TensorFlow2.0__03.01_Convolutional_Neural_Network.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Convolutional Neural Network\n",
    "\n",
    "### Author: Ivan Bongiorni, Data Scientist at GfK.\n",
    "\n",
    "[LinkedIn profile](https://www.linkedin.com/in/ivan-bongiorni-b8a583164/)\n",
    "\n",
    "In this Notebook I will implement a **basic CNN in TensorFlow 2.0**. I will use the famous **Fashion MNIST** dataset, [published by Zalando](https://github.com/zalandoresearch/fashion-mnist) and made [available on Kaggle](https://www.kaggle.com/zalando-research/fashionmnist). Images come already preprocessed in 28 x 28 black and white format.\n",
    "\n",
    "It is a multiclass classification task on the following labels:\n",
    "0. T-shirt/top\n",
    "1. Trouser\n",
    "2. Pullover\n",
    "3. Dress\n",
    "4. Coat\n",
    "5. Sandal\n",
    "6. Shirt\n",
    "7. Sneaker\n",
    "8. Bag\n",
    "9. Ankle boot "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![](https://limetta.se/globalassets/nyhetbloggpress-bilder/fashion-mnist.jpg)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Summary:\n",
    "\n",
    "0. Import data + Dataprep\n",
    "0. CNN architecture\n",
    "0. Training with Mini Batch Gradient Descent\n",
    "0. Test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.0.0-beta0\n"
     ]
    }
   ],
   "source": [
    "# Import necessary modules\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "import tensorflow as tf\n",
    "print(tf.__version__)\n",
    "\n",
    "from sklearn.utils import shuffle\n",
    "\n",
    "from matplotlib import pyplot as plt\n",
    "import seaborn"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 0. Import data + Dataprep\n",
    "\n",
    "The dataset comes already divided in 60k and 10k Train and Test images. I will now import Training data, and leave Test for later. In order to dataprep image data, I need to reshape the pixel into `(, 28, 28, 1)` arrays; the 1 at the end represents the channel: 1 for black and white images, 3 (red, green, blue) for colored images. Pixel data are also scaled to the `[0, 1]` interval."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_csv('fashion-mnist_train.csv')\n",
    "\n",
    "# extract labels, one-hot encode them\n",
    "label = df.label\n",
    "label = pd.get_dummies(label)\n",
    "label = label.values\n",
    "label = label.astype(np.float32)\n",
    "\n",
    "df.drop('label', axis = 1, inplace = True)\n",
    "df = df.values\n",
    "df = df.astype(np.float32)\n",
    "\n",
    "# reshape and scale data\n",
    "df = df.reshape((len(df), 28, 28, 1))\n",
    "df = df / 255."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 1. CNN architecture\n",
    "\n",
    "I will feed images into a set of **convolutional** and **max-pooling layers**:\n",
    "\n",
    "- Conv layers are meant to extract relevant informations from pixel data. A number of *filters* scroll through the image, learning the most relevant informations to extract.\n",
    "- Max-Pool layers instead are meant to drastically reduce the number of pixel data. For each (2, 2) window size, Max-Pool will save only the pixel with the highest activation value. Max-Pool is meant to make the model lighter by removing the least relevant observations, at the cost of course of loosing a lot of data!\n",
    "\n",
    "Since I'm focused on the implementation, rather than on the theory behind it, please refer to [this good article](https://towardsdatascience.com/types-of-convolutions-in-deep-learning-717013397f4d) on how Conv and Max-Pool work in practice. If you are a die-hard, check [this awesome page from a CNN Stanford Course](http://cs231n.github.io/convolutional-networks/?source=post_page---------------------------#overview).\n",
    "\n",
    "Con and Max-Pool will extract and reduce the size of the input, so that the following feed-forward part could process it. The first convolutional layer requires a specification of the input shape, corresponding to the shape of each image.\n",
    "\n",
    "Since it's a multiclass classification tasks, softmax activation must be placed at the output layer in order to transform the Network's output into a probability distribution over the ten target categories."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras import Sequential\n",
    "from tensorflow.keras.layers import Conv2D, MaxPool2D, Flatten, Dense, BatchNormalization, Dropout\n",
    "from tensorflow.keras.activations import relu, elu, softmax\n",
    "\n",
    "\n",
    "CNN = Sequential([\n",
    "    \n",
    "    Conv2D(32, kernel_size = (3, 3), activation = elu, \n",
    "           kernel_initializer = 'he_normal', input_shape = (28, 28, 1)), \n",
    "    MaxPool2D((2, 2)), \n",
    "    \n",
    "    Conv2D(64, kernel_size = (3, 3), kernel_initializer = 'he_normal', activation = elu), \n",
    "    BatchNormalization(), \n",
    "    \n",
    "    Conv2D(128, kernel_size = (3, 3), kernel_initializer = 'he_normal', activation = elu), \n",
    "    BatchNormalization(), \n",
    "    Dropout(0.2), \n",
    "    \n",
    "    \n",
    "    Flatten(), \n",
    "    \n",
    "    \n",
    "    Dense(400, activation = elu), \n",
    "    BatchNormalization(), \n",
    "    Dropout(0.2), \n",
    "    \n",
    "    Dense(400, activation = elu), \n",
    "    BatchNormalization(), \n",
    "    Dropout(0.2), \n",
    "    \n",
    "    Dense(10, activation = softmax)\n",
    "    \n",
    "])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d (Conv2D)              (None, 26, 26, 32)        320       \n",
      "_________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D) (None, 13, 13, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 11, 11, 64)        18496     \n",
      "_________________________________________________________________\n",
      "batch_normalization (BatchNo (None, 11, 11, 64)        256       \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 9, 9, 128)         73856     \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 9, 9, 128)         512       \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 9, 9, 128)         0         \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 10368)             0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 400)               4147600   \n",
      "_________________________________________________________________\n",
      "batch_normalization_2 (Batch (None, 400)               1600      \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 400)               0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 400)               160400    \n",
      "_________________________________________________________________\n",
      "batch_normalization_3 (Batch (None, 400)               1600      \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 400)               0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                4010      \n",
      "=================================================================\n",
      "Total params: 4,408,650\n",
      "Trainable params: 4,406,666\n",
      "Non-trainable params: 1,984\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "CNN.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 2. Training with Mini Batch Gradient Descent\n",
    "\n",
    "The training part is no different from mini batch gradient descent training of feed-forward classifiers. I wrote [a Notebook on this technique](https://github.com/IvanBongiorni/TensorFlow2.0_Tutorial/blob/master/TensorFlow2.0_02_MiniBatch_Gradient_Descent.ipynb) in which I explain it in more detail.\n",
    "\n",
    "Assuming you already know how it works, I will define a function to fetch mini batches into the Network and process with training in eager execution."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "@tf.function\n",
    "def fetch_batch(X, y, batch_size, epoch):\n",
    "    start = epoch*batch_size\n",
    "    \n",
    "    X_batch = X[start:start+batch_size, :, :]\n",
    "    y_batch = y[start:start+batch_size, :]\n",
    "    \n",
    "    return X_batch, y_batch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# To measure execution time\n",
    "import time\n",
    "start = time.time()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "There is one big difference with respect to previous training exercises. Since this Network has a high number of parameters (approx 4.4 million) it will require comparatively longer training times. For this reason, I will measure training not just in *epochs*, but also in *cycles*.\n",
    "\n",
    "At each training cycle, I will shuffle the dataset and feed it into the Network in mini batches until it's completed. At the following cycle, I will reshuffle the data using a different random seed and repeat the process. (What I called cycles are nothing but Keras' \"epochs\".)\n",
    "\n",
    "Using 50 cycles and batches of size 120 on a 60.000 images dataset, I will be able to train my CNN for an overall number of 25.000 epochs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.\tTraining Loss: 2.963158,\tAccuracy: 0.825\n",
      "2.\tTraining Loss: 2.2184157,\tAccuracy: 0.875\n",
      "3.\tTraining Loss: 1.6165245,\tAccuracy: 0.90833336\n",
      "4.\tTraining Loss: 1.8334527,\tAccuracy: 0.89166665\n",
      "5.\tTraining Loss: 1.2200023,\tAccuracy: 0.93333334\n",
      "6.\tTraining Loss: 1.9942344,\tAccuracy: 0.89166665\n",
      "7.\tTraining Loss: 2.1937194,\tAccuracy: 0.8833333\n",
      "8.\tTraining Loss: 1.8208188,\tAccuracy: 0.89166665\n",
      "9.\tTraining Loss: 1.6057307,\tAccuracy: 0.9\n",
      "10.\tTraining Loss: 1.036152,\tAccuracy: 0.94166666\n",
      "11.\tTraining Loss: 1.1800997,\tAccuracy: 0.925\n",
      "12.\tTraining Loss: 0.9550361,\tAccuracy: 0.9583333\n",
      "13.\tTraining Loss: 0.71287704,\tAccuracy: 0.96666664\n",
      "14.\tTraining Loss: 1.1355927,\tAccuracy: 0.93333334\n",
      "15.\tTraining Loss: 0.42137903,\tAccuracy: 0.975\n",
      "16.\tTraining Loss: 1.2537751,\tAccuracy: 0.925\n",
      "17.\tTraining Loss: 0.6678084,\tAccuracy: 0.96666664\n",
      "18.\tTraining Loss: 0.67579913,\tAccuracy: 0.96666664\n",
      "19.\tTraining Loss: 0.93005073,\tAccuracy: 0.95\n",
      "20.\tTraining Loss: 0.9756758,\tAccuracy: 0.95\n",
      "21.\tTraining Loss: 0.5439663,\tAccuracy: 0.96666664\n",
      "22.\tTraining Loss: 0.48298022,\tAccuracy: 0.975\n",
      "23.\tTraining Loss: 0.71461415,\tAccuracy: 0.96666664\n",
      "24.\tTraining Loss: 0.59260106,\tAccuracy: 0.96666664\n",
      "25.\tTraining Loss: 0.52771163,\tAccuracy: 0.975\n",
      "26.\tTraining Loss: 0.70353717,\tAccuracy: 0.96666664\n",
      "27.\tTraining Loss: 0.8188721,\tAccuracy: 0.95\n",
      "28.\tTraining Loss: 0.5126863,\tAccuracy: 0.975\n",
      "29.\tTraining Loss: 0.45719293,\tAccuracy: 0.975\n",
      "30.\tTraining Loss: 0.54228246,\tAccuracy: 0.96666664\n",
      "31.\tTraining Loss: 0.55225444,\tAccuracy: 0.96666664\n",
      "32.\tTraining Loss: 0.9566125,\tAccuracy: 0.94166666\n",
      "33.\tTraining Loss: 0.29893246,\tAccuracy: 0.98333335\n",
      "34.\tTraining Loss: 0.70337826,\tAccuracy: 0.9583333\n",
      "35.\tTraining Loss: 0.5612643,\tAccuracy: 0.96666664\n",
      "36.\tTraining Loss: 0.59767526,\tAccuracy: 0.96666664\n",
      "37.\tTraining Loss: 0.67508984,\tAccuracy: 0.96666664\n",
      "38.\tTraining Loss: 0.67434996,\tAccuracy: 0.9583333\n",
      "39.\tTraining Loss: 0.0228432,\tAccuracy: 1.0\n",
      "40.\tTraining Loss: 0.2526902,\tAccuracy: 0.9916667\n",
      "41.\tTraining Loss: 0.17682979,\tAccuracy: 0.9916667\n",
      "42.\tTraining Loss: 0.4235972,\tAccuracy: 0.975\n",
      "43.\tTraining Loss: 0.13609551,\tAccuracy: 0.9916667\n",
      "44.\tTraining Loss: 0.5388846,\tAccuracy: 0.96666664\n",
      "45.\tTraining Loss: 0.293708,\tAccuracy: 0.98333335\n",
      "46.\tTraining Loss: 0.04402418,\tAccuracy: 1.0\n",
      "47.\tTraining Loss: 0.5923631,\tAccuracy: 0.96666664\n",
      "48.\tTraining Loss: 0.7310451,\tAccuracy: 0.9583333\n",
      "49.\tTraining Loss: 0.41430146,\tAccuracy: 0.975\n",
      "50.\tTraining Loss: 0.14125405,\tAccuracy: 0.9916667\n",
      "\n",
      "Training complete.\n",
      "Final Loss: 0.14125405. Final accuracy: 0.9916667\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.losses import CategoricalCrossentropy\n",
    "from tensorflow.keras.metrics import CategoricalAccuracy\n",
    "\n",
    "loss = tf.keras.losses.CategoricalCrossentropy()\n",
    "accuracy = tf.keras.metrics.CategoricalAccuracy()\n",
    "\n",
    "optimizer = tf.optimizers.Adam(learning_rate = 0.0001)\n",
    "\n",
    "\n",
    "###  TRAINING\n",
    "\n",
    "cycles = 50\n",
    "batch_size = 120\n",
    "\n",
    "loss_history = []\n",
    "accuracy_history = []\n",
    "\n",
    "for cycle in range(cycles):\n",
    "    \n",
    "    df, label = shuffle(df, label, random_state = cycle**2)\n",
    "    \n",
    "    for epoch in range(len(df) // batch_size):\n",
    "        \n",
    "        X_batch, y_batch = fetch_batch(df, label, batch_size, epoch)\n",
    "        \n",
    "        with tf.GradientTape() as tape:\n",
    "            current_loss = loss(CNN(X_batch), y_batch)\n",
    "        \n",
    "        gradients = tape.gradient(current_loss, CNN.trainable_variables)\n",
    "        optimizer.apply_gradients(zip(gradients, CNN.trainable_variables))\n",
    "        \n",
    "        loss_history.append(current_loss.numpy())\n",
    "        \n",
    "        current_accuracy = accuracy(CNN(X_batch), y_batch).numpy()\n",
    "        accuracy_history.append(current_accuracy)\n",
    "        accuracy.reset_states()\n",
    "        \n",
    "    print(str(cycle + 1) + '.\\tTraining Loss: ' + str(current_loss.numpy()) \n",
    "                  + ',\\tAccuracy: ' + str(current_accuracy))    \n",
    "#\n",
    "print('\\nTraining complete.')\n",
    "print('Final Loss: ' + str(current_loss.numpy()) + '. Final accuracy: ' + str(current_accuracy))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12431.102645158768\n"
     ]
    }
   ],
   "source": [
    "end = time.time()\n",
    "print(end - start)  # around 3.5 hours :("
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA3gAAAEWCAYAAAA0DzVNAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAIABJREFUeJzs3Xd8leX5x/HPlZxMyACSsEIMS2QooBFRECeKC+rGWke1Uv1VbdVaR91Wa4e22mGr1bpardraUrWOKmpdVbCCgqKgIHtvSMi4fn+ck8PJPknOySHJ9/16nVfOcz/389xXgibneu5l7o6IiIiIiIi0f0mJDkBERERERERiQwmeiIiIiIhIB6EET0REREREpINQgiciIiIiItJBKMETERERERHpIJTgiYiIiIiIdBBK8ERERERERDoIJXgibcTMFpnZkYmOQ0REJJ7M7DUz22BmaYmORaQzUoInIiIiIjFhZsXAwYADk9uw3UBbtSWyu1OCJ5JgZnaBmS0ws/VmNt3M+oTKzcx+YWarzWyTmc0xsxGhc8ea2Twz22Jmy8zs+4n9LkRERAA4G3gXeAg4p7rQzDLM7E4zWxz6m/ammWWEzo03s7fNbKOZLTGzc0Plr5nZtyLuca6ZvRlx7Gb2HTP7HPg8VHZ36B6bzWyWmR0cUT/ZzK41s4Whv5+zzKyfmf3GzO6M/CbM7J9m9r14/IBE4k0JnkgCmdnhwI+B04DewGLgidDpo4AJwJ5ALnA6sC507gHg2+6eBYwAXm3DsEVERBpyNvCn0OtoM+sZKv85sB9wENAd+AFQZWZFwL+AXwH5wCjgw2a09zXgAGBY6Pj90D26A38GnjKz9NC5y4EzgGOBbOA8YDvwMHCGmSUBmFkecATweHO+cZHdhRI8kcQ6E3jQ3T9w9zLgGuDA0BCXciAL2Aswd//E3VeErisHhplZtrtvcPcPEhC7iIhImJmNB/YAnnT3WcBC4OuhxOk84LvuvszdK9397dDfvTOBf7v74+5e7u7r3L05Cd6P3X29u+8AcPfHQveocPc7gTRgSKjut4Dr3H2+B80O1X0P2EQwqQOYCrzm7qta+SMRSQgleCKJ1Ydgrx0A7r6VYC9dX3d/Ffg18BtglZndZ2bZoaonE3wCudjMXjezA9s4bhERkdrOAV5y97Wh4z+HyvKAdIIJX239GiiP1pLIAzO7wsw+CQ0D3QjkhNpvqq2HgW+E3n8DeLQVMYkklBI8kcRaTvBpJwBm1gXoASwDcPd73H0/YDjBoZpXhsrfd/cpQAHwd+DJNo5bREQkLDSf7jTgEDNbaWYrgcuAkQSnIJQCA+u5dEkD5QDbgMyI41711PGIGA4GrgrF0c3dcwn2zFkUbT0GTDGzkcBQgn9bRdolJXgibSvFzNKrXwQTs2+a2ajQctK3A/9190Vmtr+ZHWBmKQT/yJUClWaWamZnmlmOu5cDm4HKhH1HIiIiwblwlQTnwo0KvYYC/yE4L+9B4C4z6xNa7OTA0N+9PwFHmtlpZhYwsx5mNip0zw+Bk8ws08wGAec3EUMWUAGsAQJmdgPBuXbV/gDcamaDQwuZ7WNmPQDcfSnB+XuPAn+tHvIp0h4pwRNpW88DOyJeBwPXA38FVhB8sjg1VDcbuB/YQHAY5zqCk9QBzgIWmdlm4EJ2DSsRERFJhHOAP7r7V+6+svpFcKrBmcDVwEcEk6j1wE+AJHf/iuCUgytC5R8S7PUD+AWwE1hFcAjln5qI4UWCC7Z8RvDvZik1h3DeRfDB6ksEH44+AGREnH8Y2BsNz5R2zty96VoiIiIiIh2YmU0gOFSz2N2rEh2PSEupB09EREREOrXQdIjvAn9QciftnRI8EREREem0zGwosJHgYjC/THA4Iq2mIZoiIiIiIiIdhHrwREREREREOohAogOIRl5enhcXFyc6DBERibNZs2atdff8RMfRXujvo4hI5xHt38h2keAVFxczc+bMRIchIiJxZmaLEx1De6K/jyIinUe0fyM1RFNERERERKSDUIInIiIiIiLSQSjBExERERER6SCU4ImIiIiIiHQQSvBEREREREQ6CCV4IiIibcDMHjSz1Wb2cQPnzczuMbMFZjbHzPZt6xhFRKT9U4InIiLSNh4CJjVy/hhgcOg1Dbi3DWISEZEOpl3sg9daby9Yy7tfrufyiXsmOhQREemk3P0NMytupMoU4BF3d+BdM8s1s97uvqJNApTd0uertrBhezlj+ncPl20uLee1+WuYPLIPAKXllZz94Hs8dv4BpAZ2PbsvLa/k2TkrOHnfvvz9w2UkmbF43XaO26c3KzeVMm5QXrjui3NXMrool7nLN7Nnzyz65mawanMpc5ZuYuKwnjw7ZznjBubx8ier6J2TTp/cDAbkdeHKp+dw3rj+DOuTDcCnKzcza/EGvlyzjf87bBDdu6SG2/hwyUYCScb2nZXkZqawpbSCZ+csZ9mGHVxz7FD653Xh3tcWUtgtgz17ZrGltJyS4u6s3lzKjdPn8ovTR5GekszPX5xPRmoyi9Zu4/C9Cthvj2689tkaTtm3kCUbtvObGQvYVlZJTmYKt5+4Nz954VOmf7iciqoqTi/pR8+cdAq7ZXL3vz/j6OG9+NrovhRkpfHXD5Yx49PVZKUH+Pcnq7h76mjGDcpjzZYyfvTcPK49dihvfr6WrukBvv3orPD3lRZIoqyiioe+uT9vfr6WbTsr2VZWwfTZywHo1z2DYb2zeXHuKvrmZrCjvJLrjx/KZX+ZTVZagHevPYId5ZWc+8f3+HjZZg4a2IO3F65j0vBe7NMvh1P2K6QgK50Zn67mmw+9z4Q983njszUAPDFtLFVVTkogia5pAd74bA0fLtnIso07uOaYobzyySr+8OaXHDw4j/98vjYcc7fMFDZsL2fCnvksWLWF5ZtK2bcolw++2shevbKYNmEAby5Yy98+WAbA5JF9mLp/P26cPpfPV28N3+fJbx/Iab9/h3GDerB331x+9/pCAN648jAm/GwGd502ksufnE1qIImdFVV1/vvumhZga1kF8380iZumz2Xh6m28t2h9+PxevbLYpzCHzTsqeGHuyhrXji7KZdOOcr5Ys42s9ABbSisAyEoPUF5ZRXpKMhu3l9dpc69eWXy6ckuNsqOH98QdXvl0NT2z0rh5yggueGQmYwd0590vgvEM75PN3OWbGTugOwf078Hdr3xe4x7987rw5dptfHvCAD5atoktpRV8tGxTONZn/m9cnVjixYJ/R+JwY7MHgeOB1e4+ota57wM/A/LdfW1910cqKSnx1mzketdL8/nVjAV8+ePjWnwPERGJPzOb5e4liY4jXkIJ3rO1/y6Gzj0L3OHub4aOXwGucveZtepNI9jDR1FR0X6LF2tv+I6s+OrnAFh0x67PMBc8MpOX563i5csmMLhnFgf/9FWWrN/BkJ5ZvHjZhHC9m6bP5aG3F3HJ4YP41asL6ty7+p6l5ZXsdf0LDOmZxfxVW8hODzDnpqOZ8NMZfLV+O//5wWEc/NMZda7/07cO4Mw//LfGvarjBSjZoxtPX3RQne+lIYvuOK5OnciyMw8o4oqjhrDvrS/Xe/1PT96HH/x1To2yv39nHF/7zVuNttuvewY3Tx7OeQ/V/ay56I7jGH7DC2zbWdnoPVqjd046FVXOmi1l9Z4fVNCVf19+SJM/v/Zs8sg+4YS4o3ru0vEM75PTqntE+zcynkM0H6KeoShm1g+YCHwVx7ZFRETaG6unrM5TWHe/z91L3L0kPz+/DcKS3c3yjTsAKC0P9ogsWR88XrBma416qzaX1vjakMqq4H9mX63fDsDmUE9I9XFZRf3JTXWPSUOWbtjR6PnmWrW5jPLKur1A1Tbu2FmnbFtZ4zFC8OfX2PcSz+QOYMWm0gaTO4AloX+HjmxxJ/geq/9/bQtx68GD+p9UmtnTwK3AP4CSturBu+fVBTWefomIyO6nk/fg/R54zd0fDx3PBw5tbIhma/8+thdL1m/n+n98zG/P3JfM1ODskorKKi594n9cdMgg9i5s3lPxu17+jH7dMjhlv0KOvOt1Fq7ZxtkH7sEtU2r+s/xz9nI+WraJa48dWqO8uiflv9cewZVPz2HS8F5c+8xHFHbL4MYThnPBIzP5ycl78+ycFZw4ui8ffLWBH31tbwDcnf7XPM/J+xZy5tgiTvrt25Ts0Y0rjx7CmwvW8qtXF/C1UX345dTRvPDxSi58bFaNtk8rKeTJmUsBGJDXhRsnD+ecB99r1vdf7cihBazbtpPM1GTeWrCuRfcQkegct3dvfnNm69bO2h168Oows8nAMnefHUXdaWY208xmrlmzprUNt+56ERGR+JsOnB1aTXMssEnz74J+/K9PeG3+GmZ8uuvzwOL123n+o5V894n/Nft+97zyOVc+HRzKt3DNNgAeeafuUNdLHv8f973xRYP3+e2MBbzx2RqufeYjINhjdcEjwYT7qr9+xH8+X8vlT87msXd3DVoKdZbx1w+WctJv3wZg5uINnP3ge+FhlH//MDhUrXZyB4STO4Av1m5rcXIH8O9PVvO/rzYquRNpA8991Ha/zttskRUzywR+CBwVTX13vw+4D4JPKOMYmoiISNyZ2ePAoUCemS0FbgRSANz9d8DzwLHAAmA78M3ERLr7iddgozgOYmq23SgUEWnn2nIVzYFAf2C2BXvUCoEPzGyMu69s9EoREZF2zt3PaOK8A99po3B2W9XDH39+6khO2a+wxrn6BuQ4cPidr7Fpezmzrp8Yvsek4b1Yv20n7y1az4PnlnD4Xj1ZsWkHB/741fC1A659vt4YTvvdOzVW8rvw0Vl1VvADeLieXr+GXPHkbP76wdIGz9deYfDoX7wR9b1FRCK12RBNd//I3Qvcvdjdi4GlwL5tkdxpgKaIiEj78vf/LQu/b6qn7Ys121i3reYCGy/MXRlO0l74OPhR4/NVW+tcW5/I5K76Xq3VWHJXn/mrtjRdSUSkHnHrwatvKIq7PxCv9kRERKTtrd5Sysbt5ezZM6tF13+6cjN5XdNYvbmM3jnp4XKvZ9DijE9Xs2JTKbkZKazdGlx18Mu122rEUpCVXue6FZtKmb9yC7+eUXergEjj7niV8RF7w4lIfBhVBKgimUoCVJJMVfhrMlUkW83yOnVt17mkiGt3HVeRZLXuGbpHjePwfarC90mqdZ9kq+e6WscBi2i3dhyh4zsrTgXaZsHHuCV4UQxFKY5X2yIiItI2xt3xKuWV3uKVqif98j/kZKSwaUc5fXMzwuWRvXbVyd5TsxrvBRt3x6t8ftuxdcr/8/lajv5l00Mel23cwV9mLokycpHdhZNKBWmUk0p58L3tJJWKiONyUqggQCUpobopVJJkVaRQSRoR9a2CFCpIjaifYpUEQmUpVJBCZSjhqiLVgu2mhOqmUkGKVZBcI9nxULJVGUq+dp9Zp+UejLSCJCpJCkWZtOvltY5D5ytICl0XvL7MU9gROt5Vz8L119K6PfCaoy3n4CWcu2NaUVNERCRmyitb/0Ft045yIJhgJToWkaZ5jeQpJZTQ7Ep+qt9XhsrLSaOcdHaGE63qhCzNdgbLa50PUEWACtLZSbqVkxaqk0pFKKHaldClWdN7/TVHpRs7SaGcZCpIppwA5QSo8OD7nQQoD/VhVZDMdk9jI13ZSYAKkoNfqwKhJGhXclQRToySqfTkUEIV+TUi0fKa5fVdX5081ddGzcSsZqIV2Y637YYCbaZTJHjK6URERBLnpbkrqXJnRN8cxv9kBj8+aW9unD6XeTcf3eA1by9cx+otpfzo2U94ce6qqNuqXqRFOhMngzK6UEa67SSNYNJUnTClRxxXJ0vplNeomxY63lW3OvkKJVcWTNIyKCODMpJj2AO1w1MpI4VSUinzlFBCFUygSkml1FPYRBfKSGEnAXZWBb+WkRo8JoUyD4TOp4SOU8Lnwl89mIBVhO5bHk60kkP3C15bSXLMvjdJjE6R4ImIiEjiTHs0uJ9bVnrwY8c1fwvuG/fqp6sbve7e1xYyffby+AYnbczpQilZbCfbttOFUrpYKV0oJTP0PpNSMq2MdHaSQRmZlJFhwfddbUfwPGXh67pQ2uIhf6UeTKyqk6vq96WkstUzWEcOpaSw01NCQ/DS2E4aOzw1nDxV93Dt9EC45yuyrJxA8P6hxKv6XGkoQdNygBJrnSrBc1dvnoiISKLs2FlZ47iqic/km7aXxzEaaYlUyslmO1m2PZykZYWOsyO+Ztc+H/E+2t6vUk8JJlOkscPT2EEq28hglXdjO+lsr0pjGxlsJZ1tns520kM9Xrt6xEo9NZxc7TpOoSxUpuRKOqJOkeCZ/ucVERGJ2urNpYy5/RWemDaWsQN6hMtfmruSaY/OYs5NR5GdnsIrnzQ8dLJ6qGQgadff4IpaGd2Fj81qNI6/RWyVILEVoIJubKWbbaGHbSaXreTaVrqxhe62JVjOFrqHznWxUrpSSpo1nnRXubGVDDaTyRbPZDOZLPcebKEfm6syI8q7sMUz2E46W0PJ2TbS2eYZoaQutcPOjxKJt06R4ImIiEj03l+0AYBH3llUI8Gr3mbgyzXbGNkvt8ltB6BuUifxkUQVPdhMgW0kP/TKZSvdbCt5bKKHbaaHbSaHrXS3LeTY9gbvtdXT2eBZrCeLdZ7NQvqwrSqdrWSEk7b6vm4hk62kKzETqcfNk4e3WVudKsHTnxgREZGmVW9LUF7pvLNwHQcO7FHj/KYd5WzaXs7GiCGU736xjuQkIzM1mczUTvXxIq7SKSPfNlLARvJtEwW2IZjEEfHeNtGDTfUOfdzpyawjh/UeTNa+ooB1VdnhBC7y6wbvyka6UkZqAr5TkY4tNzOlzdrqFL+BNe9ORESk+V6et4qX563ixyftzRljisLlZz/4Xp26U+97ty1Da+ecXLZSYBtDydrGXe8jyvJtI9lWd+uICg/uqbXac1np3ZlTNYDV5LLGq185rCaXjd6VrWSgeWYiiRdIarue7U6R4ImIiEjLvffles4YU1Rj83FpXBd20M/WUGSr6GerKbS19Lb19LPV7GGr6Gqlda7Z6umh5Kwbn3gRb1Ttw2rvxppQMrfGc1ntuawnS8MgRXYTeV1T+fqYIu55tfEh68lJbfegpVMleO6OnmKJiIg0jxl8vGwTHy3blOhQdhvJVNLb1lFsq9gj9Cq2lRTYBgptLXm2uUb9LZ7BSu/OUs/jvaq9WOIFrPJuwcSNHFaHVoYUkfZlwp757Ffcvcl6KclK8GJKKZ2IiEjLGcbxv3oz0WG0uWQq6WerGWAr6Gdrwr1vA205/WwNKbZr24dST+ErL2Cld+elqj1Y4j35ygvCr010QZ9IRGKne5dUzj2omLte/qxV9/n6AUX8+b9fhY+vPmYv7vjXpzXq7NmzK5+t2srpJf34y8wlNc4Fkox9+uY02c7oom6tirM5OkWCJyIiItGrPRSzDUcWtZCTz0YKbS29bD19bB0FtiG0L9sOstjOTgL8rOJ0PvN+Na5MpyzcA1dkqym2leGvfWwdAasK193uaSz2AuZ7P16oGsMi78niql4s8p6sJlfDJqXT6ZubwbKNdeeJAowszGH20l29/u9dewRjbn+lyXt+fPPRjLjxRQAW3n4syUkW3nal2qI7jgu/P2vsHoy+9eV67/Xlj4/FQotxVN8jNzMlvEDUpYcP4vKjhnD7iXuHz194yEDOGrsHw298kYyUZD65dVL4fuu2loUTvH7dM1iyfgfJSUa3Lqk1Ypr86zeZE/reX/v+oRTndWny+44lJXgiIiISdudL83nuoxU1yp6atTRB0dTHKbLVjEn6lH3sC4YkLWGILSHXttWoVeYpbArttbaFTIbYEiamfcCXVT3ZQToBKsixbfS0jTWu2+BdWewFfOiDmF51EIu8F19U9Wax92Qd2agXTmSXjNTkOmVd0wJsLasgPaXuuWh4xBOmaKZXNbaYonvd85EPsJIaeHqVFLqoS1rT30N932fkXZMSsNpjp0rwNDdcRESkcb9qYqGARMiklEOSZjMxeRYHJc2llwX36dvsGcz3fjxXOZbPvJCvvIAV3oMV3r3OkMgebOKk5P8wOmkBKVRSTjJbqzJYHBpKuch7sdgL2EzXBH2XIvW7e+oorvrrHErLq5qse+EhA/nd6wsbrfPdIwbzz9nL+foBRYzom8PU+97l4sMGRbWvZbX8rDQmDuvJNcfsxZjbXmFH+a7hyn//zjhem7+aE0b24fq/f8xL81YBNffEzMlIwd25/+wSTr/vXabu348n3g/2jEV+Xg8kB3vFb5kynBv+MbfeWHIz627rccp+hezVK6tGAnfk0AL+/clqAPK6prF2axnHjOgdPn/XaSPp1z0TCCauPzx2KEcMLahx3x5d08LvJ4/sQ5XDdw4bVDeoUFJ35NCe9OueUW/c8dQpEjxtkyAiItK+BKjgsKQPOS35dSYkzSHNylnvXXmragT/rRrKf6uGssD7RD0sch053F95PFQ2XVc6nmgSn6bUHnIYKT0lKaoELNJ+e3RjS2k5n63a2mCdBbcdQyA5idc/W8PfPlhW41x1T1m1cw8q5upj9uJfH69g8bpdG9kPzO/CwjW7ergvm7gnl03cM3y86I7jWLJ+e9QJXuRQRIBPbp1UYwjloIKuDCoIPii57+yS8LnSiCRw3KAe/PbM/Wrc77k5K9hSVlHvar1nH1jcYIIHsHffHD5atomxA7rz7hfrGVTQlW8dPKBGnZ+fOpJRt7yMu5OdEWDt1jKSI359nLRvYY36F0yoeX1thnHVpCENnAv6v8MGhoeItiUNFhcREZHdRh6buCLwJG+nXcr9qXcxMmkhf6o8gqk7r2P/snu5pPxSHqucyOdeqDlvErXiHplN1slOD4QTk/qMaGQhjfGD8uqUNTV3dVjv7HqTmfrygRF96rY9uiiXPjm7Vl6tvm5kYW6t+zWdYGTWM9Qy1iKHMg4uyKpzfr/i4CIkLVltckTfbGDXQiZF3ev+e6cGgr8v9t2jW/hn1CWt5X1d/RuZV1cdT25G221uHqlT9OBV0/49IiIiu6dCW8O05Gc5Lfk1Uqng1apRPFF5ODOqRlFJ/D98SuKcul8h+/fvzg+entPie7z3wyPISEnm9uc/4fH3ltQ5/7XRfRmQ35XF67ZxZT3tPHb+AQztnUUgKYkFa7awfWclW0sr6JWTTnmls0ePTNIDyfwpYrXFSL8/q4SB1z4fPr7vrP0YXdSN/W/7d7js3IOKeejtRQDcc8ZoJg3vxXH3/KfGfZ75v4Pon9eFUbcEFw2p/uh63vj+/O71hdxx8t7kdU1jZ0UVe/XOpryiinMfep/ZS3bNJf3pKftwakkhL81dxaPvLo5q1miPrmlMv3gc3/nzByxZH1w05d+XH8KRd70OwAvfO5iv1m2nqIFE+YKD+3P/f77k/PH965x786rDAOiTm8HfvzOO7WUVHDCgR516v/n6vny5dhuZqfWnJ//5wWEs27iD4h51E6ubJg/n9P2LGFmYw2FDCti/uO6KlZmpAZ69ZDz987qQnGScc1AxvXOaP3xy3i1H88R7Szh5v8IG69xw/HBO2reQAfmJGfLdKRK8RHSNioiISNOy2colgb9zTnJw1by/Vk7g95XHs8h7N3GldBRnjt2DUf1yW5XgFWQFe7J+fNI+9SZ4ZjCmf3dWb6m7wTzA+MG7euD226P+Pc0ih0MeNaxneG4Z1N3EemBBV/Kz0mqUnbJfYTjBmzyyT53752elhXugAklWY84awHs/PLJuUGlwwj69mb1kIxZK5dJTkjl4cD75WWnBBC/Kj8H7FOaS3zUtnOBF9mbu1SubvXplN3ht9Ty46l6ySIXddiWFo/rl1jlfrUtaoNFe0n7dM8Nz5GpLCySH7z2mf8N70kXev7FYGpOZGuC8ehLZSKmBJPZtw20Raovb2AYze9DMVpvZxxFlPzOzT81sjpk9Y2Yt+8mKiIhIu2ZUcVbyS7yedjnnJ/+Lv1UezISyX3JNxQVK7joZb2CIVfcuuxbPOGpYT745rrjG+QsPGRh1G9XJzz59W/7RMzLOE0f3Db8/dEg+QI1eo4KI5C41NNGrb24GuZkp9fYu1b7nrjabjuuggcHkdOKwnvVeaxF9eKc00utUn8EFXSns1nQv14TBwZ/B4XsVNFEzOkcP79l0JWlQPHvwHgJ+DTwSUfYycI27V5jZT4BrgKviGEMNrnU0RUREEm6ALeeOlPsZkzSftyqH86OKb/CJ75HosCRKi+44rs6+ZPsU5jBn6Sb26pXFpyu3hMvn/2gSQ657AYBfnD6Sy/4yO3xuZL9cZi/ZGP50Vvu+H1w/MXx839klANx4wvAa7Ta2cMqiO45j4LXPU1nl4flwRT0yaywSUvv7aEx1h1pWWoBj9u7NmOLuvLdoPReFEs2nLjyo3hgifXjDUQ3e/5gRvcLvmzP4bFif7DrtQESCZ7v+fb4xtvH/z6r/Lf560YEAvHz5IVHFsHdhTr0xtNTvzyqJ2b06o7gleO7+hpkV1yp7KeLwXeCUeLUvIiIi0bvrpfkcHfEBMz6c85Jf4KrAE+wglct3Xsjfqg5Ge8t1XNbIv231mXiukVAVunlMputUx9mG/7nGqnOi+SHr/8n2LJHLT50H/CuB7YuIiLQZM5tkZvPNbIGZXV3P+T3M7JXQNIbXzKx5Y6laoayiknteXcBx97wZtzay2cr9KXdxQ8qjvFG1DxPLfsbfqiagD5Ktk9e17h5gAN+utcT7zZOHc8yIXvzjO+NqlGc0czPq331jv3rLp00YwNHDe/LweWPom1tzSN9Fhw7kgXPq9sjcedpIpozqwz6Fu+ZFXXfcUAYXdOXnp44EgoufnDeu4flOj18wlnMObLhX6qlvH8iZBxQ1uKLleeP689j5BzR4faTsjABnjOnHo7XqtyYF+/XX92WfwhyO26d3jflh1Ylxa5LfPXt25cTRfbl76mjuPG0Uk0f2YXifhufRQXArgckja/6bSPuTkEVWzOyHQAXwp0bqTAOmARQVFcWkXa2iKSIiiWBmycBvgInAUuB9M5vu7vMiqv0ceMTdHzazw4EfA2e1fbSxN9y+5Pepv6CADdxcfhbayekxAAAgAElEQVR/rJyEErvYeGLaWI6864065dccO5Rrjh0aHoJ4zkHFnHNQcY15ZIvuOI6qKmdAxOqPANcfP4xbn51HZmoy23cG9y7rm5vBW1cfXqedAwf04J0v1tEtMzU8rO6tqw9n0LXPhxcJuWrSXgA887+lNa4dmN+Vu6eOrlH2rYMH1Ni/bPzgvBoLoNRpf2APDhxYd0XGaiXF3SkpbnjRjRtOGNbgudrMjB+ftE9EQdSXNmhIryymXzy+nsZaf+9AchK/OH1U+PieM0Y3UjtoYH7XqOrJ7q3Ne/DM7BzgeOBMb2hWLeDu97l7ibuX5Ofnt7LNVl0uIiLSWmOABe7+hbvvBJ4AptSqMwx4JfR+Rj3n26WJSTN5KvUWAE7deSN/rDwGJXexkxZo3RYS9X1G6hFa3CRyOfrae35VX9c3tABH7f3EqutH3j83o/7exlgrqLV6ZdzEseNgQD0/P5FotWkPnplNIrioyiHuvr0t2xYREUmgvkDk2u1LgdrjwmYDJwN3AycCWWbWw93XRVaKxwiX+HDOT/4XPwz8iTk+gAt2XsEatHh2pOuOG8qPnvsECA6hLOqeyS/+/Rlzlm6K6vpLjxhMv+6Z/O4b+/Gvj1fwjw+X16nz2vcPZfWWsgbvETk37Wuj+nDokAKmjOpDVnqAkf1yWbmplHMefI8Hz92/xnVvXHkYSzZsZ1S/XA7ZM7/OkvN/vmAsHy3bSEryrr6EQ4fkc/fUUQwuyGJHeWVU32NL/POS8TUWemmPHvvWAcxZurHVCbx0TnFL8MzsceBQIM/MlgI3Elw1Mw14OfQL5V13vzBeMYiIiOwm6nsOX/v5//eBX5vZucAbwDKC0xlqXuR+H3AfQElJSUz6EBpbCKNlnCsDf+E7gek8XzmGy8svopQ26lXZzT14bgnnPTQToMbQwnMOKgbg89Vbok7w9i0KJlWTRvTizQVr6q1TnNeF4ry6G0PXJyU5ia+Fluo/Ymhwmfq8rmnMun5inbqRe5Kd0MCebofvVXOpezNjyqi6WwHEWs/sdHpmp8e9nXh2ROd1rfvzE4lWPFfRPKOe4gfi1Z6IiMhubCnQL+K4EKjR3eLuy4GTAMysK3Cyu0f3SX+34twQeJTzAi/wp4ojuK7im3hC13RrO/26Z4Q3iW5IUcRGzbU3wobgZtPR2qNHdIlbfU6qZ881DQdsniP2KuC9L9fXWVRGJNGa/I1rZhebWeK2Yo+B2D+ZFBERaZb3gcFm1t/MUoGpwPTICmaWZ2bVf5evAR5s4xhjwLkt8CDnBV7ggYpj+GHFeZ0muTtkz3xevuwQxg7YtaDHa98/lNk31tz3rKh7F2ZddyT/u34iBVl1e5nGDujBbSeOAGB4n2zeueZwnrs0uAhH5MbfH1w/sc68uGiYGbNvPIqfnrJrsZBrjw0ugqLPS80zbcIA/nf9xHBPpsjuIprfur0Irvb1ZGiJZ/3fLyIi0gzuXgFcDLwIfAI86e5zzewWM5scqnYoMN/MPgN6ArclJNgWc64J/JkzA69wb8UJ3FrxDTrTYipd0pJJT0mma8RiI8V5XcjJSKlRr8qdHl3T6Nal4QVHqhc5KeyWQe+cDLLSgvfITN01H6t7reubs1J4TkYKgYi5cVnpKY3UloaYWaP/jiKJ0mSC5+7XAYMJDq88F/jczG43s4Fxji3mtE2CiIgkirs/7+57uvtAd78tVHaDu08PvX/a3QeH6nzL3RteGSPGYvHo9qLkf/LtwHM8XDGRn1RMpaMmd5ccPggIDrU8NzRvDuDCQ4Ifi846MFjW0M80NbnmR6/DhuRz+cQ9o2rbDH52yj4M7d3wXmaF3TIYN6jhbQOaur+ItH9RzcFzdzezlcBKghO+uwFPm9nL7v6DeAYYC/qFJSIiEj9Tk1/lqpQn+HvlQdxUcQ4dNbkDuOKoIVxx1BAAnp2znIfeXsSxe/cKz50bXNAVgF71LPKx6I7j6pT98ZtjGmyr+sG0R6zHc2pJP04t6dfAFfDtQwZy1tiGN/5urB0R6RiaTPDM7FLgHGAt8AfgSncvD80T+BzY7RM8ERERiY+Dkj7mR4EHmVE5ku+XX9hp5tzBrjlrkQlSclKwLLtVwx5rJsjV96w93DNS9T50aYHm//yrr6m9l52ItE/R/J+cB5zk7osjC929ysyOj09Y8eHx3JFSRESknWppf1uxreC3KXez0PtwSfklVLTt9roxddiQfGbMX8NBA3vwncMGceYf/gvAP74zjlN+9zYnjOzDkUNrL/tf9z49s9O56YRhHDW8V7js/rNLyM1sfsJX/amlsFsmNxw/jGP37t1g3e8dOZistEC9q2M2ZcqoPqzYtIPzxvdv9rUisvtp8jexu99gZvua2RSCv2vecvcPQuc+iXeAsdBxB4qIiIi0Xksef2azjQdSfk4lSZxf/n22svuuJPj0hQdyyu/eafD8RYcOZOyAHsyYv4bkJGPcoLzwuZH9cvn8tmMbvX/tIY7njquZKE0c1vr9zJpKvjJTA1xyxOAW3TuQnMTFh7fsWhHZ/USzTcL1wMNAD4K9eX80s+viHZiIiIjsrpw7U+6ln63mop3fY6kXJDqgRjU1F39wQVd6Zgf3pBvZjH3oqvc/G9G34UVPRETaWjQDtb8O7O/uN7r7jcBY4Mz4hhUfmkQsIiLSet9Kfp6JyR9wW8U3eM+Htlm7/732CLLTdw0+GtbAapLTJgzgb/93UJ0VKxty4ui+7NUrm+cvPZjLolzREoK9e89dOp7/O3RQ1NdEQ4vDiUhrRPObbxEQuRRUGrAwLtHEiX5RioiIxMa+9hlXBZ7gX5X783DlUU1fEEM9s9PJz0oLH19xVP3JWP+8Luxb1I3h4Z61xj8IVG/xO6xPdnhBk2gN75NDUjOviZYeTItIS0ST4JUBc83sITP7I/AxsNXM7jGze+IbnoiIiOwuctjKr1J/xQrvzlXl00jELHeLeGq7d2FOvXXGDgjuA1e9T92AvC4N3u+KZvTYiYi0B9Esd/VM6FXttfiEEn96ECYiItJyt6b8kXw2cnL5zWym4aQplhbdcRzFVz9Xp/ylyyZQkJVe7/n+oYRuyqi+TBnV+KqSLV2YRERkdxXNKpoPm1kqUP2Ia767l8c3rNgyraMpIiLSKsclvcvk5Hf4WflpfOQDEhZH9V/0jjx8cdenlg78TYpI3ESz0fmhBFfRXETwd04/MzvH3d+Ib2giIiKyO8hnI7emPMiHVQP4XeUJcWvnB5OG8NMX5jO6KJeTRvdl285KAG4/cW+6d0kFGp9X/9j5B/Dpys2NtnHr10bQOzud3MwU3lm4Lmaxx5Jp8QARaYVohmjeCRzl7vMBzGxP4HFgv3gGJiIiIrsD5/aUB8ikjCvKL6KS5Ji38N0jBodXr6xvRcqvH1BUT1R1e7fGD85j/OC8OuWRzhq7R/h9SXH35oYqIrLbi2aRlZTq5A7A3T8DUuIXUvx4Rx7PISIiEgcnJL3DxORZ/KziNBZ64/PZWqpXTnrTlUIG5ncFIDMlmmfUIiKdTzS/HWea2QPAo6HjM4FZ8Qsp9jTSQUREpPmy2cYNKY8yu2oAf6w8ptnX53VNY+3Wshplt504gh8+83GNstNL+kV9z5+fOpLT9+9HUY/MZsfTHM9dOp4uqYlNIvVcWkRaIpoevIuAucClwHeBecCF8QxKREREEu/7gSfpzmZ+WH4eVVF9ZKhp36LcOmXVWxhEas4+cl3SAhw6pKDZsTTX8D45FDeyvUI86bm0iLRGo7+tzSwZeMDd73L3k9z9RHf/hbuXNXbd7koPwkRERKIz0hbwjeR/80jlUXzcwlUzjxzWM8ZRNdLW0MaTvoObmJu3OxlUEByGOrENf34i0nE0OvbA3SvNLN/MUt19Z3NubGYPAscDq919RKisO/AXoJjgqpynufuGlgQuIiIisVF7KGASVdyW8iBryOHOilOjukdWWoD3rzuSva5/AYB5txxNZmqAHzw9J1znt2fuG37fNzeDZRt3tD544NNbJ5GS3HgP40PfHEN5ZVVM2ou34rwuzLvlaDJSYr+gjYh0fNGMt1gEvGVm15vZ5dWvKK57CJhUq+xq4BV3Hwy8EjoWERGR3cjpyTMYkbSIW8vPYivRz3VLj0hIMuuZv5YW2PWxIzXQ/CGfjbWb3MQwz+QkqxHf7i4zNaDtEkSkRaKZPbw89EoCskJlTY52dPc3zKy4VvEU4NDQ+4eB14CrooghJjRZWUREpHFd2c7lgad4r2oIz1aNjfq6e84YXW/5OQfuwdqtO1m4ZisHDOjBqs2lQHCe2cWHDWo3vWoiIu1FNAnePHd/KrLAzKIbr1FXT3dfAeDuK8yswQHzZjYNmAZQVFR3/5vm0BMwERGR6FwUmE6+beb8nVfSnOU+Dtur/j/pN08ZUeN4VcT77x89pAURiohIY6IZH3FNlGUx5e73uXuJu5fk5+fHuzkREZFO66NlmwDoyxq+lfwv/lY5njk+MC5tpYbmynXvkhqX+4uIdHYN9uCZ2THAsUBfM7sn4lQ2UNHC9laZWe9Q711vYHUL79MyGqIpIiJSx+3PfwLAVSlPUIXxs/LT66138+Th3Dh9bp3yf18+Ifz+6QsPpKKq4T+4/bpncsdJe7fpCpsiIp1JYz14y4GZQCnBjc2rX9OBo1vY3nTgnND7c4B/tPA+zaIBmiIiIg1zd4bbl0xOfof7K49lBXX3qgM456DiOmX3nrkvgwqywsclxd3r3esu0tQxReR1TWtVzCIiUr8Ge/DcfTYw28z+7O7lzb2xmT1OcEGVPDNbCtwI3AE8aWbnA18BLZ3LJyIi0q6Y2STgbiAZ+IO731HrfBHBBchyQ3Wudvfn2yI2B64IPMVG78L9Fcc369p+3aNfZVNEROIvmkVWxpjZTcAeofoGuHvju566+xkNnDqiWRGKiIi0c2aWDPwGmAgsBd43s+nuPi+i2nXAk+5+r5kNA54nuG9s3O25cx6HJ3/IHeVT2RLltgin7lfI/x02iP55XeIcnYiINEc0Cd4DwGUEh2dWxjec+HJNwhMRkcQYAyxw9y8AzOwJglsHRSZ4TnCeO0AOwakSbWLq1kdY49k8XHlU1NeYoeRORGQ3FE2Ct8nd/xX3SOJIuySIiEiC9QWWRBwvBQ6oVecm4CUzuwToAhxZ341iuY0QAF+8zuiKOdxccRY7SG+wWnZ6NB8ZREQk0aLZJmGGmf3MzA40s32rX3GPTEREpOOo71Fj7WElZwAPuXshwVWsHzWzOn+nY7qNkDvMuI0V3p0/VzY+g2LOTcH11X5y8t6ta1NEROIqmsdx1U8YSyLKHDg89uHEl2uEpoiIJMZSoF/EcSF1h2CeD0wCcPd3zCwdyCOeWwqVboKkAL+qOJEyotuXLik0LCY5KZpnxCIi0taaTPDc/bC2CCSeNEJTREQS7H1gsJn1B5YBU4Gv16rzFcGFyB4ys6FAOrAmrlFl5MK5z/H4Nc82Wu17Rw4Ov58yqi+frNjCd48Y3MgVIiKSKA0+fjOzX0a8/26tcw/FMaaYe+6jFQCs2FSa4EhERKQzcvcK4GLgReATgqtlzjWzW8xscqjaFcAFZjYbeBw4170Nxp6Y4U3M2PjekXuG36cGkrjhhGHkZKbEOzIREWmBxnrwJkS8P4fg3j3V9olPOPHx/qINAHy6cjPD+mQ3UVtERCT2QnvaPV+r7IaI9/OAcW0dl4iIdCyNJXjWwHsRERHpBGZ8/1C2llYkOgwREWmGxhK8JDPrRnAYZ/X76kQvOe6RxYEWWREREYme9rkTEWl/Gkvwcghubl6d1H0Qca5dpkrtMmgREREREZEoNZjguXtxG8YhIiIiIiIirdSsTWzM7KY4xSEiIiIiIiKt1NxdSic3XWX31RarTYuIiIiIiCRKcxO8dr2aptI7ERERERHpyJqb4O0XlyjaSLvOTkVEROJg5abSRIcgIiIx1GSCZ2Y/NbNsM0sBXjaztWb2jTaILebUgyciIq1lZheHtg7qEP775bp6y1/43sFtHImIiMRCND14R7n7ZuB4YCmwJ3BlXKOKF2V4IiLSer2A983sSTObZGYdcoDIXr2yEx2CiIi0QDQJXkro67HA4+6+Po7xiIiI7Nbc/TpgMPAAcC7wuZndbmYDExpYC5WWVyY6BBERiaFoErx/mtmnQAnwipnlA+1ywL6rC09ERGLAg8syrwy9KoBuwNNm9tOEBtYCZRVViQ5BRERiqMkEz92vBg4ESty9HNgGTGlNo2Z2mZnNNbOPzexxM0tvzf2ipV0SRESktczsUjObBfwUeAvY290vIrgQ2ckJDa4FOuT4UhGRTiyaRVZOBSrcvdLMrgMeA/q0tEEz6wtcSjBhHAEkA1Nbej8REZE2lgec5O5Hu/tToYefuHsVwfnq7d7A/C6JDkFERFoomiGa17v7FjMbDxwNPAzc28p2A0CGmQWATGB5K+8XlY45DV5ERNrY80B4PrqZZZnZAQDu/knCooqhf14yPtEhiIhIC0WT4FXPvj4OuNfd/wGktrRBd18G/Bz4ClgBbHL3l2rXM7NpZjbTzGauWbOmpc3VajsmtxERkc7tXmBrxPE2Wv/gM2G+XLs90SGIiEgMRZPgLTOz3wOnAc+bWVqU19UrtHfQFKA/waGeXerbV8/d73P3Encvyc/Pb2lzNe8Zk7uIiEgnZ6FFVoDw0MxAAuNplQff+jLRIYiISAxFk6idBrwITHL3jUB3WrcP3pHAl+6+JjRv4W/AQa24n4iISFv6IrTQSkro9V3gi0QHJSIiAtGtorkdWAgcbWYXAwX1Dalshq+AsWaWGdoc9gigTeYsaIimiIjEwIUEH0wuA5YCBwDTEhpRjCVp0rqISLvV5JCS0JPJCwj2tAE8Zmb3ufuvWtKgu//XzJ4GPiC4d9D/gPtacq9mt61BmiIi0kruvpoOvvpzekpyokMQEZEWimbOwPnAAe6+DcDMfgK8A7QowQNw9xuBG1t6vYiISKKE9m49HxgOhPdxdffzEhaUiIhISDRz8IxdK2kSet8ux25oyImIiMTAo0AvglsHvQ4UAlsSGlEMXXr4oESHICIirRBND94fgf+a2TOh468BD8QvpNj79iED+P3rXzC8T3aiQxERkfZvkLufamZT3P1hM/szwcXI2r1/XjyevQtzEh2GiIi0QjSLrNwFfJPgpq4bgG+6+y/jHVgsbdi2E4A5SzclOBIREekAykNfN5rZCCAHKE5cOLGjueoiIu1fowmemSWZ2cfu/oG73+Pud7v7/9oquFh554t1AHyweEOCIxERkQ7gvtCertcB04F5wE+ausjMJpnZfDNbYGZX13P+F2b2Yej1mZltjH3ojatSfici0u41OkTT3avMbLaZFbn7V20VVKxVz72r0j4JIiLSCmaWBGx29w3AG8CAKK9LBn4DTCS4tcL7Zjbd3edV13H3yyLqXwKMjmXs0dDfSRGR9i+aOXi9gblm9h6wrbrQ3SfHLaoYq15apUKPJkVEpBVCDz4vBp5s5qVjgAXu/gWAmT0BTCHY+1efM0jAatOuBE9EpN2LJsG7Oe5RxNnyjaUAPDtnBb/+eoKDERGR9u5lM/s+8BdqPvhc38g1fYElEcfVG6TXYWZ7AP2BVxs4P43QxupFRUXNCrwpeg4qItL+NZjgmdkgoKe7v16rfAKwLN6BxdLOyqpEhyAiIh1H9X5334kocxofrlnfPj0NpVNTgafdvbK+k+5+H3AfQElJSUxTskpleCIi7V5ji6z8kvr39dkeOiciItLpuHv/el5NzcVbCvSLOC4EljdQdyrweCxiba6+uRmJaFZERGKosSGaxe4+p3ahu880s+K4RSQiIrIbM7Oz6yt390cauex9YLCZ9Sc4CmYqUGfSgJkNAboB78Qg1Gbr1z0zEc2KiEgMNZbgpTdyTo/4RESks9o/4n06cATwAdBggufuFaHFWV4EkoEH3X2umd0CzHT36aGqZwBPuFY7ERGRFmoswXvfzC5w9/sjC83sfGBWfMOKreQkC88rcHfM6psKISIi0jR3vyTy2MxygEejuO554PlaZTfUOr4pBiGKiEgn1liC9z3gGTM7k10JXQmQCpwY78BiKdmMytBc9i1lFWSnpyQ4IhER6UC2A4MTHYSIiAg0kuC5+yrgIDM7DBgRKn7O3etdtnl3lpxkEFqLTH13IiLSGmb2T3atgJkEDKP5++KJiIjERZP74Ln7DGBGG8QSN4EkpXUiIhIzP494XwEsdveliQpGREQkUjQbnbd/Efmd5t+JiEgrfQWscPdSADPLMLNid1+U2LBEREQa3wevw1BKJyIiMfQUUBVxXBkqExERSbhOkeAlRwzR3LBtZwIjERGRDiDg7uE/JqH3qQmMR0REJKzBBM/MtpjZ5npeW8xsc1sG2VqRwzL/8J8vEhiJiIh0AGvMbHL1gZlNAdYmMB4REZGwxlbRzIpXo2aWC/yB4OqcDpzn7u/Erb143VhERDqjC4E/mdmvQ8dLgbMTGI+IiEhY1IusmFkBkF597O5ftaLdu4EX3P0UM0sFMltxryZFrquycM22eDYlIiIdnLsvBMaaWVfA3H1LomMSERGp1uQcPDObbGafA18CrwOLgH+1tEEzywYmAA9AcO6Cu29s6f2ibDX87s0FGkUjIiItZ2a3m1muu2919y1m1s3MfpTouERERCC6RVZuBcYCn7l7f+AI4K1WtDkAWAP80cz+Z2Z/MLMutSuZ2TQzm2lmM9esWdOK5kRERGLqmMgHk+6+ATg2gfGIiIiERZPglbv7OiDJzJJCG5+PakWbAWBf4F53Hw1sA66uXcnd73P3Encvyc/Pb0VzcNiQ1l0vIiISIdnM0qoPzCwDSGukvoiISJuJJsHbGJpn8AbBSeV3AxWtaHMpsNTd/xs6fppgwhc3x+zdK563FxGRzuUx4BUzO9/MzgdeBh5OcEwiIiJAdAneFGA7cBnwArAQOKGlDbr7SmCJmQ0JFR0BzGvp/aJhWkdTRERixN1/CvwIGAoMI/i3cY+EBiUiIhISzSqaBcAKdy8FHg4NRekJrGtFu5cQ7A1MBb4AvtmKezVN+Z2IiMTWSqAKOI3gImR/TWw4IiIiQdEkeE8BB0UcV4bK9m9po+7+IVDS0uubK8mU4YmISOuY2Z7AVOAMgg85/0Jwm4TDEhqYiIhIhGgSvIC776w+cPedoZ63dkPpnYiIxMCnwH+AE9x9AYCZXZbYkERERGqKZg7eGjObXH1gZlMAbSYnIiKdzckEh2bOMLP7zewI9AxRRER2M9H04F1IcL7crwn+IVsCnB3XqGJMIzRFRKS13P0Z4JnQ3q1fI7j4WE8zuxd4xt1fSmiAIiIiRJHguftCYGxoqwRz9y3xDyu2tIqmiIjEirtvA/5E8OFnd+BUgvu5KsETEZGEazDBM7NvuPtjZnZ5rXIA3P2uOMcWM+rBExGReHD39cDvQy8REZGEa6wHr0voa1ZbBBJPyu9ERERERKQzaDDBc/ffm1kysNndf9GGMcWeMjwREREREekEGl1F090rgcmN1WkPNAdPREQSzcwmmdl8M1tgZlc3UOc0M5tnZnPN7M9tHaOIiLR/0ayi+XZoBc2/ANuqC939g7hFFWP7FOYkOgQREenEQiNifgNMBJYC75vZdHefF1FnMHANMM7dN5hZQWKiFRGR9iyaBO+g0NdbIsocODz24cRHl7Sa3+bSDdsp7JaZoGhERKQTGgMscPcvAMzsCWAKMC+izgXAb9x9A4C7r27zKEVEpN2LZpuEw9oikLY0/iczeOWKQxiY3zXRoYiISOfQl+A+stWWAgfUqrMngJm9BSQDN7n7C7VvZGbTgGkARUVFcQlWRETar0bn4AGYWY6Z3WVmM0OvO82s3Y95PPV37yQ6BBER6TzqmwzutY4DwGDgUOAM4A9mllvnIvf73L3E3Uvy8/NjHqiIiLRvTSZ4wIPAFuC00Gsz8Md4BtUWtpVVJDoEERHpPJYC/SKOC4Hl9dT5h7uXu/uXwHyCCZ+IiEjUoknwBrr7je7+Reh1MzAg3oHFW+3HpiIiInH0PjDYzPqbWSowFZheq87fgcMAzCyP4JDNL9o0ShERafeiSfB2mNn46gMzGwfsiF9IbUQZnoiItBF3rwAuBl4EPgGedPe5ZnaLmVVvR/QisM7M5gEzgCvdfV1iIhYRkfYqmlU0LwIeDs27M2A9cG48g2oLVa4MT0RE2o67Pw88X6vshoj3DlweerW5vK5piWhWRERiLJpVND8ERppZduh4c9yjagNK70RERHZ58XsHJzoEERGJgSYTPDO7vNYxwCZgVij5a5dcPXgiIiJhPdSDJyLSIUQzB68EuJDgHj59Ce69cyhwv5n9IH6hxZfSOxERERER6WiiSfB6APu6+xXufgXBhC8fmEAr5uKZWbKZ/c/Mnm3pPVrDHeYu35SIpkVEREREROIimgSvCNgZcVwO7OHuO4CyVrT9XYIriSXMcfe8ydIN2xMZgoiIiIiISMxEk+D9GXjXzG40sxuBt4DHzawLMK8ljZpZIXAc8IeWXB9Lqza3JkcVERERERHZfUSziuatZvY8MJ7gNgkXuvvM0OkzW9juL4EfAFkNVTCzaQTn+1FUVNTCZkRERERERDqPaHrwADKAze7+S2CxmfVvaYNmdjyw2t1nNVbP3e9z9xJ3L8nPz29pc01yd278x8csWrstbm2IiIiIiIi0hSYTvNCwzKuAa0JFKcBjrWhzHDDZzBYBTwCHm1lr7heV4h6Z9ZbPX7WFh99ZzIWPNZpviojI/7d33+FRVOsDx7/vphJCAkhvhqIgKN2KiBVQ7OXaO5Zru96rXsGK7WfvvaLXi4qiXJGuCAgovXcChBpCgPSQZMv5/TGTzW62ECDJhuT9PM8+O3vmzMyZk92dvHvOnKOUUkqpGq8iLXiXARcDBQDGmJ2E6Vp5IId34XYAACAASURBVMaY4caYNsaYFOAa4HdjzA2Hur+Kmv7wmSHKYz2v3ZWnc+MppZRSSimljmgVCfBKjBX5GAB7cJUji8eD7NkQdJVvSPfT4h3VUx6llFJKKaWUqgIVCfC+F5GPgYYicgfwG5U0+qUxZoYx5sLK2FdYvz4Jn51LS/YGrFqyJcu7nJqZD4DT7WHupr3kFjmrvGhKKaWUUkopVVkOGOAZY14DxgA/Ap2Bp4wx71R1wSrVibeDx8XrMR8iePxW/bSkrNWutIfmK5PXcs0nc+k+Ymp1llIppZRSSimlDktFBll52RjzqzHmEWPMw8aYX0Xk5eooXKVp3AHOf5nTolZze9SksFldbg+fztpcTQVTSimllFJKqcpTkS6a5wVJO7+yC1Llet3AFHdfHokeTRfZGjRLdmEJ45enB6Qv2rKPeZsCu3cqpZRSSimlVE0SMsATkb+LyAqgs4gs93lsBpZXXxEriQjDnEPJIZG3Yt4njpKALN8t2MaO7P0B6Vd8+BdXfzL3oA43e8MeRs7RlkCllFJKKaVU9QnXgvcNcBEwzn4uffSpjmkNqkIWSTzsvIsujm08E/1l0DyvTlnn93pfQWAgWBE3fD6PZ35ZfUjbKqWUUkoppdShCBngGWNyjDFpxphrjTFbgP1Yswokiki7aithJfvD04N3XJdyTfQMroqaccD8vZ/7teoLpZRSSimllFKVoCKDrFwkIhuAzcBMIA0IP1JJDfeW60pmu7vxXPRIukpapIujlFJKKaWUUpWiIoOsPA+cAqw3xrQHzgHmVGmpqpgHB/9w3kcWDfgg5m2SyT+o7XMKnTw6Zjn7S9xVVEKllFJKKaWUOngVCfCcxpi9gENEHMaY6UDPKi5XldtLMveWPEBL2cuHMW8Rg6vC2/Z4diqjF27j2/nWaJxpewq4/csFFDkjG/AZY/h+4Tbyiyt+LkoppZRSSqnaoyIBXraIJAJ/AKNE5G04iGioBltsjmWY8w5Oi1rNc9FfYN1iGFpOodPv9bPjV3Pqi9N45pdVTFu7mz837mFjZj6PjV1RhaUObdGWLP49ZjlP/bwyIsdXSimllFJKRVZFArxLgELgn8BkYCPWaJq1wlhPf961B125I2pC2Lw9np3qbbUrlZ5T5A0Lb/tyIee8PpNv5pXl+WHhNk584Td25RQdVjm/nruFlGETcLk9IfMU2F1GM/OKD+tYSimllFJKqSNTuHnwOolIP2NMgTHGY4xxGWO+ApYCDauviFXvDdeVjHefzPDob7nY8WfYvMN/Cmydm7EuM2T+R8YsJzOvmFNenHZYZXxx4hoAilyhAzyllFJKKaVU3RauBe8tIC9IeqG9rtYwOHjI+Xfmmy68HvMhZzmWVMlxCkusnq3GGHKLnEEnVT+QIqebW0bOZ/OegsMuz4rtOQx8c2bE7tmbvWEPE1ekR+TYSilV3URksIisE5FUERkWZP0tIpIpIkvtx9BIlFMppdSRLVyAl2KMWV4+0RizEEipshJFSDGxDC15iDWmHR/GvMXJsqbSj/H8hDXsL3HTfvhEuo+YSr+Xfveuyyl08ukfmzDG6vD5Z+oeMnIDu3XOSd3DjHWZPD++YpOo5xe7vPss7+XJa1mfkc+SrVmHcDaH74bP53HPqMURObZSSlUnEYkC3gfOB7oC14pI1yBZRxtjetqPz6q1kEoppWqFcAFefJh19Sq7IDVBPgncXPIo20wzPot9jd6yvlL3/828rRz31GS/tGKXm2Xbsunx7FRemLiGvzbtBeC6z+Zx8v9N897PVxqjbcwM3XIn9vOsDXtIGTaB58av5vinp/DBjI2Veh5KKaUO2klAqjFmkzGmBPgO6x53pZRSqlKFC/AWiMgd5RNF5HZgUdUVKbKySOKGkuFkmmS+jn2RUx2rqvR4nZ+YzCXvl00rWFLuHrvHxq6gyOlmvz0FwzvTNgAHGu/T8vnszQAhu0GaCu1FKaVUJWgNbPN5vd1OK+8KEVkuImNEpG2wHYnInSKyUEQWZmaGvgdcKaVU3RQuwHsQuFVEZojI6/ZjJjAU+Ef1FC8yMmjM1SVPsd00ZWTMKwxwLKu2YwebS+/MV2cEpO3M3k9BBe+dW52ey9Jt2SHXi7ftL7jXpqwLOriMUkqpCgv2RVv+V7ZfsG6P6A78BnwVbEfGmE+MMX2NMX2bNm1aycVUSil1pAsZ4BljMowxpwHPAGn24xljzKnGmF3VU7zIyaQh15Q8Qappzacxr3GpY3a1HPfu/y5mT77/NAe7gtyLt3ZXHt2engJYUzGs3pkbcp/GwKU+rYQB6w/Qkvfe9NSA6SGUUkodlO2Ab4tcG2CnbwZjzF5jTOkF4FOgTzWVTSmlVC0SfaAMxpjpwPRqKEuNk0US15U8zscxb/JW7Ae0c+7mHfdlBP8htvL0ff63CudNGVY2d9/9Z3c6YP6/Nu4lZ38JfVMaMyd17yGVTyml1EFbABwjIu2BHcA1wHW+GUSkpTGmtE/9xUDlj/allFKq1qvIROeVSkTaish0EVkjIqtEpEZ398ylPjc5hzHGfQb/ihnD6zEfEUdJpIsV1Lu/p4Zdv3x7Ntd+Ope7/7uYVyev86a7PIYR41axckcOF707u1KmYDgYpaN85hY5Sc85+KkjlFKqpjPGuID7gClYgdv3xphVIvKsiFxsZ3vAvi4uAx4AbolMaZVSSh3Jqj3AA1zAQ8aY44BTgHtDDBVdYziJ5mHnXbzuvJIromYxJnYEbSUj0sU6aBe/V9ZNc/TCsnv9R83dypd/pnHhu7NZsSPHO5BLOIu3ZjF6QVm3zf0lbiaFGMwlPWc/70zbEHK6htKRQwe9+Qenvvh70Dylfly0ndkb9hywfPM27eWNqesOmE8ppaqLMWaiMeZYY0xHY8wLdtpTxphx9vJwY0w3Y0wPY8xZxpi1kS2xUkqpI9EBu2hWNrv7Sbq9nCcia7BGEqvYxG4RI7zrvpw15mhej/mQCbGP85Dzbn719I10wQ5b6QTspdwew89Ld9C6YT227isMus3lH/wJwI+LdzB/8z4GHNuUmesz+fnefvRo29Av7z2jFrNkazYDuzWnS4ukgH053Vbgl54TeK9heQ/9YA14k/bSkLD5rv5kLgD/GtjZmzZ6wVY8Bq49qd0Bj6OUUkoppdSRKBIteF4ikgL0AuYFWVfpw0A3Sog57H385unDkJL/Y4tpxqexb/BC9OfU58juVli+YW3csp3847ulXPnRX/zr+7IRRO/7JnBS8vmb9wEwc731N/pgRmpAS11hsTvgOGe/PsO7HOqOxq/+TPPuvzI8+uMKHQ1UKaWUUkrVahEL8EQkEfgReNAYEzAEZFUMAz35wTP4/q5TD3s/200zriwZwSeuIVwb9TuTY4dV+Xx5Vam0i+SBjF+eTsqwCWzPCt6qBzBlVQaTV+5i5Y4cducW4fEY7yidItYgL1v3FrIpzITtmzLzSRk2gafHreJvH/8VNM9//kojv4LTRADVfl9hbTJ2yXbWpIcepVUppZRSStUc1d5FE0BEYrCCu1HGmJ+q67jNk+JpnhRfKfsqJpb/c13PFHdfXo35mG9jX+BHd39ecl5DJo0q5Rg11XWfBjS4+vn7KP+WvqOPSgAgr8jFtZ/ODch/0xfzObZ5ovf1i5OC33bi8ZQ1AT718yqWb8/htat6VKjMZ702I+z6lGETOP/4Fnx4g/+o5HlFTlbtzOWUDkdV6Di10T9HV6xbrFJKKaWUirxIjKIpwOfAGmPMG9V9/Mq2yHTmgpIX+cB1MRc6/mJ63EPcGfULsTgjXbQqE+q+vFC27LXyz1wXuqvt+ox87/Kvq/0HsNmZbXWB7fDYRL/0fQWHPppp6T6hbBTPSSsDp3e8Z9RirvlkLtmF1rHyi108N3510AnpS7k9hmE/LufRMcvD5lNWfbo94edhVEoppZRSFReJLpr9gBuBs0Vkqf24oDoL8OjgLpW6vyLieMV1DYNKXmau5zgei/mWGXH/5IaoX2t1oHew3psefhqHUD6euZF7Ri0KSC+wu2juyS/m2McnMSc1cHTNf45eyrQ1/gHjL8t2ctpLv3P5B3NIGTaBm76Y7133zbytPD++bLyfWfaInSUuDwDvT0/l89mbGTVvK26PYeCbM5myahdOt4csO+AcOWcz3y3YxuiF23j39wOPSFpXlbg8HP/0FEaMq7zuzb+vzWDr3oP7AUIppZRSqjap9gDPGDPbGCPGmO7GmJ72Y+KBt6w8jiqapzzNtGSo8xGuLxnOTtOE52NGMj3uX9waNYkG6D+dh2rcsp1MXBHYujZv8z5Of/l3+j7/GyVuD9d/No/TX/7dG2gBjF2yg9u/Wui33f3fLgFg8dZsoCyIA3hs7Ao+m70ZwO++s4d+WMZPi7fz4YyNALjcHlbsyGF9Rj6P/ricf45eSq/nfgXg+QllcxMXFAe24Hk8hjXpuTz980q++jON2Rv28MXszSGnkQhmf4mblGETGLtkOwBFTrdfF1ZfuUVOVu7IqfC+q8ItI+fz89IdfmlOtxU0j1m0/bD2vW1fIcu2WX/L275cyFk+A/gopZRSStU1EbkHL9JuPPVo0vYW8u38rQfOfAjmeE5gTsnx9Hes4IHon3g65msejv6eH91nMMp9DuuMDtN/MLIKQ7eCbs/aH/C6NNA6XGk+A7PM2rDHLxD0vU9QsAagCUYEpq/bTccmibSz70X8cOZGXp0SfI6+205vH5BWUOyi29NTGHFRV27pZ63flWtNKfHWbxsY3K0lxz01mbsHdGTY+f6t01v2FjDg1RnW+VTwHrq9+cVk5BbTtVWSt5W0vJRhE7ijf3seH1KxKSxnrMtkxrpMYqIcXHBCS8CqG8A7CM+h6v/KdADWPT8YoEJdPof/tJwBxzZl8PEtD+vYSimllFI1TUSnSYiUhNhoXrz8hCo+ijDL052rSkZwYfHzTPKczNVR05kSN4xJsY/y96hxtJHKmf5BVb6UYRMCBosJxTcALd8KN3JOGreOXMCA16wgpKDYxf+W+LdklXp2vP9UkBsy8vhh4Tb25lstkp/P2exd59sIXdoy9s28LQH7LA3ugpUtlMFvz+KCd2Zx68j5PPG/lSHzfTprs9/rlGETeHTM8rD7vmfUYopdVqumw47wKnoL3vqMPFJ354VcvyY99Lryvp2/jbv/a/19t+wt4MxXp7M7r4hFW7KCdvUtVeR0k1tUOd2uS1weHhu7gt15B57/USmllFKqoupkgFfdVpoOPOy8m1OL3+NJ5y0UEs+jMd8xO+4fTI19hMeiR3GaYyVxHPqgIapmuPXLBUHTjbHuBzzvjZls2J0fNA/A2l25pAybwIhxqzjvzT94ZMxyVu20uld6PIH5t+wtZJg9t19uUVlr26wNmaQMm+CX94cgXSGXbM3i32OW4fEY9uQXA5CZZz1PX5cZEOw43R7eL3cv5bhlO7n8gzkAjF64jfSc/Tw+doW3C2Z5n/6xCYDJ9qA2voHniu05vDvNum/x5clrGfjmTO+6gW/+wblv/OG3L9+Bdg6mi2upOal7GDknjbS9hZz0wjSu+PBPrv8s9CixXZ6cTPcRUwPSC4pdvDF1nfecF6btY9aG8D/gTFuTwTfztlbqPYjVzeX2hOwarJRSSqnIqJNdNCNlH0l87R7I1+6BtJUMBjkWcqZjKbdETebO6Ak4TRSrzdEs8XRiqacTG0xrNppWFBEX6aKrCpoRZqTQsSFa7nwNfmsWAF/+meZNK21J3JG9H4/HsGx7NreFCCRL3fj5/IC0f49ZzhW92xDlENbtyqN9k/rc+Pl88otdNE+K593fU/njkbP8ttltB3ulvp2/1a976etT1/Hu7/4B3/CfVjBjXSbndm3OWZ2bBZSjsMTN+ow8Hhy9FLCC31IXvTcbgPvPOcZ7vyOUBYO+Copd9Pbpjls+zNidV8SouVt58NxjEAl+422oYO6vjXs5taP/1Bh5YVru3vx1PZ/N3kzrRvW4+sR2XPmRNX9juG6x7kMISA9WyrAJ3HJaCiMu7lYl++/0+CQuOKEFH1zf58CZlVJKKVUtNMCLkG2mOZ+5h/CZewgJFHGKYzV9HOvpLalcFTWTW6LLWgm2eZqy2bQgwzRiF43ZZRqz2zQkjwTyTD3ySCDf1COeEs6JWkwvRypHkYtByKcemSbZetCQTJPMLnMUu0wjsknEv7OfqunKTxVR3vjlO3nml9Uh13d8bCKPDu7Cy5PX0iIp3jtZfGmQtiN7f8hty7cI+m7nqzTIvXXkAgZ1a87HN/b1W//BjI2s21XWndLlMSzblk3L5LI5Kq/2meB+854C7v5v2SiqWQUlDPtpOfExUX773Zfv3wL+0PfLmLVhD/06NaGwxEVekYvjWjZgRQUGnPlw5kZ6H92QuGjrGH+m7uGX5TtD5t9vT4dR4vJ4W1zL25CRh8Mh/LxkB4OPb+kNbMXnM7g3v5hnfllNi+R4Hh3chahyI0INfHMmV5/YjkYJMTRKiKVt4wTaNKrnVxdzUvewaEsWD5xzDGD9WHAoAV7KsAncfnp7nrww/H2WwQZAUkoppVTkaIBXAxQSz++e3vzu6Q1AFG46SDqdZIf1cOzkaMngGMcOmpFFlIT/5T/dNCbDWJOtt2U3ZzhySJLAUTyLTIw3aNxpjiLdHMUek0yGacRu05AMrGdtQTxy3PfNkgPmeXmyNUBM6UAtvsqPdHm4pqzK4IMZgUHgtLW7/V5f8v4cHjz3GO/reZv3eZfLTzXx2tR1TFnlP/UFwND/+I+WWlhiBV1P/byStbsqfn8ewB/rM+n8xGQmPHA6O7L2c+fX/tN07Mkvpkmi9bnwDXyf/Nm/u+X6jDzWpOdySc/WnPdmWffSd35P5fWregAwYUU6z+YX8970VEbOSfPm6depCQOObeq3r/UZ+TxX7l7NC7u35L3rrO+O7MISb6tkaYBXavLKXczfvI+nLqrYwDgAn8/eTK92DXGIeAfHUUoppVTNpgEesPSp85i6OoMebRrStnE9uj41JaLlcRPFBtOGDaZNaYJXFG6akEMTsYK2BhSSyH4ayH48CIs9x7LKpATsM44SmkoOzciiheyjhVjPze3nvrKeFo59xEjgsP65JoEM04gM05DdNGJ36bL9nGGnFRNbRTWiqst3C7ZV+j5fmRx8xNDy3vot+JyBPy32DzpHzTvw6LcnvfCbt3vpwQZ3voa8Mzto+tjFO3hn2ga6t00Ou/1AO6j7x3dLA9bl+4xQ+sLENQHnWf6ewoFv+t9/WGr88nTeu85a9u26u7/E/7Nc2gq6aU8+n97Ul637CtlXUMKybdkM7d/BL+9XPl2ES380CNfddPn2bGZt2MPdAzoGtDoqpZRSqnppgAc0TIjlb33bRroYFeImigwak2EaB950FEYxsWw3TdlO0zDbGRqSTzPJprlk0Qz7WbK8aSfJWpqSTZwEDp+fYweCpa1/mT6BYDb1yTaJ5JBItqlPPvXQ7qGqqpS/d7CyvTDRmutwTureQ96Hb/fWYCOr3jJygXcqitenhg+SOz8xiY5NE1ntM3fjZfbANwDzNpWVc8a6TP5Yn+k3P2RpgDd++U6iHcLTIQZ+ycwrptjlpk2jBL/0i9+zjvXRzI2sGDEobFmVUkopVbU0wAsjJkpwuuvSCHFCNg3INg1Yb8IFvFYg2Nwn8GtG2XJzyeJkWUszsogN0iII4DIOcrxBn/WcTSI5pn7Za5NINvXJNfXJJtGb101U0H0qdSQpHbUUQk8V8emszQHTUQRT7PL4BXfg33J59Sdz/daVb1UNdn9lMCe+8BsAUQ4JOt9gXlHweROVUkopVX00wAvjwXOP5dUp67i1Xwoj56RxXtfmtG5Yjy//TMMhFZ+/q/YpCwTDT9puaEQeTSWHZApoKPk0lHzvsu9zE8mhEztIdhSQHOR+QV+5ph453sCwvh0YWsFgju/rcmlFxKKthkrBuoyD77bqGwRWZDJ5pZRSSkVGnQ7wvrz1RO9ADMHce1Yn/ta3LQvS9jFyThrRDuGJIcfxt75t6doqicISl/d+vYFdmzN1dQZPX9Q17CiGdYuQRRJZJsl6WcH/CR14SCoNCCkgWQpIJt/7uqHk+6QV0IVtJDusdcHuISxVbGLIoT65JoFcEsgx9cmlPnmmnv2cQB71yDP1KCSeAuIpMNZzof1cQDyuuv2xUUoppZRSNVid/k/1zCBzdJXXtEGc3zxd0VEOurayApaE2Gh6tm3I0m3ZREdpy1Bl8eDwthACB3GvoSGBYhrawaAVBJYGilYwmEQ+SbKfJApoLHmksIsGDut1qO6k5RWbaG8AWGjirGUTb6fFeYNBK91aX0wMxSbGeiaGYmLt17EUlaaZsnUlRGNwHFL9KaWUUkqpuqtOB3ihnNm5KUOCDAkebK7kT2/qy5zUPSzZmmXlAa49qR1J8dF8/Mcmv7z/u7cfk1akB6QD3HlGBz4Jkq4OhlBoB1Y7TZODGoQGDHE4SaKQRNlPAsXUZz8JUkQiRdSTYhKx06WIBIqoTxEJUkR9iqkv+2lEvpXusNfL4Q30UWyiAwLCEnyCRDtALC4XIBaVphmfdeUCyNABZ9lDu7MqpZRSSh15NMAL4stbT/J7bcJECk0bxHFpr9beAA/gxctPACA6Sji+VTJPjVtFZl4xrZLjGX7BcRzfOpn7vy2br+yv4WfTMrmeBngRJRQTSyaxZJqGZcmHcauR4CGBYupRQhwlxImTOHweUlLutdPK5/faSosPsq6B7KcJuSH27TzsGikf8DlNFC6icRGFkyjcROHCgdNE4ySaEu8jxpvmJAqnvU0JMd6g1VpnpbuNAxfRFNvblhCN05QtF/ukOX3zEI1HWzmVUkoppfxogFcBpV00JUyLRqfmVnfCto3Lhg9/ZFAXAJ4qN+T4RT1a+QV4LZPrBd3na1f14OEflln7b5ZI6u58/j24M/ec2YkuT06iyOnhtn7t+WKONcpefIyDIqfnIM9OVRWDgwLqUUC90oTyGaqM4CEWV4WCxzicxAcNNn22ESfRuInGRQxuonEThcdKEzcJFBCDyz5mCdEOtzdfDC5icAWdWuNwuY14g0unXSonUbhMFC7KHmUBaRROE1WWzw4yXVhBpss4cBPlDVY9OKwgFAcuUxbU+j+Xrbfyh17v9u6v9JhlaR7jwI0DD9Zz6bIHCUjX1lWllFJKhaIBXgX069SERgkx3D2gY8g8N5zcjuNbJdGrXaOAddec2JZ3f08lMT6wuuc/do53+Ykhx/H8hDU8Mqgzr05ZxzHNEr3rbj4thSf/t5KWyfGAdf9fkbOEe8/qSFK9aJxuD48M6lLh4c5V7WZw2F0xY0sTymeIAOMNOkuDvmjcRInHJzi014mLWJzEYT3H2mml+Uq3t/K57XQrjIuxg85oAh8x4iZenERTFLheSoNWF3H2/h12EBslNWvUSI/xD/o8SEBg6F32yevxW19+Hw7cdpBp7HXBAk7vvkxg4LnXJAGhJ0RXSimlVNXTAK8CGtePZclTA8PmEZGgwR3Av847lgfOOYaYqMDuZM2S4r3LQ/t3YGj/DhhjuKJ3G1okl627/qR2tGlUjzOPbQqAw/4B32Os6RxKfXvHKfy1cQ/v+EyiXN4TQ45jaP8O/Jm6h+s+mxf2vJSqPGJ3r4zxTw4WO9WseAow3hbLsmc30XisZ7Gf7fVRpem+z+Iut325/UjZtg7vs/FZtvcrPsv2+vJ5HPY6/7xWGBhln0uw/UaLm1irPdFvu2B5xd6Pb/oO0yTSf6gaTUQGA28DUcBnxpiXQuS7EvgBONEYszBYHqWUUioUDfCqgYgQU26UzUVPnEuRK3h3ShHxBnendGjM3E37cDiEs3xG/ezVrhG/rs4gNto/aDy141Gc2vEo5mzcy6ItWYy5+1Su/Ogv7/qT2zdmaP8OAJzWqQkf3dCHu/+7yDvNA8DIW0/k1pELABjUrTlTVmUc8BxbJMXz4uUnMGreVn5bc+D8pVY/O8g71YRSNZfYoVhU8NUVCUhrXNBaNdIiXYAaSkSigPeB84DtwAIRGWeMWV0uXwPgAUB/fVNKKXVIdISCCDkqMY7WDYPfe+dr1NBT2PDC+QHpb1/Tk3H39SO5XkyQrUIr/z9mN3vKh4t7tuLC7i157tLj/QLJy3q1YfOLF3BOl/BTSsx97BzO6tKMz27uy2//GsAJrZMBmDPsbL987XzuUVw+YiAJscF/Y1j1zKCKnpJSSh0JTgJSjTGbjDElwHfAJUHyPQe8AhRVZ+GUUkrVHhEJ8ERksIisE5FUERkWiTIcKaIcErRrZ0JsNN3bNAyyhaVvitVdtEliHFMePIO/9W0DQGy5fbVtnEDaS0O4sHsr3ruuNzeecjQAXVpYg8ac3L4xIsLwC47zbtOv01HeKSN+vrcfMx4+02+fnZol8sv9p5P20hBaN6xHj7YNObZ5ItMeGkCPtmVlToq3gtOUoxIor36cf+D3wNmdvMuX927tXS5/bF93ntGBBY+fyzvX9mLJk+eFzKeUUtWgNbDN5/V2O81LRHoBbY0x48PtSETuFJGFIrIwMzOz8kuqlFLqiFbtXTQr2k1FHZ5HBnbmqj5tSWlSH4CXLu/OUYlx3HJaSoW2/+q2k5i1YQ+N6luDdHRqlsjc4daAMA0TYthXUMKCtH1+AVsoP9/bz7v8zMXdqB8bxUMDO3vTbu/fgSf/t5LrT27H0P4dWLQly2/7l684gatPbOe9r/Cly7tzUY9WdGuVRLMG8YTymB2UXtyjVdD1xzZP5PqTj+bpcqOcDj+/C5NW7mLptmy/9It6tGLyynQ8Btwe/7bQW/ul8O9BXTjuqckhywNw14AOfDzTmg6jaYM4MvOsufI6NKnPpj0FAIy95zRmrs/krd82BGx/Re82/Lh4e9hjhPLZTX0Z+p/A23lOTGnEgrSsIFtUv2ObJ7I+Iz/SxVCqKgQb+tT7RSIirbX+jwAADZxJREFUDuBN4JYD7cgY8wnwCUDfvn3rSOdfpZRSFRWJe/C83VQARKS0m4oGeJUoOspBJ59ROB0O4dHBXSq8ffOkeK7s08YvzXfQl1YN63FJz9blNzugxvVjeemK7n5p7Y+ygtDjWibRvkl92ttBKViTzl99YjsA7h7QERGIjXb4dSMtdfvp7dmUmc/0dcF/0X7gnGP4dv5WLu/dmo9nbmLqPwdgjCEzr5hLe7Xm6o//Ym9BCXcN6MhdPiOm3jNqERNX7OKJIcfx7rW9KCh2MWLcKmauz2R3XjFpL5WNGrju+cFMXJHO+GXpTFu7m49v7MPqnbm8PW0DzZPiGH7+cdzZvwMPjl7KG3/rybfzt3JH/w6UuD30eGYqYN1f2atdI2KiHLw6ZR2A3zGu7NOGaz+dS1J8NLlFLm/drUnPDTjn849vQecWDfhs1mbO7dqcy3q1ZuySHX55vr/rVIpdHro8GT44PVxpLw0JO8pr2ktDKHK6D1iOL289keE/rSA9p/J7sPXrdBRzUvdW+n6VwvpBs63P6zbATp/XDYDjgRlidZFoAYwTkYt1oBWllFIHQ4yp3h//7NHBBhtjhtqvbwRONsbcVy7fncCdAO3ateuzZcuWai2nql6rd+ZyXMsGiJT9yJ1b5KReTFTQLqq+Vu7IoXH9WFrZ9zSuSc+lXkyUt/WyonIKnRSUuLz7KVVY4mLVzlxOTGnsl15Q7KKgxBW0FXF3XhHfztvG/Wd3wuEQ8otdRDuE+JgQg3QAk1akc2L7xjRJjPOmfT13Cxd3b0VyQtm9lsYYvpiTxpV92jBl1S56tGlI5xYNKHa5mbxyF20a1WPJ1mx6tm1I33JlBtieVcjoBdsYNW8rH93Qh5PaW3n+WJ/JCa2TWZ2eS/OkOL6Yk8aunCKaJ8Uxbc1u3r22Fx2aJpKzv4RtWfsZcExTrv7kLzLzihlwbFPO6tKMu75eRNvGCZzQOpkmibEM6taCBWlZ3NG/PdFRDlbuyOGGz+dxtn1P5/GtkkmqF8N5XZt77ye97IM5XNWnLY+NXQHAK1d2p15MFGd2bsqqnbmc0uEoAFJ357Eps4D3Z2zkkh6tuO309vznrzTmbtpLYlw0C9KyuHtAB8YvT+eszs14drz1G1KPNsms3JnLL/edTmZ+MTd/MZ/rTm7H4G4tOOPYpgz9agG/rdnN8a2TWLkjl64tk+jULJFxy3byw92n8tWfafyxPpPx9/cnIS6K16eu58o+rXnml9Uk14th1oY9fvX9j3OO4e1pG5g7/Bzmbd7LP75bGvA3aZUcz5MXduXvoxb7pb99TU+G/7SCISe0JDbawbOXHE/HxyZ611/euzW5+538tmY3SfHR7He6cbqNd8oVX32PbkTvoxuRu9/J6IXbmPnwWSzdns3a9Fw+mLERsLpmr92VB8ALlx3P42NXerdvlBDDiIu7ect/Re82TFqZTmGJ25vnmztO5rSOhzeSpogsMsb0Payd1EAiEg2sB84BdgALgOuMMatC5J8BPHyg4K5v375m4cLDi/++X7iN9k3qB3zHKaWUqlkqeo2MRIB3FTCoXIB3kjHm/lDbVMYFTCmlVM1XWwM8ABG5AHgLa5qEL4wxL4jIs8BCY8y4cnlnUE0BnlJKqSNDRa+RkeiieaBuKkoppVStY4yZCEwsl/ZUiLxnVkeZlFJK1T6RGEVzAXCMiLQXkVjgGmDcAbZRSimllFJKKXUA1d6CZ4xxich9wBTKuqkEvQdBKaWUUkoppVTFRaKLZtBuKkoppZRSSimlDk9EJjpXSimllFJKKVX5NMBTSimllFJKqVpCAzyllFJKKaWUqiU0wFNKKaWUUkqpWqLaJzo/FCKSCWw5zN00AfZUQnFqE62T4LReAmmdBNI6CVQZdXK0MaZpZRSmLqik6yPo+zkYrZNAWieBtE4CaZ0EV23XyCMiwKsMIrKwIjO/1yVaJ8FpvQTSOgmkdRJI6+TIpX+7QFongbROAmmdBNI6Ca4660W7aCqllFJKKaVULaEBnlJKKaWUUkrVEnUpwPsk0gWogbROgtN6CaR1EkjrJJDWyZFL/3aBtE4CaZ0E0joJpHUSXLXVS525B08ppZRSSimlaru61IKnlFJKKaWUUrWaBnhKKaWUUkopVUvUiQBPRAaLyDoRSRWRYZEuT1UTkTQRWSEiS0VkoZ3WWER+FZEN9nMjO11E5B27bpaLSG+f/dxs598gIjdH6nwOhYh8ISK7RWSlT1ql1YGI9LHrONXeVqr3DA9eiDoZISI77PfKUhG5wGfdcPv81onIIJ/0oJ8nEWkvIvPsuhotIrHVd3aHRkTaish0EVkjIqtE5B92ep19r4Spkzr9Xqmt9PpY966PoNfIYPQaGUivkYGOmGukMaZWP4AoYCPQAYgFlgFdI12uKj7nNKBJubRXgGH28jDgZXv5AmASIMApwDw7vTGwyX5uZC83ivS5HUQdnAH0BlZWRR0A84FT7W0mAedH+pwPsU5GAA8HydvV/qzEAe3tz1BUuM8T8D1wjb38EfD3SJ9zBeqkJdDbXm4ArLfPvc6+V8LUSZ1+r9TGR7i/UW19oNfH0nPWa2TF6qROf++FuR7U2fdKmDqpUe+VutCCdxKQaozZZIwpAb4DLolwmSLhEuAre/kr4FKf9P8Yy1ygoYi0BAYBvxpj9hljsoBfgcHVXehDZYz5A9hXLrlS6sBel2SM+ctYn77/+OyrxgpRJ6FcAnxnjCk2xmwGUrE+S0E/T/YvbmcDY+ztfeu3xjLGpBtjFtvLecAaoDV1+L0Spk5CqRPvlVpKr4+WOnV9BL1GBqPXyEB6jQx0pFwj60KA1xrY5vN6O+H/ELWBAaaKyCIRudNOa26MSQfrzQk0s9ND1U9trLfKqoPW9nL59CPVfXZXii9Ku1lw8HVyFJBtjHGVSz9iiEgK0AuYh75XgIA6AX2v1Da18Xv+QPT6GJp+7wWn33voNTKYmnyNrAsBXrC+vLV9boh+xpjewPnAvSJyRpi8oeqnLtXbwdZBbaqbD4GOQE8gHXjdTq9TdSIiicCPwIPGmNxwWYOk1cp6CVIn+l6pferi30KvjwevLn/G9XsPvUYGU9OvkXUhwNsOtPV53QbYGaGyVAtjzE77eTcwFqsZOMNuCsd+3m1nD1U/tbHeKqsOttvL5dOPOMaYDGOM2xjjAT7Feq/AwdfJHqyuGNHl0ms8EYnB+pIeZYz5yU6u0++VYHWi75VaqTZ+z4el18ew6vT3XjD6vafXyGCOhGtkXQjwFgDH2CPSxALXAOMiXKYqIyL1RaRB6TIwEFiJdc6loxbdDPxsL48DbrJHPjoFyLGb26cAA0Wkkd3MPNBOO5JVSh3Y6/JE5BS7r/RNPvs6opR+Qdsuw3qvgFUn14hInIi0B47BuhE66OfJ7js/HbjS3t63fmss++/3ObDGGPOGz6o6+14JVSd1/b1SS+n1Ua+Pvurs914odf17T6+RgY6Ya6SpASPSVPUDa1Sf9Vij1Twe6fJU8bl2wBqJZxmwqvR8sfr0TgM22M+N7XQB3rfrZgXQ12dft2HdDJoK3BrpczvIevgWq4ncifUrye2VWQdAX/vDuxF4D5BIn/Mh1snX9jkvt7+EWvrkf9w+v3X4jGoV6vNkv/fm23X1AxAX6XOuQJ2cjtX1YTmw1H5cUJffK2HqpE6/V2rrI9TfqDY+0Oujb13oNbJidVKnv/fCXA/q7HslTJ3UqPeK2DtSSimllFJKKXWEqwtdNJVSSimllFKqTtAATymllFJKKaVqCQ3wlFJKKaWUUqqW0ABPKaWUUkoppWoJDfCUUkoppZRSqpbQAE+pI4yInCki4yNdDqWUUqom0eujUhYN8JRSSimllFKqltAAT6kqIiI3iMh8EVkqIh+LSJSI5IvI6yKyWESmiUhTO29PEZkrIstFZKyINLLTO4nIbyKyzN6mo737RBEZIyJrRWSUiIid/yURWW3v57UInbpSSikVkl4flapaGuApVQVE5DjgaqCfMaYn4AauB+oDi40xvYGZwNP2Jv8BHjXGdAdW+KSPAt43xvQATgPS7fRewINAV6AD0E9EGgOXAd3s/TxftWeplFJKHRy9PipV9TTAU6pqnAP0ARaIyFL7dQfAA4y28/wXOF1EkoGGxpiZdvpXwBki0gBobYwZC2CMKTLGFNp55htjthtjPMBSIAXIBYqAz0TkcqA0r1JKKVVT6PVRqSqmAZ5SVUOAr4wxPe1HZ2PMiCD5zAH2EUqxz7IbiDbGuICTgB+BS4HJB1lmpZRSqqrp9VGpKqYBnlJVYxpwpYg0AxCRxiJyNNZn7ko7z3XAbGNMDpAlIv3t9BuBmcaYXGC7iFxq7yNORBJCHVBEEoFkY8xErO4pPavixJRSSqnDoNdHpapYdKQLoFRtZIxZLSJPAFNFxAE4gXuBAqCbiCwCcrDuQwC4GfjIvkBtAm61028EPhaRZ+19XBXmsA2An0UkHuvXzX9W8mkppZRSh0Wvj0pVPTEmXAu4UqoyiUi+MSYx0uVQSimlahK9PipVebSLplJKKaWUUkrVEtqCp5RSSimllFK1hLbgKaWUUkoppVQtoQGeUkoppZRSStUSGuAppZRSSimlVC2hAZ5SSimllFJK1RIa4CmllFJKKaVULfH/LDp/q2p4UvQAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 1080x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize = (15, 4))  # adjust figures size\n",
    "plt.subplots_adjust(wspace=0.2)  # adjust distance\n",
    "\n",
    "from scipy.signal import savgol_filter\n",
    "\n",
    "# loss plot\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.plot(loss_history)\n",
    "plt.plot(savgol_filter(loss_history, len(loss_history)//3, 3))\n",
    "plt.title('Loss')\n",
    "plt.xlabel('epochs')\n",
    "plt.ylabel('Categorical Cross-Entropy')\n",
    "\n",
    "# accuracy plot\n",
    "plt.subplot(1, 2, 2)\n",
    "plt.plot(accuracy_history)\n",
    "plt.plot(savgol_filter(accuracy_history, len(loss_history)//3, 3))\n",
    "plt.title('Accuracy')\n",
    "plt.xlabel('epochs')\n",
    "plt.ylabel('Accuracy')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 3. Test\n",
    "\n",
    "Let's now test the model on the Test set. I will repeat the dataprep part on it:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Dataprep Test data\n",
    "\n",
    "test = pd.read_csv('fashion-mnist_test.csv')\n",
    "\n",
    "test_label = pd.get_dummies(test.label)\n",
    "test_label = test_label.values\n",
    "test_label = test_label.astype(np.float32)\n",
    "\n",
    "test.drop('label', axis = 1, inplace = True)\n",
    "test = test.values\n",
    "test = test.astype(np.float32)\n",
    "\n",
    "test = test / 255.\n",
    "\n",
    "test = test.reshape((len(test), 28, 28, 1))\n",
    "\n",
    "prediction = CNN.predict(test)\n",
    "\n",
    "prediction = np.argmax(prediction, axis=1)\n",
    "test_label = np.argmax(test_label, axis=1)  # reverse one-hot encoding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[855   2  13  13   4   1  92   0   4   0]\n",
      " [  0 980   2   2   0   2   4   0   1   0]\n",
      " [ 19   1 848   6  23   0  41   0   2   0]\n",
      " [ 18  13   9 926   9   0  23   0   1   0]\n",
      " [  2   1  74  35 936   0  74   1   3   0]\n",
      " [  0   0   0   0   0 963   1   9   0   6]\n",
      " [ 92   2  50  16  26   0 757   0   4   0]\n",
      " [  1   0   0   0   0  17   0 966   2  32]\n",
      " [ 13   1   4   1   2   3   8   0 983   5]\n",
      " [  0   0   0   1   0  14   0  24   0 957]]\n",
      "\n",
      "Test Accuracy: 0.9171\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "\n",
    "CM = confusion_matrix(prediction, test_label)\n",
    "print(CM)\n",
    "print('\\nTest Accuracy: ' + str(np.sum(np.diag(CM)) / np.sum(CM)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "seaborn.heatmap(CM, annot=True)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "My Convolutional Neural Network classified 91.7% of Test data correctly. The confusion matrix showed that category no. 6 (Shirt) has been misclassified the most. The next goal is to correct it; one possible solution would be to increase regularization, another to build an ensemble of models."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Thanks for coming thus far. In the next Notebooks I will implement more advanced Convolutional models, among other things."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}