^WXW1E|vEX
z@;0-c>ft1X^O|Pj!FRx4;JkOafkqydafOWR~O87>oce7
z)-NsTuoF6*DcdzV^Pf_JqqUY!RLIDt^VFIREsp|F2@mJtFOe-
zjjv7*`I3UTOxVjpVQRg5>RR>Vs!DwpWepsANCskB9r|&wu(?TphOHaw;9(Br
zY|2SWe=TjSv17`d*6Z%i(HK2k6Lh*ZpS6apg)V{?2?Th=`j_6s_ZF-;GRl#7-NtqF
zl!p^e+JubpVQivQ^cY}qUSL(9BceNh$nakaO7ubHVHf=p%umzB%4S5{UUz~Gsxb^~
zAU~9_lUUQ_;rlG?EfrROMNQQNOPHG-rONd)X75OFF=CwHP1I`PK-5;V8R3mwsL8X+
z;V2379r|eM)>E|{Z_`KSaJtiOUEll1l`KzbnUNeGc_edr`%-m}!|OJrn5>Cy#w3Fo
z`jA`#R$N$^lod*{x{;Lk?$pIS8umdJV#b^xt$C3UpkC@UuL|wqkzQ2C^|58TG%!>=(bs&N!K#oxN#f`YM5<
zN=lJZqP96u26y(lB!XZLHh`|b%DTHonGubfA{O80-5i;I!Vgp!(l8-9ky<%8z~!k#
zFk6#P>mkP=W1#D?jSTh)pMvQYu{L=}cD!N%|C+a6pyi!^5w2Ivy*X|le7T{Zt0=F(
z`dvAre%j3#1DsM|oBLWLI~&YipO2$_&<07%*{Jz!lDD8FDR7EKxm)I!y4!whJ
zG=u6$aFks$$>EX*llNO3YjnZ{{;6edbeRT!)`~SYPH|
z7xOO5{?nR&malCw#aG!vRYcVh*UzL{XO^L};(CcZl(%rnVH7tR-=dQ%OHJt2<=JR!
zmmQsIOPCVx1>^6E5IglKCz_XXkFp()vA)VA(dUJs81-MHb9&c*kS|s24kU|Wdf>~F
zFGZwTAyImD{c&S(HWxC0p+kS5{W(5wF=p<0m5kqiQ#OyBel6u^-*kl%f1J8Tp@2Az
z>@pzkL3ncs-_SvP>D8}&vLidZ;H2}54j`5S>#)tNz%TPS8eJGZ8o6U(i})-qHH)>3
z@h{|wbb|O6>zwCU+FH252glK=mbQoztlryM!JwGnATN?9ro0H2#cXTho-w)Z5G=nO
zT0@wBkF4+Gt#t<=o{cJII;rOR>h%WJrh*##EcbU_LUKj%Q#C&W95t^Y>~Eq%uR4Oq
z%JR-da*Cb=rNR!MycL?IX1|-r)u-r~4AN+E^_&f9(;up`b>qeI@gJSM)jB}FdaTV!
zTP|SSs#(Jz>R1|yVvBT<^pP6}_UR^8*{o%Mt$lRE!F%xQxhMddyhpi>=22w0%S&VF
zH*T1ham_<`G(LDy<2zg!@%k3F%hUlBz@Q5Ej6Fvzo?3WUCLn!utX+1A!jV0?6@(r6
z>EZ{z}?Z5GR`mG_&Jq7a005)GkiunpNJ=a-CZ@{Ics*qe&7xSvE0$
zQ*X9Fb85q*91+4F-(6Zm$|l6hC3AV_re@
z8A4M72;XL^BnOm7FI~3T^QE;F2PeFL;3D3N$~=@E=bkI_dy2}3fP+oetY&uXD7-k2
z7m^aX;#>pqJj(!%4NX{?1@sR=@Y{&rDbJD!+T$G)2%Kt!4+G|_>*r~m@w9l%vR-S$
z?g~hM-QmwV8sFu5zqZ3I+&FzMvlY%I)DViZ*D{8+O?>Xw8iB8nBAe}^+Gde|Hn7S&
zJh%KbzIKvszk`hXdf3s%xfZlhIdUs`+Sa8iDbY*(9oO>wO|668-eZw|)DjBW7(Pt<
z(z#??owJ1IDTXQb)mD+8Y2$YoG;4$7V~NZw+}Szp!`M_
z{7`8U5sLpa6M&gwI64h}1(W-WZf1|Ix(dY4K&_WX=1FKn{yzP1$j&%ySt+i`qT~JO
z5}S#^l7*9_;2oTK2gJw^9|icy(=X6muJVq{7Rm|?Fab{X%+Zb4bjJ8JB?|t9eL^r*Szi;YWA~P
z0W~PXBcJ!5Ha=3Ndwvb_xoMuEMh~#f+8g;AkwzefV{)AxOlUp#c?*Ev*d$gOTfb1z
zYc}JmB3QQ@*$D7|{NgZhGV5vgSvRp7wG#j%z{D2n!TB)BSQ{YV?URmHW1T_53Y&z^fSP0Y7hAXXJ>L
ziYaobS&^&RG`3&frP?*cLL5xv_cs!8HV6^Qo0IM%j$bVvd`e4TKHNMVB
zh$~|bbs2v9)hO~}Q*KyX-<1>*tLxlZ&XB0aUDpyb2#^=``<`-XyD#eMintQj49_SE
z79@)*gLqzl<$ZbzaWOhS{X0
z6j$fMea}K_)$O9!tuIS9mEQXlyOBP}xSE2K5?)+mAFx}HJf(Jqc(N!$Mmx&(u>Wms
z7uZ{K92F_t{(*)IFaTl#j($ohY(eM5MeDjjpOev3$j?
z_(I8lW;~Cfg}6Lm=sC#HAHkZLlAg<-#&@|Vcn+T}_EKOd7kdp6@byrZ*q@+(OuQi7(C8%dTJc{rOqh)pSZUo!*E&s
z?IDOxwBHPKbYK|=ymq;p0I!=g1=OUbY%MfM%!(=1Y$l(trw5qLN)Xe4C(X`Ybvymt
znJVUUn&L7$R4tO_rC>6D8^E1zcSjMm+O{bJ@S?smEVLtN4A+_}|J^lOv18_*xfuF?
z+npDHJ`t%`(-AIXIG=tPGzhvfUmVxQ*UZNZXPD09^SvNCwoox7H*FYd7S?zWNRm;5
zQ^?=LVBMWu)uH;3UoT$qC2Oo^V-ByNLJ#sGh&PBWyFt1eGf3kxsn~O;#D?XXyQQGL6VkDP2ZeMn_Lf)Sm)Z3T&mc$
z89mfULBa?%QRuyoO1`npGYSYru*i1*SwQOLn?(x#{JM6;KGz)8w&&>}HKdAWMGv_^
z*RPaeQJ5g>p^h6a3HW7l(^{;=2L(t4zJ_r%tG>yg!2h8MG-v96$kP>B
zeBxztsHk%%9I8*BZ~4H)<$3dZ_7&M4{U$?xc-EjJuvBpObq(tLYz_M^<3I>!h2V|V
zNqgwPr&vzhxul^wSy|(=g{qliu59U3o7xZk3>>(lEH0!DiqV`TPXUoM_g`*$hK?8#
zb#OU?%Z7OLN%-up9TUl-IcofWk=5d1oQT|~d7c`Z`=#r%o|+afqwHh2#vUqjAoFKj
zsB%YHqJUgPV;_ss>=$)n0;P%v<0c+2ZyN^@HWskPddr#WGLm8mzO&jaJ4@N=e0wI2
zQrXkP4KGYfrU^RnRG!SY^z^W_vi@QUOc1uLCvE^y=y8!UlN5O#*Y#{SF?B6f~x{Q6ty
z+znOe;(y~#PC+6KS|r_A3n73-a$%1Qt8W{*?Ymn=p~s-D=MQwAlyRs{ASi$`oZhUv
zqcYQ^Nl)y$z(l)XLuD`kYP0;yJ3EBc89axI6|T%e-?i6F!?xb9
zc4B6CXB-_kLhxGR&fK3v7|p!7eRMO77!Q<)MM9u`G}|m-tMteJvqmMJmnGNEI)5~R
z!w=ze`0R_nEb6|e=*@G1GV1?rzn2Rpb-6c9hN~@O%Lpp5vfp<%TXkNpRs>_Z*A#yhl==#VC~Mo`JiJ#i
zvvWLQ^s?i5VHyUIGpHtcKRNVs_pMt4XHHB!q%p`{bb^a+E~JKv8fWi`lf;d8gwdBj
zbAt;aDgW^)CkexrQMXhY^MV`!2+E&Wh%1tM_tiE8h@FdnD(j+9gTKlLr;V=f`?;&8
z)wuHj>dq}32Gkuk#YWk%Zq&Qm01gT12Ybg{Y}1&8S;fJ>0Z5X*Qml7I@d;7aIwgUL>){&zgN1slx2(1Q9iHCvb1~X&%ZW8d?+FJ5Y
z<7U>&wBomar+Fmxyzt1rhfpWh1YAns5E5QHetMZe$j|%aa99vCF39A9p%`^j_?u1U|lxG9oWHGtYIC)5WP%
z3BGq;A|ArUK6u2)QrTraaE`lDI+$)og*%g<#wsUU(M&>rzf*bkcW4ucj4TeZ?Q~)E
zVnE`_%a*PAdHK+J3O;&H%Rbsnw=_0u{V?5j$f9c
zj*UKVu#38crJR-FnQQ=6V@BuZ%ihdx>i1ykOAwiNJ~*Rh7Vm01(m96{-j#1*bUriV
zV7>pe_9`4-Zj$nggvO1y6EpgQnDW4-R%Jqe&=`zj`0g?UtYMxOtDxP#0qUAapHnHx
zy4$7yVl3w0cL)gBfp{NuMPA8yd)VbhwVA)^G(GE?tt9BDnCLpph0jl|i}UvmX!8n)
zXu|!Cb``L{@ywz%lyzrLe_y-T`^(YIV}oxT$}|a?>C!^giNVa)qQ@N5!)e9q12gx3
zOZM)^wHV=wE%+qC?qxtlt8*(QkQa;*i)ST?n?5v}l(9wgT@-p$aXuB7fm$)(Bs-%Z
zE@#o)(0m+Oe;p%@L(IC*AZi?_O(*{5%=QU;{C4GdEdci!P`3-kCfK0kvE=%pdXhZL?FT)qDvSJ$QW1G
zvlN!Mj{MtgfB3!#x4Uk5v|4AXs(f=Xz36-Hd%&n`=4nFAPyW*TtC1^b+yl`o(C#i`<`H-`oD{ItbFqJr*w}By;$@o)$eElFb
zy-&^l(vB0zsC5@1SKWS1@g3eyTPgzUyV3~J7eoHoBRAC&2PTADWa(q>6O
zNL>*oQ12Q_%MV_sNb{AMAgIn)2s!$B?;5)!ia!NAC%?4l-m*gdg>zG5X)Qv_(Oct)
z&_L@ZhRIEAX7}o4ciCX6TsAE5<6dlvxF!e&+4DNL8c=oWys(v$Bh%}DBa6q&9MKJ>
z0ixO};AJzl1_v-kA%T3cCmaQU%RBz_nO0tn77_ecP1}nA`G5wcv4aH-tuERzEp;bH
z2-HH>);-?c5g%K;$+av;j-X2*ULMK2Rk}oqy^LZA=Nb%N-YY**g
z_hF{55d9zYq47q=ad~EcfrHh*Iiz-h8y_|7Dag94&dHiye4ps=TD@=qZHCSM3j)S;
zdDx{`@hYzPX12%>a#Mibi#Z2;D~#F6;VTj&NGOYWhsbw*nI~kq}@Q$YrmwO<4K2$5;#Jupbjs8;65BqkUh8g9Ca=7
z8487H^_!(A_&xJl`fQ_QcDmr@VJtESyK3aH$T{-LrqZYKQkM=BuqSZb@8JToFS3lS`tYFnl;rU>PEWZZy&3w
z3LA+{`p@Iea8JEwoc6Kb=1eN#4Evv9GTxh5IQN(zRu)v?G%&Vbi#0NHX>MTbeF#FR
z+6NGYdt(N>POb2KQ6Ob&{OgZ`IpscV-YV!<>)X_5Nb`Lv>aN*7#Jj{ex>pH3Y?19H?2
zuIu<`{GYLZ3-!Q&5*neOYiq;qv~Pj7e?BXmd3OM3K$yP^o6Lxy!>T^C)9=xEofCK{
zs|&ZrYu>SY@Wh*xJTJdVMpTO?(eJ@5OJ^!GYr0!^22OwjLHU9OXkJBHP|HkXa%XVb
z*gzyVVnaHZ3HKiLmjlGMJx&!#88|&Ukq%#5h<+Y)k*75^f5DxYj=FU`_S!#(w81-p
z?9GTYEaGqe9V%ftIx`egq5bU3t6TPvJtoY9s|Q-vn+$<_{AU08wg^c2(VC%aYTo64
z^tFCLx|CVI!)#txV?4owq?S)kvh^&FNiVZ*_TJcG8=mAx`qwq3Z9J%9e}?yn|JgCP
z35!fWbRK`Fe?rCG3mV0e*0t6BTE(0M#V|v8G++R|gBwR;R?HOn)!%2FPv9LqxWJO9@S=^S_#?tyN
z_wMnj$*?kF9yel(>XvN9Tdvt~cZn?{jxUR6P@SV#f9P&vw$j_UJM%n!wi9w9m{(RM
zXYHoz1PmM5TL0#6rWj>N5&rcYf)G1Y3~A=j!e=^1PA7nkSs^U7zblv2DboNmzu<04
zZ8CL@BztN4QhUH4hHC%C=$f1koLw<2ltQx4a@*bV#HrVhOE?|6TjDq{`ddtCJ`~cE
z20QsYe>1IRzxyusi07X)UlHEo*MHWH$h7ydeS4wB>SiFwSLp$4F5|u7kah51cB;cm
zUUU8zAz-Xx{D1<*VzgeIJmwyk)bJ)cYR#ZN^$J}iSw8=ad*-ur?g6n!OEzz7RJd$oK8zX_I>Xj%Ixyk7gp4r=AOyVcUcfbgf`
zrs%ELXuc^m!i~t~V2u73WD5~^`jv`HmzYv*(<50sP0I8+Cc)fJ^mn3zg)+}WE^zZ8
ze{&iR4;JSx_`p4p52uq({_!O=p*@p?8ZO6$wPACU#URd>r_MctsmYU|ugBCrC)P%+ET&KN;2K14DPk8|tV#Lcq
zp*vV==fNY$lPmI1z&4*U-D@r@c{Uh2
zSy9{WR8)gd)Clx_)hsVBB405-h$7O#+md52z3TV{Y)5o$DXl9x@5HwS1m?kke|07I
z1efZQ4E4RXcUJ8*e))~!0;khMQ`e+(ClUIQ0QrjO9yb?Qj*NkE0h`#B*c{Db@9INx
z`pC2c@}-QtaMd(nVwrBp2|da3iwwdHp)`7QOZ|A98A#c46y68g6*
zEh|!uD+@ietaLBii7pm#i8K=r&@m@9!|4?vz}vxIc6z0)q>>E?-vst8e^iUj|7#oO
z(C?{wbHX4Jh-Yy|e;}->`CbKmiz7`-EACLAuI84S^P!D8$jzORxQ>bmYC{Hb
zJfyGndHK72JLN*Vmdyokem57%xo?kre=J#nADrmhr~5Tnkp%=b@S$9^`@xSM>qSyc
z;~rXy=q}!uGv)E|Ui`8wqS+Ui>7VCMvYfJ3ga%nd^_JD`UKVs1cdIb2*W#)A=_J|I@b1K76$;|FCf
zf*HzalZ{*|CGoZf62Y`^(W9}0qmK0vkqqge`5S9pw
z_Cx&^BsN^BsB4#qf0$I@z6npBcY8kU!`aIX@!9lKC(GAHIOaeLWH;d56TVrrRJ8Fd
zDkCyh7EUvO(xjYpH3O7BT(dsR|Ks{tD~4xeQn(L9(>b^P%
zf^yZxpp99-Y;Tr|%8$ih97{wLjgV^U4IoJK;)%Yf*C(m9e-ZeOGck;Jg)Ig&7nPJM
zWBzn2i*5%c9WwFkrx%W6Zd0=r1!LwjD~{d{4xJ1E0&uA7P-r@&tb|&Lu$DUx>Z<1~
zxx}Qi{yTh^R7kE7jTgD~MePHAvo~$3J)%S5poJ(7gor|v)ow?i>R^duU>sb(5fC&G
zFB;EC0zlZDe<5uoC;1otarj2csfn&SYq>7S;fq^vR|H35_AUm^j%UXzyf!ewx(2Ccx%*`iP7_*&I$b-5{^*y11ygHDi69#Ll$g0TZ+
zbs2>*Az!$(0Y^9&vm+zr+~clGuXmt~fG*2zY<=cfBVBiR|MBI=2e+fxHO@*l`A03t
zvRGP0a2n<-Qonu@9x;bjP)-7=Nv^{gW@NMQe=069voaX{A^K#UM-xN&8)V19(YNs6{bNl-`xJg7ZQj=ChID>7dTpjl9_@f6OoPW
z#OgUaepCM?)!L1XhmxBFnZP;n=HK;Cd?;NE(5gZO=5U?
z8#j_zNxX2+a8XAVH
zc{`8~oEjI8CetXmNBn;ltPz8;{PQ7q|_=e(TeLxx}QbsY^
z8D+FbQ3@AB>e=Kg-4b+sa%=wEYE!{QU8~2?)~UfB#N74IrdSI_#^0C4xD2R=e*^Cw
z>~8;!hR%_N4ITz9qIPp_KMKWLJ13>ja9c6xJ}b~dSM2dZ)2SeNr5z=2tsQ23AIF4^
zoVKoDP}eXimp^iz`MQZVbS|tx>faPna+P;}fg}~L%bv$Kr)qc6=yJkYWhYf9%sRu1mBQc)_K@(d$qg5OI!E`CCsgzyyLF20@J#
zbwvEXB)NFf-5xo}!!XL1BA~u0c}S_xR{E2<#WVM1f~rDrp>hn2;c?u}!iKDBr)!Lz
zj|sVYPQNdg>)JiZw#B2%;IPUmb;!x!C)0Y(aQL*mwvOeUVMu(~<&hu(H&h3nmWHSc(sgdk8iAML@X=j{+1niBV)_FdmL|JOt3?E2d#
z6Q_q=`#9v3Q)cMuoHsm=*`jv*!*FG5Y}8v3hp-$=)5M1gDAXBUT=Pu&!wV~Gxjp?>
zf$vkzLLmO^Se`89mN~Z^f51iBhypn435W4wP`z^(P;ylAvaj+~KrIXOQb#d{`ct<<
zB8#>AnUVp*a<8YCw?xAt(2@^9&YLF1WUCThpm)w4G(-zEJ@A`6ozv~a<@fdgDA3(0
z2AH*qp;&im!?6~hvMcQZVVuOMaFi3Psc=XBjHg0rdj(%(?8a#Ve=5fk%U(M4h?gkF
z9uf@BP-)#Sxt9}P^Bt?mn_%XgtQp9b0V?*#4{!*pBJLUcx`<~}e{_k*m|i8@1kP9v
z0StQ9>lQqH(Q+KcfySkmcq3~VdElV(;ukDaBm|vWjn`XNJ0dl7Zv@!ibfQn`+f*g5LbN%?PRGfYtyjJ2dHj3XcIYkH9X^$^Z-CbhUgph8
z2zIYh|FTq`l(iL3dyUa+NTSECInJ3eIsk!{Oy+nLWe$;@f4CzWJQY9&S@3L^u*>H{
zW~Xs!sVQYojfaB3j^fzFapaVkIf7ji-pW|Ni{o;qvg}i_7m@YKse3H91wo0uprKlE
zCxVO@`R$a}aEN+5y?Jce#~4}`^&b$|7$fHfF7vYy_e8vrbEs;x6_mQ
zMnS{Hx2^?;pfah_T2uScZHh80QwfjYAOaGxSt=
zn#71Ie^=`83-^--u1o&5;qMmXN=sDWMr1PY75#j9ulTW=(HrydE?>EJc|>wE%$S*2
zRp`$2w-24>I=Bw&4Mzm-i9PiFAh_3wfY6g4lZ>tTW4x(g
zS^3$aS;H6j7h;CI_g?$IHCe(*mJc%Ct|<{YD&uZ&V)l%_SBPrlsP4f?ba=;_fQGrf
ze?2o%HPM$&e#U^i9PNwlw-9~
z&cDv^dGrdlLy;v(6B1QNp_a&RvW|W=H?yK3{4+`aHwOZrIe{#!m
zp-l*dzi5nG_da%_SDc?l@PHHxalG(!hYk7N=I;S0X93NB@da*jl29}}G`=(!MKI{f
zR|Fuj=7|dI*?9lb_uuJ4KeeVz|H2u*5CeLB!yrEg?IK}mz>X&gu^~Ier#Xh5R74@Y
znHX7LeX!AyE>h)K&0-8?#6&3Jf0mPm`NoJZCMGA4{)i)npk&od^f)xzIcW0?aNHz*gmA&e7;6BEW}z5X
zg|(4{njHX6sdyL>as=gdGH4xvclBEC_mgUqW`lZPLSO9hdcyQ%
zQRiD;cH;qxv>Gc;Q=uV-e;Ho-H$Yen>skAbwShPfAVaueR
z_a+QrT#&O2oWXabH#4jvh=cQe%*Nu0+un@ZEOZ4JS3Yzf_)?@%HjHhgu;bn$A(AHD
zfPOcD>KrNm;no*-(B(~pI)WYGMsGWPzZiNg{BvUQo&B=LqWvN-ea!t>5Q6TJO&r|~O|OlfEWJqEwM
zZ;Cw1b*Z{6)w&xRfLr$AJzOje0D400ElCo3JA>~6ixx1pMf1pF#)`S55+zy??wn|Zv
zHLiyS74r*|1S&}tRCNm;!zTnk4P0IANHC!~T=j!aymX>fG&}-MI(RP`(=uM!)6L4x
zuYwigs}nb@!-(<_X_q#hS3ijdON*iyOrU;W<5gpbdu74|PuFH8r}BMibxD8Tzjm4V
z7OV7;h{22xe;ViF$*@&;d{A)204y{^htC=&auwAcZ&OH0PlS(6CD-nzZ3+(oV%z5$
z13^rne3eNhy+Xejn%xmo-W;-7B}-7p*d6fERSUm+}DOuI0f(2#_U6GF!+^MW$
zoqu#hf9^TRo7Y=~R=3$Lh&Gdrn!6PXsD^m7e{UQmO-wdXvqf7Umx@Tz>Fy{W>!Df>
z*c0Zk9~Y*ebB
zwF1U;6S39pd*Fh~*kV*`0xeSp4uu{7)YV{t<8
zwTD8ksOe)Pi+UG}f+GB()5%s7)K;d0S*!#LqP?u+uEB|#Qh$!h<*myiG3L&2*I@k&
zf7`-MB>K4|OIVEecHO2Q(?EjLo@1UX%b`rReH{4+)A=QGbAP`R11?&=SvlUm7}W37yhg6
zynZXo$7#l;lg^VYe&pSfV_S={TRg=PeV#
z7#;`s9Di&J537DqsPc3RGxjDE^ww`BO4x*|4PyjdvJGrQSI>~3){z6v&Pp5(kG>X$
zqiak5+E{@hP7{>|b;yUt99xw!NHC%1F0a*!wZp?PTW)pi#qN8>c}Cmn=VdB9e~K0S
z0Xz~&R_8#Y20&I@A$#qWeI-nHIl35z0bDt3;?X4v?mFa
zMDZ5HD;<$D)jB`IQb((T;N#yQKRygM6
zxss2cqk_^a%|MaQoskzgLWIfC&X?W%rT7dxF>x0PLo8zEKc_u_ilVkWf5G?F7NM>w
z=_(93<*FN|IeOdR07NKV4lC82AAZ4kCT|xA+myX++--lBR~Jpq-TQ;K-X1flsD-Dr
zj`0Nomhe8V(8z%nUIX;nMUE3{DH^A~1r_x#@KBgbt)y`ry*TD_BBmVA
zT0i^Us6-W9%wvCg*8*aY-_~WGV)mr}^4rO(T!6bm3RO1Wq-$`${*vZ2(&}t{L^$|?
zAlt}jCs1Q%%;#R&SjhT$lnee)j8cV7`m+TbKSvlBrbeX~V=okaf1s?Sjv*T|7OFjH
zqKa`l54P}x6&WaP?#9i})}-W|Cz|O>l25_I>mlK$T(^&ct&0`ZXLn=62`XSae*;o8
zx$@J#QCzZG?7Q>M?)#sZCsw-%%!Ro4OhSdMbBepSx3bIW$R@(?v!41@7EjI{ms#8b
z?xJ>pwuF?F`jsrBe|+OgInj|~4+(V17jK*#A8d|;%rFe~RJ2rvIhuH{VFQ%RaNfh2
zkuyvwDwX)PC{_3jPc*#&^%u${)vW4y>sItlUEA*HlSJMfkyGiKk|?zxtG+52p2rk~
zz0wR~P=BrfjP27^UOl7BJBaEKo4>}n3O64{u+%kD^zXb`e?L?5W#g*)g(!dusJo_ZL-J;O4cjSwgJBqqgwh9A2XT5S~{uB2v(SL4bplg>Y}KC`|P^*zjA8*<)jL+TWdWg{@6HpgDFX
z1e11MV!f!(9z%s^*v#3XEMsp>e`i<$D1=GMO0hcKRFQkuo&ewV
z;ug76B%LPShPs{E{T#p?(K7~q1aO+D2_NW^Ci-*i{Y0-8MD3ixxAi#o4bu#GhmB-X
zf0IF-Tz`u}G@gsYq)hNi@t6wNuTz^wPF!?{H42!(-HSOSG9j1Ya-UjIMR{Of2%(dQ{Z4L?v;H6UEnHHtS*8+WEnJt
zz{0tnCSC_>5gp%KjZivG)}Wl*`)7FZ103QwZQH0m**5E-u9O56PP9~!5dz%OAgg&R
zAI=CKr15Hsin}W2%4Xl!BulJ?goJr*2F1MVZIyI0<|3=&O?)4D45t7CcP*>^FwRqSwHg{%a4?Zjv9hAJHjBH~U*Lg2Sp_F3L?1C=_1MveK4*y6;T_ueVc4nkRv>>h-R>UVlxe**h~
z3bWV1gC@l?3Wb9I;?RXI3+1bN9pPNk!{%qKlRy;fSyk$jj%t_*R}opUMG(^faSgJW
zc=hFZiqHGUg~$xxlt}Pp4m=UY`xsg@691wqX6k*P&P
z&UgqPXDGAKVTv4T{RJKmQbK0Ye{rOQ=&V|wP_K-r6LQphj|R!(QHlY}Nr3h`n|0#U
z7YnDA$h!+QQb81C4~Nntw+9fpa@=g&niAeuh(;iKL_Q&ul|{4
zZdv8oqWbx_p?;cjJ$e(v4?2&+EbtQ0mZ5`oNS;lx;!WL`!haYO5MR0Lf1foXn!UBR
zH3m1WO#My;?eFHCz3qrQ4Gzm-{HFp}1wCKZZxkTE0y0CtHi|EDaScg|TlV0>rFhrA
z1+1n&Ts=DjfSU|@B3uY>&^;=!=KdvqY8>aQv)kYr0$V*65Pr9^J+8JOmctNb5p?!}
zAwFYaW1sP2^b~UEk*F7we_fPLcv)yXJY>v3p8CZlf>$7$g?of4Pv+|_eO!8l&Olqc
z1FL2Iz*gxOr;tIhqzw=0>DS$y7Vi;q%Uf<1!e)(i+5~znDYXuBdN|$SL090|0z~`C
zB#}U&4-=B#AIjEK19&y4m}z-J`zg!^zYZ7C3X_?qGxaYCdk=e$D{t!3Oo
zI=u#_Ee0CRak)_mFoL7I=U!@%3moSIUDf|OAn1N-Flc{a2hV-Gk>-(xLGi^Oni0J=
z=N>MW49CI3)c!iqf1X{V!jcr^5k&;@zoYV{e!*W^XPN$8?z820al#UDcYkkER&)R|
zVOq|-C~06It*xOgar+b8{=5B18cC=nTKnNhkf#Is!K9|cH9v*T^!oPV+1D6N^pzzt
zyyGT+ewQLQU=>br8-38nIZ8~kPzQjz#!=p%kQ!EBB)T50e`b1gk<0Fc9WAh{ycUBZ
zV|-nz73OdvxVzr7#-Z>p2@RZ+Zznz^j%ro3wey?Yd`wlB>sk!d0*@*Mcw!DgtQxB
zRTqQ)fBKkBe-0qGuy8&O3%gM-@{uPr{tw-Tgh#Tme_ug3H+L;5SW_ku17#4Oghe;~
z8}~iKj)ET}+c#-mVCK6nrtP
z(;P>l5T;nT48cJE*nL#3TPHLu2kl-3b7YXtic*jhe*+WDdbt32`l8lG)6i0hH2Qy&
zXH>oIyrz74JPrsYMb1xAmn3i8vk|^DVl$$b&UXt|o|l(}Kn|6p(Xrc2J`@N%K{gQ)sF7#WZgH^oVv~4#r73M4j3CgP=y*Zq*EsCkS
z_4s>6e;0vQs=B}0u=|s^txegu)Xz=6OWCJ!*P98M^q7k7&0}M}Gts4?OKkQ)FGjb4
z5#KBmVwwgbIC+3->#V-kLP$NPeRW;*VNYp*!&OYB%lMM#lR#wi?N)>RJ|DHBdtd<_
z4&UphYr+pOI@K#rb6?KMWlPf)3MF1v(H5P=e`l50u8`XwFxbNv7&r|=I*lRd@dPkJ
zG!F=}4j8YySJn6$YoQ=8aA36Nb;#LHq;c5OsI9(-8i*=i*c99ThQK_I$+}4e`7v5I$_=VF0
zSY9Z!uxnT8YnxL+@L5Y)pwAqRg~@95GA45Abikt(Gb^gJMm->LWTJ_B
z^HYo2!opIL|6sO|eOfm)0!Ac3i@dM3rCa6m8xs{`T>Gn|@u$heg-Jnjz9Ojze}L1L
zJ`&4C?bWP;HJv^-^RqsOmG4mrE*F@}@>tJ1B%X0X`4EUbu1^Y&1e8!;
zOExv467(z$SGmD=rl@gX4wSBuzl>Tz6=Nh`_{sW+@S&Y6fgp!{cChsq74&
zu}v3IL5#4l)juC*qQ|qdc22tbe|jrcZaE<)PsT>4r~Vsb`4h<3%paAIS<1zzz;}*J
z&wGYg^Eh$lI2g&YhwL`mZ-7eves?s+%;Cd4N1+}}@D3jTTR5iu+5R+)?sm{ilhCte
zZPj0k>NO!EUZ*m35=Br~?+t0q9OWmq*ioA`pp<5KP)l93JuABvMYzBEe?X$M`9-Yo
zSP_-er3P(JIcCpR7d!;56iryoM$o!PCSNbAB>%Pp_Z`;(EWMMf{2`cyFZ%u`F{;t$
zaly5g!Ak%~g0>)0V4gWYG@S0BDSDpGcVKxU4YV%_q*vl+E>X19Ys>yDMf{*MqNu}c
zForC(GB4sFTA$1sHK)7pe-l17AyRB9$jD8_f7}YsO>0ODOV&>AHgQfOIh%GLwI+_w
zWpxtpHmow^t}C1y1()6Bq7oVbXLVlkJetyFU6rbv_t$km7-+irz)LnLJ?Pf0?^J^<
zd@H#IHvnO@si;OA&^)REmX_;lcY;thSO6e*c_iaAcG0YgKER%Le@#`!cjovRHcVM1
znPqbmEZnzHz^F^}R4g<;w14d(3lI4>I2H&4;L5*M<^iLImmeqElw+_xDpQGAO3cG*
za2??A^|3a_)BEOFf0`ofXNO}AkNU+3wB^MJ#hj7Bi?m{vTM7
zb_CWiD-uW$A%n{{26-U_MV+_kqY*V_b0ssngA*X`dfbwp&a-UVUvY{S;|t2zQcXJ;
z8$+H3@{r)>cd!Ml0e>@e=#Oj9O`!D_vI)sPCJ^jmQ(2+vfBUV@MX3g9l7#X3WPR@-
zx=Ot{R2{G-A@VpgRI{O=)nL^DacL$~`H;x3CRxw&XyIELCj&&5=$X3CrH^5F$sUCD
za4Y)Zc3GPdi)5~d>zWRxAXa;(m}>RPk-lQVEcmtzh*d{R-JZTRiisZ(IEih8^xC_4
z0~q#&9l&QLf7+&DcSMstl7a_+=u@+TX_WGd&R6RBwOruO*I11)YsSW*pEJ;#u?El?
ztPP9T()P`WjKw4Rt{BoDM}`4iywxCPqu^}$QEfOH-|6GokTv{h>#P$Ww9S;kBK&Ho
z-h66z&B&iLiFOmaDuTcMC_Gj!i3Wfj4Bsz|+vk2@e~S$q93m#l&~V*PRbfuvwzB=U2a_IJ
z53;>46or26N~%~7+*#{j+ge}51F-6r@F|2d?pWIipKDkpxryjGj?&KzIsBD*U!hrr
z9>I>se_}PrSmv3T08^~Jg9ZQ^)ciotlBs4mw^fN>e^w;^RZ+$|?@mSo#N2SGl+_-_
zKq00YUr*)?v27396KOz?)!SJ7dEy`c2!YCx5DWT(5`tQ|=6XA-h+Z!hi>e-MPn8eP5^#cvm;PfUH^4GPQQu8e?%
zj_oyfp-2|DUpVZ7+y{Bl{ZIv#6ENDiEIj}N$QK=y$<0aRX4KkvWgvqtp70cz<}Yg8
zYdLwbS)8wBulb2rya7xF!Ctg9@SrB-^GEQ(OM{Rm?K&gxWGZa(5Y5DD-q4YY^k~N0
zf0Z)wNik)JYK#g1Qr|3Sn^lGFf`c4$gId%MLmz|ye^EW`
z!GulyoD|s*l?t3XQH@@ToLj8Y
z&xVc30fzm@`{IhneP9SKdz|nDK{3=X0wl?5=Q}$;t!lxeFml(@1i1?R%Fl2C;B)6}7gXdqDkn
zUsuq!sz^HEfsPClWQ+N{e^fL`*w6Pz`+C613-N=?LWh6cX4=wa0JT#UTYRC)Z~fK^cQ@o#L5AvGjEe!tMCwwhqY;#=2F2C)E56npRpB
zf(@i?KeW=SvS#1|F8V&$oX6$3I%h4uztiltiIFV&5uur`(3#0#(2AyB69D)j2pvgO
zxLn}1X|0NjtEMHMe;E8Vh%J!!k_MD(%eO%RnuV3p4AktdfCN)eb9;=-xOG?HEx6^+py^d<`-AJ8GY
z+zlwXJnl*n5XM12gG=>w?Qi=Mu>Q1a#%v`{Jn~m+_9C~Ce^<*K7z)+?&-d!cX9S4C
zHDgfh)Da;ptK9y?^?5uvDd-olm@ke6g^1fB9l2=-R@8;Avm)8h6d8y*XivAIBa++w
zPej1}0lbmdmr6)?QfRIi9=Rxs}SnDv0$)gG?|X}_27iwL{MJI(-3pL5=(bT&E+1bOBOl<`UGGj06!U_Z7+^BC^gHLd!nW#~&
z@)e-{E{uRcywPKiSO1O|3$MQPRt;fZJvStEGSlvte_hr5Ri>p!nfmH)8|Ab21+W+T
zBUR6jIB_47OsKZZ5eLllE(La?i^#w>6($64qKB7WeBhYXbi&|%!-TP#+m38v26X%;hXqCU<%BqYZ
z;ccD_e@!3z+Y^`ty_hznkwqfeFr4FgGDK?Dgmz!_d#RY3x}|C^#XmVq6oPnG$%r4E
zFcVQY=mabR=~4n9f^=I*-A!Pm7{|~;NqRDB$CF&5RV(>6F?7h=vr*(&%X23m`&B!y
z&i$4h>wwZ*<`WxBO$0CtTS!d7jc3?tF~uIOf7GaR%6fAv554K!uMx?xFaP-=V9w=m%4*
zZONA75ml*-KQ2HgBe-`8i||+llEauGe@xMM_Xl>E5vVjp!Q%Im?wMun&{w_FYP8L5
z(Q7V+F@HUGL9+wmwFN7#PpfFi@B^vl{3XNd2@GOZ^8dzu~p;nPBW;3(OScHOk!a7D$MqID7r
zw3*^bx0^)}qlN`-LmD1qzU$|6e#+6WR_@;n@y%-N!7sutV+Rw-fS7ILQIb*QKb{V(
zJ$bazw(yh?mfF`bJyLfWg%H%gfA)bVXG}D4ETP>X?*UJwfBpV~5TNzTcW*wt1k=un
z;0fHg>p?-xwQ|1^(or&d4L7744G#^fAq2c1gQ0zT`H2-L-s~jj0btc5zT!!K9oB^|fdgI?ouE6;nfYDz%eVHAepBN1Guzj4xD;3i_oLOGL<&WF0HG^W^e{
zs-_EzV*%(WC>8X^7Oi@g^dHai8u%?oN+F;gKe>YT`v7tf%-gr4}oUSlkGw1SAh5~bruXM$g0@<~S$;Bxg
zwQBtYTLdz{;8^<1&7Vm&t!uAl+^qR8>9MON+7C6Zz!H?R4e0ZF
z=;5Q=SX>2k42V6JfbS@gZZ0CejK!aIOM``(1@=0cau8$Gf=z4nf2UDN+D+{6mAnem
z!@Tx*r1Lyq|EVF`e{l>&eC%x|HG`(y6r#ld&vqu=>ctytyUIL=JH#~@{0MdCM%?lf
zU!ihy!T`gc{%}g{8M|$zbMi5i&Jdx
zE6P-ZVu7?DN(HDYe}A?-?KXH>TN1$H{IlwaYY0+Ko)8>#mwL*4y-0nVYjThtmMu_W
z#)`rOaaOg}XqjVWVt6-)`O8VBp>HCkEDlr_0m(ac?q`TL%b*&>aB2*4dhR2PJZ_}O
zk^YCl2NReVS+U%Km5d~3cY2QdK=0wj@@NvhPsQOWKZ-wOe>kZY*!EbslNa!{W6=d`
z24}I6#+k(2%(WNBD^^n)a_p>w#PMaJ1l6npGn7ps$!V@Esrpiyb3$5T4~tlq#D)>$
z8}RKRj*E8@0U}_t;t2zG?I7&W2a|EGvtq|u9C!;sme(84Ai(TdHYv4B!`xb}}=
z?rj}0d^2H4e+Yix_X&R4;Cx-EsTZ;GwcKUOoBWnWFnkH{_wVG5B%eFIa=|RG>T&`4
z%z8w1lyTb~m)Fiw#gJOL8|XP$OJOse&&CdqoP6UtLFF<2*%T3r=pDL`RB$xU8$82M
z2AK$WF93IrE1dy-n<|Jht3y_Oa};v}n}uFolau1*f63ttxFtIPct{bLw<^&JRn;5e
zUv`hH<^(hoyDwUazAp>EZNM^sV%|b{DCc6}H%=Wr4IJ|iwgKcAReHU~SC#wt9HZrC
zXsPd9ckZniqN(0>JrhI^Hy9aa?83Q$%0CmrxvgXDetUEhIn%c|2(`8gaa|dSMbYZ9
zRM|waf1UTwEp8%M$!;i#FmAj5+9>h5u7b!#EiJk~-`zR4bdPB_8SF1tt*IMKju
zSm5pz2L37Xsr1A~EZZz$Lr{q=LFPL7#$GngnRp9_fgUMFTbt2xl<~~>v~%AlFh_O{
zOqM(Y3RaI0x62jbP;+4Q*OtO(eU9rP26VADe^U_@5`M$61r+OOy3JeGfW;9a?Zpz4
zDnXN^Iv{NvgL~heR!CBEj}9_Ran}khyqs8W^f(m*_?{XfJ4wv-&Jm{07LF#-xjIf%V!;E4bo(U8#f^lw9#t%+&uEE%kJ
z%mnyE{FB*slJ9&JD-`CWSwJyrElVk2mtYS9H^xsOErS%&+399y%mnCvSyf|+Is7Nl
zYHDEDADL_4sWs<84c2Yxa~o{L
zgC0xt1G`S=)^q3jrP)yFqqAP>ilDnrK~vCcJ2;hEnMWS9{o-J5*w14(4S|*df9f_a
z5G;?>{Y@kUVsl6%zCdo(wl7{XM7176vaTbi_9S3_7$Z9(a+b2wSoaPx!
z>(avR)%QGdEbAXhVTv<&ETL~f%0Z#0MR9SDhe!!z
z-Rc_BXOCf8=#9A&hcjga;A(=Qlxw*D?f>L^u6jXVhX4rjzrW#|NjwmnV8hChnCYM1
zXsa5)>-D!43^Sbz$|Z?KgNoJB5nA6Z;EMONf3pjAD)!Ka
z7JFt-4G-qn$1$V+jPimY5WQ#236NY?`dbm5)Os{Gpf2^E(a`Q3p;D*RJJN+%AZLuk
z&QN?OU^4zn+U=Bi&*nFP_1VX01dY6MG|os@?~?b9@M$QirXoJH*#pWpBBl$>B1D78
zv1#~Kkdh>zJ^WOno4p3_e|gi5G3)PSkMUSNaHuB+D5$$u
z0798t#)avjUEia`1KFiL;cd42VLA|C`6@??v^?nBixHe*JjJV8o|Bzl0EXc|j5GJ#
z=DD5<7E!A&{HQTcIAw2mx2j-HygERg0yy2zCHZuYoP2@I=Gh!afA|*$S=IKhVB51%
zIu{6t+Tnit2J3Z&jU=UyXDQ8UpMrz7HPuC;JavCQdVe>#9$27zXhy3xNkS0_oN
zjTO>~YjeS5!!2(`SE{Bl{Gfb2E#%cotB`$KoI3t9I~H>>SOs8T2ph5
z@s`+Pl?Z1Wf4tAiBmo^pH2TRE>BoL`d2+OCwN}8b5jhRlSHiR&?H5~o@N_$Ai~(7;
zG&mWUrb+t`S}(%siA-dZ*M^TJyW`n;1M|x#>
zXAF?Zn7i9yYN1b~jSXv@nM0pyx9e}ux{XK%9z!-Ee-v6Z7;`q}J^#80tWnO5Nyo@R
z23g}~Z2WQEd0)FZ(l#ku&}T(ghxjy8GHq+~K_{?j0N3AQmxAISYaN(H&!Z+BDVBm%
z9oW_JIk_6|)-&e^jmfOS39*|T{(I?y!NoJOm5Vy(``m8s2iAN6`c
zNx_58vCwZ0?|J2L0KOt${w2%(FIm|unal7JYoHSl?HAsvAY%{_?AGKPB7B6!TeEQ6
zNVWlf<0FR5YmBoe@tTCU$nt~g>8kcMU=kRh9-fOcLJC38m5UwThY-rUn4w45e%d96
ze=uYXh`&3e<3YlG0sOBnYo7;j4ksO1sng39azlDU=MWIF5`spNASlK!TPx-Vs6Osi
z`+a5Cf4T)Hnav_
z=dx1;(Efy9j!$J<0|n<;4T9IEy^{5RZKU9ld_9hN4$r6`5o=R);X~FY7O%{ubk75k
zX{!e#3{RfgqMYF+*E7+V)sNHyrEiN1Uh>E+-7T)tmqLA7zp@PCPT6jLa{0)|e?w>F
zseDl-Oe=0hG!?1f@owcb<%{;tzCV!uwbxLF-CPPN0l3FP;iQq`3cn-k
z+zC(n56J^M9+ma@p~yz^kH)iwpX>|>Ce}pWTtMlZ4=6#4EGf4Bw9%WnqVVct?4o(@
z$y{jY2N^#mXP|yQUBseco(=&ge@&X8*9~x}V%$AGWG2)~?{QX~UuLtc23AcVtYI7j
z<#B-@B3u}MnI^cX&$d8!mB)EaV243ky}q5xWWA8%*TCg;*?1gbtKYN!8m!u|=tbvR
zt_*YAH(U>3odk(UyGbX6<8=RlwFl4k+3<#r>|vY_H7AGF_U3eD1Wt8c{o&U8&_GPQc9bxQ?i?3>_d`$
zXEK9?a>krtn3*$XX)GyZZ&9{H60)YEcq<`=_C+KVvX+Q!Nm0Hts;}kx-ao%THM`Ti@UqK%K`;X|L5G~<|Ze(A7?Vje0!7UH_yDItQdR^GGTsPUh2@Q*MJr}DAlh($`#nw6SFOE
z46T=Xf(9nDqmc;4#q7^Zh;^>>$-6>zKY%``+eFWTyNP7lz9K`Br8Wr~S^1z2p$eF|
zlYOf~tsy&(Ve29T^)8}@)&qW$1y6cl-`qU%EA&ZlS2MIQrrrgCc2vVggP3ovXW{b#C90)McJX3!UzFbrPC
z>148(**Ik`wglF5P@PW5KM)h>oiWoi;W0z=4i$=F?a4s(q*7rr`zhn_CV*2YY$#Fx
zO>mc0$2L-DKbEPx2^D=FIdUp=>{kXP#KKteeQL};n5VB$8Gro6#C_r4g~E-Tp)m(D
zq8fxb1SpU;D^Gsl_xR#bD)@?t5Qr%Qq7)J_Ihv?a6;WVh(6uEdpv@^S@*p2+ZQUl?
z_9bfYv7hMQ<;_1d;4QJOraw3!@-sNQ>|jl2zcjAc7d$Cr?e8$;n9+TYk8jzBSB@Z_
zu<>B_jZj2rx=36deO2i_!QJr;K8D%V5zqWoMv>j79J`{g{SFXj=~43&AM2qpf`q!d{6$OA=UKX{aO;ENsvA3MiIzD8yXz;NIz
z@S!7a`^_6dCBw6&u|2T@PzDGw^4{eAf)1ZU-oqkhrr*im#|)5ujl?PX`I?qgA(fCH
zc96dnmi&SgD#s?>$fE^9emj&{BZ2{cZmSu|B9N(&6?QK~!on+%NFFWt?K`3x*?dpC
zPwz~VSTk>|Y0lAFRb)m;SL_&MCQ=hc-CTaSvV{>n|NVRLEDhCaX!?=S7>+=MA28bk
zLsx;SqJ%63KfSPk`qn@D`l_g(s`ER_H`@aV#V=QMMECa_+_??)ue{FIpmKJHlp0Uatd*YOkM^
zBP}jYNw0YHYrAS6%?RJ-2yIChFPR0vCnwB12U3N@S(&HQEIgqY*$#Nrqdxcj%!Jnw
z!cJBX`L@HztT8yck69;V$eG6lXQlL2b()!*E8M|7l@fxsl6$I}rTKKz
z_mkB`Vz^q?e)n<7r3yPKi&@x>@qeae7ffp14;dbeSZid&)J;g?&Ae6U(x$?9%1M)F
zs#;}o+WxccFW3+`hyvb?(I3RVT0WE`nEl~7=Z3In``nQVY}4^P6HE|jofP+ZT6*nt
zX4Ra%7g8V)PTwjOm8?#Ll*|*ROArNvj4Hd(D$F*X3Vzm=4Bmx;#1u_ae!j3!tl45V
z$X(C@H6tfSug~oLrlv@laHt72bT>f6<=
z)Nw*d28Bn5-c@PowCr&QU^P!JQRv5KPlW6f0S}0(MeaB;rLl7>xz(lRk7@9)%DmpclO-hjAAY6jN1xLHQa^
zLK^X1pP4(oAP~_@3QdxV{$YeK@!>I+KBW-ZW#}m4S~0zlcXpnSIstzt?N_hcge-sA5Z=#Z{6eQ5<>UW*j05ldFZ#B75WaCE+1FJ#Vb_@7rEl~
zdi(G$42!F%l17^M0|K=0@+v})4G{+Yg9Y%urI%9p*Q0%=LVGMZX@%K{dY~DH|3U40
z<0c$K>!|7*^T23y9*Yf;P(;8zx@*0cMeQ9PANkNxSReH;dW2u6OSG29y
zvj1tBteN#9mdc?M5)85UZcEP92BRLnEL1
zx$h07fTuQm_GW>2@|pT|lZU`Zd2`<$RPzMrcHa#~Onc4d+<=5@E`5IBb#aV2h{gv8
zs{^aQ3@vWtt0eOst(b>j%YzreXJxsRlI_z1U+s9&=5x5yn4iVE?+hi09
zCmOy=^Vp}Df4Fs^w)vZ(`j`#k;a3DN)h-j_~uYMU2$ewZ(*5-3B`K5g@9RtRd(&vEtJx@DD>tYwEV|&pPs*E|-uXENn9?>966(OHXp;eMQqcA+
zbIbb3e+dw##$a{T&su->V9jZs8T$7zh!<_Od+M_$BE7dy5vT~W2;>#n2A+gpDot`Z
zFt3o%ELg0q6;oX&Nl^&UN$;xuuDI?j#B?UaDsWWM*8G);UPDvld~#?3cCH1{f-v0Y
z&sYHTOHw1wC!>Hg=cv?}Q*{8ix`tg%Ct7Cs27PcaTM${5C-JpB4c=X{YK1H_d*ak)
zvXIfx@|ohxsVt{TyL$Yz(TwJHO}C(!%)PDl?q$}fY24cF;Yx37u%U_DxOXlg&YC_=
zmC@K#Qg^wA%sR2K=k0ak6uWM~d=0FIhV)KOX*}jlJkdDy%kFD0R_hnAE-n~CP~dbw
zy$Ap#1)DuF1{q=!3ok6BGqHd|T#j>?v+?qUqPAF!KAAtJYG>P-{{`SL7gyIN=D5tM
zs}T2wAM$J_eIuxx;C8m@RJKmFcWrUGz>HK@(_f#(Z7=lYPDY$6*fiJAQk>>Y+{(6M
zW#kzeEt+{{H89jd=gb*j>4u1bV#TBs4tPj~IGpH5PzMHb^X8f1W{N839~LzKuI+Xt!h
zD8+(J1^FM`3L>5%WxERM(($P7Tm4yVnhoRQJb6-|6A{<1W=u&ueW^@Mj`GjT$Am;i
zZ%chgeBFx}&2Fno&uiv5c5EH3b4Sc^I?qx@bzdU3o*>zztatfT#hXs!v91V}Evcm$
z0j{p)`BqQJ^vPh-yF{l7merDd_F{&vRfX$)p$+aGgT3^TGn87ZB3#RTRzajIA|X3;
zVYTxTE)OMMT}RLGe>N~}5Bm?*!QCS`hb|K`vyQZEG_D`gC&J|UQyc@G3_WZ7C%4i4
z`HdwHx&7gbvYZu`8r|3GUGb1Ht()lC!i4KrM)IYqxi|2t;u?ye!RPv}z6w~CFRQd{T)$dhi>HiuYav+m7WvjKF^uU+fb+c8mHf?#@bL8Lrv|ag&
zHF2@$1poUXF^OR7_~6yHZ!j^~&YOxX!y~Y6;XFh?9Y|+qf5Itw@1G+7K1XcqxU;6D
zBik8X+RB^Hi){yF)qfb}?(dq;4369Be=9HqeLisNPJbfWH;LUnGKe5F~C
zu`$>h|*Qz+j4lqh9@-db1#6E2QLi>sE}|VXnZaKtaeq52mc90
zu^BT+;T8qO*1_KX`0pV3NqFm!|)v_U<>v_L%-04U(h`Dr?d2Z*%nf6hJLKdc^2rI_)
za=UKyp1c@#+{Reff2YMfXqL`diJ$S&N4TKFfqh5c5beNNnK&#yBfaRz%4
zPN>+?sHMw>bRlM{dq|gK=W$}HCJAXu;;mHO;bn<@ScAcZK}Fv;cGFP=fp}6bjnU?)
zhDx;O@r0XKdhV1a6R*)@I#otJEv(WY5Cj6W#$tD&0e)Iq`DJp8w%7{{Rdj5)9A(;F#vd
zPQXtDXx{&j80=<_VSoT43?E;L?F=9onS+Ceot=%Fn~#x=J>_EtfcO7qC}sh_pxIMM
zeP9R)xOos+r7RttoXNQO|1X!}WaH=kZ!XiRqpM2Tis}1QJ-z4HBwEm}*4U^n+DA?$
ziu;*d_Q1FXKDB;N-dolP{N^fes>``o>mWvqITSCIQr&Ia*k+chAS0H#FTsG26VC_Q
zQNz-GwNM8|E>0+9ngtjs8L0-x
z8ki0?WKq!#g46-@gdTT%sEIH?d~$3i#mJy{vf3FrXZ(3)q3tAxPv03#zNi^_j1fZ_
z56Dq}b{k*-(Cu_$Kp|}Qqxn%qIo|(*|+l
z&Mk9eF8E<+pv?N=g#2ZC4l$>rr>B*AWpbH}*;E9ZEWUsL+ry%PL}XiQN~b%K*N%%j
zw%jjwOo;pGled%TPL2_Y7f*aElZQ~=aK~U$0>h9|ZQVFZVz8JspFl&hH$)>VweB(z
z&UMV;R~N*9Ad&DKDaU|w%*`K6&9qZ(_zF7ub;w5Io5kjJS9}|Q^v40$FnUdV#S@{H
z#YbOCIQeQIAA9rfXnp34h&-W|Ru(2Q{*6mG@|Pn~Jf8dD85o}O2LUm~?{Uj(dEUr;
zK`6#HFq93-UMq*-~Hfr2BjC)p<^$?`E9&E+H@4tHZpS)k;+&AHMFMFNc2A*F`c{v=SjK
z@bLZ8w`O>mYb?^;12g(yhSao6rM49##5FWelA?k=rf#rb*#dFI*><~r2`tiPWbBSD1oA|VDSh%k&S4d#
z`N#jwmhMjS?~Zb&FmK-$p8uT8TwN*t17*%V@f(50GrG*>u%xZvv75hr7jSHnOqJy_
zzZP6A6ACLFqbjP8L7vE%In!?UiPWJm3!F1iBF=3I0KLr62G+ovuv^ke{KL0PUY
zG4;pW_07c}W9P&Q92KzvG(j9*hN5qgnwiEAT(whz-xjSGm+?qgJ9bw3NDU63dU|l)
zw%>MXuf7!M>#Y3vY5d$Gc(#R8^LDy9$EsBCi*`W^e&vxzN*8wRg-loPeq_FUZ1v9I
z-rMpKRs;KAEZvZ?`n$bj{R>fdR(y-dxt;Eaaq1D=_wDMf12RYv#g_qH#m(pEf=pb0
zNbz`pKKK@2vJNHDvPD9+yDb6@c(R>C_pDDwPefMSwYXy{QAd?5mPqYv4-fN}bVG0b
zaSj8({{YPQZfjzV1|4`uqyWv}wbPom_p|<+;U}jZo^v~dQo7H>Hxzya_O~g)oEJr;
z*3-Z6?~w^?%N)z+w(nmq*1Ay-XID{X8T92CH;_takGToo(0L%?TkCkgpe!&}w_VGv
zo6wgByqHkY`)z;L6gCS$0f~vb!C-caF&iVW{G?64AuGErIE^`_!t6|aACa{enuqW1
zojyja$Lv}Uh5lY+cqpw<0(zMB{~A!;UzvK`pcFHdPoAs^s$dlW6TGN~xg
z5;Nsz#?3e9%te4k+W#mwytflG?VS<*<`Bn>8a?~GIT=9_SdcgJa$NcP4E5I;tf2tD
zQ@&AKCElGF(ykso)9`(q2%UO`4^=ndg(KRwCYIo?;pH#-%lZ}6E?$OckmSr*-2QQL
z3b|`9`%RdM@rfAq>if;<6#f%te3Kw0GFC)|AdzrXu{wXVV>53GJgr4FSfo>k&!#;T
zD$bQ!=`$WCt8ziAven8d*J@q%X!bOi0{4WwvqZ*3qor>t@UkWa?b4V$tgJT^e#2U<
z%ibVFGKi2A@c;P&+B$ALqV@8)_jtMz
zXC-}yh|&#!-l4AZh-OSM1TZRhM=<)1Z-Equq~QlEMUG_<}wE52?8I_JbrDw0eKZ+HQuh?cVu5e64T_|6g3{=QZSQwW}JvQ9_W^9O(YCe>!wUWg#)$AzF}C63ZOU(=A_`Qxm-^T
zI_XfVL(p7pz(04aq7M#XuRQ7R1mvxXmZM5OlMfuyHa^e97+Tf)APH~=s!vKX&JcO)
zd+JjW8W!E!LbT7ui;ZTD3w|3aVB_8j|6Ne-SWvU;87vVC*>3V2U^`Fyy5q?g!oL7Q)
zEPpoj8FbJSMjSYLg=K)NM9&}ur;>|09U+!b^J{>0`x&j_rRiEn8(73wOb8E+w+o~7
z`9?&yyt70u2x|XdMI!&g&58@eUgv(a$K$^#7^F~@1n`Tdohk1S#Aj1IG|0`9Bcri{aR_ZF;6%|wMS_%
zeB|$iTK9C~!{guK$k*>xi8?xf@z-yoC7=
zvR~tiYkqQR^MrZDdQr%WmG_`r-7Z*9HB<9#x|Y%wFvf2bL%D&d5e)$I_aePcjMi9}
z1{|WKNe~&}m(D^lJi|AARIX|PR8+O}Hfy31E|(QPWZov?+Kp0ITj_CsCC1Y^K#s;D
zp}*3aAF_8_h;>gfXaqVn(=s@ck0EPG^b=5Dzm_d?$sr8($zGq&Un6tA5X-lFNuAby
zinfgy+_+ZyXe-TZ|HP?bC~6Gy$w7#!MA`nNOW4EQwCLY&lG)fN#pJN%IuE5ZZwY6g
z&jK%sx5$5zB&AXHox}wq
zUAL1UqWbWy=Ag(#1Bj%+1ucO$jb^t)SleqXf!I2B-Gznv=b)9~e#rVu_s-g!F>3)S
z&C0X2&OSg$c}jZ4Xp2=Emw&T+arO!0f6*>{{3!x!0CF%tn?53|s->qH8T)^Ph9;RV
zI~gY#`@f`$ot-loJ0C9@8<`3stE{cL-M@(M|3ngGxMc{wC`xuql|d3YqH
zxFo-d^GfiF|0iJQ0FwRxOHltc`k!eu85bWr*?)^0;@Jt+Xl4B2n=`i@#fSSxI)nrI
zJX&F8`j}V>@`%lmcBa+DH
zpF;t2b5NSL5E`BlhQedTPpYqtW+#?*?@6uVaIzt&zCte;
zuCX13>Wiw^4ANoeQSS_sxNq%5+^(Y`pE{?WUalN(mjRM1$A;Ce{kB~@rKECorSY0x
z#%Slkz-2T*-5`zK3Va|NY-lc%uJ-LIG=2y;kPT*dyXO4zcQcW{GJaqI8^T_lm;R*n3z=@AoL(E_FQGDMz^FAfr7_
z)y;UZC$VjfH5GI3H?0ue8klBF_TU^+h)(?ZfrwJt`JJ@V*Hf?4QZY|*C1p>}qf!-<
zm)mvKqrR-De&1R1>O)a0}(yu#bx=J3wSDo1{
zo8h21xt?(>PNi`n_Pc>b89+}9)}CUdpu-oHFR9K18W0EYTSqtTxrmLmZdl6927m;K
QpPiE*k(ydkNec150LLjNW&i*H

delta 100637
zcmV)NK)1i7?hpFA50E7RHkU!00Vt795;)DttF^PWk-e_9a?0jFaUcz2J`|9Yto`-*
zx*M2>G%8
zBS*(3&%**O3$lEbVL{<nfU|SZbp2PQG7u#DpK4=|VXF9NI_XSvSGhG=F)z&t5QBu`1yWGHHIQ$^Yek
z!jPayb@YCk-8hW>T9
z3KJ&)PQL#gi8f?nx_S}qWHh_Fm3z~u9j3V+8XLpVX&QYVXW@@S_bk+Q#4r}<~5@*@nWIby)f+e+1cI(F_gPl{d6I-2v<_#*atxe%4S`N4h;pp&Q(-*6Wt^C
z$5w1mj39^SIT@FRW_B{O!&PdxrTC)ZJj%j96O&T1vQU}27Y*hAR(R9cEtJW)t}cqG
zo@B>U@Fmgkkx-Lm`@zOHY#kCn0e^uTASQyh^h-qn{i@KSXsBl6om2pj&fZz)tgl3L
ze=beDkRl1vtd!4=%(Wa_}Zi3f7RmUPB@Y9Kk^0#U5hitR_u4f`5;|WAFfT
zpsN5p@c)*#GI%7SGA-CA2#*>{AQiQUf%>*%)h|Semm26ShR;L;2ngvxnD4`tk4B6)
zb1iNaO!Vp=7M0>C+A1sjp3bMqI3QGB8Zs^8LZOrcv`pd+`KxuLwc(g6qCj)Z5w&fn
zW|<}#Ok?jCuqE&nn}O)ZX@3^VTl-<@Y8Pbq2%%J=28Sbrr`&NfbJpvLicJ=5W(8q}
z_P>jTu<0j~;yq_H!DBl#?E}|E44bj{w;8NhR`~Uyv%~1i)q4s^O#g#~y1{lb3=_Gp
zyCAMdme_q<1zm++`dG+Suf_|Dfr@|qE{j6N`YNC~#vt#^M3uz$#s>+28rs8~23
zsWdO>%&hB2XD@{Nv>_=Ykb*gmz4Y
zbdv@a^3cGN<-O2UV1R&m=RgiNruCtSY)8(Ql-ZR#mL2B_*AP
z{YfatzO76Jk?{qj4Oa4!$4~9N6G{$E36kt%JF*8cvfpcpB!52&P(t*s_C3m;GjQ<-
z(OXYY%{6I;nmg>fOVU8S3d#gEATyz8PkbOO@U*k%98@!tdhUJcB{ml^0Mgv|2nvFO7*kuaw|i}fehd83;O-HhuQ5Ki}pa68w1s0yW^@*QAbhXxR(W8h%YsFJmO
zC@BOuxd|-!6@PHa{amSeZzX$mKo%KHdljp|5;N$~Gs%IE>kI(bEu7b^Q5Jm#*&Ln7
zIuWtg>7{2>iK3Ev&p11v7*4(viN&aHJuI_la&cXx)%6^oVkVlF_NbEi_A))-LEYVt
z4guMFSqG1&LQ53+cZIkW5sLU6#m;3K1w1)^E2|W}e1D3bo)JbcMfyBRulSt0R^k{p54{HiJbx*aA^YLZW%bg{QT3dJs3lu;-vOCsl6`x7$b3
z$YFGww_reJRh$|VLXYD%Eg5{5ZuUzt<
z4Ors@HGhavH)ts-wi~_!fS?&^Y^V6#?Nh>MRHedO_hx6$xiRd>a==}Fx&@EgVT*SC
zE7{MT_h>j$qv1>eHYWmt-_YA?MasT?0bS|1LXMwa;D-SfjyLd=E?%gNoR-FuPM>#F
z(TU!}`W45jLVW3_?kjd~IKXjw^5|F-an}3-Tj*ElOE6~%tY@dbn_*MYExorgn
z$7T#6GL*)ui2kx%jzRhU>1eZTaVlj)%7BA(4T`^00oRxewkXarkZQ&qQT@
z;fY{{%Ci40PE}R4p`TO;8IXSa-~K?K{Psu4U+Cx$+ga4rt*lspKyEL%a7O1=W)0i``)^yWqiYuG)}~2Xcef?51*oH*EV$KiFrQUy!#(ISB
    IF~`20VsdnTUnFiwh?}xU!g}%#e#{GppOT~l}*L2RGfIK*5%`YD{)86N~BD3
zJiGhr(~Slxf`SgujQx-&Vn`f~Mju~)(4>p|MY{O=O*(u2_3q}=SCTJsmK$LfcW)P5
zS(b6JFgjyG>c!o9@l*11%Cp;F?!Fnvefo-LlR+uVExdnE7kt6BWhT>y@jDp*uXbIw
zw@Z=dNxvznQC9b5eJk?hw_9zJV*7B*6aVSwRHsF~jt{Uf07jLozG*x9Qmj@@yRPc{
z?FSgj>19>ZvpKBYyrt*5T(u=ES9D9<6-YgkU1=sPi4L>gE~S*y<+Vn6%Z|--5-4(s
zO}SWd%Vd9QuG1rDDc5shgkgE6NSbfUau4Fb{+oLHfm;0rw$gMZPW3gMxho&mY{zTb>|epGiG6YFwdR_1gDoVmE|zR
zA6uHnCQX^(YUp%Aox=`<5=`1~;kRYcKeSGRu!F9uS0yzcQNzAnzK0Wst$_l4zg@B2
zQj-p$t%|;CYCjFyy;ley=2#bf5x4Xd48!=d-IQ&~jx-OZBlCr}hRH0Wj$gg)-BD*5
ztWtmBx|v0pGq)bRExLE~vUzCX?0=(Acr@L9L&QB9U
z!yg&_(U$9n6{?wt+l%e5916ot+gaCCo+=bUWh>uu3>)vCbb%XISH*I!ZNZc{dG
zxjSz`C^Nj&!ZNG8$ggj$a2i4v>bi%WZ`Y%S;SX@fo?4?_L3>jKD#i)uu`avnzNYrB
z=-cW&$)!}pp(L0!2N46QemK7on3(0Dt6Kc`aFZp(l8x~6B3&e-mJC$|@TOGd6oM`#{=k5#jM*ww>F
zl$^3JS-lLAz$m56a%Qv+rTpofF7~FM
zvq;_wwf|`_a`h;jTQ<1BeQUa
z5Mht^)Y%r)&n0?AO?p1DNi9eJI)iM3?P2wa+(mzjCre6~?*W?%Z%7=}=w;SMzE1
zEtV2(NmZ0XKv6V}=8|D;14>Ojw)rL!NDNa}9*$9`5gU}uW_sxJI<^`FpwXf(#_%5_
zVn79!i-{OmAp%PLl2m^NAOv!bVOZjf5%K14cQ?P@jAcWfF%6xRl3%}
z!$)*<3DxhT>$-z*;7Z}`;`^I_Q+XY1W}FER2=VkJtVk8eY^7O%f(6Xydi@ab0nIWC
zNkMR^L~`m@MHOIDh1fLH>iWYtm4$pdRKa>dn9$&Bd@<=XWTJmitx|#3IjR*51!DCX
zwNmyA^co62PL)Drd4l|bzM$mcSt!5ekxWYJf?&l}5D_ETU^(ga`DLJwNf8cTX)#HP
zM+)TN-qJ$c2KXSrexIShajId}>v2d-gVXI~9L&Ln`}y)rHsTh1Bb^O+jo2woK1L)0
z4^AP*MMOGTjHiE4g@bzqD|`akKtj=kW$+p4u#necc03=WzKKrod!&N8NKcQz@lTK>
z6nDiLyiYKv47DBpeL&;XjldIr@!$#j*O5GZNc|}YD8oh1=WTr&3{7^UI2e?i=?O~>
z6w_rlY+Nib$4mxTQsusq`!NPhIOFDp;GTX#WO4yB>H&W<0u)in>kTDjm8=^m)_8JH
ze-TpvdHQjZyIvPcdf=ek!zu!gZ`ZmV=TGRD+$X*Z_u*-V(~)
zzG^*jL*m2+;y4ICfoP|4#>JE|*|QWG6mH?XlmeX4Uf5mJ^=C+NX%$Nq4`K3{wZKsov13j*r<^RlSSFs^4>_++f-=H{UksEmi@
zoWF{AgYNNRTHDDqbSe0N3mpRgLq80ymcl`20~Q;+;XAUxhNh)gp`95W9&2YC$aIvn>HDL#KJ
z6IKPKOd8kb1c^fifr2HN*(C+*k^&reBzq!&@VKJzv_a!;YrSp|iG-8*q))7~I+)!o
zv7gB10=5ql`cyOI{S)ECU1agupl>c%mC=lcBQPULG)m%ROfo!)dzn#dVv4)nP?P+l
zrw@*+R3!|d
z|Nn4bGi}h%LuN;u2bHFgGD=lV4MDiyGo*nL%uOjYg5PJlm+dJ|i>VkbCIb%^G7D%i
zZ<7Mflb^RqnF9P1;stDSoaq{y)Guk1`dXV*mv03pO1z`8Xt9JZ1HW9w?$03k8L%D&
zMtBX^_e6`AGm959hGrHTbIX4(xrDmPCF}*4$YLIm2Ogo1^*UQtCUD%)W`OCF|CCnWuWZ;N<=HcEHlw}b!dP?}2Z2pw8`5&e2KPKMUQ(hV)_Yu=E
zlVfB6@x)+uQ`9UE2oVdkq8=dgV_Wt9?fRh%dA<=@AK>ZXmnc%k0iS=sAjY#flF4W4
ziKd=lSFk1MKIwDX!gv7%F5W;hFOZFS{lR>lgudBH`(a2~R#d;YA6-r(!^TMaGjCVz3Ex
z{j7|qFJ2@b>H5hjk57L$cQWPKm?2%%cEsZ+yT2xFd=^U^!a&8ub(l8(N$kpaixJGJ
z0T8F-q%k5SXZFzH@LMa-`dI%v<1xvMtf5LC>UW;G@X2Xkw8gIM%QpHW5esoi^K++i
zY~5+(SR}N@Y0l@N(Ygsm%s`Aoczad6@%|3T19&b98cLVQmU!Ze(v_
zY6>_oATS_rVrr9SVh#Z~mq42VDSzEsTW{OQ6@K@x;3I`rvz+@dMS-@9-L}CN*me!<
z0@*y!5@RzfkzR_jll=OgGZ%?NjV`v60_lr293IZhxqas{F0$LK$o_s+_}{->Uw!jE
zqZy^7l#*rF?=s3c(aK~>YQnh8u9w-H{Ks$qI-7GX^B?;T1~l*9x#vaK&VS%>Xzb6n
zUe4ZLzy9WXs%j?-(zyf#GjOqLV|M;1<
zn=z9=x`%h&wq3g6aOVlO?Qck<^T*ar+*f_Iu7|qcxU(=bK8rsH7lH`EyAC#FQ
zrRD}d5JYjF%_$M3aI$DTm46yhF=A*53tNKXv<$dZgewtmO=fc~On!Y=Z#-X%uJ7y3
z{cH}bbvv%RJ@>H5j-8#-m?Ea6Zq0kt6$tNry%<0-bIv)eYNxt>Iiq>gRztnP7b34(
zPw|NIu03_ty0?#gGt>!jDnl5{=1daY6q$eRsrZ
z_0^lVMYe>Oud{;i(qxa8be&03P-JbD{dDzj?@nU@g{5RE2o>6GdG8L+$NXzz@ZgHD
zvUF6xB&<;qZ4}6x5`RX^*c``FFhmssomxwAxzs--HdI6^2~wrhFLZ6CSY&f}qjhwI
zXk}moUNcY!Keu|qxi}pyC5eRRNT)@SFh)BH%zR3i5GV_@Ibwz-&NMHGfKh_K1CKWo
zkDpp%#Q)e3?au-{uxmF%f@8r59XU*qS|79v5PjN^!P1E`@_#ZzF3kx$4l6}=+$QW8
zFpaC1*l~o`1w;!O8S*i`(l4o3b3(7L5@=sIVOnGshG8F_uY>u`t-&
zg;0l2!(BCmk*9o^NtIW9Q?&~}W25kxCh
z84!L==O-g_e
z0}DNkZ&Q(38|IftOloz3#ME;n7V*jn2MLad-bTSx1AqG90W~j=gAjePC*E=g^`B;l(KdP$xfj#KZE^(i)|VQA7Jnf
zHegVHxVNwYs8J0~*Sc9SX_;56ttSGoagvT=Zb=HRtq@DPHxp?^$d1@;IzK;ZFGDaGmJSW2I_TsfUa$81fZZN%2+S6w8-
z1JxZ9BRywq=6u@weC6Apl;eV*6nv-RwAVuRhkz6L)Z<+p!qmP%6SiJ9`NODW+ipUs
z=e=9x@2bTRrYb+`w!IrHtD*80VAx;*5fB!H3V*s@5{L7W@B#y#Z4&WZV-Ei(&`)ZZq{9>e9uAoC$<
zIjHyIxuFioNy#(8C83ZWJwMWgg@9wD;%r;QRU?J^4QhTaLY$k!kyFu*LFi#3m0|Fx
zhJPvq;Kzt6iDU(lR6P~Rj;rF#srk86yJB;E>bFVo)JmUrJx*=q3_b|Ol?sTq_pV)k
zYBw|7;lp+lOuQOWQe}V?7>7AB7A1^FlSTD^77RxobfWWO1S#NaZg5Bb5)TS1X>4Pb
zr*lR5C2~daIT{h=MFs20SdRQh6>BnxM}Hpuzc#sJU@l$^=4ub-sTD?m@U)(zIh8EE
zI@1LAEdee$k#P^43+m%A4p!@KgYjAv7(4s=k2sclKoWEcNHEjg3{~SQdR)W(^cZ1R
zA-9K#8{1c|2irKe?RH!C!cOa^ZPv{v7udz9)f{*6vD*x7Q2!Fe%
zyzcw1cSF}-hJnihgdWHLI-=2mM%hduMY*nqzWG3AbEr%6f7ufQhxX8^wm4@c6hNgr
z=)UUlr-NbR@w@(g+c;oe`6S5WdgtiY)rY-s#ue1$h`X$3M3QR9WA;V7Nkz9X1z?RP
zCKPSgb`dRen_E>Y_pll2`!n4qcYoi)47UL&YCbVzK)&165{Dr3zIkXMqxpfBx63UCEz6XpRIa5XV!ZWKBQFx)|r8-r==yvBa)UpaR@Avdle#fMBbI_}SDiR^gfB?K)U
ze`6gLcD!-@4ENwVtk?n_?ti-5&UQgpM{P}LBEq$tJ2=XMYY!6k&J%1ZfGkYhAm<+w
zef<4}lCfHg)_(s!ArQRl5YGxp@dd&(B!$f;X^4c>2fMUtl!my=on1?or8=N{u`vzM
z3YkH~8WzAtxdiXHNwZSICliUIMJk&AbM0?FHy&$cFh!?;CE|T3F@I$a$b|!C;~hNz
zP~G-bdz-!_Wl)}p(7j&U^nEon_p7GvcNMmtT!C5~_oK!)h=(l#kbl_W~UxwMi}zgKvH5y{y#cwr`>r#
z`~S>o@{Hv?w&xis!@t**(FcHm{{>ovuZfdPVkQGKFfo_$=L0E!jal1{+_({a_g5%S
z)&MiIL{gNrK^`1$yzwFzBOWXQ*o#1;R!?)SmU>0C$FpCbDppaq){OnqO;Nm770)?E
z#ld|L2S2@t`TwuCFJ8SBA}FFl3mx3MYP{773XKx4YnDczdX?ZbTl3JJZ+I
zV7itbheJ6)DJ#N%WBJt_%^$V(FSx^(E3Lw^H)UIK&8IkxyPnENbCrbcX2+^)?a-gA
zo2y@Me-7eE#Yu3JBvFy_nrIbLBw^k6T@S-1^k^*om+rL)Yn9^4*bm4CXr
zQ8EqR?QvIW_;4liu%>cT!@BkZJsk1Hxf#r`TU%cw-0I+}K?aRu?o@Wdztm2(uDRzu)b
zFAq2;sFifb6n@cJT9d2qvQkl>|)C
zi$2Ct4J-*Ey+2xdB67)@h~T*&4&8m%mJQX99)WjdyQ4~GqLEQc&vQAChZ~1kxYY_x
zc_i;Bx!xx*lfy<_G|bf3cWC;@d9bP2g^+jrI5gB#tieXkJy+b>x~F!yOiQ(OTM>AF
z?;LVNjkC{uhFxvVefw%(m&3W|2%J;w0^BaV1^+qo>wmYr1=yX8JV;@?1tJ2}gc#!%
zSCPW`bHtO;Ug3G~4qECR1BDI^M@2p)p6T#Z+GjQ(6Jnjk%-=w3CbHyN^yEf=-k2L2
ziTpd6frk;_H`h&~GXQM()AW~p3%;d)P9q8MvDkm*jt3a?&e@%fJfX+VzlE2NR4z|9
z7aGY0Mhw$tG%NtBY40(zga^mO`?gqSz>51(MGnBDXhRcvTqDHk(AJJ#USJ=8(8oc4
zANb;betHpngq`Z}pXdH`;4_Xo(~--eJPmbE6^V!h4?`(|J2D_zwO9QqsD4|Wt!3G-
z-ErO+Ow(ZQl>_N%{vv#QY5hS=F7}y&)=U{ohx#j8RA?C4QlV^Y$0htrA!Q)6)66AU
zVork+L#PfvfalI&Yv4Aj_~j~p7mmy5@kbzz7Lldl8@EM?iiL8Zf$CpFBt2TMq@T)ZknJ*m336HFQIhkK
zoy{PW<{9u@wY8UASU=9-o={J{fAg9Yl2BRbUw8__@*X{f1tNE7b?ht;$8(&{jc>Fw
zaA?s+Wn%^!WsUXTPl*O;ORMke-Kg7AND~|M*>FI2r*R&>H9gKKQxO`LLs;DL*373^
zDwH(XmpLH`4-TCuljys&KGXd26>wJ9Ah_JqY@LVA^q!9UHX8UD_QRG5@Xj`OZGo~p(oy~iL
zB=m0gyT*9B8%o1Un36h=RLpo=U7TKvz^BxB5>~lY
z#mzi`T?CezI1|!;BID%JrK1IifHY3MG&w;)ms?;Yfdez?n{_-Xv#la7e_)gxb*E?4
z8IExvr#ez812(`B_5%TJE%3jk2T$p#-ew5e8sVMyRmK7r-aTUD?u^S!#a+6$@C2gQ18
zQbeg*`at=ydRsSps=>4FBme?i>(S2N5kc06OWN_?6dacl(&!q|;hJNEW5nziQ}`I=
zk-VOVVq@-qddPTc%LDAX_T{jBghU*rIlBi;cwaB>8Gz?;afGi=9u&!Ec?P`2I$q;E
z(vofUZ%*>&hKGmMWgC=NY_(g!TQ*A|vjv2K^w~gz>v77Ar_?Okp~8`AJ?X9|P!sfO
zq%+RhHfY;XAf;TfXVuu*iOw885YM#Srh6u>m$8O_09()cjoc)`%Uk)xN-fPgB_Z*c
zC?dsD3__*h#m<}wI#T2kmr;JGX=*?!3+Y2;6vSb>(E-A=7SQY1k0Un9W#*$
z_wS~E1y`K=+Ab5X`L*%fBVIT^7#H<~7|5D)Q;$a
z@BRQgYo?rAPfmMMmAO}yGMh2Gdn`{U*3|daManY5|vpl)OG`>IQ$h2>dink<=Nd39LoWd$>x#Um1J~(
zOOHhO*v%!|@V_Mdr`(YgSgXcMK)MLVZ6S63K)!grXWGggunfcKiLnQpJx4LyTJYo1
zNd&Iz7Wd1UH`_H;&zLx8GC{9aRc-Cmk;x=h>m6hT0C%yW!Q~c#(Th=&s@X;U)R>Ye
z1Z}(eP6da2`XWGHn*-34bX6Tq2-4nvP7^&5Q;E);u4NJAzghDC0VLxeez=C}W_OI&
zbPZxVF1Spr=g*xagZRxv^tlMS;iKc?2dpM5fM$?i#-MoGdbO{6Cw`yVe#
z72kVdcsx=CsIUFGp_V68yVIKTr@$2D9>x=dzx&d{b(Y$&JV3$c1SgoAE(z6t`ZgSD
zdZ*k?M^7;PQ|?eQN;JEpF6ovAw=S2KQfo^Ul(?b1nl8;v!%+k6znGH7;b3@5vp_ne
zQoUNp*AsGV9MT>yi2@qamkGx+|JxWpi&Qq=g@1vSAgQ_ala)3Tt~{0BJtht2Z2T;d
zDTQCKR~sC^F?c3jeroZ~kx3+8VvetO18dxNXYj_&dQCeD+KvgjpR#@bDQ0ZNaP*RU
zuRo)BP{tAyjT6uv-Ez#PE58~$&aow7lB9@4dKp<#ejB=?<;|?bAwdIx){1ifZ9a(?
zjsAA~;{SrdpgfZ`Y$pRaF))`wn*k_)tyx=-+r|}s_pi`PQ3~Cia4x(B=tGm#joSt(
z>TQamE{a)dc12hs6_T>!{Q92TP^4DcwSm5f!{J=c{e0&rT|KVS)t_Fa^8DuR)mN{L
zUTLkWEHkUShn2Qg6?wVJvqBl0t?o9fe<$DgdTn*`etm0;EEzj_ZUYa#^=%!0dLDo0
zN5B5h-JidDt&3Hqs@&vstV(qSccv>%SHNs;OI14fx`!M7xh_qD$C)CjeanNkTf^^1
ze94nO9M-p{OoAVIg0F`K{vJLQ2`&DBKQe!R9A&{_t&_v92^)SV2H@ETzvIycsJ`K0
zBbSd`*whS>q(Khtj=h`?yN3~fSw-f|j5iXW$yHftX0DF%v1PJ>Q{ghTzIAz(d>{J2
z^B>p98;|R5e>jfvav
zq3$v!duaB}&iBkS*@$&fs1}T|y&Y=5BO|g*u3}4HdQiFog{jgQ_MIQPUfhKAH1`|i
z>7uKo**TrgsZv{}^G!Q{jKLEbCg~pd>HGTzg7PdrhhL2bJzD7FR@`7ue0-;aCr
zmwPd><%yeSA6iuUAUivL_F;3Z<&aH>r+jSViW`{PkD_WSV8=h!VmaDWT`ZNVw1SYZ#FzQ9+sE-
z$jMWk81REsbTS;@r8cjdKxB+j31Ci-A}TT2o-KAR=t(6X)>#IMY@(Z+LP=;cW7cb_ZU0vplQ@?;3(kep*w+iWU;0_f}8*AT3qpL@lIXF_`RR27U)>
z1r|Q>JouF?kl>PkB_EpM2&;bek;e=eFgcGJnGu=8NP?=(kBdF|fPlzD
z0X&p05kCY@TOLjK#(5s>A^z^gaX89djtUm->xxUe!m1QBLsTzFjHiyV@z<>h=kZHtk
zF+1i3n-Z)T1Vz=&!zVT*m}L8cRV`6)ZuGP5hrVg+=0Io>rcVF_^GTrXRU
zKZ{rvV6`sJB34Ne%Zh`56@!DYcYp+MF2qQN$T0i7wE_-OB?xnI>;K(a7bnemhPVM*
ztCrV)M;7lH%`+Q5W(%8jUSuH)C@?cR4w6DrmC81@j08-s3!KF|_3KTQnG`7L06XpZ1+xc#kBD3T9%{;nBnX^o_J88!`Az8Y=A$rM8g
zo$SND>p#usfaHF|DN9F$lAdmR`_p&)$?qP!9wf70-T}c)T?;`LLaDYQ>(^MW0+u^}
zwtftNPv8yR6w01Sp(ILDs1z?aKs@&b-Vnv{%tXn+?^`crEc*t`h#!#g7g+!B$6Y5e
zZG(XiK+in#T%sRxhc4Yzk@9K70bv6SW2IBf%Q9L)ihx1GJaxPpkptM!?fO;&g;-%S
zJP*PLyCy7r2$cP4il1*p9B!co_y9qFbE%_sh01@KQ8TN~jCZ`lnUehL2KiLCV88CG)<2@0%$I
zalP4Tc*d2*ncD$`GRf>3V&t^Dpe*0NOPY4JrObXlofX}?T_
zxlvFiog2?*@NQoqrZAA+FR=bBka0y0R$!LfZmwE9;7DEN5zIxL4_TM1(`z+bsg*-h
zn{R(Xwz{koRFKO(FUVGZR~7<%vD|aTJZ+Z0Y@V}4?L`$qF#de&GysL2EuXN2)|nO0
zTBmVFpM^p@B#10>t?s&N&geo(CPTXRF6tM)smc@lq
zEt}iTrCPsKSvoCfZB@}+D8)^RHA%{UeQr}rH+7*PWu|6tn3~tmEpt6J;B`(FH0RPaC
zCgrwy+}>il4Y?X(`i2j~!E?!4k-tq43Erp9=R@%0v6oe}0o~z--$U!~cT)qiC8&zE
zt?wt=^K;^{IMUKTu6KSI8bmMk)Cn$PtzOW-NmMEi(>E@EQgQV5c$gaHJo6LwGebRc
zF;^r%c*gwzp9gBe0`fx{AfNGnFfV&OKM-oqh(D(t3!_{FzX^52bU{j@(yM}+s@-@b
z`4dJ;E;aGV|4q5-q>4!oI6KK0*NRebzV3RclVbx3!66;i@mJ5Ug2Aq3LXG86D)MFx
z1t0tI6APz*HVd+VMH*Mj`K%c)4H$?lj}4m;OcS;A7o`TID6Oeu&;JBLwmrtVsh2MY
z!B)QyQxD*YsIdgfMAYnoW%NL4X{GbVP)c1o9OK@gvAQcK`ws|8*05+`^URi{{u>@-4K^C^aB$Z
zFf<@AAa7!73OqatFHB`_XLM*FGBGqYmqD8WD1W6`OLOG55x(!Qm|Hki8ZHSCeE7DG
zc9Tt{jmuK`V&?#n&hW&(UG%*Benr|1tGw`_8R5
zQh(;rV?WTKv3t`wSkSE<*Rb!E=!0n{e_)+P@p_YGRdm0zd^&t<-L|Ho@juPbnU=;s
z{PhzJeo0jF1T*-iF_FHG>wn+>W0l59`iAdJ#iAs({S%^!`p409G~A*fMMUB>MAIXE
z4})=Lc(&b^=5;%{(G2q?^tC9B8+BJ3UVqVW{SdA=Q55XFLXMND=MApkn-SFIf&k*2
zfA~tldNFBYV0
zh_Kh4eeBx?tyCn@mvxyE%dlpC)%!zhP>?KzcMP$9&bLToSrFTBQ`@5-yfCves()GI
zhNhGA%9B}i2Rf?llN~oVMi1Rzm<*l(3*Zdw1{p}Bouktvjf@WdPBVZb?uqim+a)|b
z?&zTIkl~MC1TdS%l~Tg{{0jUaE=zEg23Wqi`Aoac1EZ|Y1XYEKlS+`JHzXGNF#G(C
z?LZD37t!ZED&c{+Y#NIaZD|Re8h_qIpI(MT0o?YWVh*h5wU>{tCUHrzwD34_uylRb
z&JQmFa^r1R7GdkB$m+M7>mQld1oUs}HTd$bv;Uop59|~}(yw_r
z;FPB6Ejtzwy@|;uVCJtW`5GpX7sJ7ZG|PcXGi{3zBBXaEzr`D=
zAD;Nh2U78S;>ao%f(`OZBD2APNTPz+G}=vr*P5t6>pedCLKCCO5;PkJ?a^u8YdZoe
zx`;dk^@B6elpZk({mGt)c7Jp<@Yq2dfRF}tvfUlMxgTb`RK+Zbi1S3Q6vUM%M36)!
zb=<}1JXaK%x@$O$cZ6&*YRN?X(~K9KTO=Wt-+_>|@t%zU?6VE64aOsPhwP1-9T(^x
zvFRN`f)OzVE(8r>DDq=jhX&bj(5a{cA&oYeR!rZqqZ&I;&%Nowy%^1FCkhiRe6eP|6(X9ln*wU
zakAtjSa|33SG4g<{TWW6(wlhM8>TW21;q3$?LtscI^+vj@h;}@AO?Evq=3@42b
zptLyKF2rIxjwCGj&9_rK+C?=6Ppb+9(_A&c1Rd=5`p$#-<$r|7DwHCRb8w_gKtTmz
z?N#VggJD3sSoQ%XzMvb}mLYW!>o$3$@Bp8H~XtS}$x(lY9
zSjS^R40tl12mt8f|EKZ|DUtN9{(#=Uf(z)wJ3%$$V^zjUk*{Qh6&@?J9t@$dl!~xU
zMcY9ib~JpLqkl$03DtE5tEG{FOA-8Y5U2&(13!y#IUML*?X(*~M!)ba
zxg~c)eUx=1B@iz7i9k+%+CYdmP!^>}4y})OZ#(#OvdcFPj_3oa*cCWXgt@Addp}3w)
zmOmL;Z=f|h<4gc<6;QZPZ#FkXDuTGH2k@4T3Kll_Cs5C;3ugB%*)=YI3Ndl?30;p<
zX!2&D_ARN!VjW?5t~}MTYq)yP)&elAD7ml%Jb-f*_%zq~I6f)nkzjdKiv^0~O&-K5
z6@Lni%Ev_w$4J`I&X@{I`d7o}xDciP07qEl8_!#mz%vw6rDW}T*{M_$IY*^97REw^
z*ML~-ZeHaP<1$Sk$infS&Nc?M+VQ%kkMBQTL(_j~b;mFDWYnIvQE7-xQNV>a%Fk3>
z7WuO7!0sB9c5D@i$Vo!^pDi9PN-Ts!a)0)mbe`97*~Y8+cg*67^M6C0_+ndb;i(@r
zhlOj+VCCL53DmijV3RzC_TPgW7sl@7-vPMm1O3W@HP_P@oxX}b(Iotq(=^?3XEy&Y
z1KLTk&O-84ltM246|A`T#mrQkiy@%G&egHrK-WkSZ~a|RX&`iVp<_#ze@W6s7;he9
zFg?Z~P2E(Zv+sD3vz(^S-E`pkob;APKdnp1uJzat7a>rjNu03SZ@CDw9T%MlBzHgE
z1`r9qt{{|=N8&mQaXL8{dOv$!y}~8jV*)4Ba9*AO5SMY83BXN}#CgV*Ybs85eZK$j
zKNm&1SC=vL0}}!=IhTRQ0~E6pov9`QGclKe#{&}+GBPnZ3NK7$ZfA68G9WTFH!(1m
z@Yw+r1u{1^#-FtV>@89QeV0cwmb-lZ~x<6;m3>lH43Z0ONoe@yn&eoZZiJp-gASXKeytre|big(D*qaReGVTiDr(8ae~H0UXZe00m>`k3dI%05c;a7aSQt0%!|#
z{1`L=7gJ)##VN2PF5BGLt7JoG`$=>K;F*n!)O7Zw6g^m0nH7qO#ybM
z09BwSKutwVSp^`Wte~c-e?mq7F|mq^y}g~|zqN>{sH#cO0z`%6RmA{6by|Rgnu_Y5
zzp6mn5B+Ae0D0Aq@jrP!9RGyNiKz;yYAK2_G5lErfC=CVbab-#llFhKk$)%y{4MPx
z)YQ?==C1?*N^@stdu|2>H#awWGZ!akdOJrmdVA}?^r@O#I04-3e;lmS-0ok4HXo(@2>P%)|3}zIA)NorY5gyEfD;h-
zUunz@o&JiIRaBG(*ce*aIsUC==Q%mW@zi;|!bWs?(@q=wo3D{CI)$
zax!!U0-POPfIeRT9Qbd7W8wgqSQtBhH2KFtf%|KADO*!J0N20FA4>jj=-=B<`PU((
z`Z%E`cDB|Ye*hDpDI9~mo%2UCQ2zgO#{MUlxQn&5yrB({^1loEzg~tm7S_T}J>VEWdVraY4dBJ}(aI)3_rID1z(8+n=ll@@uy=9x0hro3!u{Du
zc1{3;=pWO+hzr0V_8;P4{>c0{u>csP{w7ucgY@6T4q%Y|n>YXra(~lD2KoOGCnJDC
z@o)Obf1vy~ePmGin?5qA{)f0e<~019KISy~n?6(<|A&|uKT??d1DOB}z<=OJO4EPf
zM^3YU;73yP{~#;d2YJ|=18x7|@nN(02Y#&F@*nsi)9N4iq0{;w_#xEhZ)Exq`j45J
zK7`u+13!e?|1IUi*WU2sKC}j!I{$5D`nUC8e<$oe<_}T!Kt~I^f0TjgV`Yy2z>j4(
z{R2NDod1Cz%W(Mzek{ZFZ)E<^;{Ff(5aIC;{LtqLaKX!oaT4m5^aUa&Lf39?LX3@Pyy
z%bO~NB;%q#Eyv{7l#YW~PMd1S{VbZjO$-#s_iy$~C(pN&mE_j*P4u@d)FcbSuzuTU
zNvN3oZK5Q&W9TdAi;E;CG?t}KuNIIlf4AgU+DS}xB%PJ3Rm3`z-Wu0R08kq{7U62|
zdW>fN`Hp&>M66rdx|Rd!RxroZ^vTg0&3y?4rvrCsssq&d{qa*wlwq&XGWD!MZV2X>
zG?aZ$7m7Aba;~a=edWTT2~<>74<4?K3Z*TAn!A719Y*>Rc!d-J>vzX~VMKajf9HYl
zDZvTm;No=I2Wma*)w(=$Unr58Ty+=ke6y1|{J?&wNPP|5b(L>kUX&EPC!C!|gaKFw
zQ)>F)^5x@PtH0r2v+^9PW0C=&3qkVe)R;Fik&dkM=Z_k&S@~Z?unbR
zu&g*53PjNba93(T-b$|WJoQ(B3F;#$QS)lU*=5p}A|X#LX*?c;+Qd?GU5jUN-@s^xw`uMGXK;?)q!58
z9PH=Ul1KDy)b+`qiHH?=f71a_iv^I=cb6S)eS;RX=?B?zg`w?9n09ZH$m=~751Y(U
zW4~%pD7LgTj)nR!yYHlg_Oqcxc-*IbRO}%jbSH=(TWtIRle-8_g|9Ko#PPG7S;Dzl
z3WmjVMFBZN;96``3#&Ahn%b%pVW5=uFUdQv6}U<;z2xe<8;T^n
z-R(mKGFfIW!N%nF@k5>=70)aN0(Hf%L-xb);@T?L*Yo~pufNh-@}V;kVuN*+NV8~(
z1shLszc3|>PkvX5eh@G8PvkNsxC)YZspL}jikuW>|H
zd!1HsfA!q{DiARg>k5iGnNN@!QW
zngO*vHxT7EXC(3+&xtI_VyBXvs}t_0jbhcYo3g)D`>U@n9LbOtMg7y!e&}7Y28;d-
z$L^hR$&$frkVrPUY%l3%Y{bq{+ip4B@|V^Q8ueHp0v^U@3Z7=Op=8nFe6b~x;Batky;rW_w
zm;=3bT-w}wgjDou+!ct(yFq)|gvAST+GWoB*b0We#r`7de2e18qPjCgM#hOn27UwP
z#*6}?f9UOh+w(kK*{dVK)uq3ZM8IgZYCxopmI`I5*;VC<*VJelTo`)BE3GpVUlHU6`lYa39TOty7g+yO=y`z0k2|Cru{zhh8Fz*$ZCUFE=*B-p3$&4fpVgI*ytm6m&ig#N@+SIPak7l-FXo&?-w?yGp*
z*3phC(-Bcv#GEw^ZybhjU2b_Kfnc(XvP4+0|!T
zT8Z&ji!x?xMK6MV3`kL-jVjvq#AZ6`Z>7jUgyJX>*M(iN1kK|L&f4hT*
zk-bsg7JIHdopkPyUW=5*WFmm%*_Eir6inIPe=cjdFD==KEkl-CnD-%7ma5gtSI;wtyM_sh8tMly
zc`>3PwOJV{plne&=qaQkjDiI0g~4ybs5Lj)auZ$%QbZK$PIzMxWev})?)2idEvY?8
zuW>iVY*gYop)L4uoMVNMvzik-@sV_D_)Vvgld}Ay*CQl;jq#LUzB}KTAo51$66|qCJ|2!>X3wR5wtT_bjsa9yI`eu
zb6aP>%*1q)5(M}_!4$S2!HoTuW;9Lvy^SzGRPwbkK{O6mM$VcRF+4OhpBGm+@|zJA
z$p&jtxAxn3Xh^h#uD%~ef3Os
zB<|awDKCW=f9@uo`UjL@TH$y*
z-x3$kh6Z6DYN1t#B#rp#X
zojifw%Q@$WeJzcx(g}w;^Wdi-h?(yY9-oF1tB3O7Ge0lc5~QAXDEi&vp-Sn%imo)M
zIC=tPfjJRyPqeVDfBKwd6yy;tNl|W-rHX_kRbxckZn)7y(omzq4|VidrUz)Hpp2vZ
z{2Gv?qFnNgVLsSyJ(u~98gw%`WQw&51V=Q&nFvHI10wBBtdNST4~yY?Y&jGtCeYH#
zJ)F-AHnx1Jyu9i^W~)#At724l&7aoEXL^q-2A14#{e+9~f7TVG???2ggnV(%7xjWo
zR~@}(%~;v77xJgXo=7W#T*1i-JA?h$Ijj}5DvPmTkRZf4H#xH>hZ&&0dC1la)wUEs
zFPCzx=&hoDr!>O^BD@xOUzmsQi>5WRp4Oi(YecasB_$qDG1f%W;C_#$Z63YrY`%^v
z>Ygzxhxt_Me`d+xga={k#R@oVe1>3YUlf)ZT_UY;TMD!%3fF;2r#exUf}m{3_w&m>
z(gvBjGMz3(&4UT1Bnf!_th&wQ*=D$W9RO{H?}UFzu?S;N);!h1g}Vrqk3Bk0_*@os
zHb@QuEY$d(QK67?v7~ZoiCP`zQob^FR^;+)&b-Zbf5V>6U?cC))!k!YfQF$C)F`6T
zjwv#?&!oi3JsE0xR}{FQG(%z+CO{~By2I6&P?JGV)1ttPs4i$Hp9i1Wwu;1D~qA@u|AmKKQCa)4z3!OKu
zj_H-#f6(=S`H)btfo^M39HVWf(v}|VMS6-_HO=>Z!sV;}XbhXnIm8K|UB+^vy78|)8Xi_pdhx|cbmOZ9
zBD7D+kVu!j$njObVO6iX-MZ+EY=}VS-Q}Wwf5oM^(Jl@Z>|-}w`hTiKQqP-3SxAh?
zW&MWFsBi@xoowjh!k!;9RI#U;0ve!~%o|~avkkkLpa=~^I}bfIlzvxZ=bGJF4Q4U1{5Z+Zz_M_Ykx*d
ze}PqR->$mX4Uvx2w5W7oXpp-F(r;2kcElrblRwl=!1elS%^PG``6-8qqY#gB<
z*!~QwT!Q7y!!n2@TC2Of_$@jxH6x%D4S6~a9$#pi1u<=TrH*!%)*pF?@E&|3IUB`H
zHRdlbnvqhM1%mvY=4TX@uRM(5Z5`$_f0eOp1?OeUaLw-`?pm$$traP1La5JQ+wvH~
z+8d;cYZ}#i>FIS0L1R^|;xRsspjvPj|E_4!HE1oe$KluI?#z{N+sWrt(G%a}jiC-u
z`T%YgmWjLW4ko%!yrUbd1W$hn`N%tG
zy({M)ZJ09*5pE4m9kgh7uc?-)e{DII>fqerlaKPN!;8xt
zBbI_^ft$=dK5IM(F~P(Tb~A2pJo)Oe(U|RlC%%%WVfxKk1i{8x4dHy4CpQfFCLcRx
zNz8=2jV?a2PhvKN>(Mbq)(v13>NwtPL
z{0$2`mtwcq+mt_*_;82qe^dIz5eb#EXvc@=UMK@hgU|Xg$m7fSGt0?b1xnFh&@9I-
z`cf_Q$U9iPzcd0nSiM>g&K`sFaI~gpvr&&BPVQHkOoAc(K`Wc1U-Uy`y9?avq(=sCuf9q8(uWzW+e&}zv
zlnU=}Eh>ekaS4QV$^0o)#m|`9WZ~&J))d9wM}}MLWrdhhr35lLB);#qzQbzfVw1^G
zWKvzhX{@Ik5x-yeDphGo)_lCuUijOqCVh>50`eRkRV
zu`F0`sV>9vV9E?4e>?jS5d$!@nCjp(X3&NdluWj0`Q%2L0ds#E3%U1uihq11+&{h<
z*TsflHqgOZ`t)=l6mQYb@KyCf4c1C=WWpbOG^R(PiWIqNKVj_+omP54jW-A^=E7g;*O$QEVay6?lk-z=BV~&gzHGpf2^?qNFr9t&|GZ3<9owU
z4$rUlw`!Y>=Et~r#-lPZt88Y7HXda^3SB5c`(>Q&$w51s*Fvq>b0(y8kJE5pjj2!tUOs`pg0jHioB-wdnBW(=xMnk9H$Ml1>Oibo+aR4x
z!xjkPBQgVae+uE<%j4INJ%cnbkiMM>KPpR_bsA=ej+=I@ww*c$`#im0E1WUT7-SUR
z#b|Iex+fi7MmkL7`&|$u`+?9$+!4gTKLylZrmmh{Hjig98C19urD+%eJk$z#&QeSL~x=;>z3W9sAe}{DNh4l^{v1cMjnHheRYd-qh
z1Z_FhJ<)L4!f_vSONSom?;Gx$ExM^cCEw@
zBvs`&Xv=lh=jy*;Q6E;gcxpRLSmPdNR=rhR!|
za7M&}IIFsE;Y?OUq(EJmzerc{T+wPYauoNPT;&eEJvand3L}Dpb@8PywKH%Rv9$eF
z26dR;PY%5~J2M{O{)^WnZ}wfLRP^(>^}43b^Ko+JZ0PJx$hznhdN28i;K>GTZ0}>jT8Od+lll*CmhS1;KivpwEOx6V^XU
zjuwS=e+MQ|b2gzNed|oX>aQY@ck^c21oF?pkPRst47EmR!$_DCo+Dd@ro3j`O`Q`L
ze@CYg5+M7l(M)-$VLZ(~U%I8-bIpIjX(8XcB7x*wwyc$u&%qXWYWCSHhkb?2aiw8@
z0-K}7_^pQNb1qXS@Ls3Ah|z{P(MHGCfWZ*;@Tmn6;-FEM<^!=zGNULZF+s(CdC|
zsA}X;VTEgD4wk+yGL-#9b~;R=+7K#f)4v
zr|#UD5%VmVspI`Bvwvpzddq$0JC((IpM|9DZQ(fQsHTbGC#)LN*xM0Wb2~J;f3h{v
zt8V$#r7NoC;2Z(QD3rFlG@{NDvM|DU_Z|*;eH^*k6CNE-u#TBTyFqCwVksLoFC%EQ
z&kZ(kN4HfbUSBTZ*~VA(zM%_cCfcnwRI|BuevXLPGre+LB2M&F+I
z<-vJ%j|WH)Dd`r~0TShU3AjnWI79S(Lm1b$!jUEVf!wI^yYk!YMp$rPap`7o%zPO=
zEo-~k(O6n+I13|lIEaFuEf~$)geK5p*=45vU0be_+JurI1&rKN&^-5M-;0~#$vqSW
zbGUo;^y|E*qf8w=Ir21Xf6N6spm|IwHT&?|){l|L0k4bW0c_OzdZMO%_M%4$A>3OA
zJt{$^JjnzJ=hm7oBrtT3T&5GN%C-eL{5uzUhL5AWCbZEeb3)gzuhaEdGGdB^&k*j@
zvWz$a^E=)H1(H0i&LcA^T=Z7~0Ot(NIe;>xl;GJC>LV799p`yce*mK!b}wdA#9OVc
z)+n7AaJ2?d4c!S4TGPS)e6Hqj
zW3Ys9PkjzC3$qj`f9&E+hliU=pA!ya5q*?k)V^EbO>YBN8I*bK>EsYw**Z~*@M&IV
ziuD~_8DpK1-LHi~f3T(ld$SSHSPujR_YV$|e?{DS2(905030jIhpAbzN`v~~Yh!6?
z9gtC)ksE*S-u=>)GRV>jy|6@=l(XGI(6ayj<9Zaz#my}~e;V(EgLZ?USQ%={1&h`$
zD|4?JVGYO4^-G<2vIN}F;rbbA0eMg$KsyL;q-~vvnuSN49|hK|uKg?DUQypgatryx56&L$W&)4%;#`ZrCwurahax=0b!<
z&J186n%u34e-&<(?&E6Y0ia0=h6%}bLB)K>Cw
z?27JX!yx?L9R0bc#9zf9%TvbbffPzjO*0Jj%QB+=xobuED>5pmk{4Z`kzul1B4&;7
zRe!^>Ov?r1Zv}Y4!XF^18Q<;)VonuQt(eqpyahz&e*-d2!OK9fvFjXK;q&`mbRtp8
zRg5n7^C<{vEK{|sEk1<=a(n@+3=)Qa84x)1XDwd``s@f+aDV5E@*3~tV@RSU`(#Gg
zDW;H})V;;>Q)6G!cxOf2{jB+q=zrsc)`@UCx8+xoc`p@5(N3>2KgNx%9Epa3Aj^L>JT&8Qhr0X5ZH<
zXb4{Big{gXo}gZ{bJZQMgAFH$(0e4OhG?~w>;ssIwt2H
zy5YILPwcEYZa(kOK`nvY(!&H?Kmfg7iNv5O7BJ4L53R%(oi?$xw6ulnzWJC+e=@22
zEMkKk29d=a0T}TI5NO9mf38*3>?b~ynDE?;AUk0f6t-=xd((r_*eZ-Z{PmVkGNJvh?=Mh)|>
zRU;d%%SApR@h~AUhBh@GV)yA&e>ILHOW35WyzcJgR%~yKWvjD#Rgfm$v@a53m!Cmp
zg%VVNoOARK7@hLOE!O-)BWAz$M~^5n6;)4Qd?;^67lXSrZ4)
zow^nmxj!sHWUfOYUnRC|+1z}RHwhE>JXg6C3*-O+S@yhU_?MqB%Q+z9!c_S?0wvtY
zt2o;r+n1QkoZApc!s%Kd@Y!gU)yc8_ycByDnGjT@>mvtAg6A(Xe<{kh2=Rlh7+5Gy
z{CkC=_I!ug+t2$Oe(<1Q{8njyoMV^RsKHxrVT1X%%AQ-WIAU^#^X
zqhq7eVCr%mAP;GRf7KTYt5i?6nuokUPXwgvtOYvfjTo5oZdHA)gl8eqo02!>e=;#)#teN$3PsttYe*)~Tmxta
z4_|N;kfKq1%GFCHbej@PJ%O!Pvvp5N_@AW{_e6i!ttE)D`+&L`jKEy$AS
zHxs|P#swfQ?=7q1k+#05T;x;7=A=maFXNI+K)|H>rHS%_K+u{K?mV`wXU9>fnyAQ_
z+vGjJe*ot|X#|d|O^ugb6VhZR;8l<5v6agoqs?Gg+N3+}~sO)U9dFwTrIu;a`-Am`0j0Z)(OAB=r>@8_*w{h#6dx1nT$gv@-
zI0{l7k7sJUyMC8iG$btihuhE~5Pk~L;-lJZfNYJ6u>0Ynf#lsw6}z5{nqu41F&`#&`yR-B1nJQBEoS|8
zvK#DK2VYaAlC|1_@NG)9QMuhZss;kIQ|;3csCy#aPTEFdQl-YF%*`c
zCq}~WY%AA^8uFbBP{CJ7l>~E`{j6r1;HLFzNot}rp1ioDL8uIR-1?gvMK42zp#52_
zikA@#hCP|vyH}9f5Q5(i=mvdjf8trbnWHPl0jj@GKwfEta9F3)r6Gcg?(Z4-81m$*
zBIaG1d?&SEJpN^8N_!X*g?AsTZxr&R#Y$ja2q|-g1C<)~X_TCT(kjuwwF4J*y73!7
z1DboKpgtq)20QvP&cy@IkDw38L
zA2JuDT-GdpI(qS@t_cqCR9_ZGc}w+BDtlOkNMo6kgjndr}b#
zwN~Z=BLc1lRUtzq=+W?uNNMz=)3b1{w*2S1`OLD5uEsX25wK*re}26WG~N=3g52Om
zm@Za&eTE
z!qK@!uPR&47;)wEfA8c-ZGjp+4Kk(uyXi4ok;k4q(#&YCCVje&$)Ncz?tKp(k}CCz
za0j+QP>P?Iq~ZSB2^_s6tSMtsxkd4ix7jSSfX%S>p2r;c&EO8vQCwbse)981bcKa0
zZSAxPPR%Tro)N)rAcAWiGFR?w(U7F69OBJN^?-N+LzoxLe=)~te=f~c0y}L8ZLF-R
z_^x7KN9AZT*9P;NKVY9x41^9s#(@KoD6!gEPZQJ7(`InSR3z0om-S*o#}+A3E!0%^
zA=pmm`g}W9q!KkAG{+4rBj1G*%JRiDWCz!Dsb}$UrO;2hQtub%VIKPD>5DFSD31}L
zKDr=$X|;`?e`K0cww7w&VD8vW|4OPxOz>XR}Q*g~h8ZkH&Ll
zwF|_tAd^;a
z*QZbJY;W+B179@R
ziO$Tze=?$N0UR7(i^i9VDZ=tcD|LK$3M6F*dqFotFRvhnQUVq*F_Hsx(&25mL!Kcp
z`DabyuI0S>jAh7rCzjqj9=HtMmUt|;C3M7+P((VSJYWtlWDYGCpl2S99Mi=_xDrT>
z73zZ&gWDQR9Sh$kRH$3!D(_1VlB(5^7uzRlf5f5gn{lNwb+9;sNCGJ1?+II?W3=Pi
z?icr>e;D3(KWqyY2rd-gDf=w6_b2zYol>Yv^UHZjutOD7bkYg%9(qe}TqcJNhkkvW
z#_l#c&1k}3tq0fGfR`i|jlaImsh$>Czu!m8kf?;9tc9hR|5{ReNApD|sK&tYtZGHD
zf5~ac$hT;Rv5e@*Rb=3uIcbnvgT|%o4Bfz}$%?|+I%AlkW-Hv~F{QqrAEQ{_=Wt`~
zU2Xwc8ZFt}@iyJKD_@KN%;4=B1s>-wnc_tU|I
z85`9Z+HT=rLG8UVEP}5hHf4?yO1$Mge`AIiESkFt5~_LbY=K3pN;S%_d=}&{X$H=e
z_k@dvds*IhQCRid8SRx@6PtyIctU$K%=np>Apt>~ETBcu24<`p41V@fCqX*%=bepW
z7|bmPU!>ARQwyklRnvCY33D3m_!CN~W?$5L=u?Az@P47Qyl3{T>@wy|!wXA?e{Yil
zT?04^Ef)!&Z$|>D{e`*0#uy_O-~aiM2$ycXmNdk02|O^uuqn8?@-3qkRL{Xj#3S%)
zm}^C%;F(Vqi#lva6xk@j4=9K4{Kk8*Uc5qnQnCC3>|xq-YOpW-rdy@!KiNG@nrbXc
z&G?%Fwq4G;wuU`Mk{2qKn$IoCe{VFK!G(UHV20{4^fL5RPp9`tfumx;GF8q>dA1gb
zfY07m2?m2mtP_BkCy5@FENqi7c4;s?_?EQ*FNPDVTm3EjHr7L&W3S9-Ils(2RkbH{
zN6)XmGpkunG$Pp9+)bqJ_p)IGJOQ@%gVjqtun0yET7K;b-+{&(1EO~Ce;?9$=3#am
zYKtPn!|)00($;;f#LvW4V4o-S6sI+HQ}CP_6Rqpni|;XOTjOXjv{bwKlJgS^%m+PB
zuQDL9e`2w!>
z!yXpI`rG36BlTkJ_oNTJo-W2yu}|@oY}RYRe_zC#x`~aEI|Lq(yOO)H6_hpOX|?El9t)R@%Y=AIz74(+sL7nW
zfGz$-pA^!ST=<};lHBz<%(f1-nX%4^+VQ>4l;A!y#DLbWGhd|GpA&sgjRGTddv`K+
zdDbv1JYTyl{bD827|Np6cnh0}^Mayldb_5OgGA0L!GrEmf3m^8q)Qw`X+>*s{rxlNDr@%LgY#jVL^5AAU`Z3r721}>x5({%feh9kK6F4?
zyb(NZezLU@e`~GHYvG+W^$Mo-bHcTJqWSNu`CZ;hwBgS%>$|B9tzKJszI6*(GUh)$
z)e&EH5k5KI3CiFUf~SvWOvXmk-MPHSphFI55uEijDJyk7XxQQX80g+3;y$stYMn+Y
zZoLUpr+Qd#=i`AL@HwVyN9GEGSGBb9uVYrbpFMB!f8R5FJ^XfstwOigg;)gP7wrY(
zo}GTRRzwsmW3~)u=a@pFB-F;9gz0~3e|S0DB~7|Nv&ENQB8Fid&;`d`E8gm;VDVOS
zEU%hLCLXW!O`IyPLW2OB3aa8LDu=HURs#NI@&dt#_O0@(^nu{=;J6;0y*c!LVIU&Z9~jVE0E2#*^4DQJQdxIe^2a`H7A*+O3jAnBa)i3VM;yqrLG-O
zE0@j2>GLx~YSgzvI$v2<^FTeDc1k{d-q;OHEpN4SBxRNjds)#HAy*Y08YCKi%^e3~X!hJChum->;u5fUw&RJY~o{XXH>02UL2
zn2kPX&H|mO+xSh=(dZji?trUn(gRXle51oLY-rjdGsr_M2o|SP%TOUh5p!CFBPocz
z+ZP8ZxOIJtVASqW5Zt3Vt2!tZ5m4tDf90JgfFsQmco|*$6vnBOU+FOtD}(AUUIEDv
zJk3OWQg3Z~KzQIL%{+O~FEnL0Qo@zRi*sy0Auh#p5sr%S9Rn+zmPpN3Z0aoTw6&LO
z*f`4RK6PC?{obXNP3`sqiV^go_7vR>WUSuL37*MRFR2o;Aw@x-Gx1b&AjLV{f8~XE
zBxsC<7C6dDN`T<8s1>sM#3{DfFP~%b5URsq*Xb?sJlgnBuMZVP)RAw_@JtryJuJJH
zD$k9Onhq{$QF|I_6B`>X>8_rCrDFWHj_Vb#R6je}DrNko+(=j`uk_Wn5Ki$4zh<-v
z8HrKPcf5uP89p=v?|X7`m+bSzqB8D@)%kv86{k9xo@w6+e4bP?{1j%;
zuHnRq`bG${K#U^*<#vFf4PNl&3#mo^lX-vkkR(g!_rxz?!l%i4G$&sTe}xq16;Kpq
zg1cFi#TJ)rhpmtCJIpCFXg^miB1$0x@)=bBvf4Pn?AG7blE=jjrSw+-_&lyAb$!H9gkL=ButS{rZCgZ~aRL2=4^Wj15WWe@hc=GQWb=5%~dyQd1LWNKZGH0**RCPVfK@t)U2VKfR3z
z=ecGg!ngCOLjx7BE&X8^i;DsB+l;aj`>}-PvGN<7Zygc!eeh58t@(_+3R2&vw2AI0
zn_@pxt^0$?ja&Nktv-uOiastBb_wa-`Ui?OiRa%@qb94aoXCa^f8NcELRhKUgH&$a
zqMM>Wd+7I@&;Dq>*|m6_7JA9WO56>`BdcWpQj{;I(w)s9;cIBY_Qj_jCN?Y=%echL
zPw-ago$m6Wf*_SqvB5w!0NJ5bH)r7rU7bKDgf~V`AK048LDd
z*ov1$T_I606w5Q$R@cs^MP0|+igfOkB6zAzlI1oe
zn4?OlUZqy_>1=r05~#(Y!R^P_2z8ocH%B$MEG0Za!IKa)(JQu~BWhClXHz{4V6%!djulGSM~P_Ad5GoBC3^n8pw`?F3N)89C>!?5_&#dWJ7!BeM*10YlO%
zp+=eD;kr5rrueEFu69mwEtMSy_gh=qH^J35e>b|+Ld=Z*x7^4pd_$?P(2@($Nvwhs
z30irouHcpPUJg`0IGmmO0}@f!!*`u`_D$}fswHMk>u??oViA%F4ULy;e(x8_MddVT
zvue&?*|@wXZat^N#mHYgX$)APG8by1-%XLd>*D#^1Zw&MW;jb?B7A&ds4TLYs#MT6N#Oy11-Ik1
zA|_%gKn%Nbf+eXDU_C-<1)y_t@kLH%@B*y8kbvM}#%Z#~i6GLr=kFUl0i*cXDuVj=m
z3v8;sO~*iQe{D#+NpLpcDDtNKd0Tn>jDbGVz-IYlDT*9Z1P($Oo0p9pCYK}>^DU|~
zWhRb_p{{K`BXc!0XN7o(f7ymwzYdK@zk0Sv7@BsN-I-8zn~PeB_=R^}t}0z~{O(XN
zO_n#i+ulFur-PhjXO&p&GGB^0w=gQgC)QXCF~3Y!9Eqf7TCff{L0GawuM$_;UHFxK
z;8Lc3JQ^aGSBEzg+mWY%oFHF*P-ZN!{`9pO=?Hyew
zPP;Pg;SMmg%{#7!?OOU6_^LZDvC9_Sv#2
zx7|7PH5xNIXT5-23p7-)VIeFUnLKTTa>i9ljYRu(yGHLjjW~OcVp`ck5`Mw9zwA>n
zF6_Bp66PDUb~bJdtHTa2iJnLucWuZeyg4~`H;Dp|71d4V3Py#tPYSqx|lyPU-
zoR4o2D<*ZUrW-C;nfs3k4B1Dq3lK>hDg%$_m&UeGtVM~H*I$^_ZYvkRaKKh`hcP8n
z$4?H7rP}I3lD>t#bwEp&hmj)1UEzr@~A@VS=SvW7xw1J!f1s^Ce#tsiV816Vu&|
zh}@(pf6gbc*iGM&Th
zzaP<_-O?(_PsCqtee3iKkzFxAlZAZI+mNJ7)RqVf^j>3-*@%@BtW4$Tcriy#RCdE)
zf19^1s3-s4x2Nu0bIG4DKPJ^8PQ}9k#S&9LQM#3CnPKvEL8^Qk8?MQZk6E={i!t>1
zoynP?e^#kr+W0Q#0mKEO!4C8XrtEBd8WPpbNp@ig#FlhCpK|E~V^Job-_$>GcXK08
z`+Wx!mGvnZ6zNmM%HF%iH}NGC-#fIYf1sNFBsX7Di2|eOs2u7(L&RIS#%f={>lV}rc|0-FxMK~wfKBUq7l*m)3l?XbZpi#Dg=R%27)Zve;REV
zNiHqSC}Y=GdB-#vi>%5G4f)S)XdW!B8^
zXDKbZC1RMg@0$n-vh5*hBFbiK+l}}6g5`g?nW+n+*Yjj;1Na7qbSR*+wmcw$
zlbjS_AvTr$RPf?2A3NTRy3~?GvW6z*123Nk`vuU>nMn#f`eIYgZwsE($=+y$kbhpZ
z^T7-*N({3`Ur+(HOAzb(7@?FXc&h9w1yeIVOBOoNbc?NpOgTHZeVZb#
zYBv5wj0>ErK)1W)ZKOhcN(p@>9YiS^o_G~@yqjvJLwWN(tPb^hamJd$u6)kdA~$LX
zrSa%{-J}N*@HQ(cvaE=FMl2=;nzlx^)UD!=6*^IOsY>3Fs!!FFWeqeAl-N6$C7^E@2~b?p7H%|C@a$h8G)Dg4mgfTm!6NQ
zENhBw{2G~NII$jVFDnkliX(cU9aX
znC8q-Y(eu)|JmBKy5T6gcDf+WyX@*qdo5o=x1*Wst`P<&z)OS#@Ch|k;HbY5mWxwz
zf9su4m7H8Y?i_k#l<%!zzuB{f3)1}5Vr^gL=rd@;o&8(yIyM<+twm$>3
zj3>Ek-GQ{giR>nB0O9vpLUm^?AN2SWz?0P=CUsaKeyUSS>`0?A)PGQb{=42V9jYW<
zMW86UUF?gbFrwQX{)nid1$|a$s1<_j32l8W6%VdYOf-!r0t;tV7xQVcFga39Fz3_I
zy`QPvde&Y{Qn#7#?HXuquj9)S^*gR92JHv=hfIwYdhIK-scGtE5mH7-)X_^feT_~I
z@#@VPLbYz0?CzyX8-KVN5ZC$iEzGDI$+aGHGy4$2tYA$AI$<*LrCG#&Lh~dH0&;nt
zzf7OT0H~96XOhWsqu{8R)u?9x+mK{RIw~~PeMtuhm^%i?v=?NQeiW{bQ7xi9oV{B*~6KYwTu9ORO7lq7|Qx=*jbT+-auH+phf-`U7xAQ!XIK5d1rKO
zTRL3AJJA2KPIDNNA*)9pq(n&-aO?3eaPo#foV+aK<}VM$pu>cri7t7mO7Jrww1sFILAKYrFKyFWV!d^~wK)`!cbsK@=CD}P&73WE7X?b!GZ)(x-#5WmBUz<}!r
z@*$x7UHEdM@L;Z5O?$`E^E(5cuLKh%frt7)+46GiD-&egITfjH}C`)9%>hx`
zRDa|l4e4l6@C^b*NJlc%6_djl6nUq{r{_H2Uv{(HCtQ{8U#~iChh;XkR|oNsJZk$_m?PKK@g~_a6txXW8%NcoOh?_5dUwWq?c3G+{;FXRAj?)b(i;{h5N&do02fC`KGx2v*xdh&R2LpIdEuyQ
z0D0~uxM~bsMC6zmBQWX86WPOg_SC;?n5_S*1rAZwY0kRN8SEMG5I|*avVN{B-Y}>g&1yue9d-JRm@lW7XUj7+fy-C
zx{Z>xmr^9G)m=dnOEcImYDS1}HGk{~8(-pGbDiwn&FzkxsXHYG3T~=tI>JLAQzVT5e~kx@`BtLP^?aEM8n8V~GI6uZV(OL{2;l
z90`N8NYnF^N-C6JT~>&2SL%i2egu9w9>rwf>gJ~mwyRk4)MK7
z|Cy{_j>_R9iLs*s;cI=hmwyX|qhCUlS)A4d)1b39kOkxZDIwD!BDnp##=FmfQmdg)
zss6w)SE>>}3F*=JrO>+UpoHGeLu1>VJ`cP_t5
ze<%|feE9X1@l{%?W}GOrcg}zw`dw~ag1eptISuZ|X~sEc_mf%CvXh2tYqsswAh4@~j71m|;zILyXZwi~JoMKNk_aROP)B8ePr+0S7@O26?7xt4*`#&Gv%Ko}$#^#;a
zVXJJ&l-xjyg8@tGA>QH(L!$5!!MwMWF9!q0NJFp%Y=V*_H?;(!K_VzTp){H~v-P5-vB=7R
zmMtZ>&e|>KV5x{`th8LrjKVh$;%9>ker)W5z`8vu_+*^rs@&7eGi$&S{V5`catUo|
z{Z+uq0Hgp2gMSv(z=;RZ44o4Rd15W_wOkVYk~%`5Cukm}+!qQa9_4euIYvB%*4Pa$
zKW;sn4=F=UvKQN+eNcRU+lquY_?CgEgf@5ku2Lwv=
z9^@$w6By$}I_{70DAb_R-n;?PEgW$HCCP^L8Xnfwt$&EypKr+3aLj`-nL7tuwRR4*
zjo|>!ZtK@srfLyB24Xn0QLHq17fZuwh@xXN8{0}Db9e>N!OX1k&;gWE3|O`cV^+C&
zP%ufD7Lx$_!}p=74G$
zD%b-#WpvZsn$PqA8_?Gw_clKSE;Y0Z0_AM%&wo+Pvhkj)q8h8p#1fX20=bG|sEV{g
zMD8@V*Ab_WNEW5I)fnvRx4WGm8v*R$-|fZU9^
zso6;#^eTpUzO!97Ddn}rm}0P3C?UJxRd&;oz1uQ0h8O7yfzlypB5+lBZJ20oeA*V@
ziPiwYzBZW$v8;^lA6ur`V2?JPWtTq3GJhIQnJ}_Ta+pkZ-0_Q|y}h#w%W~*S`s2zR
zHIkLb(w1?RuhQ8|toGoX-pWXP8Ti`qeMQ-VPw%TdFalxSHT_ZPmMX&@<8O$4
z4E7pVR7A58fQ12a7nTM41HeK#mw&22=>79_cbf7xGiUtW-*^`(aT96I-J~R`;hXIf
z(Kp}Y@EBS%>15w*2QKt*6%-j-L#P-g*tnd*?pyf9^6!AGrkJ}TJDS9M7mg?%@T`~&
z2)wH&N}(~AZsxr5@x&N!{x*G|BWzR*XjC9=l?jd`R2pU<3A`_uy~DR5+kZ4vI+nxG
zH}#`61>+m1jaCpzrT7o+s%pxGhYeH|o-!w5+CNei&sVS^cf6QDZkp_`Wzu>3kxc@=
z;pP5|7q{j>y-xwV+bf({7BuIHrNIlq=`@Jk!gCZpbviY5B{q=KV@MIJQA6aI{7D%G
zA%FY}JeYMGjtr7f-ArgAZ+|9y0vx2k>71cLL`l@GWJS_AIUt-R3ns)1HLQYSJeq{V
z0(#%~+v2T=$}(ZaoctcZL5h~`j)8I399Uo`h8oqj@Hw!bl(lt@>B)zSrvU+f2Bb7?
z&&ftPK_=~QpZi+as}86j^2DO-Dd0AxU6`&yu7OjGJo#}J~aiy>!J$7
zB;K`Vcem{SyTyS~uLZ6%QUGxt*+t=tOS>^6ZR18zI3AYSM}E=EGb%l_FEMyD2&v?3
zSv#UD!G3ezz&m$bZL}Vebl{x@%jhJU6vKN|#Z1E_D?aMk0Y0VG#(+}HTo4X2!41YVvJ8G0yp^i+fZ|$1;
z^LF#T5D<4fKnvLd1~bBBf<8^oTql8lo>6tctx2iIU$@)Vlk6^J@O2>1q~qPALuOFj
z+J#iv3866r$4o3B<2I{!5?Z%tX%+_nmTvqG68U-J>$61TPgcLzv20;%5Ln9ej?*2$
z&RlX)Eg5f(X@BOqJ?WAYOIMRSHT{eu8*gs)ozwK79du3|$_B095j(y~km3+Mu{f)H
zrMMid_T$+42)saU3|Yi_z^CPF4UVSFw`;!BEl)A!#!GrRf89ZoFDe$xuu=$4Kj*g#
zmaiBXh)L%Nq~>mw>zLZeVgBbGc?Dj(BvaJ1UuP1uc7MB#0xR&OY8^8H3C>)i6oHU!
z5m|6as(XbYw`FDd_(kD_%mtJfjl_2wi3INJ%p<|o+M?cU0GI3=#Zv3(w7%Wl@_P>h
zV0oDv@APbnwtDI$Tmj;Rv@EYXz%4@xM+t>`*x9t8A;8w`
zwTqISQMteA<*~u8`PWcZxIhKXDJ|vqdcHNX*ES5q_|RjT_y(NsI>;WFQ}|HW5T5_A
zX;d9u^wj7&ER!nRP-#*9e?73b&mSyLXA5CrcwS7)Y9Kkh6}JudpzhvoDnfR*HEUvvzK2k2ZvF+_X*dx3
z2}l2RGlx`)6l1oV`J9H3q<$l8Eq9X9@1RaESp>a5A2I^sdJ~nwITU$Doy*>#@T;7;
z^?y9E1RdH1o4g#+Fj_mLvl}Vgu5^JZzsX7e*M^|4=IZ13bs9Y7&mP9C2I6BR6Ch?(
zpOwpZvn^^*W1S0u?hhIpE2bOw|9YUi$*GOx3^XMFG@KU|K*cbg&*R>S($I6&+f(nG
zRz?Q3rAlScQ6aBvKlgU+d6vcNpu;yM8-H-wB5|u1kDciu2es`5`j8-nr8St<(5P)Tn2=B(Q_!O_GIy8o5bfMB*)r
z@h#-OlPSm-7#Gv=-ib-|HZXA#gFrJ-pau-Nw2k%T0n6)JAH`EXhskJWlrW;+N`Dev
zZw>dvfO%jidriM7bv8+ovzczAwDSx=5Me>?zo1Oo|J$hBi!4Nk-k!m7D#1Wzr?k;D
zzyreNWo_B`jec979$_?7N-&C-yQgz)5+9Z+-Cmx$mwz`X(cvtS2$7RJ*Fk^Rag
zQNJhfJUFjZhu;UGW_7R
z47UwWem+ui<$dd#CgUH6Vdj?we*rgRVw>m}@KgE5r1619RYRtxmMF9bHnFPvt)kv<
zIn7y-HLxd?#C;b@mYuIVa?hOHnIEtWas3+aNKhx6!~chs8~v@A2?`xE48d?EW)*ad
zVIgv63vfK*8;bUA)iHBzOMjX)8()KjWiY6MD6#a+0EqjkgAwE>%y5cW>YEIhDbh<;
zjc8{r7<~sa~IL3vvxI)=wFWWchR@2XJ8=!uCY;sMF%N7WM;oquTX>gLzuN|M(8
zGB;yfSe9@%#{iG^eQwf$y+~+d!B~6=&#+Q4ZB3TtbU@I|ZMce3;VL*WPpXvi#2oc}
zz`3oQ>kkgvgDvWu8Ui+HwELZXfrwor4ms=qi06R%m51M=18S(5rwz!2fl}#Fz{B>f
zdsn%DfAooSlE(l4uu_|+UrkR$@QSEZRNd{2q*
zX*({5IoZ8eq}F~3QBJd=(%kfuXRRnBY(EAiLhk*{qS4L@UC)BShKas5%+kc<5I^e}
zKz8^RtMhk*#~w0+*Phq9+9vykzevpt%%JO7p3~^B!FVZ0qkne(^}gEI3LS#RXzTMv|*}+)UI4rR~n4tpZu8t-a7w|KT9Sli&IMNpjlYfAapx
z;^a36QxEh;^?!>u$2{_ddV0GvK*|({%DSA+8LYnMlfp-xjl#%zn2`KT%}hW1@!~tH
z^CICD%Xwvek=Af((QP}~OST8Pw>t|E)&!gpytkoj7mfC%FHI?*XBLY0<{udaYti1f
z{xYQU1LCjr-PeuUgTgnjR!)aXjK%;>1Q|5CBK-xNNPk%Qe0k7>lC0yZK0KcC?q#^a
zlxpb@k0SLoJGw8i>lc*~6_%pR4GKgz*eZLbZ=mJ0(Dl@>djxF}@LQrmTG~yOzY%Og
z?$Ev!*C^N6HLX2!awub{B7{ZmW|eafffaurHBV#xIXd0aOauk~lO
zh+qiu@?%Ul2E~*#pB|O)O3lI^eYJc~LiQ!!<1Divw0pwi*p47!srH<-b59>{(8p
zUCb^^9Ah5TN(h?0_p
z9)E^R&7*bJ1D}2~!#h4}yTuSWGaba^qG4qj%jnlbr%ScSUc0NB4Yqp^J=LTuQ|r%@
zbfoi$P=2w+krb8OUS?k00P-wUvL~Ryji{tL+jW!aj$RCP8F4+~m@zZ>hbyn)mXX}w
z3(iRGZ_wcGoM$aD;L?=95N!Xx+0yQpdw;VF&PnmU_NZU;b!tTQmiPU#c+cQph+aC!
zeB8n*?Iz`-SUik+O#%vz>@9%&XQKwxibCGAd+Y)<#g$aG`=g=7`eaWodqHYZOYw{QeRL*p~d4g;O)hA-s%i
zYjeS5!!0nBR-gP@o0Ue~f5lEu34hkN`qn@R#IIOI!^e5Ah)4yCCEg6v#@xt>Bzpqo
z+jX&SfD>C`WCipmSiRE7%q(w2K$KE-%SuL)Ie$M^8pAnX
zZJUbZ`=#m4*!IW48fUBP?I*g)6;l&s=@8@obdFNFp!{G3|3H_~8Kj#*Kh@$Wj@|zr
zd-1a3-xI{POq=~t0sz9`KcG;hGOmqra`WVIIsWm|Juj0A@v~v#sP!jjtzjr{yQ)Ve
z)|g=<07Z%w95f9Gz1{HSS$}u58=KJ^DD$BW$-muSwxvMGHpIKG+tDTlzIAOHdD!_<)SAXHWQoh#eC>Eqy
z$heRbtEtSecu_RkR3B0PFjWBaz)@|lec31tHrAMPQNkhi$pYDNUY%b5SQM!9;lB%h
zHQxOX;3Qfo4s=clQ}dW6ja3R1T%rQX`of?xZ+cW820U=#L1w0Tjb~ul{^egS0_&p8
zN*8M$;tC1}3D(Ku3V+1=0{H(sX%#J$$a+Gl6pp#odk*lcdk%(|DWW@`Bv+#*>7~hf
zXjt*_n9T|1et{k7n6G2vNZ@+b2+;{DTZkm1(_GhC3&8Af_jM*(B7{jmDs52Z>xW4i
z{E78g<|`^E!~y8JzMM}A*W|0F%eXf<6vf*m0>s3r8^?(J5`R%^$Xm#oBf`g2P%}s7
zFUgA5>YFUg4yiYld5O>c#l>>kQSTh(Neso_F?H
zJ75Rl6vkM~eEQdi6=2`R;ZE-MXomO&GGR0$e%CoOQ2f}Z_+=-6<^jsr$r4n%5uCyD
ziLj>1oJm-SOv6|Ox=bR9~yt56+c}n)S+y^c;OiuuSk%tZ|CTO5=_
z-@3m#`-tyno=m-nO-YoeI{JOQ2V$oCM%(2Y%tqE2aeIez}UJd=G
z-)acHzJ-~8C7GF^_!~pTFI)(7RhZDh#~Nqyjg-@F>^cBC*?xAJ9n)<6-68n%cVn})
z-q|{8X{_U$D4r1T&kk40qar@&F4V5Kvyhv~wtrQdz`jH)mc0}~a|0J2pHyn%U_VFA
zD{^2hDvx$rX})@xzV^t3R8g2tpdd{cjpk9=Shi}P6Ge%b2Y`+Glom2(2YS1w-}P_8
z@nRtff|HKy-W3wm*V$#no1j(BPwXevDU-H44TM_Rap&>!QPfvBj(k+e$qZD&v;T2!
zHh(L4FkJAgWzxWBf@)`&)_Dd-Y;v)b%DvPXpy=+hcYe3DW6J?iTwg4Cq?{!IEDr>!
z*kN4pk^Od+JJUo?F4By5Uocq*D!LfJ{lAagcv(dP_Rr*zsq+Z-DHG=ELC+x1o<-D+
z*=HA5)lcINI(W_Q{9IylyR?LT_`+3Vwto!jeBN-D&A5`X-gZr?}X
zWj~N0P^X|>2u6=hEoJOAe9VAkebL3}xBl*ts*Ni!_>wuO;Mx-$Z#9As+|v~^z*)YV
z4f(yH*XCA?!^)UKc~P;tmO)Fc(AI*935D>i{rF!y&*40Y4#T`RtGl_`6stIUg2P;q
zp#R835rkHv{jyJ+dd{iW34fBo^Bu;_{1W}>(QZt7C@E#p^Mf~BP}gFaK$srI<)MZf
z&-Sf{-+6M%y
zuO~v$EP`9o##`ehev5MFEw>%qo^cr~UjS||2JA?+-_Ja}8A^QeoYUS$j0V?W4Ndyk
z#z`XbY|87PKhRvAiGo?qf3(Os8&Wp)GcUB;|3dCQ;>}Wd&sIUNavMeK^AvJB*UT$kOq@7@)(H!)B=f9T-N=l30_`;Qm%!+tga*XB4cA
z!^LVI8ZFAK1h!(&C`MYuScfUZwe*sl?SShKcb>thl~GMVK6idqfl*_{Z;w>6a!HF0
zDKM8|D9;T}kAGd9MmzI#k|-#rn?&#n=MkD`ecHq*S0xW+FlrWNOFp$9<`OdhllAms
zey;<-Nm29IZK8T&OrIyRy>K!1X&B%#l2mW3k*S%dy
z_;}0qX?i3vZ|Krq#N$Cjw$&z`Q=N7T^ch*G+-iO#l(AOz>=swK$Z|2TXF3pQiqHs;
zuh%Y*wSNQ@@~^{#O@4e@Hpur$&{k@3!{*AH&9D|L+U}h4Nk)DX@yj%pVNz>yykH(G
zD$>SXIy+?Rd9R`yw;gS@Cd!&?e-28I@hJpRm~rFnq~6~oRI;iiw@X=J*vv!wwwbyE
zv|~AQy)({XE7lMP7k?D4vgOvRl3}HzaT%i@vAF7{T%pLQ&C5;CddzTh+XmzSqz#V%$V90{))M`we2v8Rai))p
z0`Hf)P0`H~pAeS`$7#RY)+PHQn2*2G89?5JR5-s+NR?mXcgC9ljMyUIQM-B+u0_%$
z!GF(^F=UcyDf+ixSGbtPU;CiE@b2JP>|5;4_qC8)fQ>}ZUPYw7BjF0DK|T`~5qM$X
z9B_o=pcbbg0-EYXY9>7Ny=NM-NbMA(kr6IgqWnOYtSTc$FK&^q6E=vIfO)g})Tr`O
zjI(1i$wrv9o3crJa@@(Z&p_j)PZ3OPet)cMnt?RL=wH%74?&2CnZ}5E{`2q39~qns
zH(SXs_u_U)LR7q`)_0!>xFZd5ivz($-8fa=x2)%N#d9lMMbAgIM-kIivECgr6kR!r
zWyEkqA>9#75aIZ4xkh-V$mqRFm*)89l{}Y+M@FcE-|OGifG|A>I+eUNHwqsro`04P
zhHYmcFVY5(^E_~$9T`yniqg{|zxEy{e9hM=FnX7Mx#~_Ts^|4#ZN`~Oru$iru{i|X
zpdK&+o_x^xcEG2ft&cR^&_UI3MH{W~>jLUpRw)6V^}^|3;5Xa=kMpo0VKgS0U$%{H
zh-K((&M-cRxGtcX3lbw)x;d1?71qi*Wmlj
zy>*daePoss=v3IGtKQ#K07hFc&O$}rOOeQZ`;#9~3T19&b98cLVV64e0}28&GnawK
z0~5EUE&*L^0y8z2fyVFN|5g4$+hz+;VU4jbItsqL5g3_JR
zQVJ*y3L;2~A_(8m_j&Yrj`#iZJ9cdM&R?F_?>evR=DlQK3|4kP+rc%^C>$6n43PnJ
z_4Hskr>i)aE7Bec0UN;`JzQZJ04gjBfl%`Ds$$@OFdPz%QiI{(G5}Np=L8tT-EnX?
zJ2(cwhe%WM0-A6X9E0~e0Cqlr9vlZV@o|Sk0TApbFhFB*U^^HV?}nosktjGnK1CJn
z?t?)(I^j-hTmgemE1hPm2m{(Mdl$49)&&W`P!52$u%0lWkM_bFkpKvd0_@;UFjoYC
zMgS&%a0|fHSl!4N&@{ScYGBMSj4y2L;qH#c{D+IGv5Bds5TK^4Z=w#s&4d6=Q)83U
zUlTYA&)-o9&^N*RPy65lPt*0(O_WV64b-6`r)vP9fF~S-MV_|(7q`oJX25T1_*4W2
z?e?<)0CK|N++{>Wyu7@G9X+r(VKl~3*xmJiCqEM>Bo^>OV_X3I69adJ{~U}5$^k!4
zoD=+4fz#aqbdmOO6c&D(q=EkBbi#uMC3y1%sjS~#}GgsHZ
zKo@X>AyGIu3Wl=B2jXBj4=iB&(}w@T9r%6`gafJ`7|dx8y}w+TzuWvpT?LI_nU$-5
zUjWSO-#Z3Fd0>72n9V%V+RtOn8>?qGn#**pCj@-O?Z
zQpzYtS2zHMiU~umNQnOOn4X?CSG=t8R}=E|ay#`){L6<|ti1~gj>Q6Eke@C%elq`J
zkKc|{?tqAi?sZFZW5NF+rk}xo>L`1(0}|y3h>A-9FboFfLkYnvL{wZH@Pp!o?*RAy
zDIq{a7=^~+Qvi1lTmXPTV<=CzSyBuT(LOc(LehZ9wZDInL8AXy4HUoL
zKbS-D%zv{G!w(AU3d1`65rD6O`vcdO8UEt<_NC*7e^p6m{crZxs8xZ^vhT`q`&tLy;{-Xi!@7w2(Y!wx>
zw;%Y5C;%1}$14jWCIv`}LjwMhY5(hD|9N@f_52q;y;T7?+#7CBIrRo@FB|ThT7A1H
zQ2ohpF)6RK@Omkuf`xWG$yCO0BO9~YqeU*bVs1!naOUM)w62za%+0{$5LCVeZ#aYN
z@oZgE*{eMVL!~8Hpk5#wwYqYDwwbVLXr|s|aB(vi|3~fY$Cd?RqnY*b^&Eg{|C*|F
zW6ME|=+Fu6P70S*as6}-saM`Ow1ytzdck}0JZlr%`A*EwcyiR;0&GBvnZ+FFS%D|z=K!iDCr0Y
zx7eQvFE$%Q@3ws=PeZT-o8MDSIfVM`5_-}SaI;^g>9b$apC0NVUv8qTHZe@z6^Xi-
zek|oRC)=xG{&ukm@bvxoHo!t)ju)+M)D`rQ>z=~JuIa>oAG7y;*egDKXP&RCRFQJ>
z5fJP>>pgYgK+5ydV)*#*mQ2PuggDhibZoDBzI-%U&&f^1!o-w)jad*MO|9qX35VzQ
z5$hz8#xmWQWC1fWdu5%RMC)i!=lZ0&Xyl~JOTIRx{K~d{9%*Ukjh@uk8r|P`Xt_c;
zqmDCk!~{uyv_C$PTTUr?J@M7$V&fR}QMB8*WWjv~b(y7#5`@F6ri8jQVON^#WtY84
zRIaZZ$uMiB%Ycb2KU|B~6IU1kt(oK2(2=u0MjD@7S60ZIuvS^BLKtV_=$}&T$g!9I
zh`+*D-T&Fo#?InZ-96A<0f1#p?c)`jypgCNN(I+{aY!+IDKAqRO>7ky9%*ykT2jm7
z*hcWGnJ+3jg4J-c{+jb{RQ<`f7@`WqZV=VN_&C72h$3+hU))fTu&XAxvt(L+{5(Zd
ztLvLYr(~;2wa(I0pIMJmeT?Z)>khSB62WL@&s3wx_I%8JBAP2rYvfPvTE`8JO5Juv
zv~03}$-;KbbO$HT4r^H;aCIv2?=ViUES*qOmVAfb21R+y1)Ty3eK?g~vvQx!zOXsz
z92EB`HF^j54%L1EO+7~WU*qa*(bfA}`f_unyTT|eX_Ep^FJ$Q#E9!c4W>&&gqXO^MpmtKL1{pB6o)xP|u9Fy)e|*a+N~(O=2Z~
z+w*G(JGHV>5f-v%MTNBaqi=3=YigCn_Zyx#ER5*NfNt9KH;_yObS7JeGsb3guL<+^
zba^d{Mj{EwDj0G|HM28@-?_iJ?D2Voq8;L_9do2phYt%ddj17uwL0_+$GZMH*z%m8=GZA3?Jm)r7}{L&Tal!$6hxpeV;j>Gn4l
zzjJ8APDjlU3L-C%iJ1HzA+9f@rPO>U!*0y5`{XlFSThA@6?fLd$zj7hc<1$)pr#mIg)2TW~
zb#K``%Of&ro6$2Kkhjr)rIjT=ZTxMk&Jb!&+S7>5joL)K7l*=)z3!3vQrmnHJRe=)
zU$D-$@lZ{#(t{3l_&q~^&@ym`o4n=;w<&D~O=r;{!nzORMSpiS-l?st#j1yS_q)l)
zEYm%uoQgM#2LuwnpTMU6=ZV?Uw-
zVTL}$_gPnJ9?zR5=RlCgW_RQ=5opOSo23(U&h
zc!8A{*JM?H?JUL0)(JXLutOtzj$@zEUOSeRNjYcBxg$qouOJu!Ay7Fq6iTQ5vMr`s
z#6VM$JzxScq?A^;Onv{%QwFb_%5HB}g^rzWjv8@;OX{VJ@L@re#8nnp0??`05pZuJ8yiujdJ(&&~_J|K#~XdEyRe
z`ydrY=4MF})}fu^#fIjbD|X&ch37E{`*sqYCk9{I-74=bq|TMiigxXja#5ztDam=I
zVSP>IR!2B_Wf%>~%%0xJ%yLBDi8&jjE}J*8b-sa8XETS1|1J@k*rI)tp5U;pK?t>O
zHLEv&wCmmVTgEmOWK6P_5rm_U*q@)dE<{&j(*IG>8v<7sSizo&Jdv9BrlY+u*>T-k
znZ!&?D%%v`?Y`V_NUj@0k>N4LCIjPWa(O6j;>g>1a7JO#mX!ers3LV(aG4{i#
zs_~M_@>9|)hm=LZ)E}H}lT#xzoD%px%_y#a`}({!Pv22aLc@IXxRJ(uoLdnd>>R*5IrMEVjF!v`%{!HpYMpDw^u
zKXdZA;*xJf7e=GXW43+1z0m)0{*a3)diKJlb`zHHx)?d9cV9M(uxwjiyl04dI_gY+
zsNtVlh9hM2y+q>F5HQL&2wnl^yEl!k5g&ZxBNQtbrodt^b|wZsh6WdiZ(zi@hPby;HZ
zK2*F_wJ}BPa%{vG1Mgz#(A3D#nfvjXm%*kS`f=`hhEy9*$!O&qI=D_n&5_2|rhW6b)
z^YhjsaF%yh4J@{Es3~nC#%)C=63Aoc`>Zo|3mlkcny22K>0C6eU^INrQQeXUl(YKz
zz_JD`kA13t+`4LY_Zds145^MzmyMtMlD&UsZY-8cAlm=s8PO$*0;y$xK5)H)f}X@G
z$laR0g=%HPYyGmmoUI(;y?g{K=V8Iah9ef@6ajieor3*|$9tiS@q>>zv<%CqSvg8a
z{Kh_Gmgf;Yy@gwin>o%2(R>j*?&Soa;rB@&*%GBC=G~_5<~F{cq78xGcf9i)
zrVT-`9ZXhRo^vs(qzH#oyA|Ikynmu{@k_=H`yFOgyX5a>9{p;&Q9G9c!ZZqrnWwHx
z(9(%=r`TQbOcf(}7W$6hxZ!xqFnDa|kvG?})!@k`)h@QUBrcMFAOYMH6aRPHMdZt*
zS6&BLH^d>G+Yv%fGoq(pPVstWGS7IsZ^Gj&2&+F7RYtC>U-rS>tuP8M+pA#{l@G4c
z$PQ#&^M5Ne`^AvB?uMNAhV*EZiJoU%rESBzqObZPv|jvDxxP9woaNCm>MeZc_lG|D
zR<(Z3Is=;7P$G(d6Y|-?kYg+64|UDttY`<%(Gh!Tx-Sdy67fH})WpX2s(d_j!H#x4
zk@!dMxv&rMn#Z<2nr|l9kCTMTM<@pQ`Pw2a#Iq9i-|N=AEWhJLG_9h*mGNd1lu=u#
z$}*Qma9QlGDPonZeriX4LvNbp)7ZkUKPB787H7z-`X4EO`~#>wu>DF+fnasYb$)?u
z0+3YOUdLgcpAWb=oF!styJu`tH!VRe-ReQ?*K%@qWcb66QuPbR`qjy0w@s45={s4R
zqa|f+2Nr^7ycIb#$?9S~{U-0X38_i8NZV@=f7wT{
z7E3HCM>S)*_O|HGhm+2Q&bv3=0DU`1%Yc;>P8}Ipgz-z9)j^!O~aWF{s<)CW%e-3+yg$
zhm5jRB$bA=u3vtxXp6OVTKBr-l$_U)X1u3w-~!pxcjc8Tjm@vKE`PUb?_REyXc-p@
zxP#$;lP7t`0@D=yZ+*9sPDbPZa#(zKeC9*i$_@XPr77pV39|cMx_WhgU+i`gIu)npXxkh~bq@|jP&Z<^$Q9w~
zk&sj}|G6M>%=%aLz#6(&{v#${N4<0u&Z8HV0A}@*5a|v{6Z-3A!z3=x9)8)~F&gBJ
z+TptAzn;+U|K^tcPL?>yY}ygB)O0L|&78rO(@?ynI+=!m+v0V?k;HHoXk4FFQ@4MA
z^U0-8d1Q6*8eOIW_9?ZZPo1)#p+IJ)6|mC0h}6C_lUaPw+cf&=i{9GhZ>omaqXuHz
zTUl2~zMom+Kj@m2&?aT?_Aa2#LqVT{#sRG?DrtA}
zHi36rDUW~ZadIOAc5gc)ogjn>dj}bRkY%VA*V?BYlju$nGltxo94;C(=uh2W9XU6h
zjD_8~(&hIp%*%I#W>MSCJx79%eCG<8WpzQO%HS5&5qZ}6x1d^j((}4GJGbZ-!(~K!
zlZ?*y@P-oU=0g+YBYPLAcGkekn@TE#n&cI#Hg
zD?sl6RzRu0Y#NDfxG2!B!o38tco;<@8pf&x<|k{*;>V`nd=0kdJa+!@oZ-q2uS1{!
zW-Q2XGkK%*mO=lsl8thC?jv?KvULI72UDgA+uE$fq8X7ni=|0K*%-4*`|hRraa+1-
zbnKMV(2K)mh5gCovllmC2A-4~(zp6_e=BkajGk2GklT@76wq1hL9<6m<<{gWzTw?h
z`D)>wHBuL_1Iqa*#1ES6Suear%CQ@pYp0~mp0=s$B3i)B4yCX?Yw)(UbSP86Ge7(8
zceXaDB>VJH+O36%$C_;g;@|E}3;-Z{o~)a)C7|JNvo@1G#|7Z|eUlEtw&BSgf1{!*
zx-Fi{lZF6#okFIe2gkTB$s|5|X3av{Ws$Grh!+jluu=JY+4+$JmC7qLu}1e-wqiG*
zM#fSYidx+?7j8hOdWhMP8aPu(`!w=y1QW4`Oly7T&+J;PG9`V<*4uM*^==+zP0gLR
zkpi>y_C@?IHy)uu=XwQXL+n#lu1iAs&h7Ef92R}a)pC(lJ6o-P*k+w#)S9}G)Tl}ty6
zSqdyw+bxpTaNp~ErpTa#rKkfFGlmq%OM{Dt)DvilpOFko
zuj*=hKbwF418onpG7~|fLCju2INMpu;xtP}T_M{(FZToiT1)vDhFaB`R3>$WB9_5)}URrj?#u_h@ZtIMVFriTW6H_?3b;Onk&mfpZjatU-tFz
z&8@t0I8X8V?#&+>WY1EyJc>0vR+e%owdzzsclDB5kORVUYpI8Of9p&R(e32Dy^1SD
z0b0jz2-DbGY6sbQwbT5c-Tb%=&+_+=7fDcBnA1oy>z}RO-j5G>!Xjf7
zys$AxoM04J`5*=In6Ezm)0*U430>QMR2|y{>$Jb_>|ut9$}h5>mASPh$9Jqf;K7
zz@4yT+$wJpf2BtctD`r%HzgmC2vT;w*O;=o8N>jNa~Sg@VBgh&w~wku$Sp5%dfv=B5A
zfd&K5Mt%L_t>Kn<7ZX->xD|42ZF6VTlq|L?C%Eukb6z9XYWUo|Jb2e3#X2u8m$WIM
z)(&%Fe|ueMB+RoSd5eSlkh-zS{1s>UBQyDH0x!PhZQM3%V|=>!*upiI4Js5go9uZz
z_YS|dBoBw~IhUkuaIc4BM@k*{?&mo$2vVdgOOIq+X
zmPOjSmmWgT&cvMij4pI$kf3BrR(~z6jdxBY>+5S;q
z+cACB+=M+83m)QjzH~eT6sT2lBo@SuW))99Sa-LLz)ukVf8YHP1%M1{$q84PjMaA>R^U~OYV;?OKPBnF+Lt!4*$
z2QWx10Efn7vBD^n34aSDap*8aAaOuF0H?*F0rnu11NxCc7J#Ja3ZnpX5CU09oC=Ub
z02`1)atL99I6#TCXxPGRjvAT7M#3QEO@}~bq{IYfhOp?~G!Czax|$lV6R+AB4Oo#V
zzHkuRmky91Dqw}SK?6iM2yxN@B^Uz8AdSTE0$?w|0dxTz?SD<}>;ZGTy^gl_%4npq
zeE^dQv;M|nV(;K+z5^f_5gkkc&}j!??r878`*r{!1i$wVfark4d2^6NUb&5_gOP))
zttk$}I|G0N0znp=&Kvt1w(STq;0H9M)QbiCEe-&bXdDhx4}%E`3PO7ausLX$<&9=C
z7V$aI=xiVeW`Fqt$S(_IfQ!TgKvaZI4h>vtz`Gp4noa>BHpnY7gO`GS2$e_?;^+JW
z3?YKU>&aM32iPF^%NQDoy;y5)YikYok?0Tygh&ttN#u|?0c^l?(TDtjRK+ELAYc-}
zV)16!d=Ihy8S@>wF^rtdjS&`33i|!VNKgPf^e1k9J%2U@hS+pAhrQGh1ia`BkcXbl
zyE8hp=(8~*TAG>KJE&PB+Jn?=V1ykA%?ai#X7l+I?bZQDwO<+Gi
z1T|ZjM<0QX@Wg>xA((&Hw=V<-L1F(T=tYO9Uc5V?1~4&>5Zyljv?Tmpg1Cf#_`E?5
zfCWH*0Dlao&@jBgi@M>taXdFdNO%|%W&&O$1{(~gdx6N0a2T5u2m%~d02m(jbKqA?
z7^e+T=@bqkd1P&b7rR?RUNE4$Mh##92V>gI@77hoBvY^a!eLI_y87>8V>OzpYQ*<|1tpe
zV@rOjHa3QX!_;tq8eS9O0H=)uv^BBezkd``mR_GljU&W=w|T1%fM7645$^7UDf)+f
zQg0kBjxfz@D_MooMZYhV+T~)E#MhnPR<%i*kn>6&G|WHrM^wi4eAwDT&pqPIA*j#=
zb$C7F+lyN%Wvvs`eY*!q5jGK<#7vEvvz^e6(HS;fQ6+cemET!q=elw=Ix=o2-G7z?
z9Ggc>bgOD+PvTn^){LK(cPqKwdtp^j!4urgbu5NtaF>Kk^`@@2>gAk;*>wr=q*|kH
zmB*gBN2Qvr_?dTW)()M&{dU(YIHJY&W7NoioM#{heEaBl{#qmHs@0hLZ$*zPW%Hk{
zeI{*{jCq!D_t0p1K&i{~s@c-i@_!am-7r^}YBLi}p=qle#R?=IgEnjmeAT~osMXKR
zG`zKRg``u0S^W%aD?5JqHrmy{_@p@efFwPK)PAm{C`KS9?eNNA&mEomJ=W6CJ==q~
zTbFL(i)c%zpxu2H@1|
z3S>_YZ{tZiN)%fpJ#An%NYM8V?~My4H9nutwA9Re+Se{=zI^UozJ8X)J(<=vl~d<+
z2@SJ7R_D>%Hx_+(NbsRVf_B6Abw#4mw@CL~F-g;J)0xP>He(W1t6cO@>>M=m{Vv?!g0;
zj*h!lq1)$E;KoqK9oL4C&CkAQeCz%)oauZo)bC2LOjmok);A;8$Y|I$<70dq^$lxM
zDvUn?rJ(`hk3@J^4%L}HJ5_vuEu=9{Kxr7dUWeY#wp
zvA5s$vyN}HlIlx&^EPF@=$g-VMk)(2e1Z3b4$qpapIN`Gm2%Asqj2o$8+z)?`<)bG
zf@hCn(~}abGK20ZDLG2zs>yUA;Vb2azMEX86fJWOn9z>iy?@h_a$ye5dJ!3gF}^+^
z%x}{AO2Pf2_*3O!aoG(EGOW|-pXyfX$5rU5m*^!<<%ac+#E1&EU;G${IdLjNNrzCL
zb?^$Glju>;kDl5o(h-LJrdm016G6iim4=RYK*^l=%;@;)*iyP+TODD}iFMQIw!-s$
zxfgKPrIHQY8h_35F^9>c51zdZ4Ll;{Dfam5!V$2cfTno+MLhjOq&!sfMZ7_1(Cez|
zar`z#Mbxo|DcK}$rW$GG32yGSq)zIpL(c1B$QpOsE~7DSFK4#Is#AuvPbi-Mvkjc+
zJ;+%n(el^DH3_Gw**&{9CQ0a7TFQIl^A>Ts?;=y*Io1(2B){)o(NmBmC?8Y2{xRJDsjjqRd8)K;
z;HZpF9@(mFp;m?+Vm%oS6;y(0In|pPr^_^}0YP1UD|z`h@AdhkE4);7ii?}Rna>Qa
z(JDIDc7I1GuAkMubk)g%;<+`M}S
zIe%iyGFNTFNtfQV?T#Xscf=9ih(!=syRK~uG|wG%Bu|Js<=P3=l6&YUa@q}Qhy)n+
zS^3(x;eAphw&g2d?cBOm1$yH#M(kV5Ro+)zI(c$|5`H29eJ<#!W17?CtiU_1%iMYS
z`?L1p!{heiacdaRzD8%L|27*APmZ_Piw%VzyRQsRynT7PSL
z&X`=`;NZv0SVcu+ai@Ii^6?@<_4$p3rA58+;rul(I%x~s8n^xJg>`LpwomU0lz*nS
z&@9-WViyD>x>jX2cyuR*XEt|v;!0F?a!yT7>*D7<<8~S|{+Mw%__4s>Ab3K%=go4@
zx!U;6Tffd2ysVVIt7q@u?f9s}oDWs_DNZW%%e8VpjE?WL#*Q}q`gCvC`$2eF=9k8{
z88e^%_#aT#Gl)2B{@Q<{HgCaDz<;O;FMO3dJ=uL>(5TNOR{l^)Fy+$moi!F4G&)H0
zU+K15}kacO50ikNY+LSpl20ga4}Nxj?lIFHO29=d
z%YT+bZhlaL(i5^#g!YF$R4faV*P%~7kg6Mcx+3Fibbw~Rw@t4F0dMAW&%M|A!bRL_
z=PT<5hBcpXUh@q3(=Z-#%P+sr61|#8_QSZ!D`fPAWXBmGIDVF&;PW}LO
zFn2ohv|v4GcQB$M*G{xvYHOZRK$akg>F+kZ+$v68~QE~3LFw3_|J!3NRGrXsquXJTK-kChl5o7Gt37Oy40
zs%9%WHYxJk-tAd~3P!uMY%hw>1N`JNozROOKw9DQ+t@3~Z2YP1q>OjgC8e
ze1zGaG^W?nwP(Xf`0;3@cSy4CxT~m1(r!Vo!*BMwH;*N)JSC{rh;q1$C);RNPV*I%
zY|KsbajV!0&lh$}DBqC~oPK%!vbf;WvB&2xG;J~tFPCuv-DXyfm2+on!(>v9jK9JD
zxi|g)0H&6JXqPec0}}!?F_(eI0~43s*#Qs(G%z%m@Yw+t128Z)m(eZ(D1W+jSQKiv
zK1@hUcL)R0-Hmj2gX9214nuc`gmg%ErwD?ygh;1!NOzZXeB<8dyzc#;-@o6*Mck|I
z^{lm?-~t*-HBA_x=Knze9Ne4$SXB#VP*WWR?SXh2X`t2>j3Qz!=+c|i+*x3Mp
z5DS0;t1>G<#lZtAv;ojNKmcYSYarMX;9v>R0_g#?HKjE)0kRsZ+J9=A46M+=nr@Dc
z4$l9Ri=?KOwk$J1N?b)t8UWH^2FPk_YW@DI1%g2Hw_*mUXhHSAd7y^B-Ib-a#I^L*
zq&e7ruK~aTa0fZN*!-sbuiR*%nF0Q!26eS`cCi1806=f;>gp)S#^&MS!D{8^;>zma
zY{lvb{wqH%Ya17Uhkt{!9RT{>83YFXRTwvj1+<*5)}X%w{O$@s$;KQ6aRL2yl5zN3
zX%8(W)Cnqg{ZC@hLb(2p3I5w0-~s~ui^dx0@|UlYnwk>89%ut`1wnuibEu&!(ACWa
zVEUH~`U$e2`#V7pK+?_G`8S91|Eiq-L-W6>OE^GRW(fB21Alt_cgKJbHy7`J)aKvI
zHg|xy*tod5{2dVlu(Sb#erNCUyJt3#zhuhdD)KVYnp!MM(7}VSC_6x_17UUbbp6Zv
zcRX<^B|!i`J1>AkfE&OL9a3qCg`|VMJv3_<)ZgWkvVqpg)xp_|?f(w89mK%{;`9Gd
zSlU1=EPvPD!hg+?O&en4$*&C72w$8RyT
z5Pm+64vqjzAlL=uXJZM1ena(f0lI?#uFh^CKc9ap{*6#M_y86*=C06Dhi(_tzoN@S
zEFAy>e~Y1+{IBYNhJgOBt;zu1rxp$nuou7rWQoeA;(y=@9SQpXe|Fk`aml!W!74y|
z5dHtD>3zuiF4Li|_8(CYur
z7r>?|rGKw1qrmik$L23PX^6Rlg$=|Cz{$f406IGZy-?Yqv%ty21MuO14z>lz^RHn6
zu(3iMT%j%iM>khLfTe>o>hFf~@&MSxe~bP`yZ|p{(y$}AQfQB~z58{CaGj{+(=j4AS+}yt<_V#}WIoP4=wfF~wQiA?OfGz{{F9r^t
z-+u=77C;y4KMngWa&iNL|0CgoCJ3DjFwp)Vk)fru`~x|msao3nVdDA?-TzQ?a6uEZ
z{1cMrx4VOz^FIPYU9A2Ap=eGpJbppyMX^WG8`OGrauI~hsxF38GrPT
z$%EF!)x+T*4$v5Ge+C`O`PU1>#oWR9pM`~X&HWz`T6m9t%p5eR=RY8n)$1P++7$0U
zIYT3PgPi});NRz`+V3~SUza2Mp9A%O9#?;1O;=|JJCLr81@vM04-;h|bh~;QutRS)
z4yYXZ@xMPA{fmI+&!zkiZwUzpPk$d4Zs@jP;S_+L!`#q{KnID>?_a*=f4_77x>cbk
z|Nr9Oj|>0^p)^QudIr@x!oMEoMi+}dMA2_%3f|yK)89HRFKF)}
z8AQ6YqmjY93ZN#du#yGPC4U*B5_n{MQzrZXA2Egjn^5HP)h(z%eq53+CuEyOStUjf_YvZzJT7L{fz&AD`!d(|qNQ}k#wpL__&Qm6;`Z_7`t_w%5&3Kp=6+b~G@6A$nZHP8
zS|JJ?m&>!8JfWn86mIuXHnLpN9T$rpOVsnl_{5vl%>=l^^%?5C;^S)=1broAOeE!B
z#_-w|Tz|ct=hMGh);lhZVaLbqItHBLY^)*g(`}PKLzw%tv)BH7Q_G1{{)%fJ%?KAtT
zqgVztIx#2GQ3ORjKD@s{$!hyK^1NoemVevoh4o(FCQ?VM6V-&h_y%@SI+cr$h)+-`
zE+q<2tT&y1pnI(HA&S0M#s
zeLY{&3ii$54MFrovz2-GAMa
zyHVs_;s8|>y%N|$!$9ftLABSI;S@=Z#)jyGjAO3@5_Sj`8ZkNIn6QN#bcbucr|*<*sB53_%0HA5Kgxb`SN4+n>SY6OfNVG
zQXaNlg!o7=Ou2as)^~QEGAX$?PJh03*%g06@pF#%aM79&W+Lh$ho$A
z&w{qx%R_c*eQSq@?>*MN@RN<9Z-z%q?N25xlgs9$7}hxPUP2&J#;2N_jwgDVn$20<_%s7<#6Ct46?jz
zmpJgjLQCfoff3h^e6KRqW4d-8BDTU;;kF1okY}hg3umb*nDt5{aQ-z*7fL+-Cb5Cs
zY*eh<$P?3aG5C;Lv(G+n!qQBQHFM3B8|mCk)7_{R4W1MxMhJ+?-S=sUPPo#Z{(He=M6tbSw7Zt`8znPE-~C3&cc2+l3-H3ubNixTxoc;D|)k3O%^C8)>}k{
z{p6j6Kda?3#Am={TGEVY<|I!T;I_Gre1$w%Ce|L}v4(+nOeoru$4OYI$VDv1?2UwGy-U%R{W8?Yc%gjcbePR1rm?s@_(IXwbFcy&C`~3cj
z6WtAtqe0_7{*3BVH$UOaWA~`yFYPlE>KONSC)C6Pq<^ok;**W41{06tfp_1^)|x!P
zHc#mQ|7KfOFjeY7n9f#C@Lbc}{I}iiiO_8j_{pA;#JVrq+p7Ss_3aCM=agCY+b1@m
zB!UNV52MTTh(1zvpOIktj+`1@WD4EI+gTM~c0vmx^AzF2N+aAM1Q;y4oaqS&n)Kt4
zm0N$9W`9M{O604{_~z;_AJv>#kj1z`GBNoUO)SEq9rk2CDUGz8IWKq@FsYrR&P+V}
zcPj@EKAD8`NT~K9Ep4N#w-qtd39(Af!C~W?M~pHqlDfnzET;w=ky;!5x5
z;8)G3&A>GQ3DSl#r*V2P;~xa?T5&&T(I)tuO@G{7CEkt|MHf>L)*tf1xsjIrj9h;e
z`$+hG=`$ELe~RL<}$1UwP8
zJ8OiD!V2f0WC>22whojss(sGob(Cz0Hb@DDEgT6-wWk!T>b(;+HQwk~z$$N4mngb{
z&~0_-!ro5JZ4#nhKXwSC5U{>i?H!TrQhzJ!N)y9=G=-=*qn%
zO5Z8&kM^TOzsFg$g{>>Urkt}@?k#>e?}M=*r07UNna_YlVsWAo-P;(x{vqu^^MAtS
zgu0A%>1b3fJJD`>qvvY7`}h}=jOwkT@edF6Bj(SbOx2g19vNi!pQ^*iQ=g>Ju-ig`+)
zU~s1=$4ajCcQ){WO?Z(bUpp=%34aoH+sMOBJlB6?O8RX7P4)Tv`=9pr+z1yIDD@t~
z>HE|LB;p0D*`gsPcXk$~C)o;!;|sOMRj#`}jhoSGsgH;L?#Z~+JG-JFiN2{EL`>48
zZv)?!lnfBqL?wvzSfrCPqXoT?N@+ef2ALbps9nBj!TYv1Ii2L5rYm&UtAE(}vh$bn
z4V)7F6$#@>&*L{21^&%68^UjHwKpaAHLD`|@i-|l$kfS0{g#_ppydu2^&0cg&ygF|Zit3LKBEF2}`3GGKnns-_
zXU01eT9Re%U1E&+%hC_2p~o!V|En*;eaw
za%^+>ICL&C8AJK@O0M-b{^RiQlfJu~q6zeyXqVXbQ*oVG^a&MLo!ReXO*;G{B$cOt
zOh+u`X_wou^O8V4zJKW^Z-3DwCOhrbimxoACW9TC>aS>@(t06;Jz2Kjn-2QBop8Rm
ze+~dd+@qFq!cXH_5i|KXM~J?x}Z6vrtu}_;G%_@J=2gXOXy0
zs8tR*Tlt`CX@95u6&5MlOYHTOQN3I|a3@fY(5K-+-Iwi=+=gtC_bO1ub7BAC{2Kfv
zg+2ABrPyG}h0`xN*9D_XdXlr3pS1Csw6djhT8_JQ&d@E}JQwwF5{`NM#L_}fH6|7%
zeu8y=q#9$KU6>RYXFgh+Q>JiSh{v?%h%6o%(c0wnWPgUnuw~eJZdU;{^tR0herVqx
z=J{;jCEhYT+msDW=onr!*}l+j#(72R^~rTMQaU#bGquI6rEM2{f#VqQB)HVIONqZ-
z-$FAXYD8>s@0zMkmqySQ@Z%`4wFG3nQ{Owen&dG%;%sfj(k`-`NFZ+G?&*u8|y?+wVQvcJlMXOO#7W;rZf2Zb!+#6d)
zne@&5Efu}Q-j}uI%&&G^xOJbXQPaF{bxjP~f4IHV-)H;Mq2z_8jl+e(pYDWvs4K*0
zcx{wBCTN=@na2T7a2`fwwcAkySqu;ePUd*blwp13Pa_s#Oda$9J=%*wX
zWQW-yIqgq1w|RvV2<&qFJhw=K56PxB1K^YshNIT|=(`#AO4LOBoU~~J$3
zBYJDBt;@&Ksp&hr^OXwZR;@B-i(=l0BbBuDsl-NIv}z|Q2)RyZ7_xjGIV+Yt<1@7A_9I0M7D?0&6$
z?SOss#6StbLT$vqQ}SGU-?Dx70^ztYGYnuz5wX~Be;sGfYd6%~r#&Z`L#WbSD#XLlcLbITTS
zWPHa#=6X>0wwu$bRhN?1#n|~x{XSgQd(c+diGHO;`g`dn8mUw}inlnzB+;r2TGMuM
z)Xmw`XJ7aUS=w;@)+)813V$k8>!PRaKN5lJ!IZ08KZf|U27}0Gt(wCbPU-^7jE<1m
zz+d*8IfW>v17pl>mps!X%PA;(Jy>Z8&7Ifs$@HduKYT5-2b;3dO-;Bs{vZ}75-vh#MZhnmgL$P$Zs9Qsl5gBQiKWC)eQ&OJuFFV4EhQ2o+`?p%{j&z
zPnqlsxB?Aw1E8`1s+w_>8*7EkhKqIWpD04A*e}>11|v2ihr@BLvCJ#vzzbG&QX|1`#AzR(`^9``ZbAv?!O~>h~QK
z_((z6R#^9iQhy_>twzZVA@Ds_9SRagGWu!8D|_4axG(gSqA&I;BZuu1)cuP-KW;F1
z4$Vhqa3C$n0UthDD9I4$g3#55c>3h(4Hn+s6W&mcTRo$hmc-H?I)2WK2s+|ecbdN_
zM~vaDT9kEovGu$^G~jROlO?X|s(WCmNM)AhEdSlKG=DC9(hL2q%sAt4$3Cna{hMl5
zGm06tyjPO^2w!UFdU|BVi+Scjn{k*56L?4vug5g}AvJ)Mop8!?R{9G{i(angXY;l%
zlC-tlZo08YJh0<7(Q9g&cIWl`%YGjq#OlQ_1xy@yhcnn!$G>aOMYypWR7__m{m5xF
z?LWmSza|Nb}V*)3Bj_G&L^?`#50rG(-)(DTDq37P(6{Ijv^E!r5KexUCK5CTi;Dd
z`N(;iDHocW4a5zew_C$e@t@TU5vpd7=o=M%U7W2#9pp+l0>!gax`vL%t%9M4$&ardLz|W$pw>{^?@5GnlpgaCGxuxE>8PZD~ofdVh;eU7l@G`3q7mEpSdfe?)`YUX?k$Z&Y!A
zwvcxX#EOWjl)XgS4WT3<{9Z3_3M=P`<^6)rH_Gut)dglqbVi5P!OZbZ%1qzNs8(C~)Z1J?36uSPc0tNP&wK3TmYI$jX|Fq|r+uOR!e;u0?s3`FhUCpKt$&RjBO*of2W?moV+p~SZ?44TesQFg(+&cd02
zn`-i6T1jM;nZbwqZZOef#DA>I;O>SrFBFZg55Y=X5oc*~qS1+a*CK3)cj_btoP7O;
za*oyH2gT1py>OjgZ<&vH+c=M^V_d0kim&-$TIlbux?nW%QGJetHwd*R-C45I7FtMr
zxTR&tI1rW^(w(2%bX$e4g1*(?Q=^z>^(kNR
z95romHpW+21<~`Hfc3Sa2q!6%=2eG=J~@^2C|WN?zQIu;B9gq
zRBDYwxCkObPxk)|taCA^oLLzK;0k{>T1pWR?|>{Za>MV~&3DzZ)1INufSI-Q$=*%+
zNIA`SH?p)0=uuc|zVFp^pxEf^b0sK>U<_&UvLXq0evR}qYJXQ1bSc>w-KQ}Z51X+h
z;jG5_BNUd@Cjf+yPc9>W$!wWNX}XX)35ZXaOCs$ib4W>ys2zwC49aC0lIFre686uh
z`W0I$#(Guyshbl|E7-;j7QoKPM+cwpw$QmRT=g++!9^`78}I4O4=OvS4e$EH&_ksx
z!f!dwMN^;c@qgbK5QfCa`$bnuh_l34jXw~&`bCub8mgfz&{HrRc{a!v&GE#Az=}M5
zfJa43;j9i!^j*rbnTeccFtP*eQrP5lGx(o4l8Z%bJ%Ec*>AR6#F>#g0^*_
zVNssP62kcwnHQ50Thmj6#{=0h+e$&qWMPfHKD8442uZYY(iLcI=}Bx9Nd|j1qMR
z43F>rI>Nk>1PR9+{hmae3d2!b@|p#q{So68vaCj`Nj}Sfm(v$`f!ei$I^;}+5ewWO
ziq{4Ey?@pOEMDXQ+Dz>?;1rD71svEE06uPY=G0@mij8UJ%$Z|}A8ve}uJY_KU{$rH
zHtqKMR%_G!?N?TrVb$^?dZN;n(4Ro}+5tTA65QBNNlnvFsU3yKxv&^YNu6;cdGA2}
ziSx0UTx{Y+v@_?0@6{J0yDf1F0z)Ih=XzHj%%yJ9YFh
zcljV9b7rHwp-7KctY&LIoR%6|pY44(5trP;%!v2;$@V~#*99WGc*T+PZBJ(GLVTwY
zp>a?Tt2aagX}2$Tw`(t|s(sPA#&;|>-swwrg&~1?g_xM)qus0k`VcaC{BW#e2KidA
zhJS|7jB(eFA7^3ET%5=v3+jsvRP{7TDJqU7FYYmwv{o^M7qgyurknHBL~xBv*o-fj
z4J?nRZYeDsbdVXgiu(ArikfS2-^xg0u{@$a%vm#S4PsZcXwe%T5DmEQGdZ1+@4Ona
zR3ok5+YsycJm{FK|H#^S7PA};!{Oa~Tz|364I=8}dJzosiX9i5uab&U-Pc@maa#SI
z^dd!=L4-(QAJz*c_(E@Mbc#T;l-64biHIxugaoQUSf_ed@$3ZUmE}9$qZ@tseYk}5
zmjH^QpO-m1jw0Dl><^;YlUjD$(#sSsZ@%N1*Pak&r0jOJ9^CH=QTk_U%&12|`hWR6
zmK2jMO@m_d;m_Pu&m%QOlvqK1Bid#ck2{N1=}WjR4kDCb{h^km0CihZH%{-fW;wV>
zFT_;7e6_r(Z{N+b_DrK1cLFqE`}1Uohy&*SwuUmJX_A>!?u~)lElesQ%g4*1-vT~`IfIhV7i_#)vGkB2+_Gb
z87tUS?8;OTiV}Q&BH7jX?r-8ZX)mTo61sn_r>}NipN;pw8I1_l9e~+o7JujDU|?jK
z4tex?`{NP8*hi2|_q3Rx1J>N%TKuiBPW)EVc@YX*!B=$wM&fX^d@j0%=@+4I=BU@Akd9KYPhq-SULz7BHfJ!)En_ci6nR
zvD+83FnEz>5E9nOX>e?R!hiFz?s$YOyX}53QT8R{f?uz-t`XKkxYTrWtf>a&I
zC<4by-%w=r#QD7#aoG2;ch#Q?=5lRgFB&>cro>U1cFpVjE-&i>ez*d+TruKZW1bze
ziSedw)g|h7jMHc6CDXH=zU${kcBfHG2ZgPG^orfK-6j$S`KW$ue1AX+K2S9uQFEG<
z=>!DvwzJ7I)~pTa@(ZUE;!PJ@V0eJm8RjJ2oZgpnY~A60k!lLHFsjK_sg2lJdEKE(
z-B^=b>%At(fQ*x^PlaD#E>@FDwrvclW3VTo_2BKh%6L#?PHRMB(AjL_*6Eh12t5yN
zxZLu2eqx95d;PN<-hZM;hArm0AwnRwWMGc}+xWc>0XtY%(_2i_1
zzlWtfmoT*0xl{{tLAO6x>Uc*KEd^M-)HC_RA5ysI@`3K{+axD=UtF+H_QYXjw=j*pE
zR9|4e=B}iazW=27I!9%-1?5HK(C2F|(0i`*)Je_o#q~kopL#D5OP8i~CLj0tuQKA5
zM~-#&J)e-RQh%>i!%gW+U&R<6_q?wq@AX12FCjG5FXcXM1k)kHtZK+DlTj|&J=pFk
z)O=`0y37C5L_{Y0{|_>oFT9PAa{NncpT#|9T}oD)zsx|
zEPMG$lYdk;Yon{wqU+oH7xkXC`e}CZT<8uLS;LtEL<20nXlBL7ubXg>%i#b*K)t`L
zFF&5@y-o|QyNsJ%f3ONu&b<01H5zS*YX2VncXEmSM*Iq*n!L>dqb5SRu@9AEQGj;|~Pr?(sPiaA%-R6Y^c;w^fZ>6*c
zG^alwTAsY(OE6EnkKWT>V|gE0eSlqBPS+1eK#2AJq#t#0saHPU;I33j00>xJkJ}vA
zNy%vHl(K)ks^jFB-zsx$t<;%Fwl$3{02#9~#teVe;t}P&T>Yr>Dfk5%)rEiD!PK@A
z&Q+FU6pB{=xG1AZ)-16r76q6;tFkxqH7Ps?r0QG_;cN}+8{6-<7xjlsjW
zrF}8)j=(fH;qXJR5Z3N0G~~wmXtKk8nF!3~DsBSr4%FeW3tU0L(m~nhN5H$1+^a;_
zeaHQiSyEL9$ok&zoT8iT>o{U738bp~_GQ^xM2Aiw5Z1sZg2i}^&_1`zrFW^r1N^1o
za0!23#_H{wHTBQXFN1vAXOxgrgywOm^qZ2}ZD;bT*jOQq1E>xDH)ua{KTPWC#^KL
zO!nz4fTPY4m-1#ggE5^cra^61E{cBvLYe308C-k%2x`7=f@#lm*;adQe6|1hj`ldb
zd0yWIo3b)?Z|`e{VyaP<28dFMmvU@X__@ONyb2(qH-O%IwmbPqGdHL|JgG;9kq_tf
zOYN5h_LNd~0Y0dX>%eQtCjSh*ka=x`m|ng|z*{B@rw)m(Xpp~Ta2N?O{9}I}Iq(%@
zmLY1LR#=r2tisv08H5)QVqey3xItRs%(BSaq9ZXXjx`dkehOR~NJT)LD)qEtgkvG*
zSxO0+ox&j`L1^LU5Bsz>X6c_$r=j;)D3hq{tT_4R7aQc0w7FL?Y$TkPk=2X4cMMt^
z2ehBHxcDa?6DhasFKhcih#Y^9DSM<3SjE$@NP4|ro85q@h$8mK@UJJ)C=Frvfyt&g
z17Q>1c+QbOmH3Nnb#+UghPLwcQ6Xg<<|gD`GPg(EUhcPc&C*#8;-Bk}4a{{Zh<1``7@esLv
zEf5)iWdm)cN^!i*>!!Lh2xOYj(
z%ChxTYMwJhC&O7lXZI_IjYJrQd5|>6_ZL{IrzFuT)GXSy=B9~{&Wz-~YLHzU{jw{X(qw^eQCXgoD2qaJ)HbZ|_BdbCvO6TGE6Z1Bx#WL4>M|QW^k*@4TjQ^`
zjorzth*XJEo;iJvq0yQhzZP~H6@T3l^j6dQu9+s0!=a8(`WUhrN(^577)Jm-H?uxE
z$~)ejwH0MKQP^R4>8q0E%W!^bY!5m(ve}Ke&tKP4UTwu^M1j3-TPMVHGqwdZZw^k8
z;f`^5*5q_x#V3Cn<=0bOY#ad$s)pjN1YHm)4l0)qXebB$_fviJ?j4H!9PVd7^~I8A
z!6SbwC^>3=#pt@m=)9nzh$+}wT=p6#9sW{CmNfJEI-y77H0@%yIs&h)gLS@}u28}B
zdVKU{wTs)@8~0Z+{@DlX9$G>^VT)SSZZG$Bv*DAmyY_#uU{O#NJddaj1DFc8X3W*9
zFx#(fwiL08SFSzKwNFAgRrru{?4*jr=3YL1rjj`WW0L4AyA7Z6epc4J!IJA~dW`-=
zzHWWYNLB^=xxMpqiMbw0kFvkJ1{Z|o$48hB{tSn14>$Dj-EydwXG+KXWXN{w6xyyB6;GMiKy^~Z
zC{Tv`AeT^|T0i
zHV%J8!f_GPBk%@k>FDh~n7X%q>Goy890Ab%EXO`&&h?w%$r*-+Dwa^;_e}tzTp_2
zsp&6vuX}_y*tFll;wZEc+);6(?!2u-oc46(u*}#XkL1n33%MwK&#P}6~i&WD=4O(MzO{rIEa7A
zz(=g56`Z^r-!sgZ%#9@EWhj0XYm#=0LOHzQL5p!}%V7tT`0x;hAIU8F!rB7%Q^hv%
z^^IB$*@GKAQ-6Na%Ti6Ei(r8q`bxxyQ$*u(1Y@%a_CN~mHE?IY`=@c|{g%cPU!BIN
zgEvx*(P|uE;Fl*krLEwDfLiAv;$we?wjUYJ3Y&1s25)U?fNU1^IKj(=E$`s#X0PD*gKDr9bDtG66tuTh^+Vx1rJ?b8>njsHGd4v_&ty542ZZ
zS4)pZudcDNJnz7;dR$C!`G-zc`TDe4@fXr4CCN5GD|u5)A8oZURr`PP#^Vts1{KD~
z`LVf7b%k$=S`|V#;tWOcVeTGjLxP+ks6forXF8Xh`1GQNsD?NEbeAByG{?br<8^5b
zh`|ShjMwkjGs0TM)^r=^&ZW#Vih0ZZwDt~+80c<(It}MbjI}xKeEW*Q^bc;rHtm?TOCZ
zIZE)3z)O`!7lmzm1BDOHz=pC4fRu&Vuz{l%BJrG$tOSw{iw>sK8-X@_?OnDLLlK^%
z+lCd|koz54uz5U3i6?6kz7W^j;rh
zn9|>FX-yQR{VZ0oWqC@SNnS*0&xvVne3#V(VdB=5Z#I{iw1EasxGm}=AdMq$HJ|Tr
zAlfd>ECqQuPf&lT&aO6?yO`cUn-D&kaEu0m^f@;=6lDa5qpN~SI;WCIP~(FAv{Nul
zq6>LaN$JSJ?`z<>ksG#Mdp-$GW>W8@*g#0OULlBBNA{|%F3NyZk5nU#{QFjZ80qFB
z=XYI^Tx?Fkx;=|BiZsJX$CTu)g_##;x*Wx?XGgIxL??d~SoqD1w_*ia=mx&5D
zFs8jV$$BY!T+-#Q`tzn@J*TDmAme)ZU4j)>Y!G5hhemYBHQf2p<)#tlLBN#AEusgC
z?ZES?_NVXxxvld_I^@EddAem$796tbC~N
z%;BEa&*OimrT$o?Mx)5{`7py03`HBF<${=tBOTr5O?X6gWyKVJFC>&%qxYQlos%m@VR>+XMEW9QkGk~+Oy@?16iA%od})57ey
zBQ-v7*f4UZhtpa#TSH%$7KIf`j8z{hKsD*VMk`LX0@jcd_{dGkoh2V0DO@J?mtZRlF@^)u~kx7h<_uF*n6{4A#oC3eiL+k9E?|5X<&TRtv
z-L}b(G8)f3WU0NNx_TU}X?9Bj$7(dTi4T8x^Q+rCxnV0UL056nuEwk4XLk6}xIAWbhH&242dZQ}um`^#YI{OVk_Xf6MNp~SbLm4^>7^_0vjDiPsemZ_bL+l>#~
zas3ILX~6~NAKa`I8D%@IOSy`_tp#*Da&BjU^rC05@gGtg#eOpu5HoVSFnyhp-}r4J9zsu7$u@@u
zeUp*!J9AHGG)_3TYgBp8Na?!9q7MeNCHIx-M?@L*9_7VQ69%6)!lodGRN=_fDUqg`
zz*}Z&fw2(-7YEjU?hy18XYNQlzrwVS7JL=GDCp)|u8@cCWwl&S9igq$?{ii2SdCbqhNkRWD%qqT%p>wWRl1E3uR->l?l{s|?B<
zV{F~ZSyD1PwHzG>DkbvLHizVmG%y8HjDpr^)3!jO-axs
zM;0Y_y*btO0fA!-)B4~GOgw)~H*AKQctL>E=e5LWXM*@mdpyk^Kt=qC^Gt!v`!Yjw
z9Um)I$O3zE&X4^Gd)BtuYIe2lC^m!eWUeBL!Pg_0Z`T6Ez*L8DqZXsN*P}x_JfvJJYOj!
zGbMg)Sjdpc*Qv$DhS`5uijbf$4}m@l>4@8h3h!_Q(Jya}kQH07CIm0SU
zmR!`etqK2VhdQcw(WIPV!rZK!!mP0tU)JA69(1gClyJOR8T4MLt%6u
zV~Y!oNt|z>XLcC=uU@}W?#N+@LX(r|olUr~LE|wS>aYcT8B2e=Tz57I-|&ORa8ieSySGp$7r@M{pQ
zT*tR3!%OKC6b4y^o!IimLgN7%Z4yqTydjz_vv2lu4{_G8d$0(vezWtL9HoD5^|Hh(6LmdH!A}?LPK;b$BlQrYJu%rccX+C`k4ubh
zD$fu`R#>s2g-6;r7xGAZN3zn^bRaCzrj>a3^w7rpki2kTm>m5zLT7g9ej@E6JIs@|
zQzS?A#SDL)p+VbP&1UrzPtdh{wfLxz&laeNp+Vkd?_HEsZ>wJ%aNRfm?&yxL2%g)q
zRPx<4yg;)4sh1RNxnC>hrLx6ImTb)peWRVG-Kp3R32VgAsfCvr5TsE3+)d)*s!@8i
zN$kp;>BH}p{4@FBn|_YdhhMi2g|KFdk@G5SakPI(PisuqA$i637Qv_>o`@6kR8QM9
z1geTRO_s>rpGQVTOP;*>bTsPMr+b$2<3o;Yvy-;8(4BuRH?Y82*lU*z(;bF{)}9z)
z?S^cQ23nvRyxcxt><<071=}vM3W-HXm#NQBz42&Y{8Xy7Kh0z6dM*$^0VZXdbOLzVD>
zJm$io+SJd+|B*oV01P%d94=Cinb;DarTzdQe|KZLb-Ps!M*pCc4N#^My~7q;aOh^g
z*L}Nz%xZGNC!Hi(s#cQfv^o3C6E$mduoZtTO1`^*9mYUn%iUq@7m9biA+UzhRjJqa
zJXGiYGp@4xbn-!Z$wJ3{Pfwe*7N{ndaDd;7LieN^?Y)`1OssW4-E*}+=zd_Vs;<*6
zQ?(z+`HIni8qV_uxq>UB*_CuLBK1OsOXJ02F}Z-C*o}Ajx?Ot3wdbL)(Xs*O6ZU_v
zgP@Yf7WxbV*FMoi?;w@1O$5>A4-yetep6d6-R7eUh{THO@o*2bp@k%ly+GxXmvxJh
z6y1k})Qe*fiE%seIN#nW5lL=&zEkAbsB=p`J=L9`
ze3}*}8G^Yk;;QXYsWqd7H>(=xp`=)=5GOl1M
zljD&Cg
z>Z(h4CDFN(LiBb+<{JHxBO+Gb9|y19`Ro73);R==qVPca*tUOd+qP}pW81cE+qP}nwrykH=AW9yY_rTFoiEj0vHd-M
zta#u4!+vd(Uh1UAWy8D(#Yu}3-lMMFc*Dv<*vAfG#DMFy3-9ZKP!C^8(@^636f|7
zh3m|Ofcoi7%91R8j(4xDVvkW){T)FEdTi^oS1y;2b7&%*hVSh+S8W%pKv`1J3HS!FYD>Wu`LaRu)2
zY%pNSbD(m->dImH=x1on#lrb+1-ff48Q9N-`SK|=EkOa}I!x*cye;V9Es
zwEb3+({VzM#xYVi&qHd<&3nYalcnJT<F2o0HPiq2R-FLbWRCTNf+~^
zgxe%-hXH>I5akx#en)e~1gEkBh`~hNYg1scm~61RS&X-az_0mv@)D9-D_$c27s{-U
zT1C8lL6qZUokj?Aym<+6b(oKK(f(%Nr`xyJl^W)}TPS#yCqFHaKd^K>bGA>Kjvw&t
zVFXde<@;XXnhk54nx`Z;z_X0XxiJvT?|LHHlH7m4;1>`)=S`iHl#U;oPBtdrGnezi
zz^jAdCthoAC=1niK7{D*wJJoAwJQO0^ahyZg{zNA{9YvJm0bHsHhIDYEKOd2WN=!m
zxiq)MrE@*zCLgAj!DpVt;WZEQr*LH;XT>4#2L9%p^RlCGR&J++6pWDP{iz@SmWAILRBy;#dEt()q&jkavKiFDjuFGmA6&
zHUVYns3XM>hSfKuK=|sc@{E0&w^5J5v;1%wGP5-$G2}SBSVu>*2$E*J9G07gpkoKk
zYSx5VjqP2{^pkZ(15+F}dDs>~N`-?f=vaSbuQf|KrIIq?u!$$Y8w4ZJ#jIAcu7Hki
zlrpGGHLRLqiQ_-Elb>NN!4a{h{PVTTuA)EmMaW@HJPe(6*tO|snFh18ITjb6fkFYH
ziGgvE0=s=|?Sztjw)o}NdI@iq%
zGWv<=ieHmKEk+6@H`pp4&I3YM%*`Dwr7Umh;TnKuYc`R%!!6b+QF5Cv=J>R`$Yqb&
z6!=rv`?BP3%pXYY!pS_j%=JI2fZ%_IdLcvmz)3DC-k6$U+nTP+o%D#>q_wX_aiaU^
z0`I~EFN|*|k8BsY<1Qu@P4n*dHG{~Nw?JqnCP+jMv45sMH|-eB@<22_wHw$9Z-AlP
z6>SG=N(5bjROb+5r+uS*Cgw;`lBjcscS3=2LODO5kXoq0ApwuL&EGiO2Tgx{oswN8
zxe2lIMOKCL;!P)X=0~UQ^XY=i!PgVVmxfFU4N>`hle@6Lqbc
z?#t2LO5{i3@;)f`As4N|$9W%0xXGSXW20gjWR}`HR2Pu@;e6stRE?bu+BXj7H$3&;
z2oiALu%s_`xYNia*l4H;%k6(WgJZgOg@Yi(D$9_K6XKUCS7n{0X#8^@hjya7oq~mg
zuc!oq}d|0vpRw0|a6#b)~&hJQ;M`{t*
z7OxhJOl`Gn5flB-9z=>4q8kzL5)U+8`-O4Let?ZZX?OtF=MYKC*E7UQZdy@)I0f&6~)a`8qRxC
zU3c~VC5G@qy2K7mO;FE9!Ifmb@;fs5#FVy&nWSORmkq+jWZbNf5^PPVlB4*!N0vs1
zeI`iBOsUNQB2e5^fJlGLBfg#P#ZA&FAo(G>6R|SkwidQlI70j;r@-=_J%R_uT^g5>
z(U0?4>5WqbN?37}Wb=j52H^_3YV>LG!y9%5Y}4I8Q`S?nxM6Z!8bM9ze{pOmnX&?_
z2x70SjnrGB=T6`s^Z!QqyVBLpCBj3IN@pl3?-93iEjK4FP@aEOdhl^iS+%`~O#j>)
zKxq%ukVlRijPI5^z&@ov1UhFU+HBYEGe|w~){r90=L+a)Ij9U36b=cb9Lz+P7IIW)
z*Wdz)d=pGs*qH3Qpq(*asFSk4~?-$X;=C5Fhr+{mJ`h12=K;3JKMKRXQP-JIEbSCE1;%Rnq-CvX(
zWrd6Y>3bgi_nt55b$}c2{3O&oP7(*IU>cU5LjQt0w7!2Wmyw()_Y-l0BNnA@;X%jtpIfIsn0K3uAqxNK
z78SddMdSwt2IZ>)h(UT0qo|$vrx7a3F5Ns)TdVO%x9(EIW~H|ZE|lff@W6=yDqeA>
zMeOVrSIO89PFKloB4_8#5VJ***~4F^Q`U!D09bzsPG3;?XMpvl)sVIYnhV7
zvm&bbz`m9X;HqR5eY)XFR_$QVnZhrX)ZXFMT|cC=01}{|rGOe@4XHBsqQjVCc$`T$
zx);zN5tzD7*e4CQO^0(p{;5^et>WxRn*`QQF35|)$CCV*Cw{^+@Z4i%v{@#QSRViY
zap-^PIr__2#A5bBXWU;-meBP(F=JN^)}t*;%HJ*B#g9i0yseRT|I)nte!Hdih`iLJ
zum#tfV*G{lc<*fOzWt$5`Uaa8Y450b?d`4=Hj&$v!jv}_18Gew-a>+2c`K+7U#!0!
zYH6gGRsY%+<%UG@@dvyThW!iVa{zBuS(K)hvXA-j}-I}FR?309;tW-K$H
zDMCa`c8nn{H8FLi7$o&9-aYcF52{73kxLy{A|EMCjT$~&J&A2(QWSyOh-JGX5IBE9
zRgLd-ZSSuwKsi2OcSxb3RnL)6WeyvuIN`Ndt4!{Ze-lo{?cT
z$2|i2EZuNNlQuP5*iLc!V$+avDZYuIcyPX3#`J8~z3JRnvwsxIpWxIf_zKli$@oTS@_*CqOy9j8Y=Un&S!
z>uo6)*q$ZmuVt>_9-vctHpRF@Dz3CzXyNRF)(u|y^F7f5YBpE58yONJrAY71#YHs4
z#~h3!<2J?a0$K^W2TDuvA-4qLHC1*OQ;FjJ#3t>!s9K~fw=BemwnUXqjn;pu*E1N;
zh$CoyD=U<>FD8LgzGsJZtW{}~bNE8G(dvOcI!_A1=+_qV^KKG7&d
z*Hh&xvqv92%EQt}^nbg(m!#~pMqY&d5Co04)P?!D-*QwHONB2RXtufeN^(l9g?S4R!RY}f7MIX!YZ^A|7J^X^9*VztyM9!eQ
z?e{kGOGVBAGgB_ptjnXgH$OUdI~w@qDk0*()eolc21aC&Eje$SI7xpP{pv~Mm=ndA
z4wux)we3VFgRoaq9@8c~KA6A2z1ciQP`AEZo065i-R~WJ0nX19hElcR&|p)#Q_Ai<
zC7PdOLqDe@qiigb1Ky;mQl2l!8^)^g6@ys-Sy_xCzRsauJGk+@yr_#mRC2SprkxMA
zTf*NcQ|I#(Moa0=*ARa(IUAp~3rTGHk>d%4?YwrrX8Se)lq-S4m=rcMlJg4w>X&l5Br~sx0AaLm>F2s6sLddPV74
zy+@QW+%QBo7MLUa#$Da0Kq;e=vl@8hl9-rB1ytamsv^*(+e1BWjfJ)WQjBvLcJq)!
z$Ev$`jhQmO~oHgBm(QlCL_RL95s{{^y7IDp$OaJN@F(2
z0EgoiUB&qxQ5q~%S3?wBJ!3G4#Hei@Ob7B8yLNvBI}U+2pr{JgSaqmHufj9~=eQe;
zs|pyZu^FX<=Wdwd
zB62QIhU|g09tbsx)3H$-MjDY|{|;t^#CB@pBSKG3u(H-VEy4CR8I7UkNAn}z}T{0MN}v3C`7EqKs2kdC8b
z+A6eGXn|~)tP|&b>0y*g8&i82;`ux5O9~)c73QWz!#Ue!hfELQQovLm;rns@*uS`3
zDVDKrGf{aj6*nYvsp+m&4_7+tf`5P44~a#zi=C5R5smiRb0?Ms21ux(4UjmSttegm
zrW*UElF&+$o&=P-2dW-HG3Q@Ghq2yIZTaH_SzpOO@!dcrPZNXekOyeFMHtw}cEro(
z<3__quNvu{=7wAzHxl9fnX;rP4>Fl4dunl43?;YM3(urhsR>5#H$zXlGn;>Z^S8pF
zbL#h<`goLGiBd4_$cm)J=s9F(@ya;bBhHTm>rE@e+PmDA0YzHt0sY5s{U73eeT4r|99e*GQWgEI~q@{`+77TAZsr2hece))IvCrACHj6qWJy_?B(hK
zcGQ479fti&Ew?Ze@cUnHNJ
z87Z}7Hym_tc*?iybA55I<0EVpnB-ZAT1P`TEYbuomLI%M4s9GWs*d%eS9+zhT&&2QE!T}wWV
zD^m)7h{Mja{@hL5b(-61I7D=nb&1?fZ`?HSE54BCU)XgL+8*l)rW
z-c+URQi`cvE+e|dp_oFWKwzIJn|};0d1kWjqt8f>Z0wu>q9og3BT=q*8zhR>0~nbv
zgQ%wUL%)9q-wg^2dS19tl*fKJ|AnLO!JVbxd~?69t)s(@%CYM1e2K8g^eMIAgbZ2d
zp!mJa{*FOd7OeUdfUq!|r!xePO;qG=>m>;(w2O8UyAG72!4u;#1RLIsY
zPs46G0uQ0wO(qs7hH9AVo~gbE(I+QH-mhc+uS^^yG(;sG;2zC*N}X>YU65sr*vuQE
zkGFq(_a$-8omEx$^IQT6Y?XLaW5m~#^1`L>4~enlFLa>{F|7DMjH`c3!^Wa^^y#6%XPrQ2+Bb6_rjRFWOo!O5
zzY)G3-pkI-mIpEbC*#X_d~(97vY4g;lMhL2Ey_S~N1E=YA_XhdIzRm#@H)%;&6$c^
z$YMi=HY!D7IC1p`fDjewG&OfuibXM{4S4l^SQo_tWMXY
z+5?n4rb#N-np)4D@x~X?FWnb9!;WXC9GX}EK)l_(Eosjm$xen~Rpe(3woBLNx5POi
zs;KU$zH|9~=fX~LkgZHUr)z)9+G1l(A@Ch%^_pSUh~S!mtjoF|2EV*Fd@JuOhq)mY{1>Rr>}Wfd2uH=?H-MA!dW1#M
z(>e*pqZ5>eBjJX)vw5x`eSrgInLwJdN+ypeEG!_*4@!tk3q9xiQi(NLF}OQ{R1i}ZgI6(87%86W(|R8H5r|F?x+!w%zq+{kqt4umdUhZg&?bx?Yd
zbGP_dAwmzc3!>#bb
z3e6AnfvTIBeoVr?$s%Umknk1Ikz>J+*N(a;IoHoX5yqP*<5GVpja;DXy@-=J&zX{l
z7E%C{Gvdzs48-p8cnC<48bc0Ku)(|T-APmFeI|Ny!!GB~&?pK<`d{8mT@mwm@s({&
zc>G;o5dSjY~NOaq_MP4vhIA
zsaQUJSS}>NJT`AG+zUjYnEgz$c!#YrxY8qjyUmC@PJi82`M=i`HtC_^LpX*&K8;yM!
ziMHe{xQYVe$G!BIm$oq=*rx_(#VV01K&%dX*QK_AMy`K%7J=3&u6H?L4k@Uh8NAzj=J3G`EgCqz#DrY*>9bP-{_@C#|q5Z@TTf=z!Zb$#=AkCbo>zE
zjvbiF6X$EV=4rF0QPbBo#Ll!x`mb|uzv8+(QSPO?k`s{BZs+u>*AvZ;s?2Qn*ZlWGSo0vk#A(AjCi?mq?%`hD*W?(
z`9ob(!P}g;LyYGg9BAFFX1aJs(m<@>{LGyI3pBKuw=Nw{Ju*VT61r5V;<`rNe7w1S
zmePN|phrthz?cJFob$mm!WJe@&Z%bY)lfuvrHYR&vem-F&B(}
z@}mr7Qv5FAk3B8w%>uskODcznO?c->c;x^r{Vg#D1r$&!$%7@1-6o+`?2+L>EAJ);
z=ON?m!-Vr1PrCE`<1BHS#DN|cddtz<k@M8OoKm#c6+&|j<5$BuSdhU9
zPba|c3bHWFC`Y}PMusA5l3_zl47xbrRUapZ*b|eFiCUPJb}ViQZB^{}Z7Mt&GC3IFwuMQVBVE#2vx(tI!B8XEGSk{f^;LyJvVHRh1F<;X
zmwG0OHV>)1lTs?j>y*~Zb&D70K4bhttFHak3ddH40mSEh5*;Fxr}~&ql2Lz!F@`HP
z4}-aawEY=Mgb$W1$V?6eOOkI3W4VphNOmT-Cy>%UE(!da$nesqLuHSja_gEBr(P2p
zwh+F~jTw7FGF+MfR4sv>NNUx5w9CYV4W=cz38E-OUoV&0j+oeOKG^pNT|)}+Zpos}
zCpeB3mgQ%Uy}ng0Z%C}283upHoA117dk1a3q!_>(?c&l_c+mxG(4q;t$-wq${d`n2
zuyk)9*bd}}uJ^LG>6x(MJClU@JY7%f&|7@%e@|^~8}j8+RVBzN5|@w;A*LO>q4P3q
z6r1Q!LBccd&QU4egdeY3ct1|@bz+5xAKOF9b;3ThTzcQubu?1y{1t!kGHtH-D_{Y+
zr3xb&U*&KdX`%Udt;L!c+hJ3s&xOnwBL0xk_|q*he;cu#lENEggiUD@8O7u8`;68W
z)vil$DK&61BNr;_#RHnw4pp;Q_D~jIk?=u4PC3bolHmP0=B@9WyCp;bX~qDEF7z1@
zV&!=)R5w#SeKSHb1$KXx61tY^@}iGx!@~^Pn6#wL(;XG_N&&Zvd44`AV71?>9_>gx
z!spdMwDcT{!wvv-WAlU!jGC|ao9WSZ_f|f4!9Qe69zyzWu1PLDr{8T86r#{Snmt-C
z*}AU9{oaAMa8Sl#$}QDYK(mCkms-Hx>&r}
zI7A~US7U#YEkQLU-WIt+rsl=QI=Dw0>E`cJRHtth^ycc`Ds*sOL_-!c`s>z@b|z=$
zoc50-;{`rnSbVaDORLh8hN#s(7}5h(0`RO`kG7_2(-xrC%L-7gR)Buc=CfSwRC)`5
z7&);wTnLNOUTYYQ_Y9kNWJO+_*Q``(GKqr5Kh(1L
z(|kC8OoBNqF+~qZ=sHc8`pSE~U~<6|Ojs=J7x|JWFmgN-f{6;6!w^LdBq~9qN5}}u
zMkasPJjnN-WZ42Y=XT{jekQ$!S0LLuwfY~#DlJxlOf^O5`<_f0ec+MLXBq^t1+!MA
za`Ae?4o@+d@(Ld$kMaN9Xh(uuUZQ$ZNnT+3dT4A$)s@o8wp)9ZhJlA7gmm5RAB-&yN0JJJy(hwCRDHy-^A
zf#2HD;w~~HztOnZt7Z_D<)7axe3>y~wtJ9*F?Sy?Pn>H}r+@uoCw&~S^UY^FIB?KS
zOQ*Ai&Lb-A!9zVDXr=x-HtCF2hp}gX;Wb%wfn|R+k%5>_amvue#+spg!9F;r5kh}(
z$Fz>*!=3_{!1QFk92E24&ydv~9e0mEtu&kM
zmyX|0?vGIQua6W=kmaoZYCs)ORG@#M+G2oS{e$IN>Ck}3PKX2w!Ub63y0udi=n_8p
z^|(4y}6xgyT_0_`PA|$4FUh77|*;GUA(a%F?w8f~A4
zaGuf=ou7kG=3Fke<6-lAe)ddfMwacdrWB>uJSq9!H91p9*jWea#R7jo<|&G)s{8i)
zPaYSA+Igly7n{35o18#z>7HD6`l}vSx4vrIGoZ4b*>RHipqBs`$W-b^MTM8UnIMFV
z`xwrt+~@bJ%+JZny8&RmkQEd1B*d1u`DNiFVD-6_zS;Ij9syzua@w>zLnv?rnM6A$
zUm@}cV=U6Q^xx#Q2=y6i@(FtI_G-4SHGI^-+F`5n+8UDj3hdLY^gIt+saRe
zA_V>YRC;P$3!xQ84Yw6Q(4Db}L6nyIvn5IkwGe?+pl5M~;IVDmiVkb_`)D|ENjzt}
zw!89`cMMWE?dvoF2WJoNX?PO**vqA;+58^l9Fa0-LE{fvkt^Ht2%6#poSg8*TLGh=
zXdB+DLEP^v7I+Ysq7e>%3X&ks`QeMh+K72a-D(7=na(o$Q~4Y!VWg{b1gGx~5dpGy
z-c7MF_?nF_pJ`tVLxyL(k{#2WlFjrm-s>U$|f=_W8&+-?t4J0>P5A0lxe2|%`DawF-{usD7@6s%WzhfAa4
z(xj(_3D6$pF@(~E15{dJ=scNYq1hF2o~o>Tw;I`RoIMw8y_w8X5i9Ckk!=sMs+PGt
zLpguKcU}0__Z+8x$UyOZHBF~IrM6ZE!IXqc^8V568oj7)Ex(U#U>~8|zjQOcX&FBE
zlR*pyQ8Ma_D(iI+aJKMtNVv!X@*wDW)z~--kt6%4o=%PQJ+B@_kC>iWJGTsqfPfu?wJRF~N94KFdfzDXzI)+~Tf{k^J8&7mHqnIs+2z#J-w{4;
zq@f@k=JhgvfFz1vC(h`xCdaElRTnu(gQ~!7s{65b$6ViZ1Zc>&k`mB?bEN?Morfv|
zwMppXBKK&(K4imRXourt9HGg02g$ph4}euPQ|Pq3s+l5x_of|3AqPFlFzUeH`7~q+
zl_;!F0K{I(XF$XchY@sOFtau5F-P}sTJn0s%>9yoz5B9Dlp~yuP|t`Ie-s{LGn`CF
zeAbUIOj<=6PAur6SmAq!#ULw)7%G{bL}iswMaV!hWu)?d?PsCWZ!-v)HY;?#oAKXm
zM9h(arZBD>KS9-0a_T*yVd_t0>6V`}j_#7YwrJ40shb6S8(UQX6h{pk6i)COm||>9
zl1jLLUq2==WcPH5exW2_Do}yIG;hw--sp-Va}b5!zIh+FGTCk_InKg<_AV1;BM9DI
zO$fPJ0|Q~Locu3uIjVOYBvpD@X*n=7EVG73I8B3R5brTOAa*<=ZKl;~i1ou~F1Y@7
z1VjuY6CEV%MMj#YIS#{(rhn`FdaSbkTP|pS$s@ZUQruhiDK;S6K)|#Cr@mo@81p@R
zpGGLV-+k?1LxzJx^5t$CPF+=O3FCB#QPl|Fq5kN39QHoz;f_1)g9|5x+-TTas!?gy
z{*8^S;O?SAe$d<-n9HlfE9=G>h;Eq_LCvgRq)AusMq?!o4mcEW#5%|(^LNHJxcJW{pwYJU8|9B
z0QvnGrAoqfyOO_V%1FUHz*_CiCn~)+caj+3qZ=RS%DNqC8-PG81D`}GdHEIu?&tY!G1W9W9T%RXd{%OLm0%YuuXE71BZZVoqIr8v@&2%X5DmL0hEYq<(R_r(%8&(|6b+dH^&+thnppkmnXg=9
zW-RyG=!5s@Un~AzN%2^$kd2-B-Z8yv7nl13XfhB2L7K;a6_p`wK$uc5A858Y$7wG9
zX7Ts7-y?e~<3T#GBypixUj0qTc}jCd&*O{y40&c0+vTpU--rvE1XXW`>A>
zwA+kvl(2?W_qUNdFq$v)m2yasXcibrZdk2dXTvCTPFh1kltj2{2zf{%FPBr$=h$2ilHUru+8vvVY+-IxKYIt(be@>+9ix>X>uAOr$*ThmBqr)75pko
zOd3c=7V>-tg&Pzy&}UBTj{_G%%82$#{mhmiMc(NKpD}O$*j4woMx1BGPO@G^00Vj>riblCE
zj00~%a4;Z;$$6q`Q|sWHi$ZbCidRB(XFOPFnhYoSPhyK3c9H3^A2z}#k`kfSf;hl#
zO1o?zl}-l95tCcr9B4Bb{{!dN_d4VtZV8T$NsnHC^^%}>r+w4iquf`in;3g3SPnYq
zmRM1+fn@L(*Rwv$B)9MR{s*<+rHZ0cFwOM~bQ>Cu5w@^)<)}f8SUijX=qZ3+&*y(v
zgdnuX^^iJ46plwxd%x3TJO9Epy+zGSMEd*iBS~A$fVjyq1JFBXEfTI5!dOy1At!GP
z5HB%*q>={&kW(TuiwD8@kq~+UsN)bTwPBH-5s~s#<&M0y1QDfq5#|*i2cxCXp7d3d
znrqqZBi{CWT~-1(WSM{HoV?WUo4V^jwiA
zw6Xniq%^T)#ukPK(+X}A@>+oR5p`hwcyr*3tt0a&j9<%Y3OwWg6%ar50w0w+3GV7S
zO6|6CY>918l|p%(CyCXd;s~k#T*MpmmwS)Jy;MeCfG$6mFlwFC9`hV{`|)dHt+jH0
z({^V$RO}+{S(Q_|0O)~#u3@38Tv15y>=@k2S4KTzTVJrZCO=w0E`kn%(((y9J$F2{
zETM6=bcwgAGTZleZivO{!R)ggq3`Nm$X?BRBOynjETj)M2h8`oT(6NGR?23it>Uy~
zZC>|>^{^cjf0i}s<$AzLsB{{;YrB7c&n;!my4`+3nrQPHWh2{PGEel8`(!aAPTIGp
zh!FZfFD;`|3#wK-i+M|c0or`D)uJ%SXtr1K8m(XUj#X0KnNwy9{a~LBGSYW3HE25?`73hz#tvl1L%Is$04b5$(j3$0Xb0Bl35I{0><4$
zv-^s~Zq*0=q#WQ1nbS9`$g{1qdD%6|&k{~6BPR{xT%SuvHZ^a!XP=0FE5xmVM*+3W
zLNpr@=w&#Yx)h49LFZ@0{CL9T;*$#Nsf64ZRYMTKZypkp%^Y`g*0Vyr6!@$wBT|@h
zu5|6&0dJt+k9Wan)*dc`iO!+mYu-y?lP8~HI0XZE>7{fUuiA|plR$3V^e!1bh~$@J
zB=MzD-Kr9RR=y|bSCW%|!*+GVH_4YfAn2U3N78K|s~yKMLy`FKo4;}mwWl2_O)%IZ
z@EPR5f%(a%`Xnt4>5TT?ks~()%e0w1SLnG=T(kf|ab1}_%{^D4yNLX|?knx1CGr{q
z2gH%LbdYSD?f+>+n@~&21_G!cgTki^^AVbys1xDwr{nq#gzLaGLN>~XdXV}enl^Iy(eG_GYv#OKpOh*E3~q7aqrZtJ{eG
zFH^98!+7=zEzNIf+c&Z`W>8PxJu{cl@z4AGMj3gaRCSJ77vhHwJKG0RG6Pye!?Btd
zO!twD5Mj-)%R3Trg&VlpjB{O~kl9aF^{T*6i4Oo$OwX*>#XMG+2r;`~fEkBwPi7z=
zWQ)D_zqruJ7Spnwvpy1FNXqLL*b`70=?6xCDtW6s=o4D`rnndp9Z^pOXU=O3Ime`>
zc;>8Xg8jeG^5&tIrf~zAI(Mt+4t*%s;bD^5Ri=QaD>MM0K#|RNC#tFyE5OxQ@kyYl
zgeCB0unWj!i=l27xyl2Nq-)Pa+;+v@^C`gqBd;=hMbIn3=ao!aeRzu$_x0t`_@)$p
z{waBaAI+#IM2I0n1nkRWU0S<-utg5;N%Fp06q7w+-|nu}0bk5m?ykgkfJso+JAlK{AVsjoTQwyV=M$A!U-mx?V8RKd%Hs<
z^~?fFArh{pHEJy*evF!o&@RHbOTnCfWeP8&;E6nKioY<&wRW1F0Uk!0cAep|D@U{L
zgG!z|{=f@m-L5VPobNjiGc`^W?P(u?q;@B79E<2m|Ne=n5KtC|I>{
z^7Yi41a&;Xx$>`G9;AD%fv=ztIk>HKnyWD}5}eAS>}%IMblEt~VO%Kn5W-@s*~3Gy
z{eff6-!-JcaMx+&%f;yPzzBkEuE-n&7{3;;HB-6s*yb&SdX%G8gR`XVr39VNRH1ZQOz}9t47Cvf1a+PV{}j`r)0=hH&EW
z&ZWoHVDi2R>lx5q)^tV)hv4J-NwY5U`vsFlMQHLs<%IH*j4PTXb7IdGu`FM@e;^I|
zksnv!G)QiL_%pL|QBD|YMx#K%m$U|#jXCl7ybzp
z?Q4`NRKV|uxrFSmx;j`;Cn<3#A!R^&b1yBo`?M0?eGIyeL~7@A_VbsL7~@smNBikkkZoK_!3Ent+tcM-b5`rWyqQ#q&pmfisi;KiNh+F*X*G5
zHjhggN~;~nllyvqO6o=27{Y$_pgeIK@PtGEOGA&mSb44#I<_6
zSmt{S_qEM|c9k7FO(PGn(470+PVPw0vH%!;RYL3nk
zLJ<5a5uT+1*;BBFCV9x(B41)pFpo$ERD}B0Ym2(60eIwu2WuvyCxJ&w
zPWA}h^dF$#pOkcl`g%p}0iWMZ6%>8xdCB+iFQsn5GiH6>`+u81TDls}*evg<7vNQ+
zbVI{7$L_ub@4Q2t_kpV~syztdzaOmv6yr}}VI}u}I9=gBb_|DJ+_F^9M0-Enjs@mG
z2ZEfX5IfNV%~1Jk_Fd5NX_^W_o+H()_FVIcBJs%#gFSiacQj)6Pho~!EQ$)7?
z$lIwf83rD?h6AV;7z*Cds!we(yMqnuY(U)vMDttz0-5Zqj0C!ktHVOHI>qTji|Bl)
z{utzce79}G=-}dT@VejlR!PlQkyGufjo~1a!$ne4VjRlHPK!uz9WnBF9l3;{DB`_Q
zuRoW|CN(;%uLqsm@NibZY|*2bM5tEy(2kLR=AK}-Lt?zrSbSjN1WhB<#UI%Yh{SC2
z`N4wIhE@sKVupSLh9dlrQlg$pqsfw=vk--U^u7wcgUS)GMk8-zF3#1Uw(8Z&JjrAF
zA6mqM6AH1C@DBu|TeUs#<%EX2o;(?;?@(w&J1HeC$iczsV@L%Ch=w9XlybQF1|MI4
z`1GAJUclvrhziAk8os(YBW@?1)HjgUP2O*~eA;Ei`xFRZJL`FXrvXlQW3A9a>~chZ
zAyg>HdYpx}7F|&k-u7E;7``Xhq87LmkM!C;5pdO!U_~&FYqj#7
zu-4PxGi1;m-Gf)u7GFs`>EU
zDkq=}Qd&cQLy_m1D}z#_ONfeNb4SYMC@##nX(zy~E=@HgG(hxVC_ojT>Ri|9~=RbL;a^_&*>w?dPfNhG_GDZ9wL!
zH{ky8vBh7QkcAt$k3lbvTOt*Al9ERc#qdOgC%`)0L&^3Tx)Q5osqR#;QuG@ayPk8te%>SuO_xcgg3(#f`sJ!N$xFAn%GM1@&@ZJpV66CpU;
z2eWzv&}O+cr|#v^9w44#cdZe&PYdy0assn0fMjM$2=Bt>SzTQhV_>92JoJ>j7Z(c^
zWI`~4t)gBO$nv3McezMi9@1<`9rvZ`PT1&e$+(N#
z%|ILRo*TN*o3~fek4inzwI^-NLKjxnVe-sS3o2Pp&d7
zW`{3dmSz*wYa#ug{?%-QM(SLSmbd%%N5=5s?jL-J$tf{2EcWtP2;;cVHvud&$@jsp
zK5VJ7lND{}jihpa$i_2y8A`Ko)R7sXz!cyQH7M@Ye6T$GAqWB2@-_|%6AFc9#$aeF
z=rs2LSkqm%X}%OF=Jj-ytT&UflQLQy^s-1TMq5~*5ZenplogfnEcTd-#Kcdv2eu;(
ze&j<8fG_i>oihNr>OTN@dvQJgdpYHGi;}jG-=7zeie8C-ivK`@nYYB)SI}L`Y_AnT
z{mXLc7GU+dNHsfP&~-&J__U>iDrNzuoSr>36(;P-aDHKH{W53{@G?~_jtvTa&6?zy
zP2`bOe0YZi0+Y8-wOUucm&<55+Tt4;_{qY<1T=q_rbX|#x*)`Fy-1j&{r3icM(lepFTfMeQwew}
zm+Q{01jea`&o1sD?k!WxvF7GmCLfXRDn40hu4X
zsjUqO#2KuY$~iv|adGQi4~+{t=A>N@xI{PkBus996E#&(Av^p&^^GPsBt*=Tb-lQb
zil30@G>(i4x?prsV7D`Btu5ZWODZNG(qtel4TQzzx(CHzmoy5oawxLDA&+
z{@QeZ@?jZ@TaLvd^QWXA0!I88d_hrbu^RsUe?xlXXWPaqkFV?Vg+Z+wv9rP*u4vus
z-b6o$5%i*OVyKmCdq@8NIxYLCy8%$^fP+T+ADdZQ_J0k5G~=d#1>D_i8z^cQP$qJ*
zc~L+f^X>x7gnB`+K6Ds`3oN$PJIV&db(IHy3gAWhg)e5|;si=pS`s7#6Y#by+zKG5
zw~l&;)!+mde(vLu2r?Izwa%Y^`#b(1yE!KvB8H6=Ip>Xw?b1nu!2zMw8~h%-lVlGp
zl{c)I;bv@hXJoi-^EAlt|4Ya`Ve6Wd04sy_?=W~v#Z~-VWkw$5Y6NdxeK5|?*6&+R
z)Z|ZGRbIQEk^lG`BN@Itdf^8^aB%<{hSuJOk@I4I+G@mC=W3<3NK7$
zZfA68ATv2OIF~`20VsdXTUm1(Hxhp5ub9W(N_lW!TdA$`8QU9svzC+iC_m8BSmK5v
z6_QGv{q_5G10*pbDUW13l`0E^0U8b9>k}B-HnQ=JQB7);R*jO4@}^O-(b6|6HQM4&
zHah7B-Nv{kG$x?iH!jMNj*SnxF|rBD5u4yrV{{WG5jWAI(KdfEVt{KBI!wSE5i?>w
z1wIxNYOO{VlWGrY%xYw6Ea;pz4W?HftQ{@Pfgb%q>c)XSVX$*p9{4(sX8>KmI#n8h
zRMSqd9F2~~Vg1^v#w%98^m|*&^#6Ccx;(iWo65X$CfqITa92&QXQ7
zlBmnkUNAY-hSqN4NNG|otBH(u8Urg6O-_0ZEy1Luc!rk9%cAXSgPk)rEO|vLX(aZd
ztfo;k!eW1f3Eqsp{k9QDjreZxqHDwt=
z?$2kdZnj!Bk?2O^MYo(^E{?k;%!nJFbtjYYz4^!Htt3kXULomsBMe_)9CVe*3g_*7
zuU@~QrM1DwCP>#p7PHIg^xYv!!<62KRzV{MkXnDES_32pkecM%1U(!;s=a9)6mJNr
zH7#jGP1*p@>3nf?F+T2~bUfyV@oLpAW>7_befV+p-O*~i>Y$?3FcOa-f`-KQh;^*U
zi9NT11*wreUPYwn4@(
zsVTLDdzyl%k=FG}n-B<$`y9!e9sgkb?Idr)V9EDMi{#1f(=b{ab!%+;YQR?j!~fb{
zut9@W>=O<=ytR8w8Zp4S9kais(Cm)T>@9zVW`_&S?hynMnKocwUrP`bu(cp+@N1#;
zz^|>R_P+;wZY>}fg;Sp+AZM=%$m!b&$eFXR0$3aVAn45GDGw?3sowA$Mi+2kNv
zq|0qUg!xJVb&$(EpfEk_u9fGLN^(oZ@cOD^czMfwObt)S)bK#Nk5(rrk<@>-yFGs-
zz@9)hmU_iKR|4!=_D0CJ72uBR^}}3mP*(*w-1YiBt{1{Qm5vk5&_3Ka8(RvQ1aA#y@|<<<$wu-YoByI-}%{(omv%E3Khx?4fFNg|_I846W5^
z7_pO(TJV2HPToU&WBH##d|Nb)xH$cAu{eF-yxpuvJm`L#9Ct6iyT{kpZ@u?z0Ln0(
zCcr3SDNT%RqAv^OJ<}y?TZbf1w|d=PkEHf@qqi|>j5P))orddZvxoNrz#bq#_*1@II+?&gaXI%y-dwYUkdCDY-v(1KdyU|JRs{r
zW|6rbDwSo@f#rKd}^4pe20D~*L}7ST!#b~e%HxIpX!eK&*n
z5mj;)L0hVFY@-vCh1$<4gTt}
zVZ&ita|U6fsCXp9rHgZ{El?{IS2vzMx*+NIEG$Ka2eYNGfCPVNv}D3QN)mUFq(0IR
zpS&kau9^CRV>)Hnh{=F_)EnbVY*K`XqmPK7r(R!NOkHkK$&SKd41Fd|Bv!*5wPd78
z^fuB;>D#Ugp4>!M0_aYr^9Wv%W)9@zCGm)~eI1msBtuT*LDPbw
zr_Dx85BT6U>65i*PePPFI}(`;TLx*P50Y7+gbSTN3{4^v
zq2Z|4dDd4TnE0b1I7$mj8;ITmF=(=lZ2g8-7?fdA2ZnzZjICzR8;NCBB!EZCqv!)N
zshuGhy#p;nN%l)nhEFJQ9hI&xS##F5F?`qnrd;fAUC(#Ltgn*QK6o#w4z5rh@cfhw
zl^Donrr@pScw9y0UTjX0arbN%p3IkBTK0!QvBVMyK4!nlW!5(sIwVMHqBM&+vi*fb
zVb&)ObZmdA(UQ;f9Q25RXa7h`AgfI=PsM%~nuhSJ?I*)t4Vh&hvRkgvue}JX5)`X$
za}~~<-f&>Lx!0{dZmqI)=qVx~uPgIjxz!u8t?d6Cp%Su>0bTJtiYSWc=+`rhEc3~2
zJhpJpi4H45;lD{_^k#au);qmkWVPj;z{AYy*qVRqCJS{#Q2@-Xddnl$S=xp=Qy9*q
zOk#dnAlqJuN=ZZ%hl++yQQAwJGXn;hBZi$wo<%wMW1Obk_Z^2N?e-tEG$JyPAUt#
zrzWvd4KC~SaI1_4kt$ZLxz@AWuWZx1{{Js-)7uS{flmGvY`Scq&b)bMGVw^}V(6o1
zlj|9z&nYvXT!xg`q_(Avl!=sJXrW84mkfLGeGKu6o#gn#?3~54Ibo6XSk~7R;*?8k7MOHFc
z=1)1h=WL$3v3IXIr{9)J@p-;HY-XYu7s)s%%6}+b3WN#$T9R!xl*}e_-kdXmz&A^V
zJ%OR0U(!Z`q@#*s$Kl9o=m6?zZ?}IdjW$=x7y{k5R7U|v_W+3?;BQ~BR$pSC@{DlBI>$78M_l49O-U=%sQ1qVtYtS@N$vakOK{$DSlfK2Wo7p-@PtDT)LP
z7@bp9sYl8fHrFqpV9%Q;Rx0Psc@O=TghTC?i=~)^|GV;~Up?;Iw~)^X9Og
zw&0Pk0Awid$;Hi0(Uz9_^Fhq=fh$X9+m}(I8vVIqU=%VqZ19ytE*)?=U~`T1GEq|i{COGO{^5oGr5{eAH!zVZi`0N3Y)fl
zmR*N-90mEwhZ$oH=T5AOdV@l4O@=jYD-|BMq0Rl)(JYon*ew
z2J9eKyd4GKIxSO1w;Whe^pn0h8j-t*dtMPPle0O#_?a+dTqg8kQ6Zy@N
zcvXfW;DgKActKaye-rn`6Y(!`Bz_jJ#2c{`zl)FJ5Anah=QTZ=fRoB!8~UFP4sY`D
zn>YV?{^pzSpS_r$k7s{M-nloQp6uVNjUJm#PysS8EYsAxO}*XH>*3wX?OyG{y*>~R
z`HSgO@x6E^o{JyE3+()*7>oDfSe%GXd=RJNOq~63an{YmMEqC$BBtV8%*0$=hzt7E
zG3`FAa&19Vh>LD9nV*OSpj(Mc@!KBwjJgSYFCX2#`}}AFpT2(%zCiHNkN90+_(BC=
zyrL|27<@7vgzv7n2X!G3AFEkDF?gQ9o-Q@S
ztKRFD8NY(tVtg_=9#4tU@)9DY9|@dHmKW3UAAMoC8)^CXZ(jd=RD_X{`jjM)duGQd
zZTy5|Gy@%@mj|Ws6skRDWnQcC>7pC2x&_pCDyHM}_a}d2G3}PiT9H?i@$_W!;X~~e
z)=IqBGlZL9_Vd4gd~x)!hgrBDGF1>$bny8ihy{HRv6ls_3Gle+Z}Ud+R9(
z{KFLXcGZ7f^2DM$KOw#8e!Uz|MfdR-8b^+l{!Ik``wgCTDn86F7l(q@Y%=Shnsbzg
zqZgsE*4K;a=6WH_Wh{eB3DY>TJ<51>fkZemofAM6O+8d6}PjV0?J7OG&q-m#{(0$lIQ|3X96`amx0Fv6Svn-14U*6
zH8Pih#{(1*I5IK{FHB`_XLM*XATl#CG?(z%0Tlx@Gd7n&n*l0+dpuP6-=|QB(n@mc
zOU5N;7}qANBKKRC5kkzEGaP2l8FS7wE|Dnw!cP<_Md^lYp=^>%q*~iugj}AGbX!?e
zO4f6RZQFW1`{(nT*O@cl_viY3e=bAP^z3(&kVu>P5Cex%#EyYO>jBjQ4+q{54~Mu+hz|%AnhXgzLI}hs
zrsA%>7J7=iQ&lg%hEh7X5@<02oXWU|PU$l8_y
zTo|E|s1S>U0Rv$H7m7Otc%VW8gn=H4045a9;IIM826V`O5BPZ7dwBy#uPr{F-ue{6
zvA2N7L-~K@ve}#N<7f!%Y&_`p0NQE@9DTg$qHj8c5c!800uMSN7sVkIMdt4IbQ`+A
zr#;n3bO%5MJ0Ly=i^6`Fn+}l~_?jAF%I2foxd1>f9LIU^F=Y-
zzlivMhxwAaElS)hkP{cr5Pn}V1|q;>|G~|-*A7L05e&v~Y~CXT*f0kYWsiw!1|xHr
zyN!plgS|JM>_+q+LUu<9b`T04jn8R|{Mp#Ktp^q~Ge9LCQJ>l)tj#EwOJt47i0HF}
z37&A2A7k`y9Uh6GLL}~A#cUX1u|;)Y33x_62plDVobCQHA|RQsSQvx@V*o_~C^|IU
zNECd3u6IN*RRj~H#K-Ya9$+!(h)A>t(yhcR|Q0Ot#!__%)xzCFoMEddJ-#fi2j
zMn`7OyEDQ@0c{>8GWjCb{HRAmDI3
z7+gs2Uzq(~oxz1UG5?prcZ$A{D04jzl+R^csG<=@sdE6Oi5OZOJ
zgmEB1rkYWVO)X63B|f5w;}ES)tS4Bs;6!qZ?_@;RhDIU~h5<9}@aF|D2@^8g;gdz{(!zfF%ejq+Ets>dCY#`$O@7h>Fd<}!`jvsjgBLm&;Xfc
zLC{I1S^!H+^Z0KpL+4l8+|UqJ{BjoU697TcP^e7zV>HxiPsE{LQc4r-Pd+G0N}U>~Kh0;o;@t
zPq!3=1j=dcDD0&VXAdEk!!sj`>TR!G)3$dneoyCy^41rb~X<>c)
z9;;hz2KNGVH
z?!8F3$=y|UL-DS!g5FKnld9HSjlm;+(wA?P+H|!+f}$%oc4*-b3Q@5YG9A`G7Q_aB$)f|ED{VC$`=ZE9
zl({AS^hHO90B@7n9PQ5yWT;3vb|cFW_@zo&sp|H>5UM|&9xHXP#9AV5SpPJZ5pA`0ShD8K0rSD&d|#_i
zH(eIc$yUkR>cU62$S7L$IlT-u%R;w$veC#{{-eF8r^D=$eyA`xtm@3W5t9@6wv^o8
z(O^3ILU;RrXmZj4{Q(V+%E19PT%~4h@yF*4jl5)Xo4M@D#*o%b^}3K}8TY#f%B}h)rs0y+WD5mVC&9|xi7YH|<^Qbv
zc}w4t;OTJxz+=VS4~F;1#|CgtyX%J9MC-1q@galb>SLUdcEgW51?7doO5vnIz6^k9w=MU%JPd
zCRZZR|BX>d-CX;_=Yxhi>exPh_wVoai*J;5T#<4JNX#Z2N&V@e7(FE}uLIIa4=G(|
zch$Z+GoYYcRuAfMIjxXyShGl#v?rAOl)=J(9`<3O!E3ps*{p-hY_ki!jt4BCvi;-n
zl^Gq&1j|3~=$N>cH%zsvn?9dN8)&+fTwq)zp=s{Xej#~;GA*vPNUArHCH?kQ#2?JD
zyCq)lo8LdMdi%*P1q@a~U6K=zvf8|&4vB>{^G0_*Tw_Vj5o?NgJ9K7x@${3LTgsPz
z8@paAHl(+*)u4&;!GZ1b|!tS)Ui;U
z0%dvEwqrcD)`8{6UwbXNWYhAgMlW=KYwoz=iP~}X>a*J_k6vB#B3{hBMe4Tq@&D3B
z(>9d5DuD=$E|%cQ#ENI@dLe@be#?8zkwb
z_Zo~Xl<%DQeIKj#P(pRH{Ld+Ox0Xgv)em>x*_dSc#`I3mvR7_y6~{;J%*c{|t`6y`
zyVYmBXPsFvURSjimX~qXe<(HHkHIfnNd@oX9n3x=dIqmmKP+hcZ6cuQo?o_O^2PCe
znZpO%1NP;OE>2ro)!M5mcQnyS&ejK8m-=|R)T(ZKYMZRei*3UdL3REQ`;+^$^|&1Q
z3#_2o2R1dh#NNbvoBJB#v>x|=ckVnk>1A@xthgoN320ogwpdeDw|Fd-@Y*!=Y>Uv0
zJuZE0SK`SXQ!`RCwB-*jCwY-0dB$t~necBP4n6m}ko(JPF4J^HkNlLp?_wnN*MK*t
zmNOQtdqrwe>bW7=M>kJ-b5wKte`oIb%V
zGLrR_uIB}eQ`>lX(c?4Q7Kb^eO&@HIY5KHY-u*OVPeYziWmG8yI}3N0R9I`OnTqYk23GQ6!u=hwm3->OlIhx&h7!q(1(OA3w&&hF
za|ARAhUDb5Cy#B3^cpv>*=^-~)+6Rh(iYt}(374Otw(%kdWy||e>>%uQzfZsm$|{me+&l+6}Gc
zZVkKr{PMNr{{m+NHPpG6G4uly0yQ+ZfyVE~^E(C~~
zPG>A%#&LX$+h{E)^q&eCmW6!4JyBw;t?dLo>?}mF~fAI6pMe?usxi=i
zr#y_K;m&Ll8{+j~kelf>QRw!y>LWPC5D#HGFgf`E=KCUyA((8IR^q1KNuVjl{me~qzryukGeYUbQ+a1`2LpXvHIE4!c
z;U0!COs3NiMqpr^EK+!Yho||2nLz}9e?bgWn87SUdXj>qxk;mB1Sw>Y7hEk?uob+A
zt=ogc^Ye=(Tly8Y#By#3Us$$+f6bQm*0Qw+#|2wwa1H~wYI6n?q%5kKN?=}bvw$D)
z)S$)p-b2gY#o*x9E}4@se;7PX?~^#A-go;<&Q4*Ex@`yWpR{S)e~&I?o87R>IDxc)WsUh)!BbS`?Y
zW5qY-wB+@a_Ti$L$qFgkpAC~yG@IZ+(Rhs0c{Vf|`I9j)(@~f`5T;-}hH(Vr7#||S
zIhe;`98U2VU>8CGj%)b7NN_b23uqonh}X9Q`nmMURk~AK0hJvJ~)P5+^g<@^M^~<&k~%QR4cA$nC=cQ4IHnPq;gL
ze$hMZNfaqf1~9xA%4RrpEo>Snni%T**&knG>?s!=8*&eWDBg2
zWpfl|^EiC6n-ZkqC>n+_j{e3pDd6sT!u=J5iRBUrIE@W@-9^oXYQk`m1uH7%eo
zrsEik`^+Tr3n`
z+XS}mONI+gx9TbC(5R|Gqnm)%eWG`u`E)%rMr+XcCZP3?4nN&3e@~D5kGYv;DBbHS
z%Vn#w4fg7bvQ=10e_=Ho-`(B}Dv0IzWi7s-g^zZshG+rg?+18uc6Bk>Zva$lre90d
zeBDJL^EZL-)ti$$3%QH1p?99F5hz`No_;^T*RC)wK7Sp)dU?G~pnL7w;i6M6I?C(x
z@ftq834HE#!Nt>(e>+*9Z$XVOFZge_gyudLm-P)nz6IHNI58RC@REi7g{X(^v^a6fYOQ>ZMtG(y
zLXk2eUeby$QGS)`fFs+lZ@(1045Z3Hs+gZ7D96&Jk$U{KfACf(%Xno;?61G|F7_^s
zFF))h!|g$i_k%4`>H+yM$2k9^TSV8J-WJ|o>IUoJ*A2`@s9Vrbo~!WseH=}r+!=4_
z)xuQR`4--JYth6O5%speMg00`atrUOwYI``-9OkO%6HK$M_Rh9hN3ZdT87HZ&|cnt
zv;5G&wINBue{+rKDY$DyZv2)f0x5qAX+cB;4Zjx7mzT&>sj>AQmaG<9WX#vD(b6Vd
zBXTEw*NB*u=b~-kXgGI6wiCz2!H`NNT_aMd+}s#;514C2Dg&2vhL8f?iI{CQeb)d+
zI<;FphIDFomN2AKx6(j5bt?^|Q@bU?ELbgvNl^w4W*@?
z7k`wM-<1$oYlT+J!~RLOM(|b@YdL_b0t9cQZUx4&!nD0AS7N9Iz%3R1s9JIT*bVW-
zo=ke>D(>>KSC!?mfA1Ad`}QOm8Mw;K*&>glnca%c=4Q523`d@S%KksgoO`*KG4uly
z0yj06M)U(Ae>XN=3NKS>dSxInH#Q0{S0Gz4ATuB_T?#K!Z*O!UF*h?HFd$M2FG+4@
zZy+%@G$1e_QVK6gL?Bx{Fg8R%F-AEtFf&6$GB-sH#tE;IWa^&4qXZ_Ol59obZ8(nH8CKUL7M?6
ze2s-pi*YFKFlDLNxPv2
z4nhMoLmkxeUIj$h0Xr$|fql@zGlnrIf6Gn!@>LWLkaj^eut#LK$fBu{r9_x4BgMbXdK3Te-5
z;hgNnGk@hgj6xAEz(u$OV{ikm$X1JuOM~MNiApn{S>TN$