From a81eacdf923e7e1bef5d5b3aca06afd092e78065 Mon Sep 17 00:00:00 2001
From: Cuzmin Simion <120457811+russian17@users.noreply.github.com>
Date: Wed, 13 Mar 2024 03:25:43 -0700
Subject: [PATCH] Add files via upload
---
article-under-overfitting/README.MD | 19 ++
article-under-overfitting/article.md | 233 ++++++++++++++++++
article-under-overfitting/src/photos/1.webp | Bin 0 -> 604954 bytes
article-under-overfitting/src/photos/2.webp | Bin 0 -> 27788 bytes
article-under-overfitting/src/photos/3.png | Bin 0 -> 48744 bytes
article-under-overfitting/src/photos/4.webp | Bin 0 -> 409256 bytes
.../src/photos/goodfit1.png | Bin 0 -> 72628 bytes
.../src/photos/overfitting1.png | Bin 0 -> 45977 bytes
.../src/photos/overfitting2.png | Bin 0 -> 95591 bytes
.../src/photos/overfitting3.png | Bin 0 -> 40974 bytes
.../src/photos/underfitting1.png | Bin 0 -> 68887 bytes
.../src/photos/underfitting2.png | Bin 0 -> 55434 bytes
12 files changed, 252 insertions(+)
create mode 100644 article-under-overfitting/README.MD
create mode 100644 article-under-overfitting/article.md
create mode 100644 article-under-overfitting/src/photos/1.webp
create mode 100644 article-under-overfitting/src/photos/2.webp
create mode 100644 article-under-overfitting/src/photos/3.png
create mode 100644 article-under-overfitting/src/photos/4.webp
create mode 100644 article-under-overfitting/src/photos/goodfit1.png
create mode 100644 article-under-overfitting/src/photos/overfitting1.png
create mode 100644 article-under-overfitting/src/photos/overfitting2.png
create mode 100644 article-under-overfitting/src/photos/overfitting3.png
create mode 100644 article-under-overfitting/src/photos/underfitting1.png
create mode 100644 article-under-overfitting/src/photos/underfitting2.png
diff --git a/article-under-overfitting/README.MD b/article-under-overfitting/README.MD
new file mode 100644
index 0000000..1bd63e2
--- /dev/null
+++ b/article-under-overfitting/README.MD
@@ -0,0 +1,19 @@
+# How to Mitigate Overfitting and Underfitting in Classification Models: Strategies and Techniques
+Welcome to our guide on navigating the challenges of overfitting and underfitting in classification models. In this article, we'll explore effective strategies and techniques to mitigate these issues and enhance the performance of your models. Let's dive in!
+
+## Article Summary
+Dealing with overfitting and underfitting is a big deal in machine learning, as it's all about finding the sweet spot where models can make sense of data just right. This article digs into how to make machine learning models better, especially those that sort data into categories.
+
+The main points are that we need to use tools like regularization to keep models from being too detailed, feature engineering to make models smarter, and cross-validation to make sure models work well with new, unseen data. We also looked at using methods like SMOTE to make sure our data samples are balanced and learning curves to help figure out how well our models are learning.
+
+## Source Code Explanation
+
+- `photos/`: This folder contains images used the article.
+
+## Further Exploration
+
+Embarking on hands-on projects is a brilliant approach to enhance your grasp of machine learning, particularly in mastering the delicate balance between overfitting and underfitting. By tweaking and building upon the source code presented, you'll gain valuable practical insights and see directly how each modification impacts the model's accuracy and generalizability. This hands-on experience is key to unraveling the complexities of creating robust classification models.
+
+For any questions or further assistance, reach out to our community.
+
+Happy coding!
\ No newline at end of file
diff --git a/article-under-overfitting/article.md b/article-under-overfitting/article.md
new file mode 100644
index 0000000..d421409
--- /dev/null
+++ b/article-under-overfitting/article.md
@@ -0,0 +1,233 @@
+
+ 
+
+
+# How to Mitigate Overfitting and Underfitting in Classification Models: Strategies and Techniques
+
+In the dynamic landscape of machine learning, the quest for building accurate classification models often encounters formidable obstacles in the forms of overfitting and underfitting. These twin adversaries represent pivotal challenges, where overfitting inundates models with noise and irrelevant patterns, while underfitting renders them too simplistic to discern essential data patterns.
+
+In this comprehensive exploration, we embark on a journey through the strategies and techniques designed to combat overfitting and underfitting. From fundamental methodologies such as cross-validation and regularization to sophisticated approaches like ensemble learning and feature engineering, we delve deep into the arsenal of tools available to empower practitioners in constructing robust classification models.
+Through a blend of theoretical understanding and practical implementation, this article endeavors to arm readers with the knowledge and insights needed to navigate the intricacies of overfitting and underfitting. By elucidating these concepts and demonstrating their application in real-world scenarios, we aim to catalyze the development of more resilient and dependable machine learning solutions.
+
+## Overfitting
+
+In the realm of classification models, overfitting manifests when a model becomes excessively complex, capturing noise and irrelevant fluctuations in the training data. This complexity often arises in models with numerous parameters or high degrees of freedom, such as decision trees with deep branches or support vector machines with complex kernel functions. As a result, the model learns to distinguish not only genuine patterns but also idiosyncrasies specific to the training data, hindering its ability to generalize to unseen instances. Moreover, overfitting in classification models can be exacerbated by the presence of redundant features or insufficient regularization techniques. Failure to address these issues may lead to inflated performance metrics on the training set while diminishing performance on validation or test data, thereby compromising the model's effectiveness in practical applications.
+
+## Underfitting
+
+Conversely, underfitting in classification models arises when the model fails to capture the underlying patterns present in the data adequately. This deficiency often stems from the use of overly simplistic models or inadequate model training. For instance, employing linear classifiers in scenarios with nonlinear decision boundaries may result in underfitting, leading to suboptimal classification performance. Furthermore, underfitting can be exacerbated by factors such as feature scaling, imbalanced class distributions, or insufficient model complexity. In such cases, the model may struggle to discern meaningful patterns, resulting in poor predictive accuracy across both training and validation datasets. Addressing underfitting requires careful consideration of model selection, feature engineering, and optimization techniques to ensure that the model can effectively capture the complexity of the classification task while avoiding unnecessary bias or simplification.Conversely, underfitting in classification models arises when the model fails to capture the underlying patterns present in the data adequately. This deficiency often stems from the use of overly simplistic models or inadequate model training. For instance, employing linear classifiers in scenarios with nonlinear decision boundaries may result in underfitting, leading to suboptimal classification performance. Furthermore, underfitting can be exacerbated by factors such as feature scaling, imbalanced class distributions, or insufficient model complexity. In such cases, the model may struggle to discern meaningful patterns, resulting in poor predictive accuracy across both training and validation datasets. Addressing underfitting requires careful consideration of model selection, feature engineering, and optimization techniques to ensure that the model can effectively capture the complexity of the classification task while avoiding unnecessary bias or simplification.
+
+
+## Learning-Curve
+Learning curves are essential in understanding how well a model is learning from its training process. These curves plot the model's accuracy or error on both the training set and a separate validation set as training progresses. Such curves allow us to diagnose overfitting, where the model performs increasingly well on the training data but poorly on unseen data, suggesting it is learning the specifics of the training set rather than the underlying pattern. In contrast, underfitting is indicated by a learning curve where the model performs poorly on both sets, implying it is too simplistic and has not learned the relevant patterns in the data at all. Through learning curves, we can gauge if additional data or a model complexity adjustment is required to improve learning.
+
+
+ 
+
+This graph displays a classic sign of an overfitting model in machine learning. The training score is consistently high across the number of training examples. This suggests that the model has likely memorized the training data, rather than learning the underlying patterns.The cross-validation score is significantly lower than the training score and does not improve with more training examples. This indicates that the model is not generalizing well to unseen data.
+
+
+ 
+
+ In this graph the training accuracy remains consistently high as the number of training examples increases. This indicates that the model is performing well on the training data throughout the learning process.The validation accuracy starts off comparable to the training accuracy but then significantly deteriorates as the number of training examples grows. This divergence suggests that while the model is learning the training data very well, its performance on unseen data (validation data) is worsening.The reason this pattern indicates overfitting is that an ideal model should not only fit the training data well but also generalize to new data. The large gap between the training and validation lines, with validation accuracy decreasing, is a clear sign that the model is fitting to noise or specific patterns in the training data that do not apply to the broader dataset. In other words, the model is learning the training data too well, including its anomalies and random fluctuations, which do not represent real-world patterns.
+
+
+ 
+
+Here is a graph illustrating the learning curve of an underfitted model. In this graph, both the training accuracy and validation accuracy are low, with a slight improvement as the training size increases. The fact that both accuracies are quite close to each other and low indicates that the model is not capturing the underlying trend in the data well, which is characteristic of underfitting. The model may be too simple and unable to capture the complexity of the dataset.
+
+
+ 
+
+In this graph, the learning curve for an underfitted model shows a different pattern with some increases and decreases in accuracy. Both the training accuracy and the validation accuracy show fluctuations, and a general upward trend initially, but then the validation accuracy dramatically decreases after a certain point. This could suggest overfitting past that point, or it could represent high variance in the data. However, since both accuracies remain relatively low overall, the model likely suffers from underfitting, as it doesn't capture the underlying data distribution well.
+
+
+ 
+
+Here’s the learning curve for a good fit model. There’s a notable initial gap between training and validation accuracy, representing potential learning and adaptation to the training data. The gap diminishes as the training size increases, showing that the model is generalizing better with more data. The validation accuracy closely follows the training accuracy towards the end, which is a sign of a model that generalizes well and is not overfitting or underfitting.
+
+## How to deal with overfitting?
+### Regularization Techniques:
+In machine learning, regularization is a strategy that reduces overfitting and enhances a model's capacity for generalization. During the training phase, a penalty term is added to the loss function to assist regulate the model's complexity. Regularization prevents the model from fitting noise and random oscillations in the training data by imposing this penalty. Finding the ideal balance between preventing overfitting and providing a good fit for the training set is the aim of regularization. It steers clear of undue complexity and pushes the model to capture the underlying patterns in the data.
+
+Regularization Techniques:
+
+1- L1 regularization (Lasso)
+
+ from sklearn.linear_model import LogisticRegression
+
+ # Logistic regression with L1 regularization
+ lasso_classifier = LogisticRegression(penalty='l1', solver='saga', max_iter=1000)
+ lasso_classifier.fit(X, y)
+
+2- L2 regularization (Ridge).
+
+ from sklearn.linear_model import RidgeClassifier
+
+ # Ridge classifier with L2 regularization
+ ridge_classifier = RidgeClassifier(alpha=1.0)
+ ridge_classifier.fit(X, y)
+
+3- Elatic Net
+
+ from sklearn.linear_model import ElasticNet
+ from sklearn.datasets import make_regression
+
+ # Elastic Net regularization
+ X, y = make_regression(n_features=2, random_state=0)
+ elastic_net = ElasticNet(random_state=0, alpha=1.0, l1_ratio=0.5)
+ elastic_net.fit(X, y)
+
+
+ 
+
+
+### Feature Selection:
+Feature selection is about identifying the most relevant features for use in model construction. The goal is to improve model performance by eliminating unnecessary features which can lead to reduced complexity, improved accuracy, and reduced overfitting.
+
+1- Be cautious of a feature with excessively high importance in your model—it could overshadow other valuable predictors and lead to overfitting. Such dominance suggests the model may simply memorize patterns associated with this feature rather than learning to generalize. Consider removing or adjusting this feature to ensure your model benefits from the full range of data and better captures the underlying patterns.
+
+ from pandas import DataFrame as df
+ import matplotlib.pyplot as plt
+
+ from sklearn.ensemble import RandomForestClassifier
+ feature_importances = df(clf_forest.feature_importances_, index=X.columns, columns=['importance']).sort_values('importance', ascending=False)
+ feture_importances = feature_importances.head(10)
+
+ # Plot the feature importances
+ feture_importances.plot(kind='barh')
+ plt.title('Feature Importances')
+
+An example of a feature with a too high correlation:
+
+ 
+
+2- One effective feature selection method is correlation feature selection. Correlation measures how closely the distributions of two numerical series are related. There are several types of correlation, with Pearson, Spearman, and Kendall being the most commonly used. Each has a different approach to measuring the association between variables, with Pearson correlation focusing on linear relationships, Spearman on rank-order, and Kendall on concordance. The Pearson Correlation selection is implemented in kydavra python library as PearsonCorrelationSelector (Kendall as KendallCorrelationSelector and Spearman as SpearmanCorrelationSelector) and as all correlation based selectors can be used as follows.
+
+Install kydavra library:
+
+ pip install kydavra
+
+The code snipet for Pearson correlation selection:
+
+ from kydavra import PearsonCorrelationSelector
+ pearson = PearsonCorrelationSelector()
+ selected_columns = pearson.select(df, ’target’)
+
+
+
+
+### Collect More Data or Class Balancing:
+Collecting more data is often the best way to improve model performance as it can expose the model to a greater variety of the underlying data distribution. Oversampling the Minority Class or Undersampling the Majority Class involves replicating examples in the minority class and reduces the number of examples in the majority class to match the minority class respectively.This can be done until the minority and majority classes are balanced. Undersampling the Majority Class can be helpful when you have a vast amount of data and don't want to increase the size further.
+
+1- The Synthetic Minority Over-sampling Technique (SMOTE) creates synthetic samples that are a combination of the nearest neighbors of minority class samples. This technique can generate a balanced dataset without repeating minority class samples. One of the main drawbacks of SMOTE is that local outliers may influence it. SMOTEENN is solving the problem of outliers using the EEN algorithm. Before applying a Balancer to a dataset, you should first split it into a train and test set. We will balance only the train set, fit the model on it, and test using different metrics on the test set.
+
+Install crucio library:
+
+ pip install crucio
+
+The code snipet for SMOTEENN:
+
+ from crucio import SMOTEENN
+ smoteenn = SMOTEENN()
+ balanced_df = smoteenn.balance(df, ’target-column’)
+
+
+
+### Cross-Validation:
+Cross-validation is a resampling procedure used to evaluate machine learning models on a limited data sample. The primary goal of cross-validation is to define a dataset to test the model in the training phase to limit problems like overfitting, and give an insight on how the model will generalize to an independent dataset.
+
+1- K-Fold Cross-Validation is especially common: it splits the data into K equal parts (or "folds"), then uses one fold to validate the model and the remaining K-1 folds for training. This process is repeated K times, with each fold used exactly once as the validation data. The results are then averaged (or otherwise combined) to produce a single estimation. Cross-validation is employed to obtain a dependable performance estimate for unseen data, to verify consistency across the model to counteract overfitting, and to maximize the utility of a limited dataset for thorough testing. This cross-validation object is a variation of KFold that returns stratified folds. The folds are made by preserving the percentage of samples for each class.
+
+The code snippet for Stratified-K-Fold Cross-Validation:
+
+ from sklearn.model_selection import cross_val_score, StratifiedKFold
+ from sklearn.ensemble import RandomForestClassifier
+ from sklearn.datasets import make_classification
+
+ # Generate a binary classification dataset
+ X, y = make_classification(n_samples=1000, n_features=20, n_informative=2,
+ n_redundant=10, random_state=42)
+
+ # Initialize the classifier
+ clf = RandomForestClassifier(n_estimators=100, random_state=42)
+
+ # Define the k-fold cross-validator
+ kfold = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)
+
+ # Perform K-fold cross-validation
+ cv_scores = cross_val_score(clf, X, y, cv=kfold, scoring='accuracy')
+
+ print(f'CV scores for each fold: {cv_scores}')
+ print(f'CV average score: {cv_scores.mean()}')
+
+
+### Other tips for overfitting:
+1) Use ensemble techniques such as bagging and boosting to combat overfitting. For instance, Random Forest combines multiple decision trees to enhance accuracy and mitigate overfitting by averaging predictions. Boosting algorithms like AdaBoost, Gradient Boosting, and XGBoost sequentially improve model performance, reducing both bias and variance.
+2) In decision trees, pruning can remove the branches that have little power in classifying instances, which can reduce overfitting. It can reduce the size and complexity of the tree, and improve its generalization and interpretation. Pruning can be applied either before or after the tree is fully grown, using different methods and criteria.
+3) Hyperparameter Tuning like grid search, random search, or Bayesian optimization to find the optimal set of hyperparameters that minimize overfitting.
+4) Utilize class-weights to adjust the penalty for misclassifying minority class samples. By default, each classification error carries equal weight, but assigning different weights to misclassification cases can address this imbalance. Adjusting class-weights allows for a more nuanced approach in handling classification errors, particularly for imbalanced datasets.
+
+## How to deal with underfitting?
+### Increasing Model Complexity:
+Addressing underfitting often requires providing the model with the tools to discern more complex patterns and relationships within the dataset. Transition to a more sophisticated model structure that is capable of detecting a broader assortment of patterns. If your initial approach was, for example, a simple linear regression, you might move to a polynomial regression, which can capture non-linear relationships between variables.
+
+ from sklearn.preprocessing import PolynomialFeatures
+ from sklearn.linear_model import LinearRegression
+ from sklearn.pipeline import make_pipeline
+
+ # Sample code to create a polynomial regression model
+ degree = 7 # The degree of the polynomial features
+ polyreg = make_pipeline(PolynomialFeatures(degree), LinearRegression())
+
+ # Fit the model to the data
+ polyreg.fit(X_train, y_train)
+
+
+
+ 
+
+
+### Feature Engineering:
+Enhancing model performance can often be achieved through thoughtful feature engineering. This encompasses both the introduction of new, predictive features and the refinement of existing ones.
+
+Introducing new features involves identifying additional relevant information that could improve the model’s understanding and prediction capabilities. This could mean including data from external sources, deriving new metrics from the existing data, or uncovering hidden insights through domain knowledge.
+Example:
+Imagine we have a dataset for predicting customer churn, and it includes features like total_purchase_amount and number_of_purchases. We might introduce a new feature, average_purchase_amount, to provide more insights into customer behavior.
+
+ import pandas as pd
+
+ # Now, df includes a new feature that may help improve model predictions
+ df['average_purchase_amount'] = df['total_purchase_amount'] / df['number_of_purchases']
+
+Continuing with the customer churn prediction theme, let’s say our dataset includes the features age and years_as_customer. We might hypothesize that older customers who have been with the company longer are less likely to churn. Thus, we could create a combined feature reflecting this interaction.
+
+ # The new 'customer_loyalty_score' could offer a more nuanced signal for the model.
+ df['customer_loyalty_score'] = df['age'] * df['years_as_customer']
+
+ # Optionally, apply a transformation to normalize the distribution if necessary
+ df['customer_loyalty_score'] = np.log1p(df['customer_loyalty_score'])
+
+
+### Decreasing Regularization:
+Regularization techniques like L1 (Lasso) and L2 (Ridge) regularization are commonly used in supervised learning models to prevent overfitting by penalizing large weights. While these techniques can help make models more generalizable, overly strong regularization might lead to underfitting, where the model cannot capture the underlying patterns in the data well enough. Reducing the regularization strength can allow the model more flexibility to learn from the data, potentially improving its performance, especially when facing underfitting issues.
+
+Decreasing L2 Regularization in Logistic Regression:
+
+ from sklearn.linear_model import LogisticRegression
+
+ # Logistic regression with decreased L2 regularization
+ logreg_low_reg = LogisticRegression(C=1.5, penalty='l2') # Increasing C decreases regularization strength
+ logreg_low_reg.fit(X_train, y_train)
+ print("Low Regularization Score:", logreg_low_reg.score(X_test, y_test))
+
+### Other tips for underfitting:
+1) Sometimes, the chosen algorithm might not be well-suited for the data or the problem. Trying different algorithms might yield better results.
+2) Use cross-validation to ensure that the model is not being assessed on too limited a portion of the data. An example of it is provided earlier in this article.
+3) If the model is underfitting due to a lack of training data, one can try collecting more data or using data augmentation techniques to artificially increase the size of the training set.
+
+
+## Conclusion
+The struggle with overfitting and underfitting in machine learning is like searching for the perfect middle ground, where our models get data analysis just right. We aim to create models that are not too complicated or too simple but just right for what they need to do. This article has walked us through ways to fine-tune our machine learning models, highlighting tools like regularization, feature engineering, cross-validation, and learning curves.
+
+In the specific area of classification models, where we’re often trying to sort things into categories or predict which category something belongs to, keeping a balance is key. Whether we’re figuring out fraud, grouping customers, or understanding feelings in text, the approaches we’ve discussed are crucial. They help us make classification models that are accurate, strong, and flexible, ready to take on real-world challenges.
diff --git a/article-under-overfitting/src/photos/1.webp b/article-under-overfitting/src/photos/1.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ec4c482d5fddf214ffaa6301cc1ac867f75e4b0a
GIT binary patch
literal 604954
zcmV(~K+nHYNk&EzI|%?+MM6+kP&gn+@CN|Uk`kQ(DgXxn1U@koh(jVFAttJKd$`aB
z2~CiPM|b@m9ljI&3qtK%_ug%P*ZMDkAIA`y1(-Mkv`4-`=T5E9-uxw|EJJT
z{m-pm9iP&AMU)eu{yX}w5(}Eb9}nLpu0Nyyisd8Ef3|f1`iH^_2e?x?{zEx%LG8o&Uk=Mftz{ALe~V{$c&s{`dF4au4Xg>p!e}R{w$f
zE$p-UKl#u6U+I1Tzg&OYe}eZ7{y**qqo1$8Y<5ANvpXU;I8}|6%`E|C9Ys?^pYe|Nnsh&;PN1y8i+GcmFr}zyJUKzaD?!|LOl5
z{ulrMnLp(J?0@zDfBw_|ugH(;f93!6|D*qB|MUKL)3^Q4{=e_P+W-IgVg2L&5B~4=
zzwkf*|H}IX{;~dF|3~_N_aFcNc8VZ?SG*ETmRGjPyhe_KaOAH|LXr?{@3>Z|NsA=l>b(L=l@Oq
z+xGMS|NsBcujjwp|K$IP|409~{SW{D{(pcU_5b_-y#IIq|NpnzpWq+zfBXNm|B?Uy
z|0ml2@DKBU`~T2?tN;K1WeHh>)-7V~F&G?EU436MLmB;4%sgx3O!{Fq`ZyXEF}Jg)
z=C8Aoy0b3Cr|Sh;jX{msfJiI)0opWm&&m#lHMUi=aT0dq`E>%fZd`mP-O=&wA`6Xk
zY7siXw~lx@D`teU>b%Pzr}1K(xSF2^{|oW*sqbL@jijnyoxmkm(d}q3cQEipt$^Jz
zKn|<%)lTKd=f!ywv&}{nO#X|9xxU-vH46m{?u=hz`SW`qK>)?B;T&{0jZ6U^0t^fi
zQiKT*@u7Iqjg~2%eYW2!IjB!U_UCB;(15-o>=k41>oEbW!~kX$>)xzdHWcS9?!@LX
z#E<9mEK<7$l|-;z^o7}a%HEG)6_|@?=gb~K2}&-JyF6CzAhddG@Yhhy(9`I}6PVx9
z#@J=t`6f-sZZ*05A^W%Rxis&rw*Z>VRz6U?#}zv?87ntis3Di6qLRo}D$YeR$wpZ=
z=J}Q;+F8Mr<@q<%G*phO4c02?X)W&m2dR+bHE+WnLHvUEl%&VDn6fmHkW3=~FH;P(
ziO~i*ce2Mz*B
z0ynMx+ssZJhzgWH{v#n8^!uMpzH{*a
zu-XRU+wQ4=AN^_YWCmXw4N|}-Rh`H;06x7btdAq&w@+xxPPoepY%;!9Vhtxy?!Txp
zRK0g*m=9My(D-kNmY|B>%^#EOJ6H@4fY|&yd$vSn_6VJOy&>r}NSI
z$+DJ;BMOMDu``Fo=8>5o*4bsUCb~@tL@pWafyn#uZTprZG;LBQY+rvXX?~yTdRQJ{
zbg8v1=Cem@cxqDmZ}+};`4(+T|Hv?6GfA-QpO;9G(u{$ckn``B4I^h>+lY$`H(wg~
z7ytj?jRNAS{{k|kw%VfIN{ug=eL|?Q14kYR%U2w>)cCc=i#^r{c{1*-rYrUyF^0UV
z-Asf66neGehYz7%H{HxH7)g@}QL3(`2n
z?N5%@>h?Uc$ULArW|WNtdAx-lu1W0URcap;cDm9n6VUf|<>B4l_QzYtgK_Q#(-4Q2
z!q6?u?fq8f9R!h?`p3n$K(Phzz{xOmG1DW~48(AD^61Z7ZO?PzNhK6}!M*w5EI1Zo
zp1&3^y_>(UYRD|0td+Uz_Z|L-89WR8kJ=RvT{bQmv+lPkssFI|n=$-V^Js%#ZCu-W
z6UTzvH?-Ib49Jo{`xB3Qh9N<~HXcXJolB_?ys8$Im&BBX2I)n=jGf_tJwDiN2}h=NflE&w`3a-NPjf$Ho!T@>jYc24
zTQaO`@gG(}!8P)BtbM87zqud^dro|<=8m75?7Y|V7+abW!ReTq2^8uQY(AgWaO<3%
ziDAadpA!m7FD>|xNf#aF^$!NLE&{`$Xz_`%)ks?-P=ApRlwV{Jw;^K8;LusLTcY2?3!lfUEVpz9CUVVmxAq
zXz=xLW0=%oDXKMRs`W{z0;^lzVWrDE{1qXE4(+pJgdx1#6#Yh<-b=@q`vRkaP#4eU
zuWvcOfGAk(v|tk#DXNd*sK`{K)SZMj?_UTz4rz~)S`$l6h%6GrSdZ
zc=X(`YjT87p*OULn<`qEx8>#Y=o*=zL+*gsGc31ZdOOSCM$jYFc4nW0SM(Jrnpn7#
zmIm4$F#>$ey=uj?CGGAgr9C%k!EvQN{M|*DEqw@OsY6mRbKw$HuWsPWIA_J@_9|Z<
zbT%fqhQpb%`81=$??O_Q%#X~0kp-&Y)nCsJabUkBt#S{N6va*d|M#hJ1sCL^PA43i
z7~_cIK7Cw0fB2;?k(2h6$6O8X
zwk@~LLWI|lhIZ4|{-2lXraPPVRXC941_|oOhD$M_t>%a=RLnLNb-8pB86Ps7%T{TX
zjfR-<+3HnPo+nuu*Pp`GgaG2n7+Vy}5uTSb{a>W(f4Q=|xx#a6@H(c6g}6-`C9-K5
z?Aqvp`A?@ffFU}g8XF)vl{dt|^ibdD;4j1p2E+GOZi;a&mpVm7ogeMc{A-oExM@`t
z_FeUcy^-xi69)jNPP5_^ubd%og8};?Q)ZUQd!J>nbK(xcHxH%Y5Q8}DzC#KmF^gHW
z3)j|2z4B23T|9LzY?ilO)0(WxGjpb;q7U;2o>xofCNa-x=bmr(&L}1K&!(gM)Tf>&
z)Soqqdv?z(QPZz{4qso^0>S5Vn0mK(_yL~Bv-0ompOsDf;9pNRNyPWM|Ky%G9FXE3
zuKo~z>*6xbLU2nvw-M%16?2{$1<;^9+$S^fcuEwodA(%~IJT+ZE}QYmQ4xFnFVqO=
zNO@-cQEdkqI1DFRX02lt6m4o(Om)A0+)nvWaNhK{sdLh+3#0;pgbeY$D03tNo|^`By^_9it>IE
z)RZ2}H1Q0zJooSH!fyDOCc$cO`&))%=lJ)&l+y)H=R_*bO}qD@Mm)k}lo(z*EuP!s
zdr1ToR>stsLvMCmY&-|esPf?j+NzzQ-oM~{&abaX|JB-Cz>W&*4=9lnyGQ+_v>iFW
zrzh3c+DOlVSk80?%-eLb1k;m>1_FDw7bWW4*Nj!ZK%)Cv(#g=HP+m^q6m_QJImAo&JRy0~*>WI$oS?2%w
zj}_DX4l#K-S|k1JlX+OM4_KINzjO2mC#Jx3p`xYmikiW8lV2+J%8en9P>%nD{AzKPp-ucZbuqSk_*BCmv
zew)*E4SSM{2^^lrWTU-#-!vSfGv=k$dVGz(5!d&ZHPGIVsuHsbhn!OOtxJB%cH)j~
z%QWKm4r}bce?{Vd=4Yn<3?@j+2c3W#9b4v-F0r#;Z8J9>QDRz9X+^P%0I&?}_t-fl
zbX0gKA^Y1^AN4Hub1g83m&6aI!;BoUd6>wj^Z+HecQ)f?>D}27^rFJdjK*rrtEZ@V~1v9e_dAJ508JEko}n|_z0I8ti~MN
z2tJh?^ARU}Sg#S~@s6n8K>N`CCj_nIoy2h=Ge%?ETf)?>^QC+ztZj}W#%H|^Q-3Y!t^dhclQ;L+(oc?_9I=7
zSM^*fm5(1NeiptaZOBkl=ecUy$~QGYiq4r8&Eh7+35?iT_NbTdf|OOneB5zG(E9(>
zQmHbx9#2;P70BGfv}vKtJjb4PFb!8K?%QQuv%a|Espo(meT!A=8;sjs@=$42*OEOJRu-6C*l~+Jo_uBIopG%{Ry|s$iib4?goy=6awbg
zSREQfZO<*gX_Zz(`xfc~Y+BCr5oM+fuyimImA%jOfPKPx}e@Q*=wv
zUJLr~;c3MApPT7z-9^r=-4|P%-_{u|1I6AV}og|
z|CDRLMd*L}))Qwe_2=phn)Dg)iqWwZ*VlwSIHS%JqLoVZV_Qg0-q?kuCkXc~w0FddEhv6A7p$fw|~`y
z*0YDN@dAp`_Y5H0NMc8r`-wj0TUM@(Yq
z6P|`Pau;j%XSCB8Hi55quJC@ZZb+cSx`Nv-BRdV(I3W}6eVi?wxC(ZwWsZ7z@sljC
z*I_Rx31Gaq1miK{$23AwjcySH9uZ*G)S8b`xM&0xE=F}fL;cH2*}9#(iuw^El!@V+
z2*WQ$%t22lvVX}OgJbt2=FAoxU*)G*5>k9&B*?~evxWt_*OaUq8kt6*NUY=I%Qp$y
zqXrN@5gkL7GG`MthyL}Iv+D*9*E2tc!=V1n9?ZTPjt{omInO8hR9-toA{d_G~5yU%{Y
z{Bm}`z~Go*S&S=Qirq%XaI@MjAb`NMz7*3i5ckAw)9lJ92vrO%3lR)Vq08PQOD9bz
z!_)U6@7EZ?!bERT?0#1|?Y-h(_bQNm^;ALnjlRb>Tf3>a?<*)z&{TG{a5Vf>1Ro#cw&dIrYoI@McUQ%dzh=c|U@Rgk`&33*3GFF5N9
z&B)@*V4ZNyqhLu`n8{Yr_{t+Ia`(OtD)`>D?h*5eqBL0z{*a~%SJR*H?+1B3D8@L`
zsd0`h(U4d6*V@7g3a}#K0JbOu0I0|lN8rzsXE1h?-X>l(O^4=Md
z!jPof`cxCQmO4tHxL9LAv)^X@EYhx$PJI>|n6Zjam3vM9brM_#lyO8n`pZ9IdX(4^
zspis=mn^|OCDlv(px>rkp1mCsFm7%(H@6#E>dhgdf4}NET3Fp&ZOfZ{P*nr!;M~Y3
z;IuVMvMRd$I!+?`vLE&9<24P*=TbOWW4``9Y<@+Ml5N!TXp)$RuKO=-HN1g@*$^;vXsHWZH
z8P^;B+Ogu@4PWw~`t9aq4T<*x<&>$z9SKkSWi!gDf
z8#4CK?D3k8RUJ3tIx(l+q#&UU!ul3GU77t(4su(ZS9c6o_R=HNq0Kwkt&>!}*EI!J
zUdE=c3xhx{1n*kIXf|z+-)C7^Q`^K;Puh$=5Z*>JctHCu377>|8s01pH
zV6t&Dixd*?|7tT*CDmb+xyk@qc$+LK3_R6OL*P8|%D>=%dvPx|E9OiAh9Mb=%ZL&`
zQb~N8qy8e3a|Z5p!0egJLTLV}K!>_k%9@z(AP2wOwgb;euy7ZB4e%(NX7RxzVBF2p
z&*LeWZZ-wJ|o*Uc`MXi?R>Bux(D8JK#^12O=Sh
z*R879wZ?2*hVg0tqLo!*8rL5$jznVE7)QH-Rl}vr9u#?@`ziOkE9RmuF}uts?>qYs
zE%MVd_GPuZB4Mm3OWYqP^uPPRsh$UP*8m@6k`DK!k&6Xvh_M2H(s1da?3c4zGu$4V
zixOq-exaWjH42^cu2&fn%`O!wnGlHNEfE;19t(aU8>LiVG
z@Bf@-3kk7l?BmCO>Tulj8%(47`|!y)=1^BTS>qyE@3lPUhEOhJl1tCqMfM{MHwdagUQdEq
zck{B#X{b21Eka|FERk@!v01>f0GV-~fN4GIIQj`+?fwFTD4VD4u9Z^8bwSB2ZKJz}
ziV^liEaSZ8TQO1^<35sr{hbG1pbsYDYa;O)h22(GLWt{cI)#tOO$G;~trj7epfCl|
zbbvH6qffu>JWEw$Fha1TXHXV;Dp|5XF^jE{lh(-pY|5r}e}Z5|-3kLIjfKnthuWIu
zmpT%a>(=4{K^k`b%bYT){Y7R3rqB>=%SVDCV3f>H5;3+dukUiq18B(y#11kd*>a*K
zi#)18tPR=OBl@1;vRn5xjAN|KRF5>}hvY
zoV&4V_{P}tv_~pBUk<%>Jl>qCAO||ldHM+HUS`?26#tL;`DSRb&j31S+g!t`D{luo
z@;q;C`(I=c-B{&Zb_^*O?iQ>L`xj2ZcJScNcV02-^cF2SPENrZS)1<@@vK}=yV0?=
z5zRODNrvkCJ|yH@-^k)(MA5+xzD78fwrPlepxwTn2G27w&;#*pgj5kxHg2-03iRxH
zkKzq;1o4srYu2;)%(m-TGRs%82P8wdgWJQcg^$PINCa+P5;O`4<6sp_)@!+gC9M(-jeg%)Sz$%`t*+v1o1jq3;zn
zdVlOg%ok^&wy}-VY*r_El-<=J@AQ{%1?Q#JcM;D?Fuvr~#&kRSNls~c1laB8_8_@J
zMsw0cfqu8jDzRmpQXos91rOgx%i;J@EYBg{!YPjF+|m|!!@3`@Xfj9tOMN{T?q;n6|U=&Y&F>-RcHa9%7SEc
z>@GQQ{*b8HWRhaphGHo|nCS^79ODoXa(XStL?RO6qegnah|8rrU
zE_2_ZSmIa-{(hCqNEJFBFf|Bzc~`AVv9yy!_b
zMhqo0-y!o24_$a97>#gdRx5Cq*xd8TT@QanH3brQin7&Avd;Bwyn4*EXWAe?@c$M3
zXO%lrTSz}k;Y`-_=Z8ccypjw#ps6t+bj1iMdJb(TBy^lfrX4MpIarn|)tbk0Koiro
zPhA}zO%jWYT+}U%G*_7ow%8Yb@i8cITT#c!Y^JR5KN{mx|DFsFn!=%K3J`F8
zaS9_SOYY;AdCrNXrfab&;h)htStfLgAKG*pfvX+*Cd3MCyL1za3sg^qd<@)seqLo{
zF1Hi&`l=O<=|zh@@LUmf`m(c_8M@w$iR7%g5yk}!3jTkf<%e|lp$L1;ZvicUit)~~
zdS*7;)&QdNp!uvN?vM-#B_0lj
z^zS(JisNeEy6qmxO_1s(xsS8JIs9V^@dI9KVJkABK@ZFjvouP31Ps=hL>Z}@vQI0&
zs&3gv4a&2RaO)1*u_(2OAxAs>5SCE>4JO@}v&yUDHXW;~5i*i?R=alphx96<4^_Rq
zathW*fczo-J_`4K#DMG9!j(yH-D#AMgwR-)9L9@DXS^f~k>=izlj{t}*{ec^T;&61
z{MB3k&lW<23(516_B+5yKHZ%^qbJS5q16;fBKG=Yf_A1dz_M
zTDqB?ylV3Pqz;Q2;~hZ=&Vrt>;vDYwjXDQ&UepEdTC>&XN?{sKaUj$jUjPS!10`Mq
zMW!NN057fz4z8_!|tj+9wkh1v4NL*IfDi
zB8x8Ue+K+0nKcb-oZ_H1br$1lGn6%@d{JhI$Zuche+*g0ROWXc6J$
z4(E?Vfw3f5sfI*(GHS8@^sGD%@dmg9VYetV6CtdQ?J9c~Iz1TBy`qj(hi|C*vIQ3i
zc-miG9G$Gj$5_|Mo>{T{vP4d=0`*7i{DqhV(40C%a%(4~M6!KWab2z7ZFAy=bRYIn
z5zxH3iX&oyAGsV+Dk069{&7>(`%NB*ka6+=c9prCKl%`N7|e8ss>B8P2n{oxj>*#W
zWG}FsWSS&m#(ySODLaPDFGyM61$m(*UCbbnU`J#wn__tDFw
zmo1fQtnf2!2yVe&&9k9cLn?s<$OMb5*r7e9D*EpvVh)l&+
z^k*esu}#BfQ<3E`=VD`{-6oK=g_Z8#?#L)8vM=7@xWY_oDk*^bBP_IE#3YE-*}Se0
zgscws%y~+W2^z>LgovnQEi~1ftoB1+i@u-5Xk>I@PvLa?&XKoI#@LLLWk)%a8ay!+
z`w!JbbH@8?mVzbE+P;y%s5w&lO*Rbq%V;AVOz$-+$}Xy(x{pwQwOrQGc%
z;J`Vu85RtF5zGD8G0ev6$|r|o0Jdngo|7z|S?gnY=bHsfg_R)e%6YQedG&Ql{0MtE
ze?~RVX1Hxm(AeiLWcX!TU;%A1gI5_qwB5ZjTY=t8Q2q>$mosF>hASYJ2nRnfUJ3@2
z!PrlTSpX8=!t7B91Rb_Cs*hNH*-=LTo;rj-YTE
zlm7vL6`XxIiM-p!m^LNW^8smCbgYtn`?){AMJp$Rsy!~eia&4?(u`plK$=ZI%Lv8d
z?@pTJ7iLf$DHcVt?hR`SBa5qPpUh!Vm6h^N)flm&2Kq)V-*mGC58YKJTY**Hz(6`^
z`BpI?9041;IliPN;Q?{&G`{&>olqa63d7hE$Z;WGu-O^wTLoi;@4>?o@9Rt24p*MqZLvCE>!AmAucMsU}CRKv${H|Rk%6TOU
z6VvbJ^9-ryIATMx%@#J-QS{M3^|qmEgdxijXI3S-DCq5SZ0W?2M9q`yUy<1%ipgL26*r=8*C-V5xTguWaDkhT2gp_qTs>Xol9GA
zahKTU^z6RWWx#*=+nvcanTR#NO4^>~2-kyN#Uk=#?t<_H#4#e-45N}9hudA%tCsh-
zqHrV6qyQkJO5IbqA*v)qp6rXUNBkxk;UfWxG709_B$pXoIxxE|7rj4WO-J4zx06f
zRFBk)ICw?nu78%LU|42YY_t|I0%Aw8Qq*Z(1f2=eo&&BvhR@USezt7~_2~N7EDNlZ
z8D2qr*wxH)Lt$JI+%G5oAb}bKoWwp8hKcgOOi=8)fB(1k_0LYDEnc`9(i3jy;2;WoIl*V`qg<5Y7#I>I=v5uP*I$PanecP+Gp
ziV6r}$@&ksPO*xt%|BtN&{yux&&ssEMFM9qRh&{Rwd{I4j)=A?UA=~B7xF&iPy&1U
zzbqC2J8<`;uXb#s!S=wZ08Qb9v>4C9uy6D7i$r6pq2VVc7Lup$aB1te$TkA|p9l87
zIrJYkiXP`=gJ#IeE5;kZX92H#=*~y4MnkOobu)6>lI8roj%%mS`5@T>jndL4eB!il
zO72gygk`S>PtC14XlB|-c_zL;+gTL=AUGRAIT4)bqno_yJqeE02RoHl>W+;90&89b8Y7N~
z!Y$Qu@`sAIoO9L*h;tM3O)mM;S$
zs-B}Tue|{nG8K2Kqeb}b_8(NnCwMS0dV;|#s{qSQ2AZz=bakzEUV9aEPkZzl><=eN
z5Z7;If*?C;H$w~Kn)uH0Uhw8$LE(BvBOo97%To^>8{o)1`~;ofVUN&}l*~w(trKRa
z>3^Jc^^z;Io0SJRTlcC2jLr7p(TQp3p1p4eShJn{iH>ITt0!uwpAAN0M&v5dB46;b?MoG&YJ^xHPt{352nSTn@9lLx;0{HLL;F$P(4>{bEXBx`4=lK&06vL_1SG
z8h^FoO#g?Qwu6TnBt>9ejbBK%1
zv);#J<#J{oY&;al<-k=9E=LYP-)17jys=$H9kCtOUExdws;ALmX1r6izxju(qUq$7
zhRdD$MZ``VI)oy~13t4gA-U?XvOY_5w}9JoLF0~yFf%8N(9LQC-R#&#){|i)E4OG;
zme$W~`M{{W*S=4YH=KQwgz3Md&q2|6qu8q6i9WrcWWO^n!m@*CFI8v`MD_J@NOu6r
z-hbLhSnvm~hTG4t_>P(_#6&Y$joS-%t`mexpitY-7(pOX@-=GA9$6m(Fcj_3B|0(5
z&1qMU#?TWSXR1ZZ;&sy1KXTu;a~fTyK#L3wx2kW{7c7gY5javu^k)y<#gIx+MzcHE
zooBNfx_wOPJP=0^20%F*ptkFx!}s3Por$(?hpLkCDq%Ox$&z!Jgfd+oQARc+bt5QK
zS9xX#niM>Ty91|4e^brI!6lE$UmD1On%|6WR*U;kk8n|Z8li1E>tN#4FK_v$e^xe8;ml
zP&YR1VUnptdjHPBMX+Y~jK%LAU@vNXr&Zq&(g3L(H}JO
zKdtG8_O_XPdP8>v>|ab~@r<61qp=h)^bh0saU6I@f|Uc(R8cA>lGZ@=bvC%-D!mW?
z&WcxL8&y)OXKz&oj3T25;_Wikc-5-0(^O&jmX#8$+~=;ZMs&P`Ps
zxt#P$elzD=j6`dlH}r;TiGoXN&IL`Wd;%uaXKt`Fl>S8WHG|hPD!CZRifscw+g6*t
ztKX}eDqP>$3eGAS<}2M(Sdyn{;iCe+l2kUWU;)ORZ^7e4pVcA}a#WmgNg#6+D!9`KxF;O?y{Rf}_3cltRdJYh@*nm{2FO#PQ
z!$92WSr6pAHv=6BGyoywC*-NXZsv`caW6!Fa(mmu@$Kug6Cv_PL=7Ar+za9AmJP(
zYiiF)-XxQA)fu?o#K!bN99{&I5j#L1#(jxmp>0r1C@)8M?X;(dFa%DYCBiYc3MNZG
zV^cX8w)>|j)z_ut8%~?lwT24M?)aZ57hBCXa!}bp1K~CBXwsFR?DTbCer%r5%Px!B
zCy?(Z0SjjRSmfC%pMIVQl`p8oEP<-fhKj-eZ)U2R*eK}q_sFVFAq@}pWdYYk_A`m{Hyf~zlqH)|KsddxN`mgqT&Co
zy6lHr6!~Q`hT@?aHS1l5mb-IjUJU91x1rP^z(Mb)7E;NvGB4-PJA11!UA4v?Azk(D
zE{qDB^Xkx{4ZPbiTWZ|he5^MsdhD`|yL1el7L1~kUVK2!BPV81?F=C&<1h~4(;iq&
z?$!C+Wcfndx0n8}b^Q?tGc%J`4cQBg0{lHz6;sa+N*cUi{aO)NjQW=hg_;v)gw3$5
zC+=Fy-n5|!3;1%tiYfnPw$-wv>5Nm|<|-Cx7}z%O;yzGL_--bbv1sRoQveSUA?&gr&LivZ$55PAAdgjtL~o5N#+grx=a7r1=fH6{}ntrd;i8K
z9C?ADj#1tr6`u;ZAUd#CV{gXCjm~X;xBrF0AOqe+29ifi{%^Ur9$beHJdQm*hjiY*
z=c%O>wqKBOR2Vlc;Ve(as%MQsqOVZ=s$@;qUylF-lho`$Hi`s29ec`9FEPHito<6k
zH?9}|ytYKz7EgHM6+_l=!>FJzBXB6je@%nR;;KxJO7;af73Jz*o$l3m#$0DYiEtCp
z1z~4Ev!QIeBnB#`N(;yT{&2ba>i;}Qf3LGIV9PjbtfwI8$8SU$y?q-G5hmCOePz&2
zh%!!tI+?Z-cD`v%kmc7cnPQgp;feQ1zGky`$ouXgpnfTN3m_ixX>mHxNx0ye*ws<8>_jkG23>gThMVA4&I1su;>ByBhvDzBOE={0)b-M5sr7DO
z-{{pw%gx-^36I(7BNw|O|JsLkVee#qS~O3*<=RR}Wr*NCvIE`slRyN(u-*;Y5G9E|
z@yEH9lu*aM0AEq9pm!JWlQ|&vGBFQ2s8aKYP$0^sD2|LerHDz
zCGFkc^Eu8sS@4$4Xfc^_E7xu4#O!mM<+)?Irm>L{<9GyVg4
zC(1JHDij_aEoU(2qQ)@aB5ALG=<&Pge~0!gLT}4_dCd}iVaDN6gT!x(fHKQO{zmhQ
zg-B!IeHy#&-!l=a`saN>^AN8onAN!UF8GDjr7iR$+XXgdl_Y2)N*j6wh%?2I@JbS6
zVpZ>GHTF;QQJo!(Xy^C-(*JmoV+Cf@=uEo0I(`y-QF#%(ltk+ct1^M=&a!4`(mPsE
z?I{Q?p0;s)9;$i|XD-CUg$=6wv_?x5S29a@8t1uiyhlA{=Gj^1`cD7{M)CyV7}OPX
zbNy?Bg*xFcoLC~3JN>!nED^s4p`q3(o9yZk(YMkcRM?x+
z>hkiu9&`dvyPwP3q$GIu&7)lQl_1ihY+8>7f;4H{A2pQjwXauz+L*n4uVIy8xiZpP
zGb4<40#RBd{XXUahpx@|9k>12U0B~T#Nbb5%9YM6^Xw7PG3i@NJ?KLaBZ))TB(F2P*BFUrq
zgJ((3FOa4va4%Q%)?dGC!?L
zcgyk2F7G^&Xc+o})szytpxx;9R{S9DJyAPJ#yNzf7FyvDR!V690)$@m1;6mn^pu@(
zsCRA`&YgW!*w8Ka@F_6P*JFD`#y8mMhP=LaH?ALZqs#?C}f@euJUm=MKlTRW+$u
zS~BUaHt^$}%TY{ic}It}#_HknZqGcu35(d{c;>A0d;3>2TUMX}(nu0f?#*h!+X$n*
zLLn|jv%`r;b02oGq^em%!9=lHh#PS+6hscqiykST;`NjRV|V<=nPjI!Z!%s)xfh
zHX~g>l#-kX!ZvIss?A;L|JGy^C%IR+h{|LeR%_1EfV2a@g0n33O+WCJ$Nath)L$;X
z6(z?yL!*Nyq3x1G*7Jswq%%ito2DP3+PaycUDZm%1YT$h{vdW6D6-30#BqUCkR6}-
zTJU0)nS9joKgg`95KV#qu@&hn?rvRh_|OdXRxT)`BZLzz)J2;LGtkKAOSvv+<)&uy
zddc4LChpC#zOCHqsuFhD*d}BO+u-P2>KFt=CH)tDPXKGj8%u1p|itRFW_RBOISw~I;qU*|)90j;>
z#&jqEJ0jaJ257y}xpd%`5x6`pt?dv^a;2BSc-(5@pK4b&^o0chl=gSsBr^
z^9D~{`Ak5{_oehW&Eae`Nf_QSzp4u!pQJW@Rvrx(wt)Z&M0%dTGqF8~Zrg)k1o|Eq+Iz8m=nWoSSg}69D
zcfpHynT*%KZD*o-ouO}fyLaAIbne`*Q+u6U-zFM92fn=Gyk?Bm*?JAM@aX{9dH3Z%
zIIU4syOa#jGqkq9cei4Vl|&cEA&|;CvaM?tL-YX1E{ia=)?ltq!Zi6soZ&yNRDnA7
ze>ysSDwlO4Trw$^MD=eSDL95*`OjhePJUdO+{QW^#;^MAY`7b7ag1hZpN{=Vod1(~
z;=TGpT)z0Ve}#I0zGA
zWBx;|We~AGZB#Z8^iDK1W%Hwj7D(&$eTn3}!^oLllD`EWP7l(PNx&7FVjR|pk
zc<8G9Xg&x?R~vrPYEPp17U!FABdOOe=hjSd-RF{cxk)0Dr{@MF!?%!^vikpzHr*rXGm=Y=QI>bkY5+?lU!2E
zhvkjFz4TJtnz5T|wG*W~qfGhP`@$D%&UeaHx~K%S+~e1#O3q5pj7leBxg8JH^WL8A
zD@y-R7`WItA9M)pToK&pg&Da(dw81y(GE%ZjP9+eSieprF+DoR{>EuW_IA8KHAzBm
z=5LC`Hmf4?Bl4Y^5x3|h$*cB#%mz`Her!G#{_{T=Ub_hv!TlO3Fv(wQn(q$3|d=v;n)fdDjY|m
zQSGefskka;60@xjH7?PsK+|u)d4e^Fu|JPizxQ>papW#&N-UNX{}CPW|CyI%kQUwZ
zjiJ20_VR+B9sSmfdCL*Y$9~vKX~Pew=J5(!0c&_RCgbKm2;
zw0FiwB~AxsBuCxN75~d2iH*a}!$bcR_d}p$a8MWD8-#}I1mlGYmed+uh0s|4UPD3N
zg50%$AAh#QV8Fy`JTa)Pmv4nGTR%u2%`bK
zLI0GmF@cLR8=3YQqW=kAD
ziH{Ki*d2DrGB<(G#C90{IJht7%#3j)zl+BhCRi2!m4-k)M2M7Qg_H!
zLg?@%uFs5`Oq#lX$wIHVN$lK&H}exB-J+NQJ!^yJ{=;PuI5sC*R_bvz;Ynli3&aW;G<%@|NIhicBYZ35!sTO
z-ZK@_mB)2c-!B7N)Hd^wdWPbdnn6A@05
ze5x|8f}3pdzV*4$Z15YDx1v)nIxv4jB52JN!g3klhUs6&jWrbil*c8tUy=bT87R`p
zxsSYhXhd{pGk+a)#*$!L=}|Iqz@OuY%`xNCi?T$moGEb@GJ`J+~pC<-~@BO|v
z8uu~9;xM>oyJO1}mW0@}rlNHRkOmYFBEzq!)C8Nd?PM*PWq>4(5Z|_XbIN@%M@rOG
z)%ojdYFoPaED}FKrh`l1W#}BKcld@PBReirCxo}J9(-Ma0)q{-3zE6CJ+9)@z(!97
z&3#XR)S>oQ$z7MN!%e@rBg|DbBFH}f8w|@>O*>u2oTq1uO|{e+cl^yS1*vzgEWNuZ
zwR1ZROys>Pu}Yn_+J$+QCv);tageMK?_MbFP@=EDzi+kNs8Pl|792T9!j@GHq!6EY
z%5E4E&Tl$O%QYdhYfh?S8Rvy;ZWxy8H^d))qLCnDDndzf(=Mz;g<;230uOqb%+Ckr
z3of)jJb=dI9wKZM!kUvyg^1Irqa+R)8VB)Qws8x~|GiK8mb96=1c$gi+*9Kwap0sz
zCfOm!0M14=TfH&>L>{8%eS3}6tA;!jK60zG$gDLOpv`j#WCgFkgA*KNSks0_@MSU{
zv0nHyf7YAcqe*KPiwfG#)!UdIKat6N8gzB8bC##l=V;(8eS-9_!pwL}!p&GMJ~>mN
zFrlvNGk1G(Kv-r;q>X<9gE
zdyM6(ZCCW4ib01bG1is#gG_+WvmPOB?W-*iEyT9$XEjlw&n&Gp%lFU-(bMM2=d8Z@
zw)O`2S!~Zq-xLBjgq`tT&|swm?9Y;+9&4mZBoB9ffBI)3yC#V6R&%xjX9EZ=G~Q%w
z566sts-_j*ZsIwO2V>DxvuRjS2@2x+a~|94UbDDCUXv1~LU}JZn#~t962RXq}OB(l)|1MY04-Z`F2$jxtQnq7gIR*p^bLvlIR**U=rgPb3asx@++G{-IME
zk({-u^ANOK&6@G%Y>Mcm
z&W5CX+g{djHZWMOQ6)|CkUBJ<8+H-g+bbY(HNa_|x=|`VjEBX^DL(x2byf@Tw1F!j6Kgnq1F8!pC{6lK0BaHYcTvDTgGk3ZK7M)G}_bi*0UPtQ0l&TBY4~pP7G`D
zfH|6iZEUjepaaNC%hnX6`tIle7=m_2rM
zbToB(k
zuIe63)^X+ixN`_zJ_|eSI6jkHjI;s%BDPnQZ-lC$qU%p%gb(xV(~JV133`(y2d5x8
z`zbTef*%{f6kVt^L!g=;od{#*Njf1l3lO@K+@#%gHzXFx%wwPDa5rFFFsQ8i*GStH
zA15Wj+g_sm;~XhxiCC{qB{$YCegt>XHKFUs+yuj&X(NOm;?L_K;Q?xLwM_`JZqnh}
z{$m>cl=PqXKFTrkm;cO5<$OGVz5obaaac?JBFhMgqUB+d_CX;X_#V-_&udw_vst+8
z(yk{yTuVY;Z*E?j21%%G>j^ghgh)RJQAP6+lmBl&`8K!oeD}G?gz+p-TbAi1p_Gbw
z1psm`=BR6My(YLMpIv;HrqtTIK4eU6idy$L7}my|D&DhDu4qlGcj+K|34|CCC(cF!
zs|&xj9wD$6Lt+jUL#w%!^G#r!KMMdkF&>PlyfWbhqi;$V)!!PGpTWnWyiQQdj
zDO{Y~^o^(e-AYtIuM6bf3HNjWXB4$>-C}26C{*=}dx^ZJ5MN^TbmJPy<1VEha@R3q
zJu9p_>_U7+7=$)(76sGW>p=X%l278r
zcy`D=$88(=d+GmvQDdBe<-S{NlMCFJ4=eaw%(6d`dW-1R8?S)WErL_Qos?#-@iFbT
z5!Buk+5(jAH^&?B3$M=Tpl&#OI{JHkH{l7oM%bTEdx#}kI`(kSuzdpmsx|u4ycAjL
zjP1hC{Z$e(4hGXf4+$D>p0AIx^wp(ADi?4b$a3Hs_QTxnZX$Y}sZfBe%}K~t>RHst
zec#=6F3e8W@9v{g^!^*hhr1C4Q}Euk-A&&N4UwP*wUa-~sHWKIGLI`EKXq;Rk_&sP
zcQ_B)!`=jQua!FKbobsU@ayk&0PUt77>ciMe}|_1Z=yet&YvbTYa$HF96@!uT>`MU
z448qeuuU??Naty9nZm;=SBi#NF%*1*uuYLv#SPvX)NKxmG*X-FWMOH0Qw8t~94RJ;
zW`I`SZ87@RZ&55Y-d7Ove|Sn(t4nofRQ86;aZ*gkfBt8Pg}r6ruWu=_ijD&yY(9MK
zVY7?W5G+)uDw{p+s1^3I$_Y3Bh0`sBMdTd6{#gEcV_~tzPH}8gA^yQQwKqLKIGBW>
zDC7qs($e=>rys2PVJDG=(Lp@gdnBdNjHANtTvv0(w61~zrhPYY-!B-feW#cJyTUkE
zAwE@6m6W7(w<^8VHA)u`g3@h==|*wKj|K^{V1P(Tq&GbSmHJNeH#Aq;Cij9zC+Mxe
z7(ku65ncJv5t~vO1}2q-j9oPOwM`@$icMT+Sj+9iuUSRjHnD##)PFt0xjOY@w!YmF
z*$fRE5m_SpVkF2t`5al_PzT8Q&icc88xVEdfqk|TF6`XbU26ER&9n0s!AA_{s@}>h
z`C0p$cE!f#*^GPn~UwHzt_Bpx!3e?zg)2iMUH08%-
z1~$eCyBeMUmiNxNM#^6oo5|8B`L75nK;08pY|g96>sK%a65u_}PsRY-s0(Yc1Nr2d
zT#myo*n&S}YiZ`f1wE@jLgc0!-jyz{`FwV*e)-CSXZiPgdlW93?4VRFsj9yzJ=TJy
z1+r|sVi2etv5)ymLz0vjNO5&&6RRZydO}jz$>Ad7H~Ek?PX|vkhGt>@oc(t~z#Q0d
zE5B%4kgdK!l?5+}*Z<+emVxa^KwB~uUynBSF@G)Bw$6(hOCb*d_aNCIvbDlrqeNO>
zgIx$^_Grc}j*w68v-v#*j2+j@aQdt3@AlV;^1r|CZ$_JSe^`-TsbC$(=4RvnQO?xs
zMPrrq`DTFlg?fpp7L;EUiJ+n~es&{FwMc}mHq~EZAEYichTHazeB;kOtFPj?5*UzU
z5@;Deno2av{_H|*_L^b2OfZ5}s=cg>%#l1
zpz!8W#l^{gyS2EKlHbWxPmR}Pjhm`AgTls*5hcr=!^$3(v`*aO@V}P@Yb%sqb)aT`lDX$D2xLzp
zd1g*JkOI=)0V5Lzq-TVVbT^(HX&YtdM?k8!%C)osbaDTP?%8uV;1;tvL2Z=~&ed33
z?{d~6OuIN494!P#1DsXwAJ5!UCQ1#hu^)pzl?Pg~|K{m=RPB+(X6+i{AeHLU{-vfu
zmf^0z+}y-IUfNm!@qe#Hb|ZEHX63LJeZ8^a9Dcs|Ct_(-!*1!M%^sF>RqFFU3Ib|_
zmhmRr0(>t~*^oePH9f`2)U}zWiK7)v?ZcjI;_(&5!z7*Vjx!qsFN~UCj4xH2qh3PUpHP_i?xFM4pRauuakyZ`W201a1Wt=4)Ag6b+`4(*?nO
z?aOpsjtX0OMZyiwb)#f~eiJ!mW1Q4%mi3|{NurrjXl+G+
zQ5Cyed2R-V@Rom(LMI*EO*PDiBV6L+>1da$UN~TWKPaX?){h9#k4{K4F
zpP;vkA1WOGb|!n|04W7|_qXBzGUwX3g3FfiDHIE(^_i$ktelX^=fbd+Yd8H5+CRQn
zYP}&R@sOy$OVPNaSyy)4r$2Kwf!~aWx%>*1B?O6-7(5q#G8SoyLTUvuwC;&pw^t=G
z7!AZmxpj^Kv*G-JmO9l@Y&
zX>jo}k1k9kCChH|nrH@kEODFTN2^JkjdIJ|zMUn7-&3BAGY4+EPQ6eLu)
zJ+h>FFPDN*3B^j;yoO@T{8n9%ygaV9e$H?&%_n}$p~|cD)@2(#Jv`JXP!hUy&>w`*
zGo@l~4*(Lt-CR)%#8bp9uO`J8UCXzcwRCKIq&{x9a|!0~$d>>VS7l8C;El!nPS4u>
zj}Ukt$KUPZr1tn$b?N?v7t_R%Nv@!mbiTF~{?PH+~drmgA4MMX%n88
z|JHvMLgl~-%!jBpkS0(*=pBD3M$nLRetGesNK~{8l|cW<74pk3_8dgRlV~X8(g2uw
zbCGlMWY%TE)(*R^_%i1Qk?;P(jzyjKW1^bi-yzqP?ANC864wOC->ns$^i09o6O2`tOzK!8!-^Nzi!M0OKCz!GwT%y`cx(_Qc`e8G+R-nk92
z%GslAg(^Cag%X{=haayey3Qrr#g
zkO8eg6UD~$aA^$E5`ehAchGwX*WjChjwz*4Di!4gzo8sDyQR-n(@B}k;KZxaiuf!
zb$hy(o6aGWaHws_wT;TwLElAkS4W2dj0AoEeBHjT-{xv|+bqh|=A-Am?Lm)@fFvUU
ze8E}!7d#9*bHGYNm^Pj87_!r^e_Ft&(bWyzqOW0%;V!INb!xK&o~=*qr&3MV!C;=f-N=0smmq~vefcm
z<(s|uO}py&f5~G&(cIm!_g#OyTri_mkhs8@rhak`fu9$n1*&jL$|XZ6zlduzB$Uzv
zd@kHmP^>eON?F@R?4oDCA1>I{CBs)$S^-iDT!vtd37->$UjB)%u`62PON_lNq?n01
zH-d2ODu#|DOKz`@qF9F7XO_nF_fB)l`JVFLlDz+QP>nvK?zr$(>Yx_ql=*BRQCFDf
zmL!?VF-3xfwt#)p@>&(IJNsng(E=n{mgb~WuHi5O#6cOhB>Ko+?8s8(}P0F%B
zW6)OJ4Zk-4$T<5+%Y|0s#Bh9OP%}qvB
zajW`kyV{K3o&eWM+e_2{oz#`Z3=n0B`>E1~3K(77!?wz|0NwtQSHhbwePsOru_HVY
z6U@i>;yH1fT+gOhUGaV0S}viAq&7Y+ljlx9Us(*AZoh4_CYN_~y|FD-qx_-VuG<#RFq`q7Bkp%SvsIOi|E>bVIpI#)rUiH>1k5Q4>
zdD9batpCBKwSQ`_a)3>sCxz2!yxbpAG*Z|$?=LMnUy?s}Ui0ixy#2Dvs8zi{tdk}A
zay#)omwMxt7@BS+(>2!4q&fM{z9(c-Ae2ofZfQlC0H3@BhN-H@`AYH)W{N)2@ZQ03PR$LbmU5IbYP{en>SS2ZrAVV
z5pMrfp60!9bXO(Kq;{&n;!@vCkQ@ccnTEw$1;Q{n%+iVAo-DWunfOSr#!fjyCLxTY
zd$5b+8tc6|%?ZC5^crpSW@AL!j)$Wy1$k50%Gd#+AxKshwxze6z8iGTg9Ql%3cWB2
z-GcsGT_y;UO@~kJnP#Rv7ttpEj9tY18^_{vvIc&Vgvc#R{AKFxi+c=PUx}KTu#y%g
zudTmq@?cao#n@YMU9Qx98i!B#&hVhwJ5M<4Nf~r
z5kPV;Eq47(^9sD`QedX3={JHrc8USeRJ+$Q#*Wr|3PR;<(14?ui!%~G$L`EN{4chX
zme^h_|0G$j?!%Sd3}U&$%BvMy{XY9K4|^G+D?@ONDCfoNL%ul`r+E5XW2}(HZCdKQ
zqR%=ko&o+OWN`{CH>N=DcZo0`Mqd_IKUgtgT+IaK;Xi0RPB6FYzMO!Q&A^DgdxU
zOu--mNfS8=R_%N%z}|X5O4qV%Df>W9(3P?do`_BI)l@L-9b2SD^k_4O9Ih0CN8&mv
zZ463;kczyXdZb1e6v|}xU=67}#MfbKIDU)T{OQ!(3}V&B#(z68d+<4$;IJp#mGXIv
zRfuG%$ReMli&oVu`j|iO>Bv1fz(uRwxAW^Ddi446V6ylqXQ)Jen>x-a%ef2J>#^2T
zpcJ@n26Zn=2<@hwm8{XPkbYAh^~h(^g2otJfbUm6auB)V`Zj&qK_=V
z0giI0O&Qh|U6(n7_#KZXq~1vnndNzzb{=Z*szgH~l95iP=@Ehr(em1)M!sUo4%U{Q
z!asaxp%v^cto>+ggMfWg()!F9HyGON!59EB3?En80|CCJ-m@i1U!0%_PoKWe1J8`!
z)0T|ZoH9oL6-eS_JA#!Nuj#pOA!(^MpkP$5^2OmdeO@Vz2SHwn6|3duU&sk7=KHOK
zhcJ4P%?IfgSYlH{mM=f$Tn|GbM}>J!kQxU^LaC-py(^I@ZGk2JX7_LDYPiR7!kpC2
zBmQ~;b>cRf{Vsqr#wEtRJ)t*_;tT4?z5}4Rxy$BdK5wTie1F~d)ytz?^15sIBm;|z
zwFh0hMPTkzn4Fdyc1nSFg@gX&L=3H!FYuDzs&MKVrq*B<`Dn}D+tKLg;sAndeLFfD-sFAk$$?Y<
z%#!-iwSyGB-g%ZuOH-aZOIV~=l%QERVQR%)Xb-?EbBb*95#1@R%k`N}B~slSAA=It
zy?XE;r|m~M7J1U^6fUkYL_q9qAM7NDOv7AI-=n-gyLHs2F-1x~ob4zq7E*in%Nvb#
z-US+4V&-P?j1`UMGfMVEvu_2#-xI~6fpZu}9nhcIb6P#;16
zTr7y(ZVoq+d_`q7dV-Y?2<@+8*+CTP7t&Tx%8|7*Pop6<%acvOSy8B@m^3d!L6OAB
z_Dw7W@g&H&fjyD;W6$Fo^}cZ0F>bs`hndSYCRD(GO{Su~%nLd=!)Ma}^2!x8{_Nhh
zzNp=wZEb%j=aYG!2?PpPs0CrcW#3CsPB|e7`@?gy#WiR85mOhzkOMXyw@9$JErEwF
zKtZ%>RF1zzC;PL2PU>Av9t!YF-JUaqB}GU$UKw9TeazN8K5RIoLu*B!&jeceMUq+CN@Oqj|@|Y-koa
zmhgscd1Y??B}$!#u&nN;hEWdOv@`YR_M95;*=evPbL&($P&E;Wzm}gzSHHuKJ*Rve
zLgynm91Po#usLqeE`fsHmHIKlKr%Z(hF{CB231|YnuO^O+ueUAnoE&qJE$h-URDMD
zJw;B_yx=7!^oX+U!xC|xm{$c1T+A!}|EkZ{
z*8eYj`8Tb%_$SZNFJZ&5r1U`vvk(rJ3v8!xHIrtHt`07`6|~bqfH+H?LF$5l$vZ6v
zw4&vLn5$lLSOYrEO)L|U3mM3E%_TaC$GCiyd1lN{JheZb3YrP$SE
z@lDS8cielbZi?Vp?C+l`Ea647Rym>pp_VBC^#Gt@%IHl8YgL
zz?|WYJOi^%lkUKAiGyg8;U{Tf`0{y@%~E
zBmGXqeF#=5V@Dvhw6%bR=*2~+-(9Y2RE;F)+hm`1fWLeM6-ZeHKd<)?~B9VkIYzYgtpC`8M2lg;O$2*HxDAhYvZDd
zHG3aiK7x9ESP>X=XfB+`6vSg)R)k)Jl2%x!(mVf)98k~93ZBEOA3mI#CZDERcwBqQXO*cVVzV0^l1&kXeixrUme)v&MU4|yf`DGeV
zjq(S7utwBCd7PrLVF@yQF*N_BLfCa(_jQAbV$dP^S24LWnmL70Zl0WH1fd(b1*fF+
zR>YWBh#j3Kvct4-yB`Zunw_9n)}FKHgjd3WY$oM}W#_&Rtb%c0#;Kqx2yy#j&0C=5
zJ(Zbh02GIpG?XUEzszn?H{}Pam!eU^v3wyI$&JG98*oJ&XqWQ;)Zvk4MVr!lI?AQ_J
z!C1ifQAWdGrxX*IXwS&86{0u_PbfrcDteDAiZO}q-mG{n>;J-K
zxD;fd;*rfSfmzEdT{HKt>sak2p^KlK{B;^`HE7Ux7PpTv^gD5z*8j9x%wto)qDcFd
zF<&u8PZA@;y_7Rn!qy+7l}75Rr}ht+3N=&lryt_^yRlfH4b*1^+~6(=!TS4sAK)lI
zu#nV^#Qw7680o=8uwJG+o;PnR37Mc(N;_;OZ4Gy(ES3JZ5$N=<$=2Sw;Lp8BMKrzi66wV?#m{!)RRaP{
z-)jHNp3bws_j?Bcr`Ik@8QKk9a-17(^qDcJ@s4RwGS;j%jZF*gB$T$FnEgq$NtRh{
zzPL+#osjxE2@+{t(@V)GX#zw`9p26@9
z!$wt4ZDp9coL?}{4upewE;O_1Ye!RgcA*`(KKvOa>Z#P+i`+)66>*I%zUPh3#)eDS
z{4Avdd3YzM#_TC;Q&CtFdKUO6QT$Xoev1;y7n64zvH~P~StWRFbG|ZBsX0@8-`Hjx
z6ndIPx07!l!%YPHhd4`-o?;^F%*ySwRgs#E}n`Om@i~?kN3I3iR
zK>B5Gb4yn1XM0zvq9p>!GebfNH2<5-CjwOfM`lR(_i1)9(13kZ>St2V>Jc7hhDIJ)
zWtjOINu6EkTpJd5y=;m9)IcrRA*^8=L;1)`e`n3d1B96lD0+QTtQKq5`A8Drq2}zA
z`C#P+@XxS1Z5jJp`2BSZE+l48GlvR^GT8_0YN&#a>lXqaujnMx6L&1J`>#d1s&kd9
zO3hEjTr2s_fM0GBQ7V%9X0u|nvnK$Q=1fJr=SxurYFiuYl6B~l8f6|6-D}4}O6Df>
z7x3V5iCRh7zi9sIc=jbRVa=vgHG2(C`sgLr*d(f_`N&C~EM4PC+m}F^QNx|cN9@GjZ2??+FR7j^Y*>13tU%N=bonqvVC{s*
z{Vi(5+t$4?YP0JO6)M+lC_#mDo%0h5DrpAe?5UHUY?Xs0`4ZrvDMO72VEk5ooQ
zd)(peQU@>&InxA}9gyfNuFiV)sL^*=1~7^*@E^2K7>fSt$wQvV&LK7BQ-|GnCAt3Uat2{Tgbq(f=e
z;Xr8&kyrKe3FUo=izR?gH1AjNA)>dJP+1iB_)Tz9UC6M`dY(CBe8^zsm*~#Lru`Kt
z`=q#KI$#IEfv1~v)r-tE`=oDWSYx(IY-WbS^iLPXxH$a^8bsMpTr|Z%auC0A2QD94
z{*pN|o~_5|YL;JewA&G5ap+PU->GCrKUnV1Db)UYcAaIIVPILP5Nc}OWYOZtFgED)gnUPf3@{H?%AmxZ}$f5N_d^xpd
zgWA@!dBVr}v88@HQu|q`xtehzdtBYu9#hF)AM{{T
zd|8bFYe~Ew<8|NioN^CuyofQ59-eDS*N!#M-N$XmZ8!^4T;@1d4w$ik-lr^8`Xb(M
z(_;Jd{ptxT45RYcgbjdVXw%Xz4(yU5;cONZh7bLludU-a&E#o}qkX?>Op9yu+49r8
z#!G^6B@-7B)hq*ptg-wTjWq1k-hZFZ4z8Un{=IRX+vw*|nPxaqS+!yrX@`602C;*=
z1h>5Y+{F_Xn^pGjt;+lNE~3&TN8XM0FmOy3T7ep1t<1N%?lCnHP!BO5y*7~xeu<9K
zzDmvR6VMtVkchF&ReE>V;wNB#B^L>H^U}D{`<;Ep-z9?}F4kj)saFI!VUi{Y7n2{{
z>4x+A29O`8a_EqYUtkm@_Ac@yhKJH;VIc4xP%?ef_qJYx6{mGPHCdvdngyl7V_0`?
zXG7o&{Sx9($XhR|;2X52s4v6T=zFpO)??`_t6w_NFN-sVWsw`I=Bxd@o8$bhwN~>P&{#;-4eq~NE8=oAOgH8xRGi)KHu9!jUuI*eQw@6IV!pA
z0t45iF=-_&{`>(su&O^hSMbgk`b>&D8^(wa;qo)|gj{YJK!SKsDrx#jut8SQ>~MKQ
zO^&b|Xulh?Buvxf<%>hSAT2u|`SbT!bNj#h$Y%a~=zQ6CjmlZ}JH%P-V#P81;pce^
z=Q$%MWT8$@(Y-dJmD~srT4o=XB;O4hej-B1kU(K8O0-}+bF{(2Qe1@*bt_5A8GyaI
zEM1ih8Z^AGtwa?XnRt3TmEQI>uODh3Si+@Z9xC{}N_-e?>7g!=6^~SAzPOxUF!~o)!zC)*mNxiawZk)bc-fdT
zz{|h}&u+TWE`%QXL=HuKW2S$6Go2>=MBw|wK&Ad)=s|<;`SaD>3?vP|LKY%l!S
z6XZo5ImoK%mm~m!TS-n%I^+(hm}ifKuZI+MQlCwCqi~sXVpbDN}`C
zDB25y0lO0Y1$6Xc8DWNrd2T=v{cfeXobw|VMF)XhJr*2aeYE(9N6R?gmDA(z5nitT
z1Zc3FBJT&DM^uyd;1c&l;;8GRPwiE6uM;C*KQsGP0mE*C@G203Ow(-PfYjl
zD5_pKwj!8RTEe#
z>R=}zzRhGntp&lU`tYNkyvnC|&*!zQ?xs}V!iH$jGeF(=2e(x1)`!Hkzq)kn{!qc0
z8g>Yp=4b-bI*Z-p8g%R@dLBZ|;*XGh_O3YLQR1Xb^1cj_(s34mKDH9HMmq@jgc*B^
zEdaOfYX2(Hplg;aq*ZU+&FvzX_t|&BowtszcMFZ`ZJYyYPDyP){xbf)u30jO&-%H%
zj5d|tUVK>2swM%a@o|3j`;5${O>9p>-wMF^YzCC{pLa@&eNRo;kg1YDyc|2-p~OfM
zK052QC@gpo7noNYUH*w$&f>EDHi&Z3nf^RIzMTf0bg!3X=~sC^iuUOES(X{>dg;S1
z0Q()ulBzNP?$Lk{@WV{p5x!CcOWJTuW38COu(b9(KP0(VMnNo{HkYt9O67L;7J~3M
z%|Od%!VOfJ3{k01ln_%wtEbbhKXEUfMmc%2R&z83A&P1a<&c(8;;_OAk88CR6u
zePdrX`V$ps*MU;0lNg8!WMZ_{@KBUgJPydK`e#3VwC83NHp&1-!8DPuA=H!l`R9?S
zUd7aKx!!a#{>VGkxE;J44MaQO;H-%6Ukm^KD0CCgu=Hs3)Y|STlq|IQqg9t1T5_FJ
zZ&`JA)bO3a!_85}KD`dM7rCNTOOVo^n-U{i7NYrnX!CKhK^#6)n;ROL+n$PPIdUvh
z6k6|6LuGx@qJ*C8YOf=yv-(OASX(U
z{k+!bOO(B(wb1>^FJ4ea;Th;B`d6tD3!%EA7c`|E&c(6dLPr}|pDt$!Er
zBHl=e+*=USluLC})WuEHrauofYgRP0Ms699lCQvj(Bk!FCMZ^OiG8i-0Y$$J7*DO|
zpco4%>s-vwsXp7HVNL(N_I_X(MW`8L21EpK!fka<^yJW1!m_XMEUg!WR#4kZW9*BN
zNAa3cg-7*ToW67ocn_oKXd$u$#(JwZN
z#;+MT1lgMYQKs-OW!r1ag#;agme|2T*l*>ubvR8o)B5@Y70W@flKwKLa{whzPM&3=
z#-dqxop<{!#J38Vp0A_OXn1O`eblGWe4!k733*}pYT{Dr+B+UD#9Fqe=^~f%-sq7YH64r5Fw!wU3egNZh*n~{kayFn
zAMrZuU3nt?x}4Uqud)Sg1qJ%>1+wptYWcaT+C8MH|2}E`%n;i>CqZGCtfm*{dtk%?
zPkoophTXYW_YQsW1{ah>97Q{{zQdAbfKi}3_0D+b83j1Pdo`~>u$v%W>JmwgSqIW0
z)_0ikEAEq{bHr?$E&4Ra9+LM68rygn{&K%|dOtc8(<~hZKq)|J2TeOoC;o{{Mwpe5
z!$U&QN}qT-aWAWuBtx};506xeoBd}Hq;N03%cwl%so2p(IUxw?#X@jEfXYHjQ_Zrt
zkC|waEz$BYLYtHI2PRe+c`)~qM)`rZ#OF1!nB9e__CzvV=IRo-%r6SFqhj~og>=yl
z$79geb-!z~)Y%-%LYJ>t4S|ko+U|41zvX*v3C{N#MB;rZeC*%cx-AGM|6jFdRVW02
zLeE6GuCX(+OlW)C*er>ua$MetPnQR#0!J7Vt|Kewp0$jw?w-#$k@m0&{?zlTQnF&_
zfUG5aNUs3#B34W^x9p~{#ISzZmT+l=?~$wKR%N1Lea-0K!avcp1eEm7AR+UfT0|rH
zz7`r4OwuC|jra`fGeUFe6E{RR@83Bm>k)ch6IghN4u1SG$pHTE$013|ALne1Q45J~
zf|`81D#-4w@ruSOm!t9`KgVv_LsrFLR5ID&M43KqSwj_)B2v2|8DXmXbNK+P&!R)&hGCS{0V<|mN;C4
zk_wB}BHgx&)@Y8CY|j}kZ~bXI&r@Sr@p2ztFjaYi0)%~nV6S7P?ldr&apFlevqy>4
zM)qF$)9B>ErK1V@!@z$Zr>UgsS%~0bw0(zN{q{_G(DaTD7}6{AB&QpXeAfwDZ;_wX<}p$ZGHqvqeol+}Qj}-+6q|6VI-sS&
z{7zDSMh&epzvCmqRgXvPs{lDb#=pPWgYECkkX{r#ZXkuu?kba10>SgbD&nq^L*-Vc
zOS6bONgmOKB~XEo_H6wjD8g%BT|0$(z%rD_D)X&R=!J2wHcrXH?e347cXQD5b(upm(OxJ0Ia>gP_9
zPR&eveL`Am$)9*X
zQ=!9ib|F}P-tIuNj2YSC^q4aF;=bRqY^0H0H8g(wpWuKze9zgxY0ea|{V
zTipY~Fwn)ZEG^I82WTzo24!5d1#%Xt%La&wXZUfnbFS@1(H@czLH{nnyfa0-D%u~+Dq2@
zj~c!Vp$4fZeCXyJy2(OZ^DR!E=>W~!KH;?xsH4_k835qW@dR_-38P|5Ut-fbca#%u
zCArmf-@?h_x1r$dMyrncnM44T)bfTjU&##wDsRcMa0s#ELC2hT5;h%B?-OhhHEyQw
zUcz5~*enUvr@`*^jk!Pb)+Z_XMeaY4=EP9htN}C#mPQ=JK48pHdkX(fb74Lk^;%KC
zgA5(j6Yh4qV1buE^B%*W%UyCQU^N(5Vmj3(bJ&L08E>yvAF-2)!o-NeII+F;^@(9!
ztwja~+^d4cH7{KS+HKRa$WtTmPN^P0GliS!A~s50Pmvmn5OBPQz|7*6#yqq&@4^cY
zF`(iH!>{@$-<-Bl9{B`~bt#^L+)9-HB~?HknfHv7|LM!jLs@*7zBTfqo8I=&{m=ND
ze@w5G*dg@RH=!lka9-wrA~$ri9TPFW$wQjNXP1yT14Mfn?WBan*jN+EYeuY(A~qDJ
z>9&z7_{*DlpL9q@(6E3Q!M~L^82iGTiwiuh6)o&I#*&%6~u4
zsCO#!+g0b~um+CJs5+B`H}jb7j+&A5(SJZ1IyoFFvz&M-!C$Rb2rW^2iY30M(L9N%jz=3vJuA+-VG
zNy?9M(v*I(g#(&0+mkUT<;hW&<5orSzb9g5^9@FzYe&FH`rVgJXGvxU)IP`&XWZOj
zu_Da7{&`d)MpeyCs#)QC4`hAO8$18%AW+Yt1+wwpe~BYR*`n_lxIrC&@yMES$t^5J
z3ld)CmO5W=dle_aqOjWGy|4ZAb|G6a0|mdP}9w&4}+8I0o}(dvz-E!SfP
zZ<^92KxW0X8k#IAa#b@Hh2FYIKT{Yq2{(^X{J(@Om*OBlWXU1c@$5Nay#NpJ1;m&E?d3u{1x{bbK?E(*v5}VC^u($jHgKR7+K?n$3??
zY3W8Xw@|3V;1|r+sw%($kIv5_;;E4UJqBMh8>W#W9;VI$3KnIkk^Oa;Sqkbz?Qx9dp*L~wDQnDbh+D7N%HL(aQ$pWJ%|UGAD*db-ck>{#;ET$
z)i+Hp@%@w|AvHV|{B>#)@PWPPycTbi%+68XV5C{=5C|%=u&Ilk!ZM6dK3ONp_2<
z+8DG(RD2Pvobo_uCdNH7?bEtQ%BaN8Ix~<<87ZR|Npiy*V_wY&#*fx>EN<&&^Kp00
z@hJ~cg%)kHk0b|qfaN9A#hQ(vqkLE^V)*k@`}fCDJ>5j0WX%mKI)IOfZ@8PCG5+t>
zK$TTjzW8nqI2KpfKe5A}N;c;E_B>?g!5o~Celtc}Dzw#;!e9cKB@$9iSa3TJ6Q-d>
zd3-E-W7Mk(>>E!*m*@+NhAtbN9Qn0JUpK5=^)KY!_sAJnY-8oePrxoKX$
zNZ*7Z>7X(KMKZy>8g*2Uv=VkUTtr#CBZ&1%1@?Q&!_mUT_w(dDhvMa`@iT{Jhm_a?
ztiHC9>s#|>uHZU*MwQpEXY1c?4^)*0qB%&z4lK-1e?llQd-ag!I$g84D~wLhxZ_6V
zc&$JSKD^t^YG7f8VmAD4|F-S+86H2Ow9uh^bD!P~aKVT}O;Tba_2Ac~}
zw$H35UYSkqbOy?xYR<%1nU6{Dd2PqrS9`b}`iU3Hz{Pm6d253odF}|Y7DSb7&Hx45
z#fXHm$R#I1g8WZH_GMT(!k}rM3Z#6Q-NVMeqWBo{_-Qx3uC8#a{AUfGp#8|oq)~?
za->70!e#aI20t0ipKCPallKmVr&U-1>W$-aRPUM|dTeHFMDT6p1{SK2Irt4uo=z`a
z2mE`rKEBZhbr$WCZoJ8Mf0F4)H5m?*{UanUuFWcPWIvd1bAafrA#C6iT`!bjZ7iYNQYnNSSe}~laI|C
z-vj8oj7C&k7((Pb4|9z#BraR6T}8MM?ASzAMugr1TNVi+cJ3+CI%z@dA|pop*IqNE
zUpPd6u8R|fl#*IqLkcR)$Hvzq0c@-exztL#5aQZAugHJj|LhLAbSlzt0un((6;QX1
zG%v8qTC4js8dN6URGhX2iNZ#-p({r
zbJ#N~z*U59129EMJ~>zJ3VNeGn3GwN@axO@yo%6SSL5#yFPw~p|DDX|v$MHSNX)YZ
zi5*Q|1ghm8G2p@WbNr7cYn;dbI5uRcmlksW{vN##nYUr~%4<~nE`#Vv&eoiK{M`EJ
z8m0!W>gBynJg@)%_yvNR{#UE%RnF138-RyG0ZkQ!=bQ0}O}&sqe$Jlbx{n4Eb%uGd
z?!^QqZjiemdfVl1yqVbQGuR~UGKfL#xJl%48Ha9^{SNhU4{L)w%|TwezUR1C9izRC
zW+pmKHzQu%jVl`f)?UEz?3d79U2;fc>=%7XDS_1TzE&COTTRJ-kAwY
zS>5NwYoO*zUrJSk$AS!j&$;}s&eS?Q^$S@G%F2yDj6FsEB(_F
z^zvECdp<+rNTsCqXBU*B!JQSW9n!Y#^;i^+eP2CK8dbn~K$_1MfEf%(ZxbF<>ho{j
zom$-$%N%SkV`x7)gW5oxgQ{=#zrmz8%;hWpO2PgZ$!CJG(cyHZu{>aNFoEGytYyOkL+vRd|g4H_42{`
zwR;rZ<`Kn?%oKNsUNIw7bLM#qV{(2KBmsfrsB4UDQ+;ltW$8Y&I40@gvPlsY%FglV
zyFz_9g;`b)*9lLC3@$!boeJr@c;x#LssjRE!x?iE4(_Oha}3aNj?w$N)%J2+jA41@k15JIR`+3YXcz_6LjC&)F1dKqtM
zUw87*yQdm1;_)kGZy(Hp2HiGq0q>S9e01Ni^JOA?K3uNY_-Z}U$1zo|+-pz&aq0`T
z0e{$2yT+rqTI6E#rexUk4%^cAyh({47;8YF^A4VPkmBxaingvhSCLcs(
zGV3q7|8IFu|N6KbjeXir(4b1E1tY_q(jWZ*dN31)Ot^#a101W&d&pd~|Nri`n6pFw
z|NsAPW_)+hmt0_ccnfA2R>?&S3izp|kwXOpW=Ae+pu=4e1KOdq>-_>VxORDqAIw};
z(8wXg3g_XCV!B9|Ljib0Nr{aALT%W#r^+#bm@DyUtRRPg_kxHs@m1ho^q?-gebko;
z^ojJ>xThIMt!XZaPdyZ_d3C)~r(EiLST1yebP&r`FQo9JP&EEKZ`5>xL?8N9_FE6_
z>=({&v*JCb7JcHw)
z-sRbxt2Lh+y+3q?sT1Qz2<5`a>trpe<&Dy_`ZwSH&|Op&>(;A0&`}MGMWPsIEpi1J
zz90AvJk{3&?G-Ck;g=bx!UJsEjFSIGB8w3K=rhH=@gi_GJ^vc>&*d7>zO{Cwn&LJS
zUaSEm^0hQO!aOf*RKMFR9M}feOz^#d4_Q?#x6-rQiGp)Xf&};f{3}36a9)7wdk5|#
zi!*7gsPG2>@2l*GtJ-y
zHT4!7MD?NxS0`o`y;cuI6F8D6fxl4?P+qt#(<{WL&T5dM{n@49icJ9f+7@$#Eh>{t
zdcF2AYe;6kl`bGb6(Yp@7r|S(@zH@?V_1)XzW8~!;2{~n_WgPjDStbDkAXp7?sbLI
z@@q8npQ{-P5qH$N=+XiX8w9?L16?4%gg6P~rzzrhwA9JE_UD~&$t-8Es^(S@_EC;uRnz)eeH$BkqKLM5G*rA33Fq!
z@%*a-LvPUB^QH_WRZyw<1%z}0OvEl2e)t7v9Z)&H#$}r9q
z5L@s{0Ms^B!Un*Z+!uU-tMv|{XMmo}H?;x!e11QDN*y7NW&^hgNSVR(oQ*CV12U-s
z#bV(GbhJze%AgZh38r2>`}e-F^BU;8-$n~>O%-q#ff)m)W2y82ZWR`1PZ@+}pMQYA
zY2|9HbP`Z*M`zH9yL`@F>blJDX*3}dM2Bak8*De1bSD$sa_UgM2(YDqc4j|WsuE(Y
z1lhMmq)!`8)Lemycbm21?QruuSieLe=uk9NKe+AZm_;93U^=EkD^#{x(*zSlMtq0K
zHW%lnZ#wfp=?e?-F+~Qe+RA@J-i{RtT*HM~k7095wXxA)tLbe>?0#rrw}99-19VAw
zz;OWu2t$umjtQ~T74d};x*?zWgBC6`c56&b?WU8Pf(BmIE(`VgG|`j?~lJf
zr|XB_U{sZ%V}Pd#9v5V8Ph|*E7@VRm+1IlNGF2!cB}dMhv;ub`Mv*ODyaI=^yZi_c
zJWBxbOQn~&BXhfcY4CArqXR1WAFBIOOf(L_prg}bKIkNAM*3EPx6+8L#TwLxXgF6+
zv0qP^QCS>brC>cW-!Aq9pj5(=1FFoyNsPNfr4aIXL7T{;oy!cJfj0Nt`GezJFN>Fe
z5m(2eTw0PTqtI*>9<=oFP$UYyV>SD7vUYg>jC3G-jmE8{v7>7$wk>VY0`_8VlEEz!
ztZtw_iIh@5G^>aKOw#(Vx|t7lh3>qG>Lmm|LquZCquho%rY+msH#EWqNGZ8>U)_TBC(XBvA|^U-WJSik2^j(!oJN-?a`7|z664T3p@Moh
z|8fEsi=5T~J)WU;8kz+{>9bJ};bswQvP0cY|6W3y)Zh~-u5pc7W
zl|}pdt!QT3gJg7GRXg@+4J`gK&^$#1$Q0xgMaAIeyUlrD
z0sSYO7nY
z`M7*5RwssMG9xW1(t0IC7+B}h-g45ft2KN341kk6ZX^j}D+`^E=%Gq|_+f5#{ENS)
z#pW#2n1GAWI?3g{D`)DCs%67$4+(zqfghxH-}nD>?l
z!CD-O2>J>>VoVAJ>C#l32Z*#fP7B<0efY*txVjpBt;g-Gf^(Jf&!f-)c~}Zd=U*%0
z$3Ds8;Jv_2e8i1{G{CG(nN*>!Vkt|Ni$aZb-6j0Tr~BM!bQJXjL-J~>^sr*7;5Rr!
zhRzQQs6cb7NqLw}>*c@2@>2I{RUtARNzPL1-`iRbixN!nChcu@^(`9|UBAvn%LyfL
z=ybfi)t;Lin$gIA%wMOJh+|6l8tU5>9WDxqi}l-%B#$~R#}=wIkVo*M8w%0(Ps?|POneCuc%}x!NxM5qpJuSTeNrOjzdq3A1
zpUfJU(T{yes441_1E%WpNai73?0aQd+}oN
zC=Oa#KtOG3V})KQN`q0Xuad-Y4&X)kd6yhO8H5r@kN`AK5k2)A*z8Ht^wOkTw8DOy
zGl5QIGXn#!m;DV(Sl3x=zG{)U(NHtX{2|?e99dV*K`_YX2agu09sa^Xm7ps^c<6g?
zo{M$f#Hi~a<#44Om!YLH^p-oGkH*QS6AiCS{aUjR&byAt`=HM_x};Bj4cQW*65$VV
zDZ#MNVbH-*6w1O{)9mC>*WxMlcU&Z8c%FnOq(BJT;Jh`+Fz$nI3FXra;<|o)Gt{Z^
zv?pEi*gM2LgO~8E;iu?|{Z5D9fCY_IP@~v97ge!Fl4MMGNLi_0h@o@pS#dl~ZsOpT
zt44txps}p_r-ewe-L-ls!mrfU*V7McbWB-ZDYrp$yyA3?_XMupQMO)$9XZ0U^up+H
zWX>c<&R|)_k|gm;Z}N9kf1=npmOE+n9|<6VZ0n+i(|}WEesLeO$gZg(sM|~2OMdv=y&Vlx^KD`7GK3C+c>5y%`G
zABr3Lrs;s}Mx?Gdg)5}XxEVqwi%Yj~0^0M<|NH^#QBzIq{&U$DJ4S%!>ChRNEB02d
z=}8bkD-kTke%!wT;ZOy&8$cI#$w)lYaPT>T14dRSV?(w9Jd4a#2?G@r|(0W`~bVdBX*U*>;zXv
z^Iipr;NlYpbBD>mte_Awl~(J5A>hp#26>FFmXK0Dqsg&rbDQL9-TH0cJ;iEj&-_tI
z*ywhQXedLzRsNhLyjw*>T=;IK?@(RoSXKrph2zR%xS25YA;E$RjHt)w|ao-Ur3*4=}{4oRk`^=0OJSXr-`dt=Ld82tTqV@#vKm})n9$RYp
zrH8`MzsKOJ<0%z;Bb;6R;K(7JDf;m6i1R$c%AZ`Py?Mf2Du;7iEC5Ka0OHt$E2lF|
zbMd#Lj*%63g=tPkw3thUCvWX+?LzxE2nGvMxioClt-azMJDumQp%<_MlzPUC`o)6A
zH9Giq^=}%N>>M^?s2S5Wxo3k@0E0pn^!`C+4}c!^@>NM4KcZy*1Kt)kI7HBvxGl#Pc#jv?ScuhSe8$!XnOvZU^8p
zwR5b(9tQ7rl{4*6^1lRON9B*$Le#|Zc61(|aTkANNj-dU{!#P3;>lCT8KVIM%F
zQGf0mXn&2!jO1shw4)6&GCU~=E&7(OnP){V5+?wly}y9zQI5~*LlF2eJI{)@2%~j;
zUgR7sc}F|3Wv!|=35pvjXdf!|lnw(5S_K!(;XS{iYqI?2k53`_#1u2mYzq))w9YAo
zILmHPKK(}D(WL?3D};p6l8RPAg;rG5-t2-e!{~g-Q<;V;M^5+ovj#dvALXbv>x>#Gt}EW&W`wq4DD7{a(+Lftde?lS!2&z=f2N=AZ
zzOK8>vC<1Z$nzf*y!Cg9p+!{@XIMy%XRJ6?HGLDtRA{P7M9X9FS}sJD8=n<;V>}8P
zQ7W`bC|W}YDn3a8B=6;a<^MT{E^o&GMOVQUpd&b?e539lcyMJPP{>1FrkMSIYhVGv
zauestUJh|JnMIsLM{nA)2g?*^ZSPvfQF2ttM2)8!V)dH1#|6C{QI$1C9u
z%{{y<*8CCGh?(s^3C-r(3pd~0W!z3jx;(2f9WXKs4kCqg@VrBP~~{__SpML-REA#PUNLr;yrM4*^TMtO
zc0XvMR{MZ`=*-rx1&?u1>w*`Jz~1@`VBRtTT*P?Ory%E(?sE8%@rZdZz?B67<#}c&
zeztdMNO!u`_xCk>&w_1)tyM@F;TfJ@4nYnlmE1aNHo*zt;^t!v4~5#1Ky6#HkoYWm
zcv9*S0u9xB6|VuM_x~?+YHC@Dw7oo0cGWDsgB>hp+uA_@4!;zFU@CyUg)m1lN1i7i
z!YpvytDeq;7#kyx03z3+2Xu*pP^2mV@#0-n;ZMsW%Ju1$M#{a%%bd^6crtWfsLiq09&toITGIIz_5jXK0nffgRW_i|S@eCQ0%!@eL
z-Bp_~fbB*7f;iHz45VD#=>?gCycI{EVA6bE9?Y-e1=XC9Ql?Ru^Kl~%w;xZ+Q8Y>w
zxGXI3s5V&A=ow7Se2s9k(fc$Eo
zR{Z6ogm4udO4CuV30}Sih$PeuhhOdkOOLpJ`%R2js;WdUmGn|LQWW_4(Zr%WyXHK4
zvb_1$wAAL4dx`r4Gei!PE^$n*EJ9D8u-E`SMpUAif++R%u#AY(ma<*&PMKz~)T4cD#%!!;W%vb#(
zL4O80H=%QK!>+rl;-@J3L7}Y*;d;uXC+XJUqaTp|jZy&IrT~dS5vJ~03xG;&xWIfn
z-_H6w@r2Ef!W`$m@hC;|Ul(Zp#=_S4L*
zeW561Qp@P0-L|098PWy-D%4`t_rkOx%6YUlW8(@PkW90l6wfd$<*vZz0FVJE2ro3Y
zwaWLo1Kgha`;&UrGXlpSV3qW610V`+xvHVoJ-E5U3RM@wS
zinHvADP{s^<~t^LG>zpS$-y_OB(kw$E-xR$Idmk@MfV2+0r@jVl0H3q+0{DkY$n9N
zl5y_ii3`A%B4*Pil2D3>wEMPM*{gR%n|dfkd_fu_PPFNxth_m?eJdJ(EfDCyV2<)7
z#pXufCW!B`f1)-ebA-w=&D|3R;w|xB?;g(S#N#ilst;itPal47^=mYw{kYf#y0BP7%$Fh05ql4}O@{J+NM#d;snl4t{
zigBv6N$7|jGNIaC7(*;X5D5i#{!KZ0NCz;r5K(Z(z-i8duz*R^F;%6p<-QNRTE9!p
zzB!dUjL?4Wp$2RIMk%
z!v5?@P&c~NLEv(21^6Xh?TC%|SD@FTsuV;oa(B9-IFkY44okbRjD6u0P>cW$33QAI
z*Z_{C_;9Vz{tKSgWuya>Ju;y{fwMtCGd|7t^M>zuSe9^G(uyTFRTi0a?}D1=cQO;p
zQNa@_#s?wk^(IjFX2%_a65KsI_T&8W6|n4Vm6k~6L=`vKuyUt-db#$JRk9!vUv9AKxxA)DJH2jIbgG3&LzVNbk1+$W?-|g
zHApPAsu)KQbgHEB!2z%gKdT7xY6=+7yJfQt#S5h!K>CKgHxsJ{U)t
zw^E?oLNRoGo~gjD|CU6B`ao7Xc0`Z}f-=Q4m8|1Y+1Zg9R8iLJx!w58PFTV+ExTN1
zttN2sgngfF6F|&-iaSQsQOgwlFyAZgw*8MX!;GO8%oG?<8(9ttx%7em0xF^
zAr76
zSItK!Z`Etng^LERp8`?I)(a~-@?@JN4@2(0yO>D#hgp;3uDcOMni>d&9i|ZtFrv#Z
zXnr3k8dvQlKBAg-$+Uc)9dFjC_NwW{v30YOwtYh!=^*<4e1vM^OA5&-`iZ#=n
zAR^E7=b2BpA9EXlt(5i@q$;zoc|6)JOjQ2s=SSq5j6i?iI9dt+#@&!g{eI8-O>{R%
zvaJHEqhysr>32G%1vBN5)?rN+;;{H=9kS)$y%s5czY{x?>k
z0!ceJNRcB%Hx@zZEsBS(g?SPjmjcXyj*|desGk4%SMA~H|K1`qqZ5C(_zvBdCTA_#
z%!BjXEaenke+7MAcP=y>x;y?vHI>Fr{QPq%mCQ+E`&Sb__ZZM-T0BgpGa$cozV4re
z|C3loC|RFoO7}LNNFQ~E
zw@E9IfWKaQRz_ldQ5_x@iVsOQ{_5Crmk}j|*X-loFJ9BNnlryO0UR{}x-K+)uIu3#
zSYt%u1#TJK-BYZMi%lH?1Dx7mYwv34r#6-G5Cnf@;$o-3GozHOr@$a`UOig
z=El#TLk=B=I;bB@4Gd^t&c7W*kpOpUrp}W|^sm&5VNoTv>QP2inUFJPi!d9~6iPDt
zMGMZn;VH`hf0SK?v7k>(USzX8x*lmY{)`%{$)N#qR+e54+@IsoSI&o%;(a?l;=@_SkDSE*JAOm{E3KFk*>
zbAOGbgAj|$%`LBcOG=)237V}8dwjn4)%(SWpsZT_N}tWvaWN?Rs0|8hcG^@x5aUGR?LCg>4({=!S7D^>G7w9}yRir5(W$@8{*
z_mS;)RY3_N`wK;$PFX(z=Ca$?ac>uDRj4duDctvDBHR+}{oB)wzTL)R(C@-D0Sh}5
zLkN{AJV7Yud4)=fu^2{Sl0sdpR9xN(P}ie1zU5@>$oVc&5TYF@mMq!C`}Lj9c=3W?
zCN>~qP;HiBMui}mkXRyWyKV{|dZSMY)n
zfY4q_wYaVgkq`yqC$#V(A-jNSuA8poW^xM`kUM_Z4S}jqw~aj^*IQ*&=U(})_X^;9
z#hvxV^!&CEDcf$`4iABad6yP&qyQ+`yRkG$NW)x-2C{oD=ZP^pEuLCjw9jd-z40Gu
zIE(_PI@cb}N_~BCbD69$LrAAfnGhatF7qt}`mL%mo%a~rcI>ZuNe5>}3{%<76t{0A
zXnC42==pCps?ji};5Q+KpG;)t$0UFNMS;#Y1Xq%Ts;{Tr{lEP*Qfn_L0|JW{9-R6>)spE>3oMQh>ogt@RlTb@NFQD=2j))>l`?h{TM`
zM7Ual1YDIYsR(J!?%KA^ftTrMr#+1zOPm#Sj57~|?vWBpA={gdjr9nlryKaT=FP@s
zIpFx(*oOx1>ssUA3g+qn<~Vn|xQ{}GV8%yjoojk5Dro~qM<$~%r4@(^Gr(VPzy+3YCDa2|K^Wg^dK?|%
zW`75ue(ze?%=k~nDx{IgmNIK0lvOS6PpI<_Ndah$?2KULKbKO)9JOm$+FWWYDExmw
z%q>M!R{MY&v=&-+r6>o8+E@FzmCbx~1eN*qO?s)7Fh^j$BH(@HZ7%xSz&@Vw<#%@y
z3&`rX&+~n?-Q2h|Qrn{k4=LSu&}y`3mw%!9~_Co
zjY5BwP9Wf)xpWm|ykgHsobg#v=Z+gzJTbhKH(A7S=c}QIv`ScaDtFKGJI@U*_Wgbj
zQ%qdobhgS}{i_6ddZ`-Mc}Ep}R=o1P>*&wo$Jl57j1SnLY?&w>{f{=Z&radasmTuF
zhd9NPMX#IQbNKOvKBnz?z4QjZ$Zw9;6EP4(^6JFTcCrBv!`yBbre)VR*`)Mu%`5MuS{GH&AnL8~X}FsGXTu
zL_F~CB89ev*2O-LnExbFaAM~%t;Be4MOs_s1d9l=<~B`29N3c5)dTZ>yy_0cmvRmY05>EZ!T)Dq=Jxeajmy_tgNi)
zNQXa>y4gI=KO}2#y);}M{PiDlC5c0-bSIkfYqrVc-xPy~Hm78PQA=yn0gyGCbCZ?i
zCG!#GazPxEgIGs`u&^$1*KOk9KwW0{bBXf=xlVYERvcjj5|(ca7R@Q8ttc}F#R~L;
zN+WWNKZqc^r2Y6=$$M$h#OAOY7hrDdzlDr!w$CC*$BUl%P|MNR+Gm_R2dpaF;znM>
z5)yerScR=MPy?@Vm5xrjWWYV7Xo*IB>lFJ^E#=}lO%HARxGunmojv2k;8TqzbvtAe
z)EJB=gKUCc$~=50qCs`p(V6_|)ug_A^aqppC=@0WvpM3k$uFtfszz$gPufcIuXVz<
zMDAcG52_A#)!1zG%UcX$`!}jWB#;Y0E*K!-h^|(kyMv)(88(K&gAyoEMbXRNBOg^V_RpPqwqJ6P&W=?@dIClZ7Hl1eJNo3xo0&x
zF(LL_-!C`g)&Er%-vn&HPKQxKt;R~-AfrmM;9YbaMQtUQLK{GY1XPp&-`OO?1V0Q45{Q$Rv)|H=`{yRi(
zZ_UGbL<}Je?XxAmL1MyQSM}TI#KYk0d_R`!PRsXR=P*Z;Meq1a!3#Q@Rv@XY;8A#o
zLLwJf;=z_r*|WpDulX|2MLTfVZYpss*>Ev
z{QnttBU*WyPCh|iU)v4YcN2DHHq8kixuyclnBME?Y}LV;FG8932`mI(%2-_tMHe97
zSa;7nv{Ik@fd?=Y7!|4!AwI9nnc)3uwXP~_B^DaUhN8vh+{H=yrL{WUsyo>Q&1Br2
zhSjAD>CVDBM29(z%H2ZVsBZ&-FG-f-5AiB;D>qSk79@qm2_A(&UOUX*IF`^W9Vx9?
z8d0K+D2_LfOs5=C>w@V~&|_Y&N+nA%Xo{47_>?`y8cySxR-Crk70~5k3u!He;!^e-
zK&6O#0|MPUYl7m~w4y4U+JS5Glk^)i9uh6l!fsN9P_7H44e?ajegoKABEAe#^c1FZ
z2Bpp}P_bGAz1oY-)y1Uc*6w_@q{Js1fj7{o9$@TdiZI%XB