From cab6108db22272596b81ae7b5b62760e19bd383e Mon Sep 17 00:00:00 2001
From: "Jangberry (Nomad-Debian)" <matt2001@hotmail.fr>
Date: Thu, 9 Nov 2023 14:58:54 +0100
Subject: [PATCH] Knn while I think about commiting
---
Artificial Neural Network.ipynb | 40 +++++
README-tutorial.md | 2 +-
README.md | 279 +-------------------------------
knn.py | 82 ++++++++++
read_cifar.py | 36 +++++
results/knn.png | Bin 0 -> 21241 bytes
6 files changed, 167 insertions(+), 272 deletions(-)
create mode 100644 Artificial Neural Network.ipynb
create mode 100644 knn.py
create mode 100644 results/knn.png
diff --git a/Artificial Neural Network.ipynb b/Artificial Neural Network.ipynb
new file mode 100644
index 0000000..1f366ba
--- /dev/null
+++ b/Artificial Neural Network.ipynb
@@ -0,0 +1,40 @@
+{
+ "cells": [
+ {
+ "cell_type": "code",
+ "execution_count": 2,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import read_cifar\n",
+ "import numpy as np"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": []
+ }
+ ],
+ "metadata": {
+ "kernelspec": {
+ "display_name": ".venv",
+ "language": "python",
+ "name": "python3"
+ },
+ "language_info": {
+ "codemirror_mode": {
+ "name": "ipython",
+ "version": 3
+ },
+ "file_extension": ".py",
+ "mimetype": "text/x-python",
+ "name": "python",
+ "nbconvert_exporter": "python",
+ "pygments_lexer": "ipython3",
+ "version": "3.10.12"
+ }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
diff --git a/README-tutorial.md b/README-tutorial.md
index abc9128..a1592f0 100644
--- a/README-tutorial.md
+++ b/README-tutorial.md
@@ -229,7 +229,7 @@ Experiments will have to be carried out by studying the following variations:
This work must be done individually. The expected output is a repository on <https://gitlab.ec-lyon.fr>. It must contain the complete, minimal and functional code corresponding to this tutorial.
To see the details of the expected, see the Evaluation section.
-The last commit is due before 11:59 pm on Sunday, November 6, 2022. Subsequent commits will not be considered.
+The last commit is due before 11:59 pm on Friday, November 10, 2023. Subsequent commits will not be considered.
## Evaluation
diff --git a/README.md b/README.md
index 27a0e3d..fbaab16 100644
--- a/README.md
+++ b/README.md
@@ -1,275 +1,12 @@
-# Tutorial 1 : Image Classification
+# Image-classification
-MOD 4.6 Deep Learning & Artificial Intelligence: an introduction
+## Setup
+1. Download the [CIFAR dataset](https://www.cs.toronto.edu/~kriz/cifar.html) `cifar-10-batches-py` in the [data](./data/) folder.
+1. You might want to create a venv with `python -m venv .venv` and activate it with `source .venv/bin/activate`.
+1. Install the requirements using `pip install -r requirements.txt`.
-## Introduction
+## Usage
-The objective of this tutorial is to write a complete image classification program in Python.
-Two classification models will be successively developed and tested: k-nearest neighbors (KNN) and neural networks (NN).
-
-### Before you start
-
-In this tutorial we use Python 3.7 or higher. Make sure it is properly installed. Make sure `numpy` is installed.
-
-We assume that `git` is installed, and that you are familiar with the basic `git` commands. (Optionnaly, you can use GitHub Desktop.)
-We also assume that you have access to the [ECL GitLab](https://gitlab.ec-lyon.fr/). If necessary, please consult [this tutorial](https://gitlab.ec-lyon.fr/edelland/inf_tc2/-/blob/main/Tutoriel_gitlab/tutoriel_gitlab.md).
-
-**Important note**: In this tutorial, you will be asked to write functions. If you want to test the correct operation of your functions in the same file, you must use the statement `if __name__ == "__main__":`:
-
-```python
-def func():
- ...
-
-if __name__ == "__main__":
- func()
-```
-
-### README
-
-Your repository must contain a text file `README.md` that introduces the project in a **short way**. To learn how to write a readme, visit https://www.makeareadme.com. We recommend that you include at least the sections _Description_ and _Usage_.
-
-
-### Code style
-
-Your code must follow the [PEP8 recommendations](https://peps.python.org/pep-0008/). To help you format your code properly, you can use [Black](https://black.readthedocs.io/en/stable/). To help you sort your imports, you can use [isort](https://pycqa.github.io/isort/).
-
-
-### Docstring
-
-Your code must be properly documented. It must follow the [PEP257 recommendations](https://peps.python.org/pep-0257/). To help you document your code properly, you can use [pydocstyle](http://www.pydocstyle.org/en/stable/).
-
-
-### Prepare your directory
-
-1. Create a new blank project on the ECL GitLab (`New project` then `Create blank project`).
-2. Fill in the form as follows.
- - Project name: `Image classification`.
- - Project slug: `image-classification`.
- - Visibility Level: public
- - Project Configuration: Initialize repository with a README
-3. Clone the repository.
- ```bash
- git clone https://gitlab.ec-lyon.fr/<user>/image-classification
- ```
-4. In this tutorial you will use files that should not be pushed to the remote repository. To ignore them when committing, you can put their path in a file named `.gitignore`. For simplicity, we use the [`.gitignore` file](https://github.com/github/gitignore/blob/main/Python.gitignore) recommended by GitHub for Python projects.
-
-
-## Prepare the CIFAR dataset
-
-The image database used for the experiments is CIFAR-10 which consists of 60 000 color images of size 32x32 divided into 10 classes (plane, car, bird, cat, ...).
-This database can be obtained at the address https://www.cs.toronto.edu/~kriz/cifar.html where are also given the indications to read the data.
-
-1. Create a folder named `data` in which you move the downloaded `cifar-10-batches-py` folder. Make sure that the `data` folder is ignored when commiting.
-2. Create a Python file named `read_cifar.py`. Write the function `read_cifar_batch` taking as parameter the path of a single batch as a string, and returning:
- - a matrix `data` of size (`batch_size` x `data_size`) where `batch_size` is the number of available data in the batch, and `data_size` the dimension of these data (number of numerical values describing the data), and
- - a vector `labels` of size `batch_size` whose values correspond to the class code of the data of the same index in `data`.
- `data` must be `np.float32` array and `labels` must be `np.int64` array.
-3. Write the function `read_cifar` taking as parameter the path of the directory containing the six batches (five `data_batch` and one `test_batch`) as a string, and returning
- - a matrix `data` of shape (`batch_size` x `data_size`) where `batch_size` is the number of available data in all batches (including `test_batch`), and
- - a vector `labels` of size `batch_size` whose values correspond to the class code of the data of the same index in `data`.
- `data` must be `np.float32` array and `labels` must be `np.int64` array.
-4. Write the function `split_dataset` which splits the dataset into a training set and a test set. The data must be shuffled, so that two successive calls shouldn't give the same output. This function takes as parameter
- - `data` and `labels`, two arrays that have the same size in the first dimension.
- - `split`, a float between 0 and 1 which determines the split factor of the training set with respect to the test set.
- This function must return
- - `data_train` the training data,
- - `labels_train` the corresponding labels,
- - `data_test` the testing data, and
- - `labels_test` the corresponding labels.
-
-## k-nearest neighbors
-
-1. Create a Python fil named `knn.py`. Write the function `distance_matrix` taking as parameters two matrices and returns `dists`, the L2 Euclidean distance matrix. The computation must be done only with matrix manipulation (no loops).
- Hint: $`(a-b)^2 = a^2 + b^2 - 2 ab`$
-2. Write the function `knn_predict` taking as parameters:
- - `dists` the distance matrix between the train set and the test set,
- - `labels_train` the training labels, and
- - `k` the number of of neighbors.
- This function must return the predicted labels for the elements of `data_test`.
-
- **Note:** if the memory occupation is too important, you can use several batches for the calculation of the distance matrix (loop on sub-batches of test data).
-3. Write the function `evaluate_knn` taking as parameters:
- - `data_train` the training data,
- - `labels_train` the corresponding labels,
- - `data_test` the testing data,
- - `labels_test` the corresponding labels, and
- - `k` the number of of neighbors.
- This function must return the classification rate (accuracy).
-4. For `split_factor=0.9`, plot the variation of the accuracy as a function of `k` (from 1 to 20). Save the plot as an image named `knn.png` in the directory `results`.
-
-## Artificial Neural Network
-
-The objective here is to develop a classifier based on a multilayer perceptron (MLP) neural network.
-
-First of all, let's focus on the backpropagation of the gradient with an example. If you still have trouble understanding the intuition behind the back propagation of the gradient, check out this video: [3Blue1Brown/Backpropagation calculus | Chapter 4, Deep learning](https://www.youtube.com/watch?v=tIeHLnjs5U8).
-
-
-The weight matrix of the layer $`L`$ is denoted $`W^{(L)}`$. The bias vector of the layer $`L`$ is denoted $`B^{(L)}`$. We choose the sigmoid function, denoted $`\sigma`$, as the activation function. The output vector of the layer $`L`$ before activation is denoted $`Z^{(L)}`$. The output vector of the layer $`L`$ after activation is denoted $`A^{(L)}`$. By convention, we note $`A^{(0)}`$ the network input vector. Thus $`Z^{(L+1)} = W^{(L+1)}A^{(L)} + B^{(L+1)}`$ and $`A^{(L+1)} = \sigma\left(Z^{(L+1)}\right)`$. Let's consider a network with one hidden layer. Thus, the output is $`\hat{Y} = A^{(2)}`$.
-Let $`Y`$ be the target (desired output). We use mean squared error (MSE) as the cost function. Thus, the cost is $`C = \frac{1}{N_{out}}\sum_{i=1}^{N_{out}} (\hat{y_i} - y_i)^2`$.
-
-1. Prove that $`\sigma' = \sigma \times (1-\sigma)`$
-2. Express $`\frac{\partial C}{\partial A^{(2)}}`$, i.e. the vector of $`\frac{\partial C}{\partial a^{(2)}_i}`$ as a function of $`A^{(2)}`$ and $`Y`$.
-3. Using the chaining rule, express $`\frac{\partial C}{\partial Z^{(2)}}`$, i.e. the vector of $`\frac{\partial C}{\partial z^{(2)}_i}`$ as a function of $`\frac{\partial C}{\partial A^{(2)}}`$ and $`A^{(2)}`$.
-4. Similarly, express $`\frac{\partial C}{\partial W^{(2)}}`$, i.e. the matrix of $`\frac{\partial C}{\partial w^{(2)}_{i,j}}`$ as a function of $`\frac{\partial C}{\partial Z^{(2)}}`$ and $`A^{(1)}`$.
-5. Similarly, express $`\frac{\partial C}{\partial B^{(2)}}`$ as a function of $`\frac{\partial C}{\partial Z^{(2)}}`$.
-6. Similarly, express $`\frac{\partial C}{\partial A^{(1)}}`$ as a function of $`\frac{\partial C}{\partial Z^{(2)}}`$ and $`W^{(2)}`$.
-7. Similarly, express $`\frac{\partial C}{\partial Z^{(1)}}`$ as a function of $`\frac{\partial C}{\partial A^{(1)}}`$ and $`A^{(1)}`$.
-8. Similarly, express $`\frac{\partial C}{\partial W^{(1)}}`$ as a function of $`\frac{\partial C}{\partial Z^{(1)}}`$ and $`A^{(0)}`$.
-9. Similarly, express $`\frac{\partial C}{\partial B^{(1)}}`$ as a function of $`\frac{\partial C}{\partial Z^{(1)}}`$.
-
-Below is a Python code performing a forward pass and computing the cost in a network containing one hidden layer for the regression task. It uses the sigmoid function as the activation function:
-
-```python
-import numpy as np
-
-N = 30 # number of input data
-d_in = 3 # input dimension
-d_h = 3 # number of neurons in the hidden layer
-d_out = 2 # output dimension (number of neurons of the output layer)
-
-# Random initialization of the network weights and biaises
-w1 = 2 * np.random.rand(d_in, d_h) - 1 # first layer weights
-b1 = np.zeros((1, d_h)) # first layer biaises
-w2 = 2 * np.random.rand(d_h, d_out) - 1 # second layer weights
-b2 = np.zeros((1, d_out)) # second layer biaises
-
-data = np.random.rand(N, d_in) # create a random data
-targets = np.random.rand(N, d_out) # create a random targets
-
-# Forward pass
-a0 = data # the data are the input of the first layer
-z1 = np.matmul(a0, w1) + b1 # input of the hidden layer
-a1 = 1 / (1 + np.exp(-z1)) # output of the hidden layer (sigmoid activation function)
-z2 = np.matmul(a1, w2) + b2 # input of the output layer
-a2 = 1 / (1 + np.exp(-z2)) # output of the output layer (sigmoid activation function)
-predictions = a2 # the predicted values are the outputs of the output layer
-
-# Compute loss (MSE)
-loss = np.mean(np.square(predictions - targets))
-print(loss)
-```
-
-10. Create a Python file named `mlp.py`. Use the above code to write the function `learn_once_mse` taking as parameters:
- - `w1`, `b1`, `w2` and `b2` the weights and biases of the network,
- - `data` a matrix of shape (`batch_size` x `d_in`),
- - `targets` a matrix of shape (`batch_size` x `d_out`),
- - `learning_rate` the learning rate,
-
- that perform one gradient descent step, and returns:
- - `w1`, `b1`, `w2` and `b2` the updated weights and biases of the network,
- - `loss` the loss, for monitoring purpose.
-
-
-MSE loss is not well suited for a classification task. Instead, we want to use the binary cross-entropy loss. To use this loss, we need the target to be is the one-hot encoding of the desired labels. Example:
-```
-one_hot(labels=[1 2 0]) = [[0 1 0]
- [0 0 1]
- [1 0 0]]
-```
-
-We also need that the last activation layer of the network to be a softmax layer.
-
-11. Write the function `one_hot` taking a (n)-D array as parameters and returning the corresponding (n+1)-D one-hot matrix.
-12. Write the function `learn_once_cross_entropy` taking as parameters:
- - `w1`, `b1`, `w2` and `b2` the weights and biases of the network,
- - `data` a matrix of shape (`batch_size` x `d_in`),
- - `labels_train` a vector of size `batch_size`, and
- - `learning_rate` the learning rate,
-
- that perform one gradient descent step using a binary cross-entropy loss.
- We admit that $`\frac{\partial C}{\partial Z^{(2)}} = A^{(2)} - Y`$, where $`Y`$ is a one-hot vector encoding the label.
- The function must return:
- - `w1`, `b1`, `w2` and `b2` the updated weights and biases of the network,
- - `loss` the loss, for monitoring purpose.
-13. Write the function `train_mlp` taking as parameters:
- - `w1`, `b1`, `w2` and `b2` the weights and biases of the network,
- - `data_train` a matrix of shape (`batch_size` x `d_in`),
- - `labels_train` a vector of size `batch_size`,
- - `learning_rate` the learning rate, and
- - `num_epoch` the number of training epoch,
-
- that perform `num_epoch` of training steps and returns:
- - `w1`, `b1`, `w2` and `b2` the updated weights and biases of the network,
- - `train_accuracies` the list of train accuracies across epochs as a list of floats.
-14. Write the function `test_mlp` taking as parameters:
- - `w1`, `b1`, `w2` and `b2` the weights and biases of the network,
- - `data_test` a matrix of shape (`batch_size` x `d_in`), and
- - `labels_test` a vector of size `batch_size`,
-
- testing the network on the test set and returns:
- - `test_accuracy` the testing accuracy.
-15. Write the function `run_mlp_training` taking as parameter:
- - `data_train`, `labels_train`, `data_test`, `labels_test`, the training and testing data,
- - `d_h` the number of neurons in the hidden layer
- - `learning_rate` the learning rate, and
- - `num_epoch` the number of training epoch,
-
- that train an MLP classifier and return the training accuracies across epochs as a list of floats and the final testing accuracy as a float.
-16. For `split_factor=0.9`, `d_h=64`, `learning_rate=0.1` and `num_epoch=100`, plot the evolution of learning accuracy across learning epochs. Save the graph as an image named `mlp.png` in the `results` directory.
-
-
-## To go further
-
-### Unittest
-
-Your code should contain unit tests. All unit tests should be contained in the `tests` directory located at the root of the directory.
-We choose to use [pytest](https://docs.pytest.org/en/7.1.x/). To help you write unit tests, you can consult the pytest documentation.
-
-### License
-
-Your project should be properly licensed. Since it is your project, it is up to you to choose your license. In general, the license consists of a file named LICENSE in the root directory. A useful resource to help you choose: https://choosealicense.com/
-
-### Deep dive into the classifier
-
-Experiments will have to be carried out by studying the following variations:
-- use image representation by descriptors (LBP, HOG, ...) instead of raw pixels using the `scikit-image` module.
-- use of N-fold cross-validation instead of a fixed learning and testing subset.
-
-
-## To be handed in
-
-This work must be done individually. The expected output is a repository on https://gitlab.ec-lyon.fr. It must contain the complete, minimal and functional code corresponding to this tutorial.
-To see the details of the expected, see the Evaluation section.
-
-The last commit is due before 11:59 pm on Sunday, November 6, 2022. Subsequent commits will not be considered.
-
-
-## Evaluation
-
-In this section, we present all the items on which the work is evaluated.
-
-- ( /1) The function `read_cifar_batch` works as described
-- ( /1) The function `read_cifar` works as described
-- ( /1) The `split_dataset` works as described
-- ( /1) The function `distance_matrix` works as described
-- ( /1) The function `knn_predict` works as described
-- ( /1) The function `evaluate_knn` works as described
-- ( /1) The graph `knn.png` shows the results obtained
-- ( /1) The function `learn_once_mse` works as described
-- ( /1) The function `one_hot` works as described
-- ( /1) The function `learn_once_cross_entropy` works as described
-- ( /1) The function `train_mlp` works as described
-- ( /1) The function `test_mlp` works as described
-- ( /1) The function `run_mlp_training` works as described
-- ( /1) The graph `mlp.png` shows the results obtained
-- ( /1) The project has a good README.
-- ( /2) The guidlines about the project structure are all followed
-- ( /1) All functions are documented
-- ( /1) All functions are documented clearly and follow the PEP257 recommendations
-- ( /1) The code is properly formatted, i.e. follow the PEP8 recommendations
-
-**Bonus**
-
-- ( /1) Project has a license
-- ( /2) Unitest coverage
-- ( /2) Deep dive into the classifier
-
-
-## Contribute
-
-This tutorial may contain errors, inaccuracies, typos or areas for improvement. Feel free to contribute to its improvement by [opening an issue](https://gitlab.ec-lyon.fr/qgalloue/image_classification_instructions/-/issues).
-
-## Author
-
-- Quentin Gallouédec
+To test knn, simply run [knn.py](./knn.py) using `python knn.py`.
+Otherwise here is a test result: 
diff --git a/knn.py b/knn.py
new file mode 100644
index 0000000..9d67d29
--- /dev/null
+++ b/knn.py
@@ -0,0 +1,82 @@
+import read_cifar
+import numpy as np
+import matplotlib.pyplot as plt
+
+def distance_matrix(X: np.array, Y: np.array):
+ """Compute the L2 distance between two matricies
+
+ Arguments:
+ X -- first matrix
+ Y -- second matrix
+ Returns:
+ dist -- distance matrix (shape: (X.shape[0], Y.shape[0])) => dist[i, j] = L2(X[i], Y[j])
+ """
+ return np.sqrt(
+ np.sum(np.square(X), axis=1).reshape((-1, 1))@np.ones((1, Y.shape[0])) + # X²
+ np.ones((X.shape[0], 1))@np.sum(np.square(Y), axis=1).reshape((1, -1)) # Y²
+ - 2*X@Y.T) # -2XY
+
+def knn_predict(dist: np.array, labels_train: np.array, k: int):
+ """Predict the labels of the test set using the k-nearest neighbors algorithm
+
+ Arguments:
+ dist -- distance matrix
+ labels_train -- labels of the training set
+ k -- number of neighbors
+ Returns:
+ labels -- predicted labels of the test set
+ """
+ # Get the indices of the k nearest neighbors
+ indices = np.argsort(dist, axis=1)[:, :k]
+ # Get the labels of the k nearest neighbors
+ labels = labels_train[indices]
+ # Get the most frequent label
+ labels = np.apply_along_axis(lambda x: np.bincount(x).argmax(), axis=1, arr=labels)
+
+ return labels
+
+def evaluate_knn(data_train: np.array, labels_train: np.array, data_test: np.array, labels_test: np.array, k: int, dist = None):
+ """Evaluate the k-nearest neighbors algorithm
+
+ Arguments:
+ data_train -- training set
+ labels_train -- labels of the training set
+ data_test -- test set
+ labels_test -- labels of the test set
+ k -- number of neighbors
+ dist -- the distance matrix (if not given, will be calculated)
+ Returns:
+ accuracy -- accuracy of the k-nearest neighbors algorithm
+ """
+ if dist is None:
+ dist = distance_matrix(data_test, data_train)
+ prediction = knn_predict(dist, labels_train, k)
+ accuracy = (prediction == labels_test).mean()
+ return accuracy
+
+
+if __name__ == "__main__":
+ print("Reading data")
+ images, labels = read_cifar.read_cifar("data/cifar-10-batches-py/")
+ split_factor = 0.9
+
+ # Split the data into training and testing sets
+ print("Splitting sets")
+ images_train, labels_train, images_test, labels_test = read_cifar.split_dataset(images, labels, split_factor)
+
+ # Compute the distance matrix
+ print("Computing the distance matrix...")
+ dist = distance_matrix(images_test, images_train)
+
+ # Evaluate the k-nearest neighbors algorithm
+ accuracies = [] # List of the accuracy
+ ks = [] # List to make sure the plot starts at one and not 0
+ for k in range(1, 21):
+ accuracy = evaluate_knn(images_train, labels_train, images_test, labels_test, k)
+ print(f"Accuracy for k = {k}: {accuracy}")
+ accuracies.append(accuracy)
+ ks.append(k)
+ fig = plt.figure()
+ plt.plot(ks, accuracies)
+ fig.savefig(fname="results/knn.png")
+ plt.show()
\ No newline at end of file
diff --git a/read_cifar.py b/read_cifar.py
index 86d78f2..29fa7c1 100644
--- a/read_cifar.py
+++ b/read_cifar.py
@@ -75,6 +75,41 @@ def split_dataset(data: np.array, labels: np.array, split: float):
return data_train, labels_train, data_test, labels_test
+def test_split_dataset():
+ cifar_data, cifar_labels = read_cifar("data/cifar-10-batches-py/")
+ split = 0.8
+
+ datapoint_size = cifar_data.shape[1]
+
+ data_train, labels_train, data_test, labels_test = split_dataset(
+ cifar_data, cifar_labels, split
+ )
+
+ expected_train_size = round(split * cifar_data.shape[0])
+ expected_test_size = cifar_data.shape[0] - expected_train_size
+
+ # check dimensions
+ assert data_train.shape == (expected_train_size, datapoint_size)
+ assert labels_train.shape == (expected_train_size,)
+ assert data_test.shape == (expected_test_size, datapoint_size)
+ assert labels_test.shape == (expected_test_size,)
+
+ # check types
+ assert data_train.dtype == cifar_data.dtype
+ assert labels_train.dtype == cifar_labels.dtype
+ assert data_test.dtype == cifar_data.dtype
+ assert labels_test.dtype == cifar_labels.dtype
+
+ # check that the first data and label are still in the dataset and match indices
+ # pick random index in the training set
+ i = np.random.randint(0, expected_train_size)
+ random_data = data_train[i]
+ random_label = labels_train[i]
+ # check that the data and label match in original dataset
+ original_index = np.where(np.all(cifar_data == random_data, axis=1))[0][0]
+ assert cifar_labels[original_index] == random_label
+
+
if __name__ == "__main__":
images, labels = read_cifar("data/cifar-10-batches-py/")
print(images.shape)
@@ -88,4 +123,5 @@ if __name__ == "__main__":
print(labels_train[0])
print(data_test[0])
print(labels_test[0])
+ test_split_dataset()
\ No newline at end of file
diff --git a/results/knn.png b/results/knn.png
new file mode 100644
index 0000000000000000000000000000000000000000..d8c38a4c3a32574ca5e317d4803aa69e48db53ff
GIT binary patch
literal 21241
zcmeAS@N?(olHy`uVBq!ia0y~yU}|7sV0^&A#=yW}dhyN^1_lPp64!{5;QX|b^2DN4
z2H(Vzf}H%4oXjMJvecsD%=|oKJ##%n9fgdNl7eC@ef?ax0=@jAbp6|09PJDY44efX
zk;M!Q{D~mUxWayUCIdrInx~6nNX4ADcPlGGzTW-!u{v#&(4?6T9!!dFycrrf8auSJ
zSf=a>-}v^i)9SU^xf|C;t$vZ5n<ZXf;`sK_w}kA#ZttmG60VMc6B-&EB2-%@@NSyy
zGim1M-^w#NC))ZL6+hb(e|_revro><JpVDp$o&4*G>JAv1_p+RHrpGl3=9klss$Ms
z85nfh7&sUh5)4@!7#JA36&aWq7$SHW1sE6(Bylt_FffQYGq5l)Y>;45U|?uCDln>r
z*btl1kszLs7Pmh9=+UEH{r%I=rg4{9GB7+~k!V|Z>sHj)*Vp~acAMTl#L=*P!Hyjf
zFJHbqsK?B}P;zLVZS|G?_5Ty53yQED2o{i+*I)TnTtiV&@x)aIh6drl;NZnyzLflb
zzC%=y!M}0+ZF5D%Mg|UshLaJMKc<;l?*m0@50xU-faP!bVY|D3*Ke(Q`q}oN9xDSw
ziy~73$E~QS<>uE!v)|5Jl`x%&f#Hw<qXEm#Di^OwtN6@VcM35u2=w`gGCV&&f4yTf
zo2HtY+s&Idqqb&U4eXEIk}+}9rcF<Zc8Z9LFF$j}M>k@_g2RU9cS|Pcm_@g*mo!c*
z`I;&dUd@ohSES0opdjPQvb#ZeYr#XOvonp|{q6s5*|d4HY5qMM-}!d2tgNhFetvdu
ze#GsqdfCV>7vSdBwrJ6!FV~{;GrzsL>E-8l?aRx{e~+J?Vd#9`_WPXwzhAs~k+*YR
za^W!{21W*l>xRDE4}|?~7DngoblseO{@RO+i@$!k?7!N*U#{xg&Gi0?KUY=;mp+{u
zo^@qKpp;d~ig){d=k-dP`)P-*c~KBkQnDo|IQVaTXJ_Z#Ld)Y%UJ0stuF6wjU}9)E
zVC>3}la+O8mTC5s$&(+~pUK;<t*-8_9lp-y-Dlpz0daBjxVX4V-tBx|_T%QJ)YDVu
z%vm#0+5OeF+j*D&{r$}x_y1(kQ6n}128I&Z;1b60kdPCPKWeC|E<OBMQ&aQSrqt7c
z^TT6f*KXNjvh!}9r3|0#&nJ_2y;#)!$9}$Ha@)FHue5mOY$B}7-ds3e|F1Z<=Ht<*
zokdG!_>Lz&FW!B(X!qTJ@3XVAm<meUKld`O$^W!s<<ZqEID(&)Ffb$tE#qjID=II)
zzV!9AmFw2|RaREcGEVPn<B?pnGI+Vyq;<RBS^eLbe7x`Kj>5+#B`*SWqqq5p$5lK`
ze17`u*{jO^HkZDZ9(OsFHkoJfR_Pu7H6M<vGBGrKNbFj+bjcDGOBueY)25YG?Afv<
zWaY}0?zQvo=3U>AczDXpnJcT`?R3BY=b8D`*|V#!JZ|TaT(o3~O78uAe>?2=|Nr+p
zG&c6_|67}vXNzT(-N^Oo$joJ3lea)xf`Nf$nkG|1NXQg5zd0-R?6LWu9Jf9k<Zh$X
zQ=IdUzy4bG;?Un+ixw%py}v)c_V+j2N9PwWUOaWegbNQ>uihZS!@%%DYuQVtdB*8|
zd3!#(rKYB4etL3pmU;fXSFc~2R)5R!yu@pMXTg>&CiUF$d3ozLZQ2C#_|Cg|x^a6V
zxW#lX1pC{ndahc%`m%Dr&7#%o_eHH-x$@S=<o3F6o9AcV*pMhYt&>wMykFz(z4{5D
z49axQgt6e-nn*2u{p-&Bwo4{W5=u==do{QG-ba5PO&y&nckkU>5E&o8{&xQUSXWoq
zuHN2IH@CL9nvbq$XPJ6WdQxDqrS5Ol-VcYkSFKtVw7>4JN$xEZ+g~phD{d59P_Jg-
zeUO2nfnl>`TO!MDj?Jr9uAF%JWk_jh>4DwN@Bdwo+PiUG#QqtUUgq<SRTeyFV>obX
z*Qd|7b$krjE_B7bVPj%o*u}wU;9_~ZjCo~}m;nO=hXGSU$0Alxk`!f1==h||e7Noa
z1B1c^jt0f9Qc!*hS7d0@NY_aQv9BvIv}r_vSYXlnDidxU6Wdy`$8jAS14D~2Q$k14
zHdY7ERhc=QNuZ2=M}WaOq~-9sA2Y7a{k4kEh);llfrXFJpylUr#!aC3;ZR^o;3(dD
zmStU<5(fhVn46u?xP%)N`Cx@=O(EAVGjT95FtsyC9Fkz>U|`tA#Awj6wI4~<MUVop
z@Xuomn`C7e7&ufw27k?X%d`ySITj(1rMns=LHe0G86*zHZ0uoVVPI%<XW&Wni|0;o
z-n=JH8C)J0I5O}g?n_PJabA_1!&zj?z@QK#z~CGa(h^%3$mf~E&JJ<Rqd&%tkz8yH
z3@jXs21kB=XHHWC1*ia1LPxQzksu2LgMt7@gW|dUvNExtAaww7Bu`5?Ffa&2upAJ0
zp3j&9jslR5&u`g+e7O}E7&t(@?0iNQaDk)1(6-~7VM7TcGeg4x1%@^YaNsWxWN?m2
zy6vX`@q!ZrPhtBu2B|i|Cs_;(4i+GF_rnV0)-SXw_nxqtk%8f11B1k)BqiCd&YZoz
zI&8cQ3`~s-8BBU*%!~{Sj4li&j9^wHh;`q>0c?qa0E2JCR#03k2r~E{FlXB^>#hR>
zg9C#AgYSX+OdjGO&oVGA;Iw$j#0E8GS|cL^1IU!dBiqCdWP!Pi4h$xY>fq$V05a%4
z>lttmFt9iXXUu24qjK1wDTbGU!N8hn0p|~<gw4k<98~iMJ3Gyt!KCqqNBbg~wcES$
z5*Qc?R&g|_C2YRQH<RO#4~U<|(V+IQD>s^}ZRP<6h6f^o48Dhs<|!~RC^&#*Qp^Py
z7z7kp942SvGp6t;Ffdq{vN%jm0I@hkm=<{MxWUH<j=&2lF}5|g4_;*DVPNp%WxSx0
z0ZtuG3?`i7n;RGz92x`|e48Rc@fZLyxo0UjiLx<XP^r;vcqg`AYPKyn7tD}kyr5F$
z=6)h^yXPWN0S1OSQcMdxZ{(Q$PrA^<%)-F1yon)0$!}T{DC84XF<8|9tKpS0@h~(r
zJaz8ewN;_3`>Nh7a_ycnb*k$0(@S^kut-fy%R1P^>Tdh*T6F%@vuT(2RDRyI;M0Ku
zLqV1oTe_^aUwgAI#oCo&f{5T7Cci6(TDeWr&&k;S{c;&(;GW9QCbhpxL`6h2EG>U-
zcwGJU)lzo33I|C^$*!)hOVRmzm)_i*u5D=<d3m}2`+_&w>-R46xBnaR_t)3g>rTHE
zxu2(SwS<vbJ=MK(Zrb0c!J$`ns(RKwT5PJcba|Wl@$85q|6Y&I>mQ!X^Xlp5Dq6~L
z!PZBZVSdFU&bvjYb!XXDZ&P{sa{2sgxAXU}O`l&I_Wf@8{lw$`bFH=>{$75+c6(gK
zL)Nts8y7h<+}T~e{@t$Edb*LDT#VDty}6Wa5xRw8R`}fAuKDZrubrMRzH-gBK>Pg_
zQ}pC+o8L%{JT#A){p;NQwadD!wkIjeW@UU^vFF;_Pf^?L?(N%uc%!H7Vk?$Ii6=iw
zO}rGmx7v+;Wfj8<sbw4uzP`Ruo6~yV?f+l*Ke=potfi%8)Xt)(h4b?z+V0i=uN4&$
zdGYbM{CE3zU#m+0e!c#Bx&7aj$9g1-?|j~IH*e?Ny#L=#O-<D+RhKWTwNGGLASuzd
z&~&%<)`G8cd-o>J=QG=?f4}&){U_7noOs)!pAxEqdb@l)OBe6^``te@=<<rhleSww
z{7%=fy0q|ZYU!=8sMWdcfARxUo3&oe3JENCnE%e^!b6+Ak7^EX*m|{~qR5}uzcW1O
zyvEI(Z3?ln3?=h4nG#%GTW#d}gMxyrZ01dxv`9C4o6OhBt5>gjd3#U3{dU^nhZiRM
z+pX-@+jZjq=i~DAF@l1EzaH|}S2&eEdG_qpM0dHDcU)N*elyN?j?7&w_VoSL)AQe7
zJol;U|6}J#k4oMyDY<H5@>*a1?Umco-!h+``&;#7li{ypJNB`3GOS)v{Mx$ye@Hc-
zNruCExnIZ6-uyRh&zX|xMweq{`_`9l&v37tyKU;V%<Rau+ooU3jE<c9Eky20<1e+f
zw@MvWmzDCgt&^K58^v{A`l7q@uP<-bd9yV58M`vPS@YvZ#oFlYex;?Qr%s=~Jjb%w
zB;|yF?_8_U>}zW-%K!Vo{`C3t`5)Nd|NVY{y=C#UAN$v@Tp3tZRTVz9ULu|$g(u-f
z=EsV!j*dIU*4<mo&voj_N2y6m<O2E5%G?g$thO}!(E*qL?{6Jl{`IPV{9mVNudLGR
z-q~Nj{i?0{(slWY`n+og99CG}Sn_&)<WkwEUtc}%xoDqvF3ij6jNyD?aoL51e`BSt
z-@p3sW1E_Pwt>#-tJf}FYuK@OP1tUJnUJdo)7Ji56|{Q)swpb2H*$ZyyV+d&z4frD
zxa_t1IK~-sH2*oNs;ZuRSg>NtmMw=LFI~FS%h%WT)#o>5yW4oB)jUs~IdkRN+1XQP
z&tAPUczM>(PfxY9wJ$d?GXFVVZ6kNL_`EHs;j-`7tJR0kiOb5aeevQ!t;6qa&8yce
zXEg|UqqF|<>G|R>@BehwTAwQ~XOP(74f64`9d)lNTNDho?GMhsSGnz3{f5}j_BYm*
z@6)-seD}B45514S_;jXj3){S=TgTGw{<xEsdoO(Pl6e=aK0eD?`u)|#m!FTiS(Zf2
z3SPd{(%R<UjXP32+oxa6%+6f<sXQE%Fn+yuxBmKUSCZ|I-iy0Za<sk+Id~YkGW2wH
zT{_Y!Jazi?<=X3ZI0e1_v3JiNo4b|I=c?Dpt-t<S{&4x<uh+HJ)t4VWzubTRy4UOX
z-?P7{a>annz(gh|>*DR#-(NmG|NTdF-I8mZEaAQ{)@Q4IyZ&LykH_=WwGOS5k2i6;
zd~C^!<?Gj6Ush}S>*Ar(s_fd0a=VM(6h^Np*wLoezwi5@=f9Htr^eVSeeIvUdG{J`
zGqp<#4gb!Oy1d?2V&DFyA;tIq1Xc5yyy&Zc_Il1kPf!vs;b!KrT*lEb$t3-p%-xE|
zy{Ar{y7c7aWKiei>h<ft58r||7@t0S=9MM8e)`m@OBXt~pL$!C{qfP!s0|5?^QvBH
zdIr6}xA*Gq^7nk}1tld{Ubzy|D`UB-TK^r(&ostx`6b*tc1yM`+*-Zr%cobizs`#v
zU6nukLgIGMV0X3Gx7L)zT4+U1?3)}PRqSKG{N9ewI<|Xr+W-CdXSR3yPM_6fwJX~7
zWEZbiUvjajr0I^XTy6cq=TTc<H#}Obyyd~}n^H5Y{wT<Kh^H)hab}@sRnbj8koIZ$
zt1t9ap8neWx$VWbm4`*7pT4+O#qc1*N0=e~?5w4qJ{8%1x!|m&sTr7-w(MTzbJ_n#
zCLB8Cbo1?7`F&r%e%<x=+wIVhkR>;6M5Lyqyf`Wze`QVN=B854<!9_UTf`3J1qVLA
zmYpkXwQcIvpYrlocDJuP>2=*){TsiBckGhG>vrqOtjkOe`g%Eh|Ak+Rvi?4qs{iHM
zucW7OreU>nH!lAp^>EG2$5-e6k6Hcnww8$W%L~Pu)iZPIpMRWxgEdJ>c2@C4j@Z(w
zS<m+UTxxeGYMuSw{juxU-d)|UFS~foMa|>(oD3y~vlAuUHgp@To;6F#a;{%!c(}Hr
z;=<GM>whI|KKB3p?nz6U;?D(?u&m{bU3m2I)la#iQ9G*o@7<1BHEkQGtl`A6Sfg$0
z(^8jae|iuXd(G(jqmQrFpDxe3cPTgPQ-!F!?85k(<SFaUdM!8oz+n^TY#+D7FTXBu
z%?gg)^LA%WIwv(v<jp3Pt=nYue{OQ=|C2uXTd29$q#`%f>~~9({W4>e8iFt1oqvB}
z2SWwduHWJ(UYnYHX*_DglX;a{T+4EvVN&C^tm}+_KP=s<wKDUIy_wzJrIC>@e;xa=
z&#R;J&9<m*)33f-zJA^HTZgnfFIgU*{N?rP<C<q(vlsMV&wlaGReH;Z&$e9~Zhx8n
zTK(0bz=hWjg}u+aD1ZO{vh->1gSxK=l(fW_#~S!-pJUxpe}d7`fkA~~-lT*-OsW6g
z?_OnfW6e@~yYDS(QV)(?-5FhSQCL1JzI3rrwacqhmw#tpTk<7O-n~qt|N4c^+GQ`x
z-;15N`zSPKiu3n)Bd`1OY#hpVZ;fW=J$ym=|IH=O?SFce&Hl1H-}l0q!i9TxpW1dQ
z%J#=GxBpI;ohLo|a^h)n$+-`9j_Y_fh%qfF`J{YJFy!Cc2NLWO1^#<>hyGqAJ@uY;
znEd_nM!8)jB>{)$t607<yIb^5;P3Z$#ZycT0>Apo{eASc_t&}m_Lt`|vo}XYZ}?m0
z**Ne26K&5+`TH#Go*bU9!t398-Fy3-h4wMiSK58Ncq=k+J*ZH({x@-b?N#1`di)Ko
z0t~*#UPg1Z-F0Bfat_=&?|;g4m6y}QdqVqHT$m+qy2RgK_s>BQ+b_KryE?aIKRx$%
zmfd^hy&q17f2vx0y~F;0^IK5f^1QS-e9p3bzc=zvy8PyxO}qSbX6C8g&uz}yWlwlx
z_VO+7?~;e_?XSe!-&&Z~Z+=6SX#)e(f|Dx0wl^nh&Tz>5SW)^v`}I@PC8v*lzg(&8
zwAuRH#cijLOp>3w`pv<;B^ypI-kD#w^7X%OEBCEhVgHZw&=F(pvX|Svxl}yu_p3Nx
zpKoYzLbTM+$!E#4I4igRJ74GcN<6r9P_4giYeje3zrvk%w|Cm#c*@98%;GTl<9b#P
z@qz<??w`7QYsYpsPQ$gEbDi%VxK~*9a+>j%7f+L~D4$vwx|=WKn7rL3W9_n^+ZS+$
z>#x<ivN%6?lFDC=^d0-^-(64$FL`zBOYy6#(dVz^uajFIR~QHyJBqCT8v@D`sat!G
z+cPqhFfm?8i78=Z4=D_7uKS;toqefRyx-{TnXKJ+m%U!ja|z_d&23-9k~fF<@7wBm
z{a@<RvT4_4|Fful`E_gWub<h{DxUZ4%RAofE2x;D%_Xz@?y_w4rLX^1pS!$$Gy93O
z$#K!U!^Qu-a|T68qWl}i5__fvCx3irUdC;q04gV5&9*ruHDkqVwWaL4=S-e<Z*_az
zpSW#xrB5Ti_Vittb$9BnH|OQEe{TDfFkLL*eU8pW+x0byhbOGFPhRkDzlE~Sb(fN_
z9hcUxlUaQ|-`+d<@|;zr+cVzHj(@euetke);?s}*3=HZW3>ijm*n}*+_vu``<hvA<
z0avd+HmfXBboSY0`RV)@CdqH_<6Tp{J>y&P>aVY6ElQX!CUBbTRLOPkpPwptV{I4i
zm)n`+6VbNcx!1q^<yY%-;PliogQGSwX7TEIe_ZbWxLjoW<CojsZQDHJL$7p)?qMhi
z%kV8dQuvc8!8G>3@+If)F4~!2qv`qUV5PH(63?{|pW>rg2fuInx;cFRh1ajUzD^F`
z7f=}5{BBpmj7;Xsr(}PBZD!5B^2X2v)IP1-TG{;TWYLY>UH^I-cm3EB^7q&N=r6Bk
zDdpaYUff%qz2(8>yOWm06$V<_-d%c`eTPY+!=htLcV97{W&fI?uYbRY+T*TVY33Ua
z;vZK2_`91`Yh~t@hB(`j>#Prrr|D=NTBpA5Y)E|gnt4&%oc(p~u6neJ?eC|q%OVdA
z!kg>s5|{4#ekf~a^qyS?%8~!`!<M~VrzkL~%I$Ue^`7gtm)$h#%cp+g*50~pLy+|M
zYcCvXCx6;DMKj-frpn2@$!<%Iy#B)2`7gjiMf04|LdgT2$_%r5F7M~z(QTN!b>9jv
z{Vlo*e{M`&uBm3UXWOGqw{v%uyS4q=nPmFCxA)ieSGHPfGq23wZ|hj7mMQyd(~c-}
z+udI5vu&)hBl6GwKgZ(g8lnG3cGY_0FyrG#e~IrY+g`C{<qo%O?%IVLysB4AZCSb`
zEZQsU>!npro@*bs8Q+fL^p~Ho;!Mf(dL2WC*G9*eGyeH=`hWm~_0pq{um0^eJCW@X
zp7M?RweAj3Dhm=$0wsfA3l_v&|NAxAwm3NK#QBnYVNt6=Mb)fhVyV17SMKMX;MzSw
zj^m<W-?nM{Tx!FcRXtNpdr#@Fxpd2RYSixCUa!B0t=zRRQa(@jdVBP3bv9n_cPlw_
z8$Wzu)Z_K<o8BDebRg&k_YV=4Wc`E8UwvnNcoF1zY1?tVuqCX!kN6$EKmGj`{rmTH
z(sP!4dS~hHX}h>NHfVZubi~Tp@jqgIzn*=?-Y`tRrhHS|x0jM{HuYtE|5p|J`HR=o
ztGi6i&aSM#Gihy$MAT}l)i)nsomRVbZR3WuOAEgLJG@=sjil^CP3P*@8@2tpf7UB7
zG#-dMu!UikpOepT-PM^HVfoxFGaPP4UtOR+EpMvI-TJ~d)kE`D=GrL){@?f3=kAv~
zTD4pMFIZRiO8D>JHl0-7SE6q&S>4#XKkoj&+v!@XOS4nF_WXOldopMJa*a!g*Khpu
z>wWwF3U6?f{O$PX*P`OJ6BDQIyAiz9s=Rwzp4Qak*)L{FZ}zdy<T%i|_~(ffC7H6L
z+glx`HCYz6C1$*nmM-0Ja<ON#P-*V8pV$6ai#AVs^yU5Sb0KNFyhGV%e|@{Ze%b%L
z8{(i!o-h5!)kkM{z3A-UvVE0f^W;Zc9y~h<>i(8xCax?$pYgQE@1p<m^cVBD&sq4D
z-~amlef4dNWXksLU+R-z7bx>ah_Qs3X#wZX85~8Irqx6)`ffH|^HSpF_3O5r3sP!-
zb=Npb|87y6p48k$%f!~_{ZHRFFLI-H#@8LuHoty<nrrpGT>iy+`L`jHkFVBru6iwR
zo_?A4X0J@y>~j*AqRZn9d?tN4QE!%6KiB$1DC3JP76)OkoZEcsW-U$d^T<u!{A}ON
zUC-{Xea?IJgF(dS^_!Cx%T)cUS9|r#nC+Co68HPJKv*T;JkK;uqU~N({F*uHcAzA(
zW>xc9sd;yT&)=J}wEo_twRP)m<gI3Xn`fK9!YJr*>4JTp|8H-*Z`AO?mBB>uz7XFQ
zhIupY@82D|)HILx)m`WEIrcg?p0dn6oOk(uoSobE*g1-xRcS}{?tZ-#WolNv>(AEk
z`cj8*yPuu!{ydL*Yk6UT-N%ck_GYd5SN+{`mx6NZ#&ky!(ZuVI!}}-SRRaxlR?hZr
zjumRqJE9^WT(os7Lzoal|Lbns(5l3z;ggoc#~Dt0>T>Hwj@jL+djVfdkG)V|U$bz#
z`J3R#+{OROoTvT&u05e_%Jr8Y|L0xYJv++TwO^xsU;R6QyG7>&F0b*;V10Z!cGu52
zir^m3foy5h$^!kP#)9k|-dh-E1^kX`C}EWE&Od8^c)~gxTi<kkf37l5#pj>SnC$&<
z_heUp=Y?16wMxGR&(ApDcz8PJ%QanA+vRF%7ccv}zyHg-WkEM{FKt)jzp!b(&(;1S
zH-Udn7y66cG7{d)flAQz|97k2c*^iXiN!(qlRs0U*at(~?;pc&ZCUTOPHy6_qh|sS
z>Zzy9wKVZuf3IrUFU#aDi}%EV!p=-2bp9nv=~*_vKL5UL@^$-ZQ27n2X=Y4ZqiSyz
zzB9Zxw1{c4jZ<jO6z84cy;D{GKHlt9yQ{c<f;Q6yQH}<+BN7K5Yp=3-5&!x9bMgM9
zy+<BP%#>R1x_$oAo+}lhS9Y4dUa$St(eb9|{Ns;JzkGULUGne!w3p@Y^FrtDc1^xK
z=ht)Hn^n)2c}|<xkgS%@=L_=X1eLqr-+#SpWGCYAU5R1Vg!~>c-3IqGi8i^a$KIwU
z)h3k%`mvL*+TT!}>9oq&ZT%f9*K~ewP&+2C?qRWROl-xW^ZyI7cY<0pb>Ci9O}%2G
zbyjpWsC8%l&cZvLZ|U#5Iq&wXIBRlV{FU|J@$dfc+t_x?Ffh6?m?$1&kh@)PKJCR{
z-F>G2{gTsOdUo`8UfFE!dnr`-(vy!;yT0F-{&M8fTB~b)6K8j2m;5!J|MFORc&4}K
zTA?JRFQ2`Qrj(zr++|~Yc2c%pTXV|{58E==?C=lw?QX67>Kb}bm&3t9kioaZu)#h2
z@{Kj^x3B%#o%_mw*L&INk14LMH@^PPKK6Ruw${e~Gq(HMSC{UcWqXhB_}kTa7i{(G
z7tXAYS_kU$yxj1zI-UPoyS;7vefx3^<8_A}&b<(cwc0m-Pg~y)YlZ`RW;CTcy<sUT
z*&iLWU{lm~Hua+~zpZp`j_u35nPYa>ICse}tK_c_q@}-}nx!=Lipi^$|7U6(Hu!R7
zlPoBk?Tn6MK4^aI>(%qi(lZ?1%Q>!-X*BzO>0;EzwK1SdY&w7Wi*wv<?7LMMKqIz_
z={c_@4$L~1R=KB5reve%q-$bfSqGQzR`u+UEtJv?VEpj*4Wn17+v~piHK4xKuB3m*
ztpe^ZVg9=IFuxlAtEIp7PhNP*EYuvMxyvi)e8$<mB|UXp3g-Tsu|3w<&wFXAEC+K5
z<Ln9fVjB{-<Q226{5iq&|GHTdmKnDjzElX%JK=Hhko=s5OXdaMiVIuL{MA?LlDc30
z-nvC%{~quhn$H?D@23Cr{x^%&`olp%YIAo-sY1g#M+TFHD>!ynZPQ)gZC<cCJ^an3
zX_KN<zja*EPwV3=vrJ5nsh=)(w`>~7ruewRS=0W!i1C%4e2PW7d&A9~syC-Lc6AqN
z?BC$UeXHy`V?rE@!{krvSx>kKytTZr&*$iQeaWS9g^l-5@my-SXZz*I)|tQN8?$L$
zIyj$i^~>j7S8Jzl(OdCvwkj**T|+j9?~m8-@cgOP6cSaVIpdn?f3sB^Z-ScjU(R-}
z`N>`~^}UbWf|JcvFB2xa+t=qFHk{7RByfwPLG3Ab!&R~OOmTY_y2y)6(LR-UJvqE@
zskhJzr?TC?`}b9?S)}wPYTsI?|L5*YznR2PBE)!MNldPB<7%BdOc8f4tZD!MIQnz`
ziZ<2nJVmY!Kcp`$UtTxI*lE(27tymr!fs1E>0&5hW4y3rjt<+}`+7GQeTv%NrgHS<
zx6hHE`CsU5pJN=ce5vY|f68leAC@vab7C;jykAxjHkWlzTKUp{A0uDeZS9Y>47(@@
z^2BtG_SpV-TR*<tzTM_iUd-+N-4rX-FxiE{MDxFd1NSxtv+6xn54K)2{lCw0<Njql
zZ)M+1VmK{%`TWZ0^U_M5RUdlwb?xOO4WyYCct%$8Jh(r7`}&{XKW|IbWGLCcTTf!0
zV(p`lxYyT`yUkg5%P^=nGGqjW$udX8TwP%QWb3-0-@9inj{MBev1o=zdtLEYi<>!@
zK*L~<--;g8<2=Br#4u}0KF@-yyJGMCSrpNGHqKXO#){QyOIUZy9Qd?}?dv7pbe<_H
zceiKk<ulvLvQw#K`lefMe;@A6{dL>j`s>ed_A58uaF%&ma0(vOW0?E+)x~S9IUB_e
zES>G8>KQ&~-mzKr4jn&U_VDQ54OwN&F7@Si`Sw=_^GoiR^~9}TsNbC}BAt4D!jgHL
zZvWhU;cApkobHd3?ei+iS7+UQ+h~6&@$!p(oL03Ci~QstTxwzLd^KO<^!--?t_&tC
zUE;+*98{~^+ot;O#nkJ&rp8BTdY+P;D)S_Z;m4Prz6;6Wy;t8n4|{#HzdZBVj(xKX
z6gGOt7sr=2$Nk;ov3`waP~z=vC7gzWr(`Cq;AD;gSz`2i_n*V50utF};ja!}H_y7f
zPs;Ps4U^@+tdgh1M=b5K+P+U?)irU~t5yFdUwb^c%I)<RM!k9knd|pg$_2Xg2ONks
zSlO~oFJZmnRnx5+ifVuUpV+&9X-M0{-_63(Z;BXX9{rhka@VUbCc06(!@uv_6_@|_
z!<z>E^H;7s3cdF%NcT@ozum1Uz5LsLC53@}l|0U-y}G|LvUl3uih5UnZo`v#pdJP%
zb9D65HudAD6C3Ld*l#udQv3UB|Lm`yp1V$p`o8%8%P$juU3<b<bR*Y@$Juo%Z`AJP
zUO#thZ!PnmY9<=KdeOPHOJ{ri-D}Rb+gEbs$F&O$Yy*Ya)UpEJ%WV;2V`RLr1T+cO
zTe4!|&Bd#JrFpYB+?Suw?Y5j{rx1htrw8V4uMe+%dMhk!xw`$0h4S$xN(#^KMn$b|
zUH$#V)~(g03-;AMZ9dstof#IkOHXe0Dv$P9R@JSi^w(Uubx7-}cv#l=f6buc>TPOi
zRjO}x!AnWmg&$w~&7C*z#x<j<CsH;gf2iX9oBdVpZ`w(*|GPF{_j$Aax8Avc_wz3P
zdz@o7`%?ASfU5O!yK-vmJ%f@ryT^-#YuA@gz4mTp$iH`5pfdgRBPmeFP~1K@_O$6n
z_mu4Ig;lTrO3Z&3vWL~~a?{zg4YSLaU%hhW#LJRZyLLr!a&q4N_ba&Q^Op6y0`u=x
zPCNFsy!7|$^5t!6{YL-4$L@^Y^XvWH=NTdQ-~E5j```ztwP*c*XV^r(+5d{K-#q2<
z_D=QIZ5t#1|4f=MR=3YR{>{oy_VZ-q*REFoc71E)OQ|^`<!6fDtS`>KJ5Q@#<4=<R
z)K|0l=C3eVQO@rxaY2s5IOwtH!-T#kZHu!0KI-lL_y3>d%8$REiZ0j84p{Fuzc{|s
zIolyWTlLO{?NPqE_xO(A$hp+<q{3I?#x1WIwo9h7eY5+v*DmtlzdwAeprtc!;ulDM
zwgVOFfA5#8zrG__yX)`Q-IsUAPkZs3``U{7q&3t2zp2^9-5`)E(Y9}YLhrOZt*bx&
z`RVTZyGHn3ktvV!>N432r{{}jer&7F%9tH0mAL(7Q%}m-SyR@1cKh<O#Nu-WZ?5jz
zuiFLI>8`ob_vzQ+zV6<VjGX%CiVH6+eez?5-`-`Cq4GJ`-rv)`&^`a{w8SmjW}Vxa
z(ULdA$l$@j9V?Hn_Whl;W!r|JpRJ)$(Ge>booeGdUh{p;!KxRR4*2PCeS2Q6v4H!=
zkr=Tk?hDud&NkL+{lMO0F#GH>P<>=~cV~RONa*XG{pBy86}QQJwoEnctuo#3vA!m5
zNv!DG#+;1#79huJ`@dgr_wnL!i^H9TH*%$}nmaE3rS{_XO=gv?eoLaaFFz&Eb^Yzi
zpxor{u(wtVLU;51m|y)db0w$)%qF}1>culWhcE2^{9UX3W7^yN*ID~*HqV#akuqhH
z@Gdn5ch?(JXTA$c+vUCad{*t6HQNHSxo^Ms{5Ij-c8f<ZI2zQR-R50lb*7y8V+Aj$
zPczA+_hY@o32)Dii_Ft|F4cCMy(oV#_F~aR&GyoVrqz3@>b+;i#qD)Jy=~g*%Vvug
zyH9bR-uHTWw))rWEsUM9|M=~~`!#%JXRcYP933w9p^Epd#`<hth1|#2woSYGYQMDi
zmXhF=|DN|W{(9JY_;B8x2h%1=frdoZ{##YLv~^+aqc1O>E`NC@(erMccTr8xvVPC2
zUkq2?{dX(OiNVCMo;zWNKHJ`H8-jkno;^iCc%f(W&*=&YY@5}VzV0nO8nv@{|GVGc
z+5Uc>mUA;_+ci$f`R_08|NMQ`qH}NU?XnXVtvl;={Y+u4sJQH&ezo0qmp#<{C^b)L
zeYW~i?@3%<SyO-CKN}Smunsir`9Cgr$?bFUtKQEujWC?pIQQk9yRYu(&e~qGdfx9R
z+Le3Sl9XjZe)zg~GrP7{Q{jY%Jq#I1VK@2CUH@C{=(b5oat7~=rn706*RPjb{qnQj
zrBic%fBk&jEGs*6?X@@SzQ|9$V(b;}|6+airc0(eJJSVL+ish<@?%xqx0KpFd!73C
z?v0%N?dl8j_aZNjPTgI4e#^vNpT3w}UVAhP)REiozrUt>vR8Ig^N$|Gl~o@uUGys6
zcxIZY0E6$D>#Q>GbQ}JCtS|Y`w9IHj`J6hV+$D8CzoqTnva0kXfBW5KpWVM*G>v?@
z?6>>f$itJqT#HSA{k63E>uyz6t#x@<Y}eN;Ob>f=3DiK^++JQ^o|*neWBuhj=KHR6
zZQDDW{oA{R?fTM-Kf7-_ux~2F!kD`P`;r;9PrsU(?z&e=)@-i$Gvm391t!1O9Q^fX
zwO8%d{c+P%zg@m6{7ZXjx0*-^`|i8TR;zuvuA2P&)z1F?mnO^SfTw=ur`uhwRQ~ni
zpSO|K-Y>r4*%_Z+_`lD;e1F>ekj=*fLRw<C`^&D1J};g1YhBJ98>R)FJChFA^E{}Y
z9Uc9D8INyLV!(kGnX=jQ_SGeAIL)zRhDUqeY^xAZ^60O;{pHc}-CrMGE7H9mwk}(3
zS@yXtf{MZ=yoWDb@?Bc>#D4ac?W;;Vr-B0K@bq5JAHS*$_?Z@TMkewo?5=0vNZtEr
z(U%XY$t9O(Hh-C|8oi~$xf4{5f+h~4&q=O%Ki@pWw%KImQ3IY_{nf$m>bL(05ArBJ
z`f8!Kd-ko2X;(H`F)ip+{l@X$u;J;}yr<&Z_$1Ct9=##LurNCyXwJMp=Ump_j9nS8
zdvA^DZtbtfzi)p%e>K<F#d8FA<potuf3FwxKJTKqYv@1IFW26#3~`NpEwIp|-FN#O
zV;9gkf$gFnFW4C`T;Y>t&X91s@i+V3U)`lfJefJOf@|lV@nRPErN-xTW%YKyi<9)d
zzx=Rfy%id`K7P->!1J|xyuZiVcv&u<#QbUdEd!pnt5$q}KW}1u#L}ioQkDC+Z4X~^
zc-iK5)t8Mk7duxiGu<W6c;QIR=7!m}ZLenAIKEDMq2+D9wJh?&U+x1wIcBrd=3EOG
z@AJB7Z?Ant{?oQh_sYF_Klh&pb-8T|PHOb}m(RUonJqQ@>(T3FI+j82{q56x{N!h>
zxO_QwmVrUw;@@fp2btYP;yMo|B`Lk_>e{e+<<Zrvntc;R8Kh^~L~Yy~Bg81sVCic0
zX4l6vj=PGSX3dM+=6rt5jisTx`CfebDgXM?$G=}Tzf_)jZ}suS^Pj7DbM@C}=ia=1
zcbD?XZJVql<FoH}_if0IsbB6Ve?jVppZtUsA?X=DHjXd2=O0OMG@I?a?uN(ItGhx=
zH$?swZjKFnpmRTL-LbUwmtr=a^62F+vt+oC$!R$8*UcWDzvm^Hzy2=Yo^|n-(%<LZ
zNxC;97D+GTxzd}?m&IMP_U_KtT&qJ~=UKk7dAT>qG|{2@b(~uk)9%s*mpK~Lp1-;%
z#jyHtj@pNXo9hqf<|s2{zk1$v_5FI;_3!7|dhG1po^kXW*WcQsdVgz=t~=b<+gp-R
zQ<eStUJGOADWCF{{xu~6tK4=dG0cjwc*pkWqtriD?qwJMax0wadHMWG(|g}*?zgq3
z?5^0nZOWA_kDxh5>^q{Of3WQ^F;d-kI-U10sDpBSpIE|=H*A}b%hYO6^PIzC_S(|&
z#l`LFC+|K=UBtFa_WvG(odtJzk6iy-z3=2VPS%FWshe+d)N^pz#rm$R)jjZmyWz|~
zR)z)H0t~+Ap5EtGvDwv7KIJ;3OjvtYnSnv!1IX7roAnanU*Dgy(6g(*lLxdYt3{n@
z!O5TBnfWH`CDeb{ul~=(a3Phn6{`I9e>NBy8HtFCFF$g`<>`|rFE*dId;RJ(hsJ-2
z0|pt4I_pcd_iqTAH0AIQdj^JGJ()}o4nbDeoSkbO{`2QgDf_xT%lzkq77i-Q-CrH9
zZ~Nf@GiXWXwe|7qZ*ES{{`Tf()TWfq|1$)Y-Cmrp|F_vJ=f(n2?XZjs3mj*e<<2^F
z`t<JCoBU?3=iXt&bBv*G!&09%#a`dyRqKt-ZaigR5ZIQ}_VMdoR_EZdvTbU<vzFNZ
z|9O6r%KGc8zkDgVTY5cq*ZY0H{a(ljtNC0wGt>C%i^ctNAD^3=nig5jt$M%rd&$FA
zak=j``yQ`ey?W||2?dYDg&ka{vX_2)EU%$;XkC+g`i!dUJ)n`^_qlpk=XAThXcy-L
zCDsN5QLg#-52kgA9ay$ZZEfUcH%m*)f5+dH?Jg=R+EVoN)MLB*20TxnK3%zBfy488
z)$ao1AFr?fyLyh_@?-t-@*8gLo|P|F5ZoQ!d-k3}_~lAoU%9n6?%nz{%ed=ZJOjh7
zNt$I$D_5+TaQtzgm)G%x36t=xVOeq5Fn{mYX`mIVvNja~wZFezy|mPO>Xa!ZM?UXF
ztjJkj5ZWjx{xqQF?lQ6U3s<$N9d(<z`E|+9ZAGSL)l;8!U0t#7<3T-khK3CwZ=LJv
z@4vpQ^!1jiuUWR=ZX|;i!%owUzV>j&=kxaQudl7WtUkYHk)(0ji(A?2ZC~v;bmvZt
z<y^o2;Z;>t<}pV%-pn!k?;>EV&-(e^vu{V|pS{z|%g`V>Ly19673V6CWoOR#fL3z+
zdS<?V<()fs_>LbrdK9#d)ar4M@s~5k=QWg+oX*>PJ|lR({QuwYt9R|X^?z&n`FX+n
zcQ6?Ij%tuwzi`jlwF|4IHrF#UyeL}6(GVCIILo4N(Wg(J{`6PwjeGj^Y38XZn&34S
zw-<7MPwDUM>kD&oa%x{6@Iab9;%>sZnFq4DwYToy=gDa(ILl;5_=eMr3<_^%Ch;s#
z0j<V1zgMvsv<|lNQ_9)d=Kf#5eub>zYLm_jh>DuEzg6GdJUlWo60~soPTqDg{kS!(
z++r_QE}u6^<?8k8uQwi-yPP^bcG<q#->Viaa{B!I{PlZ#t4q(CUT1l0WqmmHlbFN(
zN9UiXt7u<J?5dy4z>tu&jI&{L+S#DW%AImmFBGRvnWABA9Q=OI=f3(RkHo}<Z@zu|
z5D<@d$!zLrG4PVv6)OVX-`i`Nc}Yb#eqT)Tu^!G-re)JEo3mY5vb#~oHsy~z1H+4|
zWs&R#BCV&aoc{lw?_2flaqO<Idlv3`|EOod)r*qWr(T0*Y(thDUiODwKk0uwgTr+0
zspsD(EZVQjqw;H8v++X-hxqUBr!4er0Tl|Ij0QY2DkUZ28TMp^6iv`(WH?~O(U5rH
zrWyB*oZpP$S9U`B?uj6~@3t{Wu${5u{_*#0Dy*J;cIO}q1B0;=LmT7tdY%P4nI7z(
zJ>xGogM)zp!{G+)ZCkl-_~<r(^G}?f=P5~71{Q_|)`AR&8}eljw59cf7al%4f~an!
zuA19{%8mq0mIDR?^0JD%*M9wfeZ~^$AA6m}>bKp@aRjZ_%~oV+W88gLEqzU88K@!u
zG*@r3if=O5m3<5nYB9EFjQI+36PJFwZmy}Q<}{h@I~QoTLj)hAfzOX<2Hy>03-Uic
z|7<)-^yPza0fvTm$Cl^M`0$=BsY&m_(%F&!SQ#1;lo;AL?<ZZjs(6oe&*D9P)~8>K
zzTEI~k6>+6EFS~IG#3UQ;eUn=ybJHLzIpx6HrXTp=y4aeol*=8YdRSu)M{c6OjW$c
z^a4^vC{%DXBp%|4Js?_|_f$O0Yf=$tHwz0dqk+$ig(xAq5tPXpPJ=BkD_FRP@4@ce
zyZhgO!n8n;;c$z#*%sym=gIEc+opRb{S;?lXzGR7c2#i~--E8VqFFiLK*P&{wU3*@
z?sQ||5#A?uz^c4^mhC--@ZudyXM3q^-S!W(ATOMmp<$~a!(kVvoZqvm7GC3e@lop3
z?HILAP%-Dk$Y|hW@y1N`h>Hi;-ogf?oM!{cc?UjmG$byml}X6ZWqo}6_xC?NDXvyO
zrZO`q-1uR_r@T%emh0H|-|TnyGcdHUF(pWR)Mt}*Vk~g4oLe^Sy7r|+P}7%Xw+sV=
z0#ic7$FnT!Y$U8=5A3_P-?;5dSdGt<!ag<zhP#an5^goI2Ns%e|Jb{Y!J-%JHa13s
zo|apA3ZkWXYc<zQQF#j53dAAFa=_rxmtIDmiDC!7n@>M!u;lV3p)1!VrQZ}WFgQpv
zB}ioEGxGErHoX6p_P@`I7qp?{K>*7E10k<ucw6iAeg+1CERKdml_}R4V>a$}>OC8`
zbZ3(3s{9&Ial_EAz|ba`z9wJxz=Lu|{ktJbE4O|<pSGD56!irfAa8DExMs5L#ADN@
zu=eBG8u`UDRd&@gGH|$pT-TEupRrS|<1xeaw=1vw`=zFO#3kd?gE+_4%nS@Nl8gpD
zE>>?|Ny;i}?(jbNBkljZ`4iW8u<uV|U}%xDn|s*jx66bFz06;7|GRzHXJA;^$sm#T
zhta@8w;^?F@2$u<{oTG+<u3CVTw!2f=<8*WNQ-ft$GK<a4!21r7MgyXh7*rlN-;37
zFf$sQv1r~S_TX9V)=5hOYlWLW{bFEv@PnfvaapbPjt49Ue1AtpGjKLAFtq3~CCuRb
zR#q_kHq(O_b)bD&PNIwkXIP4P5|TH&Tc3Wd;<-4Jk%d7aMUdfek8#K525zscr>0Bp
z6$X0s^}qXeot=R}y^TR4t;v$D;Irc6|2O`NGc;UqW#BPhC$~2Cz(EP-8+ME;`)<ce
zGce4t0NM7;Pr-+Qk>P+0M?>PV>#Q+%FX*hl{$ktpuT{L)7+4q%sDQ-HIgey}Y<)K^
zulnwyQ*2jvMvFY%kexAq-raAz7#JQr5o9>*Q_|vWyGUlzl2>_iI37v=H@8{3TuJs%
z6eEMcsUJ)DocHBDIF%>E{(`$<&3|?VhZo8WZHno)eIC4v;Vv+aKETM3AjNXP;L1L+
z1oLeSV!{lJ3<lgx2@<c~{cTt+J!|XsRgOZSwed3;84Y;0u4lbr%UIwY{XKt{_BDHM
zNl3^^6dl{v%lKv)!;a36^$ZL%xEKw1envA$?@H%cP|q%~t$~^0fCxuJ;)(sz9_HH_
zHnGYuFg$Z$;BkJd+Yl?yFxxgRS9k4X7tnI`G)ItdGQ;kD+!g7SduK)M0gd+CZgii%
zUygynq7>xRxu*}v>|)q*H}CbQ>&y%c%ljB4+JvgNoM$rQS?nF(E$hI*U?9(wAn|D{
z!?p_Et2?82ec3t#v{wJ80K?$`pC!&01=nuhrhT2sgr`~e0Rw}9B2&VQSAB(`{@dI2
z`IGLd_4*|8IxsMtX=0FA7ISxlbjSMN<pn`=k`~LvezRg=P}soHka+4Q@9xH&d*RyP
zV&u%R5XORke|}1tWK2j(N}4iv?%L4RVX|LuxOR(aYHNGzP4_M=ECekeZDQsAW1pXw
zcWv+Yd#g7lAJ@{;^ZWbj>t)cY#Yc}Gz1uNG@A+-sCDu_4GeK_NIYW&>FJ?zT@v}1*
z@9Zp|I)DE9kB^UM|N8P$OIzFe(Z{`2UoSE9+bo!>9iH{~7q}&mc7EPkRqtsYllbIp
zR{Z%>WBcPl^QrUauP<`#4haZYa9VHo8f*R4r%R4YI819`WH>M<$1Pl6L|ozK&6~4K
zGAI4}^W5Hh(w?~Wmo8r}y&hZsa^LTF)(eY$W*9gcr=PpBxZkeo$o%zjd!r_+`(Jx^
zceiQ6friKP)-PM;=C?c;w7%JcW39lK)kkN4z20B;YFaVd>&BUjnd|-?OY3*D|I)Xo
zZOg0Y;#!u9Y7f>kGB|{DJU##ZT@vw)*#6b))=fK`Hu>~Zt4HVW=55c)%Cf3y)k+sH
zXB3}u<y5xE-p7lssN3!Gj$>wEDB%reIndR`RcRx)bjgyE3L8~bS5;M2_u9D9(ydkB
z-mF})LgVZl%i#K7m*<yUbQPEVc;nbHx4YHv_x`%JeP8L7dfq3`o}GGG@@xO4%a^U|
z17puTozHT~l<~ubck#1(!HGvWnB@SDhHBZ3!n0@3?)r8sJ5{1>Ue&9Wi|4)n_xpW#
zY3bJg+2`lk#zy>I`B9fS_^zvN1Gp&Iqtq08pwq+4YtrF|0hyVZ?Z>^ny{FEaRdwb0
z&9`N|(q=xBK%2GB&NkOJHxCb9?)O#xF?ap$^7q$%zu#YfW+TJJix<24%-qY<lf?Cz
zFW+N2!E}LvVMecJ6vN_0i%be1xjcRP^r!sis=c7a+fy`y{eIZQtq;%Je*6Dk{Xe<4
zxB0%``+Z*3_jj@vb~T^e&X<rXzJ;M`Z(BX@fs3_s8SU*E7+ypz%V3<c#IpF=i(}IH
zAz@+D)cohIsr>wG<-UD!o72xvQ@LAknAgkOJ2)?I-I8k`t1ud(KXzBlo;~~P+5EcT
zpFe+Ay<WTh$;XO6k2yix9#*Va@gOFnVMo=Laz^Pk!3%{|u2Xqe6>rx#m3V!ET_K~j
zJ_ExGIl`?`wzI!}Ro$)o{Z>m$D<~&tjggVjmq*?DS=ZJ?PFk|J`unT1=J%&uxn*0k
zg5U1P0%!A$l0U_o@;+5uz3?Wmf5m~juY7;6QuS00s@A%c*yWbOz)-+(c@48fms6#z
z;$Z`mN8<lOHy^*C{yla<-MSkMbANX(jsALkPwuX?r24<#W_<*^^RD6)`~ADQ&zdKm
zO51$$rkSwCg>1%6%p431&)gVzlGo|Y;@q=xhH3%>gMl$qg2d-uW}YWe4Oe;3-ip4e
z;3+lL3sed8a5N;Y(Jo{DSo(k2&ZNKIXBik6iWM2!Hm;LkJ^||Tp6O?hIJPT^XMsLL
zUmCjr1A|XDL_ndMftewpj^%*CvwTLL0EV1z6<fD%@q-lRpep$1w&v9~5;OE3%zgSb
zE9>K`^pciK!qqHK<}ffU_$|nAc*c6x4OR}D8)_f_Hwt>3x}7~QKaqjq!d{Ms#51A~
zA`i@#1l1=I3Jh%<9nLe|J0fJjz|g`9@`>)1J2&2hCU1T|<y@*nTRpc0Xdg4!(;qJi
z7H#RhYJC+;VcWwXaV*Dd_MDB4kzB{F|FyQz^xGKN7rc&_fq{je(csL^Ka3x)9An(3
z&+yjf=`02Y2T)6K&SiVSCs(2v%-^q`s?wXpZNb3c)6XFBjA^TE%x}2^;9T{FqapF!
zcIgu<byzRocysaGr)#<oKn>Fw@{9&&Sd)Z296i_WX}e;r=T-Y?F>|OC0|SQ;Q^E}I
zFWVW)>=^Q_QrHEcdEJ^RVaDh8Y(-p~8_YL@VluInLE_mh=7i_b;L_TmLXn|uXRs{u
zi?34cI*iT?EDQ=V0t|=e<ck%gZ*S6Gl@l;`w`(BZ?zS(T+$ot13<@3`4T<kU%9(C%
zSsyjYL}@Fi>fXh`Ai(#-g}->8VZ&EnIn%8b{=a_CuDQE^Wyyag1_3>ehD1F(wV-`d
zqpdFU?pe8`mQ#X(K|l`V%f#)Pipz4Xclf<wI|Pmod8P%dC1sj3UbUsYU7Iz*1>~YR
z0t~tbuAW?)P_wo^b6v#$A|nQd1AjOg#2&nPQdR)AV;)C?*oB)*6F}9bLxmDUltYsG
zafUp-kGYKE@*hBf|3QF3cfnrnA4j5L^|wmzyVq|f?E*y%NX~e&L)o<Jph3QQ7X@F<
zm$)6el#!u9L4hHvAzyOF<VcQN=8P)Jrg3Gv*LuHVVqiGv$grj1+#2Q`nexI83=ACM
zEDqcS$L9TkR8k5xf(*J3Qlc4n4`;+PwJ~roINVSK871VAo7^2`bMLWSeBA46|9+|6
z%)N9X1XQmn+!0{VeQ+W7PL@=3U-h2+Kldlt*@_Cke6b1UMukI(;WZ^KON4LprU*a&
z&%nTu&En8q@aq`kyaPc7v%g>Un&h?-9L=2!8G`?|GWaJbomloL%M+BW6k-G!bd@&d
zTJJdZxP;N%ta{7pMR7um0t^iWt_)i^o^$50hX({{uUAz4+7mSY$BTCwybKH_#!L%b
zclZ_XJXk2~kYfgF87asJGUy(<x=q^SYvHbjTI+@jpkB}qC5EUYoZA`1HnZ?BFg)yI
z$PiRFV!QC#bj#8s+gU(o1wgC`zpe8ic`JihsM(5hcNeYT0Hv(g>n<@cIG8IlM7jL7
zYH*il-jNyq`d7VjP?grDgX`HC7#?={S9TX1m8<7@uxy*G10<4_8KPS9wRgN~ORL;_
zOt3!@v|sa@1H%@M`ziv5CuB)S_f?#ERtstwuW4Y&5WFbJ_qC(QW6~-qK6!S~l7_vZ
zypuq=s>PaVLF<<H>@`0u9O7Tc&(gkjKZ=8a;gB%n1*cornQZp3Uf9p<VIKtw?s66f
z?m}53!6&alZfGbKV9-6XU-r(#T@Bthm?an(Rx2<>br^5UHJ`Aa^*~yGc(nH`70>|+
z8<?0DwEl=@*lz}DCECP&?dc2nf8<8O6gCEiU4o1koN9JAhSaZR-B9~?f5pwW+w#_P
zGcc6sgPgXVp^T?tb_9bI1B16BLzKsFi35GH2mYxhfLtHJ#CXA}?+R0f1e^B(&=C#G
zI2%9%BnupyMTCVHzj>3hHfrmu7030Wws=HFMi%Xi>FVwdPD@)RrV~+c!|?Jl-{AN6
z_NH#Wxh3!JF1>wC4?eGFt~eoa;8WE0Z`aL1D`J-&V_$8;z;L10N032WFJ^^h@w1F~
zcXomrXg@waHm&_tGIi=y*^jWj^%qV})!th7x9aRX+i1{M`AMqYe~w?de*N{E&F8N@
z?zg`eJ6*6uFJ8RBc{9)930Lg?<$KJZ*VS3Vvet}&;X*CAK~eFjQ@!N0?)HptZ*ERn
zvV4ABRQ~?I+ve4Ni@aNYzqVc0zVy`<zWeuUzu&ET*ebpy=_nUym;a=azhAF=PXcXc
z(pzgg<wq2QZJhJgcVenXT#P_FFS)sX5;^ji7d}nu6KGn$GiUG9eR&Vg{?1Z01R0^t
z`BMzNc^17b$MV(ZRjXG|oi<Hv`)%F!!wc2t*93{jmR!`l`lxL8)3V)v?_a-iWzMCa
zO$SA#mri=-7Al()R;8aAT)1_~HB$ygh6dA?8t#PjqsNbD-`!OjwYO@kT=g46(9qfG
zX}Z=+i=QFdbj-my=2j)<6WY}J%R;YsPjdUd7_=FAHOEh}151|R+%m7?IdkUBTYD-u
z7Zn%tOUqXnfp#jMooRghf32@??@Olzn{&-4tZF{1rLlJE^$C-z+%zri)J{xgWN3IN
z^hp>0kl^oa#)}q1w#c8(;fv)iFb0)}8<tGS$JOq-XRrMOI`~&qT&%ja>}pzSYUYaz
z3yXH&joMSOQA{Ud!SZ=kr|Qo!v-7Rkup!{+(W8@=G&M1K24!WvGK>~Gdv3dk&)0)$
zQ?G7gOHylKU`Wti#?e4v#P8S{oNcUs$@gkL`_?y1S@L?_Zof$;<?mu_|9m)nw>Z)9
zK|Z6+jWc`-3=9!d&VE?S(K9(LI9Pf4<(02qWnC^|x8IqwyY91JBWU$k`F<VHg2ng$
ze#G>8B#Id@9FRKm<ELzfqCJ~H<I~qn_jYWLDt#yB>UzUeMN{uYC@6GY7`8~ni+{Lw
z$;$K6i76GX7C%fG7!G*-c(Pq0{I-#Soel#pL&H}A2Hj&fAHQb2Fe_P}fnkFY(}JTa
z-#T7>um$INTNa1z!rHa02F;)oKAs9N=&G4rTDW4Hr~F^x<iF8}85tN}D=<X$oK5@O
z&kP>0yyn8NMd7|r!m<C13<<t04&1NiOK5vq)i!AFXk}n%SYU88?f}P{_&Ce3*IcVh
z?seFK5A|SSWW3OHDQ2Tr@lm$jJPZs1Dxgi|R6YsDg=IH`u`!SHq>$iX;U``U3<o+t
z)#m-SJHEDqp@J(bJ6qNB<;$0j>sS~V=B&AOE6Q^2InW^}W`|FG{4P?tH%?SsT-(kr
z&TO{sji)6uXU&pYx_mjP@8ad*!BJ)zd&=v&R^;odo!363E`9wpiutes&(&+!R>kkH
z+jM)arBt8Tf#8Jgx37YaR=JofBqVg`AAeu})ODeou6-?9`|z~a_0wDDS*_mo=VyfR
zH_$OS60l=%EY`hvkpY^-xp?v7LA|A?BVR_TX2<GZm)f_cWN(Z%Q>&)}_@o@rj(i+v
z<V3~f<*frv$^7_Hp{lO_x=VXqfTQC<!}*{8d_I3VZFA|5uFlSv%ewr}9_F{NiFhk1
zDS5o+oc8)Xi+Xx^jvhIZ@#e<Hs67>h=AqSDY`jt>H@4o&F}quMT-M#@p3nEMi~H>^
zEu9{>s_yTvlWCh@e!U+5{zLI$gD)4H`B&!O-v>H4!q2a7Tkh?$JD<<Zw~udT;|=oh
z>Djb-^VY1ZtMqm?O)g&ZIv3|iK=QoJ#_wC-73`_5zU*r*8}_|N(zvhsUcjF1$9G>-
z-dmq*HoNZk?fY_H=Pay>_kXf^*~X25<#!6(tG>TmJI}WI)vML(!!k2p8lJy>^Sc#C
z!?f=Y+vQ*NS--ome16@krQXw5?c5m&s;_Em{~FGpK4HRw+TY)P?k{_P@2;=t{9T60
z$4XYz9_<o!xBWKR-|ph}eP4BJ-&fzid~0iV=~2<}7hT%xE*$HX-o5eBxyZd$rR}d{
zc9pDDc{wY4o#m_IGez;QU%xi3`%`gtzWx6W`~PS2>lWMp``BOg>*eyVH`C{9>goAi
zUhW^BocuWO{7&dGGsVx(h1S*WJJv71{^jN6tJkbqvs&+Qi|p$vep~|?Z+1Mlo%!9V
z-16SusY^hI*sS=ybb6fCGfU^M_vGG3@$WrW`}^C=o9XjGoAR%Q$Jd_u-1ex}cH;S2
z^8bGv|8>6p-{SDNN>};19}By>yGyTzhRc55Gq-a6;$B9XMQhfanUGTR<6*nI?X#WF
z=NZrOOaCq^>QMaroawQC`T9f0<E!6pJ$7cMv9-;#xpQ;>JYcuyC`gTc|8Dntvo=22
ztctz&l>gmX^F4IZ$6ML!fA0TYwbv~xYt_fw_y1k{9}yZlRZvzoch8qg-tM;NpPZcB
zYyED=;s1M&KQ{dJ^>zA_XV1z!@6P@D^|kG!(A8m*lX~v%E|>p0V{)xN%i)RZz8;g#
zH__Yw=hI}Xe#v)lJMUaSt-t@yjt5QLeU&|u?^Y+u{EG|@4NY}%bCX)X=Tq1J8E0pk
zOHb;VsvR!(QD$!K`o*HU4KEh#+7&hF%Ifg-$#UW0;maGD*|ltKW2eX0?cDe0)9H5_
z`U=}}Z!fF+`z!Or1jS$H%J(fkbm-7xx!pxiy`DaKQc{ug>&wfiokgkP(<K9ZeSNDw
z9u@z3_kCTs+3dF~j?etL@7J4KTfJ3PRb}7JzWg$ztn6Fh`{>}{$=B9IUVdl^+19^o
z_w}IHzi*^%KKZmr_VbR(mHI7{Qcq9Qw6u&=cJI5AyZ!F6zrVlBf4xz#bNkYzORIjr
z-Trm&``Y(Aj?a3yukP>Ww6n8Jv-0xZ70fp>GBQax(6IR4@^gExU+iPxuwQ?Dwfx@~
z?pCkYY%aN#xm@<^4$x6QH`C`|mCoO@P&~dSFm8Q#`n<|zU0q$fHr_gS^4z&=pU>N`
zFTYpm?r-~bO4-|6ON-B0x^LaO)!k;_k|i#hnwqk5-kzR;o}QlNQ=|>n@Bg=J)v8q`
zp#1q&!ubAm=b!J+p1iZY3pC<1Z^Mr3y}iA=zTd0<x;DSoo0F5%-S%C|eeQ)x$9g1x
z?f?JV{^`@FFFziazuqoix1w8rUxZx$^N!=~r}x%AY!w%gm#@F07ifIvx^nHcDO09s
zXlcFLP`^dywO#46y?te%)vgO4gR=9nUg>`8xRg^<G-ny5cK!cteBLH_v0Lw@_<x_o
zLC4a>#y->B`QuUd-~aV@b`+jEckbGOM&_@dPV3LVuz$_jS2NS+T|6otzozc*FIkWu
z!nb7Q<z3r)Jx&|6GCkN|_H)GMed3^r-T(jPt;^ON>y=)<(78S9=BCtL9}aOhM?_u+
zo!Rtt?)#eOJA^l%DTw#~vY9QmTg<^@(z`vM{q}yn7X9l%Ge78ft;tn+DJe_*?EjYJ
z-ru*kd(~?zz3JY{Zao%RZ&$8d*?!hZTUR$UD=Vvfinsu1RIB>^-sQ9N_x)TlFa6r@
zSG&#Ci|<dGH0eo!g-`AJ+K2hy-`$PcRr1m>`hVY(iubJnrYR=`K&fr(yMi6tPfynm
zcXT}Hxxansh287jf3NZM^$iUUp1f++Dl3`#HNUUS{BCtikU{sH0S{=;d)3>m*T4Mx
z{XYA{gM;d|eEr84^Gcho$iKfYHX!jj=y;(%S?j*2-%~V$r%a!I{o%%?-qXu6G`UO9
z+kT%jY4T*#;%8?jSIvuj|N6nf=Bcx0-Fmq1>Xj>?#R}H1YP%sj?X7K&zy7MFuYX?T
z`z`T*pq%ynUbX+;>hE?hi)G&L6<F2P)wQMm|G&l8a=%q;Z{7C5e~v}q<KzAIFQ2!a
zs<pkBe)dSGu&AWu&4+fPA|fj`Z1}L^`%2^Yr7}t<CI<+lY`<GJ`(cSyZ|do3lE1&b
z{ai0=Q*ojG%-8Gj`FG0i*ZzAiU-?8Z`Q#+kV|}vLJ*L^$B(BGn&lMCFPOka+bo#*^
zEt8AR+kTfgJx$lR?Cq`8nm-?pyZQO`O`I|%Wy;j4l5K@Qw$_2fTwGc{?!N!`+>Jez
zn+?m~#V8vYUAnO$(RqX3!zC3qa>)l87?17!ey{q*`7<@~+1c5>TeGiAWnW)ssv_xc
z^RZ=P+1sf9Yp%!t|8>3X^i!jI6_0t7|Ni=V?CJFQJRUil8yOiHALc)9;?^_x_hq^L
z&;9?e@Bf$T;_fbe-v0ldjhUC#9{;}of8WERojsE#P0HK9Z@u-kuV$vElG9_$W?tN1
zU!Qz;SE=m1<9XX}*Y~ZDfB*3w_rany*=yIXHLd$JdA@{p_&O7i_x60hSADF7Q@H2h
z;dbfX-rj%3GoId)d;k500ng%HyKcR>xVZh}m0*9_k1}aH*Q<YuEPlOqyA(4Uj{)fT
zk;k9U+nX=*nVIA>(`f18O1D0lOr5ALC;tEK+;hD!g~g$Ja><eP&*xR2t3Olud~W)k
zipRbG-q(LwEH4Q%VbP*RJyWKLJf7sOXZY#!=iU2%_{}iQo_2A2e!Q}l)+rBfZ{xJH
zvr=>J@4Nf|^ZNMJi4w<7>+jF$kuW^e(bZ*itVc3=&);vi@5Yoq*NxnCWI~FFh=|NL
zW8?eR`R#r@=-9t~-|LT6dyh5q+hr_SvP9zI#f!&U#p5b&)HgRWvr82}Klk>-jrsq+
zsQ<5TJ~>%kdVbxnO!L@_8)aWtmA}1Jdgc2^cKMnI_Q$WU|6cLFHNo=QRymuB3p31e
zXI)(GFAs`s35$XSg{O75%ZNQWQ<UHTy*fvt?c?<Qe~xza_L{C*wW_azZ}Xk&!avt!
zW@H%T?f+Y*Y-o596rzPsr-n=Y{`S`TjZM4VbC-LsEsQ`%0$kss%f!H-AYuqQcWuZY
df%x<P9*@s8PX4JM7#J8BJYD@<);T3K0RYLGCSU*n
literal 0
HcmV?d00001
--
GitLab