classification_prob2_outro.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-11-13 10:59:01.197892: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory\n",
      "2021-11-13 10:59:01.197918: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.linear_model import Perceptron\n",
    "from sklearn.neural_network import MLPClassifier\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import (\n",
    "    confusion_matrix,\n",
    "    balanced_accuracy_score,\n",
    "    ConfusionMatrixDisplay,\n",
    ")\n",
    "\n",
    "import tensorflow as tf\n",
    "from tensorflow.keras import layers, models, Input"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Load Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train = np.load(\"./data/Xtrain_Classification_Part2.npy\")\n",
    "y_train = np.load(\"./data/Ytrain_Classification_Part2.npy\")\n",
    "# removed x_test,  loaded only on the final Testing steps"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Multi Layer Perpectron"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_4\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "rescaling_4 (Rescaling)      (None, 2500)              0         \n",
      "_________________________________________________________________\n",
      "dense_12 (Dense)             (None, 1000)              2501000   \n",
      "_________________________________________________________________\n",
      "dense_13 (Dense)             (None, 1000)              1001000   \n",
      "_________________________________________________________________\n",
      "dense_14 (Dense)             (None, 4)                 4004      \n",
      "=================================================================\n",
      "Total params: 3,506,004\n",
      "Trainable params: 3,506,004\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Epoch 1/20\n",
      "295/295 [==============================] - 5s 15ms/step - loss: 1.0423 - accuracy: 0.6086 - val_loss: 0.8721 - val_accuracy: 0.6682\n",
      "Epoch 2/20\n",
      "295/295 [==============================] - 5s 15ms/step - loss: 0.8068 - accuracy: 0.6967 - val_loss: 0.7535 - val_accuracy: 0.7191\n",
      "Epoch 3/20\n",
      "295/295 [==============================] - 5s 16ms/step - loss: 0.7709 - accuracy: 0.7008 - val_loss: 0.7027 - val_accuracy: 0.7476\n",
      "Epoch 4/20\n",
      "295/295 [==============================] - 4s 15ms/step - loss: 0.6985 - accuracy: 0.7340 - val_loss: 0.6554 - val_accuracy: 0.7564\n",
      "Epoch 5/20\n",
      "295/295 [==============================] - 4s 15ms/step - loss: 0.6680 - accuracy: 0.7442 - val_loss: 0.6227 - val_accuracy: 0.7693\n",
      "Epoch 6/20\n",
      "295/295 [==============================] - 4s 15ms/step - loss: 0.6598 - accuracy: 0.7507 - val_loss: 0.6084 - val_accuracy: 0.7802\n",
      "Epoch 7/20\n",
      "295/295 [==============================] - 4s 15ms/step - loss: 0.6184 - accuracy: 0.7704 - val_loss: 0.6086 - val_accuracy: 0.7761\n",
      "Epoch 8/20\n",
      "295/295 [==============================] - 4s 15ms/step - loss: 0.5907 - accuracy: 0.7836 - val_loss: 0.6821 - val_accuracy: 0.7408\n",
      "Epoch 9/20\n",
      "295/295 [==============================] - 4s 15ms/step - loss: 0.5926 - accuracy: 0.7792 - val_loss: 0.6188 - val_accuracy: 0.7700\n",
      "Epoch 10/20\n",
      "295/295 [==============================] - 4s 15ms/step - loss: 0.5731 - accuracy: 0.7863 - val_loss: 0.5991 - val_accuracy: 0.7917\n",
      "Epoch 11/20\n",
      "295/295 [==============================] - 5s 16ms/step - loss: 0.5690 - accuracy: 0.7923 - val_loss: 0.5582 - val_accuracy: 0.8094\n",
      "Epoch 12/20\n",
      "295/295 [==============================] - 4s 14ms/step - loss: 0.5578 - accuracy: 0.7946 - val_loss: 0.6430 - val_accuracy: 0.7720\n",
      "Epoch 13/20\n",
      "295/295 [==============================] - 4s 15ms/step - loss: 0.5435 - accuracy: 0.7979 - val_loss: 0.5680 - val_accuracy: 0.7951\n",
      "Epoch 14/20\n",
      "295/295 [==============================] - 4s 14ms/step - loss: 0.5429 - accuracy: 0.7953 - val_loss: 0.5452 - val_accuracy: 0.8033\n",
      "Epoch 15/20\n",
      "295/295 [==============================] - 4s 15ms/step - loss: 0.5320 - accuracy: 0.8069 - val_loss: 0.5914 - val_accuracy: 0.7897\n",
      "Epoch 16/20\n",
      "295/295 [==============================] - 5s 16ms/step - loss: 0.5220 - accuracy: 0.8065 - val_loss: 0.5365 - val_accuracy: 0.8073\n",
      "Epoch 17/20\n",
      "295/295 [==============================] - 4s 14ms/step - loss: 0.5365 - accuracy: 0.8023 - val_loss: 0.7230 - val_accuracy: 0.7374\n",
      "Epoch 18/20\n",
      "295/295 [==============================] - 4s 14ms/step - loss: 0.5223 - accuracy: 0.8053 - val_loss: 0.5350 - val_accuracy: 0.8094\n",
      "Epoch 19/20\n",
      "295/295 [==============================] - 5s 16ms/step - loss: 0.5084 - accuracy: 0.8121 - val_loss: 0.5695 - val_accuracy: 0.8066\n",
      "Epoch 20/20\n",
      "295/295 [==============================] - 5s 18ms/step - loss: 0.5172 - accuracy: 0.8145 - val_loss: 0.5725 - val_accuracy: 0.7890\n",
      "BACC(MLP): 0.6815777629857059\n",
      "Y_hat shape: (1474,)\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWUAAAEGCAYAAAC95YRPAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/MnkTPAAAACXBIWXMAAAsTAAALEwEAmpwYAAAz/ElEQVR4nO3deZgU1dXH8e9vVvZ12EFAQY2IiKCCGgMR4xIjajRqTIJRgyYuUUMSNb4JSYxbjAuSELdENEbjRsANcSOKigiyK7usgjAswzrMTM95/6gaaIdZephuunvmfJ6nnq6+XX3rVDOcvn3r1i2ZGc4551JDRrIDcM45t5cnZeecSyGelJ1zLoV4UnbOuRTiSdk551JIVrIDSHd5rTKtW5fsZIcRd4vmNEp2CAmjBrnJDiExdhclO4KE2GU7KLJC1aaO0wY3to2bIjFtO2PO7tfN7PTa7K82PCnXUrcu2Ux7vUuyw4i70zr3S3YICZPZo0eyQ0gIW7oi2SEkxNTCV2tdR/6mCB+93jmmbbM7LM2r9Q5rwZOyc64eMCJWmuwgYuJJ2TlX5xlQSnpcKOdJ2TlXL5TiLWXnnEsJhlHs3RfOOZcaDIh494VzzqUO71N2zrkUYUAkTWbE9KTsnKsX0qNH2ZOyc64eMMz7lJ1zLlWYQXF65GRPys65+kBEqNX0GQeMJ2XnXJ1nQKm3lJ1zLnV4S9k551JEcPGIJ2XnnEsJBhRbetzTIz2idM65WjBEhIyYlupIOkzSrKhlq6TrJbWS9IakxeFjy3B7SRolaYmkOZKOqap+T8rOuXqh1BTTUh0zW2hmR5vZ0UA/YCcwDrgJeMvMegJvhc8BzgB6hstwYExV9XtSds7VeWV9yrEsNXQKsNTMVgBDgbFh+VjgnHB9KPCEBaYCLSR1qKxC71N2ztUDIhJ7n3KepOlRzx82s4cr2fYi4OlwvZ2ZrQ3X1wHtwvVOwKqo96wOy9ZSAU/Kzrk6L7jzSMxJOd/M+le3kaQc4Gzg5n32Z2aS9mtktCdl51ydZyaKLDPe1Z4BfGJmX4bPv5TUwczWht0T68PyNUD03ZU7h2UV8qScIlYtyeX2q7rteb5uZQ4//OU6zvvJBsY/lseEx/PIyDSOP2UrV/zfWoqLxAO/6sziOY1QBvz0D2voc8L25B3Afuh8cCG3jPl8z/P2B+3myXs6Mu6xtkmMav8NPXcRp52xDAkmvnow48cdysGHbOaan88gO6eU0oj466hjWLSwdbJDrZG8DrsZcc9SWuYVYyZee6Yt4x9vT5PmJdz84GLadd7Nl6tzueOanmzfmroppTT+45QvZm/XBcAEYBhwZ/g4Pqr8GknPAMcDBVHdHPtI6CcoqT1wP3AssAX4ErjezBYlcr+VxPIocK+ZfXqg9x2LLj12M+bNhQBEInDJMb048YwtzHq/CR+83pwxby4kJ9fYkh/8k732VPAf+6G3F7IlP4vfXHIwD762iIw0OnW7elkDfnba1wDIyDCemj6X9yc2T3JU+6drtwJOO2MZN1w7hOLiDP54x7tM+6gDl/1kDv9+shfTP+5A/+PWctlP5nDTiMHJDrdGIiXikdu7snR+Yxo2jjBqwjxmTmnGkO/mM+uD5jz3945ccNUXfO+nX/CPuw5KdrgVCk70xe8/h6TGwKnAlVHFdwLPSrocWAF8Lyx/FTgTWEIwUuPHVdWdsP/CkkQwTGSymR1iZv0I+l7aVf3OxDCzK1I1IZc3672mdOi6m3adi3n5idZceM2X5OQG3VMt8koAWLkol6NP2r6nrEnzCItmN0pazLV19EnbWLsil/VrcpMdyn7pctBWFi5oze7dWZSWZjBvThtOPGkNZtCoUTEAjRsXs2ljwyRHWnObN+SwdH5jAHbtyGTVkga0bl/MwFM38+YLeQC8+UIeA0/dnMwwqxGc6ItliYWZ7TCz1mZWEFW20cxOMbOeZjbEzDaF5WZmV4d5sLeZTa+85sQOiRsMFJvZ36OCng3MlPSWpE8kzZU0FEBSN0nzyraVNELSyHC9h6Q3Jc0O33eIpCaV1NNY0ivhtvMkXRiWT5bUP1wfI2m6pPmSfh+1z+WSfh9V5+EJ/HwqNXl8CwadswWANUsbMO+jJlz37Z6MOK8HC2cF/6kP7lXI1EnNiZQEXR2L5zRiwxfZyQg3LgadvZnJ41smO4z9tmJ5c47svYGmTXeTm1tC/+PWkddmJw+P6ctlw+cw9qmXuHz4bB5/rHeyQ62Vtp12c0ivnSyc1ZgWecVs3pADwOYN2bTIK05ydJUrO9EXy5Jsiey+OBKYUUF5IXCumW2VlAdMlTShmrqeAu40s3GSGhB8mRRVUs/pwBdm9m0ASRX9Hv6NmW2SlAm8JekoM5sTvpZvZsdI+hkwAriihsddK8VFYuqk5lx2S9DlFInAti2ZPPDyYhbOasSfruzG2KmfcdpFG1m5OJdrTj+Mtp2LOKL/DjKT//e0X7KySxnwrS38486OyQ5lv61a2Yzn/nM4t935LrsLM1m2tAWlpeLMs5bwyJijeX9KZ75+8ip+/ouP+c2vByU73P3SoFGEW/+2iIf+2JWd28unDpHqd1uKxHBhSCpIRq+8gNslnUxwh5ZOVNGlIakp0MnMxgGYWWFYnl1JPXOBv0i6C3jZzN6roNrvSRpOcPwdgCOAsqT8Yvg4AzivkpiGE1yZw0Gd4vsRfvx2U3r03knLNkE3RV6HYk48swAJDu+7k4wMKNiUSYvWEa76/Rd73nf9d3rS6ZDCuMZyoBw7eCtL5jZiS376tvQBJk08mEkTDwZg2GVzyN/QiEsvn8tDf+sLwHvvdubnN36czBD3W2ZWKbf+bTHvTMjjg9dbAbAlP5uWbYrYvCGHlm2KKNiYuv9+hii21D0JGS2Rbav5BJcglncJ0AboF16m+CXQACgpF0+DauqvsJ7wJOIxBMn5Nkm/jX6TpO4ELeBTzOwo4JVy+9odPkao5EvLzB42s/5m1r9N6/gOs5n835Z7ui4ATji9gNnvNwFg9dJciotE81YRCneKwp3BxzXjf03IzDK6Hrq7oipT3qChm5k8vlWyw6i15i2CL8U2bXZwwolrmPz2QWzc2IDeR20AoE/f9axZ0zSZIe4n4/o7P2fV0oaMe2zvhWhT32zJkO/mAzDku/l8+Ebqdj+VneiLx9wXiZbIr463CVqyw8uuhpF0FNAVWG9mxZIGh88hSKptJbUGtgNnARPNbJuk1ZLOMbP/SsoFMoHmFdUjqSOwycz+JWkL+3Y/NAN2AAWS2hGMNZycqA+hJgp3ZvDJe035+d17L/457aJN3HtjF4YPPozsbOOXD6xEgi0bs/nNxQejDGjdvphfPbgiiZHvv9yGEY45eSsP3JSaZ+1r4je//YBmzYooKRF/G30MO3bkMOre/lz5s1lkZpZSXJTJg/dX1E5Jbb36b2fIefl8vqAho1+eC8DYe7rw7N87cMvoJZz2vfWsX5PL7df0THKklTOUNt0XsgR2BIUJ8n6CFnMhsBwYCYwCmgDTgQHAGWa2XNJ1wM8JBlYvA5ab2UhJPYGHgDygGLgA2Aq8VL4e4DDgzwRdGsXAT81suqTJwIhw/XHgBIJLHwuACWb2uKTlQH8zyw9PCt5jZoOqOsb+fRrYtNe7VLVJWjqtc/olj1hlfq1HskNICFuanl/M1Zla+CoFpRtrlVG7925iI188KqZtLz30wxmxXNGXKAntZDGzL9g7Vi/awEq2H0WQsMuXLwa+GWM9y4HXK6hjUNT6pZXsv1vU+nRgUEXbOefSixk1mfsiqdKj59s552ohONEX98usE8KTsnOuXkiFk3ix8KTsnKvzjNgmsE8FnpSdc/WCt5Sdcy5FGFDqJ/qccy5V7NetnpLCk7Jzrs4z8NEXzjmXKszk3RfOOZdK/OIR55xLEcF8yt6n7JxzKULeUnbOuVQRDIlLj5Zyenx1OOdcLZTNfRHLEgtJLSQ9L2mBpM8kDZTUStIbkhaHjy3DbSVplKQlkuZIOqaquj0pO+fqhTjfo+8BgvneDwf6AJ8BNwFvmVlP4K3wOQRTCvcMl+HAmKoq9qTsnKvzgqk7FdNSnfC+nycDjwV1W5GZbQGGAmPDzcYC54TrQ4EnwrtaTwVaSOpAJTwpO+fqhVJTTAuQF97tvmwZXq6q7sAG4J+SZkp6VFJjoJ2ZrQ23Wcfee492IrihRpnVYVmF/ESfc67OC2aJi7kNml/NnUeyCO4Deq2ZfSTpAfZ2VQT7MzNJ+3VbJ28pO+fqvOAy64yYlhisBlab2Ufh8+cJkvSXZd0S4eP68PU1QPQ94zqHZRXypOycqweClnIsS3XMbB2wStJhYdEpwKfABGBYWDYMGB+uTwB+FI7CGAAURHVz7MO7L5xz9UKcr+i7FnhKUg7BTZ5/TNDIfVbS5cAK9t6f9FXgTGAJsDPctlKelJ1zdV7Z6Iv41WezgIr6nU+pYFsDro61bk/KtbR4XhPO6HlissOIu8xmdfdPI/LpomSHkBiqm72RQU6rPZ8lzjnnUoTfo88551KIASXeUnbOudTh3RfOOZcqzLsvnHMuZfgk9845l2K8peyccykinSa596TsnKvzDFFS6if6nHMuZXifsnPOpQrz7gvnnEsZ3qfsnHMpxpOyc86lCENE/ESfc86lDj/R55xzKcL8RJ9zzqUW86TsnHOpwickcs65lJIuLeX0OB3pnHO1YAaRUsW0xELScklzJc2SND0sayXpDUmLw8eWYbkkjZK0RNIcScdUVbcnZedcvVCKYlpqYLCZHW1mZTdQvQl4y8x6Am+FzwHOAHqGy3BgTFWVelJ2ztV5RtB9EctSC0OBseH6WOCcqPInLDAVaCGpQ2WVeFJ2ztUDwYm+WBYgT9L0qGV4BRUaMEnSjKjX25nZ2nB9HdAuXO8ErIp67+qwrEJ+os85Vy+YxbxpflSXRGVOMrM1ktoCb0ha8NV9mUmKfY9RPCmnoLz2uxnx58W0zCvGDF77TzvGj+3ITfcvpPPBuwBo0jTC9m2ZXHP20ckNtoayc0q5+4nZZOeUkpllTJmUx1Oju+15/cpblvCt89bx3f4nJS/IOGjcrIQb7llFt8MKMYN7f3EQn81onOywaq3zwYXcMubzPc/bH7SbJ+/pyLjH2iYxqtjEc/SFma0JH9dLGgccB3wpqYOZrQ27J9aHm68BukS9vXNYVqGUT8qSzgHGAV8zswWS2gAvAznAdWb2XrntHwXuNbNPD3iwcRKJiEfu6MbST5vQsHGEUeNmM/P9Ftx5/WF7trnips/ZuT3l//n2UVwkbr7sKAp3ZpKZVco9/5rN9HdbsXBOM3r22kbTZiXJDjEufvqHNUx/pxm3De9OVnYpuQ1Lkx1SXKxe1oCfnfY1ADIyjKemz+X9ic2THFX1gtEX8emtldQYyDCzbeH6t4A/ABOAYcCd4eP48C0TgGskPQMcDxREdXPsIx36lC8GpoSPAKcAc82sbwUJOdPMrkjnhAyweUMOSz9tAsCuHZmsWtqQ1u2KorYwTj5zI5NfyktOgLUiCndmApCVZWRmBb/wMjKMy0Ys47F7uiczuLho1DRC7+N3MPHpVgCUFGewY2v6fYFW5+iTtrF2RS7r1+QmO5SYmMW2xKAdMEXSbGAa8IqZTSRIxqdKWgwMCZ8DvAosA5YAjwA/q6rylP5LkdQEOAkYDLwU/ky4G2goqT8wENgAPETwIVwt6TZghJlNl3Q6cDuQSdBPdIqk44AHgAbALuDHZrZQ0qXA2UAj4BBgnJn96gAeboXadirkkCN2sHB2kz1lRx67lc352XyxomESI9t/GRnGA89/QseDdvHyvzuycE4zhv5gDR+905rN+enxH7wq7Q/aTcHGLH5x30oOPqKQxXMaMua3ndi9KzPZocXVoLM3M3l8y2SHEbN4dV+Y2TKgTwXlGwkajeXLDbg61vpTvaU8FJhoZouAjQTJ9bfAf8LxgbuAxsBHZtbHzKaUvTHs5ngE+K6Z9QEuCF9aAHzdzPqGdd0etb+jgQuB3sCFkqL7gfaQNLzszGyRFcbxcL+qQaMIt45eyEN/6v6VropBZ+Xzv5fTsZUcKC0V157Xjx8NHsChvbdxZL8tnHTaBiY8VekJ6bSSmQk9eu/k5SfyuPq0wyjcmcGF16yv/o1pJCu7lAHf2sK7L6dHUjZiGw6XClf9pXpSvhh4Jlx/hr1dGNEiwAsVlA8A3jWzzwHMbFNY3hx4TtI84D6gV9R73jKzAjMrBD4FulYUlJk9bGb9zax/jhrU9JhikplVyq2jF/LOhDZ8MKn1nvKMTOOEb23i3VfTNymX2bEtiznTWnDU8QV06LqLxyZO459vfERug1IenTgt2eHtt/y12WxYm83CmcGJvSmvtKBH711Jjiq+jh28lSVzG7ElPzvZocTMYlySLWW7LyS1Ar4J9A6HlmQSfGbzy21aaGaRGlT9R+AdMztXUjdgctRru6PWIyTt8zGuv30pq5Y2ZNw/O37llb4nbGH1sobkr0vPn/nNWhYRKclgx7YscnIj9D1hM88/2oUfnDxwzzYvTJ/CFacfl8Qoa2fzhmzyv8ih8yGFrF7agKNP2sbKRen571WZQUM3M3l8q2SHETsDi/ES6mRL2aQMnA88aWZXlhVI+h9fHVpSlanA3yR1N7PPJbUKW8vN2Tsc5dJ4BhwvvfptY8i5G/h8QSNGT5gFwNi/dOXj/7XkG2flMzmNuy5atSniF3csJCMDlGG8N7EN0/7Xuvo3ppm//l8nfv3gCrKyjXUrc/jLjQclO6S4yW0Y4ZiTt/LATel1TKnQNRGLVE7KFwN3lSt7ISx7rLo3m9mG8EqbFyVlEIwZPJXgROFYSbcCr8Q35PiYP6MZZ/Q8ocLX7v11zwMcTXwtX9SEa7/br8pt0n2MMsCy+Y249szDqt8wDe3elckFvfc5z5XyanDxSFJVmpQlPUgVXSxmdl1CItpb/+AKykYBo8qVNSn3fFDU+mvAa+Ve/xA4NKro1rD8ceDxqO3O2t/YnXOppWzui3RQVUt5+gGLwjnnEsmAdE/KZjY2+rmkRma2M/EhOedc/KVL90W1Q+IkDZT0KcH4XiT1kfS3hEfmnHNxI6w0tiXZYhmnfD9wGsHFG5jZbODkBMbknHPxlyYDlWMafWFmq6SvfIPUZFywc84ll9WNE31lVkk6ATBJ2cDPgc8SG5ZzzsVZCrSCYxFL98VVBJNpdAK+IJgfIubJNZxzLjUoxiW5qm0pm1k+cMkBiMU55xInTaa0jmX0xcGSXpK0QdJ6SeMlHXwggnPOubgoG6ccy5JksXRf/Bt4FugAdASeA55OZFDOORdvcZzkPqFiScqNzOxJMysJl38RTBDvnHPpI92HxIVTZwK8JukmgvmMjWAS+FcPQGzOORc/KdA1EYuqTvTNIEjCZUdyZdRrBtycqKCccy7elAKt4FhUNfdF+t/B0jnnIGglp8Al1LGI6Yo+SUcCRxDVl2xmTyQqKOeci7s4tpQlZRLMpLnGzM6S1J2gi7c1QS/DD82sSFIu8ATQj2CqigvNbHlVdccyJO53wIPhMphgkviz9/9wnHMuCeJ7oq/8lc13AfeZWQ9gM3B5WH45sDksv499b9yxj1hGX5xPcNvsdWb2Y4JbazePOXTnnEsFcUrKkjoD3wYeDZ+L4H6iz4ebjAXOCdeHhs8JXz9F5SYSKi+WpLzLzEqBEknNCG6rFOt98pxzLvlqdvFInqTpUcvwcrXdD/yKvdcItga2mFlJ+Hw1wbQUhI+rAMLXC8LtKxVLn/J0SS2ARwj6SrYDH8bwPuecSxk1GH2Rb2b9K6xDOgtYb2YzJA2KT2RfFcvcFz8LV/8uaSLQzMzmJCIY55xLmPic6DsROFvSmQQDH5oBDwAtJGWFreHOwJpw+zUEPQurJWURdP1urGoHlXZfSDqm/AK0ArLCdeecSxuy2JaqmNnNZtbZzLoBFwFvm9klwDsE598AhgHjw/UJ4XPC1982q/pi7qpayn+pKjaCjm0noeyYRhamlciWgmSHkDAFPxiQ7BASovlTHyU7hNSW2Cv6fg08I+k2YCbwWFj+GPCkpCXAJoJEXqWqLh4ZHIdAnXMu+RIwr4WZTQYmh+vLgOMq2KYQuKAm9da9Jp5zzlUk3S+zds65ukRpMsm9J2XnXP2QJi3lWC6zlqQfSPpt+PwgSfv0nTjnXKqKdeRFKswkF8sVfX8DBgIXh8+3AX9NWETOOZcIaXI7qFi6L443s2MkzQQws82SchIcl3POxVcKtIJjEUtSLg6nqTMASW1Im/vCOudcIBW6JmIRS1IeBYwD2kr6E8FVKbcmNCrnnIsnq0OjL8zsKUkzCKbvFHCOmX1Wzduccy611JWWsqSDgJ3AS9FlZrYykYE551xc1ZWkDLzC3huoNgC6AwuBXgmMyznn4qrO9CmbWe/o5+EMcT+rZHPnnHO1UOMr+szsE0nHJyIY55xLmLrSUpZ0Y9TTDOAY4IuEReScc/FWl0ZfAE2j1ksI+phfSEw4zjmXIHWhpRxeNNLUzEYcoHiccy7uRB040Vd2vylJJx7IgJxzLiHSPSkD0wj6j2dJmgA8B+woe9HMXkxwbM45Fx8pMgNcLGLpU25AcPfVb7J3vLIBnpSdc+mjDpzoaxuOvJjH3mRcJk2+c5xzLhCvlrKkBsC7QC5BDn3ezH4nqTvwDNAamAH80MyKJOUCTwD9CBq4F5rZ8srqr2o+5UygSbg0jVovW5xzLn1YjEv1dgPfNLM+wNHA6ZIGAHcB95lZD2AzcHm4/eXA5rD8vnC7SlXVUl5rZn+IKUQXV9k5pdz9xGyyc0rJzDKmTMrjqdHd9rx+5S1L+NZ56/hu/5OSF2Sc9B+0lav++AWZGcZrT7fi2dHtkh1SzNo2387vLnqHVk12Yib++9HXePb93lzz7Q856WsrKYlksHpjM257dhDbC3M5re9iLvnG7D3v79F+I8Me+C6L1+Yl8ShqrnGzEm64ZxXdDivEDO79xUF8NqNxssOqWhzvZm1mBmwPn2aHixF08X4/LB8LjATGAEPDdYDngdGSFNazj6qScvKn4K+GpHMIphX9mpktqGK7V4Hvm9mWAxRarRQXiZsvO4rCnZlkZpVyz79mM/3dViyc04yevbbRtFlJskOMi4wM4+rb13DzRQeTvzabB19dzNTXm7NycYNkhxaTSKkY9fIAFq5pQ6PcIh6/7kWmLe7MtEWdGfPa8URKM7j6jKkMGzyTv742gNdn9uT1mT0BOKT9Ru4aNintEjLAT/+whunvNOO24d3Jyi4lt2F6dNbWoPsiT9L0qOcPm9nDX6krGC48A+hBcCempcAWMyv7z7ka6BSudwJWAYQj2goIujjyK9p5Vd0Xp8R8CMlzMTCFvbeqqpCZnZkuCTkgCndmApCVZWRmBX9NGRnGZSOW8dg93ZMZXNwc1ncnXyzPYd3KXEqKM5g8vgUDTytIdlgx27itMQvXtAFg5+4clq9vQdvmO5i2uAuR0uC/1ryV7WjbYsc+7z316CW8OeuQAxpvPDRqGqH38TuY+HQrAEqKM9ixNU3uvxx790W+mfWPWh7epyqziJkdDXQGjgMOj1eYlSZlM9sUr50kgqQmwEkE/TUXhWUdJL0raZakeZK+HpYvl5QXrv9X0gxJ8yUNj6pvu6Q/SZotaaqkpP6OzsgwHnxxBv+e8iEzP2jBwjnN+M73v+Cjd1qzOT83maHFTev2xWz4Yu+dxfLXZpPXoTiJEe2/Di23cWjHjcxb2fYr5d85dgEfLuiyz/ZD+ixj0qweByq8uGl/0G4KNmbxi/tW8tfXF3L9n1eS2zCS7LBiotLYlpoIG3vvENzHtIWksm+ozsCacH0N0AWC6z+A5gQn/CoUy41TU9VQYKKZLQI2SupH0J/zevgN1geYVcH7LjOzfkB/4DpJrcPyxsDUsPP+XeAnle1Y0nBJ0yVNL7JdcTugaKWl4trz+vGjwQM4tPc2juy3hZNO28CEpzpV/2Z3QDXMKeaOH07i/pcGsnP33i+ZS7/5CSWlGUwMuyzK9OryJYVFWSz7stWBDrXWMjOhR++dvPxEHlefdhiFOzO48Jr1yQ6rerG2kmPo4pDURlKLcL0hcCrwGUFyPj/cbBgwPlyfED4nfP3tyvqTIb2T8sUEw08IHy8GPgZ+LGkk0NvMtlXwvuskzQamEnx7lf2PKQJeDtdnAN0q27GZPVz20yZHDWt7HFXasS2LOdNacNTxBXTouovHJk7jn298RG6DUh6dOC2h+060jeuyadOxaM/zvA7F5K/NTmJENZeZEeGOH07i9Zk9mTzv4D3l3+63kBO/toLfPf1Nyp+eGXL0Ut5Iw64LCH7NbFibzcKZwYm9Ka+0oEfvxDRM4kk1WGLQAXhH0hyCnPOGmb0M/Bq4UdISgj7jx8LtHwNah+U3AjdVVXmadAZ9laRWBGc6e0syguF7BvwSOBn4NvC4pHvN7Imo9w0ChgADzWynpMkEF8cAFEd9e0VI4mfTrGURkZIMdmzLIic3Qt8TNvP8o134wckD92zzwvQpXHH6cckKMS4WzmpEp+5FtOuym43rshk0dAt3Xt012WHVgPGbC/7H8vUtePq9o/aUDjh0JT8YNIuf/v1sdhd/9UtGMk45ailXjRl6oIONi80bssn/IofOhxSyemkDjj5pGysXpUl3WvxGX8wB+lZQvoygf7l8eSFwQaz1p2VSJvgJ8KSZXVlWIOl/BAl5ipk9Eg7YPoZg0HaZ5gTjBXdKOhwYcCCDjlWrNkX84o6FZGSAMoz3JrZh2v9aV//GNFMaEX/9TSdu//cyMjJh0jOtWLEoPUZeAPTpto4z+y1mydpWPHH98wCMmXgcN579PjlZEUb95BUA5q1sy90vngxA3+5rWb+lCV9sapa0uGvrr//XiV8/uIKsbGPdyhz+cuNByQ4pJulymbWq6NpIWZLeAe4ys4lRZdcBNxDMz1FMMI7wR2b2uaTlBH3I24D/EnRNLARaACPNbLKk7WbWJKzrfOAsM7u0uliaZ7Wxgc3Ss9VTlciW9BkFUVMFP0jJ7+Jaa/7UR8kOISE+Kn2TrbapVkN0G7XrYj0vurH6DYE5o26cYWb9a7O/2kjLlrKZDa6gbBQwqpLtu0U9PaOSbZpErT9PMMjbOVcX1LFJ7p1zLv2lSaeAJ2XnXL2QLn3KnpSdc/WDJ2XnnEsd3lJ2zrlUYdSJSe6dc65OqBM3TnXOuTrFk7JzzqUOpcmFcp6UnXN1XxzvPJJonpSdc/WC9yk751wK8cusnXMulXhL2TnnUoR594VzzqUWT8rOOZca/OIR55xLMSpNj6yczjdOdc652MT3btZdJL0j6VNJ8yX9PCxvJekNSYvDx5ZhuSSNkrRE0hxJx1RVvydl51y9oNLYlhiUAL8wsyMI7vN5taQjCO5S/ZaZ9QTeYu9dq88AeobLcGBMVZV7UnbO1Q9xaimb2Voz+yRc3wZ8BnQChgJjw83GAueE60OBJywwFWghqUNl9XufsnOuXqjBib48SdOjnj9sZg9XWKfUDegLfAS0M7O14UvrgHbheidgVdTbVodla6mAJ2XnXN1nQOwTEuXHcjdrSU2AF4DrzWyrtPeG22Zm0v6N9/CkXFuZGah5s2RHEXfaviPZISRMy3Fzkx1CQkQGHpXsEBJj1vtxqSael1lLyiZIyE+Z2Yth8ZeSOpjZ2rB7Yn1YvgboEvX2zmFZhbxP2TlX55WNU45lqbauoEn8GPCZmd0b9dIEYFi4PgwYH1X+o3AUxgCgIKqbYx/eUnbO1X1mNem+qM6JwA+BuZJmhWW3AHcCz0q6HFgBfC987VXgTGAJsBP4cVWVe1J2ztUL8bqiz8ymEDS+K3JKBdsbcHWs9XtSds7VD+lxQZ8nZedc/eBzXzjnXKowIJIeWdmTsnOuXvCWsnPOpRK/m7VzzqUObyk751yqiHGyoVTgSdk5V+cJkJ/oc8651CHvU3bOuRTh3RfOOZdK4jr3RUJ5UnbO1Qs++sI551KJt5Sdcy5FmI++cM651JIeOdmTsnOufvAhcc45l0o8KTvnXIowII43Tk0kT8rOuTpPmHdfuNpp3KSY626eTddDtoGJ+//Uh6LdGVz9q7nk5JQSiYi/3XMkiz5tmexQa+SGPy/n+FMK2LIxi6tO7QXAFbes5vghWygpzuCLFbncO6IrO7am159mXvvdjPjzYlrmFWMGr/2nHePHdgTg7B+u5axL1lFaCtMmt+Qfd3dLbrAxuPFn7zOg3xq2FDRg+I1nAzDsopkMPHYVViq2bG3An0efyKbNjRh47EqGXTQLKxWR0gzG/LM/8xe0S/IRVKA0Pk1lSf8AzgLWm9mRYVkr4D9AN2A58D0z2xze+foBghun7gQuNbNPqqo/rf7yJW03syY12H4QMMLMzpJ0NnCEmd2ZqPjiafgN85kxtS13/KY/WVml5DaIcNNtM/j3Y4cyY2pb+g/8kh9f/Rk3X31CskOtkTeea81LY9sy4r7P95R98l4z/nFXJ0oj4rKbV3Ph1ev4xx2dkxhlzUUi4pE7urH00yY0bBxh1LjZzHy/BS1aFzPglE1cfXYfiosyaN6qKNmhxuSNd3ow4bXD+dW17+8pe258L8Y+0xeAc878jB9cMIdRDw9g5twOfPhxF0B077qZW2/8H5f//JzkBF6Z+HZfPA6MBp6IKrsJeMvM7pR0U/j818AZQM9wOR4YEz5WKiNuYaY4M5uQLgm5UeNijjx6I5Ne6gJASUkGO7ZnYyYaNS4BoHGTEjblN0hmmPtl3rSmbNuS+ZWyT95rRmkkuDnwgk8ak9e+OBmh1crmDTks/TRoL+zakcmqpQ1p3a6Ib39/Hc8+3IniouC/WsGmnGSGGbO5n7Vj2/bcr5Tt3LU39ga5JXvOmxUWZlN2c+egvLIbPSeXzGJaqmNm7wKbyhUPBcaG62OBc6LKn7DAVKCFpA5V1Z9WLeUyYQt4JJAPHAnMAH5gZibpdOB+gp8KU6LecynQ38yukfQd4FYgB9gIXGJmX0oaCRwEHBw+3m9mow7IQUVp33EnBVtyuOHW2XTvuZUlC5rz0H29eOT+I/jD/R9x+bWfogxjxPATD3RoCfetCzfy7kvp1SVTXttOhRxyxA4Wzm7C5b9ezpH9tzLsxpUU787g0Tu7smhu02SHuN8uvXgmp35jKTt25vDLkd/aU37icSu57JJPaN6skP+745QkRliFxPYptzOzteH6OqCs/6YTsCpqu9Vh2Voqkc4t5b7A9cARBEn0REkNgEeA7wD9gPaVvHcKMMDM+gLPAL+Keu1w4DTgOOB3krITEn0VMjKNHodu5dUXu3LdsJMp3JXJBT9aypnnreCRB3px6TlDeOSBXlx/y5wDHVpCXXTNWiIl4u1xrZIdyn5r0CjCraMX8tCfurNzexaZmUbT5iXccH5vHr2rKzc/sIi0uYqhAo8/3ZdLrjqft9/rztmnL9hT/v60g7j85+fw+7sHM+yimUmMsDLhhESxLJAnaXrUMrxGezKr1Zx06ZyUp5nZajMrBWYRdLAfDnxuZovDD+Zflby3M/C6pLnAL4FeUa+9Yma7zSwfWM/eb7w9JA0v+wcriuyK3xGFNq5vQP6GBiwMT+K9/04HehxawClnruaDycH3zJS3OnDoEVvivu9kOfX8fI4/pYC7r+tO2U/hdJOZVcqtoxfyzoQ2fDCpNQD563J5f1JrQCya0xQzaN6qJLmBxsFb73Xn6wNW7lM+97N2dGi3nWZNC5MQVRXK7mYdywL5ZtY/ank4hj18WdYtET6uD8vXAF2ituscllUqnZPy7qj1CDXrinkQGG1mvYErgejO2WrrNbOHy/7BcjIb1mC3sdm8qQEbvmxIp4O2A9Cnfz4rlzdhU34DevfdGJZt5ItVjeO+72To940Czv/pl4y8/BB2F6brn6Rx/e1LWbW0IeP+2XFP6YdvtqLPgAIAOnXbRVa2UbApLXsN6dh+6571E45dxao1zaLKg4Zhj+4byc6KsHVbbkVVJFW8+pQrMQEYFq4PA8ZHlf9IgQFAQVQ3R4XS86+jcguAbpIOMbOlwMWVbNecvd9WwyrZJqkeurcXvxw5k6zsUtatacT9f+rD1Hfbc+UN88nILKW4KJMH7+yd7DBr7KYHl3HUwG00a1nCkx/N4V/3duTCq9eRnVPK7U8tBmDBzMY8eEvXJEdaM736bWPIuRv4fEEjRk+YBcDYv3Rl0vNtueGOJYx5ZSYlxRn85Vc9SYdfAjdf/y5H9fqS5k0Leeqh53nyP3049pg1dOm4lVKD9Rua8MDDAwA4acBKhnxjKZGSDHYXZfKn+04mJY8xTn3Kkp4GBhF0c6wGfgfcCTwr6XJgBfC9cPNXCYbDLSE4z/Xjauu3NBlQDXuHxEUPdQvLRwPTzezxcif63gMOCYfEXcreE31DgfuAzcDbwLFmNig80bfdzO4J650HnGVmyyuLqXluOzuh4yUJOd5kiqyp8ss8rSk39Vpx8RDp0yPZISTEtFlj2Lp9Ta2yfPMGHeyErrG1vyYuumuGmfWvzf5qI61aymVjlM1sMjA5qvyaqPWJBH3L5d/7OMH4QsxsPHt/XkRvM7Lc8yPjELZzLun8ziPOOZdaPCk751yKMCCSHjMSeVJ2ztUDBuZJ2TnnUod3XzjnXIowoNSTsnPOpQ5vKTvnXArxpOyccynCDCKRZEcRE0/Kzrn6wVvKzjmXQjwpO+dcqjAffeGccynDwPziEeecSyF+mbVzzqUIMyj1pOycc6nDT/Q551zqMG8pO+dcqvBJ7p1zLnX4hETOOZc6DLA0ucw6Xe/n7pxzsbNwkvtYlhhIOl3SQklLJN0Uz1C9peycqxcsTt0XkjKBvwKnAquBjyVNMLNP41G/t5Sdc/VD/FrKxwFLzGyZmRUBzwBD4xWmLE3OSKYqSRuAFQdod3lA/gHa14FUV48L6u6xHcjj6mpmbWpTgaSJBDHHogFQGPX8YTN7OKqu84HTzeyK8PkPgePN7JraxFjGuy9qqbZ/LDUhabqZ9T9Q+ztQ6upxQd09tnQ7LjM7PdkxxMq7L5xzrmbWAF2inncOy+LCk7JzztXMx0BPSd0l5QAXARPiVbl3X6SXh6vfJC3V1eOCuntsdfW4qmVmJZKuAV4HMoF/mNn8eNXvJ/qccy6FePeFc86lEE/KzjmXQjwpx5Gk9pKekbRU0gxJr0o6NEmxPCrpiAO4v3MkmaTDw+dtJH0kaaakryc7vngpf5xVbPeqpBYHKKz9Jml7DbcfJOnlcP3seF9i7LxPOW4kCfgAGGtmfw/L+gDNzOy9pAZ3AEj6D9AReNvMfifpImBI2QD7cttmmll6zA5TTvnjTHY8tSVpu5k1qcH2g4ARZnZWwoKq57ylHD+DgeKyhAxgZrOBmZLekvSJpLmShgJI6iZpXtm2kkZIGhmu95D0pqTZ4fsOkdSkknoaS3ol3HaepAvD8smS+ofrYyRNlzRf0u+j9rlc0u+j6qyy9VcZSU2Ak4DLgYskHQ3cDQyVNEtSQ0nbJf1F0mxgYLn4Tg9jmC3prbDsOEkfhi3tDyQdFpZfKulFSRMlLZZ09/7EHI/jDMs6SHo3PM55Zb8Kws82L1z/b/jLab6k4VH1bZf0p/C4p0pqd6COpbywBTxZ0vOSFkh6KmxolP37LJD0CXBe1HsulTQ6XP9O1C+jN8uORdJISf8I614m6bqkHGA6MTNf4rAA1wH3VVCeRdBahuAyzyWAgG7AvKjtRgAjw/WPgHPD9QZAoyrq+S7wSFQ9zcPHyUD/cL1V+JgZlh8VPl8OXBuu/wx4dD+P/RLgsXD9A6AfcCkwOmobA74X9Xwy0B9oA6wCupeLtRmQFa4PAV4I1y8FlgHNw89mBdDlAP0bV3ScvwB+E/X5No36bPPKHVNDYB7QOuoz+U64fjdwaxL+breHj4OAAoILITKADwm+gBqE/z49w7+3Z4GXo/4tRofrLdn7y/sK4C/h+sjws8oN/243AtkH+jjTafGWcuIJuF3SHOBNoBNQaYtIUlOgk5mNAzCzQjPbWUU9c4FTJd0l6etmVlBBtd8LWzkzgV5AdF/ui+HjDIIviv1xMcGkLISPF1ewTQR4oYLyAcC7ZvY5gJltCsubA8+FvybuC+Mu85aZFZhZIfAp0HU/466pio7zY+DH4a+c3ma2rYL3XRf+QphKcCVYz7C8CHg5XK/N5x8v08xstZmVArMI4jkc+NzMFluQZf9VyXs7A69Lmgv8kq/+e71iZrvNLB9YTxV//84vHomn+cD5FZRfQtAa7GdmxZKWE7Q+Svhq91GDauqvsB4zWyTpGOBM4DZJb5nZH8reJKk7QSv8WDPbLOnxcvvaHT5G2I+/B0mtgG8CvSUZQWvRCD6PaIVWs37kPwLvmNm5kroRtKzLx7zfcddUFcf5S+Bk4NvA45LuNbMnot43iKClP9DMdkqazN7PvzhMdAfsOKpRm8/1QeBeM5sQHvPIONVb73hLOX7eBnLL9RkeRdCKWx8m0sHsbdV9CbSV1FpSLnAWQNjSWi3pnLCOXEmNCFqO+9QjqSOw08z+BfwZOKZcXM2AHUBB2M93RpyP+3zgSTPrambdzKwL8DlfnRugKlOBk8Mvj7LkB8Hxls0ncGkc491flR3nycCXZvYI8Cj7fv7Ngc1hQj6c4JdBOlkAdJN0SPi8ol9B8NV/r2EJj6oO86QcJ2GL51xgiIIhcfOBO4BXgf7hz7ofEfyRY2bFwB+AacAbZeWhHxL85J1D0B/XHniqonqA3sA0SbOA3wG3lYtrNkG3xQLg38D78T1yLgbGlSt7Abg5ljeb2QZgOPBi+BP/P+FLdwN3SJpJarSsKjvOx4HZYZwXAg+U22YikCXpM+BOgi+htBF2EQ0HXgm7wNZXsulIgu6mGdTNqUoPGB8S55xzKcRbys45l0I8KTvnXArxpOyccynEk7JzzqUQT8rOOZdCPCm7hJIUiZoX4rlwzPX+1vW4gjsJVzvLXDiXwwn7sY89c1bEUl5um5rOuDZS0oiaxujqNk/KLtF2mdnRZnYkwWXFV0W/KGm/xiCb2RVm9mkVmwwCapyUnUs2T8ruQHoP6BG2Yt+TNAH4VFKmpD9L+ljSHElXQjAdqqTRkhZKehNoW1aRqphlLrws+yrghrCV/nUF8zu/EO7jY0knhu9tLWmSghncHiWYY6RKqmTWt/C1+8LytyS1CcsOUTCr3YzwuPdrNj5XP6TClVKuHghbxGcQXOEGweXIR5rZ52FiKzCzY8NLzt+XNAnoCxxGMIFSO4LJh/5Rrt42wCPAyWFdrcxsk6S/E8yAdk+43b8JZvGbIukggptefo3gKsgpZvYHSd8mmJazOpeF+2gIfCzpBTPbCDQGppvZDZJ+G9Z9DcFNRq8ys8WSjgf+RjCPhnP78KTsEq1heAk4BC3lxwi6FaaVzQwHfAs4qqy/mGAehZ4E80o8HU5k9IWktyuov7JZ5sobAhwh7WkIN1MwP/LJhHMEm9krkjbHcEzXSTo3XC+b9W0jUMrey8T/RXDpeJPweJ+L2nduDPtw9ZQnZZdou8zs6OiCMDntiC4imNf59XLbnRnHODKAAeFcDuVjiVk1s76VZ+F+t5T/DJyrjPcpu1TwOvBTSdkAkg6V1Bh4F7gw7HPuQHB3l/Iqm2VuG9A0artJwLVlTxTcHYVwH98Py84gmKy9KlXN+pbB3ulbv0/QLbIV+FzSBeE+pOA2Yc5VyJOySwWPEvQXf6JgUvuHCH7FjQMWh689QXA3jK+oYpa5l4Bzy070EdwZpn94IvFT9o4C+T1BUp9P0I2xsppYq5r1bQdwXHgM3ySYBRCCubAvD+ObDwyN4TNx9ZTPEueccynEW8rOOZdCPCk751wK8aTsnHMpxJOyc86lEE/KzjmXQjwpO+dcCvGk7JxzKeT/AUG6gkGcpBTdAAAAAElFTkSuQmCC",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "train_x, test_x, train_y, test_y = train_test_split(x_train, y_train, test_size=0.2)\n",
    "\n",
    "train_y_cat = tf.keras.utils.to_categorical(train_y, 4)\n",
    "test_y_cat = tf.keras.utils.to_categorical(test_y, 4)\n",
    "\n",
    "# trainar o modelo:\n",
    "# Input scaled from (0, 255) to a range of (0, +1.)\n",
    "mlp = models.Sequential()\n",
    "mlp.add(Input(shape=(2500,)))\n",
    "mlp.add(layers.Rescaling(scale=1 / 255))  # scale 0-255 to 0-1\n",
    "mlp.add(layers.Dense(1000, activation=\"relu\"))\n",
    "mlp.add(layers.Dense(1000, activation=\"relu\"))\n",
    "# if softmax is chosen, the model will have more parameters that need to learn\n",
    "mlp.add(layers.Dense(4, activation=\"softmax\"))  # output layer\n",
    "mlp.summary()\n",
    "mlp.compile(loss=\"categorical_crossentropy\", optimizer=\"adam\", metrics=[\"accuracy\"])\n",
    "history = mlp.fit(\n",
    "    train_x, train_y_cat, epochs=20, batch_size=20, validation_data=(test_x, test_y_cat)\n",
    ")\n",
    "\n",
    "# plt.plot(history.history['accuracy'], label='accuracy')\n",
    "# plt.plot(history.history['val_accuracy'], label = 'val_accuracy')\n",
    "# plt.xlabel('Epoch')\n",
    "# plt.ylabel('Accuracy')\n",
    "# plt.ylim([0.5, 1])\n",
    "# plt.legend(loc='lower right')\n",
    "\n",
    "y_hat_test_cat = mlp.predict(test_x)\n",
    "y_hat_test = np.argmax(y_hat_test_cat, axis=-1)\n",
    "# test_loss, test_acc = mlp.evaluate(test_x,  test_y, verbose=2)\n",
    "\n",
    "ConfusionMatrixDisplay.from_predictions(\n",
    "    test_y, y_hat_test, display_labels=[\"Caucasian\", \"African\", \"Asian\", \"Indian\"]\n",
    ")\n",
    "\n",
    "bacc = balanced_accuracy_score(test_y, y_hat_test)\n",
    "\n",
    "print(\"BACC(MLP):\", bacc)\n",
    "print(\"Y_hat shape:\", y_hat_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "BACC = 0.809608093469514\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y_hat = (mlp.predict(test_x) > 0.5).astype(\"int32\")\n",
    "bacc = balanced_accuracy_score(test_y, y_hat)\n",
    "ConfusionMatrixDisplay.from_predictions(\n",
    "    test_y, y_hat, display_labels=[\"Male\", \"Female\"]\n",
    ")\n",
    "print(f\"BACC = {bacc}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Deep Neural Network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 168,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_77\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "rescaling_66 (Rescaling)     (None, 50, 50, 1)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_132 (Conv2D)          (None, 46, 46, 12)        312       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_132 (MaxPoolin (None, 23, 23, 12)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_133 (Conv2D)          (None, 19, 19, 64)        19264     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_133 (MaxPoolin (None, 9, 9, 64)          0         \n",
      "_________________________________________________________________\n",
      "dropout_53 (Dropout)         (None, 9, 9, 64)          0         \n",
      "_________________________________________________________________\n",
      "flatten_66 (Flatten)         (None, 5184)              0         \n",
      "_________________________________________________________________\n",
      "dense_132 (Dense)            (None, 64)                331840    \n",
      "_________________________________________________________________\n",
      "dense_133 (Dense)            (None, 4)                 260       \n",
      "=================================================================\n",
      "Total params: 351,676\n",
      "Trainable params: 351,676\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "def make_model():\n",
    "    model = models.Sequential()\n",
    "    model.add(Input(shape=(50, 50, 1)))\n",
    "    model.add(layers.Rescaling(scale=1 / 255))  # scale 0-255 to 0-1\n",
    "    model.add(layers.Conv2D(12, (5, 5), activation=\"relu\"))\n",
    "    model.add(layers.MaxPooling2D((2, 2)))\n",
    "    model.add(layers.Conv2D(64, (5, 5), activation=\"relu\"))\n",
    "    model.add(layers.MaxPooling2D((2, 2)))\n",
    "    model.add(layers.Dropout(0.4))\n",
    "    # Flatten\n",
    "    model.add(layers.Flatten())\n",
    "    model.add(layers.Dense(64, activation=\"relu\"))\n",
    "    # model.add(layers.Dropout(0.3))\n",
    "    model.add(layers.Dense(4, activation=\"softmax\"))\n",
    "    return model\n",
    "\n",
    "t = make_model()\n",
    "t.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 150,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Callbacks, \n",
    "#   Earlystoping and\n",
    "#   Learning Rate Scheduler\n",
    "\n",
    "# learning step\n",
    "def ln_scheduler(epoch, lr):\n",
    "    if epoch < 10:\n",
    "        return lr\n",
    "    else:\n",
    "        return lr * tf.math.exp(-0.1)\n",
    "\n",
    "ln_callback = tf.keras.callbacks.LearningRateScheduler(ln_scheduler)\n",
    "# testar com outro monitor = \"bacc\"\n",
    "earlystop = tf.keras.callbacks.EarlyStopping(\n",
    "    monitor=\"val_loss\", patience=5, restore_best_weights=True\n",
    ")  # restore_best_weights = False"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train Test Split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 161,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_x, test_x, train_y, test_y = train_test_split(x_train, y_train, test_size=0.2)\n",
    "\n",
    "train_x = np.reshape(train_x, (len(train_x), 50, 50, 1))\n",
    "test_x = np.reshape(test_x, (len(test_x), 50, 50, 1))\n",
    "\n",
    "train_y_cat = tf.keras.utils.to_categorical(train_y, 4)\n",
    "test_y_cat = tf.keras.utils.to_categorical(test_y, 4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Class weights and Data Augmentation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{0: 0.4112326931665922, 1: 5.400293255131965, 2: 1.3763079222720478, 3: 1.5231596360628619}\n"
     ]
    }
   ],
   "source": [
    "# imbalanced data class weights\n",
    "from sklearn.utils.class_weight import compute_class_weight\n",
    "\n",
    "class_weight = compute_class_weight(\n",
    "    class_weight=\"balanced\", classes=[0, 1, 2, 3], y=y_train\n",
    ")\n",
    "class_weights = { c: w for c, w in enumerate(class_weight)}\n",
    "print(class_weights)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create a data augmentation stage with horizontal flipping, rotations, zooms\n",
    "data_augmentation = models.Sequential(\n",
    "    [\n",
    "        layers.RandomFlip(\"horizontal\", input_shape=(50, 50, 1)),\n",
    "        layers.RandomRotation(0.1),\n",
    "        layers.RandomZoom(0.1),\n",
    "    ]\n",
    ")\n",
    "\n",
    "# Load some data\n",
    "classes = 4\n",
    "\n",
    "# Create a tf.data pipeline of augmented images (and their labels)\n",
    "# train_dataset = tf.data.Dataset.from_tensor_slices((train_x, train_y_cat))\n",
    "# train_dataset = train_dataset.batch(16).map(lambda x, y: (data_augmentation(x), y))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 162,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(5892, 50, 50, 1)\n",
      "(6155, 50, 50, 1)\n"
     ]
    }
   ],
   "source": [
    "# Duplicate African Class Images with augmentation\n",
    "africans = train_x[train_y==1]\n",
    "africans_augmented = data_augmentation(africans)\n",
    "# plt.imshow(africans_augmented[0])\n",
    "\n",
    "train_x_plus_africans_augmented = np.append(train_x, africans_augmented, axis=0)\n",
    "print(train_x.shape)\n",
    "print(train_x_plus_africans_augmented.shape)\n",
    "\n",
    "train_y_plus_africans_augmented = np.append(train_y, np.ones(len(africans_augmented)))\n",
    "train_y_cat_plus_africans = tf.keras.utils.to_categorical(train_y_plus_africans_augmented, 4)\n",
    "\n",
    "#class_weights_augmented = compute_class_weight(\n",
    "#    class_weight=\"balanced\", classes=[0, 1, 2, 3], y=y_train\n",
    "#)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 163,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "263\n",
      "526\n",
      "78\n"
     ]
    }
   ],
   "source": [
    "print(sum(train_y == 1))\n",
    "print(sum(train_y_plus_africans_augmented == 1))\n",
    "print(sum(test_y == 1))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train and Evaluate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 164,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/40\n",
      "97/97 [==============================] - 7s 70ms/step - loss: 1.4174 - categorical_accuracy: 0.2478 - val_loss: 1.1658 - val_categorical_accuracy: 0.5075\n",
      "Epoch 2/40\n",
      "97/97 [==============================] - 7s 76ms/step - loss: 0.9263 - categorical_accuracy: 0.5971 - val_loss: 0.7455 - val_categorical_accuracy: 0.7429\n",
      "Epoch 3/40\n",
      "97/97 [==============================] - 7s 72ms/step - loss: 0.8047 - categorical_accuracy: 0.6725 - val_loss: 0.7884 - val_categorical_accuracy: 0.7198\n",
      "Epoch 4/40\n",
      "97/97 [==============================] - 8s 80ms/step - loss: 0.7549 - categorical_accuracy: 0.6923 - val_loss: 0.7619 - val_categorical_accuracy: 0.7049\n",
      "Epoch 5/40\n",
      "97/97 [==============================] - 8s 83ms/step - loss: 0.6608 - categorical_accuracy: 0.7241 - val_loss: 0.6373 - val_categorical_accuracy: 0.7680\n",
      "Epoch 6/40\n",
      "97/97 [==============================] - 8s 82ms/step - loss: 0.6093 - categorical_accuracy: 0.7423 - val_loss: 0.7031 - val_categorical_accuracy: 0.7320\n",
      "Epoch 7/40\n",
      "97/97 [==============================] - 8s 84ms/step - loss: 0.5817 - categorical_accuracy: 0.7582 - val_loss: 0.6342 - val_categorical_accuracy: 0.7687\n",
      "Epoch 8/40\n",
      "97/97 [==============================] - 8s 83ms/step - loss: 0.5358 - categorical_accuracy: 0.7729 - val_loss: 0.5885 - val_categorical_accuracy: 0.7856\n",
      "Epoch 9/40\n",
      "97/97 [==============================] - 8s 84ms/step - loss: 0.5053 - categorical_accuracy: 0.7894 - val_loss: 0.5282 - val_categorical_accuracy: 0.8114\n",
      "Epoch 10/40\n",
      "97/97 [==============================] - 8s 86ms/step - loss: 0.4739 - categorical_accuracy: 0.7989 - val_loss: 0.6193 - val_categorical_accuracy: 0.7626\n",
      "Epoch 11/40\n",
      "97/97 [==============================] - 8s 85ms/step - loss: 0.4416 - categorical_accuracy: 0.8132 - val_loss: 0.5615 - val_categorical_accuracy: 0.7897\n",
      "Epoch 12/40\n",
      "97/97 [==============================] - 8s 85ms/step - loss: 0.4366 - categorical_accuracy: 0.8107 - val_loss: 0.5593 - val_categorical_accuracy: 0.7985\n",
      "Epoch 13/40\n",
      "97/97 [==============================] - 8s 87ms/step - loss: 0.3962 - categorical_accuracy: 0.8307 - val_loss: 0.5281 - val_categorical_accuracy: 0.8060\n",
      "Epoch 14/40\n",
      "97/97 [==============================] - 8s 87ms/step - loss: 0.3746 - categorical_accuracy: 0.8338 - val_loss: 0.4992 - val_categorical_accuracy: 0.8175\n",
      "Epoch 15/40\n",
      "97/97 [==============================] - 8s 86ms/step - loss: 0.3521 - categorical_accuracy: 0.8426 - val_loss: 0.4340 - val_categorical_accuracy: 0.8399\n",
      "Epoch 16/40\n",
      "97/97 [==============================] - 8s 86ms/step - loss: 0.3476 - categorical_accuracy: 0.8435 - val_loss: 0.4660 - val_categorical_accuracy: 0.8372\n",
      "Epoch 17/40\n",
      "97/97 [==============================] - 9s 88ms/step - loss: 0.3297 - categorical_accuracy: 0.8525 - val_loss: 0.4595 - val_categorical_accuracy: 0.8318\n",
      "Epoch 18/40\n",
      "97/97 [==============================] - 9s 88ms/step - loss: 0.2975 - categorical_accuracy: 0.8640 - val_loss: 0.5080 - val_categorical_accuracy: 0.8195\n",
      "Epoch 19/40\n",
      "97/97 [==============================] - 9s 88ms/step - loss: 0.2943 - categorical_accuracy: 0.8642 - val_loss: 0.4282 - val_categorical_accuracy: 0.8494\n",
      "Epoch 20/40\n",
      "97/97 [==============================] - 9s 88ms/step - loss: 0.2812 - categorical_accuracy: 0.8731 - val_loss: 0.4777 - val_categorical_accuracy: 0.8236\n",
      "Epoch 21/40\n",
      "97/97 [==============================] - 8s 88ms/step - loss: 0.2760 - categorical_accuracy: 0.8746 - val_loss: 0.5180 - val_categorical_accuracy: 0.8182\n",
      "Epoch 22/40\n",
      "97/97 [==============================] - 9s 89ms/step - loss: 0.2666 - categorical_accuracy: 0.8728 - val_loss: 0.4792 - val_categorical_accuracy: 0.8338\n",
      "Epoch 23/40\n",
      "97/97 [==============================] - 8s 87ms/step - loss: 0.2665 - categorical_accuracy: 0.8780 - val_loss: 0.4653 - val_categorical_accuracy: 0.8372\n",
      "Epoch 24/40\n",
      "97/97 [==============================] - 9s 88ms/step - loss: 0.2466 - categorical_accuracy: 0.8840 - val_loss: 0.4933 - val_categorical_accuracy: 0.8290\n",
      "BACC = 0.803\n",
      "Y_hat shape: (1474,)\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEKCAYAAAAfGVI8AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/MnkTPAAAACXBIWXMAAAsTAAALEwEAmpwYAABTnElEQVR4nO3deViU5frA8e8NDLIIgoK4oKK54L6b5b5VtmhlZmamVlpWWnnq2HbKylbbjm1m/dJcKkmzzFxOLmUuuSvu5pooKSCyyjrP7493QGQdYIYZhudzXVzCzDvv+4wMc8+z3bcopdA0TdOqLjdHN0DTNE1zLB0INE3TqjgdCDRN06o4HQg0TdOqOB0INE3TqjgdCDRN06o4uwUCEflKRC6IyP4i7hcRmSkix0QkUkQ62astmqZpWtHs2SOYC9xUzP2DgWaWrwnAZ3Zsi6ZpmlYEuwUCpdQG4GIxhwwF5inDn0CAiNS1V3s0TdO0wnk48Nr1gTN5fo6y3Bad/0ARmYDRa8DX17dzeHh4hTRQq3p27twZq5QKdsS1g4KCVFhYmCMurVUBxb22HRkIrKaUmg3MBujSpYvasWOHg1ukuSoROe2oa4eFhaFf25q9FPfaduSqobNAgzw/h1pu0zRN0yqQIwPBMuB+y+qh7kCCUqrAsJCmaZpmX3YbGhKRb4G+QJCIRAEvAyYApdQsYAVwM3AMSAXG2astmqZpWtHsFgiUUiNLuF8Bj9nr+pqmuZbMzEyioqJIS0tzdFOcmpeXF6GhoZhMJqsfUykmizVN06KiovDz8yMsLAwRcXRznJJSiri4OKKiomjcuLHVj9MpJjRNqxTS0tKoVauWDgLFEBFq1apV6l6TDgSaplUaOgiUrCz/RzoQaJoTi0tO5/1fj3IoOtHRTdFcmA4EmlZKJSVUzHNcVxHJEpG7ynqt1IxsZq79i31nE8p6Cs2Gqlev7ugm2IUOBJpWenMpPqEiIuIOvA38rzwXCvAxVn5cSs0oz2k0rVg6EGhaKVmRUBFgErAEuFCea1Wv5oGHmxCfmlme02g2ppTimWeeoU2bNrRt25ZFixYBEB0dTe/evenQoQNt2rThjz/+IDs7m7Fjx+Ye+8EHHzi49QXp5aOaZmMiUh+4A+gHdC3h2NyEig0bNizsfgJ8PLmkA8FVXvn5AAfP2XbepFU9f16+rbVVx/7www/s2bOHvXv3EhsbS9euXenduzfffPMNN954Iy+88ALZ2dmkpqayZ88ezp49y/79xkjipUuXbNpuW9A9Ak2zvQ+BqUopc0kHKqVmK6W6KKW6BAcXnvQ0wMekh4aczMaNGxk5ciTu7u6EhITQp08ftm/fTteuXZkzZw7Tpk1j3759+Pn50aRJE06cOMGkSZNYtWoV/v7+jm5+AbpHoGm21wX4zrKMLwi4WUSylFI/luVkgT4m3SPIx9pP7hWtd+/ebNiwgV9++YWxY8cyZcoU7r//fvbu3cvq1auZNWsWERERfPXVV45u6lV0j0DTbEwp1VgpFaaUCgMWA4+WNQgA1PD2JF73CJxKr169WLRoEdnZ2cTExLBhwwa6devG6dOnCQkJYfz48Tz00EPs2rWL2NhYzGYzw4YNY/r06ezatcvRzS9A9wg0rZSsSKhoU4E+Jg6c0z0CZ3LHHXewZcsW2rdvj4jwzjvvUKdOHb7++mtmzJiByWSievXqzJs3j7NnzzJu3DjMZmOk8M0333Rw6wvSgUDTSqmkhIr5jh1b3usF+Jh0j8BJJCcnA8Yk/owZM5gxY8ZV948ZM4YxY8YUeJwz9gLy0kNDmubkAnw8Scs0k5aZ7eimaC5KBwJNc3KBPp4AesJYsxsdCDTNyeXsLtbDQ5q96ECgaU7uSpoJ3SPQ7EMHAk1zcgHeOUNDukeg2YcOBJrm5AJ9LT2Cy7pHoNmHDgSa5uRyegR6jkCzFx0INM3JeXu6U83DjQQ9R1CpFFe74NSpU7Rp06YCW1M8HQg0rRII9NFpJjT70TuLNa0SMHYX6x5BrpXPwj/7bHvOOm1h8FtF3v3ss8/SoEEDHnvsMQCmTZuGh4cH69evJz4+nszMTKZPn87QoUNLddm0tDQmTpzIjh078PDw4P3336dfv34cOHCAcePGkZGRgdlsZsmSJdSrV4+7776bqKgosrOz+c9//sOIESPK9bRBBwJNqxQCfEx6aMjBRowYwZNPPpkbCCIiIli9ejWTJ0/G39+f2NhYunfvzpAhQ0pVQP6TTz5BRNi3bx+HDx/mhhtu4OjRo8yaNYsnnniCUaNGkZGRQXZ2NitWrKBevXr88ssvACQk2KaEqQ4EmlYJBHh7cjwm2dHNcB7FfHK3l44dO3LhwgXOnTtHTEwMgYGB1KlTh6eeeooNGzbg5ubG2bNnOX/+PHXq1LH6vBs3bmTSpEkAhIeH06hRI44ePcp1113H66+/TlRUFHfeeSfNmjWjbdu2/Otf/2Lq1Knceuut9OrVyybPTc8RaFolEOhr0stHncDw4cNZvHgxixYtYsSIESxcuJCYmBh27tzJnj17CAkJIS0tzSbXuvfee1m2bBne3t7cfPPNrFu3jubNm7Nr1y7atm3Liy++yKuvvmqTa+kegaZVAjW8PbmUmoFSqlTDDpptjRgxgvHjxxMbG8vvv/9OREQEtWvXxmQysX79ek6fPl3qc/bq1YuFCxfSv39/jh49yt9//02LFi04ceIETZo0YfLkyfz9999ERkYSHh5OzZo1ue+++wgICODLL7+0yfPSgUDTKoFAHxOZ2YrUjGx8q+k/W0dp3bo1SUlJ1K9fn7p16zJq1Chuu+022rZtS5cuXQgPDy/1OR999FEmTpxI27Zt8fDwYO7cuVSrVo2IiAjmz5+PyWSiTp06PP/882zfvp1nnnkGNzc3TCYTn332mU2elyilbHKiitKlSxe1Y8cORzdDc1EislMp1cUR1y7utR2x/Qz/XhLJxqn9CA30qeCWOYdDhw7RsmVLRzejUijs/6q417aeI9C0SqCGTjyn2ZHuY2paJaBrElRO+/btY/To0VfdVq1aNbZu3eqgFhVOBwJNqwR0TQJDZZssb9u2LXv27KnQa5ZluF8PDWlaJZBbk6AKLyH18vIiLi6uTG90VYVSiri4OLy8vEr1ON0j0LRKILcmQUrV7RGEhoYSFRVFTEyMo5vi1Ly8vAgNDS3VY3Qg0LRKwNPDDV9P9yrdIzCZTDRu3NjRzXBJdh0aEpGbROSIiBwTkWcLub+RiKwVkUgR+U1EShfGNK0KCdAZSDU7sVsgEBF34BNgMNAKGCkirfId9i4wTynVDngVeNNe7dG0yi7Ax6RXDWl2Yc8eQTfgmFLqhFIqA/gOyJ+ftRWwzvL9+kLu1zTNItDHU9ct1uzCnoGgPnAmz89Rltvy2gvcafn+DsBPRGrlP5GITBCRHSKyQ08UaY4mIl+JyAUR2V/E/aMsw537RGSziLS3xXVr6B6BZieOXj76NNBHRHYDfYCzQHb+g5RSs5VSXZRSXYKDgyu6jZqW31zgpmLuPwn0UUq1BV4DZtviooE+OgOpZh/2XDV0FmiQ5+dQy225lFLnsPQIRKQ6MEwpdcmObdK0clNKbRCRsGLu35znxz8xXvvlFmDJQGo2K9zcKs+mKs352bNHsB1oJiKNRcQTuAdYlvcAEQkSkZw2PAd8Zcf2aJojPAisLOrO0gx7BviYMCtISs+ydRu1Ks5ugUAplQU8DqwGDgERSqkDIvKqiAyxHNYXOCIiR4EQ4HV7tUfTKpqI9MMIBFOLOqY0w55X8g3pCWPNtuy6oUwptQJYke+2l/J8vxhYbM82aFp+KelZds/pLyLtgC+BwUqpOFuc80q+oUwaFVhSoWll5+jJYk2rMLv/jmfyt7vp/uZau36qFpGGwA/AaKXUUVudN0D3CDQ70SkmNJeWmW1mxb5o5mw6xZ4zl/Cr5sHdXRuQbS574jIR+RZjWDNIRKKAlwETgFJqFvASUAv41JIpM6tcxW6yM0GpK4nn9BJSzcZ0INBcUkxSOhE7zjB/y2n+SUwjrJYP025rxV1dGlC9nMNCSqmRJdz/EPBQuS6SI+YofNIV7vqKwMa3AbpHoNmeDgSay8jIMrPu8HkW74xi/ZEYss2KXs2CeOPONvRtXrtyLrn0sUwGpMTi72X8ucbrHoFmYzoQaE4lI8vM3qhLRCekcT4hjX8SLV8JaaSkZ1E/wJsGNX0IDTT+bRDoQ2a2maW7z/LTnrPEp2ZS268a43s14a7OoTStXd3RT6l8vANB3CAlBg93N/y9PEjQm8o0G9OBQHMau/6OZ+riSP66kJx7m7fJnbo1vAjx9yIgwJuzly6z9eRFkvOtpff0cOOGViHc1TmUnk2D8HB3kXUQbm5GryAlFoBAX52BVLM9HQg0h0tJz+Ld/x1h7uZT1PH34r/3dKBlXX9C/L3w9/IoUJpQKcWl1EzOxKdy5uJl0rOyGRAeklvg3eX4BkOKsdkswFvnG9JsTwcCzaH++CuG537YR1T8ZUZ3b8S/b2qBn1fxb+giQqCvJ4G+nrQLDaiYhjqSb1BujyBAZyDV7EAHAs2u0jKz2XPmEpnZZrLNCrNSZJsh26xYc8iY2G0S5EvEw9fRrXFNRzfXOfkEwT+RgLGp7GRsioMbpLkaHQg0uzl76TIPfb2DQ9GJhd7v7iY82vcaJg9ohpfJvYJbV4nkGRrSNQk0e9CBQLOLnafjeXj+TtIzs/lgRHtCA31wE8HdTfBwE9xECKruSW1/L0c31fn5BkNaAmRlUMPbRGJaFlnZZteZENccTgcCzeZ+3H2Wfy+JpG4NL76bcC1Na/s5ukmVm69lL0FqLIGWCfHEtCxq+no6sFGaK9GBQLMZs1nx3q9H+GT9cbo3qclnozoTqN+sys/XkpU0JZZAS1CIT83QgUCzGR0INJtISc9iSsQeVh84z8huDXhlSBs8PfTQhU3kBoIYanjXAXS+Ic22dCDQym3jX7E8tzSSs/GXeenWVozrEVZg7b9WDj5Bxr8psQTW1BlINdvTH9m0MktIzeSZ7/dy3/9txcPNje8mXMcDPRtbFwQO/gRzb4XMy/ZvaGXnawkEqbFX1STQNFvRPQKtTFbui+Y/Px0gPjWDiX2v4YnSLAE1Z8OaaXDxBGz9HHo+ac+mVn5eNcDNBCkxuiaBZhc6EGjFysgyk3A5k0upGVy6nMml1EwW7zzD6gPnaV3Pn7njutKmfo3SnfTQz0YQ8KsLG9+HzmOM5Gpa4URy9xL4VfPATfQcgWZbOhBoBcSnZPDGikOs3P9PgeRuANU83Jh6UzjjezUu/Vp2pWDTf6FmExg+Fz7vAxs/hEGv2KTtLsvXSDzn5iZGmonLukeg2Y4OBFoupRTL9p7j1Z8P0jVtCxE1t5Aa0IKUWm3IrN2OakFhBPhUo36gd9mXLp76A87tgls/gLrtod3dsHUWdJsANerb9gm5Et/gPPmGTHqOQLMpHQg0AM5cTOXFH/fz+9EYetYXPk75Co9MM5zdBmcsvQKvGsabd4Nrof1IqHVN6S+08UPjTa39vcbP/Z6H/T/A72/BkI/K/0SUgtijcHoTpMRBlweubMiqzHyDIe44YGQgTdCBQLMhHQiquMxsM3M3neL9X4/iJjDttlbcH/MubpFJ8PAGqNUUzh+A6L1Xvv54HzbMgLBe0GkMtLwNTFakivhnHxxfC/3/c+X4wDDo+hBs+xyumwTBzUv3BJSCCwfh1Ebj6/RmSI29cv+Wj6Dfi0ZAcK/EL3efKxlIA308+ScxzcEN0lxJJf7L0MoiNjmdXafj2fX3JXb9HU9k1CXSMs0MbFmbV4e2oV7Cbvh1AfR4AkJaGw+q38n4ypEYDXsWwq558MND4BUA7e8x3myDWxR98U3/Bc/q0PXBq2/v/TTsXgDrXoURC6x/MkrBqudg62fGzzUaQNOBENYDGvUwir6vmgorn4Gdc2Hw29C4l/Xndya+QZCZAhmp1PAxcfifJEe3SHMhOhBUEb9ERvP2qsP8fTEVAJO70KpeDUZ2a0if5sH0aR6MZGfCwqegRkPoM7Xok/nXNd68e06BUxtg59ew4yvY/iUM/cQICvnFnzaGgLpPLLhCyDcIrp8Ev70BZ7ZDg67WPanNM40g0OUB6PEkBDYqeMzoH41VSqtfgK9vhdZ3wA3ToUaodddwFjm7i1NjdQZSzeZ0IKgCTsam8PT3ewkL8uX5m8Pp1DCQNvVrFFz3v+VjiDkMIxeBp2/JJ3ZzgyZ9ja/kGFg8DpY+DIlnjSCRd2PZlk+M2rvdHy38XNc9Btu/MPYXjF1+9WMLE/k9/PoStL4Tbn7PaEthRKDVEGg2yOiRbPwAjq6GCb+XfhjKkXI2laXEEODtT0pGNhlZZp3GQ7OJYgOBiHgBtwK9gHrAZWA/8ItS6oD9m6eVV1a2macW7cHTw405Y7tSp0YRY/nxp+D3dyD8VmhxU+kvVD0Y7lsCPz0Ga1+FhCgYPMMYl0+JM4aR2t1d9MqgatWNXsiKp+HYGuONuygnfocfJ0KjnnDHrKKDQF4mb+j7LHS4F/Z8A0HNSv8cHSk331AcAZagcOlyBrX9dBpvrfyK/AsSkVeATcB1wFbgcyACyALeEpFfRaRdhbRSK7PPfjvOnjOXeO32NkUHAaVgxTPGJ/bBb5f9Yh7V4I7Z0PMpY6ho0X2QkWJ80s+6DNdPLv7xncZAYGNj3P/E72A2Fzzmn/3GeWs1hXsWGtcsjYCGRkCobLmQ8vQIclJR601lmq0U1yPYppR6uYj73heR2kBDO7RJs5F9UQn8d+1f3Na+HkPa1yv6wEM/w1//gxteL//YuZsbDJwG/vVh5b/h69vg4kloPhhqhxf/WA9PuOVdWPwgzBtiBIVOo6HDKPCrY/QyFg43hq3uWwzeAeVra2WSJwNpQEhOmgkdCDTbKLJHoJT6BUBEhue/T0SGK6UuKKV22LNxWtmlZWbzVMQegqpX47WhrYs+MD0JVk6FkLZw7SO2a0C38cYKoPMH4fJFYxWSNZoOhH8dhju/NILS2lfh/Vbw7UhYMAwykmHUYodO9orIVyJyQUT2F3G/iMhMETkmIpEi0qmw40rF0xc8vC35hnISz+kJY802rJlpes7K2zQn8s6qIxy7kMyM4e1yE5UVkJECP0yApGi47UPbr7MPvwUeWAm3vAeNrrP+cSZvaDfcmDSetMtYURS1w9hQNWIB1Glj23aW3lyguImUwUAzy9cE4DObXNU3GFLjcgOB3lSm2UqRf/kiMhi4GagvIjPz3OWPMU+gOZBSioTLmYW+yW86FstXm04y5rpG9GoWXPgJEqLg23uMzWKD34bQLvZpaL2OxldZ1brGyEPU/0W4HA/Va9uubWWklNogImHFHDIUmKeUUsCfIhIgInWVUtHlurBvkGWOwPid6x6BZivFfQQ8B+wAhgA789yeBDxlz0ZVSYnnjDe6kGKGcSxSM7J4YO52/jxxkVq+noTX9aNFiD/hdfxoHOzL09/vpUmwL88Obln4CaJ2GEMtmZfh3ojiV+g4C3eTUwQBK9UHzuT5OcpyW4FAICITMHoNNGxYwpSbbxAkn8fH0x2Tu+h8Q5rNFBkIlFJ7gb0i8o1SSr/i7MlsNj6dJ5yFp4+CW9F5/dMysxk/bwfbTl7k4d5NiE/N4Mg/SXy77W8uZ2YD4O4m/DDxerw9CzlP5PfGEk+/OjBmGdQuIlhoFUIpNRuYDdClSxdV7MG+wXD+ICJGBtIEnYFUsxFrBoXDRORNoBWQu/5QKdXEbq2qag78YOTwATi3B0I7F3pYRpaZiQt2sulYHO8Ob89dna9MmGabFWcupnL4nyT8vT1o3yDg6gebzbD+dfjjXSP9wt3zXSMZm3M6CzTI83Oo5bbysQwNoRSBPibiU/TnM802rJksnoMx2ZUF9APmAVYlhBGRm0TkiGX1xLOF3N9QRNaLyG7L6oqbS9N4l5CVAeumG+viEWMzVSEys81M+nYX64/E8MYdba8KAmD0AsKCfLmpTR2uvybo6gdH7TDSK/zxLnQcbaRd0EHAnpYB91tWD3UHEso9PwBG4rnsdEhPIsBb1yTQbMeaQOCtlFoLiFLqtFJqGnBLSQ8SEXfgE4wVFK2AkSLSKt9hLwIRSqmOwD3Ap6VpvEvY9TXEn4Qb3zQSux37tcAh2WbFU4v2sPrAeV6+rRX3Xmvl9o3Yv2DRaPhygJGa+baZRqpnjzLWEnAhW7Zs4bHHHqNdu3YEBwfTsGFDbr75ZoBgESm25JqIfAtsAVqISJSIPCgij4hIzvrbFcAJ4BjwBVBEXo1SyruXwMek9xFoNmPN0FC6iLgBf4nI4xhd3OpWPK4bcEwpdQJARL7DWE1xMM8xCmMVEkANjAnqqiM9GX5/2xiqaTYIzu6EDe9A6kXwqQmA2ax4ZvFelkdG89zgcMb1aFzyeZP+gd/eMtI6mLyh7/NGLp9q1vzaXN/gwYOpV68eQ4cO5YUXXqB27dqkpaVx9OhRVq5c6Qb8JCLvK6WWFfZ4pdTI4s5vWS30mM0bnpt4Lo4An2pERulAoNmGNYHgCcAHmAy8hjE8NMaKxxW2cuLafMdMA/4nIpMAX2BgYScq1cqKyuTPT40x33u+NVIeNB1oFGg5vg7a3kV8SgYv/LiPFfv+4amBzXm4TwmFYLIyLInV3ofsDCPdc+9/G3mAtFzz588nKOjq4bPq1avTqVMngPNKqb4iElTogx3pqjQTTfTyUc1mittH4AX4KaW2W25KBsZZUksk2uj6I4G5Sqn3ROQ6YL6ItFFKXZVkplQrKyqLlFjYNNNI8paTdrl+JyNF87G1rDP1YuqSfcSnZPDs4HAe7l3C3PyZbbBsMsQcglZDjTQPNfV8fmFiY2NzA0F6ejrVql2Vr8gXQCkVW8hDHStvBlKfcNKzzFzOyC58dZimlUJxcwQzMbKO5tcD+MCKc1uzcuJBjER2KKW2YKxKcr5PYvaw4V2j0MiAl67c5uZOZlhfkg6s4sG526jl68lPj/fgkT7XIEUlSUtLhF+ehv+7AdITjRTSd8/TQaAY9957b+73111XYMez83Y5fXICQWzu7mI9YazZQnGBoLNS6of8NyqllgK9rTj3dqCZiDQWEU+MyeD8Y65/AwMARKQlRiCIsabhlVr8adjxf9Dxvqsqem0+Hstbxxrgl3WRl7ua+enxHrSuV8y85ZGV8Gl3oyBMtwnw2NaypZCuYowh/ILfWzhvWlKTF1Tzh5TY3AykegmpZgvFzRH4FHNfiauNlFJZlsnl1YA78JVS6oCIvArssEzE/Qv4QkSewpg4HqsK+ct0OevfMFI+972SsunT347xzqojdKrZGTJhbO1j4FFMl3/DDGPZae1WMPxr66t6aVf1rgrpaTn368+nFqTEUMPbkoFU9wg0GyguEFwQkW5KqW15bxSRrlj5qV0ptQJjKV3e217K8/1BjKGmquOf/RC5yMjG6W+khv5pz1neWXWEIe3r8fawdjCnPfy1Bnr9q/BzJF8wCsiH3wp3zdHLQUspKiqKyZMno5TK/R5yewfO/Z/pG2xMFvvqmgSa7RQXCJ4BIkRkLldyDXUB7scY5tFKKysdlj8JXv7Q80kAdv8dzzOLI+nWuCbvDm9vlB5sOhA2fghpCeBVyNDQxg+Ncw16VQeBMpgxY0bu9126FEi2F1WhjSkt32C4dJq6NbxxEzj8TxI3t63r6FZplVxxuYa2iUg3jPXQYy03HwCuVUpdqIC2uRal4JcpELUdhs8F70DOXbrMhPk7CfGvxqz7Ol+pP9t0IPzxHpz4zVgBlFditDG/0H6kkZlTK7XRo0fjVkR5y7Fjx16q2NaUkm8tOLuDGt4mOjYMZP3hC0wZVIlqL2tOqdixfkvxmZeVUsMsXy/pIFBGWz+H3Qug9zPQ+g5SM7IYP28HaRnZfDWmKzV983yyD+0G1WoUnm5i4/tgzoI+z1Rc211Mly5d2Lp1a4Hbv/zySzB2wTsv32Bj6bHZTP/w2uw7m8CFxDRHt0qr5IqrWfyziNwmIqZC7msiIq+KyAP2bZ6LOPEbrH4eWtwCfZ/HbFZMWbSXQ9GJzLy3I81C/K4+3t0DmvQx5gnyzp1fOgM75xqrjQLDKvAJuJaZM2cyYcIExo8fz8WLF9m9ezfXXXcdq1evBjjs6PYVyzcYVDakXaJfCyMt929HXH+hnWZfxfUIxmPsIzgsIttFZIWIrBORExiF7Hcqpb6qkFZWZhdPQMQYCGoOd34Obm68/+tRVh34hxduaZX7x1xAs0GQdA4uHLpy2x/vGf/2etr+7XZhPXv2ZOfOnYSEhHDNNdcwZMgQXnnlFb7//nsA5559zc03FEvLun7UreHFusO6k66VT3E1i/9RSv1bKXUNMBwjvcQUoI1SapBS6qeKamSllZ4E31o2L438Bqr5sXR3FB+vP8Y9XRvwQI+woh97zQDj35wkdPGnYPd86DQGAhoU+TDNOosXL+bbb79l4sSJ1KlTh0WLFnHx4kVHN6tkPpassSkxiAh9W9Rm47FYMrLMxT9O04phTfZRlFKnlFJblFJ7lFKp9m6USzCb4YeHjayfw+dCzSasOXieZ76P5LomtXh1aJuidwsD1KgPtVtfmSf4fQaIO/SaUiHNd2UDBw5kwYIFrFmzhjfeeIOtW7fSoUOHnBVEzr2zPTfxnJEBo394bZLTs9h+qhIEMc1pWRUINCA6Ek5vsf74TR/CkV/gxtfhmn5sOhbLo9/sonU9f2bfn2eFUHGaDjCuGb0X9n5rJJGz7D3Qyu6xxx5j+fLlNG5sZHJ1c3Nj0qRJbN68GcCv2Ac7Wp5U1AA9mtbC08NNDw9p5aIDgbVWPAPfjTTW75ckOxP+/AyaDoJrH2Hn6XjGz9tB41q+zB3XDT+vAvPvhWs2CMyZsOg+cPeEHk+W6ylohp49e3Lw4MECt1uGhs4UuMOZWNKTk2L0CHw8PejepBbrdSDQyqHEQGBZOVS1A0ZWBpzbbRSXP7Ki5OOPrYGUC9DlAQ5EJzJuzjZq+1Vj/oPdCPQtxQawBt3Bszpc+hu6jQe/kLI/By3XpEmTiI0tmFw0Li4Ork6U6HzcTUaG2pQrK4X6twjmRGwKp2JTHNgwrTKz5g1+BEZRmndEJNzeDXJK5/cZJQIBds0v+fg9C8E3mOMB13H//23Dt5oHCx66ltr+XiU/Ni8PT2jSF0y+RkoKzSaOHTtG794F8yb26tULis+x5Rxy9hJY9A83PiDo4SGtrKxJHncf0BE4DswVkS0iMkFEnHss1Zaidhj/drjPKBqTUEwWgpQ4OLKKpOZ3ct+cXYjAwoeuJTSwjO8vN78LD66+koteK7ekpKTi7nbe7KM5fIKuCgQNa/nQtHZ11h/RgUArG2tXDSUCi4HvgLrAHcAuS2Ux1xe1HfzqQe+nAQV7vin62H0RYM7kycOtSEnPYt4D19IkuBwlIv3rQp22ZX+8VkDTpk1ZsaLgEN/KlSsBrJgEcjDfoKuGhsBYPbT1xEVS0rMc1CitMiuxVKWIDAHGAU2BeUA3pdQFEfHBqD/8kX2b6ASitkNoF6jZGBr3NlJF9HoaCslXY969kJMezfgjMYQFD3WlVT3/Qk6oOdIHH3zArbfeSkREBJ07dwZgx44dbNmyBYwaGc7NNxhOb7rqpn4tajN7wwk2HovlxtZ1HNQwrbKypkcwDPhAKdVWKTUjJ9eQZT/Bg3ZtnTNIjjE2czXoZvzccTRcOg2n/ihwqPncXtzO72Pu5R68d3d7ujWuWbFt1azSvHlz9u3bR58+fTh16hSnTp2iT58+REZGQqXoEQRD6kXIvvLpv0tYIH5eHqw7pIeHtNKzpnj9NCA65wcR8QZCLJvM1tqrYU4jylKyOdRS+KXlbUZCuN0LjHxAeexa9jFtlQeN+47htvZ6vb+zUkpRrVo1xo0bV+QxIiJOWyTJNwhQcPkiVDdSlJjc3ejdLJj1Ry6glCp+s6Km5WNNj+B7IO/+9WzLbVVD1HZw84C67Y2fTd7Q9i44tAwuX8o97Lstx2kSvYIjAb0YN7CjY9qqWaVfv3589NFH/P331aNAGRkZAH4i8jUwxhFts0qeIvZ59QuvzYWkdA6cS3RAo7TKzJpA4KGUyq2HZ/m+6lRDidpuTNaavK/c1mk0ZKXB/sUA/HbkAr8vn09NSab1zRP1pzEnt2rVKtzd3Rk5ciT16tWjVatWNG7cmGbNmgHUBD5USs11bCuLkSfxXF59WwQjopeRaqVnTSCIsUwYAyAiQ4GCu3Gc1Ynf4Oj/yvZYczac3XVlWChH3Q4Q0gZ2L+DguUQeW7iLsT4bMVevg3uzgeVtsWZnXl5ePProo2zatInTp0+zdu1adu/ezenTpwFOK6V2l3QOEblJRI6IyDERebaQ+xuKyHoR2S0ikSJys82egE/hPYKg6tVoFxqgA4FWatYEgkeA50XkbxE5A0wFHrZvs2zot7cg4n4jl39pXTgEmSkFA4GIMWl8bjdvzV1MmFcy3bJ24db+HnArpuC85nRMJhN169YlICDA6seIiDvwCTAYo5DNSBHJX9DmRSBCKdURo7Trp7ZpMUX2CAD6t6jN3qhLxCU7/5y35jys2VB2XCnVHeMF31Ipdb1S6pj9m2Yjly9B1mVY/VzpH5s7UVygri3mNsPJxET/y6uZ3eE4orKNgjFaVdANOKaUOmEZKv0OyFdTFAXkrB2uAZyz2dW9A0HccjOQ5jWgZW2UgiW7nLv0suZcrFk1hIjcArQGvHLGv5VSr9qxXbaTngge3nDoZyMHUNNSDN1E7TDyvwc2LnDX7J2XCM3uzD1eW/A6ccIoLxnUzIYN15xYfa5OThcFXJvvmGnA/yybLn2BQl94IjIBmADQsGFD667u5ma8LlMKViZrXc+fAeG1efd/R+ndPJjwOnofi1Yya5LOzcLINzQJY/v9cKCRndtlO2mJ0GEk1GwCK/5tXfbQHFHbjDf4fJO/205eZMbqI5xqcAdemQlGzYEO99q44Zq9paSkYDYbC+KOHj3KsmXLwHYpJkYCc5VSocDNwPzCkjcqpWYrpboopboEBwdbf/Z8+YZyiAjv9hKCq5l54ts9pGVml/0ZaFWGNXME1yul7gfilVKvANcBze3bLBsxZ0NGkvFHM3gGXDwOWz627rGX4403+HzDQrHJ6Uz6dhcNa/ow5r6x4B9q9Dja3Gn79mt21bt3b9LS0jh79iw33HAD8+fPBwiz4qFnuTpLaajltrweBCIAlFJbAC9sWfTGN6jQQMCueQTOH8Dq6q+QdeEwb6107hLMmnOwJhCkWf5NFZF6GDVd69qvSTaUbkkuVs0fmg2E8Fthw7vWTRyf3Wn8m2eiONusePK7PVxKzeSTezvh5+MFt75vfHnVsMMT0OxJKYWPjw8//PADjz76aE7NYu+SHgdsB5qJSGMR8cSYDF6W75i/gQEAItISIxDYrsq8b3DBoaFTm2D5FKjfheqZsaz0fomEP+frWgVaiawJBD+LSAAwA9gFnAKKybrmRNItG2u8LOOkN70JSsHq50t+bNQOQKB+p9ybZq79i43HYnl1aOsrOYSa36iHhSoppRRbtmxh4cKF3HLLLaV5XBbwOLAaOISxOuiAiLyaZ6n1v4DxIrIX+BYYa9OdyvkykBJ/CiJGQ2AjuG8xPLIRj9COfOD5GUmLJhBTGeoxaw5TbCCwjGmuVUpdUkotwZgbCFdKvVQhrSuvNEsgqGZ50w5oCL3/ZewKPlZCdoyo7VC7FVQzsm1v/CuWmev+YlinUO7u4ty1SzTrfPjhh7z55pvccccdtG7dmhMnTgAUm6M6h1JqhVKquVLqGqXU65bbXlJKLbN8f1Ap1UMp1V4p1UEpVcbNLEXwDYb0BGPOKz0Jvh0J5iwYuchYVeRfD7cxPxPX+UluNf9Gxmd9UOcP2LQJmusoNhAopcwY66Vzfk5XSiXYvVW2kr9HAHD9ZGPieGUxE8dms9EjsMwPmM2Kl5ftp0mQL6/d3lrvHHYRffr0YdmyZUydOhWz2UxQUBA4e6nKHDlpJpIvwJLxEHMEhs+FoKZXjnH3oNZtr7Cm6+d4ZiSQ/Xk/OLPdIc3VnJs1Q0NrRWSYVMZ3v9weQZ7xe49qxsRx3DHYPLPwx108DmmXcucH/jgWy/GYFB7r1xQfT6tW3GqVwL333ktiYiIpKSm0adOGVq1aAVSOeqA5geCXKXB0Jdz0FlzTv9BDB91yN280/JLL2W5c2vxVBTZSqyysCQQPYySZSxeRRBFJEpHKkdWqsB4BGBPHre+A9W/AwZ8KPi5fxtE5m04SVL0at7SrHHPkmnUOHjyIv78/P/74I4MHD+bkyZMAtRzdLqvk7C7+63/QeZxR07oIIsILI/qwhbbGXhonTaqqOY41O4v9lFJuSilPpZS/5efKsUslzTKKVa2Q5g791HijX/KQUX4yr6jtRi8iqDnHLiTz25EYRndvRDUPnT7ClWRmZpKZmcmPP/7IkCFDMJlMjm6S9apbOi5hveDmGQX2uuQXVL0al+r3JSDzApfP7q+ABmqViTUbynoX9lURjSu3onoEAJ4+cO8iCGoO3913pS4xGOOooZ3BzY2vN5/C092Ne6+1ctenVmk8/PDDhIWFkZKSQu/evXOSzlWOHVg1G8Pwr2HEAnC3LoA1u87IgnFs81J7tkyrhKwZGnomz9d/gJ8xts87v7REcDOBh1fh93sHwn0/GMU9FgyD8wchPRkuHIDQriRczmTJrihua1+PYL9qFdt2ze4mT57M2bNnWbFiBSJCo0aNAI44ul1Wa307eAdYfXiHNq05Lo1wO/ar3ZqkVU4lznwqpW7L+7OINAA+tFeDbCo90egNFNdt9guB+3+E/7sR5t8BA/4DygyhXYnYfobUjGzG9QirqBZrFSghIYFXXnmFDRs2AMYqIsBlx/9EhEv1+9LuzAKios8TWrdyzItr9mdNjyC/KKClrRtiF2mJhc8P5BcYBqOXGsVmfnocgOy6nfh6yym6hdWkTX29a9gVPfDAA/j5+REREUFERAT+/v5gXYqJSqth96GYJJtdv//o6KZoTsSaOYKPRGSm5etj4A+MHcYlsqJ4xwcissfydVRELpX6GRQnp0dgjZBWMOp7oxJZrWb8eiqTqPjLujfgwo4fP84rr7xCkyZNaNKkCS+//DKAS48BBrfsTar4oP76FbNZrx7SDNYsis8zi0oW8K1SalNJD8pTvGMQRi9iu4gsU0odzDlGKfVUnuMnAbYt9mttjyBHg24wbiWYs/nql5PUD/BmUCvdfXZV3t7ebNy4kZ49ewKwadMmuLo+t+txN3GpTg+6ntvJn8djub5ZKTKeai7LmkCwGEhTSmWD8QYvIj5KqdQSHpdbvMPyuJziHQeLOH4k8LJ1zbZSeqKxi7g06nXgwLkEtp3cyPM3h+PhXpbRM60ymDVrFvfffz8JCcYy48DAQIDTDm1UBQjueAum6F9ZsHkD1zcbVvhBly8ZCyh6PwMtbqrQ9mkVz6qdxVydkdEbWGPF4wor3lG/sANFpBHQGFhXxP0TRGSHiOyIiSlFAse0xDJlBZ2z6RTeJndGdNFLRl1Z+/bt2bt3L5GRkURGRrJ7924AP0e3y95M4TcCIMfXkJiWWfhBm/4LZ3cYlf2ysyqwdZojWBMIvJRSyTk/WL73sXE77gEW5/Q68itz8Y70Ug4NYdQbWLbnHMM616eGTyXaYKSVmb+/f85EMVSWFBPl4V+Py4Hh9FR7+CUyuuD9iefgz8+MPTYXT8DeypFsWCs7awJBiojk5mIWkc7AZSseZ03xjhz3YKTqtR2z2cjKaO1kscXCP/8mI9vM2OsLlqfUqoTKl1OrDLxa3URX9yMs31bIton1b4DKNhZP1O8Mv79Tusp+WqVjTSB4EvheRP4QkY3AIoxc7CWxpngHIhIOBAJbrG61NTKSAFWqHsGJmGRm/X6cgS1DaFq7uk2bo1UaVWIpjTQbhAfZVD+3iWMXkq/cceEw7FkIXR8yllX3fxESzsDOrx3WVs3+rNlQtt3yZt3CctMRpVQRA4tXPS5LRHKKd7gDX+UU7wB25ORtxwgQ39m0aAdcyTxqZY8gM9vMU4v24OnhxvTb29i0KZpz8fPzKzSVuOUl6FnhDXKEBtdi9vSjf/YeFu+M4tnB4cbta18Bz+rQ62nj5yb9oFFP+ONd6HifkZpFcznW7CN4DPBVSu1XSu0HqovIo9acvKTiHZafpymlCuwxKLecPENW9gg+XneMvVEJvHFHW+rUKCIlheYSkpKSSExMLPCVlJQEsNPR7asQ7ibcrunLIM99/LDzDFnZZji9BY6sgB5PgK8lCasI9H8Bks/D9i8d22bNbqwZGhqvlLqU84NSKh4oOuets8jJPGpFj2D33/F8vP4Yd3asr1NNa1VHsxuomR1LYMox1h06D7++BNXrQPeJVx/X6Hq4ZgBs/OBKT1tzKdYEAve8RWksG8Wcv/tcWFGaQqRmZPHUoj3U8fdi2tDWFdAwTXMSTQcCcKffIdb9OAeitkG/58DTt+Cx/V+Eyxdh66wKbqRWEawJBKuARSIyQEQGYKzuWWXfZtlAcSmo85j+yyFOX0zlvbvb4++ll4tqVYh/PQhpw6jAgzyUOZ/zng2hw32FH1u/E4TfCps/gtSLFdtOze6sCQRTMTZ6TbR8rcVISe3ciitKY7H20Hm+2fo3E3o1oXuTylGYStNsqulAqp/fTlM5x3+Sh7HqUDEbNvs9byzJ3vxRxbVPqxDWVCgzK6VmKaXuUkrdhZEiwvlfCSX0COKS05m6JJLwOn5MuaF5BTZM05xIs0EAmEO7EV1nAM/9sI8LSWmFHxvSGtoMM4aHki9UYCM1e7MqkY6IdBSRd0TkFPAqcNiurbKFEorSvLXyMImXs/jwng66BKVWdTXoDl3H43bbh3xwT0dSM7J5dsk+ilzN3fc5Y3PZL1N07WMXUmQgEJHmIvKyiBzG6AGcAUQp1U8pVTl6BMUUpdlyIo5BrUMIr1M5yi9rml24e8At70JIa5rWrs6zg8NZd/gC3247U/jxQU1h4DQ49LMeInIhxfUIDgP9gVuVUj0tb/6Vo54rFJuCOjHNqDXQqq4OApqW15jrwujRtBbTfznI6biUwg+6fhK0HAJrpsHJP8p2IbMZfn4STvxe1qZqNlRcILgTiAbWi8gXlhVDlScPSzFFaQ5HJwHoQKCVSUkFlyzH3C0iB0XkgIhUmqxtbm7CjLva4+4mTInYS3ZhxWtE4PZPjRTvi8cZSepK69QfsHMO/DjRmIDWHKrIQKCU+lEpdQ8QDqzHyDlUW0Q+E5EbKqh9ZVdMj+BQtDGR3FIHAq2U8hRcGgy0AkaKSKt8xzQDngN6KKVaY/ztVBr1Arx5bWgbdp6OZ0rEHlIzCklDXc0PRiyAjFT4fixkZZTuIrvmgcnHCCK/vWWTdmtlZ82qoRSl1DeWIvahwG6MJaXOLb3oWgSHohMJ9DER4u/SVQk1+8gtuKSUygByCi7lNR74xLILH6VUpVtiM7RDPZ6+oTnL9p7jjk82cyImueBBtcNh6MdwZiv870XrT556EQ4tM3IXdR5rpLyOjrRZ27XSK1X5LaVUvKU2wAB7NchmSugRtKzrX2jiMU0rgTUFl5oDzUVkk4j8KSJFlvgqc9ElOxMRHu/fjK/HdeNCUhpDPt7Eqv2F1C5ocyd0fwy2fQ6R31t38sgIyM6ATmNg4MvgHQjLnzLmDTSHcN06jEXMEWSbFUfOJ+lhIc2ePIBmQF+MEqxfiEhAYQeWuehSBendPJjlk3txTe3qPLJgF2+uOGQkqMtr0CvQ8Dr4eTJcOFT8CZWCXV9DvU5Qp40RBG58w6iGtmuu3Z6HVjzXDAQ5RWkK6RGcjE0hLdOsA4FWVtYUXIoClimlMpVSJ4GjGIGhUqof4E3Ew925r3tDPt9wgnu/3Epccp5CNe4mGD4XTN7w8xPF7y84uxMuHIRO91+5rd3dENbLWIWU7Dy9oqrENQNBTlGaQnoEVyaKXb40rWYf1hRc+hGjN4CIBGEMFZ2owDbaXDUPd6bf3pYPRrRn75lLPLJgJ5l5ewZ+dYz9BWe2QuSiok+062tjkrjNsCu3icCtH0Dm5dLNNWg245qBIDfzaOGBwMNNdAUyrUyUUlkYFfpWA4eAiJyCSyIyxHLYaiBORA5irLh7RikV55gW29YdHUN55652bD8Vz9sr8yUY6HCfMeTz60uFp6tOT4J9S6D1nQU/pAU1gx5PQuR3cHKD3dqvFc41A0FunqGCq4YORSfStHZ1nVZCK7OSCi4pwxSlVCulVFul1HeObbFtDe1QnzHXNeLLjSf5JTLPBLKbG9z8rlHEZsM7BR94YClkplw9LJRXrykQ2BiWT9E1kitYiaUqK6ViylQeik7iumt0plFNK48XbmlF5NkE/r14Ly3q+F3pYYd2NpaF/vkZdLwfgvMkdNw1D4LDoUG3wk9q8jbSXSwYBm81Au8Ao1fvVcP48g4wVho17mXvp1fluHaPIF9RmviUDP5JTNPzA5pWTp4ebnw6qhNeJnceWbCTlPQ8m84GTAOTL6yaemXi+PxBiNpu9AaKW7bddCAM+z/o+qDxfe1wo05yahyc+A3m3w57Ks1G7UqjSvUI9I5iTbOdujW8+WhkR+77v61MXRLJRyM7GntzqgcbtQtWTYXDv0DLW2H3fCMbcLt7Sj5x27uMr/zSEiDifiMtRUIU9H6m+KCiWc1FewSFF6U5qAOBptnU9U2DePrGFiyPjGbu5lNX7uj6ENRuBaufg8uXYO+3RkDwLcewrFcNuPd7aH8vrH8dlk2C7MzyPgUNVw0ERfYIkgj2q0ZQdZ1aQtNs5ZHe1zCwZQiv/3KIjX/FGje6e8Dgd+DS37DgTrgcX/QkcWl4eBoJ7/pMNXoZ34zQSetswDUDQXrhRWlyUktommY7bm7Ce3e3p0mwL+PmbmPxzijjjsa9oPUdxiaygIbQuK9tLihiDD0N+diYN5gzuGwZULVcrhkI0goWpcnMNnPsQrKeKNY0O6jhbeL7R66nW+OaPP39Xt5dfQSzWcEN041FG13HG8tLbanTaBgVARdPwux+ELXDtuevQlwzEKQXTDh3PCaZjGyzrkGgaXZSw9vE3HHdGNGlAR+vP8bk73aT5lMX/nXIKGZjD00HwoO/gsnL6BnsXmCf69jTti/g0HKHNsE1A0Fagl4xpGkOYHJ3461hbZl6UzjLI6O594s/icvwsO/qnpBWMH49NLoefnoMVk6tPJPIWz6BFU9DxGiHBgMXDQQFewSHopPwdHejcZCvgxqlaVWDiDCx7zV8OqoTB84lcsenmzl2oZB6BrbkUxNGLTFSYm+dBfPvgBQ7ZvXIvAxfD4FN/y37Ofb/AKufh5a3Gak5Fj9Q9tKf5eSagaCQojSHohNpFlIdk7trPmVNczY3t63LdxO6k5qRxbDPNrP91EX7XtDdA256A26fBWe2wRd9IeaIfa616jk4+Tv8/o4xAlFapzbB0oeN9N13fgmjvofAMPh2JJzbY+vWlsg13xUL7RHoFUOaVtE6Ngzkh4k9qOXryagvt16dm8heOoyEB1YaZTR/nFh8WuyyOLDUqLfc4hbISDZSZ5TGhUPw3Ujjjf+eb4z5DZ+aMHqpkUZjwTCIPVaGdv0Ih1eU/nG4aiDIV5TmQlIasckZOhBomgM0rOXDkonX07Z+DR7/dhdf/lEBGbnrd4ZBrxpLVw/+aLvzxp+CZU9A/S5w99dGHYU/Z1k/J5EYDQvuMpa2j1psBIAcNerD6B8BZQxtWbskNjsL/vcf+H4MbP+yTIHP9QJBIUVpDkUbG0700lFNc4xAX08WPnQtN7Wuw/RfDvHKzwfINtv4k3p+7e+B2q1hzSuQlVH+82VnwuIHAQV3/Z9RkOe6xyAxCg7+VPLj0xJh4V2QdskyFNSo4DFBTeG+JXD5ohEMUksYTkuOMfIvbZ5p7OYe+V2ZJuZdLxAUUpQmZ8WQXjqqaY7jZXLn43s78UCPxszZdIrHv9lFWma2/S7o5m6U0Yw/aQzllNe66UZJzSEzjWEdgGY3Qq2msOXj4j+Jm83w/ViIOQx3z4O67Ys+tl5HGPktXDwBH3WGVc/DhcMFj4vaAbP7GMn8bp8Ft7xn7LwuA9cLBIUUpTkUnUjdGl4E+JTtP0nTNNtwdxNeuq0VL97SklUH/mH0/20lIdWOSz2bDoTGveH3t8s2qZvj2FrY9CF0Hmvsls7h5gbdH4Vzu+HvP4t+/JaP4PhaI+1G0wElX69xbxj7C4T1hG2fw6fXwv/dYOyTyEiBHV8Z+ybc3OHB/xnzIuXgeoEgtyjN1YFAzw9omvN4qFcTPhrZkT1nLnH351s4n5hmnwuJGHMFqXHFL/VMjjHyFn05CJY/ZYy1/73VGGZOOm+s8AluCTe+WfCx7UeCd02jV1CYs7tg7avQcgh0ecD6tjfoBiPmw5TDMOg1Y5jop8fgnSZGG8N6wYTfi+9dWMmugUBEbhKRIyJyTESeLeKYu0XkoIgcEJHyJxrP1yNIz8rmeEyKnh/QNCdza7t6zB3Xjaj4VO78dDPHY+y016BeR2g7HLZ8WvgE7MWT8NUNcOJ3Y9x//xL45V/GbW+GwsddID0Zhs8xaiPk5+lj1E84/AvEHb/6vvRkWPIQVA+B2/5bto111YOhx2R4fDuMWwntRsDAV4x5hryTzeVgt0AgIu7AJ8BgoBUwUkRa5TumGfAc0EMp1Rp4stwXzlem8q/zyWSble4RaJoT6tE0iEUPX0d6VjbDZ21h75lL9rlQ/xdBZcP6N66+/dwe+L9BRnbUMctg3AqYehqe3G9MvPZ70RheumMW1G5Z9Pm7jjeCyNZZV9++cqox1n/nF+V/0xYxdk8PmQk9nzSGhWzEnj2CbsAxpdQJpVQG8B0wNN8x44FPlFLxAEqpC+W+ar4egU4toWnOrU39Gix+5Hp8q7kz8os/2XA0xvYXCQwz3qz3LDSqpQEcXw9zbzGWcj7wvyslNEUgoAG0GAx9njF6Aq1vL/78fiFGr2P3AiOogNGz2LMAej8NYT1s/5xsyJ6BoD5wJs/PUZbb8moONBeRTSLyp4jcVNiJRGSCiOwQkR0xMSW8SHKK0njlBIIkvExuhNXSqSU0zVmFBfmy5JHraVTLlwfmbuexb3YxY/VhInacYdvJi5xPTEOVd2NY76fB0w/WTIN9i2HhcAhoZCSty1tbuay6PwqZqbBjjlGH4eenILSrUTvByTm6VKUH0AzoC4QCG0SkrVLqUt6DlFKzgdkAXbp0Kf7VkJZvaOhCEs1D/HB30yXtNM2Z1fb3YtHD3Xn5pwPs+jueVfv/uWqvgbfJnX/d0JyHejUp2wV8akKvKbDmZfhrNTTqCfcsNHbz2kKdNtCkH2ybDUdXgzIbQ0LuJtuc347sGQjOAg3y/BxquS2vKGCrUioTOCkiRzECw/YyXzVfUZqktCy9bFTTKgl/LxMfjOgAGDVEzl26zKm4VE7HpbD20AWm/3KImr6e3NkptGwXuPZhiIwwxvuHfmKkd7Cl6x6HhcMgKdoIAjUb2/b8dmLPQLAdaCYijTECwD3AvfmO+REYCcwRkSCMoaLy7T/PV5QmLTObEH9dmlLTKhuTuxuNavnSqJYvEMw9XRsyds42/r04khB/L3o0DSrDSb1h4ib7pcVuOsDYAxDUAtrdbZ9r2IHd5giUUlnA48Bq4BAQoZQ6ICKvisgQy2GrgTgROQisB55RSpUvd2y+ojSXM7PxNtludl3TNMfw9HBj1ujONK1dnUfm78xdCFJq9qyNIAL3L4Nb3rXfNezArnMESqkVwIp8t72U53sFTLF82UZOj8AiNSMbb0/HB4LMzEyioqJIS7PTxhmtVLy8vAgNDcVkKtv4rWVhw38Bd+BLpdRbRRw3DFgMdFVK6VqK5eTvZWLOuK7c8clmxs3ZztLHrqduDW9HN+tq9gw0duLoyWLby9cjSMvIxssJegRRUVH4+fkRFhaGVMIXiitRShEXF0dUVBSNG5d+DDfPHplBGPNc20VkmVLqYL7j/IAngK02aLZmUbeGN3Mf6Mrwz7Yw9qvtfD/xOvy9nH9C1pm5XoqJtKuL0lzOzMbHCXoEaWlp1KpVSwcBJyAi1KpVqzy9M2v2yAC8BrwN6G6gjYXX8WfW6M6ciE3m4Xk7ycgyO7pJlZrrBYI8PYLMbDNZZuU0cwQ6CDiPcv4uStwjIyKdgAZKqV9KaIf1e2S0q/RoGsTbw9qx5UQcI2Zv4cC5ciSVq+JcLxDkmSO4bElx6wxDQ1rVISJuwPvAv0o6Vik1WynVRSnVJTg42P6NczF3dgrlv/d04O+4VG77aCPTlh0gMa2SFK53Iq4VCMzmq3oEaRlGIHCGyWLNpZS0R8YPaAP8JiKngO7AMhHpUmEtrEKGdqjPun/1ZdS1jfh6yyn6v/s7P+4+W/6dyFWIawWCjGTyFqXJ6RE4y9BQVZGVleXoJthb7h4ZEfHE2COzLOdOpVSCUipIKRWmlAoD/gSG6FVD9lPDx8Rrt7dh2WM9qR/ozZOL9jDyiz85diHJ0U2rFFxr1VD61QnnnDUQvPLzAQ6eK+Ma6CK0qufPy7e1LvG422+/nTNnzpCWlsYTTzzBhAkTWLVqFc8//zzZ2dkEBQWxdu1akpOTmTRpEjt27EBEePnllxk2bBjVq1cnOdlIF7x48WKWL1/O3LlzGTt2LF5eXuzevZsePXpwzz338MQTT5CWloa3tzdz5syhRYsWZGdnM3XqVFatWoWbmxvjx4+ndevWzJw5kx9//BGAX3/9lU8//ZSlS5fa9P/IVpRSWSKSs0fGHfgqZ48MsEMptaz4M2j20ja0BksnXs9328/w9qrDDP7vH0zscw2P9muqh4iL4VqBIO3qhHOplqEhLz00lOurr76iZs2aXL58ma5duzJ06FDGjx/Phg0baNy4MRcvGjVSX3vtNWrUqMG+ffsAiI+PL/HcUVFRbN68GXd3dxITE/njjz/w8PBgzZo1PP/88yxZsoTZs2dz6tQp9uzZg4eHBxcvXiQwMJBHH32UmJgYgoODmTNnDg88UIoCHg5Q0h6ZfLf3rYg2aQY3N+HeaxtyY+sQpv9yiJnrjrE8Mprpd7Th+mvKsBu5CnCxQHB1jyB3jsDJPglY88ndXmbOnJn7SfvMmTPMnj2b3r17566nr1nTyJm+Zs0avvvuu9zHBQYGlnju4cOH4+5u/F8nJCQwZswY/vrrL0SEzMzM3PM+8sgjeHh4XHW90aNHs2DBAsaNG8eWLVuYN2+ejZ6xVlXVql6ND0Z04M5O9Xlh6X7u/WIrwzuH8vzNLQn01fnH8nKtQJCvKE3O0JAz7CNwBr/99htr1qxhy5Yt+Pj40LdvXzp06MDhw4UUxi5C3mWX+dfh+/peSfX9n//8h379+rF06VJOnTpF3759iz3vuHHjuO222/Dy8mL48OG5gULTyqtXs2BWP9mbmev+4osNJ1h7+AJD2tcjNNCb0EAfGtT0pkFNnyq9Kc21/trSKsccgaMkJCQQGBiIj48Phw8f5s8//yQtLY0NGzZw8uTJ3KGhmjVrMmjQID755BM+/PBDwBgaCgwMJCQkhEOHDtGiRQuWLl2Kn1/hJUATEhKoX99YWj937tzc2wcNGsTnn39Ov379coeGatasSb169ahXrx7Tp09nzZo19v6v0KoYb093pt4UztAO9Zi+/BCLd0aRnH71ooYa3iYGhNfm8f5NaRJc3UEtdQzXWjWUryjN5Qy9jyCvm266iaysLFq2bMmzzz5L9+7dCQ4OZvbs2dx55520b9+eESNGAPDiiy8SHx9PmzZtaN++PevXrwfgrbfe4tZbb+X666+nbt26RV7r3//+N8899xwdO3a8ahXRQw89RMOGDWnXrh3t27fnm2+ulKkeNWoUDRo0oGXLYkoCalo5hNfxZ8FD17Jv2g3seWkQPz/ek09HdeK5weHc0CqEFfujGfj+70yJ2MPJ2BRHN7fCSGVba9ulSxe1Y0cRq/D+eB/WvgLPR4OnD/O3nOI/Px1gx4sDCaru2FTUhw4d0m9wJXj88cfp2LEjDz74YIVcr7DfiYjsVEo5ZL1/sa9trULEJKUze8Nx5v95mowsM7d3rM+k/s1oHFT5KxwW99p2sR5BIrh5GDnH0UNDlUnnzp2JjIzkvvvuc3RTtCos2K8aL9zSij/+3Z8HezZmxT6jhzBj9WGX3qDmenME1a4UpbmcYSSi0kNDzm/nzp2OboKm5coJCBN6X8M7qw7zyfrjRCek8fawdpjcXevzM7haIEjPV4sgMwtPDzddr1jTtDIJ9qvGO3e1o2FNH9779SgXUzL4dFQnfDxd663TtUJbWsFaBHrpqKZp5SEiTBrQjDfvbMuGozHc+8VWLqZkOLpZNuVagSC9YC0CPT+gaZotjOzWkM/u68yh6ETumrWZqPhURzfJZlwrEBQoSmPWgUDTNJu5sXUd5j94LbFJ6Qz7bDMR28/w25EL7D+bwPnENDKzK2eBHNca6MpfuN5JylRqmuY6ujWuyfePXM/YOdv495LIAvcH+Jjo2CCAOzuFMqhVSKV4D3KtQJCvcH1apnMUrq+s8mYa1TTtihZ1/Pjtmb5EX0ojLiWdmKQMYpPTiUvO4HxSGr8dvsCkb3fj5+XBre3qcmenULo0CnTaKoWuEwjyFaUBJ54jWPks/LPPtues0xYGv2XbczqJrKwsnXtIczrVPNwJC/IlrJDNZmazYsuJOJbsiuKnPef4dtsZGtb0oVezIBoH+RJWy3hcw5o+eHo4foTedf668hWlASMNdaCPzjKY49lnn6VBgwY89thjAEybNg0PDw/Wr19PfHw8mZmZTJ8+naFDC6vDfrXk5GSGDh1a6OPmzZvHu+++i4jQrl075s+fz/nz53nkkUc4ceIEAJ999hn16tXj1ltvZf/+/QC8++67JCcnM23atNyEeBs3bmTkyJE0b96c6dOnk5GRQa1atVi4cCEhISGF1k1ISEggMjIyN0/SF198wcGDB/nggw/s8L+qaQW5uQk9mgbRo2kQrw3NYtX+f1i6+yzLI6NJuHyllKabQP1Ab9rUq0GPpkH0bBpEo1o+Fd5zcJ1AkK8oDTjx0JCDPrmPGDGCJ598MjcQREREsHr1aiZPnoy/vz+xsbF0796dIUOGlPhC9PLyYunSpQUed/DgQaZPn87mzZsJCgrKrW8wefJk+vTpw9KlS8nOziY5ObnEGgcZGRnkpFyIj4/nzz//RET48ssveeedd3jvvfcKrZtgMpl4/fXXmTFjBiaTiTlz5vD555+X979P08rEt5oHwzqHMqxzKADxKRmcjEvhVGwKp+JSORmbws5TF1m5/x8A6gd407NpED2aBdGopk/O/liEK3+TYUE++NkwW6rrBIKczKNeV08W+zjj0JCDdOzYkQsXLnDu3DliYmIIDAykTp06PPXUU2zYsAE3NzfOnj3L+fPnqVOnTrHnUkrx/PPPF3jcunXrGD58OEFBRgGQnHoD69aty60x4O7uTo0aNUoMBDkJ8MAoejNixAiio6PJyMjIrZ9QVN2E/v37s3z5clq2bElmZiZt27Yt5f+WptlHoK8ngb6edGp4pcaHUopTcalsPBbLpr9iWbk/mkU7zhR5jhreJiYPaMbo7o1sMrTkOoGgkB7BZWftETjQ8OHDWbx4Mf/88w8jRoxg4cKFxMTEsHPnTkwmE2FhYQXqDBSmrI/Ly8PDA7P5ynK74uobTJo0iSlTpjBkyBB+++03pk2bVuy5H3roId544w3Cw8MZN25cqdqlaRVNRGgc5EvjIF9Gd29Etllx4FwCMUnpAORNc5SZbeabbX/z2vKDzN9yimcHt+TG1iHlGk5y/CyFraRdXZQGjEBQGZZuVaQRI0bw3XffsXjxYoYPH05CQgK1a9fGZDKxfv16Tp8+bdV5inpc//79+f7774mLiwPIHRoaMGAAn332GQDZ2dkkJCQQEhLChQsXiIuLIz09neXLlxd7vZz6Bl9//XXu7Tl1E3Lk9DKuvfZazpw5wzfffMPIkSOt/e/RNKfg7ia0Cw1gQMsQBrQMYWCrK1+D29Zl3gPdmDOuKyZ3Nx5ZsJMRn/9JZNSlMl/PdQJBvh5BtlmRkaU3lOXXunVrkpKSqF+/PnXr1mXUqFHs2LGDtm3bMm/ePMLDw606T1GPa926NS+88AJ9+vShffv2TJkyBYD//ve/rF+/nrZt29K5c2cOHjyIyWTipZdeolu3bgwaNKjYa0+bNo3hw4fTuXPn3GEnKLpuAsDdd99Njx49rCqzqWmViYjQr0VtVj7Ri9fvaMOJ2GSGfLyJ55eWcTWiUqpSfXXu3FkVatuXSr3sr1RitFJKqeS0TNVo6nL1+e/HCj++gh08eNDRTahybrnlFrVmzZoi7y/sdwLsUM722ta0EiRezlDvrDqkZv1W9Ptdca9tl50j0LUIqq5Lly7RrVs32rdvz4ABAxzdHE2zOz8vE8/caF1vvjCuEwjS8hWl0WUqbWLfvn2MHj36qtuqVavG1q1bHdSikgUEBHD06FFHN0PTKg3XCQTp+YrSWHoEzpQ3XCnltFvMi9K2bVv27Nnj6GbYnHLhalOaVlquM1mcllBgDwGAt6dzPEUvLy/i4uL0G5ATUEoRFxeHl5eXo5uiaU7BeT4ul1dawTxD4DxDQ6GhoURFRRETE+PopmgYgTk0NLRMjxWRm4D/Au7Al0qpt/LdPwV4CMgCYoAHlFLWrcvVNAdwnUBQSFEacJ7JYpPJlLsbVqu8RMQd+AQYBEQB20VkmVLqYJ7DdgNdlFKpIjIReAcYUfBsmuYc7DpuIiI3icgRETkmIs8Wcv9YEYkRkT2Wr4fKfLGBr0D/F3N/TMsdGnKOQKC5jG7AMaXUCaVUBvAdcFWWPqXUeqVUTvmqP4GydT00rYLYrUdg5ScngEVKqcfLfcGG1171o7P1CDSXUR/ImwQmCri2iGMBHgRWFnWniEwAJgA0bNjQFu3TtFKzZ4+gxE9O9qQDgeZoInIf0AWYUdQxSqnZSqkuSqkuwcHBFdc4TcvDnnME1n5yGiYivYGjwFNKqQIp9/J+agKSReRIEdcMAmLz3hDydmmbXakUeL4uriKeb6MS7j8LNMjzc6jltquIyEDgBaCPUirdmgvv3LkzVkSKmlTWv2vX5tDXtqMni38GvlVKpYvIw8DXQP/8BymlZgOzSzqZiOxQSnWxfTOdk36+DrEdaCYijTECwD3AvXkPEJGOwOfATUqpC9aeWClVZJfASZ57hdHPt2LZc2ioxE9OSqm4PJ+WvgQ627E9mlZuSqks4HFgNXAIiFBKHRCRV0VkiOWwGUB14HvLIohlDmquplnFnj0Caz451VVKRVt+HILxh6VpTk0ptQJYke+2l/J8P7DCG6Vp5WC3QKCUyhKRnE9O7sBXOZ+cMLLgLQMmWz5FZQEXgbHlvGyJw0cuRj/fqqOqPXf9fCuQ6JQHmqZpVZtzJOLRNE3THEYHAk3TtCrOJQJBSaksXIGIfCUiF0Rkf57baorIryLyl+Vfl6nJKCINRGS9iBwUkQMi8oTldpd9zoXRr23X+j076+u60geCPKksBgOtgJEi0sqxrbKLucBN+W57FlirlGoGrLX87CqygH8ppVoB3YHHLL9XV37OV9GvbZf8PTvl67rSBwIcnMqioiilNmCsrMprKMYmPCz/3l6RbbInpVS0UmqX5fskjKXF9XHh51wI/do2uMzv2Vlf164QCApLZVHfQW2paCF59mH8A4Q4sjH2IiJhQEdgK1XkOVvo17bBJX/PzvS6doVAoAHKWAfscmuBRaQ6sAR4UimVmPc+V33O2tVc8ffsbK9rVwgEViUBc1HnRaQuGLu0Aavz2lQGImLC+GNZqJT6wXKzSz/nfPRrG9f7PTvj69oVAkFuKgsR8cRIZVFVcrssA8ZYvh8D/OTAttiUiAjwf8AhpdT7ee5y2edcCP3aNrjM79lZX9cusbNYRG4GPuRKKovXHdsi2xORb4G+GOlqzwMvAz8CEUBD4DRwt1Iq/6RbpSQiPYE/gH2A2XLz8xjjqS75nAujX9uu9Xt21te1SwQCTdM0rexcYWhI0zRNKwcdCDRN06o4HQg0TdOqOB0INE3TqjgdCDRN06o4HQgqCRHJttS/zfmyWVIqEQnLm/lR0yqSfm07nj1rFmu2dVkp1cHRjdA0O9CvbQfTPYJKTkROicg7IrJPRLaJSFPL7WEisk5EIkVkrYg0tNweIiJLRWSv5et6y6ncReQLS470/4mIt8OelKahX9sVSQeCysM7X/d5RJ77EpRSbYGPMXahAnwEfK2UagcsBGZabp8J/K6Uag90Ag5Ybm8GfKKUag1cAobZ9dlo2hX6te1gemdxJSEiyUqp6oXcfgror5Q6YUlm9Y9SqpaIxAJ1lVKZltujlVJBIhIDhCql0vOcIwz41VIUAxGZCpiUUtMr4KlpVZx+bTue7hG4BlXE96WRnuf7bPT8keYc9Gu7AuhA4BpG5Pl3i+X7zRjZKgFGYSS6AqMM3kQwSiGKSI2KaqSmlYF+bVcAHRkrD28R2ZPn51VKqZxldoEiEonxyWek5bZJwBwReQaIAcZZbn8CmC0iD2J8OpoIRKNpjqNf2w6m5wgqOcs4ahelVKyj26JptqRf2xVHDw1pmqZVcbpHoGmaVsXpHoGmaVoVpwOBpmlaFacDgaZpWhWnA4GmaVoVpwOBpmlaFff/5VMuA0S4PvsAAAAASUVORK5CYII=",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Generate the model:\n",
    "model = make_model()\n",
    "model.compile(\n",
    "   optimizer=\"adam\",\n",
    "   loss=tf.keras.losses.CategoricalCrossentropy(),\n",
    "   metrics=[tf.keras.metrics.CategoricalAccuracy()],\n",
    ")\n",
    "\n",
    "## baseline\n",
    "# history = model.fit(\n",
    "#     train_x,\n",
    "#     train_y_cat,\n",
    "#     epochs=40,\n",
    "#     batch_size=32,\n",
    "#     validation_data=(test_x, test_y_cat),\n",
    "#     callbacks=[earlystop, ln_callback],\n",
    "#     class_weight=class_weights,\n",
    "# )\n",
    "\n",
    "# augmented african class\n",
    "history = model.fit(\n",
    "    train_x_plus_africans_augmented,\n",
    "    train_y_cat_plus_africans,\n",
    "    epochs=40,\n",
    "    batch_size=64,\n",
    "    validation_data=(test_x, test_y_cat),\n",
    "    callbacks=[earlystop, ln_callback],\n",
    "    #callbacks=[earlystop],\n",
    "    class_weight=class_weights\n",
    ")\n",
    "\n",
    "\n",
    "# Evaluate the model\n",
    "fig, (ax1, ax2) = plt.subplots(1, 2)\n",
    "ax1.plot(history.history[\"categorical_accuracy\"], label=\"accuracy\")\n",
    "ax1.plot(history.history[\"val_categorical_accuracy\"], label=\"val_accuracy\")\n",
    "ax1.set_xlabel(\"Epoch\")\n",
    "ax1.set_ylabel(\"Accuracy (Cat)\")\n",
    "ax1.set_ylim([0.5, 1])\n",
    "ax1.legend(loc=\"lower right\")\n",
    "\n",
    "ax2.plot(history.history[\"loss\"], label=\"loss\")\n",
    "ax2.plot(history.history[\"val_loss\"], label=\"val_loss\")\n",
    "ax2.set_xlabel(\"Epoch\")\n",
    "ax2.set_ylabel(\"Loss (CXE)\")\n",
    "ax2.legend(loc=\"upper right\")\n",
    "\n",
    "# test_loss, test_acc = model.evaluate(test_x,  test_y, verbose=2)\n",
    "# print(f\"Test Loss: {test_loss:.3f} Test Accuracy {test_acc:.3f}\")\n",
    "\n",
    "y_hat_test_cat = model.predict(test_x)\n",
    "y_hat_test = np.argmax(y_hat_test_cat, axis=-1)\n",
    "# test_loss, test_acc = mlp.evaluate(test_x,  test_y, verbose=2)\n",
    "\n",
    "ConfusionMatrixDisplay.from_predictions(\n",
    "    test_y, y_hat_test, display_labels=[\"Caucasian\", \"African\", \"Asian\", \"Indian\"]\n",
    ")\n",
    "\n",
    "bacc = balanced_accuracy_score(test_y, y_hat_test)\n",
    "\n",
    "print(f\"BACC = {bacc:.3f}\")\n",
    "print(\"Y_hat shape:\", y_hat_test.shape)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 170,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "93/93 [==============================] - 5s 55ms/step - loss: 1.1981 - categorical_accuracy: 0.4043 - val_loss: 1.1151 - val_categorical_accuracy: 0.5007\n",
      "Epoch 2/20\n",
      "93/93 [==============================] - 7s 76ms/step - loss: 0.8706 - categorical_accuracy: 0.6256 - val_loss: 0.7305 - val_categorical_accuracy: 0.7388\n",
      "Epoch 3/20\n",
      "93/93 [==============================] - 7s 74ms/step - loss: 0.7547 - categorical_accuracy: 0.6877 - val_loss: 0.8043 - val_categorical_accuracy: 0.6845\n",
      "Epoch 4/20\n",
      "93/93 [==============================] - 7s 75ms/step - loss: 0.6695 - categorical_accuracy: 0.7232 - val_loss: 0.6835 - val_categorical_accuracy: 0.7415\n",
      "Epoch 5/20\n",
      "93/93 [==============================] - 7s 75ms/step - loss: 0.6191 - categorical_accuracy: 0.7452 - val_loss: 0.7591 - val_categorical_accuracy: 0.6974\n",
      "Epoch 6/20\n",
      "93/93 [==============================] - 7s 75ms/step - loss: 0.5756 - categorical_accuracy: 0.7615 - val_loss: 0.5304 - val_categorical_accuracy: 0.8148\n",
      "Epoch 7/20\n",
      "93/93 [==============================] - 7s 73ms/step - loss: 0.5283 - categorical_accuracy: 0.7770 - val_loss: 0.5501 - val_categorical_accuracy: 0.7910\n",
      "Epoch 8/20\n",
      "93/93 [==============================] - 7s 73ms/step - loss: 0.5078 - categorical_accuracy: 0.7902 - val_loss: 0.5979 - val_categorical_accuracy: 0.7809\n",
      "Epoch 9/20\n",
      "93/93 [==============================] - 7s 72ms/step - loss: 0.4618 - categorical_accuracy: 0.8028 - val_loss: 0.6798 - val_categorical_accuracy: 0.7442\n",
      "Epoch 10/20\n",
      "93/93 [==============================] - 7s 72ms/step - loss: 0.4511 - categorical_accuracy: 0.8111 - val_loss: 0.6074 - val_categorical_accuracy: 0.7626\n",
      "Epoch 11/20\n",
      "93/93 [==============================] - 7s 72ms/step - loss: 0.4031 - categorical_accuracy: 0.8221 - val_loss: 0.4619 - val_categorical_accuracy: 0.8351\n",
      "Epoch 12/20\n",
      "93/93 [==============================] - 7s 72ms/step - loss: 0.3903 - categorical_accuracy: 0.8254 - val_loss: 0.4757 - val_categorical_accuracy: 0.8270\n",
      "Epoch 13/20\n",
      "93/93 [==============================] - 7s 74ms/step - loss: 0.3501 - categorical_accuracy: 0.8467 - val_loss: 0.4731 - val_categorical_accuracy: 0.8256\n",
      "Epoch 14/20\n",
      "93/93 [==============================] - 6s 67ms/step - loss: 0.3557 - categorical_accuracy: 0.8411 - val_loss: 0.4467 - val_categorical_accuracy: 0.8372\n",
      "Epoch 15/20\n",
      "93/93 [==============================] - 6s 68ms/step - loss: 0.3234 - categorical_accuracy: 0.8540 - val_loss: 0.4719 - val_categorical_accuracy: 0.8324\n",
      "Epoch 16/20\n",
      "93/93 [==============================] - 6s 67ms/step - loss: 0.2999 - categorical_accuracy: 0.8569 - val_loss: 0.5088 - val_categorical_accuracy: 0.8195\n",
      "Epoch 17/20\n",
      "93/93 [==============================] - 6s 69ms/step - loss: 0.3101 - categorical_accuracy: 0.8562 - val_loss: 0.4596 - val_categorical_accuracy: 0.8351\n",
      "Epoch 18/20\n",
      "93/93 [==============================] - 6s 68ms/step - loss: 0.2840 - categorical_accuracy: 0.8676 - val_loss: 0.5311 - val_categorical_accuracy: 0.8033\n",
      "Epoch 19/20\n",
      "93/93 [==============================] - 6s 65ms/step - loss: 0.2809 - categorical_accuracy: 0.8686 - val_loss: 0.4854 - val_categorical_accuracy: 0.8229\n",
      "BACC base_= 0.784\n",
      "Epoch 1/40\n",
      "97/97 [==============================] - 7s 66ms/step - loss: 0.3653 - categorical_accuracy: 0.8635 - val_loss: 0.4220 - val_categorical_accuracy: 0.8480\n",
      "Epoch 2/40\n",
      "97/97 [==============================] - 7s 67ms/step - loss: 0.3250 - categorical_accuracy: 0.8804 - val_loss: 0.4147 - val_categorical_accuracy: 0.8555\n",
      "Epoch 3/40\n",
      "97/97 [==============================] - 7s 69ms/step - loss: 0.3059 - categorical_accuracy: 0.8902 - val_loss: 0.4140 - val_categorical_accuracy: 0.8541\n",
      "Epoch 4/40\n",
      "97/97 [==============================] - 7s 71ms/step - loss: 0.2996 - categorical_accuracy: 0.8897 - val_loss: 0.4117 - val_categorical_accuracy: 0.8507\n",
      "Epoch 5/40\n",
      "97/97 [==============================] - 6s 67ms/step - loss: 0.2931 - categorical_accuracy: 0.8931 - val_loss: 0.4048 - val_categorical_accuracy: 0.8569\n",
      "Epoch 6/40\n",
      "97/97 [==============================] - 7s 71ms/step - loss: 0.2849 - categorical_accuracy: 0.8965 - val_loss: 0.4076 - val_categorical_accuracy: 0.8643\n",
      "Epoch 7/40\n",
      "97/97 [==============================] - 7s 70ms/step - loss: 0.2770 - categorical_accuracy: 0.8981 - val_loss: 0.4072 - val_categorical_accuracy: 0.8582\n",
      "Epoch 8/40\n",
      "97/97 [==============================] - 7s 76ms/step - loss: 0.2643 - categorical_accuracy: 0.9002 - val_loss: 0.4064 - val_categorical_accuracy: 0.8575\n",
      "Epoch 9/40\n",
      "97/97 [==============================] - 7s 70ms/step - loss: 0.2648 - categorical_accuracy: 0.9019 - val_loss: 0.4151 - val_categorical_accuracy: 0.8528\n",
      "Epoch 10/40\n",
      "97/97 [==============================] - 7s 70ms/step - loss: 0.2554 - categorical_accuracy: 0.9041 - val_loss: 0.3973 - val_categorical_accuracy: 0.8575\n",
      "Epoch 11/40\n",
      "97/97 [==============================] - 6s 66ms/step - loss: 0.2474 - categorical_accuracy: 0.9116 - val_loss: 0.4120 - val_categorical_accuracy: 0.8521\n",
      "Epoch 12/40\n",
      "97/97 [==============================] - 7s 67ms/step - loss: 0.2419 - categorical_accuracy: 0.9108 - val_loss: 0.4051 - val_categorical_accuracy: 0.8575\n",
      "Epoch 13/40\n",
      "97/97 [==============================] - 6s 66ms/step - loss: 0.2344 - categorical_accuracy: 0.9152 - val_loss: 0.4016 - val_categorical_accuracy: 0.8569\n",
      "Epoch 14/40\n",
      "97/97 [==============================] - 7s 69ms/step - loss: 0.2259 - categorical_accuracy: 0.9194 - val_loss: 0.4008 - val_categorical_accuracy: 0.8596\n",
      "Epoch 15/40\n",
      "97/97 [==============================] - 6s 66ms/step - loss: 0.2261 - categorical_accuracy: 0.9173 - val_loss: 0.4202 - val_categorical_accuracy: 0.8555\n",
      "BACC = 0.800\n",
      "Y_hat shape: (1474,)\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEKCAYAAAAfGVI8AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/MnkTPAAAACXBIWXMAAAsTAAALEwEAmpwYAABL90lEQVR4nO3dd3hUZfrw8e+dTCpJgJDQkiAJoXcIICooWEBUsCF2cVUsoFjWtaxrW3fX1X11V7GhP8GOiA0bLiouFqSKVJUSIAktEAgESJvc7x9nEoaQRshkUu7Pdc2VmdPmOcPh3OfpoqoYY4xpvAL8nQBjjDH+ZYHAGGMaOQsExhjTyFkgMMaYRs4CgTHGNHIWCIwxppHzWSAQkVdFZKeIrCpnvYjIMyKyXkRWiEg/X6XFGGNM+XyZI5gOjKxg/dlAR89rAvCCD9NijDGmHD4LBKo6H8iqYJMxwOvq+AloJiJtfJUeY4wxZXP58bvjgDSvz+meZdtKbygiE3ByDTRp0qR/ly5daiWBpvFZunTpLlWN9cd3x8TEaPv27f3x1aYRqOja9mcgqDJVnQpMBUhJSdElS5b4OUWmoRKRzf767vbt22PXtvGViq5tf7YaygASvD7He5YZY4ypRf4MBLOBqz2th04EslX1qGIhY4wxvuXL5qPvAAuAziKSLiLXichNInKTZ5PPgY3AeuBl4BZfpcWYisyZM4fOnTuTnJwM0Lq87UTkIhFREUnxfD5TRJaKyErP3+Fe234rIr+JyHLPq6Xvz8SY6vFZHYGqXlbJegUm+ur7jakKt9vNxIkTmTt3LvHx8YSEhESLSDdVXeO9nYhEApOBhV6LdwHnqepWEekBfInT4KHYFapqhf41pKCggPT0dHJzc/2dlDotNDSU+Ph4goKCqrxPvagsNsZXFi1aRHJyMklJScWLsnCaNq8ptelfgX8CdxcvUNWfvdavBsJEJERV83yY5EYrPT2dyMhI2rdvj4j4Ozl1kqqye/du0tPTSUxMrPJ+NsSEadQyMjJISPBus0A+Rz7V4+n1nqCqn1VwqIuAZaWCwDRPsdBfpJw7l4hMEJElIrIkMzOzmmfROOTm5tKiRQsLAhUQEVq0aHHMuSYLBMZUQEQCgKeAuyrYpjtObuFGr8VXqGpPYIjndVVZ+6rqVFVNUdWU2Fi/dF+oVywIVK46v5EFAtOoxcXFkZbm3a+RYI5sxhwJ9AC+FZFNwInAbK8K43jgQ+BqVd1QvJOqZnj+7gfeBgZWK4HuQphzH2Rby2rjOxYITKM2YMAA1q1bR2pqKvn5+QDROE2bAVDVbFWNUdX2qtoe+AkYrapLRKQZ8Blwr6r+ULyPiLhEJMbzPgg4Fyhz8MVK7VwDy16HF0+B37+s3kmaGhMREeHvJPiEBQLTqLlcLqZMmcKIESPo2rUrQJaqrhaRR0VkdCW7TwKSgQdLNRMNAb4UkRXAcpwcxsvVSmCbXjDhfxAVB29fAv99ANwF1TqUaeD2b4OcndXa1QKBafRGjRrF77//zoYNGwC2A6jqg6o6u/S2qnpacZNQVX1MVZuoah+v105VPaCq/VW1l6p2V9XJququdgJjkuH6r2DA9fDjs/DqSNjjt5EwDE7rnLvvvpsePXrQs2dP3n33XQC2bdvG0KFD6dOnDz169OC7777D7XYzfvz4km2ffvrpmk9QYT7s3wGF1Wtaa81HjakPgkLhnP8H7U+B2bfBS0NgzPPQ9Vx/p8wvHvlkNWu27qvRY3ZrG8VD53Wv0rYffPABy5cv55dffmHXrl0MGDCAoUOH8vbbbzNixAj+/Oc/43a7OXjwIMuXLycjI4NVq5zSwb1799ZougHI2eH8jWhVrd0tR2BMfdL9ArhxPkQnwbtXwBf3QKF1W6ht33//PZdddhmBgYG0atWKU089lcWLFzNgwACmTZvGww8/zMqVK4mMjCQpKYmNGzdy6623MmfOHKKiomo2MYX5cHA3hLcAV0i1DmE5AmPqm+hE+MOX8NXD8NPzsOUnGDvNCQ6NRFWf3Gvb0KFDmT9/Pp999hnjx4/nzjvv5Oqrr+aXX37hyy+/5MUXX2TmzJm8+uqrNfelOdudv9XMDYDlCIypn1whMPIfcOnbsCcVXhwKK2bCgV1Ok1MDWgQHMn1SuT5kyBDeffdd3G43mZmZzJ8/n4EDB7J582ZatWrFDTfcwPXXX8+yZcvYtWsXRUVFXHTRRTz22GMsW7as5hJSmAcHs6BJC3AFV/swliMwpj7rcg7c9D3M+gN8cMPh5aFNnaKCsGjnb7jnb1hziGwNcSkQ0wkCGuizYJEbslIhfz/k7a/x3NIFF1zAggUL6N27NyLCE088QevWrXnttdd48sknCQoKIiIigtdff52MjAyuvfZaioqKAPjHP/5RcwmpgdwAgDhjv9UfNjGN8SURWaqqKf747uO6tt0F8NsXsH87HMpyyowPev4eyvK8z4KCA4f3CYuGdid6XidBm97H9VTpa2vXri1u4lsxdz7s3ui0oAmJcAJBi47O+4akMBd2roUmsdA0/ohVZf1WFV3bliMwpiEIDIJulXV7AApyITvNqVfY8hNsWQC/fe6sc4U6OYUTBjvBIWEQhET6Nt01rSAXsjZAUaGTCwhu4tws922FmI5wPENUaBHkH4DgiOM7Tk3ZvwMIOO7cAFggMKZxCQp1bogxHaGfZ/ijnJ1OQCgODN89BeqG4Eg4+5/Q5/K6ceOrTF4OZG100tqiIwSHO8sjWzvBLzcbwppV//j7tsGBnU5waXZCtVvo1IiCXCen16Sl8xBwnCwQGNPYRbSEbmOcFzhFKemLYf6/4ONb4Pcv4LxnnHqGuurQXtizCQKDoUWHI2/S4S2cSuN9WyE0CqQa9SL5BzxBIAIKDkHmr9A0walz8UeQzNnunEdEzcx31EBriowx1RYSCR2GwzWfwBmPwG9z4PnBsP4rf6esbDmZTsupoHCnArz0k7oIRLUFdx4c2H3sxy8qcnpyBwY7xU2xnSEoDPZudoJPbbfSKjgEh/ZAk5gayQ2ABQJjTHkCAuGU2+GGb5wn3zcvgs/vhvyD/k6ZQxX2ZcC+dAhp6uQEAssp5AiJcp7mc7Y79QfHYv82J4g0TXB+E1eIU/QU2cYpbsr8FXJrtpdzxenx5AaaHH/dQDErGjLGVKxNL5gwD756BBa+ABu/hQtfhrZ9Kt4v/wCkfgfr/uvkJnJ2OOXrwU2c+oeS902cm3RIhPO+uOlr6Zd3C0ctgr1bnCfj8Bin1UxFRTQizsB9u35z6kSi2lbt3PNynCKh8BZOsZL38SJbOwFm7yangrpJLES29W2T3IJDkLvXqSAuL+hVgwUCY0zlgsLg7Meh01nw0S3wyukw7H44+XbnKRmcG/Xu9c6Nf91c2PyD05QzKBwST3XqIPIPeF45h9/n7Dj8OS/Hefouy4iZsK0QAlyAOseObOPcFKtSTh8c7uRscnY6waOyprJFnmATGOwEkfKOGdPZqX84kOnUrzRv7/xevrB/G0igU0lcgywQGGOqrsNwuPlH+PQO+PpR54Y/6EbY9L3zfq9nVNSYTjBwAiSfASecdGwtbAoOHe4DUfLKguCmEN4c3G7QQicIHGsFdmQbp2J5/zZofkLF2xYXCbVIPhzsyhIQCM0SnJzM3s2Q+RtEtiaiTTI5OTll7rJp0ybOPffckoHoqiT/oFMUFdG6RnMDYIHAGHOswqNh7HRY8a5TZ/DeeM9T/1A4+TZIPrPym2xFgsKgaZzz8rZ2rVNOfzxcIU4RzoGdEBHrpLssJUVCMVXvSxEaBbFdnKaq+7c5xVe7NzrBK6Tp8RcZ7d/u5AYian5KUwsExphjJwK9L3WKfLI2OB3RgkJr7/u/uBe2r6zmzuoUQwUEgsurCKd1T6f464giocN1Cffeey8JCQlMnDgRgIcffhiXy8W8efPYs2cPBQUFPPbYY4wZPRoKDjq/UcFB2JPtVO6GNnOKpkoFltzcXG6++WaWLFmCy+XiqaeeYtiwYaxevZprr72W/Px8ityFvP/C32ib3JNLzhtDeno6brebv/zlL4wbN66av8Nh1mrINHpz5syhc+fOJCcnA7QubzsRuUhEtHi+Ys+y+0RkvYj8JiIjvJaP9CxbLyL3+vYM/CiqjTNHQm0GgeMmzk2+qLDsFkTFRULN2h1RJDRu3DhmzpxZ8nnmzJlcc801fPjhhyxbtox58+Zx1113oeBUeiPQqrtTtBTWzCnWydoAO1Y5ndO0CFR57rnnEBFWrlzJO++8wzXXXENubi4vvvgikydPZvny5Sz573vEt23DnO+W0rZtW3755RdWrVrFyJEja+QXsRyBadTcbjcTJ05k7ty5xMfHExISEi0i3VR1jfd2IhIJTAYWei3rBlwKdAfaAl+JSCfP6ueAM4F0YLGIzC59THMczn78+PbXImfoCQl0+gUUVzZXUCTUt29fdu7cydatW8nMzKR58+a0bt2aO+64g/nz5xMQEEBGRgY7duygdWvP84SIc5yQSIgqgrx9To/gQ2lOZfeOVXz/9RfcesPVkJVKl9ZNOCG+Db8vX8Dgfj35298eIz11HRee1peOvQfRs3cz7rr7bu655x7OPfdchgwZcny/g4flCEyjtmjRIpKTk0lKSiI4OBggCxhTxqZ/Bf4JeM8FOAaYoap5qpoKrAcGel7rVXWjquYDM8o5pvEXCXAqjgsPOTdmcEYsLaNIyNvYsWOZNWsW7777LuPGjeOtt94iMzOTpUuXsnz5clq1akVubjnTRQYEODmD6CSI7QQBQU4FswQA6qQld68TIHJ2cvmIAcx+5Z+E6UFGXX0b3yxcSadOnVi2bBk9e/bkgQce4NFHH62Rn8MCgWnUMjIySEg4ogIyHziillJE+gEJqvpZqd3jgDSvz+meZeUtP4qITBCRJSKyJDMzs3onYaonrLlTWbxvm1MvUE6RkLdx48YxY8YMZs2axdixY8nOzqZly5YEBQUxb948Nm+u4lzSAS5Pa6N2DDl9JG99Oh9aduP3fSFs2Z5F55POZeP+UJL6DeO2P/6ZMedfwIpVq9m6dSvh4eFceeWV3H333TU2t4EVDRlTAREJAJ4Cxvvi+Ko6FZgKzjDUvvgOU47ioSd2r3eafeburbSVUPfu3dm/fz9xcXG0adOGK664gvPOO4+ePXuSkpJCly5djjkZt9xyCzfffDM9e/bE5XIxffp0QsKbMPPDj3njjTcICgqidevW3P/AgyxevJi7776bgIAAgoKCeOGFF47jBzjMAoGp9/ILi1i/M4ctWQeICgsiNiKE2MgQmoYFIZV0NIqLiyMtzfvhnWAgw+tzJNAD+NZzrNbAbBEZ7dnOOzsR77VvectNXRIS6fQOzt1bYZGQt5UrD7dWiomJYcGCBWVuV14fAoD27duX9CEIDQ1l2rRpR21z7733cu+9R7YzGDFiBCNGjDhq2+NlgcDUK3sP5rNm2z7WbN3H2m37WbNtH+t37qfAffTDdFCglASFGM/f4teF/eKJCHExYMAA1q1bR2pqKnFxcQDRwOziY6hqNhBT/FlEvgX+qKpLROQQ8LaIPIVTWdwRWAQI0FFEEnECwKXA5b76TcxxioqDPQXOMBUVdRxrwCwQmDpJVUnfc4jVW/exZmt2yc1/a/bhirjYyBC6tYni1E6xdGsbRVJME/blFrArJ5/M/XmHXzl5bM3O5Zf0bLIO5FGkcF6vthACLpeLKVOmMGLECNxuN0CWqq4WkUeBJao6u4I0rhaRmcAaoBCYqKpuABGZBHwJBAKvqupq3/1a5rgEhULLYy/SqYqVK1dy1VVXHbEsJCSEhQsXlrOHf9hUlcbv3EVK6q4cVm/dx+qt+1iVkc3qrfvIPuRMOh4g0CE2gq5toujWNopubaLo2iaK2MhjnxjEXaRkHcgnJiK4zGKjejtVZSOwdu1aunTpUmlxX2Onqvz66682VaWpfe4iZenmPXy+chtz1+xgf24BoUGBhAUHEuoKJDQogJCgQEKDAgl1BRAWHIgrIIBNuw+wZus+DhW4AQh2BdCldSSjerame9umdG8bRZfWUYQF10yWPTBAqhVAjP+Fhoaye/duWrRoYcGgHKrK7t27CQ09tg5+FghMtbmLlEWpWXyxahtfrNpO5v48gl0BDO0YS3zzMHIL3BwqcJNb4Ca3oIjcAjfZhwrY6VmWV1hEQvNwxg1IoEecc9NPbhlBUKC1ajZHi4+PJz09HWtmW7HQ0FDi4+Mr39CLBQJzTArdRSxMzeKzldv47+rt7MrJJzQogGGdW3J2zzYM79KSiBC7rEzNCwoKIjEx0d/JaJB8+j9WREYC/8GpMHtFVR8vtf4E4FUgFqdH55Wqmu7LNJmKqSr7DhWyY38uO/flsWNfLjv3O3937Mvlp4272XOwgLCgQIZ3bcmoHm0Y1iWW8GC7+RtTX/nsf6+IBFL5eCv/Al5X1ddEZDjwD+Cqo49mqiu3wM3egwXsOZjP3oMFZB/K93wuYO+hfLI963bn5Jfc/PMKi446TmSIi9ioEIZ0jGVUzzac2im2xsrtjTH+5cvHuJLxVgBEpHi8Fe9A0A240/N+HvCRD9PTYOQWuMk6kH9E88hdnr/Fy3Z53h/Id5d7nGBXAM3Dg2gWFkx0k2D6t2tOy6hQWkaG0Mr7b1SIPfEb04D58n93WeOtDCq1zS/AhTjFRxcAkSLSQlV3e28kIhOACQDt2rXzWYL9RVXZvi+X1MwDbMk6WPK0vveA5+/BAufleV/WEztAs/DDvWp7xTcjNjKEFhHBNA8PpllYEE3Dg5z3npu/PdEbY8D/lcV/BKaIyHhgPk4vzKMeYRvKeCz7cgtIzTzAxl05nr8H2Jh5gNRdB0qaTxbzflpvGh5E+5hwmoU1c27i4cE0Dw8q6SUbE+Hc8ENcdmM3xhw7XwaCisZhAUBVt+LkCBCRCOAiVd3rwzT5XH5hEVuyDrIxM4fUXc5NfqPnpr8r5/Ck3AECCdHhJMY0YVBSNEmxESTFNOGEFuG0aBJCaFCAtZU2xtQKXwaCxVQy3oqIxOB06S8C7sNpQeR363bsZ+aSNPYeLCDIFUBQgBAUGIArMIDgQMEVGEBQYABBgYKIkLHnEKm7nBt/2p5DuIsOZ1paNAkmKbYJw7vEkhQbQWJMEzrENiEhOtye4I0xdYLPAoGqFpY13kqpMVxOA/4hIopTNDTRV+mpjLtImffrTqb/uInv1+8iODCAFhHBFLiLKHArBe4iCt1Kvvvo8vnQoAASYyLo3rYp5/VuS2JMExJjmpAUE0HT8CA/nI0xxlSdT+sIVPVz4PNSyx70ej8LmOXLNFQm+1AB7y1J4/UFm9mSdZDWUaHcPaIzlw5IoEXE0UMRqCruInWCQ1ERbrfSNCyIgAArxjHG1E/+riz2m/U79/Paj5t5f1k6B/PdDGjfnHtGduGs7q0qHOJARHAFCq5ACMOKdowx9V+jCwQ/rN/Fi//bwHfrdhHsCmBM77Zcc1J7esQ19XfSjDHGLxpNIMgvLOKfc37l/75PrbT4xxhjGpNGEQjSsg4y6e1l/JKezfiT2nPfqC7WYscYYzwa/Hi/n6/cxqhnviN11wFevLIfD4/ubkHAHGHOnDl07tyZ5ORkcOYkPoKI3CQiK0VkuYh8LyLdPMuv8CwrfhWJSB/Pum9F5DevdS1r9aSMOQYNNkeQW+Dmb5+t5Y2fNtM7oRlTLutLQnS4v5Nl6hi3283EiROZO3cu8fHxhISERItIt1KDI76tqi8CeCatfwoYqapvAW95lvcEPlLV5V77XaGqNuWYqfMaZCDYmJnDpLd/Zs22fdwwJJG7R3Qh2NXgMz+mGhYtWkRycjJJSUnFi7IoNTiiqu7z2qUJUNYwJ5cBM3yVTmN8qcEFgo+XZ3D/BysJcgXwf9ekcHrXVv5OkqnDMjIySEjwHgmFfJwBE48gIhNxRsoNBoaXcahxOAHE2zQRcQPvA49pfZsg3DQaDeYx+VC+m3tmrWDyjOV0bRPF57cNqX9BICsVVn8Ie7f4OyWmFFV9TlU7APcAD3ivE5FBwEFVXeW1+ApV7QkM8bzKnGdDRCaIyBIRWWJTMBp/aRA5AlXl8ld+YnnaXiYO68AdZ3TCVV/mvd29AdZ8BGs+hm2/HF4e3QE6DIOkYZA4BEKtn4MvxMXFkZbmPVo6wZQaHLGUGcALpZZdCrzjvUBVMzx/94vI2zjzc7xe+mANZWRdU781iEAgItwwJImIEBdDO8X6OzmVy/zNufGv+Rh2eB4i4wfAWY9BwiBIXwIb58Hyd2DxKyABENffCQodhjnbBtoYRgAUFcGBTMhOh4O7oUUHaN4eAqrWMmzAgAGsW7eO1NRU4uLiAKKB2d7biEhHVV3n+XgOsM5rXQBwCc5Tf/EyF9BMVXeJSBBwLvBV9U/SGN9qEIEAYFTPNv5OQvlUYecaWDPbefrP/BUQaHcijHwcup4HTeMPb58wEAbfAoX5kL7YCQob5sF3/4L5T0BwBLTt6/x1BYMrFFwhEBji/C3+7AqB8Bhn29jOVb45HqWoCLI2QMYyCAqDdoMh4jgCblERZK51Ap4735Pe0MPnEhh85DkEuGD/dshOc274e9M879MgOwPceUcePygcWnaFVt2hVQ/nb8tuEB59VFJcLhdTpkxhxIgRuN1ucEbDLT044iQROQMoAPYA13gdYiiQVjwTn0cI8KUnCATiBIGXq/+DGeNbUt/qr1JSUnTJkmq2yNu9AX79DHKzITTKKW4JiXLehzR1PodGOcuCwuB45gPI2Qkbv3Vu4Bu/hf1bAYETToZuY5ybf9QxBq9DeyD1O+d421dAYS4U5nm9cp0ba8EhjmrYEtQE2vZxgkJcf4jrB81OOPocVWHfVti6DDKWOjf/rT9D3r4jt2vREU4YDO1Ocv6Wdaxi7gLYuhy2/AibF8CWBZC799jO3VtEa2iW4ATPpgnOq1kChDWH3ethx2onp7V9FRzKOrxfZFtPcOgOJ08uMzCIyFJVTal+4qrvuK5tYypR0bXdYHIE5dq1znkKX/0x7FjpLJMA0LKneywREOTcXFoke14dnJtfi2SIanv0TS//IGz+8fDT+87VzvKw5pB4qlOk0+lsiDyOCuyw5tBttPOqiCoUFTrBoeSm7rmxL3oZ3FOc7cJbQNt+TmAIdHm2WQY52z2/gcu5afa82LNdP8jLOXxDX/MxLPMUe0e2hRM8QSHhRDi4y3PT/xHSFkPhIWe76A7Q9VwngLQ7EUIiywloXu+L3BDR0rnxR8U5uYTytDvxyN8hZ4cTFHas9rzWQOr/YMhd1fs3MKYBang5AlWn6KW4DH6npzl4wqDDT+JNEyA/B3L3ObmDvH3O+7x9zpNqrufv3i3OE+buDVBw8PB3BIU7N7SYZOfGtO0XSFvoPI0HBjs3o+Ly/Na9IaAOVVwX5jtBqvimv3UZ7FwLqBPo4jyBoW0/aN0TgkLLP1ZRkfP7blngBMEtC2D/tsPrJcApmjnhJKc4qd3g4wuENcVdUG4di+UITENV0bXdcALB9pWHb/67fscphjnJqximbfW/VNW5we1adzgw7F7vvPZuhtiu0OE05+bfbjAE17MezHk5oO7jb5mkCntSnRxAeAtIGFDvWjtZIDANVeMoGvrkdufp9oSTYeAE5+YfedSwMdUj4gSSqLaQdOqR61SPry6hLgiJqJnjiEB0kvMyNeJAXiE/rN9Fn4RmtIyqIHdmzHFoOIHgvP9ARKvja81SHfU9CJg6bce+XCa8sZR/XNiTywa283dyTANVhwqvj1PrHrUfBIzxscSYJsREhLAoNavyjY2ppoYTCIxpgESEQYnRFgiMT1kgMKaOG5gYTcbeQ6TvOVj5xsZUgwUCY+q4QUlOx7eFGy1XYHzDAoExdVynlpE0DQuy4iHjMxYIjKnjAgKEAe2jWbTJAoHxDQsExtQDgxKjSd11gJ37cv2dFNMAVdiPQERCcYbQHQK0BQ4Bq4DPVHW175NnjAGveoLULM7rfRy95I0pQ7k5AhF5BPgBGAwsBF4CZgKFwOMiMldEetVKKo1p5Lq1iaJJcKDVExifqChHsEhVHypn3VMi0hKwro7G1AJXYAD921t/AuMb5eYIVPUzABEZW3qdiIxV1Z2qaiNkGVNLBiVG89uO/WQdyPd3UkwDU5XK4vuquMwY40ODEp16gsXWesjUsIrqCM4WkWeBOBF5xus1HaeewJgGYc6cOXTu3Jnk5GSAo4asFZGbRGSliCwXke9FpJtneXsROeRZvlxEXvTap79nn/We/zfHPTphz/imhLgCrHjI1LiK6gi2AkuA0cBSr+X7gTt8mShjaovb7WbixInMnTuX+Ph4QkJCokWkm6qu8drsbVV9EUBERgNPASM96zaoap8yDv0CcANOQ4vPPdt/cTxpDXEF0rddMwsEpsaVGwhU9RfgFxF5W1ULajFNxtSaRYsWkZycTFJSyRwKWcAYoCQQqKr3hM1NOGpC6COJSBsgSlV/8nx+HTif4wwEAIMSW/DsN+vYl1tAVGjZs6wZc6yqUkfQXkRmicgaEdlY/PJ5yoypBRkZGSQkJHgvygfiSm8nIhNFZAPwBHCb16pEEflZRP4nIkM8y+KAdK9t0ss6pue4E0RkiYgsyczMrDS9gxKjKVJYunlPpdsaU1VVCQTTcLK5hcAw4HXgzaocXERGishvnnLSe8tY305E5nn+I60QkVHHknhjaouqPqeqHYB7gAc8i7cB7VS1L3An8LaIRB3jcaeqaoqqpsTGVj6fRt92zXEFiBUPmRpVlUAQpqpf48xvvFlVHwbOqWwnEQkEngPOBroBlxVXsnl5AJjp+Y90KfD8sSTemGILFixg4sSJ9OrVi9jYWNq1a8eoUaN47rnnyM7OLne/uLg40tLSvBcFAxkVfNUMnGIeVDVPVXd73i8FNgCdPPvHe+0TX8kxqywsOJBe8U0tEJgaVZVAkCciAcA6EZkkIhcAVZnkdiCwXlU3qmo+zn+gMaW2UaD4CaopTgW1Mcfk7LPP5pVXXmHEiBHMmTOHbdu2sWbNGh577DFyc3MZM2YMs2fPLnPfAQMGsG7dOlJTU8nPzweIBo7YWEQ6en08B1jnWR7reeBBRJKAjsBGVd0G7BOREz2tha4GPq6p8x2U1IIV6Xs5lO+uqUOaRq4qcxZPBsJxykX/ilM8dE0V9osDvB+10oFBpbZ5GPiviNyKUwl3RlkHEpEJwASAdu2sM7M50htvvEFMTMwRyyIiIujXrx/9+vXjrrvuYteuXWXu63K5mDJlCiNGjMDtdgNkqepqEXkUWKKqs4FJInIGUADs4fD1PxR4VEQKgCLgJlUtflS/BZgOhOFUEh93RXGxgYnRvPDtBn7esoeTkmMq38GYSpQbCDwDzkWq6mLPohzgWs/QEvvK2+8YXQZMV9X/JyKDgTdEpIeqFnlvpKpTgakAKSkpFbbYMI3Prl27SgJBXl4eISEhJet++uknTjzxxKMChbdRo0YxapRTPSUi2wFU9cHi9ao6uaz9VPV94P1y1i0BehzruVRFygnNCRBnADoLBKYmVFQ09AzOqKOlnQw8XYVjZwDezTHKKie9DmcgO1R1ARAK2JVtjsnll19e8n7w4MFHrLvllltqOzk+FxkaRPe2Vk9gak5FgaC/qn5QeqGqfoiTJa7MYqCjiCSKSDBOZXDpgtotwOkAItIVJxBU3obOGC+qWub7sj43FAMTo1m2ZQ95hVZPYI5fRYEgvJr7AaCqhcAk4EtgLU7roNUi8qindybAXcANIvIL8A4wXhvq/1zjM96jN5QeyaEGRnaokwYmRpNXWMTK9PJbRBlTVRVVFu8UkYGqush7oYgMoIpP7ar6OU73eu9l3mWva3CKmoyptvT0dG677TZUteQ9OLmBjIwaabVZ5wxof3iimhTPe2Oqq6JAcDcw0zPIXPFYQyk4TeEu9XG6jKmyJ598suR9SkrKEetKf24oopsE07lVJAtTs5g4zN+pMfVdRWMNLRKRgcBEYLxn8WpgkKrurIW0GVMlV111FQEBZZdW7t27t3YTU4sGJkbzwbJ0Ct1FuAJt+nFTfRVePZ7JZx5S1Ys8rwctCJi6JiUlhYULFx61/JVXXqFfv35+SFHtGJgYzYF8N2u21VRrbtNYVTQfwScicp6IHDXEoYgkeSp9/+Db5BlTuWeeeYYJEyZwww03kJWVxc8//8zgwYP58ssvmT9/vr+T5zPFE9VYM1JzvCrKEdyA04/gVxFZLCKfi8g3npFHXwKWquqrtZJKYypwyimnsHTpUlq1akWHDh0YPXo0jzzyCO+99x7x8fGVH6CeahkVSmJME37aaIHAHJ+K5izerqp/8oy4OBZneIk7gR6qeqaq1tjYKcYcr1mzZvHOO+9w880307p1a959912yshr+DXJg+2gWb8qiqMhaXZvqq1INk6puUtUFqrpcVQ/6OlHGHIszzjiDN998k6+++oq///3vLFy4kD59+pCSksLUqVP9nTyfGpgYTfahAn7fud/fSTH1mDU1MPXexIkT+fTTT0lMTAQgICCAW2+9lR9//JH//e9/fk6dbw1KsnoCc/wsEJh675RTTmHNmjVHLc/KyuLf//537SeoFsU3DyeuWRgLrZ7AHIdKA4Gn5ZAFDFNn3XrrrWUOM717924mTy5z4NAGZWBiNAtTsxrsuErG96pygx+HMynNEyLSxdcJMuZYrV+/nqFDjx4HcciQIaxYscIPKapdAxOj2ZWTR+quA/5OiqmnqjJ43JVAX5xp+KaLyALPhNuRPk+dMVWwf3/5FaUFBQW1mBL/sP4E5nhVtdXQPmAWznSTbYALgGWemcWM8avk5GQ+//zzo5Z/8cUXJCUlVbr/nDlz6Ny5M8nJyQCtS68XkZtEZKWILBeR74vn3haRM0VkqWfdUhEZ7rXPtyLym2ef5Z4JnXwiMaYJMREhLLRAYKqp0qkqPUNGXwskA68DA1V1p4iEA2uAZ32bRGMq9vTTT3Puuecyc+ZM+vfvD8CSJUtYsGABn376aYX7ut1uJk6cyNy5c4mPjyckJCRaRLp5RsYt9raqvggl/x+eAkYCu4DzVHWriPTAGXI9zmu/KzwzlfmUiDAoMdpyBKbaqpIjuAh4WlV7quqTxWMNefoTXOfT1BlTBZ06dWLlypWceuqpbNq0iU2bNnHqqaeyYsUKOnXqVOG+ixYtIjk5maSkJIKDgwGygDHe23hyxMWaAOpZ/rOqbvUsXw2EiUgIfjAwMZqMvYdIy7JuPubYVWXy+oeBbcUfRCQMaOXpZPa1rxJmTFWpKiEhIVx77bUVblPWJDUZGRkkJHjPqEo+Rz7VAyAiE3F61gcDw0uvx3lgWqaqeV7LpomIG2de48fKmnRJRCYAEwDatWtXbvorM6xzSx6R1bz24yYeOLdbtY9jGqeq5AjeA7wnk3d7lhlTJwwbNoxnn32WLVu2HLE8Pz+fb775hmuuuYbXXnvtuL5DVZ/zDLdyD/CA9zoR6Q78E7jRa/EVqtoTZ7yuIcBV5Rx3qqqmqGpKbGxstdPXrkU4F/aL542fNrM9O7faxzGNU1UCgUtV84s/eN4H+y5JxhybOXPmEBgYyGWXXUbbtm3p1q0biYmJdOzYkXfeeYfbb7+d8ePHl7lvXFwcaWlp3ouCgYqmNZsBnF/8QUTigQ+Bq1V1Q/FyVc3w/N0PvA0MrN7ZVd3k0zviLlKem7fe119lGpiqFA1lishoVZ0NICJjcCrJjKkTQkNDueWWW7jlllsoKChg165dhIWF0axZs0r3HTBgAOvWrSM1NZW4uDiAaGC29zYi0lFV13k+ngOs8yxvBnwG3KuqP3ht7wKaqeouzzDu5wJfHfeJViIhOpxLBiQwY/EWJgxNIiG6omnHjTmsKjmCm4D7RWSLiKThZI1vrGQfY/wiKCiINm3aVCkIALhcLqZMmcKIESPo2rUrQJaqrvbMtzHas9kkEVktIstx6gmuKV6O05ruwVLNREOAL0VkBbAcJ4fxcg2dYoVuHZ6MiPDsN+sq39gYD6lqt3QRiQBQ1RyfpqgSKSkpumSJz1vkmUZKRJaqql8mOq6pa/uRT1bz+oLNfHXnqSTGNKmBlJmGoKJru0odykTkHOAW4E4ReVBEHqzJBBpjas7Np3UgODCA/3z1u7+TYuqJqgw69yLOeEO3AoIzSc0JPk6XMcfswIEDFBU5Ddx+//13Zs+e3SiGmCitZWQo15zUno9/2crvO2yeAlO5quQITlLVq4E9qvoIMBiouJeOMX4wdOhQcnNzycjI4KyzzuKNN94ot7VQQ3fj0CSaBLt4eq7lCkzlqhIIihslHxSRtkABznhDxtQpqkp4eDgffPABt9xyC++99x6rV6/2d7L8onmTYK47JZEvVm1nVUa2v5Nj6riqBIJPPM3kngSWAZtw2kUbU6eoKgsWLOCtt97inHPOAZyxhBqr64Yk0jQsiKcsV2AqUWEg8ExI87Wq7lXV93HqBrqoqlUWmzrn3//+N//4xz+44IIL6N69Oxs3bmTYsGH+TpbfRIUGMWFoEt/8upOlm/f4OzmmDqu0+aiI/KyqfWspPZWy5qOmKoqKisjJySEqKuqY9msIzUe9HcgrZOgT8+jaJoo3rx9Uo8c29cvxNh/9WkQukrJG7DKmDrn88svZt28fBw4coEePHnTr1o0nn3zS38nyqyYhLm4+rQPfr9/Fgg27/Z0cU0dVJRDciDPIXJ6I7BOR/SKyr7KdjKlta9asISoqio8++oizzz6b1NRU3njjDX8ny++uPPEEWkWF8NTc32xeY1OmqkxVGamqAaoarKpRns/Hlt82phYUFBRQUFDARx99xOjRowkKCipz6OnGJjQokEnDO7J40x7mr7NhwszRqtKhbGhZr9pInDHH4sYbb6R9+/YcOHCAoUOHsnnz5mOuI2ioxqUkENcsjP/3X8sVmKNVpbL4E6+PoTjD6S5V1bIm5/A5qyw2x6KwsBCXqyqD7DoaWmWxt5lL0vjTrBVMvao/Z3U/ampm08AdV2Wxqp7n9ToT6AFYWzRT52RnZ3PnnXeSkpJCSkoKd911FwcOHPB3suqMC/vGkRjThKfm/s6h/Mbbv8IcrUqDzpWSDnSt6YQYc7z+8Ic/EBkZycyZM5k5cyZRUVEVTl/Z2LgCA7hnZGd+3b6fEf+ez4/rrb7AOCrNM4vIs3gm68YJHH1wehhXSkRGAv8BAoFXVPXxUuufBop7/IQDLVW1WVWObUxpGzZs4P333y/5/NBDD9GnTx//JagOGtmjDTMmnMh9H6zk8lcWcumABO4b1ZWmYUH+Tprxo6rkCJYASz2vBcA9qnplZTuJSCDwHHA20A24TESOmFVbVe9Q1T6q2gd4Fvjg2JJvzGFhYWF8//33JZ9/+OEHwsLC/JiiuunEpBZ8MXkIN53agfeWpnPmU//jy9Xb/Z0s40dVqUWbBeSqqhucG7yIhKvqwUr2GwisV9WNnv1mAGOANeVsfxnwUNWSbczRXnzxRa6++mqys51B1po3b16lSevnzJnD5MmTi8clOqoWVURuAiYCbiAHmKCqazzr7gOu86y7TVW/9CyvMDfsb6FBgdx7dhfO6dmGP72/ghvfWMo5Pdvw8OjuxEaG+Dt5prapaoUv4CcgwutzBPBjFfa7GOc/QPHnq4Ap5Wx7ArANCCxn/QScnMmSdu3aqTEVyc7O1uzsbFVVffrppyvctrCwUJOSknTDhg2al5enwEGgmx55/UV5vR8NzPG87wb8gjM1ZSKwAefGH+h5nwQEe7Y54phlvfr37++T36My+YVunfLNOu14/+fa6+EvddaSNC0qKvJLWozvAEu0nGuvKkVDoeo1PaXnfU3Pin0pMEs9uY7SVHWqqqaoakpsbGwNf7VpaKKiokr6Dzz11FMVbrto0SKSk5NJSkoiODgYIAsn51pCVb170jfhcJ3ZGGCGquapaiqwHicnXJIbVtV8YEbpY9YlQYEBTByWzOeTh5DcMoK73vuFa6YtJn1PZZl+01BUJRAcEJF+xR9EpD9wqAr7ZQAJXp/jPcvKcinwThWOacwx0Ur6yWRkZJCQ4H2Zkg/Eld5ORCaKyAbgCeA2z+I4IM1rs3TPsvKWH0VEJojIEhFZkpmZWfHJ+Fhyywjeu3Ewj4zuzpJNWZz19Hze+GkzRUXWAa2hq0oguB14T0S+E5HvgXeBSVXYbzHQUUQSRSQY52Y/u/RGItIFaI5TEW1MjaqpISZU9TlV7QDcAzxQIwel7uV2AwKEa05qz3/vGEr/E5rzl49WcfkrP7Flt+UOGrJKK4tVdbHnZt3Zs+g3Va10IlhVLRSRScCXOGWmr6rqahF5FKesqjgoXIqTvbbHDlMtkZGRZd7wVZVDhyrOvMbFxZGW5v3wTjDl51zBKeZ5wfO+olxvVXPDdVJ883Be/8NA3l2cxt8+W8uIf8/nnpGduXpwewICbPymhqYqQ0xMBN5S1b2ez82By1T1ed8n72g2xISpSYWFhXTq1Imvv/6auLg4QkJCDgEDVLVkjksR6aiq6zzvzwMeUtUUEemOM1vfQKAt8DXQERDgd+B0nACwGLjc+5hlqavX9ta9h7j/w5V8+1smA9tH88+Le5EY08TfyTLH6HjnI7ihOAgAqOoe4IYaSpsxfuVyuZgyZQojRoyga9euAFnFOVcRGe3ZbJKIrBaR5cCdwDUAnhv7TJwm0XOAiarqVtVCnOLTL4G1wMzKgkBd1rZZGNPGD+DJi3uxdvs+zv7PfF75biNuqztoMKqSI1gJ9CouuvF0FFuhqt1rIX1HqatPTaZhaMiDztWEHftyuf+DlXz96076tWvGk2N70yE2wt/JMlVwvDmCOcC7InK6iJyO07pnTk0m0BhTP7SKCuWVa1J4elxvNmQe4Oz/fMcr3220oa3ruar0LL4Hp0PXzZ7Pc4GXfZYiY0ydJiJc0Deek5Nj+POHq3jss7Ws2bqPv1/Yk9CgQH8nz1RDVYahLlLVF1X1YlW9GKc89FnfJ80YU5e1jAxl6lX9uevMTnzwcwaXv/wTmfvz/J0sUw1VGoZaRPqKyBMisgl4FPjVp6kyxtQLIsKtp3fkhSv6sWbbPs5/7gfWbrMpzeubcgOBiHQSkYdE5FecHEAaTuXyMFW1HIExpsTZPdsw66aTcBcpF73wI3PX7PB3kswxqChH8CswHDhXVU/x3PxtWiNjTJl6xDXl40kn07FlBBPeWMKL/9tglcj1REWB4EKcEUHnicjLnhZD1qXQGFOuVlGhvHvjYM7p2YbHv/iVP763grxCe36s68oNBKr6kapeCnQB5uGMOdRSRF4QkbNqKX3GmHomNCiQZy/ryx1ndOL9Zelc8fJCduVYJXJdVpVWQwdU9W1VPQ9nzJSfcZqUGmNMmUSEyWd05LnL+7FqazZjpvzAr9utErmuOqbJ61V1j2e0xNN9lSBjTMNxTq82zLxxMIVFRVz4/I/MWbXN30kyZTimQGCMMceqV3wzZk86hU6tIrnpzWU89d/fbI6DOsYCgTHG51pFhTJjwomM7R/PM9+sZ8IbS9mfW+lo9qaWWCAwxtSK0KBAnri4F4+M7s6833ZywfM/krrrgL+TZbBAYIypRSLODGhvXDeQ3Tl5jJ7yPd/+ttPfyWr0LBAYY2rdSR1imD3pFOKbh3Pt9MW88K11PvMnCwTGGL9IiA7n/Zudzmf/nPMrt81YzqF863zmDxYITKM3Z84cOnfuTHJyMkDr0utF5E4RWSMiK0TkaxE5wbN8mIgs93rlisj5nnXTRSTVa12f2jyn+iI82MWzl/XlnpFd+HTFVi564UfS9xz0d7IaHQsEplFzu91MnDiRL774gjVr1gBEi0i3Upv9DKSoai9gFvAEgKrOU9U+qtoHZ1yug8B/vfa7u3i9qi739bnUVyLCzad14NXxA0jbc5CrX11ETl6hv5PVqFggMI3aokWLSE5OJikpieDgYIAsYIz3Np4bfvFj6k84PexLuxj4wms7c4yGdW7J1KtS2LTrAPe8v8LqDGqRBQLTqGVkZJCQkOC9KB+Iq2CX64Avylh+Kc40rt7+5ilOelpEQso6mIhMEJElIrIkMzPzWJLeIA3u0II/jezCZyu2Me2HTf5OTqNhgcCYKhKRK4EU4MlSy9sAPYEvvRbfhzNg4wAgmnLG5/IM2ZKiqimxsbE+SXd9c+PQJM7s1oq/f76WpZuz/J2cRsECgWnU4uLiSEtL814UDGSU3k5EzgD+DIxW1dJDaV4CfKiqJV1lVXWbOvKAacDAGk98AyUi/Gtsb+KahzHxrZ9t5NJaYIHANGoDBgxg3bp1pKamkp+fD87T+2zvbUSkL/ASThAoq/fTZZQqFvLkEhARAc4HVtV86huupmFBPH9FP/YczGfyjJ9x29hEPmWBwDRqLpeLKVOmMGLECLp27QqQpaqrReRRERnt2exJIAJ4z9MUtCRQiEh7IAH4X6lDvyUiK4GVQAzwmK/PpaHp3rYpfz2/Bz+s383Tc3/3d3IaNJe/E2CMv40aNYpRo0YBICLbAVT1weL1qnpGefuq6ibKqFxW1eE1ntBG6JKUBJZt3sOUeevpd0Izhndp5e8kNUiWIzDG1GkPj+5O97ZR3D5jOWlZ1jrXFywQGGPqtNCgQF64oj8AN7+1lNwCG4aiplkgMMbUee1ahPPUJX1YlbGPRz5Z7e/kNDgWCIwx9cIZ3Vpxy2kdeGdRGrOWpldpH+udXDVWWWyMqTfuPLMTy9P28ucPV/L7jv0cyndzIK+Q/XmFHMgrJMfzOpBXSE5uIQVFyqRhydw6PBmnJa8piwUCY0y94QoM4JnL+jLupQVM+yGViBAXTUJcRHhezcODSYgOJyLYWb4l6wBPzf2dA3mF3Ht2FwsG5bBAYIypV2IiQvjqzlOrdFMvKlIemr2al+Zv5GC+m0dGdycgwIJBaT6tIxCRkSLym4isF5F7y9nmEs9Y76tF5G1fpscY0zBU9ck+IEB4dEx3bhyaxBs/beZP76+wXspl8FmOQEQCgeeAM4F0YLGIzFbVNV7bdMQZnOtkVd0jIi19lR5jTOMkItx7dhfCg108/dXvHCpw8+9xfQgKtLYyxXxZNDQQWK+qGwFEZAbOOO9rvLa5AXhOVfcAlDOOizHGHBcRYfIZHQkPDuRvn68lr8DNlMv7ERoU6O+k1Qm+DIlxgPewjukc3RW/E9BJRH4QkZ9EZGRZB7Ix240xNeGGoUn89fwefLV2J9e/toSD+TYTGvi/H4EL6AichjOC48si0qz0RjZmuzGmplx14gn8a2xvftywi2teXcT+3ILKd2rgfBkIMnBGZSwWz9HjvKcDs1W1QFVTgd9xAoMxxvjMxf3jefayfvy8ZS9XvLKQvQfz/Z0kv/JlIFgMdBSRRBEJxpnKb3apbT7CyQ0gIjE4RUUbfZgmY4wB4JxebXjpqv78un0/l079iR37cv2dJL/xWSBQ1UJgEs70fWuBmWWM8/4lsFtE1gDzgLtVdbev0mSMMd5O79qKaeMHkJZ1kPOe/Z5lW/b4O0l+IfVtLI6UlBRdsmSJv5NxzAoKCkhPTyc3t/E+ddQloaGhxMfHExQUdMRyEVmqqin+SFN9vbYbgt+27+f61xezIzuPv1/Yk4v7x/s7STWuomvbehbXkvT0dCIjI2nfvr11c/czVWX37t2kp6eTmJjInDlzmDx5Mm63G6B16e1F5E7geqAQyAT+oKqbPevcOLOQAWxR1dGe5YnADKAFsBS4SlUbd0F0Hda5dSSzJ57CpHeW8cf3fmHN1n3cP6oLrkbS16BxnGUdkJubS4sWLSwI1AEiQosWLcjNzcXtdjNx4kS++OIL1qxZAxAtIt1K7fIzkKKqvYBZwBNe6w6pah/Pa7TX8n8CT6tqMrAHuM6Hp2RqQPMmwbx27UCuPbk9r/6QyvhpixtNJbIFglpkQaDuKP63WLRoEcnJySQlJREcHAyQhdPxsYSqzlPV4qmxfsJpAVfRsQUYjhM0AF7DmcDe1HGuwAAeOq87T1zci0WpWYx57gd+37Hf38nyOQsEplHLyMggIcG7lTP5lDEHsZfrgC+8Pod6Ojv+JCLne5a1APZ6GkxA2Z0pTR12SUoC70w4kYP5bi547gf+u3q7v5PkUxYIjKkiEbkSSAGe9Fp8gqcC7nLg3yLS4RiPab3m66j+JzTnk0mnkNwygglvLOWZr9c12IluLBCYGldYWH+67cfFxZGW5j0SCsEc3fERETkD+DMwWlXziperaobn70bgW6AvsBtoJiLFjTHK6kxZvL/1mq/DWjcN5d0bB3NB3ziemvs7t7y1jN05eZXvWM9YqyE/eOST1azZuq9Gj9mtbRQPnde90u3OP/980tLSyM3NZfLkyUyYMIE5c+Zw//3343a7iYmJ4euvvyYnJ4dbb72VJUuWICI89NBDXHTRRURERJCTkwPArFmz+PTTT5k+fTrjx48nNDSUn3/+mZNPPplLL72UyZMnk5ubS1hYGNOmTaNz58643W7uuece5syZQ0BAADfccAPdu3fnmWee4aOPPgJg7ty5PP/883z44Yc1+huVZcCAAaxbt47U1FTi4uIAoinV8VFE+gIvASO9B0YUkebAQVXN83SIPBl4QlVVROYBF+O0HLoG+NjnJ2N8IjQokKcu6U23NlE8PudXvl+3i0nDkxl/cntCXA1j0DoLBI3Mq6++SnR0NIcOHWLAgAGMGTOGG264gfnz55OYmEhWVhYAf/3rX2natCkrVzotI/fsqbyjTXp6Oj/++COBgYHs27eP7777DpfLxVdffcX999/P+++/z9SpU9m0aRPLly/H5XKRlZVF8+bNueWWW8jMzCQ2NpZp06bxhz/8wae/QzGXy8WUKVMYMWJEcfPRrOKOj8ASVZ2NUxQUAbznqWQubibaFXhJRIpwctePew2zfg8wQ0Qew2l19H+1ckLGJ0SEG4YmMaxLS/7++Vr+8cWvvLlwM/ed3ZWze7Su9w1BLBD4QVWe3H3lmWeeKXnSTktLY+rUqQwdOpTExEQAoqOjAfjqq6+YMWNGyX7Nmzev9Nhjx44lMNB5QsrOzuaaa65h3bp1iAgFBQUlx73ppptwuVxHfN9VV13Fm2++ybXXXsuCBQt4/fXXa+iMKzdq1ChGjRoFgIhsB1DVB4vXq+oZZe2nqj8CPctZtxFnKHbTgCS3jODV8QP4bl0mf/tsLbe8tYwB7Zvzl3O70Su+mb+TV21WR9CIfPvtt3z11VcsWLCAX375hb59+9KnT59jOob3k0/pXtJNmjQpef+Xv/yFYcOGsWrVKj755JNKe1Rfe+21vPnmm7zzzjuMHTu2JFAYUxcN6RjLZ7cN4e8X9CR11wFGT/mBO99dzrbsQ/5OWrVYIGhEsrOzad68OeHh4fz666/89NNP5ObmMn/+fFJTUwFKiobOPPNMnnvuuZJ9i4uGWrVqxdq1aykqKqqwDD87O7u4zJ3p06eXLD/zzDN56aWXSiqUi7+vbdu2tG3blscee4xrr7225k7aGB8JDBAuH9SOeX88jZtP68CnK7cx7F/f8tTc3+vdPAcWCBqRkSNHUlhYSNeuXbn33ns58cQTiY2NZerUqVx44YX07t2bcePGAfDAAw+wZ88eevToQe/evZk3bx4Ajz/+OOeeey4nnXQSbdq0Kfe7/vSnP3HffffRt2/fI1oRXX/99bRr145evXrRu3dv3n778DTVV1xxBQkJCXTt2tVHv4AxNS8yNIh7Rnbh6ztP5YyurXjm63UM/9f/WJSa5e+kVZkNOldL1q5daze4SkyaNIm+ffty3XW1MxpDWf8mNuicOV5LN2fxx/dWsCXrIPeM7MwNQ5LqRGVyRde25QhMndC/f39WrFjBlVde6e+kGHNc+p8QzceTTuasbq34++e/cuMbS8k+VLdnQbNAYOqEpUuXMn/+fEJCQvydFGOOW1RoEM9f0Y+/nNuNb37dyXnPfs+qjGx/J6tcFgiMMcYHRITrTknk3RtPJL+wiAtf+JF3F2+pk8NUWCAwxhgf6n9CNJ/edgoD20dzz/sruXvWCg7lu/2drCNYIDDGGB+LiQjhtT8M5LbTO/L+snQueP4HNmbm+DtZJSwQGGNMLQgMEO48sxPTxg9gx75cRk/5gS9WbvN3sgALBMYYU6tO69yST28bQnLLCG5+axnPzVvv93oDCwSmXBEREf5OgjENUlyzMN698UTG9GnLk1/+xj3vr6DAXeS39NiALv7wxb2wfWXl2x2L1j3h7Mdr9ph1RGFhoY09ZBqcEFcg/x7XhxOiw3nmm/Vk7D3E81f0p2lYUK2nxXIEjci99957xPhBDz/8MI899hinn346/fr1o2fPnnz8cdWGzc/JySl3v9dff71kCImrrroKgB07dnDBBRfQu3dvevfuzY8//simTZvo0aNHyX7/+te/ePjhhwE47bTTuP3220lJSeE///kPn3zyCYMGDaJv376cccYZ7NixoyQd1157LT179qRXr168//77vPrqq9x+++0lx3355Ze54447qvuzGeMzIsKdZ3XmyYt7sXBjFhe/8CPpew5WvmNNU9V69erfv7/WR2vWrPF3EnTZsmU6dOjQks9du3bVLVu2aHZ2tqqqZmZmaocOHbSoqEhVVZs0aVLusQoKCsrcb9WqVdqxY0fNzMxUVdXdu3erquoll1yiTz/9tKqqFhYW6t69ezU1NVW7d+9ecswnn3xSH3roIVVVPfXUU/Xmm28uWZeVlVWSrpdfflnvvPNOVVX905/+pJMnTz5iu/3792tSUpLm5+erqurgwYN1xYoVR51DWf8mOHMQ2LVtat0P6zK1x0NztP9f5+ovaXtq/PgVXduW325E+vbty86dO9m6dSuZmZk0b96c1q1bc8cddzB//nwCAgLIyMhgx44dtG7dusJjqSr333//Uft98803jB07lpiYGODwfAPffPNNyRwDgYGBNG3atNLJbooHwANn0ptx48axbds28vPzS+ZPKG/ehOHDh/Ppp5/StWtXCgoK6NmzzGkDjKkzTkqO4YObT+La6Yu55KUFPHNpX87qXvH/w5piRUONzNixY5k1axbvvvsu48aN46233iIzM5OlS5eyfPlyWrVqVencAUC19/PmcrkoKjpcQVbR/Aa33norkyZNYuXKlbz00kuVftf111/P9OnTmTZtmg1rbeqNjq0i+fCWk+ncOoob31zKq9+n1sr3WiBoZMaNG8eMGTOYNWsWY8eOJTs7m5YtWxIUFMS8efPYvHlzlY5T3n7Dhw/nvffeY/fu3cDh+QZOP/10XnjhBQDcbjfZ2dm0atWKnTt3snv3bvLy8vj0008r/L7i+Q1ee+21kuXlzZswaNAg0tLSePvtt7nssssqPJc5c+bQuXNnkpOTAY56BBORO0VkjYisEJGvReQEz/I+IrJARFZ71o3z2me6iKSKyHLPq0+FiTDGIzYyhBk3nMhZ3Vrx6KdreHj2atxFlTcvVVUKq9nyyIqGGpnu3buzf/9+4uLiaNOmDVdccQXnnXcePXv2JCUlhS5dulTpOOXt1717d/785z9z6qmnEhgYSN++fZk+fTr/+c9/mDBhAv/3f/9HYGAgL7zwAoMHD+bBBx9k4MCBxMXFVfjdDz/8MGPHjqV58+YMHz68ZCKdBx54gIkTJ9KjRw8CAwN56KGHuPDCCwG45JJLWL58eYXTbLrdbiZOnMjcuXOJj48nJCQkWkS66eG5h8GZczhFVQ+KyM3AE8A44CBwtaquE5G2wFIR+VJV93r2u1tVZ1XpBzXGS1hwIM9f0Z/Hv1jLy9+l8uXq7QQFBlDoLqKwSJ1XqfdFCsO7tOTV8QOO+ftsPoJaYvMR1L5zzz2XO+64g9NPP73M9WvXrmXv3r08/PDDfPnllwCISAbwnKr+o6x9RKQvMEVVTy5j3S/AxZ7AMB349FgCQX29to1vzVqazrzfdhIUIAQGBBAUKLgCBVdAAK4AITBQCAoIwBUoJMY0YUyfuDKPU9F8BJYjMA3O3r17GThwIL179y43CBTLyMggISHBe1E+UPb/JMd1wBelF4rIQCAY2OC1+G8i8iDwNXCvquaVsd8EYAJAu3btKkyraZwu7h/Pxf3jffodFghMhVauXFnSF6BYSEgICxcu9FOKKtesWTN+//33Gj+uiFwJpACnllreBngDuEZViwtp7wO24wSHqcA9wKOlj6mqUz3rSUlJqV/Zc9NgWCCoRapaJ6asOxY9e/Zk+fLl/k5GjSsuEo2LiyMtLc17VTCQUXp7ETkD+DNwqveTvYhEAZ8Bf1bVn7yOXzyaWJ6ITAP+WNPnYExNsVZDtSQ0NJTdu3f7fXAp4wSB3bt3ExoayoABA1i3bh2pqank5+cDRAOzvbf31Au8BIxW1Z1ey4OBD4HXS9cFeHIJiBP5zwdW+fKcjDkeliOoJfHx8aSnp5OZmenvpBicwBwfH4/L5WLKlCmMGDECt9sNkKWqq0XkUZyemLOBJ4EI4D1Pjm6Lqo4GLgGGAi1EZLzn0ONVdTnwlojEAgIsB26qxdMz5phYqyFjvFTUssLX7No2vlTRte3ToiERGSkiv4nIehG5t4z140Uk06vTzfW+TI8xxpij+axoSEQCgeeAM4F0YLGIzC7VUQfgXVWd5Kt0GGOMqZgvcwQDgfWqulFV84EZwBgffp8xxphq8GVlcRzg3S4vHRhUxnYXichQ4HfgDlVNK72Bd6cbIEdEfivnO2OAXdVPcp3QEM4B6u95nOCvL166dOkuESlvsKf6+nt6s3Pwr3KvbX+3GvoEeEdV80TkRuA1YHjpjbw73VRERJb4q6KvpjSEc4CGcx61SVVjy1vXEH5PO4e6y5dFQxmAd9/9eEp11FHV3V6dc14B+vswPcYYY8rgy0CwGOgoIomejjeXcnRHnTZeH0cDa32YHmOMMWXwWdGQqhaKyCTgSyAQeLWMjjq3ichooBDIAsYf59dWWnxUDzSEc4CGcx51RUP4Pe0c6qh616HMGGNMzbKxhowxppGzQGCMMY1cgwgElQ1lUV+IyCYRWekZbqPeDDojIq+KyE4RWeW1LFpE5orIOs/f8ueLNOWya9t/GtN1Xe8DgddQFmcD3YDLRKSbf1N1XIapap961lZ5OjCy1LJ7ga9VtSOeGbpqO1H1nV3bfjedRnJd1/tAgA1l4XeqOh+n1Ze3MTgdBPH8Pb8209RA2LXtR43pum4IgaCsoSwqmnO2LlPgvyKy1DOsRn3WymuWru1AK38mpp6ya7vuaZDXtb+HmDBHOkVVM0SkJTBXRH71PJXUa6qqImLtlBu3BndtN6TruiHkCCodyqK+UNUMz9+dOFMgDvRvio7LDq/pGtsAOyvZ3hzNru26p0Fe1w0hEFQ6lEV9ICJNRCSy+D1wFvV7ntvZwDWe99cAH/sxLfWVXdt1T4O8rut90VB5Q1n4OVnV0Qr40DMnrgt4W1Xn+DdJVSMi7wCnATEikg48BDwOzBSR64DNOPP7mmNg17Z/Nabr2oaYMMaYRq4hFA0ZY4w5DhYIjDGmkbNAYIwxjZwFAmOMaeQsEBhjTCNngaCeEBG3Z+TG4leNDXYlIu29R1g0pjbZte1/9b4fQSNySFX7+DsRxviAXdt+ZjmCes4zzvsTnrHeF4lIsmd5exH5RkRWiMjXItLOs7yViHwoIr94Xid5DhUoIi+LyGoR+a+IhPntpIzBru3aZIGg/ggrlX0e57UuW1V7AlOAf3uWPQu8pqq9gLeAZzzLnwH+p6q9gX5AcU/VjsBzqtod2Atc5NOzMeYwu7b9zHoW1xMikqOqEWUs3wQMV9WNIhIEbFfVFiKyC2ijqgWe5dtUNUZEMoF4Vc3zOkZ7YK5nsg1E5B4gSFUfq4VTM42cXdv+ZzmChkHLeX8s8rzeu7H6I1M32LVdCywQNAzjvP4u8Lz/EWe0SoArgO88778GbgZnKkQRaVpbiTSmGuzargUWGeuPMBFZ7vV5jqoWN7NrLiIrcJ58LvMsuxWYJiJ3A5nAtZ7lk4GpntET3Tj/cbZhjP/Yte1nVkdQz3nKUVNUdZe/02JMTbJru/ZY0ZAxxjRyliMwxphGznIExhjTyFkgMMaYRs4CgTHGNHIWCIwxppGzQGCMMY3c/wcFrSARuC78bQAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "## Teste\n",
    "# Generate the model:\n",
    "model = make_model()\n",
    "model.compile(\n",
    "   optimizer=\"adam\",\n",
    "   loss=tf.keras.losses.CategoricalCrossentropy(),\n",
    "   metrics=[tf.keras.metrics.CategoricalAccuracy()],\n",
    ")\n",
    "\n",
    "# # baseline\n",
    "# history = model.fit(\n",
    "#     train_x,\n",
    "#     train_y_cat,\n",
    "#     epochs=20,\n",
    "#     batch_size=64,\n",
    "#     validation_data=(test_x, test_y_cat),\n",
    "#     callbacks=[earlystop, ln_callback],\n",
    "#     class_weight=class_weights,\n",
    "# )\n",
    "# # Evaluate Base Line\n",
    "# y_hat_test_cat = model.predict(test_x)\n",
    "# y_hat_test = np.argmax(y_hat_test_cat, axis=-1)\n",
    "# bacc = balanced_accuracy_score(test_y, y_hat_test)\n",
    "# print(f\"BACC base_= {bacc:.3f}\")\n",
    "\n",
    "# augmented african class\n",
    "history = model.fit(\n",
    "    train_x_plus_africans_augmented,\n",
    "    train_y_cat_plus_africans,\n",
    "    epochs=40,\n",
    "    batch_size=64,\n",
    "    validation_data=(test_x, test_y_cat),\n",
    "    callbacks=[earlystop, ln_callback],\n",
    "    #callbacks=[earlystop],\n",
    "    class_weight=class_weights\n",
    ")\n",
    "\n",
    "\n",
    "# Evaluate the model\n",
    "fig, (ax1, ax2) = plt.subplots(1, 2)\n",
    "ax1.plot(history.history[\"categorical_accuracy\"], label=\"accuracy\")\n",
    "ax1.plot(history.history[\"val_categorical_accuracy\"], label=\"val_accuracy\")\n",
    "ax1.set_xlabel(\"Epoch\")\n",
    "ax1.set_ylabel(\"Accuracy (Cat)\")\n",
    "ax1.set_ylim([0.5, 1])\n",
    "ax1.legend(loc=\"lower right\")\n",
    "\n",
    "ax2.plot(history.history[\"loss\"], label=\"loss\")\n",
    "ax2.plot(history.history[\"val_loss\"], label=\"val_loss\")\n",
    "ax2.set_xlabel(\"Epoch\")\n",
    "ax2.set_ylabel(\"Loss (CXE)\")\n",
    "ax2.legend(loc=\"upper right\")\n",
    "\n",
    "# test_loss, test_acc = model.evaluate(test_x,  test_y, verbose=2)\n",
    "# print(f\"Test Loss: {test_loss:.3f} Test Accuracy {test_acc:.3f}\")\n",
    "\n",
    "y_hat_test_cat = model.predict(test_x)\n",
    "y_hat_test = np.argmax(y_hat_test_cat, axis=-1)\n",
    "# test_loss, test_acc = mlp.evaluate(test_x,  test_y, verbose=2)\n",
    "\n",
    "ConfusionMatrixDisplay.from_predictions(\n",
    "    test_y, y_hat_test, display_labels=[\"Caucasian\", \"African\", \"Asian\", \"Indian\"]\n",
    ")\n",
    "\n",
    "bacc = balanced_accuracy_score(test_y, y_hat_test)\n",
    "\n",
    "print(f\"BACC = {bacc:.3f}\")\n",
    "print(\"Y_hat shape:\", y_hat_test.shape)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Save / Load models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Save the entire model as a SavedModel.\n",
    "# !mkdir -p saved_models\n",
    "from time import time\n",
    "NAME = \"c2_12k5x5_64k5x5_d2_64_wc_africans_aug\"\n",
    "#NAME = \"teste\"\n",
    "\n",
    "model.save(f\"saved_models/{NAME}_{int(time())}.h5\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "loaded_model = tf.keras.models.load_model(\"saved_models/c2_12k5x5_64k5x5_d2_64.h5\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Predição FINAL"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# NEVER CHANGE\n",
    "Xtest_Classification_Part2 = np.load(\"./data/Xtest_Classification_Part2.npy\")\n",
    "Xtest_Classification_Part2_50_50 = np.reshape(Xtest_Classification_Part2, (len(Xtest_Classification_Part2),50, 50, 1))\n",
    "\n",
    "print(\"Test input shape\", Xtest_Classification_Part2_50_50.shape)\n",
    "y_hat_test_cat = model.predict(Xtest_Classification_Part2_50_50)\n",
    "y_hat_test = np.argmax(y_hat_test_cat, axis=-1)\n",
    "print(\"Test output shape\", y_hat_test.shape)\n",
    "\n",
    "# Save\n",
    "np.save(f\"y_hat_classification_prob2_{int(time())}.npy\", y_hat_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load and print shape\n",
    "load_y_hat = np.load(f\"y_hat_classification_prob2_111111111.npy\")\n",
    "print(\"Loaded array shape\", load_y_hat.shape)\n",
    "print(\"Loaded Bin count\", np.bincount(load_y_hat))"
   ]
  }
 ],
 "metadata": {
  "interpreter": {
   "hash": "e7370f93d1d0cde622a1f8e1c04877d8463912d04d973331ad4851f04de6915a"
  },
  "kernelspec": {
   "display_name": "Python 3.9.7 64-bit",
   "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": "3.9.7"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}