letter_digit_generator_v1.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# VAE model building"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Import module"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.layers import Lambda,Input,Dense,Flatten,Conv2D,Conv2DTranspose\n",
    "from keras.layers import Activation,BatchNormalization,Reshape,Concatenate,Dropout\n",
    "from keras.models import Model\n",
    "from keras.losses import mse, binary_crossentropy\n",
    "from keras.callbacks import Callback,ModelCheckpoint,ReduceLROnPlateau\n",
    "from keras.utils import to_categorical\n",
    "from keras import backend as K\n",
    "from keras.optimizers import Adam\n",
    "from keras import initializers\n",
    "import seaborn as sns\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "import os\n",
    "import cv2\n",
    "from mnist import MNIST"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train shape: (697932, 784)\n",
      "X_test shape: (116323, 784)\n",
      "y_train shape: (697932,)\n",
      "y_test shape: (116323,)\n"
     ]
    }
   ],
   "source": [
    "emndata = MNIST('emnist_data')\n",
    "img_rows = 28\n",
    "img_cols = 28\n",
    "img_channel = 1\n",
    "orig_dimension = img_rows*img_cols\n",
    "\n",
    "X_train,y_train = emndata.load('emnist_data/emnist-byclass-train-images-idx3-ubyte',\n",
    "                               'emnist_data/emnist-byclass-train-labels-idx1-ubyte')\n",
    "X_test,y_test = emndata.load('emnist_data/emnist-byclass-test-images-idx3-ubyte',\n",
    "                             'emnist_data/emnist-byclass-test-labels-idx1-ubyte')\n",
    "X_train = np.array(X_train)\n",
    "# X_train = X_train.reshape(X_train.shape[0],img_rows,img_cols)\n",
    "y_train = np.array(y_train)\n",
    "X_test = np.array(X_test)\n",
    "# X_test = X_test.reshape(X_test.shape[0],img_rows,img_cols)\n",
    "y_test = np.array(y_test)\n",
    "\n",
    "\n",
    "print('X_train shape: {}'.format(X_train.shape))\n",
    "print('X_test shape: {}'.format(X_test.shape))\n",
    "print('y_train shape: {}'.format(y_train.shape))\n",
    "print('y_test shape: {}'.format(y_test.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# for i in range(X_train.shape[0]):\n",
    "#     X_train[i]=np.transpose(X_train[i])\n",
    "# for i in range(X_test.shape[0]):\n",
    "#     X_test[i]=np.transpose(X_test[i])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "By Class: This represents the most useful organization\n",
    "from a classification perspective as it contains the segmented digits and characters arranged by class. There are\n",
    "62 classes comprising [0-9], [a-z] and [A-Z]. The data is\n",
    "also split into a suggested training and testing set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_str = '0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz'\n",
    "lebel_dict = {i:v for i,v in enumerate(label_str)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA8QAAANWCAYAAAA4AkW/AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjAsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+17YcXAAAgAElEQVR4nOzdeZRV5Zn+/XtbzPM8igUImgACaieKyhAFEVEBMSo/56TTxkYTk1/UrGiiqOluh7TYS+NK+u1ObKdMDQhGjDghIFEjgoAyCFIoMs/zUJz3D/ENL9dVuqGqzqlifz9rZSkXZ3iA85y972z3RZLL5QIAAAAAgKw5ptALAAAAAACgEBiIAQAAAACZxEAMAAAAAMgkBmIAAAAAQCYxEAMAAAAAMomBGAAAAACQSQzEAAAAAIBMYiAGACDPkiQZlCTJH5IkKUmSZFeSJDuTJFmaJMlTSZL0L/T6AADIiiSXyxV6DQAAZEKSJElEPBYR1x8U74qIXETUPSh7KJfL/TCfawMAIIu4QgwAQP5cG38fhv8UESfkcrm6uVyuXkR8JSKePfBzP0iSZEQB1gcAQKZwhRgAgDxJkuTViBgQER9GxFdzudy+Q36+ZkQsiIjOEfG7XC43Ku+LBAAgQ7hCDABA/rQ98M85hw7DERG5XG5vRMw+8MMGeVsVAAAZxUAMAED+LD3wz15JktQ49CcPXCHufeCHf8vbqgAAyCgGYgAA8uexA//sEhHPJEnS5fOfSJLkxIj4Q3z2n0sviYiH8r88AACyhXuIAQDIoyRJbo6I+yKi1oFo54F/1o2ITRHxTETckcvlNhRgeQAAZApXiAEAyKNcLjc2Ii6OiDUHorrx979yqXZENIyIxgVYGgAAmcNADABAniRJUi9Jkt9HxHMRsTwizo2IFhHR8sC/z4+IKyPirSRJehZsoQAAZAT/yTQAAHmSJMmjEfHPEbEoInrncrmdh/x83fisZfqEiJiey+X65n+VAABkB1eIAQDIgyRJGkbEPx344SOHDsMREQeyRw788KwkSVrla30AAGQRAzEAAPlxQkR8/lctLfmCxy0+6N87Vd5yAAAAAzEAAPmx/6B/L/6Cx7U+6N+3VtJaAABAMBADAJAvC+Lvf8XSPyZJUuPQByRJUhR//8+qN0bEwjytDQCATGIgBgAgDw7cH/z/HPjhKRExKUmSk5IkOebA/3pGxPMRccaBx4zN5XKlhVgrAABZQcs0AAB5cqBFelxEnHdQvPvAP2sflD0TEVcxEAMAULkYiAEAyKMkSZKIGBmf/X3Dp0ZEq4jIRcSaiHgrIn6Ty+X+XLgVAgCQHQzEAAAAAIBM4h5iAAAAAEAmMRADAAAAADKJgRgAAAAAkEkMxAAAAACATKrxRT+ZJAmNWzgsuVwuKfQaCoG9gsOVxb3CPsHhYp8AXy6L+ySCvYLDV9Ze4QoxAAAAACCTGIgBAAAAAJnEQAwAAAAAyCQGYgAAAABAJn1hqRYAVFdFRUWS5XLav7F///58LAcAAKDKO+YYvV7qMmffvn0VvZy84AoxAAAAACCTGIgBAAAAAJnEQAwAAAAAyCQGYgAAAABAJlGq9SUO58bymjVrSta6dWvJtm/fLtmmTZskKy0tlYwCIBwNWrZsKVnTpk0l6969u2SNGjWSrGPHjpINGTJEsmeffVayhx9+WLIdO3ZIBuRbjRp6iHb7pF69epJ9+umnku3du7diFoZMS3te1KBBA8l27dqVKjscbp847vyJcyocjcqaU1yeJIlkZ511VqrMeeKJJyQrKSmRzJWcFhJXiAEAAAAAmcRADAAAAADIJAZiAAAAAEAmMRADAAAAADKJUq2D1KlTR7JBgwZJ1qVLF/v89u3bS9avXz/Jli9fLtncuXMle++99ySbNm2aZOvWrbPrAfLJFWV16NDBPvbHP/6xZD179pTMFQi58rr69etL5opWWrRoIdm8efMkmzRpkmTA4SoqKpKsefPmkrVp08Y+3x1/+vTpI1mTJk0ku/feeyX75JNPJNuwYYNkW7dulYxCrmxKe150wgknSNajRw/Jli5dKtlTTz0lmfusHnvssXaNw4YNk8wVernv+unTp0u2ZcsWyXbv3m3fG8gnVyA6cuRIydxxIiKiW7dukrm91rt3b8nc+ZNz9dVXS3bXXXdJNm7cOMl27tyZ6j0qA1eIAQAAAACZxEAMAAAAAMgkBmIAAAAAQCYxEAMAAAAAMinJ5XJl/2SSlP2TVZQrMWnWrJlkrqznjDPOkOwnP/mJZG3btrXvXbt2bclcsc/+/fslKy0tlWz9+vWSPfroo5I99NBDku3YscOusbLlcrmkIG9cYNVxrzhp948ry3JFWaeccop9n+Li4lTvXdFcMZArH7rnnnsqfS1Z3CtHyz5JEv2jc6Vyffv2lezyyy+XrGvXrvZ9jj/+eMm2bdsmmSsBcvbs2SPZL3/5S8mef/55yUpKSlK9R0Vjn+TPMcfoNZIhQ4ZI9otf/EIyVyrqCrl27dol2SuvvCLZnDlzJOvVq5dkERFnn322ZLVq1ZLMnVO5Qi9XvvXOO+9ItnjxYskKVb6VxX0ScfQcU5x69epJ9qtf/Uqy4cOHS1a3bl37mm6PuznQHeOcL5ohD+bK9H72s59J5oq2KnpPlbVXuEIMAAAAAMgkBmIAAAAAQCYxEAMAAAAAMomBGAAAAACQSdr4VI24G84HDRok2RVXXCFZz549JXPlQc2bN5cs7c3mZXE3tbusTZs2kn3rW9+SbNmyZZL96U9/kqxQZQ+omlzpgivYcaUqbq+4ArmyuGI5V7biSrAaNmwomduTruxh06ZNkrkCFeBzNWvWlKxbt26S3XzzzZKl3Tv79u2z771w4ULJHn74YclWrlwp2SOPPJLqcQsWLJBs69atdj3Inu3bt0v2ySefSObOldw5x4YNGyRz+2nAgAGSueLSiPTnVC1atJDsuuuuk8ztx08//VSyBx54QLJ8lALh6OM+m9dff71kI0eOlMyV15WlPPPLxo0bJXvvvfckO/300yXr3LmzZD/4wQ8kmzlzpmRuxqkMXCEGAAAAAGQSAzEAAAAAIJMYiAEAAAAAmcRADAAAAADIpGpTqpX2hnNXkFBcXCxZUVFRqvd1xTxlFSS4whJXCuSKIdq1ayeZKynq2LGjZDfddJNkM2bMkCxfN6aj6nGfueHDh0t26aWXSubKtxy3V9avX28f+/rrr0vmyhl69eol2YgRI1K9d2lpqWQlJSWSzZkzx64R2eOKgXr06CHZvffeK9k//MM/SOYKTObPny/Z7Nmz7XrGjh0r2UcffSTZySefnOpxjz76qGTTpk2TrKySLxzdXNnh22+/LdncuXMlc+dZ06dPl+yhhx6SzBXX9evXT7JGjRpJFuELtI477jjJ1q1bl+q5Xbt2lWzgwIGSVbVSIFQPrgRr6NChkt14442SuXO5LVu2SOZKhyP8XOEKTd1ecaVad9xxh2Tf/e53Jbvkkkskc3OdyyjVAgAAAACgEjEQAwAAAAAyiYEYAAAAAJBJDMQAAAAAgEyqkqVaruSgb9++ko0ePVqyNm3apHoPV7izYcMGyVwJjytFifCFJdu2bZPsggsukOz222+XrHHjxpK50ovly5dLtn37drtGHP1ckc/ll18u2X333SeZK9BynzlXuPDLX/5SsieffNKuce3atZL98Ic/lOzcc8+1zz+U28833HCDZBMnTky1Fhz9XLHiqFGjJPve974nmSvs2bp1q2Tz5s2TzJWQfPjhh3aNmzdvlsyVyo0cOVKy//zP/5TsL3/5i2QUaOGLtGzZUrIzzzxTMnfuddJJJ0nmzolcAdzhlB26Y179+vUlc4WortDrq1/9qmSuXM+do7nSImSXK9C69dZbJbvlllskc5/hp556SrKnn35asocfftiuxxXznn/++ZLNmjVLsoYNG0q2Zs0ayRYvXiyZO7ccNmyYZO641adPH8lcEVh5cYUYAAAAAJBJDMQAAAAAgExiIAYAAAAAZBIDMQAAAAAgk6rk3f+uVMsVGjRv3jzV67mbr125iLtZ3d1YvmXLFvs+rpSrVatWkp122mmSNWjQwL7moVzJl1u3Wwuywe0LV9rmylJcwc7UqVMl+9WvfiXZc889J9mePXvsGl150dVXXy2ZK5XI5XKSuVIWV6DlCiCQTa6Ix+0dV0Linvu73/1OMlds4r7DXSlchC8luvbaayVzxwC3HyujiARHN1fQ6Yo8i4uLJevQoYNkp5xyimRuT5S37G3Tpk2SuSK99u3bS+ZKtdw52saNG49wdciKE044QbKLL75Ysnr16knm9t4vfvELydz3ujtPivB715U/ujmnrNnnUG5fuDLiiy66SDL3++UK+5YtW5ZqLYeDK8QAAAAAgExiIAYAAAAAZBIDMQAAAAAgkxiIAQAAAACZVCVLtRxXtOWKTVw5iSvQuuWWWyRzxTxllZ04rnzl29/+tmT9+/eXzJU9uFKJ8ePHS/bSSy9JdjjrxtGlSZMmkrnPpiu8+u53vyvZn//8Z8nWrVsnWe3atSW77LLL7Bp//etfS1a3bl3JXFnEfffdJ9njjz8uGQVayCf3fe32ifturlmzpn3NXr16SeaOe3PmzJGMAi1UBFeQM3PmTMncMcZ9ft0x4fXXX5fscL6/3Z5wpZF9+/aV7Morr5SsZ8+ekrnjqjtnLG8ZGI4uI0aMkKxr166SufOxCRMmSLZw4ULJzjvvPMnc+VhExG9+8xvJ1q9fbx97pFyh1/79+1M91x0La9TIz6jKFWIAAAAAQCYxEAMAAAAAMomBGAAAAACQSQzEAAAAAIBMqjalWu6GbHfj9qZNmyR75plnJEtboOVu8G7atKldoyvQ+v73vy9ZgwYN7PMPtXz5cslc+cTOnTtTvR5wsBUrVkj2pz/9SbLt27ener2vfvWrkt166632sa5Ay32Ox40bJ9kDDzwg2Y4dO9IsEfhC7jjjjguusGTYsGGSbd68WbLnn39esvr169v13HDDDZJNnjxZMle04o6PwOFyJVGvvvqqZG3atJHMlVO5YqszzzxTsokTJ0pW1rmXK/QaPXq0ZK58yJVluV+zK5d0BUUrV660a8TRz5U/XX755ZK585/FixdLdu+990q2d+9eyXr06CHZli1b7BrdPFTRxwo3NzVs2LBC36MycIUYAAAAAJBJDMQAAAAAgExiIAYAAAAAZBIDMQAAAAAgk6pkqZYrNpk3b55krkArSRLJXGFJrVq1JNu1a5dk3bp1k+yCCy6QLCLiO9/5jmRpC7RcodCTTz4p2csvvyyZ+/0CDuY+I0uXLpVs9+7dkrkCoRNPPFGyu+66S7KvfOUrdj2uqGvChAmpXpMCLVQEtyemT58umSuxGjRokGTFxcWSuVK5q6++WjJ33IqI2LNnj2SulMsdu4DKkrbwc8SIEZJ16NBBsv79+0vmyqmuv/56u54zzjhDMldc5Iok3XnWqlWrJBs/frxkJSUlkrkSPmTXMceku+6YttDRPW7WrFmSlTV7bNiwIdV6yuOEE06QbODAgZK5EjI3C7mSu8rAFWIAAAAAQCYxEAMAAAAAMomBGAAAAACQSQzEAAAAAIBMqjalWvPnz5ds48aNknXt2lWyn/zkJ5LlcjnJ3n//fcluuukmyVyBQ0REq1atbH4oV1zkbor/05/+JBmFQvgyroBg7969kp111lmSPfDAA5J17txZsrPPPlsyV17nyrMiIs477zzJ/va3v0nm9gpQEdxxxhUDueKcq666SrJrrrlGso4dO6bKyrJ27VrJXCmRO35QtojK4op5XnvtNckmTZok2ejRoyW78sorJbvwwgslcyU8ERFvvfWWZL///e8lc/vElXe542W+in1w9HPfzR988IFkmzdvTvXcKVOmSOYKIiMitmzZkmaJlitYbd++vWQ/+tGPJHNFrG4OW7RokWSu5K4ycIUYAAAAAJBJDMQAAAAAgExiIAYAAAAAZBIDMQAAAAAgk6pkqZbjShxmzpwpWXFxsWSuFOj++++XbNu2bZK1bdtWslq1apW5zkOVlpZK9pe//EWyO++8U7KFCxemfh/gc64k5KWXXpLsggsukOzGG2+ULEkSyY45Rv+/NFeQ4IqyIiLeeecdySjQQqG57+ulS5dK9h//8R+SuXKR6667TjJXDOT2TkRE06ZNJXOlji+++KJkZRXaAYfDff8fd9xxkvXu3Vuynj17SuaOHc2aNZPMlQe5gtSIiIkTJ0q2fv16ydz+BqqqskrkDrVnz55UWVlq1qwpmTv2DBkyRDJXknfSSSdJ5gq51qxZI9nYsWMl27Vrl2SVgSvEAAAAAIBMYiAGAAAAAGQSAzEAAAAAIJMYiAEAAAAAmVRtSrVcQcLDDz8s2amnnipZjx49JGvRokWq7HDs3btXsvfff1+y//7v/5Zs3rx5klEAgSPhyqnuvfdeyXr16iVZp06dKnQtp512ms0HDx4s2XPPPScZewCF5gpHunXrJpn7rLsCoX379qV+76KiIslGjBgh2euvvy7Zn//8Z8lcURHwRVwZzplnninZwIEDJXPlOm5PuOIuV3I6e/Zsu0YKtJBv7ru5Y8eOkrlyKrcHhg4dKpk7Rxs/frxkW7duLWuZomHDhpK5glW3bld+534f3HHm1VdflcyVU77wwguS5QtXiAEAAAAAmcRADAAAAADIJAZiAAAAAEAmMRADAAAAADKJgRgAAAAAkEnVpmXa2bVrl2Tr1q2TLJfLSeZa3tJyrxcRsWLFCsl++MMfSvbGG29IRiMiKpNr8XTK+myneZx7jzp16tjnX3vttZLNnTtXsqVLl6ZaD1AR3GfYNUr/27/9m2QnnHCCZMuWLZNs4sSJktWo4Q/Fo0aNkuy4446TrF+/fpK99NJLku3cudO+DxDhz4sGDRok2Z133ilZ+/btU72eO09yf8OH+1s7ymrT5fwJlcmdx7i/KcOd1zRv3jzVe7i/zcD9rR/f+973Ur3e4Sjr+HOotOeHrmV6+vTpkrlGafe3pOQLV4gBAAAAAJnEQAwAAAAAyCQGYgAAAABAJjEQAwAAAAAyqUqWarlik86dO0t2++23S3b66aenej3gaNWqVSvJnn76acmKi4sl2759u2SunMcVA7nyoZEjR9o1nn/++ZLNmjVLsn/913+VbN++ffY1gfJye+JHP/qRZKeeeqpkn376qWQ/+9nPJHvuueckc4VEERHnnnuuZE2bNpVswIABkrVu3VoyV/IFfK527dqSucK2tm3bSuaOHc8//7xkTz75pGSuKOjrX/+6ZO78LsJ/rgtZzoPqK22x3AMPPCCZK8Fy1q9fL5krrGrSpIlkaQuwyuIKr9Lun48//lgyVybZrl27I15LIXGFGAAAAACQSQzEAAAAAIBMYiAGAAAAAGQSAzEAAAAAIJOqZKmWKza56667JLv44oslq1OnTqr3cDewl7esx91I/u///u+S3X///ZKNGzdOMkoh8GXc533IkCGSubIHV2gwduxYyX7zm99IVlJSIln9+vUlK6sAYvTo0ZK5Mgugsri9c/XVV0s2fPhwydznesKECaky972+Z8+eMtd5KFcS2ahRI8nKW76C7HFFbP3795fsww8/lMyVxT3yyCOSbdy4UbK+fftKNnjwYMnccSMiYurUqZJRIIcj4Yrl3B7o2LGjZEVFRZI99dRTko0ZM0ayzZs3S9ahQwfJRo0aJZkr/Tr++OMli4h4++23Jbvooosk27Vrl2Tu1/eTn/xEsjvuuEOyoUOHSvb4449LtnTpUsnyhTNQAAAAAEAmMRADAAAAADKJgRgAAAAAkEkMxAAAAACATCp464Yr/hgxYoRkrtikbt26qd5j7969ki1YsECyKVOmSLZ9+3bJGjRoYN/HrbFHjx6S3XrrrZK9//77kr333nuSuTIwZIMrnXJlCrfffnuq5y5ZskQyV4KyevXqVOtzZRSuFALIN3ecadOmjWQXXnihZK4sbufOnZJ98sknkrkCLVdMUt4CLFeQBxyuZs2aSeY+m65o8cUXX5TMHTvc66UtNG3RooXNmzdvLhmlWjgSXbt2lWzgwIGSuc+xK6J68MEHJXPnXu7cfu3atZJ98MEHkj3xxBOSffOb35QsIuL888+XLO2edIWOjnucOxfs1auXZG7f5uv4xhViAAAAAEAmMRADAAAAADKJgRgAAAAAkEkMxAAAAACATCp4qZbTqFEjyVxhT1qLFi2SbPTo0ZK98847krkby2vVqmXf591335Xs7rvvluzEE0+U7Oabb5bstttuk2zNmjX2vXH0c8Uh11xzjWSdOnWS7Omnn5bsrrvukixtgZbTv39/yfr27Wsf64rutmzZcsTvDRwuVyRSVmHiobZu3SqZKwNxRSlNmzaVzBV8RUTUrFkz1Wu69aQtKkI2uc+/+w7v2LGjZKeccopk7vPvzr1cQY4rCtq8ebNkLVu2lCzCH2dmz54tWWlpqX0+8DlXVOqKthz3eV+4cKFk5SnHdYWOroB31apV9vmLFy+WbMeOHZK5IlZXaueKg50mTZpI1r17d8kmTZokGaVaAAAAAABUIgZiAAAAAEAmMRADAAAAADKJgRgAAAAAkElVslTLlevs3r1bMlcK4W5WX7t2rWTr1q2TLG0JSVk3eLub1UtKSiQrLi6WbMiQIZJNnDhRsmeffTb1enB0ady4sWSulGDjxo2SjRkzRrKlS5ce8VpcucL1118vmSsCi4iYN2+eZC+//LJkFAOhIrhyqrTlja6IZ9q0aZL99a9/lcyVALmixa997WuSRUS0b99eMnc8e/TRRyVbuXKlfU2gLG5PuKK50047TTJ3PHn++eclc+doSZJI5s5r3nzzTckiIubMmZPqfYCDuc+dK4lyRbruuPDKK69I5maXfCirgPepp56SzP36Bg8eLNk555wjWVnFqYf6+OOPJXP7tpDzDFeIAQAAAACZxEAMAAAAAMgkBmIAAAAAQCYxEAMAAAAAMqngpVquNGf8+PGSnXLKKZINHz5csvr160t2+umnS3bfffdJNnv27DLX+WXvERExcOBAyU488UTJ3I38TZs2lczd3D9p0iTJKNXKBvd5cJ8bV6q1bNkyyVzpSL169SQbNGiQZNddd51krnChrM/mvffeK9n8+fPtY4HyatasmWTumOIet23bNsnefvttyc466yzJzjzzTMncccvt47K4Qq/JkydLVqgyF1RfrtB07969krnyOVf6WLduXclcSdell14qmTsWvfDCC5JFRMyYMUMyzovwZdw5kDtXcp8lV9Q4bNgwyZ5++mnJNm3alGp9mzdvlswdj9q2bStZly5d7Gu2adNGspEjR0rmzvvq1KljX/NQ27dvl+yxxx6TzB23KNUCAAAAACDPGIgBAAAAAJnEQAwAAAAAyCQGYgAAAABAJhW8VMspKSmR7Kc//alkH374oWRXXXWVZMXFxZINHTpUsiFDhqRdolWjxpH/drpyMXfzPPBlXNmDK1hwhRKjRo2S7Morr5TM7Sn3emvWrLFr/Nvf/iZZaWmpfSxQXhs2bJBs1qxZqR7Xvn17yUaPHi2ZKxxxZVmuVHHPnj2SRUQsXLhQsieeeEKy9evX2+cDZXHnHFOnTpVs5cqVkrk9cfHFF6d6X3ee1K9fP8mWLl0q2Ysvvmhfs6z9Axyu9957TzJXVNqqVSvJOnbsKNkf/vAHyVxRnSuTckWjrvRrwIABkrmirQhfdNeoUSPJjjkm3fVSd943YcIEyf73f/9XMvcdVEhcIQYAAAAAZBIDMQAAAAAgkxiIAQAAAACZxEAMAAAAAMikKlmq5W7SdjeSP/DAA5LNnj1bsmuvvVaygQMHSuZuNk97Y3lZ3E3j7gZ9d/P8lClTUr0esmHGjBmSTZ8+XbJhw4ZJ5ooinLTFcG6PPv3005KNGTPGPt/tZ6CyuBITV6C1du1ayY499thUmbNjxw7JXnrpJcnmzJljnz9u3DjJXNEWhXSoCB9//LFkrnwu7Z646aabUr3v7t27JRs/frxkixYtSvV6wJFy51mTJ0+W7NJLL5XMzRCuaCutrl27SubKt8pT6Hs4du3aJZk7Ht11112SubLkqoYrxAAAAACATGIgBgAAAABkEgMxAAAAACCTGIgBAAAAAJmUuHKc/+8nk6Tsn6yiioqKJCsuLpZsxIgRkvXp00ey7t27S1ZW0daWLVske+211yR74403JHv33XclW7FihWSuHKYqyeVySaHXUAj52CuuOOGf//mfJfvxj38sWZs2bVK9hytscIUnrtzkkksukWzp0qX2fb7oeycrsrhXqtIxpXbt2pKde+65kvXr10+yAQMGpHqPSZMmSfbkk09KtmrVKvt8V8qVNeyT/HHnNv3795dsyJAhknXu3Fkyd/7kzJw5U7K7775bMsoYy5bFfRKRn73iyrKGDx8u2Q9+8APJ3B6oWbNmxSzsC5R1juVKfd0xyc0fL7/8smTuXNCVb1UlZe0VrhADAAAAADKJgRgAAAAAkEkMxAAAAACATGIgBgAAAABk0lFXqpWWKyhq2rSpZI0bN079mvv27ZNs9erVkrmSIldmVB1R7JBfLVq0kGzo0KGSuaItV6Ayd+5cyVzhyZQpU1I9l/KssmVxr1T1Y4rbE658q3Xr1qlez5VlVfXCkaqGfVJYafdEgwYNJEt7/rRp0ybJ1q9fLxnHk7JlcZ9EFG6vuD3Qvn17yYYNGyZZo0aNKmVNBytrppg3b55kL7zwgmSuwNfNONURpVoAAAAAAByEgRgAAAAAkEkMxAAAAACATGIgBgAAAABkUmZLtVA5KHaomlyJnOOKGI6WwreqJot7parvE1Q97BPgy2Vxn0RU/b2S9twrXzjHo1QLAAAAAID/HwZiAAAAAEAmMRADAAAAADKJgRgAAAAAkElV625vAJVi3759hV4CAABAZnDuVX1whRgAAAAAkEkMxAAAAACATGIgBgAAAABkEgMxAAAAACCTklwuV+g1AAAAAACQd1whBgAAAABkEgMxAAAAACCTGIgBAAAAAJnEQAwAAAAAyCQGYgAA8ihJkjpJklyfJMmkJEmWJ0myM0mSzUmSfJAkya+TJPlGodcIAEBW0Atpw1UAACAASURBVDINAECeJEkyKCL+OyKOPSjeEhG1D/zvc5Mj4qpcLrc+j8sDACBzuEIMAEAeJElyaUQ8H58Nwysi4h8jolkul2ucy+XqRMRXI2JsROyLiCER8dckSVoVar0AAGQBV4gBAKhkSZJ8JSL+FhH1I2JuRJyTy+XWlvHY8yNifETUioiXc7ncwLwtFACAjOEKMQAAle/n8dkwvDsivlnWMBwRkcvlno+Iew/88JwkSYbmYX0AAGQSV4gBAKhESZK0jYhP4rP/E/q3uVzuuhTPaRARn0ZEw4iYnMvlzq/cVQIAkE1cIQYAoHINiL8fb/83zRNyudy2iHjxwA/7JklSoxLWBQBA5jEQAwBQubof9O/vHsbzZh/4Z4OIKK645QAAgM8xEAMAULmaH/Tvh/PXKK0r4zUAAEAFYSAGAKBqSgq9AAAAjnYMxAAAVK6DrwofzpXeI72yDAAAUmIgBgCgcr1/0L+fchjPO/nAP7dFREnFLQcAAHyOgRgAgMr1akTsP/DvI9M84cBfuzTowA+n5XK5fZWxMAAAso6BGACASpTL5VZGxLMHfnh5kiQnpnjaD+Kzv4M4IuKXlbIwAAAQSS6XK/QaAAA4qiVJ0j0i3o6IuhExNyLOzuVy68p47JCImBARteKzq8vn5DhYAwBQKbhCDABAJcvlcvMj4h8jojQiToqId5Mk+VaSJE0+f0ySJCckSfLvETExPhuGl0bE/2EYBgCg8nCFGACAPEmS5LyI+K+IaHdQvDkiakdEnYOyFyPiylwutzaPywMAIHMYiAEAyKMkSepGxLURcWFE9IyIFhGxJyJWRsS0iHgml8u9XLAFAgCQIQzEAAAAAIBM4h5iAAAAAEAmMRADAAAAADKJgRgAAAAAkEkMxAAAAACATKrxRT+ZJAmNWzgsuVwuKfQaCoG9gsOVxb3CPsHhYp8AXy6L+ySCvYLDV9Ze4QoxAAAAACCTGIgBAAAAAJnEQAwAAAAAyCQGYgAAAABAJn1hqdbRoqioSLLS0tICrAQAAAAAUFVwhRgAAAAAkEkMxAAAAACATGIgBgAAAABkEgMxAAAAACCTqmSp1jHH6Jxeu3ZtyVq3bi1Zs2bNJOvXr59kr7/+umQlJSWSrV+/vsx1AgAQ4Y9Rbdu2TfXc7du323zDhg2SUQiJrHLnhhERSZJIlnafuOc6uVwu1eOAfHP7oqzPtfsc79+/v8LXVB1xhRgAAAAAkEkMxAAAAACATGIgBgAAAABkEgMxAAAAACCTqmSpVv/+/SUbMmSIZGeffbZkrlSrZcuWkq1cuVKyJ598UrJf//rXkpVVtOVuVt+3b599LFARXJlCWcUjh3JFCpQrAF+uXr16ko0YMUKy73//+5K5spOPPvrIvs/vf/97yaZNmybZ2rVrJaMECFVR2tLUhg0bSnbGGWfY12zRooVk7777rmSuvK5Dhw72NQ+1ePFiyVasWCHZ3r17U70ecLC053I1a9aUrEGDBpLVr1/fvo/bA9u2bZPMzS5pC7mq63kkV4gBAAAAAJnEQAwAAAAAyCQGYgAAAABAJjEQAwAAAAAyKfmi4o0kSSq9lcOVITz99NOS9evXT7JatWpJ5goNduzYIZm74Xzjxo2STZ8+XbJ58+ZJFhGxadMmyaZMmZL6+UeDXC6njTEZkI+9krZ4xBUs1Kih/XmbN2+WzJUruD1FWVz5ZXGv5GOfVDR3jLr++usl+9a3viVZx44dJXOlWqWlpfa9XYHj5MmTJRs7dqxkCxYskGz37t32faoy9kn15QqA3DHL7ZPOnTtL9s1vftO+jytOnTVrlmTuXDBtqZYr6XLnqhs2bJAsHyVDWdwnEdVzr7hjQPPmzSVr1KiRZE2aNJGsuLg4VRYRUVJSItny5cslc+eC7vjhHrd161bJ9uzZI1mhih/L2itcIQYAAAAAZBIDMQAAAAAgkxiIAQAAAACZxEAMAAAAAMikvJZqtWrVSrL77rtPsmuuuSbV67m1uyKFG264IdXrjR49WrI+ffpI1rRpU/t8l7uChQcffFCyV155RTJ3o7srWSnUjekOxQ4V9nqSuZKEnj17Snb88cdL5krk5s+fL5krXHBlcZ9++qlkrnyrrLIgZHOvVKUCFLfHXDnPRRddJNk999wjWevWrStmYV/CFZu4Aq37779fsgkTJki2c+fOillYJWGfVF+uLMvtpxtvvFGyxo0bS9asWTP7Pscco9d20h573PeAs2bNGsncul0R69q1a1O9R3lkcZ9EVK294krk3Oe4ffv2kt18882Sde/eXTK3B1xRXb169ewaXbHc9u3bJXPnc66I1ZUEuzLht956S7KlS5dKlo95hlItAAAAAAAOwkAMAAAAAMgkBmIAAAAAQCYxEAMAAAAAMqlGZb2wKzk488wzJRsyZMgRv4crmHrkkUckmzt3rmR79uyR7JZbbpHM3cDuiowiIq644grJzj33XMnGjBkj2Xe/+13JXn/9dcl+/vOfS1aoG9NRMYqKiiRr3ry5ZHfeeadkZ511lmRt2rSRrHbt2pK5sqytW7emetykSZMkc+UKixYtkiwiYtWqVZK5/UwpFypCixYtJEtbbOKOUa58y0lb2FMW9z3u9rI7Jt19992SuePyuHHjJNu1a1eqtQCfc591dxzr0KFDqswVFLnPb1lq1KjY09smTZpI1rlzZ8nmzJkjWT5KtZBf7rPoCrR69eol2emnny5Z3759JXNFje77362lrL1St25dycoqCj6UK3R0+8LtXeeTTz5J9R75whViAAAAAEAmMRADAAAAADKJgRgAAAAAkEkMxAAAAACATKq0Ui134/YZZ5whmSutcuUMrtBj+fLlkk2bNk2ytDdpr1u3LlW2ZMkS+/zZs2dL5gqJLr30Usk6deokWdu2bSVzvze33XabZGvWrLFrRNXjCghcMdbJJ58sWbt27SRze899blwZgiuFcGtxhSU9evSQzBVtlZW7EjlX8uUK8Sjfwufq1Kkj2Xe+8x3JXFnW1772NclciYmT9rhVGdx7u8Kfu+66S7L9+/dLNnPmTMmWLVt2RGvD0cftCXe+4krq3HlgefZYRPp95j7rjiskcmvs06ePZK7k1O2dtGtB1VSrVi3J3LnS17/+dckGDx4smSt5dO9R3qLGwyngOpQrgHXrPu200yTbt2+fZC+//LJkhSxX5QoxAAAAACCTGIgBAAAAAJnEQAwAAAAAyCQGYgAAAABAJlVIqZa70XrgwIGSjRgxQhdgynkcd0P2a6+9JtnKlStTvV55lHWDtyvbcoVXJSUlkl155ZWSFRcXS+aKYGbMmCHZ448/LtnevXslQ3658oKGDRtKdvzxx0vmSkvSlpG4Yrm1a9dK5gqJGjRoIFnXrl0lc4US7nERvoDLWbRokWSrVq2SzP1a8lVohMJxBSMnnniiZJdddplk7rOZjwItVwq3c+dO+1i3Hle04r5X0hZt3XrrrZJNmDBBsgceeECyHTt2SIajnytf7NWrl2SuXMcdx9I6nO/0Xbt2SbZt2zbJ3D5xxzy3Fzt27Jgqc/uTUq3qzZ23uWOKK0Pt0qWLZGmPPYXk9oo7Z3R7vHfv3pK5c8YtW7ZIRqkWAAAAAACViIEYAAAAAJBJDMQAAAAAgExiIAYAAAAAZFKFlGp16NBBsjvuuEMyV+jhip7cTdWuxKF58+aSuRvTyyosqWiu8GHNmjWSjRkzRrLHHntMshtvvFGym2++WbJf/vKXqdbnilLWrVuX6rmoGK5co379+pK5Yg5X4uBez5X2uLK5l19+WbLWrVtL1q1bN8lc4Zvbj40aNZIswpcpOHPnzpVs9uzZkr344ouSuVIViraOLq5gKm15h3uuU54CLfe4Tz/9VLJly5bZ5zdp0kSyY489VjK399y6XZa2AKVdu3aSuSJJ9tjRxZXmuHLP4cOHS9apUyfJXAlrWu47PcIf35544gnJ3n//fcmaNm0q2dChQyU7//zzJXPHZLdn3b5D9eY+Nz179kyVue9rpzzHnrJKdLdu3SqZm5Hcvm/WrJlkbo3unPaEE06Q7JxzzpHMzX+ffPKJZK5ouby4QgwAAAAAyCQGYgAAAABAJjEQAwAAAAAyiYEYAAAAAJBJFVKqVaOGvkxZZTqHcjdLP/jgg5L97Gc/k2z37t2SVYdCD7fG1atXS/Zf//VfkrmypUsuuUSyu+++W7K6detK5gq5SktLJUPlcfvHlcjVrFlTMvdn5QpGxo8fL9m//Mu/pHrfHj16SHbddddJdtJJJ0nWqlUrySJ8OYMrZXHlLa6U6K677pLsr3/9a6rnUr5VfbVv316y0aNHS9ayZUvJ0pZOpf0suOPRggULJHPHt3feece+pitG6du3r2S33367ZK7E0v363O/N4MGDJdu0aZNkt912m2SuSBLVlyukGzhwoGSuPChtgZYryHH7zp0vRkS8+eabkrlj3qpVqyRzpbCuAGj//v2SuWOyKxRyx/iySo9QPTRo0EAyV7LmyhvzUbK2efNmm8+fP18yV+rYokULydy+d0XGjtsD7vfQPS5fuEIMAAAAAMgkBmIAAAAAQCYxEAMAAAAAMomBGAAAAACQSXm9e9kVALkykVdffVWy733ve5K5G7wHDBgg2XPPPZdqLVVNSUmJZK6QxZUede/eXTJ3k/wxx+j/J1Idfm9QNldQsn37dsm2bt16xFmTJk0k27hxo2TusxkR0bx5c8mOPfZYyerUqZPqceedd559n0OlLdrauXNnqtdD/rjvql69eklWXFwsWdoSk7QFWu7zMW7cOMnc97Ur2tqzZ0/q9bjPqyvocUVzaYu23L5zBXevvfaaZL/73e8kc4VjKCz35+6KN6+55hrJXAFivXr1JHOFha70ccKECZJt2bJFslmzZkkWETF37lzJ3PmT20+uCMm9t9O6dWvJ3Dmoe5wrCHPHbhSeO/a444wrvXUla+71ylPo6Lh9FhExc+ZMyd566y3JXJneaaedJpkrlkv763OPK6SqtRoAAAAAAPKEgRgAAAAAkEkMxAAAAACATGIgBgAAAABkUl5LtVzpzh//+EfJ1q1bJ9m0adMku+CCCyS77LLLJJsxY0aq96hq3A31rpBl7Nixkv3617+WrHfv3pJ16NBBsqVLl6ZdIqqgGjV0W7vCE1do4EpQ1q9fL5nbj25/z58/367RFWONGjVKsmbNmklWq1Ytyfr06WPfJw1XtOX2QHkKLlB+tWvXluyMM86QrGnTpqleL22JiSuEcgVarsSqMj5HaQu93PfAfffdJ1mrVq1SvW/Lli0lGz16tGRTp06VbNmyZaneA/njitPatWsnmTtvcM91XLHPG2+8IdkzzzwjmSu22rBhg30fV/xYnn22f//+VK/n9liDBg0kc8estEV/KLyioiLJevbsKZkrEXUFpBVdJuXO29wxIcIX2LmCt06dOkm2bds2ydIeb6sDrhADAAAAADKJgRgAAAAAkEkMxAAAAACATGIgBgAAAABkUl5LtRxXVOBKfH76059K9sEHH0jmbn53BQnVlSt4mTx5smSbNm2S7Oyzz5bsyiuvlOz++++XzN20j8Jz5Qz169eXrGPHjpK58g/3+SotLZVsxYoVkrkSlPfff1+yiIj27dtLduqpp0rWtWtXyVxJhSuDKU/RliuZcL83yJ/WrVtLNmDAAMlc0U3aAi1n4cKFkj344IOSFbKIzRVtvfDCC5INGzZMsosuukgy973ifg8bN24smSs/Q2G5P5Phw4dLdv7550s2ePBgydznY8eOHZL9z//8j2Su1Oe9996TLF97Z/PmzZLNmzdPMndO1bx5c8lcyVD37t0lcyWUa9euLXOdKBx3TGnYsKFk+ShPc/OMK5Urq9DUfcbc94M7Z3TnlkdTORxXiAEAAAAAmcRADAAAAADIJAZiAAAAAEAmMRADAAAAADIpr6VazZo1k+zyyy+XbNy4cZKtXr1asrFjx0qWryKGqmTdunWSTZs2TTJXojFq1CjJnn76ack+/PDDI1wdvsy+ffskc0Ufe/fulaxOnTqSudKpHj16SObKP1xxiCtxcAVTLnOfzYiIkpISyW688UbJTj75ZMnOOeccyS655BLJOnXqJJkrZvrKV74i2YwZMyRzRVvuzw6VwxWbNGrUqELfw5UHjh8/XjJXtFXIY48rNnGlLzVr1qzQ93XfNd26dZNs8eLFkh1NZZdViSsWdeVn5557rmSnnXaaZK5wZ8+ePZK5c7TZs2dLtnLlSskKuXe2bdsmmTs+ueKiFi1aSObKKl1B0bvvvisZpVqF5wrj0v6ZutIp93rl4b43t2/fLtny5cvt8915pCs5deeM7phS0b++Qjp6fiUAAAAAABwGBmIAAAAAQCYxEAMAAAAAMomBGAAAAACQSQzEAAAAAIBMqrSWadeE5trIunfvnupx7vWy2CjtuN+b+fPnS3bRRRdJdjQ1xFUHaRsCly1bJplruaxVq5ZkrnHWPc41BromX9dKWF7uNV0Trfu9cc3OZ555pmTt2rWTzDWmuqbitL+vOLqsX79esqlTp0rmGtXzxX0OO3fuLNmdd94pmWtoL88xIG2j/aRJkySjZbpyNG/eXLK+fftKNmTIkFTPdRYsWCDZhAkTJJsyZYpkO3fuTPUe+eL2smvodX9zh2vndU3u7nHub11xx33kl/s+dO3RxcXFkrk26nycY5eWlkrmzmEi/Lp79+6dKivP31Lg5rWqdgxgGgIAAAAAZBIDMQAAAAAgkxiIAQAAAACZxEAMAAAAAMikCinV2rx5s2Su1Klr166SuRvOKa8pv6p2szo+4/5cXFnWkiVLJFu5cqVkrsTBFUe5UghXxLNixQrJXAGWK7YqL1e24tbz17/+VbL33ntPMvdrdiVAriiikIVjSK+iv+dc0c3JJ58s2fTp0yWrjD3hPnMdO3aUzBVoXXzxxZLVqVMn1fu6YzAlltVDmzZtJHNFZ65MsKioSDJXOjV79uxUmXtuVfscpT0mv/baa5INHTpUMneMcXvWlRu530PO5QrPfQ+7/VOe0qm03Mzkjlv33HOPfX6LFi0ka9mypWSNGzeWzH0/uP3s9v0nn3wimfu8b9q0SbJ87QGuEAMAAAAAMomBGAAAAACQSQzEAAAAAIBMYiAGAAAAAGRShZRqbdy4UbKZM2dKNnjwYMmaNm0qWYcOHSRbunTpEa4um1wpEKomV8y0atUqyVwBgftzbteunWSufKt3796SueIuZ926dZK54oPDKUNIW86wevVqyebOnSvZ8ccfL5n7fXDlW8cdd5xkJSUlqdZHCUrlqOjyRseVoriSnLTFj67grn379qneIyLiG9/4hmT9+vWTbNCgQZLlo0DLHfvnzZsnGXuicriyn3POOUeyCy64QLK0nw93THjsscckc3uxtLQ01XtUNXv27JHsgw8+kGzbtm2Sub3cvXt3yVzR2aRJkyRj7xx9ylMc7I5l7jzQnd9F+GNSrVq1Ur2P4/aK+8548803JXPnba7QjlItAAAAAAAqEQMxAAAAACCTGIgBAAAAAJnEQAwAAAAAyKQKKdXat2+fZOPHj5fsrLPOkmzo0KGSXXXVVZI98MADku3YsSPtEo9qrlhjwIABkhUVFUlGYUPhueKRtWvXSvbII4+ketzw4cMlcyVR//RP/yTZhRdeKNnixYslc6V5H330kWTLli2TLMKXkTiuBKtjx46SuaIiV/bgsrQFQiistOWNQ4YMkaxu3bqSuWITVyTiykl69eolmSsDOf300yUbPXq0ZM2aNZMsIqJt27aSuVKUtAUoTtrP/86dOyWbPHmyZDNmzJCM40z+uO9Ml2WR2yfuvMiVDHXr1k0y9/tanr0IHAk3AzRq1Mg+tjyFXu573B33XLHiyy+/LJkrj3XnaPnCzgUAAAAAZBIDMQAAAAAgkxiIAQAAAACZxEAMAAAAAMikCinVcpYuXSrZPffcI9lJJ50k2W233SZZly5dJBszZoxkrtjHlRZVV+6GeFcy5EqUXHnK/PnzJdu8efORLQ4Vxv1ZbdiwQbJPPvkkVebKP1q0aCFZ/fr1JWvTpo1k7jPnCrTyVap1yimnSOZKJVz5xLp16yQrKSmRbPv27ZJRFpQ/7vfaHWfcZ8uVark95gpx3DFq1KhRqd7XFXJ99atflcwVZUX4z2ta7liRtkDLHTNXrFgh2YsvvigZx4/Ccp9DV3zjPgvuM+OK3W644QbJnn32Wclc6Zr7bNWsWVMy993fuHFjycriHuvOi3r27CmZO3a4skpXhrd7927JXPnf7NmzJeN4goOVpwCrvNz3w5o1ayRbsGCBZK5A67XXXpNs165dR7a4SsIVYgAAAABAJjEQAwAAAAAyiYEYAAAAAJBJDMQAAAAAgEyqtFItZ/ny5ZJNmzZNsmuuuUaykSNHSuZKF5566inJ3nvvPclcaY4rQ6hqWrZsKdntt98uWfPmzSVzN7DPmTNHMlfAgcJzZTXus+32xaZNmyQbOHCgZK7cp2nTpqneo0OHDpKdfPLJkkVE7Nu3z+aHcqVC7r1d5spb3GfblZC5ArO9e/eWuU5UPlc4476/3He7K5BLW1jiStxuuummVM91JV0uK0t5SlVcKYrLXPGd+1757W9/K9lf/vIXyarDcfRo4fbEhx9+KNm8efMk69Gjh2Tu+79WrVqSubI49z3qCnfcd3/aAiy35gi/pxo2bCiZ28vuNd2v2ZVL7ty5U7KVK1dKNmXKFMnmzp0rGaVa2ZC20K48r1fe57vv8TfffFOyt956SzL32a4OcwVXiAEAAAAAmcRADAAAAADIJAZiAAAAAEAmMRADAAAAADIpr6Va69evl2zs2LGSpS3QuuCCCyQ744wzJNu4caNk99xzj2QTJkyQzJUmVEbxQVFRkWSuGOuiiy6SbMiQIZK5G/QXL14s2bPPPiuZK99C4W3ZskUyV17gCqFc4cmxxx4rWbNmzSRz5SSuBKVRo0apnhtRvhIIV5blikxckdiSJUske/fddyVzBRAUnlQ9K1askOzhhx+W7O6775asXbt2krlSIfddWrNmzbRLrHBp947bJx999JFkY8aMkcyVp7jnuvdA/rjvJFfg9MEHH0jmyhJPP/10yVq1aiVZt27dJHOlipdeeqlkjttP9evXl6xJkyb2+eUpJHJlidu2bZPs/fffl2zixImSuaI/yueOPq4czn0funN793ktbzFWWm6N7pzfnVO5GWn27NmSueOyO3+tarhCDAAAAADIJAZiAAAAAEAmMRADAAAAADKJgRgAAAAAkEl5LdVyN40vWLBAsl/84heSXXXVVZIdd9xxkrVs2TJV5kpWevfuLdkbb7whmStXiEhfMOJusj/ppJMku+yyyyTr27evZO7X50ohxo0bJ9miRYvKXCeqFlf+sW7dOslceZ3bZ9OmTZPMlZYUFxdL5j6vxxxT8f//miuNcYVXU6dOlcyVarnnuoxiuerBFdOMHz9esi5dukg2fPhwyXr27JnqfctT4lPe8hS3J1z540svvSTZb3/7W8nSFv7kq/QF5eP+7D799FPJ/va3v0nmijxdWWKtWrUkS1u0mJY7npR1jHGfTfcd7s6LPv74Y8mWLVsm2cyZMyVzBWarVq2SjP1Ufbg/F1egtXr1asnc+ZPLatTQ0as8xxTHnS9G+PMdV6DlyrJmzJghmftucZ/36lDAyBViAAAAAEAmMRADAAAAADKJgRgAAAAAkEkMxAAAAACATEq+6Mb+JEkKcte/u7nclT24Uq0LL7wwVdatWzfJateuLZn7/XGlJuXlyiLc74O7Ud6VI/3f//t/JVu4cKFk7ub38sjlchXbDFBNFGqvlIcrdqhZs6ZkDRo0kMwVqORL2oILt1fc3q2M/ZxGFvdKofZJx44dJevTp49kY8aMkaxTp06SuWLEtA7nM7hhwwbJpk+fLpkr/HHlYiUlJZJV9bIT9knlaNGihWRt2rSRbPDgwZK5veP2WMOGDY9scWUoa5+4sqznnntOsvnz50s2a9YsyVzx0MaNGyVzx6JCyeI+icjPXnHnRSeeeKJkXbt2lcztC3dOVdGlpFu2bLG5K4xbsmSJZK4cbu3atZJVx3K4svYKV4gBAAAAAJnEQAwAAAAAyCQGYgAAAABAJjEQAwAAAAAyqUqWapVHnTp1JGvXrp1kV111lWSufKtDhw6SNWnSxL63K8Fyv7+uAMg91z3ulVdekWzy5MmSvfrqq3aNlY1ih6OPK3uo6AKI8qpK5SZpZXGvVKWiRlei6AqErrjiCslOOukkydyecCVArtjno48+kiwi4o033pBsxowZkrnyreq4Jxz2SeVwn9datWpJ5s6fevXqJZkrD2rUqNGRLa4Mh1Oq9fLLL0vmioLc3qlK5YtpZXGfRBTumOJmDVciV79+fclcoWlFK+v7f/v27ZK5EjlXSlrVCxjTolQLAAAAAICDMBADAAAAADKJgRgAAAAAkEkMxAAAAACATDrqSrXSSlu+5cojevToYV8zbanK1KlTJXMFL0uWLJFs5cqVku3evTvV++YDxQ5AOlncK1V9nxQVFUnWrFkzyRo3bnzE77F582bJXClQRNX6bi8U9knVQ9Fi1ZPFfRJRtfZKddgX1bEwrqJRqgUAAAAAwEEYiAEAAAAAmcRADAAAAADIJAZiAAAAAEAmZbZUK63KuEm+tLQ01eO+6M+mqqLYAUgni3uFfYLDxT4BvlwW90kEewWHj1ItAAAAAAAOwkAMAAAAAMgkBmIAAAAAQCYxEAMAAAAAMqlGoRdQ1e3fvz9VBgAAAACoXrhCDAAAAADIJAZiAAAAAEAmMRADAAAAADKJgRgAAAAAkElJLpcr9BoAAAAAAMg7rhADAAAAADKJgRgAAAAAkEkMxAAAAACATGIgBgAAAABkUo1CLwAAgKxIkuSUiLgwIk6NiBMiomVENIqILRGxICKej4jHKjdzMAAAIABJREFUcrnchoItEgCADKFlGgCAPEmS5JGIGH1QtCsi9kZEw4OydRFxUS6Xm5nPtQEAkEX8J9MAAOTPWxFxS0T0iYimuVyubi6XaxSfDcTXRsTaiGgREROSJGlcsFUCAJARXCEGAKCKSJLk3Ij4y4EfXpnL5Z4q5HoAADjacYUYAICq468H/fuxBVsFAAAZwUAMAEDV0fegf19SsFUAAJAR/CfTAAAUUJIktSOibURcEBF3R0TTiPgwInrkcrndhVwbAABHO/7aJQAACiBJkl0RUdv81IyI+D8MwwAAVD7+k2kAAApjVUSsjojtB2WvRsTNuVxueWGWBABAtjAQAwBQALlcrmMul2uTy+UaRETriPhRRPSOiLeSJLm7sKsDACAbuIcYAIAqIkmSr0fEzPjs/7C+MJfLPVfgJQEAcFTjCjEAAFVELpd7KyKmH/jhPxVyLQAAZAEDMQAAVcuKA//sUtBVAACQAQzEAABULZ0P/HNrQVcBAEAGMBADAJAHSZIUJUmSfMljzomIrx/44WuVvigAADKOgRgAgPzoEBHvJklyfZIknQ8ejpMk6ZAkyY8j4tmISCJiQ0Q8VKB1AgCQGbRMAwCQB0mSdIyIjw6K9kTEloioGxH1D8o/ioiRuVzu3bwtDgCAjGIgBgAgD5IkqRURwyJiQEScFhFtI6JFRJRGxNqImBOfXSF+OpfL7SzQMgEAyBQGYgAAAABAJnEPMQAAAAAgkxiIAQAAAACZxEAMAAAAAMgkBmIAAAAAQCbV+KKfTJKExi0cllwul3z5o44+7BUcrizuFfYJDhf7BPhyWdwnEewVHL6y9gpXiAEAAAAAmcRADAAAAADIJAZiAAAAAEAmfeE9xABQSEmit3rkctwyhMpRo0a6Q+L+/ftTZQC+XNp9dzjYowAOB1eIAQAAAACZxEAMAAAAAMgkBmIAAAAAQCYxEAMAAAAAMolSrQKoU6eOZA0bNpTMFU3Ur19fsn379km2evXqVI8rLS2VjOIJVDZXltWyZUvJevbsKdmqVaskW7JkiWR79+6VzO0BHP1q1qwp2YknnijZoEGDJHPfzfPnz5fs3XfflWzFihWS7d69u8x1AtWVO19p2rSpZB06dJCsf//+kjVq1Cj1e7tzlnnz5kmWdo+6YwfwZY45Rq8xusyd/zRp0kSyxo0bV8zCjoA7V1q5cqVkR9PxjCvEAAAAAIBMYiAGAAAAAGQSAzEAAAAAIJMYiAEAAAAAmUSpVgUqKiqSrLi4WLIrrrhCsl69eknWoEGDVK+3bds2yaZOnSrZ1q1bJZs1a5ZkL774omRH043zyB9XHhER0blzZ8nuuOMOyQYOHCjZ+vXrJXvppZckc2Upbl9s2LBBMlfctWvXLslQPbgCrccee0yyU089VTJXyLVp0ybJSkpKJHvooYckGzdunGQ7d+6UDKgK0p7XjBgxQrI+ffpI1rt3b8natm0rmdt3ZcnlcpK5PTp79mzJrr/+esmWLVuW+r2RTa4c15Uyus+7K6Dr3r17qsyVdFWGzZs3S/boo49KNnnyZMnWrl0rmdujVQ1XiAEAAAAAmcRADAAAAADIJAZiAAAAAEAmMRADAAAAADIp+aIbnZMkqfp3QRdAWTe1f+Mb35DsxhtvlMwVBdWqVUuyPXv2SJb2ZvVWrVpJ5koq5syZI9lll10mWdqSiVwu51uUjnLsFV++0qlTJ/vYO++8U7KLL75Ysrp166Z673379km2d+9eyVavXi2ZK+lyxUe/+tWvJHOFXGllca/kY5+47+dhw4ZJ5kq1Fi1aJNnbb78tmStP6dq1q2Tuu/kPf/iDZN/+9rcli4goLS21eZawT/LH7Z3ynNe47+/ylAKVVdKYtrDHncdceumlkr3zzjtH/B6FksV9EpGfveI+d/3795fMlU516dIl1eu5feEel69jglvPunXrJJs4caJkP/nJT1I9t1DK2itcIQYAAAAAZBIDMQAAAAAgkxiIAQAAAACZxEAMAAAAAMikGoVeQHXkCrAiIr7+9a9Ldvrpp0vmiibcDeeuAOill15Ks8QYMWKEZG3btpWsUaNGktWowccCf+eKHVq2bClZ3759Jbvyyivtaw4ePFiyOnXqSLZ//37Jdu/eLZnbKw0aNJCsuLg4VVavXj3J/vjHP0pWnlItVA73mZk/f75krkxt8+bNkt17772SPf7445KNHDlSsmuuuUayIUOGSFZW+dySJUskq+rlPqi+mjdvLtn1118v2dChQyUrz3mD27MuK4srAHKZOwcaPXq0ZLfccotkVakUCPnlvnNdwe2qVaskK8++2Lp1q2SvvfZaqscdDjcHuBnCnSu50r1jjz1Wsuqwf7hCDAAAAADIJAZiAAAAAEAmMRADAAAAADKJgRgAAAAAkEm0J32JoqIiyVzxRETEOeecI5kr9pk7d65kY8eOlWz27NmSLVq0SLL69etL5m6SP++88yQDvowr0HJFQxdddJFkLVq0sK/pCk927dol2ZQpUyR7/fXXJZs6dapkp556qmRu3W4/Uyx3dHEFWk8++aRkrvDKFay57/CPPvpIsoYNG0p20003SXb77bdLFkG5D/KrcePGkp100kmSpf1+dAWIK1askMyd67givLKcfPLJkg0aNEiy2rVrS+aKT5s0aSIZ+w4Hc+fi3/72tyUrz7nEvn37JHMFonv37j3i94jw84I7dl133XWSuTJUd87oylmrWkEkV4gBAAAAAJnEQAwAAAAAyCQGYgAAAABAJjEQAwAAAAAyieaYL1FaWipZWeUKrhirS5cukr3yyiuSLVy4ULI9e/ZI5kq+zjrrLMm+8Y1vSOaKImbOnCnZ5s2bJUM2uIIEVzTkCrRatWqV+n22b98u2YQJEyQbM2aMZMuWLUv1Hp07d5YsbYnDli1bJHMFF6geXKnWQw89JNkzzzwj2Zo1ayRzn6OdO3dKtmnTplTrO+OMM2xOuQ8KzRUgOu7zP27cOMkefvhhyUpKSiRze8eVYkVEDBw4UDJXtNW+fXvJ0v76gIO5Iqu05yaF5MqtOnToINlpp50mmdsr7lzO7eeqVqDl8E0AAAAAAMgkBmIAAAAAQCYxEAMAAAAAMomBGAAAAACQSZRqHYHdu3fb/Pnnn5csbWGDu9G9Xr16kg0aNEiy0aNHS9a6dWvJVq5cKdlTTz0l2YYNG8pcJ44e7jPnihR+9KMfSdayZUvJXAGdK1eIiHjiiScke/zxx1M935UzuAKtUaNGSda0aVPJXCnEpEmTJFu1apVkqL527dolWXlKUdzn//XXX5fMfY7cXgTyzRUHunIrV6A1ceJEyX7+859LtnTpUsn2798vmSvQGjZsmGQR/hiVtuTRvTf+X/buPL7K8kD7+H2AEJAQIGEtu6DIKi6dSjQsClVQEai1Kjhqa1tbdYa+1tqp+9S2bh3tom2nnVar3R1BRFFBEFDUVikQQgelEDbZ9zUQOO8fte/L9LqiT8hycnL/vp/P++n01ywPep6cc3veXCLbuQHeEELo2bOnNHf/nHzyydLcc+aCBQukZesZgneIAQAAAABR4kAMAAAAAIgSB2IAAAAAQJQ4EAMAAAAAosSoVg1y4wzuF9s7d+4szQ1ItGzZUpoblRg0aJA0N6D11ltvSVu8eLE0Nw6DhseNU7kRqz/+8Y/S3FhcaWmpNDfaFkIIM2fOlLZ//377sf+oSRP9sTV+/HhpI0eOlObGi2bPni3tqaeekuYGJYC/cz/rzzjjDGlt27aVtnbt2lq5JqAq3OuGH//4x9KuueYaae7n4+WXXy7NPe84+fn50saNG2c/tkePHtLcz3o3iPrmm29Kc0NiQE1xr59cc4/h1q1bS+vatas0NzQaQggTJ06Udv7550tzr7OmT58u7dvf/ra0bdu22e9d3/EOMQAAAAAgShyIAQAAAABR4kAMAAAAAIgSB2IAAAAAQJQY1aplffr0kfb9739fmhvaysnJkdapUydp77//vrR77rlHmhuPcCNKiJd7PNx6663S3LDDrl27pG3fvt1+Hzfc5kYl3Nicu6eGDh0qrXnz5tLcqMq8efOkrVmzRhpQVW4Y0Y2VJB0aAmqTG/Fxr0369u0rbciQIbVyTcdy905VbN26VZobUKzseQv4O3evFBYWSisoKJDWv39/aQMGDJDmHu/ucwcPHizNPfdUdj1uENINy7lRLfdaKVufz3iHGAAAAAAQJQ7EAAAAAIAocSAGAAAAAESJAzEAAAAAIEqMatUg94vpZ511lrRTTz1VWps2bRJ9DzdGNG3aNGlTp06Vtn//fmnZ+svvqB3u8eCGSFxzA1hdu3a138eNcvXs2VOaG4u45JJLpLmhLTfStWnTJmlz586VVlFRIQ34MG4E0Q2buDGWJUuW2K/phuqA2tKxY0dp7udt27ZtE30991ivK+65zF33NddcI62srEzaypUrE30PNDzuZ3u/fv2kTZ48WZobm3Ov991rInf/uNc1rlXGfU33OD7hhBOkjR07Vpp7LfjSSy9JO3jwYNJLzBjeIQYAAAAARIkDMQAAAAAgShyIAQAAAABR4kAMAAAAAIgSo1o1yA1ezZo1S9r3vvc9aZ/61KekuYGLgoICad26dZPWokULae6X2t01A8dygxKdO3eWNmzYMGk33HCD/ZqFhYXS8vLypLlRombNmklLOhTh7gs3/LVw4UJpR48elYY4uRGTCy64QNqkSZOk7dixQ9qtt95qv8+WLVuO4+qA49Okib4kdD+Xk6rO6JT73OqOdLnhxwkTJiT63DvuuEOaG99Cw+NGO3/wgx9I+/jHPy7NPeaScq85krbqatq0qbQxY8ZI69u3b6KvN336dGn17fzBO8QAAAAAgChxIAYAAAAARIkDMQAAAAAgShyIAQAAAABRYlSrlrnRhQceeEDatGnTpJ155pnSvv3tb0s7++yzpRUXF0ubP3++tM2bN0sDjuUGJb7//e9LGzBggDQ3nhVC9cZRkg4xNG7cWFrr1q2lDRo0SJobgGBUC3/nhlKKioqkuRFE95ywdu3aGrkuoDp27twpbd68eYk+1w3NJeV+tq5Zs0aaGxqtrCcdZGzevLm0cePGSXvvvfekuddybrwU2cO9bjjrrLOkudcNSQe0KioqpG3fvl3a66+/Lm3p0qXSqvvaxN0rw4cPl9ajRw9pPXv2lHbNNddIKykpkbZy5cpE11dXeIcYAAAAABAlDsQAAAAAgChxIAYAAAAARIkDMQAAAAAgSoxqZYAbXVixYoW0Jk30b8/u3bulderUSZobOHKjEIxq4VhJByVOP/10afn5+Ym/jxvGcsMQbmzo2WeflZaTkyPtyiuvlOZGtQYOHJjo47Zs2SINcXJjcR//+MelufvJ/Vx3Yz8hJL9PgJqwdetWaV/60pekVWdAK6l0Oi3N/ZwPwY8Cde7cWdrNN98szQ1otWjRQtqXv/xlaaWlpdKmTp0qLekQJDLP/b168803pb3zzjvSunTpIm3Pnj3S5syZI23BggXS3KjWjh07pLl7pSrcfeXunxtvvFGauy9Gjhwp7aqrrpL24IMPStu/f3+l11nbeIcYAAAAABAlDsQAAAAAgChxIAYAAAAARIkDMQAAAAAgShyIAQAAAABRYmW6nnDLaqtXr5a2cOFCaePHj5d28cUXS3ML1cuWLZNWUVFR6XWifnGLtW4xsKCgQJpbGD98+LA0t7DoHoduldAtLIYQwquvvipt165d0hYtWiRt9uzZ0nr06CHtggsukOb+OvTr109aq1atpLEyjb/btm2btD/96U/Szj77bGkf+9jHpN111132+7iVUbdG6n625+bmStu5c6f9PkBl3Kp5ppbOK1trdv/mDreYfccdd0hz/4aPq6++WlrXrl2lfeYzn5E2b948aTx3ZLfly5dL+9znPifNvR5zr6c3bdokrby8XFpd3WfuGleuXCntlVdekTZmzBhpJ554ojS3Mu1e302fPl1aXa208w4xAAAAACBKHIgBAAAAAFHiQAwAAAAAiBIHYgAAAABAlGptVKtx48bSUqmUtPo02JBJ7s+8fft2aTNmzJA2cuRIaS1btpSWn59/nFeHulZYWCite/fu0oYNGyatY8eOib6eG/JZv369NDco8dnPflZa0kGJEPyohBv0Svqz4OSTT5bmhrHc13PDcm7gC/i7AwcOSPvRj34krW3bttKKioqk3XDDDfb7TJ48WdqOHTuk7d27V5q7Hy+//HJpb7/9tjQ38AJkk3Q6La2srEzad7/7XWnutdJNN90kbdCgQdLatGkjjVGt7OZem7jHUkPiXivNnDlT2re//W1pDzzwgDT3+nXixInS3GhkXd0/vEMMAAAAAIgSB2IAAAAAQJQ4EAMAAAAAosSBGAAAAAAQpRoZ1SooKJB25ZVXSsvJyZG2YsUKabNmzZLmRkwaOjdC5kZacnNzpTFWlj2aNWsm7Qtf+IK0a6+9Vpp7PMyePVvaE088IW3r1q2Jrq++DUq0aNFCmhsqcuMm7h4oLS2VtnPnzuO8OsRq1apV0m655RZpbuDuiiuusF/z4osvluaeb92Qnvu58uijj0q74447pD3//PPSeP5AQ3Tw4EFp69atk+aeB91zTP/+/aW517ncT8g27l554403pLnhR/dadeDAgdLcGCqjWgAAAAAA1CIOxAAAAACAKHEgBgAAAABEiQMxAAAAACBKtTaq9aUvfUlat27dpG3cuFHat771LWnz5s2TtmHDBmkVFRXSjhw5Iq2+DRo0aqT/bMINNnziE5+Q5v4s7hfdFy1alOhzUbfcIM6ECROk9erVS9rq1aul3XfffdL+/Oc/S3P3RSY1btxYmhsgGj16tDT316tJE/3xtm/fPmluQCudTld6nYDjHjNuuM41d8+GEMLjjz8uzd0TbnzL3RNu8GfYsGHS3DCfu3fQ8LnXJq650dQOHTpIcz+Xq2vXrl3S3OPVvaZy96N7rbR9+3Zp7du3lzZgwABpzz33nDRee+GjuHvqYx/7mDT3WN+zZ4/9muXl5dW/sGNU53Wk+zmSSfXragAAAAAAqCMciAEAAAAAUeJADAAAAACIEgdiAAAAAECUan7d4O9f2AwnnHDCCdI6d+4sbfz48dLatm0rbc6cOdLcL5Lv2LFDmhtIyOTIkPvr1apVK2k9evSQtmnTJmmvvPKKtJKSEmkMO2Se+3ufn58vLZVKJfp6btwnU4/tykYTmjdvLm3kyJHSJk2aJK24uFhau3btpB04cEDa1KlTpU2ZMkWaG+cDasvBgwdtLysrS9SWLVsmbdasWdJmzpwpzT3fuhHL6dOnS+P5o2Fp1qyZtFGjRkkbPHiwNPd6xQ22uee2pCp7vJWWlkpbs2aNNDdM+fOf/1zapZdeKs29BnXPyfVtKAjZwT2W+vXrJ+2hhx6StnjxYmluyC0E/7O9OiOi1RnJq2/PH9y5AAAAAIAocSAGAAAAAESJAzEAAAAAIEociAEAAAAAUaqRUS03QLNr165En+tGHC688EJpbnDny1/+srRDhw5JW7JkibTf//730tzoVHl5ubR9+/ZJCyGEnTt3SnO/cN6pUydpp59+urTTTjtNWpcuXaT96le/kubGg/bv3y8NmefuHzcO53To0EHa8OHDpbnRheoMRzVu3FhaYWGhtLPPPtt+flFRkTQ37tO9e/dE17NlyxZpM2bMkHbvvfdKW716daLvAdRXbkDurbfekrZu3TppbqjRPR+98MIL0urbKAqSc+NPbkDru9/9rjT3c9mNAlVncKcqTjrpJGlJxyXdY90NaOXm5kpz993u3bsrvU6gMu6e+upXvyrtjDPOkOaGsrZt22a/T9IBLXfvduvWTdrEiROludeC7t5zZ66kZ8fawDvEAAAAAIAocSAGAAAAAESJAzEAAAAAIEociAEAAAAAUaqRxYMNGzZIe/jhh6V97Wtfk+YGptygQfPmzaX17NlTmvuF8d69e0tzw0NuFMsNJJSVlUkLIYTS0lJpLVq0SPS93S+r5+XlSXO/mO6GUtwYWNJfpkfd2rRpk7Q5c+ZIGzRokDQ39DF06FBpr7zyirR3331Xmnu8tmnTJtG1fOYzn5FWXFwsLYQQCgoKpLlRFjcMMX/+fGm//e1vE32cG9/ivkBVucdvfn6+NPfc6H421wZ3ja1bt070uYxlNXxJnzvc6Job3HE/v+uKG3l0P9fdx3Xu3DnR9zh8+LC0hQsXSps5c6a06gxYouFx948bFR03bpw09xrm5ZdflrZmzRr7vdu1ayeta9eu0kaMGCHN/Xw499xzpbnz2sqVK6W5QeDt27dLqyu8QwwAAAAAiBIHYgAAAABAlDgQAwAAAACixIEYAAAAABClGhnVciMhzzzzjDQ3OnXSSSdJc+M8p59+ujQ3yJWTkyPN/QK7+8Vy19www+DBg6WFEMLYsWNt/0du2MGNZbm/rlOnTpU2ZcoUaYw4ZA/393nevHnSLrnkEmm9evWS9slPflKaG0ZxQ1tu3G3AgAHS3NCWG/Fp1Mj/M7f9+/dLmzVrlrSnnnpKmhvLcuNb7p4Cqsr9vL7iiiukTZo0SdpPfvITae7eXrt2rf3eSR/DbsRk9OjR0tx968ZXFi1aJI2hrYalQ4cO0twIonv8ZwM38pV0QNF93LJly6R9/etfl7Z8+fJE3wM4lhvRbdq0qTQ33jhx4kRpu3btst+nb9++0tyZxp2v3BCfG5srKSmR9sADD0h76aWXpGXydRvvEAMAAAAAosSBGAAAAAAQJQ7EAAAAAIAocSAGAAAAAESpRka1HDcU9O6770rbuHGjNDfesXr1amkjRoyQ1rp1a2lJfzm8sgGgpB/nuhu3OnjwoLQNGzZI27Fjh7QXXnhBmvtriOzhHu8zZ86Udtddd0n72te+Ju2UU06R5oax+vXrJ80NkSS9L9zj2t3zIfjRPTeg5e57xrJQl9zAjhvBKiwslHbfffdJW79+vbTnnnvOfu+lS5cmucRw1llnSZswYYI097PG3YtucI9RrYZl9+7d0p599llpbdu2lda9e3dp1RnfcveYey6q7td0j+FDhw5Jc89bbhTo7bffluZGhoBjucehGzJ0r3/cvXf99dcn/t7u9Zy719zrrK1bt0qbMWOGtEceeUTa//zP/0hz58RM4h1iAAAAAECUOBADAAAAAKLEgRgAAAAAECUOxAAAAACAKKXc8MD/+x9Tqcr/x1rkfunbjWC5sSw3quXGt4YMGSKtf//+ia6lMnv27JH26quvSnOjKnPnzpW2c+fORJ9bn34xPZ1OV28JI0vVxb3i7gE3oDV58mRpxcXF0tz906SJ7uy5x6EbFXLDQFOmTJEWgh8tcaNcDVmM90qmnlNqmhshadeunbShQ4dKu/zyy6WNGTPGfp+kQ0XueWr79u3S3Ajf888/L82Np2QK90ndOeGEE6SNGjVK2sSJE6UNHDhQmnvOcm3z5s3S+vbtK62yoS03QOpGitwAaUlJiTQ3arl8+XJpvPbKvIbynNKsWTNp7t5zzynujNOiRQv7fdxzhTu7zJkzR9qCBQukvf7669K2bNki7cPOmnWtsnuFd4gBAAAAAFHiQAwAAAAAiBIHYgAAAABAlDgQAwAAAACiVC9HtWqaGwpq06aNtFatWlXr+1RUVEjbtGmTtMOHDyf63GzEsEOdf19phYWF0rp16yZt2LBh0vLy8qQtXbpU2uLFi6W9//770mIbyqqKGO+VhvKckpQbxXL3pxspCiH5c9LRo0elubGgF154QVp9GgZyuE8yyz2GCwoKpLnHan5+vrRTTz1Vmhv1caNalXHPUX/84x+l7du3T9revXulZePrsRjvkxDq171S05IODHfo0EGaO/dUJunZxT1XuOee+o5RLQAAAAAAjsGBGAAAAAAQJQ7EAAAAAIAocSAGAAAAAEQpilEt1B2GHbJH0tEFN5qQjUMK9U2M90o23id1oSoDKEk1lPuW+6RhcSNdjhuMrExDeaxXR4z3SQgN+15B7WBUCwAAAACAY3AgBgAAAABEiQMxAAAAACBKHIgBAAAAAFGq+SUPAFmhoqIi05cAIHAvIh5HjhzJ9CUAgOAdYgAAAABAlDgQAwAAAACixIEYAAAAABAlDsQAAAAAgCil0ul0pq8BAAAAAIA6xzvEAAAAAIAocSAGAAAAAESJAzEAAAAAIEociAEAqEdSqdTjqVQqnUqlHs/0tQAA0NBxIAYAAAAARIkDMQAAAAAgShyIAQAAAABR4kAMAAAAAIgSB2IAAOpYKpWamEqlXk+lUntSqdSuVCr1ViqV+kIqlUpl+toAAIhJk0xfAAAAsfjgwPtfIYRrP0jpEMLOEMKZIYR/CiGMCCGUZ+bqAACID+8QAwBQd24K//8w/MMQQvt0Ol0QQigIIdwdQvhMCOGSzFwaAADx4R1iAADqQCqVahZCuOuD//pkOp2+6e//Wzqd3hVCuOeDj/l6Jq4PAIAY8Q4xAAB145Phb+8EhxDCv1fyMfeFEA7WzeUAAAAOxAAA1I0zP/jPtel0eoX7gA/eKX6n7i4JAIC4cSAGAKButP/gP9d/xMetq+0LAQAAf8OBGACAupXO9AUAAIC/4UAMAEDd2PzBf3b5iI/rXNsXAgAA/oYDMQAAdePtD/6zayqV6uU+IJVK5YcQzqi7SwIAIG4ciAEAqBszQwg7Pvi/76jkY74WQmheN5cDAAA4EAMAUAfS6fSBEMI3P/ivV6dSqUdSqVRhCH97ZziVSt0RQvhGCGFnpq4RAIDYpNJptj0AAKgLqVSqUQjh8RDCVR+koyGEXSGE/BBC4xDCb0MI5SGEq0MIT6TT6Wvq/ioBAIgH7xADAFBH0un00XQ6/c8hhH8OIbwZQjgQQmgSQlgYQrg+hHBlBi8PAIDo8A4xAAAAACBKvEMMAAAAAIgSB2IAAAAAQJQ4EAMAAAAAosSBGAAAAABxG3IOAAAgAElEQVQQJQ7EAAAAAIAoNfmw/zGVSjFBjSpJp9OpTF9DJnCvoKpivFe4T1BV3CfAR4vxPgmBewVVV9m9wjvEAAAAAIAocSAGAAAAAESJAzEAAAAAIEof+jvEqB2NGuk/h0il9P9L+5EjR+ricoCs16SJ/ig7evRoogbgf3PPUa5xj6G29OjRQ5r7OR9CCGvXrpVWXl5e05cEIAF3n6bT+qve9e2MwzvEAAAAAIAocSAGAAAAAESJAzEAAAAAIEociAEAAAAAUUq5X3T+f/8j/8LrWnHRRRdJ6969u7Tf/e530rZu3Vor11RT+JfDo7a5sZXnnntO2p/+9Cdp//7v/y6trKysJi6rymK8V7hP6h83llVcXCxt6NCh0ubNmydt/vz50qoztMV90vC5EZ7FixdL6927t/38X//619Kuu+46afVtxKcmxXifhBDfvZJJbvzXnV3Gjx8v7fDhw9Lcfbt9+/bjvLrkKrtXeIcYAAAAABAlDsQAAAAAgChxIAYAAAAARIkDMQAAAAAgSrpkgBrVrl07aT/72c+ktWrVStrpp58u7fOf/7y06gyWANnGDbCcfPLJ0tz4VklJibSHH364Rq4LyEZuVMsNaH3961+XtmfPHmmvvfZazVwYouFew5SWlkrr27ev/fzLLrtM2j333CMtUwOKQENQWFgo7c4775T26U9/WtqGDRukLViwQFpdjGpVhneIAQAAAABR4kAMAAAAAIgSB2IAAAAAQJQ4EAMAAAAAosSoVi3r2rWrtIKCAmmpVEpaXl5erVwT0NC4+2f//v3SGFUB/recnBxprVu3lnbkyBFp06ZNS/RxwIdxo1r333+/tHHjxtnPd49hNwDEz3/g+LmzS1FRkbQWLVpIa9++vTQ3HLx48WJpdfWcwjvEAAAAAIAocSAGAAAAAESJAzEAAAAAIEociAEAAAAAUWJUqwY1aaJ/OS+88MJEH+fGHh544AFpbnwCiF06nZa2du1aaW6wAYjZSSedJG3kyJHStmzZIm3Dhg21ck3Atm3bpB0+fNh+bPPmzaVddNFF0t55553qXxgQgUaN9P3S/v37S2vTpk2ir7d9+3ZpCxculJbJUUbeIQYAAAAARIkDMQAAAAAgShyIAQAAAABR4kAMAAAAAIgSo1q1zP1iuuPGInbt2lXTlwNkvVatWklzo1pr1qyRtmfPnlq5JiAbNGvWTNr48eOluaEtN1JX2cgRUF3r16+Xtm7dOvux7vHqHtf333+/tIMHDx7H1QENW25urrSioiJpSUe1ysvLpe3evbvqF1aLeIcYAAAAABAlDsQAAAAAgChxIAYAAAAARIkDMQAAAAAgSoxqAai3mjTRH1FDhw6VlpOTI61v377S3CDXli1bjvPqgOzSsWNHaRdffLE0N6jCIB3qkhtKPHr0aOLPdz//u3TpIm3FihVVuzAgAm6obtSoUdLcazR3n5aWlkqrb8PBvEMMAAAAAIgSB2IAAAAAQJQ4EAMAAAAAosSBGAAAAAAQJUa1AGSV/Px8aalUSlqjRvzzPuBYbgClZcuW0o4cOSLtueeek1ZRUVEzFwb8A/cYnD9/vv3YPn36SHPPCW5UEYide14477zzpLmhLcfduyUlJdJ27tyZ6OvVFV4xAgAAAACixIEYAAAAABAlDsQAAAAAgChxIAYAAAAARIlRLQC1xo01hBBCly5dEn3svn37pDVu3Lj6FwZEyN07bnzIjZ0sWbKkVq4JcNxjtbi4OPHnu+cT9/nvvPNO1S4MaGBycnKkde7cOdHHOWvXrpW2cOFCaW58K5N4hxgAAAAAECUOxAAAAACAKHEgBgAAAABEiQMxAAAAACBKjGrVY4wHoT5wozvdu3eXdsMNN0gbMWKE/Zp9+/aV5gYb3LhPo0b8czzgePTr109a69atpW3fvl1aaWlprVwT4Bw9elRaZY/BPn36SKuoqJA2Y8aM6l8YkMXc67mTTz5Z2siRI6W5oTp3nz377LPSZs+eLc3d45nEK0sAAAAAQJQ4EAMAAAAAosSBGAAAAAAQJQ7EAAAAAIAoMapVy3bv3p3o49ywiRtAWb58ebWvCaiKE088Udrdd98t7dOf/rS0Xbt22a+5bt26RN+7W7du0nJzcxN9LoD/beDAgdJatmwpbd68edJ27NhRK9cEOG5wp6SkxH7shAkTpLnxxU984hPSeE2FbOJGsQoLC6W5M0Vl/V//9V+luaE6J51OS3NjqOXl5Ym+XibxDjEAAAAAIEociAEAAAAAUeJADAAAAACIEgdiAAAAAECUGNU6RkFBQaJWUVFhP3/jxo3S3DjJkSNHpLlfdB80aJC0KVOm2O8N1IQePXpIe/HFF6X17NlT2m9+8xtp99xzj/0+ZWVl0txYxIUXXijtpptukjZixAhp7p5yo0IrVqyw1whkM3c/tWrVSpobRXHjRYxqIZu4Ua3u3btn4EqAj9asWTNpHTt2lNa2bVtpN9xwg7SzzjrLfh83StqpU6dEH+e454WlS5dKcyN59Q3vEAMAAAAAosSBGAAAAAAQJQ7EAAAAAIAocSAGAAAAAEQpilEtNy4yduxYaTfeeKO04uJiaYcOHbLf56GHHpL2+OOPS3O/hO5+Ud4NCt17773S3EgX8FE6dOgg7XOf+5w0N6C1ZcsWaW5A669//av93m7Ix5kxY4a0M844Q1rSUa0BAwZIY6gOmdakSfKnYjdO4lphYaG0cePGSXP34u7duxNfD1BXSktLbU96T/z+97+v8WsCPowbp+rTp4+08ePHS7voooukuZ/r1RnFqory8nJp7jXa66+/Lo1RLQAAAAAA6ikOxAAAAACAKHEgBgAAAABEiQMxAAAAACBKDW5Uyw1onXjiidIefPDBRB/XuHFjaZUNoAwbNkzatGnT7Mf+Izdssn379kSfC3yUZs2aSXNjCAMHDpQ2ffp0aXfddZe0FStWHOfVVc6NRVxwwQWJPtfdu6eeeqq0Ro30nwtmwwAEMsc9ZtyIiRuucx83ZswY+31atmwpraSkRNqyZcukdevWTVrXrl2lrVu3Ttq8efOkVVRU2GsEsgUDpKhNeXl50m6++eZE7YQTTpC2detWaW6w6sknn5TmzkIhhDB06FBp55xzjjT3HLdmzRpp9913n7TNmzfb713f8Q4xAAAAACBKHIgBAAAAAFHiQAwAAAAAiBIHYgAAAABAlDgQAwAAAACi1OBWpt1K7hNPPCGtd+/e0tyy7KZNm6Q99thj9ns/9dRT0nJycqS1bt1amls/dGuirCTio7h1wT59+iRqhw4dkvaLX/xC2tKlS4/z6qrGLbrn5+cn+ly3kjh48GBpbvX3wIEDib4HGha3TN69e3dpbq28qKhI2ogRI6S5x2/nzp3t9bjnj507dyZq7nPd/eTWo/fu3WuvB6gN7jmrZ8+e0q666ir7+fybAlDXCgsLpX3hC1+QdvXVV0tzi9KrVq2Sdu+990pz/3aQHTt2SGvfvr20EPwZKandu3dLKy8vP+6vV9/wDjEAAAAAIEociAEAAAAAUeJADAAAAACIEgdiAAAAAECUsnpUy/1i+oQJE6S58aCDBw9Ke+mll6S5oaznn3/eXo/75fJevXrZj02CYRMcDzea8Pjjjyf63AcffFDatGnTpKXT6Spf1/FwfxY3SudGWdw1du3aVZobNFqxYkXSS0QWaNasmbTzzz9f2qRJk6SNGTNGmhvsWblypbRZs2ZJc8MkVeGGv4qLi6UNHTpUmrtPunTpIu2+++6T9v3vf1/anDlzKr1OwD3e3M90N5Z1/fXXS2vRokXNXBhQBe5x3KNHD2nXXnutNDfKWFZWJu3uu++W9swzz0hzg5/uOeGss86SFkIIQ4YMkeZG6dzY4quvvirNDQ9nK94hBgAAAABEiQMxAAAAACBKHIgBAAAAAFHiQAwAAAAAiFJWj2qdd9550twoSm5urrRf//rX0u655x5pq1atknbkyJGklwjUKjcWNH78eGknn3yytHfffVeaG3GoqwEtZ8CAAdLcqFZSbijCNWSv5s2bS3Nji27ExA2lLFu2TJq7T6ZMmSLtvffek3b48GFpVeHGJFu2bCntnHPOkbZmzRpp7hpPPPFEaaeccoo0RrXwd02a6MvJW2+9VdrXv/51aW7UZ/369dLcKGII/jUeUFPatWsn7cYbb5TmHp9ubPfJJ5+U5p4/3ICWe6z37dtXmhuqCyGEwsJC2//Rjh07pC1YsECa+/NlK94hBgAAAABEiQMxAAAAACBKHIgBAAAAAFHiQAwAAAAAiFLWjGq50YWrr75amhtFOXjwoLS77rpL2sqVK4/v4j6EG+xxoypNmzaV5oZSgGONHj1a2pe+9CVpOTk50j7/+c9LKykpqZkLq6f+8pe/SFu3bl0GrgS1xQ1oucFE91xRVlYm7Z//+Z+luUE6N4BSXalUSlrPnj2lnXvuuYm+nvvrMHPmTGlumGz79u2JvgcaPvd67LHHHpPmXqO5YbdLL71UmvtZ/Z3vfMdez1e+8hXbgapyj+2ioiJpF1xwgTQ3eOWeU2bMmCHNnRXat28vzb3mmzx5sjQ3ghhCCI0bN7b9H7lRrdLSUmlHjx5N9PWyAe8QAwAAAACixIEYAAAAABAlDsQAAAAAgChxIAYAAAAARCmrR7X69++f6OM2bdokbcOGDTVzYR9h8+bN0tz1uIGX4cOHS2vSRP+WuV/GRxwGDBggrU2bNtLcQMLq1aulpdPpmrmwGuKGVXbt2iWtbdu20tx98eqrryb6OGQH9/PQDYy4ISo3duLGFpcsWSKtru6Tdu3aSXMDQn369JG2bds2aS+88II09xwFfJiLL75Y2vjx46W55x03CrRixYpE33f37t2JPq4ySQeFgJqSn58vbeLEidLGjBkjzb2+O/vss6W55wk3yFhZd8NY7vXh3r177ddsKHiHGAAAAAAQJQ7EAAAAAIAocSAGAAAAAESJAzEAAAAAIEpZM6pVHXl5edLcCMny5csTfb2uXbva7ka+ioqKpHXq1CnR92nZsmWij0O83IicG03YuXOnNDd4kklu8MSNSrj72Y1CzJ07V9rzzz9/nFeH+qhLly7S3M9hNzB16623SpsyZYq0uhjQ6t27t+1u5OvSSy+Vtn79eml33nmnNAa0UFXNmzeX9t3vfldaYWGhtHvuuUda0gEtpyqjWm5wzw0XJX3dhzi41xILFiyQ9uKLL0q77LLLpBUUFEi7/vrrE12Ley3nno/cfXHCCSfYr5mTk5Po+7ihLvezoCHhHWIAAAAAQJQ4EAMAAAAAosSBGAAAAAAQJQ7EAAAAAIAoZc2olvtF95KSEmm9evWS5sYennjiCWkzZ85MdC0XXXSR7W3atJHmhrHcL7UDtcndP3XBjTWEEELPnj2lnX766dImTZokrVmzZtK2bt0q7ac//am0+fPn2+tBdnLDOa7t27dPWllZmbSavk/ctXTr1k3aQw89ZD9/1KhR0txA3s9//nNpbiAMqKrOnTtLc8OiFRUV0n7729/W6LXMmDHD9vvvv1+au/eGDx8u7T//8z+lHThwQFqmnkOReVu2bJF27733Jvrc0aNHS2vRokWi77F9+3Zpc+bMkbZmzRpp1113nb0eN1TqXqe5+/60006T5p5Hjxw5Yr93fcc7xAAAAACAKHEgBgAAAABEiQMxAAAAACBKHIgBAAAAAFHK6lGtn/zkJ9LcYNXIkSOlDRw4UFq/fv2kpdNpaYcPH7bX+N5770l78sknpZ1xxhnSiouL7dcE/s4NH/To0SPR565du1aae2w7jRrpPzfLzc2V1rdvX2ljx461X3Py5MnS3nrrLWmPPvqotH/7t3+T5sbr3JjF7NmzpblBLmSHXbt2SXOjUx06dJB2ySWXSFu3bp20bdu2SWvdurU0NzR07rnnSnM/6y+44AJpIfixlG9961vSpk+fLm3//v32awJV4cayXHPPCUOGDJH27rvvSkv6XOR+pofgB7QcN4jqxlnffvttaddff700N3qEhsc9PleuXCnt1ltvlTZt2jRp3bt3lzZv3jxp7vG1adOmSq/zWLt377bdDdC1b99eWkFBgbTLLrtM2ty5c6W5gbBswDvEAAAAAIAocSAGAAAAAESJAzEAAAAAIEociAEAAAAAUcqaUS3nlVdekfbaa69JO+WUU6T99Kc/leZGi9xYiRvKCiGE999/X1p5ebm0O++8U9rZZ59tvybwd27YIekg1KBBg6Sddtpp0twgkRugKyoqkjZu3DhpbpghBD+s8thjj0lbunSptIkTJ0o7/fTTpZ111lnS3BgSo1rZa8eOHdJWr14tzT0+3EiOG1Z0j8H+/ftLGzx4sLR27dpJc2NXbnglhBCeeuopaS+//LK0AwcO2M8HqssNzZWWlko788wzpbmf6W4I8pe//KW0PXv2JLxC/9zoxiCTDlM2a9ZMWuPGjRNfDxo+95jbvHmztGeffVaaexweOXKkZi7sAzNmzLDdDcu5127u8e6e91q1aiWNUS0AAAAAALIIB2IAAAAAQJQ4EAMAAAAAosSBGAAAAAAQpawe1XLciNWSJUukXXbZZYm+3saNG6UdPHiw6hd2jN27d0tzoyjul/aBY82bN0/aJZdcIq1Xr17SFixYkOh7uLEH9xh+4YUXpD3yyCP2a7pRloqKCmk5OTnSnnvuOWluNAkNn3vMXH755dJuu+02aW4Yzg3NubEsN+blxhbdY3Xt2rXStm3bJi2EEI4ePWo7UFfcPXbxxRdLe/HFF6Wdeuqp0tyo6C233CJt37590tq0aWOv0Y0UuddP7rnMjeZdc8010rJ1KAiZlamf4ZU9p7ihRnefduzYUdpbb70lzQ2xZiveIQYAAAAARIkDMQAAAAAgShyIAQAAAABR4kAMAAAAAIhS6sOGm1KpFKtOtSA3N1faSSedJO2MM86Q5oZb6tPwSjqd1nWLCGTqXmnWrJm08ePHS/vKV74irVWrVtLcY2n69OnS3CDX/PnzpVU27JB0MK5RI/1ndhdeeKG0++67T9prr70mzY0rbd26NdG11LQY75W6uE/cwE5hYaG01q1bS8vLy0v0PdyQSG0MMIL7JFu0b99e2lVXXSXtiiuukNa1a1dplQ1oOe75xI18Pf7449JmzJghLRvv2xjvkxCy817JJPeacdSoUdJOPPFEac8++6y01atXS6vvg8CV3Su8QwwAAAAAiBIHYgAAAABAlDgQAwAAAACixIEYAAAAABAlRrXqscaNG0s7cuRIBq4kOYYd6qcmTZoc9+dWVFTU4JXUDvfncwNhDNBlVn2/T1D/cJ80LO5ntRvQcqOPVbF27Vpp5eXl1fqa9VmM90kIDfteqStuvNSNU9b380dSjGoBAAAAAHAMDsQAAAAAgChxIAYAAAAARIkDMQAAAAAgSoxqoUYx7AAkE+O9wn2CquI+AT5ajPdJCNwrqDpGtQAAAAAAOAYHYgAAAABAlDgQAwAAAACixIEYAAAAABClDx3VAgAAAACgoeIdYgAAAABAlDgQAwAAAACixIEYAAAAABAlDsQAAAAAgChxIAYAoJalUqm7U6lU+oP/ty+VSn3sQz62xzEfO7wOLxMAgOhwIAYAoG6dEEK4K9MXAQAAOBADAJAJn02lUidn+iIAAIgdB2IAAOrO2hDCkhBCkxDCtzN8LQAARI8DMQAAdedoCOHfPvi/P5VKpf4pkxcDAEDsOBADAFCH0un0CyGEuR/81/syeS0AAMSOAzEAAHXv1g/+c0Qqlbogo1cCAEDEOBADAFDH0un0WyGEKR/81++kUqlUJq8HAIBYcSAGACAzvhFCOBJCGBxCuCLD1wIAQJQ4EAMAkAHpdPp/Qgi/+OC/fjOVSuVk8noAAIgRB2IAADLn7hDCgRDCiSGE6zN7KQAAxIcDMQAAGZJOp9eHEH7wwX+9PZVK5WXyegAAiA0HYgAAMus7IYQdIYT2IYSbM3wtAABEhQMxAAAZlE6nd4b//+8jvjn87WAMAADqAAdiAAAy7/shhHUhhJYhhNszfC0AAESDAzEAABmWTqcPhr8NbIUQwsUZvBQAAKLCgRgAgPrh8RDC/2T6IgAAiAkHYgAA6oF0On0khPCNTF8HAAAxSaXT6UxfAwAAAAAAdY53iAEAAAAAUeJADAAAAACIEgdiAAAAAECUOBADAAAAAKLU5MP+x1QqxeIWqiSdTqcyfQ2ZwL2CqorxXuE+QVVxnwAfLcb7JATuFVRdZfcK7xADAAAAAKLEgRgAAAAAECUOxAAAAACAKHEgBgAAAABE6UNHtQAgWzVqpP+87+jRoxm4EgAAANRXvEMMAAAAAIgSB2IAAAAAQJQ4EAMAAAAAosSBGAAAAAAQJUa1MiA3N1da27ZtpW3dulVaeXl5rVwT6r/GjRtLS6fT0hr6cJS7fyZMmCDtkksukfbrX/9a2ksvvSSN+wz43woKChK1iooKaatXr5bmfnah4XPPYy1btpR28ODBRK2+SXqf7Ny5U9q2bdukcZ+gvnBDpa4l5Z4rMol3iAEAAAAAUeJADAAAAACIEgdiAAAAAECUOBADAAAAAKLEqFYtSzoA9NnPflbaz3/+c2lTp06V5oYmGGLIbm5kbdKkSdL++te/Sps5c6a0bBgjSeqkk06S9rWvfU3agAEDpPXs2VPakiVLpJWVlR3fxSFaqVRKWmFhobQuXbpIq2xc5L333pNWF4Nvbijlxz/+sbSLL75Y2uHDh6Wdc8450tx9h4blhBNOkDZq1ChpY8eOleZ+Bj/55JP2+2Tq53WTJvoS+sorr5T2xS9+UdqCBQuk3XbbbdLcuCpQ29zZ5ZOf/KS03r17S8vLy5O2d+9eaa+88oq0jRs3StuyZYu02jjj8A4xAAAAACBKHIgBAAAAAFHiQAwAAAAAiBIHYgAAAABAlBjVqmVJB4AGDhworUePHom+xxtvvCGNUaDs0bhxY2lXXHGFtG9+85vSNmzYIO2uu+6S9pvf/OY4r67+Oeuss6S5sSz317V169bS3DAKUFWDBg2SNnnyZGluYGrPnj32az744IPSnnnmGWk1PbTlRrXcc5QbXnHtvPPOk7Zs2TJplY2LoX5xj4+CggJpbkzq2muvlda9e3dpBw4cSPR9Q/D3yf79++3H1qScnBxp7nVbnz59pLnnHff8xKgWMsE9Zu+55x5pblTLPQe4n+0TJ06U9uc//1navffeK2316tXSqot3iAEAAAAAUeJADAAAAACIEgdiAAAAAECUOBADAAAAAKLEmkwNciM+SQeA3FiE+7jPfe5z0lq0aCHtF7/4hbQjR45IQ+alUilpbqDE/X12gwZufGvq1KnS3GhJNpg/f760TZs2ScvPz6+Ly0GEmjdvLu0HP/iBtI9//OPS3ODI0aNH7fe57rrrpM2bN0/a+vXr7ecfL/czqbJBI8Snbdu20oqLi6XdcMMN0jp27Jjoe+Tl5Um76qqr7McuXLhQ2vPPPy+tsvvseHXo0EHa8OHDpbnXhnv37pXGqBwywf1s79Wrl7SPfexj0txzoft6rhUWFkpzA3udO3eWtmbNGmnpdFpaVfAMBwAAAACIEgdiAAAAAECUOBADAAAAAKLEgRgAAAAAEKV6OarlBj2aNm0qrVOnTom+nhscOXz4cNUv7CO4wYYVK1ZI27NnjzQ3AORGj9ygyptvvimNAa14ueGDk08+WdqSJUukVXeUoC6Ul5dL2717dwauBDFwA0JuLKioqEiaGxJxz2/uMR1CCC+88II0NyBX09yISZcuXRJ9rhsGcn8OBoSygxuBGz16tLRx48ZJa9++faLv4V4TubGebt262c8fOnSotFmzZkmr6SFJN37p2sGDB6VNnz5d2saNG2vmwoAqcPe4ez5zj203GLdt2zZp7vXmj370I2nuzLR69WpptfFalXeIAQAAAABR4kAMAAAAAIgSB2IAAAAAQJQ4EAMAAAAAolQvR7W6d+8ubciQIdK+8pWvSNu7d2+ij6uNQSE3oDJ48GBprVu3luYGuWbPni3tqaeekuZGwxAHN9DTrFkzaX/4wx+k3XzzzdKee+65mrmwWuQe726054wzzqiLy0ED4gZChg0bJu2LX/yitKQDWk5lQ1lz586VVhdjVO45Kicn57i/HiOP2atly5bShg8fLm3gwIHS3D3hXuusWrVKmhvQcuOjIYTQs2dPaXl5edKqM6rl/izutWqLFi2kudelpaWl0mpj7BXZwT1X1MZwlHuOc/e4u6fcPeC4kdOSkhJpr7/+urSdO3dKq2x0sqbxDjEAAAAAIEociAEAAAAAUeJADAAAAACIEgdiAAAAAECUMj6q5X7B+5JLLpF2xRVXSHOjOYcOHZL21a9+Vdodd9whraysrLLLTMT9Avxf//pXaW5ApWvXrtIWLVokbe3atdIYYsBH6dGjh7ShQ4dKc+NU9W0Qx91nbqgFqKrCwkJpl19+ubR27dpJSzqg5bghnhD888LChQulVefx7667ffv2iT7OqY0hGGROQUGBNDdy6h6rjnud9atf/UqaG65zY28hhNC/f39prVq1krZly5YEV+jl5uZKKyoqktamTRtpbjTMjQzVt+daVJ8bOe3YsaO03r17S3Pjv7t27ZJWldEp9xxXXFws7ZxzzpHmRrXcY/add96R5gYi3Vkok6/leIcYAAAAABAlDsQAAAAAgChxIAYAAAAARIkDMQAAAAAgShkf1XLceIf7ZW73cU2bNpX2iU98QtrAgQOlrVmzRlpVfsHbfWxpaWmi1rlz50Rfj8GSOOzevfu4P9fdF02a6K1+4403SnOjbY899pi0ioqK47y66nP3hRufYGgLVXXVVVdJGzNmjDQ3Blkd+/bts3316tXSavpx7f4sbqQo6Z95/fr1iRrqH/f3eNCgQdLccJT7XPe4fvLJJ6W5McfPfOYz0ip7/eNeH9a0Dh06SBsxYoQ098XX2WsAABNcSURBVNdhz5490qoyhITs4F57udf2//RP/yTt3HPPlebGFhcvXizNPU9Udq+4Qa8BAwZIy8/Pt5//j9yorzvjuIHh+vYajXeIAQAAAABR4kAMAAAAAIgSB2IAAAAAQJQ4EAMAAAAAopTxUa0jR45Imz9/vrQuXbpIGzx4sDQ3HtSjRw9pQ4cOlTZ79mxplY2dJOVGitwvxY8cObJa3wfZyw1UzZ07N9HHuce7G3ZwcnNzpZ133nnS3OCJG0ioq8E3N8Tg7rP6NtiA+sWN3wwZMkSau08c91zmuAEgN0wSQvXG9ZLKycmR5oZgknJjQZX9+VC/dO/eXdrEiROlFRQUSHN/j6dOnSrtiSeekLZjxw5pCxculHbqqadKqyvu50DLli2luZ8Dr776qrRNmzbVyHUhM9zPcXdf3HbbbdLOOeccae6cMmzYMGlvvfWWtLvuuktaZY8v9xrvoosuktasWTNpBw8elPbuu+9KmzJlSqKPq294hxgAAAAAECUOxAAAAACAKHEgBgAAAABEiQMxAAAAACBKGR/Vcnbt2iVtz549iT7XDQq54aEJEyZIW7RokbSnn35amhsNqUzr1q2l9evXT5r7BX3Ea/v27dLcaIl7bLtxK9fcvTJq1ChpbhTi1ltvlbZ582ZptcH9mYuLi6W50STg79yAkBs7Sfqz+bXXXpOWdLgrk0466SRpSUce3XPhm2++KY2Bu/rH/Ry95JJLpLnHgns+Wb9+vbTvfe970lavXi3N3YtuNLWye9ENw7nXXu45z/1Z3H07cODARN/DfT33mpahuezmHiNuVKuoqEiae7y7+7FXr17SOnXqJM0N0L300kvSQvCDiYWFhfZj/9H7778vbebMmdLee+89aW6Qq77hFAYAAAAAiBIHYgAAAABAlDgQAwAAAACixIEYAAAAABClejmq5X5ZvaZHp3r06CHtq1/9qrSlS5dKc78wHkII+/fvl9aqVStp/fv3l8YAEI7lBkrWrVsnzQ3iHDlyRNq3vvUtaVdddZW0nj17Srv88suluQGI6667TlpVBuiqIz8/X5obUEGc3OiOe/y3b98+0ddzAyEPP/ywtDvuuENaJgcU3fe+/fbbpbnnKMcNJH3nO9+RxqhWdnA/R3Nzc6Xt2LFDmhsldY8PNzq1b98+aWVlZdJOO+00aSH4oaEbbrhBmhuD3Lp1qzQ3jnTZZZcl+jg3iOleR3JPZDf3GigvL09a06ZNpSV9beI+zn29Nm3aSHOPzRD8a7wWLVokup6dO3dK27hxo7RsHYzjHWIAAAAAQJQ4EAMAAAAAosSBGAAAAAAQJQ7EAAAAAIAoZXxUy4189OvXT5ob+XCf6wYb3MiQG7E65ZRTpLmhrRdffFFaCCE8/fTTib5PJkdVkB3c47g6nztt2rREH3fnnXdKc6Mq48aNk/bQQw9JW7JkSeJrBGpL9+7dpZ1++umJPtc9VpcvXy5t1apV0lq2bJnoe9QV99wzcOBAae55y40AubEgN7iE+scNzSV9vLq/xyUlJdLcCI/jhqh+97vfSRs6dKj9fDeGN3r0aGnuefDll1+W1rVrV2nu54W7n9auXStt8eLF0hjVym5uyG3EiBHSOnToIM0NcrkBLffc4342Dxs2TJobzwohhOLiYmlugKuiokLanDlzEjX3udmAkxkAAAAAIEociAEAAAAAUeJADAAAAACIEgdiAAAAAECU6uWo1oABA6QlHdVy4wzTp0+X5sYZ3PCKGw9y41shhPCnP/1JmvuztG7d2n4+UBPcYIMbXfiP//gPaW445MILL5TWokULaX/4wx+k3XLLLfYan332WduTcGMkbryL0ZI4ueeFb3zjG9JGjRqV6Ott27ZN2iOPPCLt4MGDib5eJrnhlqQjj+Xl5dLeeOMNaYxqZQc3oOWGeNxjZvfu3Yla0vFEN3z63nvvSdu4caP9fDeq5f58PXr0kJafny/NvRZ0X8/9+crKyqTt2bNHGrKHuwfcEFXHjh2lufE6J+m94n5eu/PRiSeeaD+/VatWib6m+3nv7r+kw3nZgHeIAQAAAABR4kAMAAAAAIgSB2IAAAAAQJQ4EAMAAAAAosSBGAAAAAAQpYyvTDtu8SzpEqZbmXaLoG7B8KabbpJWWFgozS0VhhDCZz/7WWlu/bBNmzbS3MIcK7k4VnUeD25J88CBA9LmzZsnbcSIEdLy8vKkuYXSG2+80V7P66+/Lm3r1q32Y/+R++uwePFiaW4l0a1FutV393FJVyCRWbm5udLcv1WgWbNmib7emjVrpM2fP19a0ucoxy30hhBCRUXFcX9Nxz33JP23HrjnzJkzZ0qr6WtG9bl7YvTo0dLOPvtsaW41/NFHH5Xm7onqPBbc423WrFn2Y93Krlv3dY/1bt26SSsqKpLm7h23jl1SUiKtIS3xNnTu57hblL7hhhukFRcXS2vcuLG0w4cPS9u8ebM0d35wj2t3fdWV9Of9+++/X+PfO1N4hxgAAAAAECUOxAAAAACAKHEgBgAAAABEiQMxAAAAACBK9XJUqzrc8M3+/ful/fSnP5XmRlb+z//5P9Iq+wX2f/mXf5Hmfnm+SRP9y75lyxZpS5culcbQVhzcGMmMGTOk9e7dW5obcUjqhz/8obQ333wz0cedeuqp0s477zz7fR5++GFpd999t7SVK1dKc/f42rVrpa1bt06aG8SbNm2aNDeqUlZWJg2Z5cbP3MCOG85x3OPtm9/8prTVq1dLGzt2rLSkg1UvvPCC7evXr0/0+U7btm2l3X///dLatWuX6Os9+eST0pYtW1b1C0Od69SpkzQ3CuQeC25c54033pDmfgZXh3vttGvXLvux7nWRex4888wzpbnXc8OHD0/09dzrtiVLlkhz41uon9yolhsldc8zHTp0kOZer+zZs0faq6++Ku3888+X5h6v7prdc2Nl1+PunxUrVkjbuHGjNHefZiveIQYAAAAARIkDMQAAAAAgShyIAQAAAABR4kAMAAAAAIhSgxvVctwvjLsxhD/84Q/SLrroImluyCiEEFq0aCHNDXU5O3fulOYGSxjVioMbSXDDQJUNJxyv8vJyaW+//ba0Bx54QNrPfvYzac2bN7ffZ8KECdLckNitt94qbfPmzdLcUISTk5MjzY0PueE71D+FhYXS3FhQ0r+fU6ZMkTZr1ixpbiRnwIAB0pKOalX2cz3p49px4ytDhgxJ9LnuXpwzZ06ij0P94x4L7ueee7y5cc8dO3ZIq+nhKHdPuGup7Hrat28vzY08nnPOOdLc67ZDhw5Jc0OXr7/+ujRet2UP9zqruLhYmhvodK93Dhw4IG3VqlXSHn30UWnuTOEer+5ergp375aUlEiri/s+k3iHGAAAAAAQJQ7EAAAAAIAocSAGAAAAAESJAzEAAAAAIErRLse48Yjly5dLe+ihh6RdcMEF9mteeuml0nJzc4/j6v6mIf2yOqrGjWr1798/0cc51Rn1cENbU6dOlTZmzBhp7p4IwY9PjB49WpobKHniiSekVWd8CNnBDci5YRM3gOIcPHhQ2rx586S5URTH3Ys1PXpXGTcCVFRUJM2NKznbt29P1FD/uAG5ESNGSOvQoYO0pOM6bgS0poej3NdzzwchhDB//nxp48aNk+ZejzVt2jTR9axcuVKaG0Jyg62on9zPzUmTJkkbP368NDfo6B6zbpTRvX5av369tOq8rqnsc901uvt5yZIl0hr6mYR3iAEAAAAAUeJADAAAAACIEgdiAAAAAECUOBADAAAAAKIU7aiWc+jQIWlvvfWWtJycHPv5559/vrTqjGoBx6rOgNbSpUsTfVxSbmjo5ptvluZGK0II4VOf+pS09u3bS3vsscekHT58WNrTTz8tzY3B9OrVS5r76zpw4EBpblSlpodkULmxY8dKe/DBB6X17NlTWkVFhTQ3iDN9+nRpSYdNWrZsmejj3GOmtLQ08ce655TbbrtN2uTJk6W1aNFC2ooVK6Tdcsst0srKyuw1on5p06aNtCFDhkhzY1KrV6+WtnDhQmmZGtfZtm2b7b/97W+lnX322dI6duyY6Pu4P5/76+D+ejHwmD3y8/Oluef+Tp06Jfp67rXJ4sWLpS1atEha48aNpXXt2lWaGyStCvecsmfPHmnusd3QX+/wDjEAAAAAIEociAEAAAAAUeJADAAAAACIEgdiAAAAAECUGNU6hhtDcEMilf1i+Zo1a6S1bds20fdu6L+sjrrjHkurVq1K9HHV4QZPnnrqKfuxY8aMkebGIpo00R9Rt99+uzQ3mvSXv/xFmhtmct+je/fu0lKplDTUDjcwMmnSJGluQMv9fdq+fbu0BQsWSEs6iOMeM8OGDUv0cVW579xfBzfe+MUvflGaG9ByY3j33HOPtJdeeinpJaKecUM8gwcPluYe61OmTJE2e/ZsaZl6vVLZmNfcuXOludG8a6+9VlqHDh2kbdiwQdqvfvUrae7nCuonN57phteKi4ulFRYWSnOPxffff1/atGnTpL377rvS8vLypLkzhXtt4n7WV8Z9zfnz5yf6uIaOd4gBAAAAAFHiQAwAAAAAiBIHYgAAAABAlDgQAwAAAACi1OBGtWp67MGN9bhfnA/B//J8nz59En2fN954Q9rOnTsTfS7isHv37uP+XDfOU9PcyIS7J0II4corr5R24403Shs+fLi0Xr16SfvZz34mbd++fdLcyJG7x2fMmCGtskEX1Dw3YuLGTpI+rt3fz9dff73qF/Yh8vPzE32cG3fp37+//dilS5dKc8NAbryxvLxc2jPPPJOoHTx40F4P6hf3WHKjOy1btpR2+PBhaRs3bpTmHkf1za5du6RNnz5dWu/evaX17dtX2rJly6QtXrxYGs8J2cONLSYdqHL32f79+6W5kTX3Ot7de3v37pXmRn1PO+00ae3atZNW2WOzpKRE2quvviptz5499vMbMt4hBgAAAABEiQMxAAAAACBKHIgBAAAAAFHiQAwAAAAAiFJWj2q5Aa3S0lJpbnChOiobHPnlL38pzY3DbN26VdqTTz4pbdu2bcdxdWgI3NDT3LlzpZ166qnS3HDU0KFDpbmREPd9qyOdTtv+/PPPJ/p8N3rhhlFyc3MTtaQYS6k77vE6ceJEae3bt0/09Q4cOCDtm9/8prTNmzcn+nqOe+5xYyVuAM4NtIwdO9Z+n2uuuUZat27dpLnnlKlTp0q7/fbbpbm/XsgO7mdcUVGRtNatW0v7y1/+Im3mzJnSavo5oTa4kaIlS5ZIu+WWW6Tl5eVJcwNHvB7Lbu61yOrVq6W5saymTZtKmzVrljT3M3f9+vXSkt5TblRrxYoV0g4dOiStshHWX/ziF9LefPNNadkwplfTeIcYAAAAABAlDsQAAAAAgChxIAYAAAAARIkDMQAAAAAgSlkzquV+Id6NgbzxxhvSduzYUSvX9I/cL+jfeeed0twvq1c21IU4pVIpaW6gzX2c44ZDMsmNSkybNk2aG3t4+umnpZ155pnSko5qVTb8hczJz89P9HHu7927774r7f3336/2NR3LjWq5557zzz9fWosWLaQNHjzYfh93f7vBt5/85CfSfvjDH0rbsmWL/T7ITh06dJA2fPhwae7n7ZQpU6S5eydbuZ8NbnzONTQ87me2e82+c+dOaTk5OdIWLVqUqLnBq6TX54YanTZt2khzw3Ah+LG5mh4ezla8QwwAAAAAiBIHYgAAAABAlDgQAwAAAACixIEYAAAAABCljI9queGDbdu2SVu1apW0BQsWSPvv//5vaW5Qoja4Pwu/rI7j4R5LmzdvTvRxbpxh6dKliT4uk9yfZdOmTdJuvPFGaZMnT5Z22WWXJfq+y5cvl+aGNVD/uJGoRx55RFpdjBa65x43ljVu3DhpjRr5fza9YcMGaQsXLpT2X//1X9LcvYOGz71+cuOE7vHKuCdiUlZWJs09fzRt2lTas88+K839vE462ulej82fP1/aa6+9Ji3puGoIdXceyka8QwwAAAAAiBIHYgAAAABAlDgQAwAAAACixIEYAAAAABCl1If9wncqlUr22+A1rKCgQFrLli2l7du3T5oblEj6S+2ovnQ6nfy3+xuQurhXLr74Ymm33367tLy8PGkXXnihNDcokQ3cgES7du2kTZw4MdHXe+WVV6SVlJRIq+mfIzHeK+4+cYNSxcXF0oYOHSpt0aJF0l5++WVp5eXlia+xJuXm5krr1KlT4s93w1juz1LfBvJqEvdJ5XJycqR17tw50fdYv369tMOHDyf6XNQ/Md4nIdT8a6/GjRsn+rgjR47U5LdFHarsXuEdYgAAAABAlDgQAwAAAACixIEYAAAAABAlDsQAAAAAgCjVy1EtZC+GHepWkyZNEn1cRUVFLV8JqirGeyXpfeKGtlxzY1INeWAqRtwnwEeL8T4JgXsFVceoFgAAAAAAx+BADAAAAACIEgdiAAAAAECUOBADAAAAAKKUbJEHQL3EWBYaIsayAABAXeEdYgAAAABAlDgQAwAAAACixIEYAAAAABAlDsQAAAAAgCil0ul0pq8BAAAAAIA6xzvEAAAAAP5v+3UgAAAAACDI33qQyyJYEmIAAACWhBgAAIAlIQYAAGBJiAEAAFgSYgAAAJYCTwVbWV0RKXEAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 1440x1080 with 25 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "np.random.seed(100)\n",
    "n_samples = 5\n",
    "n_cls = 5\n",
    "n_label = len(np.unique(y_train))\n",
    "fig, rows = plt.subplots(n_cls,n_samples,figsize=(4*n_samples,3*n_cls))\n",
    "\n",
    "for row,cls in zip(rows,np.random.choice(list(range(n_label)),n_cls)):\n",
    "    row[int(np.floor(n_samples/2))].set_title(lebel_dict[cls],fontsize=25)\n",
    "    for col_ax,img_idx in zip(row,[i for i,v in enumerate(y_train) if v==cls][:n_samples]):\n",
    "        img = X_train[img_idx].reshape(img_rows,img_cols)\n",
    "        img = np.transpose(img)\n",
    "        col_ax.imshow(img,cmap='gray')\n",
    "        col_ax.axis('off')\n",
    "plt.subplots_adjust(left=0.2, wspace=0.02)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train shape: (697932, 784)\n",
      "X_test shape: (116323, 784)\n"
     ]
    }
   ],
   "source": [
    "X_train = X_train.astype('float32')/(255/2)-1 # for tanh\n",
    "X_test = X_test.astype('float32')/(255/2)-1\n",
    "print('X_train shape: {}'.format(X_train.shape))\n",
    "print('X_test shape: {}'.format(X_test.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_train shape: (697932, 62)\n",
      "y_test shape: (116323, 62)\n"
     ]
    }
   ],
   "source": [
    "y_train = to_categorical(y_train,n_label)\n",
    "y_test = to_categorical(y_test,n_label)\n",
    "print('y_train shape: {}'.format(y_train.shape))\n",
    "print('y_test shape: {}'.format(y_test.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def sampling(arg):\n",
    "    arg = [z_mean,z_log_var]\n",
    "    dim = K.int_shape(z_mean)[1]\n",
    "    epsilon = K.random_normal(shape=(K.shape(z_mean)[0], dim),mean=0.0, stddev=1.0) # reparameterization trick\n",
    "    return z_mean + K.exp(0.5 * z_log_var) * epsilon"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "image_shape = (img_rows,img_cols,img_channel)\n",
    "batch_size = 128\n",
    "latent_dim = 10\n",
    "epochs = 10\n",
    "learning_rate = 0.001\n",
    "learning_rate_reduction = ReduceLROnPlateau(monitor='val_loss',patience=10,verbose=1,factor=0.5,min_lr=0.00001)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Encoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "image_input (InputLayer)        (None, 784)          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "reshape_1 (Reshape)             (None, 28, 28, 1)    0           image_input[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_1 (Conv2D)               (None, 28, 28, 16)   160         reshape_1[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_1 (BatchNor (None, 28, 28, 16)   64          conv2d_1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_1 (Activation)       (None, 28, 28, 16)   0           batch_normalization_1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_2 (Conv2D)               (None, 28, 28, 32)   4640        activation_1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_2 (BatchNor (None, 28, 28, 32)   128         conv2d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_2 (Activation)       (None, 28, 28, 32)   0           batch_normalization_2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_3 (Conv2D)               (None, 14, 14, 64)   18496       activation_2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_3 (BatchNor (None, 14, 14, 64)   256         conv2d_3[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_3 (Activation)       (None, 14, 14, 64)   0           batch_normalization_3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "flatten_1 (Flatten)             (None, 12544)        0           activation_3[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 128)          1605760     flatten_1[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "z_mean (Dense)                  (None, 10)           1290        dense_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "z_log_var (Dense)               (None, 10)           1290        dense_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "lambda_1 (Lambda)               (None, 10)           0           z_mean[0][0]                     \n",
      "                                                                 z_log_var[0][0]                  \n",
      "==================================================================================================\n",
      "Total params: 1,632,084\n",
      "Trainable params: 1,631,860\n",
      "Non-trainable params: 224\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "w_init = initializers.random_normal(stddev=0.02)\n",
    "gamma_init = initializers.random_normal(mean=1.0,stddev=0.02)\n",
    "\n",
    "img_inputs = Input(shape=(orig_dimension,),name='image_input')\n",
    "\n",
    "x = Reshape(image_shape)(img_inputs)\n",
    "x = Conv2D(16,3,strides=1,padding='same',kernel_initializer=w_init)(x)\n",
    "x = BatchNormalization(gamma_initializer=gamma_init)(x)\n",
    "x = Activation('relu')(x)\n",
    "x = Conv2D(32,3,strides=1,padding='same',kernel_initializer=w_init)(x)\n",
    "x = BatchNormalization(gamma_initializer=gamma_init)(x)\n",
    "x = Activation('relu')(x)\n",
    "x = Conv2D(64,3,strides=2,padding='same',kernel_initializer=w_init)(x)\n",
    "x = BatchNormalization(gamma_initializer=gamma_init)(x)\n",
    "x = Activation('relu')(x)\n",
    "before_flatten_shape = K.int_shape(x)\n",
    "x = Flatten()(x)\n",
    "x = Dense(128,kernel_initializer=w_init,activation='relu')(x)\n",
    "\n",
    "z_mean = Dense(latent_dim, name='z_mean')(x)\n",
    "z_log_var = Dense(latent_dim, name='z_log_var')(x)\n",
    "\n",
    "z = Lambda(sampling, output_shape=(latent_dim,))([z_mean, z_log_var])\n",
    "\n",
    "encoder = Model(img_inputs,[z_mean,z_log_var,z],name='encoder')\n",
    "encoder.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Decoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "latent_inputs (InputLayer)      (None, 10)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "label_intputs (InputLayer)      (None, 62)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_1 (Concatenate)     (None, 72)           0           latent_inputs[0][0]              \n",
      "                                                                 label_intputs[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_2 (Dense)                 (None, 128)          9344        concatenate_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_3 (Dense)                 (None, 12544)        1618176     dense_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "reshape_2 (Reshape)             (None, 14, 14, 64)   0           dense_3[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_transpose_1 (Conv2DTrans (None, 14, 14, 64)   36928       reshape_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_4 (BatchNor (None, 14, 14, 64)   256         conv2d_transpose_1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "activation_4 (Activation)       (None, 14, 14, 64)   0           batch_normalization_4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_transpose_2 (Conv2DTrans (None, 28, 28, 32)   18464       activation_4[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_5 (BatchNor (None, 28, 28, 32)   128         conv2d_transpose_2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "activation_5 (Activation)       (None, 28, 28, 32)   0           batch_normalization_5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_transpose_3 (Conv2DTrans (None, 28, 28, 16)   4624        activation_5[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_6 (BatchNor (None, 28, 28, 16)   64          conv2d_transpose_3[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "activation_6 (Activation)       (None, 28, 28, 16)   0           batch_normalization_6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_transpose_4 (Conv2DTrans (None, 28, 28, 1)    145         activation_6[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "flatten_2 (Flatten)             (None, 784)          0           conv2d_transpose_4[0][0]         \n",
      "==================================================================================================\n",
      "Total params: 1,688,129\n",
      "Trainable params: 1,687,905\n",
      "Non-trainable params: 224\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "latent_inputs = Input(shape=(latent_dim,),name='latent_inputs')\n",
    "label_inputs = Input(shape=(n_label,),name='label_intputs')\n",
    "decoder_inputs = Concatenate()([latent_inputs,label_inputs])\n",
    "\n",
    "x = Dense(128,kernel_initializer=w_init,activation='relu')(decoder_inputs)\n",
    "x = Dense(before_flatten_shape[1]*before_flatten_shape[2]*before_flatten_shape[3],activation='relu',kernel_initializer=w_init)(x)\n",
    "x = Reshape((before_flatten_shape[1],before_flatten_shape[2],before_flatten_shape[3]))(x)\n",
    "\n",
    "x = Conv2DTranspose(64,3,strides=1,padding='same',kernel_initializer=w_init)(x)\n",
    "x = BatchNormalization(gamma_initializer=gamma_init)(x)\n",
    "x = Activation('relu')(x)\n",
    "\n",
    "x = Conv2DTranspose(32,3,strides=2,padding='same',kernel_initializer=w_init)(x)\n",
    "x = BatchNormalization(gamma_initializer=gamma_init)(x)\n",
    "x = Activation('relu')(x)\n",
    "\n",
    "x = Conv2DTranspose(16,3,strides=1,padding='same',kernel_initializer=w_init)(x)\n",
    "x = BatchNormalization(gamma_initializer=gamma_init)(x)\n",
    "x = Activation('relu')(x)\n",
    "\n",
    "x = Conv2DTranspose(img_channel,3,activation='tanh',padding='same',kernel_initializer=w_init)(x)\n",
    "outputs = Flatten()(x)\n",
    "\n",
    "# instantiate decoder model\n",
    "decoder = Model([latent_inputs,label_inputs],outputs,name='decoder')\n",
    "decoder.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## VAE"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "outputs = decoder([encoder(img_inputs)[2],label_inputs])\n",
    "vae = Model([img_inputs,label_inputs], outputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "beta = 1 # 1 --> regular VAE\n",
    "reconstruction_loss = mse(img_inputs,outputs)\n",
    "\n",
    "# reconstruction_loss = mse(K.flatten(inputs), K.flatten(outputs))\n",
    "# reconstruction_loss = binary_crossentropy(K.flatten(inputs),K.flatten(outputs))\n",
    "reconstruction_loss *= img_rows*img_cols\n",
    "\n",
    "kl_loss = -0.5*beta*K.sum(1+z_log_var-K.square(z_mean)-K.exp(z_log_var),axis=-1)\n",
    "vae_loss = K.mean(reconstruction_loss+kl_loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "image_input (InputLayer)        (None, 784)          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "encoder (Model)                 [(None, 10), (None,  1632084     image_input[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "label_intputs (InputLayer)      (None, 62)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "decoder (Model)                 (None, 784)          1688129     encoder[1][2]                    \n",
      "                                                                 label_intputs[0][0]              \n",
      "==================================================================================================\n",
      "Total params: 3,320,213\n",
      "Trainable params: 3,319,765\n",
      "Non-trainable params: 448\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "vae.add_loss(vae_loss) ## in this case, we set y value to NONE during the fit\n",
    "vae.compile(optimizer=Adam(lr=learning_rate))\n",
    "vae.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "os.makedirs('best_weight_ldg_v1',exist_ok=True)\n",
    "best_model_weight_path = os.path.join('best_weight_ldg_v1','ldg_v1'+'-best-wiehgts'+'-{epoch:03d}-{loss:.3f}-{val_loss:.3f}.h5')\n",
    "save_best_model = ModelCheckpoint(best_model_weight_path,monitor='val_loss',verbose=0,save_weights_only=True,save_best_only=True,mode='min')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train_or_load_weights(flag,weight_name=None):\n",
    "    if flag == 'train':\n",
    "        hist = vae.fit([X_train,y_train],\n",
    "                       epochs=epochs,\n",
    "                       batch_size=batch_size,\n",
    "                       validation_data=([X_test,y_test],None),\n",
    "                       verbose=1,\n",
    "                       callbacks=[save_best_model,learning_rate_reduction])\n",
    "        vae.save_weights('ldg_weight.h5')\n",
    "        return hist\n",
    "    if flag == 'load':\n",
    "        vae.load_weights(os.path.join('best_weight_ldg_v1',weight_name)) # load all the weights for encoder and decoder when loading for vae"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 697932 samples, validate on 116323 samples\n",
      "Epoch 1/10\n",
      "697932/697932 [==============================] - 197s 283us/step - loss: 76.7209 - val_loss: 68.9462\n",
      "Epoch 2/10\n",
      "697932/697932 [==============================] - 195s 279us/step - loss: 64.7492 - val_loss: 65.7158\n",
      "Epoch 3/10\n",
      "697932/697932 [==============================] - 195s 279us/step - loss: 62.2805 - val_loss: 61.5968\n",
      "Epoch 4/10\n",
      "697932/697932 [==============================] - 195s 279us/step - loss: 60.8709 - val_loss: 61.3175\n",
      "Epoch 5/10\n",
      "697932/697932 [==============================] - 195s 279us/step - loss: 59.8790 - val_loss: 60.2467\n",
      "Epoch 6/10\n",
      "697932/697932 [==============================] - 195s 279us/step - loss: 59.2089 - val_loss: 59.7529\n",
      "Epoch 7/10\n",
      "697932/697932 [==============================] - 195s 279us/step - loss: 58.6744 - val_loss: 58.9852\n",
      "Epoch 8/10\n",
      "697932/697932 [==============================] - 195s 279us/step - loss: 58.2446 - val_loss: 59.4655\n",
      "Epoch 9/10\n",
      "697932/697932 [==============================] - 195s 279us/step - loss: 57.8677 - val_loss: 59.0574\n",
      "Epoch 10/10\n",
      "697932/697932 [==============================] - 195s 279us/step - loss: 57.5495 - val_loss: 57.9839\n"
     ]
    }
   ],
   "source": [
    "hist = train_or_load_weights(flag='train')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "def training_plot(hist,model_name):\n",
    "    loss = [hist.history['loss'],hist.history['val_loss']]\n",
    "    plt.plot(loss[0],color='b',label='training')\n",
    "    plt.plot(loss[1],color='r',label='valid')\n",
    "    plt.title(model_name,fontsize=18)\n",
    "    plt.legend(loc='best',fontsize=13)\n",
    "    plt.xlabel('epoch',fontsize=15)\n",
    "    plt.ylabel('')\n",
    "    plt.tick_params(axis='both',labelsize=13)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "training_plot(hist,'ldg_v1 training')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_five_reconstruction(input_set,label_set,random_idx,encoder,decoder):\n",
    "    fig, axes = plt.subplots(2,5,figsize=(15,5))\n",
    "    for row,flag in zip(axes,['in_img','out_img']): \n",
    "        for col,idx in zip(row,random_idx):\n",
    "            original = input_set[idx]\n",
    "            if flag == 'in_img':\n",
    "                original_img = original.reshape(img_rows,img_cols)\n",
    "                original_img = np.transpose(original_img)\n",
    "                col.imshow(original_img,cmap='bone')\n",
    "                col.set_title('idx = {}\\nlabel = {}\\noriginal'.format(idx,lebel_dict[np.argmax(label_set[idx])]))\n",
    "            if flag == 'out_img':\n",
    "                encoded_z = encoder.predict(np.expand_dims(original,0),batch_size=1)[2]\n",
    "                predict = decoder.predict([encoded_z,np.expand_dims(label_set[idx],0)])\n",
    "                predict = (0.5*predict) + 0.5 # image back to 0-1 from -1 - 1\n",
    "                decoded = predict.reshape(img_rows,img_cols)\n",
    "                decoded = np.transpose(decoded)\n",
    "                col.imshow(decoded,cmap='bone')\n",
    "                col.set_title('reconstructed')\n",
    "            col.axis('off')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Traing set reconstruction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 10 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "np.random.seed(100)\n",
    "random_idx = np.random.randint(len(X_train),size=5)\n",
    "plot_five_reconstruction(X_train,y_train,random_idx,encoder,decoder)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test set reconstruction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAz4AAAFMCAYAAAAKvmk7AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjAsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+17YcXAAAgAElEQVR4nOzdd3gd5ZU/8O+RZMkqVnO35d4LuOEKLhAXmk0nhCSElpC+v2RTNtlUkiybullIgxRIgEDovRcDBmxsbLBx77Zc5SLLlmVJlt7fH3e83POesSVsyboz/n6eRw8+9565GqGj0YzunPeIcw5ERERERERxltbSO0BERERERNTceOFDRERERESxxwsfIiIiIiKKPV74EBERERFR7PHCh4iIiIiIYo8XPkREREREFHspceEjIktFZMpRnpsiIqUneZeIjok1S1HDmqUoYt1S1LBmU1tKXPg454Y452a39H4cISKvikiZiFSIyPsicpH3/FdEZH3w/AIROSvkNTJFZIVf4CIyXETeFZGDwX+HJz2XJSJ/EpEdIrJHRJ4Uka7N95XS8UrBmt0gIlUiciD4eMF7vreIPCUi+0Vkl4j8Ium5YhF5VEQqRWSjiFztbXvUemfNRkcq1ayIdE+q1SMfTkT+PXheROQ/RWRTUHf3i0i+9xpTRWRhULebReTKpOfOCZ6rEJF1IvK5pOfOFpElIlIuIruD2mfNpqiI1e13veeqRKReRNolvcax6tYFjx/Z/i9Jz/1IRGq91+99cv8PUGOkUs0eISL/FvwerxSR5SLSP3h8SlCjyXX1GW/bq4JtKkVkrYhMDB7PFJGHgvMPJ97FXqrWbEpc+KSgfwPQ2TmXD+BzAO4Rkc4AICJjAfw3gMsBFAD4K4BHRSTde41vAtiZ/ICIZAJ4HMA9AIoA/B3A48HjRz7veACnA+gCoBzAbU3+1VFczXTO5QUf0488GNTXiwBeAdAJQAkSNXjE7wHUAOgI4JMA/igiQ4JtG6p31ix9ZM65TUm1mgfgNAD1AB4OUq4B8GkAZyJRV9lIqisRGQzgnwD+E4m6HA7g3eC5VgAeBXB78NzHAfxGRIYFmy8DMMM5Vxi89moAf2y+r5bioqG6dc79l/f8zwHMds7tAo5dt0mGJb3Gjd5z/0p+fefcuub7aikuRORGADcAuABAHoALAexKStnq1dXfk7adhkQdXwegDYBJAJLrbg6ATwHYfpRPn3I1mxIXPsHV4tTg39kicpeI7BWRZQBGJ+X1Cf6qPDKIu0jir9dTmnJ/nHOLnXOHj4QAWgHoFsQ9ASx1zr3rnHMA/gGgHYAOSfvZC4lCuMV76SkAMgD81jlX7Zy7FYAAOCd4vheA551zO5xzhwDcD2BIU35t1DRSrWYbcC0SB7bfOOcqnXOHnHOLg/3JBXAZgO875w445+YAeAKJk06g4XpnzUZEitfsNQBed85tCOKZAP7qnNvsnDuAxC/ej4tITvD89wDc7px71jl32Dm32zm3NniuGEA+gLtdwnwAywEMBoCgVrcmfe46AH2b8WujExCxuk3eb0HiOPr3pIePVbcUE6lUsyKSBuCHAL7mnFsWHBPXOuf2NPIlfgzgZufcXOdcvXNui3NuCwA452qcc78Nzhvqmmqfm1tKXPh4fgigT/AxA8D/veUWHCC+DeDe4BfgnQDuOtpbipK4taf8KB9PHWsngm0PAZgHYDaABcFTzwJIF5GxwV+9rwfwHvTV7m0AvgugynvZIQAWByeQRyzGhyeKfwVwZlD8OUj89f3ZY+0npYSUqNngc5SJyAtJf90GgHEANojIs8FBdbaInBY81x9AnXNuVVL++/iwJhuqd9ZsNKVKzR5xDfQJogQfyXEWgH5BPC743EtEZJuI3CMixcH+7wBwH4DrRCRdRMYD6IHEXyaP7HN3ESlH4hj9DQC/AEVBqtdtsolIvIv+cNJjR63bJK+LyHYReUREenrPzQxOlJeKyBcauY/Uslq6ZkuCj6GSuLVyvYj8OLggOqKDJG5XXy8i/yOJP4gi+J1/BoD2IrJGREpF5Hcikv0Rvv7Uq1nnXIt/ANgAYGrw73UAzk167nMASr38JwAsQeKiIasZ96sVgPOQuFI+8pggcVFTC+AwEm8Xjk56/hIAzwX/npK87wC+D+B+73PcC+BHwb/zkfiF7YLXXgSguKW/P/wIrY2UqlkkbgnKBpAD4DtIXJgUBs+9ENTreQAykbgNc13w74kAtnuv9Vkkbs9oTL2zZiPykWo1m/R5JgI4ACAv6bEbAaxC4h3HgmBfHIDxwfM1wdfTH4lbNx4GcG/S9jMB7Ahq8jCAzx7lcxcjceIxrqW/P/w4an1Epm695/+KxEls8mMN1e2k4LhcCOB3AD4AkBE8NxiJWzPTAUwAsA3AJ1r6+8OP0O99ytRsUCsOwNNBXfUMjq2fDZ7vFNRWGhJ3cLyOxLuSCOrNIfGH/85I3O3xJoCfhXyeUgBTvMdSsmZT8R2fLgA2J8UbQ3L+DGAogNucc9XNtSPOuVrn3LMAZojIrODhG5H4q/cQJA5QnwLwVPAX71wk/nL4laO85AEkThST5QPYH/z7jwBaA2gLIBfAI+Bfz6OgxWvWOfemc67KOXfQOXcLEr02E4OnqwDMcYnbK2oA/AqJGhuEhmvyqPUePM+ajaYWr9kknwHwsEvc0nbE35C4oJ4NYCmAV4PHjywWUwXgTufcqmC7/wJwPgCIyEAA/0Lir/GZSNTut0TkAv8Tu8TtHkd6LTOa+OuippfqdQsgcXsTgCtg3w06at0CgHPudZe4fagcif7JXkgcp+EStyltdc7VOefeAvC/SPReUmpr6Zo9cufRL5xz5S5xW+btCOrOObc9qK1659x6AN/Ch3V1ZNvbnHPbXKJX7TdIqtljSdWaTcULn234sJ8GALonPykieQB+i8RfU34U8jZxcu6zYldgOfLxUU7OMpB4mxIAhgF4Mjhw1Tvnngv2eQISt2H0BPCGiGxH4iSwc/C2dU8kfoGfLiLJt3CcHjx+5LXvcs7tCYr/NgBjJGlFGEpJqVizDh/eKrQ4iMOsApAhIv2SHhsGXZNHq/cjz7NmoyclavZoJ4hBrf3QOdfTOVeCRD1uCT6AY9f0UAArnXPPB6+zEom/dp53lPwMJHrW/D8AUOpJ6bpNcimAPUhcuCc7Vt2GST6Of5TnKHW0dM2uROKdxsbW3f/VlXNuLxJ/bPooNduo125RLf2Wk7NvC/4cwGtIrHpWgsSBIvl2sb8CeCD49x1H/t2E+zIQiV+Q2Ujc6vYpJIpmZPD8Z5A4WeyNxDdwGoCDwXYZSLxteOTjUgBbg3+nI/HXx41I/CUnC8CXgzgzeO07kXjruyD43N8FsKWlvz/8SPma7Y7ErW6ZSLz78k0AZQDaBs8PCGp0alCHXwOwNqnu7kfir+u5wevsAzCkoXpnzUbrI5VqNunzXB0cA8V7vBiJPzYJErdLfADgc0nPXw9gfVCXOQAeQGIxAwTbHUBi0RgJ4jX48NaOS4OfiTQA7YNtF7b094cf0a/bpOdfQKIh3H/8WHU7BIlV3tKRuA3ut0ictLYKnr8o+LoFwBgk/gjwmZb+/vAj9WsWiUWJnkJiVbYSACsA3BA8NwWJcwhB4gLtVSTelTyy7c0A5iPxx6EiAG8A+EnS81lInHeUApge/FuC51KyZlu8QEKKJCf4JpUjsezoN48USfA/cQuCHoLg4LAGwCebcF8GIbGgwf5gH+YDuCTpeQkKYVOQsxzAp4/yWlNg7+UcgcTylVUAFgIYkfRcWyR6fnYGn3sOgDEt/f3hR8rX7JDgYFoJYDeAlwGc4eVcGnzeCiT+Cjkk6bliAI8F228CcHXSc8esd9ZsdD5SqWaT9ul5JP0STXq8PxInfQeROMH8ekjOj5G4wC8DcDeAoqTnrkTiYmk/Er+Qfw4gLXjuK0icfFYi0Qt3P4AeLf394Uf06zZ4risSfWV9j/J8aN0icaG+MqjLncExuV/SdvcFx/cDSJy4frWlvzf8OGp9pFTNIvFu9v3B8XAzgB/gw4uTrwf7cDB47jYAbZK2bQXgD8H+bwdwK4DW3tfqvI+ewXMpWbNHvnAiIiIiIqLYSsUeHyIiIiIioibFCx8iIiIiIoo9XvgQEREREVHs8cKHiIiIiIhiL1IXPiKyQUSmNjLXiUjf4/w8x70tUTLWLEUNa5aihjVLUcOabTmRuvCJMhEpFpFHRaRSRDaKyNUtvU9ExyIiXxaRBSJSLSJ3tfT+EDWGiFwlIsuDY+1aEZnY0vtEdDTBCXBVMIRyu4jcFQy1JKJmwAufk+f3SAxC7QjgkwD+KCJDWnaXiI5pK4CfAvhbS+8IUWOIyDQkZvZch8SwvkkA1rXoThE1bKZzLg+JAaYjAHynhfeHKLYie+EjImNE5G0RKReRbSLyOxHJ9NLOF5F1IrJLRH4pImlJ218f/FVwr4g8LyI9mnFfcwFcBuD7zrkDzrk5AJ4A8Onm+pyUeqJUswDgnHvEOfcYEgPI6BQUtZpFYjjkzc65uc65eufcFufclmb+nJRCIliz/8c5tx2J4ajDT9bnpJYXtZoN3qX8hogsFpF9IvIvEWndnJ+zKUX2wgdAHYCvAWgHYDyAjwH4opdzCYAzAIxEYkLu9QAgIhcD+C4S0+zbA3gDiQmzDRKRPwTFGfax+Cib9QdQ55xblfTY+wD4js+pJUo1SwREqGZFJD3Yj/YiskZESoMTiOyP+DVTtEWmZkNeowTAeQDWNCafYiOKNXslgHMB9AJwOoBrG/M5U4JzLjIfADYAmHqU5/4fgEeTYgfg3KT4iwBeDv79LIAbkp5LA3AQQI+kbfs24X5PBLDde+yzAGa39P9TfjTvR1Rr1tvPnwK4q6X/X/Lj5HxEtWYBdAlecwGAzkicRLwJ4Gct/f+UH837EdWaTdr3AwD2B6//MoDClv5/yo/m/YhBzX4qKf4FgD+19P/Txn5E9h0fEekvIk9JohmwAsB/IfGLLtnmpH9vROIXIwD0APC/R65sAewBIAC6NtPuHgCQ7z2Wj8SBjk4REatZoqjVbFXw39ucc9ucc7sA/AbA+c30+SgFRaxmj7jYOdcGwBQAA2H3l2IsojW7PenfBwFEZkGOyF74APgjgBUA+jnn8pF4q0+8nG5J/+6ORLM2kCigm5xzhUkf2c65txr6pCLyJ0msvhL2sfQom60CkCEi/ZIeGwbgaPkUT1GqWSIgQjXrnNsLoBSJv3DSqSsyNetzzr0G4C4Av2pMPsVGZGs2iqJ84dMGQAWAAyIyEMAXQnK+KSJFItINwL8B+Ffw+J8AfEeCVdVEpEBErmjMJ3XOfd45l3eUj9CeHedcJYBHANwsIrkiciYS92je/VG+YIq8yNRs8DkyJNGwmA4gXURai0hG479cioFI1SyAOwF8RUQ6iEgREreMPNW4L5ViImo16/stgGkiwgUOTh1Rr9lIifKFzzcAXI3E7WJ/xodFkOxxAO8CeA/A0wD+CgDOuUeRWPL0/uBtxQ+QaChsTl8EkA1gJxKNZ19wzsX2ippCRa1mv4fE7UP/AeBTwb+/18yfk1JL1Gr2JwDmI/Eu+3IAiwD8rJk/J6WWqNWs4pwrA/APAN8/mZ+XWlSkazZqJGhMIiIiIiIiiq0ov+NDRERERETUKLzwISIiIiKi2OOFDxERERERxR4vfIiIiIiIKPZOqQufYM3yRq2U8lFyG3idniLiuAwwHQ/WLEUNa5aiiHVLUcOaPT5c1a2ZiUhPAOsBtHLOHW7ZvSFqGGuWooY1S1HEuqWoiUPNnjLv+IhIekvvA9FHwZqlqGHNUhSxbilqWLPHL/IXPiIySERmi0i5iCwVkVnB43eJyB9F5BkRqQRwdvDYT5O2/ZaIbBORrSJyY/D2Xd+k7X8a/HuKiJSKyL+LyM5gm+uSXucCEVkkIhUisllEfnRy/y9QlLBmKWpYsxRFrFuKGtZs84v0hY+ItALwJIAXAHQA8BUA94rIgCDlaiSmdrcBMMfb9lwAXwcwFUBfAJMb+HSdABQA6ArgBgC/F5Gi4LlKANcAKARwAYAviMjFJ/TFUSyxZilqWLMURaxbihrW7MkR6QsfAOMA5AH4b+dcjXPuFQBPAfhE8Pzjzrk3nXP1zrlD3rZXArjTObfUOXcQwI8b+Fy1AG52ztU6554BcADAAABwzs12zi0JPs9iAPeh4aKjUxNrlqKGNUtRxLqlqGHNngRRv/DpAmCzc64+6bGNSFzBAsDmhrZNio+VCwC7vUaug0gUKERkrIi8KiJlIrIPwOcBtGvMF0CnHNYsRQ1rlqKIdUtRw5o9CaJ+4bMVQDcRSf46ugPYEvz7WEvWbQNQkhR3O4H9+CeAJwB0c84VAPgTADmB16P4Ys1S1LBmKYpYtxQ1rNmTIOoXPvOQuBfxWyLSSkSmAJgJ4P5GbPsAgOuCRrIcAD84gf1oA2CPc+6QiIxB4j5MojCsWYoa1ixFEeuWooY1exJE+sLHOVcDYBaA8wDsAvAHANc451Y0YttnAdwK4FUAawC8HTxVfRy78kUAN4vIfiSK7YHjeA06BbBmKWpYsxRFrFuKGtbsycEBpgERGQTgAwBZUR3KRKcW1ixFDWuWooh1S1HDmj26SL/jc6JE5BIRyQyW8Ps5gCdZIJTKWLMUNaxZiiLWLUUNa7ZxTukLHwA3ASgDsBZAHYAvtOzuEDWINUtRw5qlKGLdUtSwZhuBt7oREREREVHsnerv+BARERER0SmAFz5ERERERBR7Gcd6UkR4H9wpxDkX+QFVrNlTSxxqFmDdnmriULes2VMLa5ai5mg1y3d8iIiIiIgo9njhQ0REREREsccLHyIiIiIiir1j9vgQERERUUvQLQoiDbfZOFffXDtDFAt8x4eIiIiIiGKPFz5ERERERBR7vPAhIiIiIqLY44UPERERERHFHhc3aJSwhsKG52CJNHxd6Zz/OpyvdaorKGiv4qKiTg1us3fvdhXv21fWpPtE1JCMjEwV5+YWmBz/mFhevtPL4PGPTg3+z0JxsT3ODx58ppfTWcXbt68z28yf/6yK6+vrjncXiZpc486Lm3eBDr7jQ0REREREsccLHyIiIiIiij1e+BARERERUeyxxwdAerr+39C5U28V9+g51GyzbdvaBl+3b99RKs7JyTc5u3dvUXFp6UoV23vggb17d3iP8L74qPJrDwAuuvwLKr7p258yOf4gu3tuf1TFf731h2ab6pqq49lFohC277FfP328m3LuRSanS9+uKv7N976pYr9Xjajl+bX+0X/fpqWlm8fGjZul4qv//QaTM3X8SBXnZmWp+Ok355tttn9lvYo3blwWskc8Z6Dm0bp1norbt++mYv/3BGB73VevsnVdumVVE+xdAt/xISIiIiKi2OOFDxERERERxR4vfIiIiIiIKPZ44UNERERERLF3CixuoBsTS0oGmIxzL9LN42POH63iSSNOM9vsr2q4Ubx9vl7MIDPD/u+urK5W8Z4DB1T82hsLzTa3fv8HKi4tXdHgvlB0FHUsVPGonj1NTiuvlvZfc66Kn7jvL2abpmwOpFNbVmZr89jgweNVPOzs4SanS4kezpvrLfjCxQ3o5NLnB1279jUZBQUdVLx8+dsqDhu2mOn9fPTtM9LkXPMfN6n46mmT7euk60URMrz4E1MnmW0qfqQXtrnl6181Ofw5O7X4i2v4iwkkHmvM0FD985KZmWUyevXS58v+74UxF4w129RW16g47SH7nsyWrWtUfCJDTvmODxERERERxR4vfIiIiIiIKPZ44UNERERERLEXqx4f/75aAOjTZ4SKP+sNzAOAay+aruK81vp10tNO3vVh7w76fuL+nTubnC2rS1V8+29+YHIOHTpgHqP4Ks7NVXFGK3vvLdHx8u8RnzT54ybnO7/8sor7depkcpZv1QObWafUkoqKOqr4Sz/8scnZvXW3itf/fLGK6+pqzTbnnvtZFd/4o2tNTrro84r/vvXuY+4rAEyfeZaKR/fubXImTdTnPH8oaG9y2OMTJ3aYtN97U1Skj8XV1bZHvbx8p/eI7QPyezvbd+hhcqZfcoWKB48frOIzhto+++3l5Spet3i9yUmbrX9e6urY40NERERERHRUvPAhIiIiIqLY44UPERERERHFHi98iIiIiIgo9iK0uIFt4GrbVjf+T5lylcn54s3Xq3hsnz4mJ9trBEsT/bnqQ4Y9+Q7X1ZnHduzbp+KakJzcLP2527dpo+L87GyzzbWfu1TFH8xbZHJeeeVeFZ/IsCdKPX6NEjWVrEx7zOnUWTdRX/5Ve6w9rVs3FdfWHTY5b8zRx6p9+8qOZxeJmkSXLnpg6Yxp402Of6xd8qYeKl5autJsc8MPr1Xx6mUbTM5ffvYrFa9b9/6xdhUA8MKDesjpr/7xa5NT6C10k5dX1ODrUnT558EAMGLEVBV/+ns3qHjT0k1mm//5/rdUXFVlF8iaPv06FU++corJue7iGSr2z3HDFgtbWK/PT9sU5ZmcpsR3fIiIiIiIKPZ44UNERERERLHHCx8iIiIiIoq9yPT4lJTYoUf/9tObVXz5+VNMTre2bVUsIb0Rx9PTU1ldreL7XnrN5Dz+x0f1NgfKTU6fwXq40/dv/oKKu7drZ7YZ4A01HXzGcJMze/Z9Kj6RYU90ctXX6foLq8fG1CjR8QgbSjd8+MdUPPb0QQ2+TlnFfvPY3Cfnqriy0h4TiZpDq5BhuRPOOU/FfTt2NDmtW7VS8RVfv1rFS15fYrZZPHeZiv/881tMzqZNy71HGj6mr1ihf37emvOeybnpEzNVPG7ydJOzfPlbKq4L6cejVKXPVzuEHK+HjNHnhOMH9FdxeroeSA0ABd6g28LCDibH7+mZNN6ee+ZkZurPFdLT09JSb4+IiIiIiIiaGC98iIiIiIgo9njhQ0REREREsZeyPT6tW+t1vC+75nMm58YrLlBx2MybxvD7JfyZPFv37jXbPPTMbBX/7oc/NjmbN69QsQvpy1i6rIuKR804Q8U3Xmjvz/V7ktLSOdMlqurrbe/Vqvf0/eE7KypMjt+7RnS80tP1r4HpM682OZffNEvFQ0tKTM7m3btV/Myb803O7Nn3q7im5lCj95Poo/DnUU2a/HGTM+XjU1Sc7fUnALZH4azTdV/uY7972Gxzz+2/VPGePdtD9vCj92n6vc7jzhxmcvy5Kf3P6G9zcgpUXLF/t8mhlpeWZntxiop0H9r13/6ayZk6ZYyKaw7rHq7tG2w9XvqZz6q438h+JuesEUNUvPuAnfXzxko912qI97uibZ6d0bOnslLFB8orTU5T4js+REREREQUe7zwISIiIiKi2OOFDxERERERxR4vfIiIiIiIKPZSZnGDoqJOKj7nHN1ge91Nl5lt/MUM/KZ/oHGDHiuqqlT8wCtvqPjR39vmxfnzn1bx7t3bQl654c+9Z4/e7tX7XlXxpWefabYpyslRcU5+rsnJSNdD1zigLDU5Zxc3WL16gYrLuLgBNaOMDN3Q3alnJ5PTtbhYxXUhi3L4ixk8eftjJmfv3h3Hs4tExxTWBO4vZnDL7T8wOad166bisGGL/rDyp158U8Vz5z5htvF/rx8vf4GGGZdeqeJRvXqZbaq9RvbNKzbbnOqDTbB31NxaZdjFNtq3767ikaMHm5ye7fUw0u3lelB0XoFdYGDSzAnHfA0AePO9pSpe/8EGk9NjsB6o2qNdOxUXhCxCtnZdqYpL19rXDVsI6njxHR8iIiIiIoo9XvgQEREREVHs8cKHiIiIiIhir0V6fPyBeQBw0eWfV/H3fvoFFff07hMMU+sNHgXsMNKV2+y9t3fd8YiKH7n7zyresmWV2aa+3n6u4yFeX1Jh+0IV+8PIwrYp7lRkcrKydB9QdU2VyaHUVFOtv1f+cK8wma10T1denq0JogR9/OjdWw9BnHqh7Sv0j7/LtmwxOY/94UEVL1z4gslpquMmUbLCwg7msYs+f7mK/X4eAMhI171B/vkCANz30msq9oeVh/f3fnRhfUp9+oxQ8UWfnKHisPODJZt1T88rTz1qcng+kKq8Y3Of4SbD7/Ma3buPyfHrIts7P2g7ebzZZs3OnSp++Y0FJueeX/9Jxf5AXQC46urzVNy5UA/L9XvQAGDhiwtVvGzZWyYnrB/6ePEdHyIiIiIiij1e+BARERERUezxwoeIiIiIiGKPFz5ERERERBR7J2lxA92wFdYQNeESPTypuzeg0W/oB+zg0Qdfe9PkLJ3zgYo/mLfI5Lz5pm7+O3TogMk5WXIL9KIEmem24dF5Q1m3rLaNxger9jftjtFJU7ZLD/N67dm5JmfKoEEq9pvPx06aZrZZvlw3DHKo7akpM1M3vg4erBtde4UMrmvlHYfmL7MLvqxYMU/F+/fvOd5dJGqAPh/o2rW/yThz1FAV+wsZhAlbtOOhW+9X8ebNK7yMhgeVh/EXeRo7dqbJ+cx3blLxmD69VewPVwWAF199R8Vbt645rv2j5mbPabMyW6vYPzYDwKBx+nd/a2/hgjB+7bfKsKf+c99ZouLXH3zd5NTX63OG3qf1Njn+wNIMb9GOsgp7brp5zXoVV1aWm5ymxHd8iIiIiIgo9njhQ0REREREsccLHyIiIiIiir2T0uPTtm1nFX/5xz80OVedPVHFaV5Pjz+UCwDuvP1hFfuDRwFg27a1Kg7ra2jKwUgflYi+9pS0hq9F/QFQ+/fanqSW/JroxBw+XKPiyn1h3199X3mWd89uXmFu0+8YRU7YsOgO7bur+IbvX6Pibl5/JQDUeMccv3cSsMdav46bU6Z3b3xubqHJyc/XX1dFxW4Vh91XXlNzqAn2jpqaf05xzb9/yeT069SxwdfxB5Y++4IdnDhv7pMqPp4hvP7veQAYM+ZCFX/+lq+ZnCvHjzvm69774ivmsd//SA9Y3bt3R2N2kU4yv9cSANp36KHisy6baHLGnq57fNJDzhn98+fKGt0LtnHXLrPNaw/MVvH7779qcgYO1PXYd2Q/k+MPS62qrVXxujI9KBUANm5cqvf3AHt8iIiIiIiITggvfIiIiIiIKPZ44UNERERERLHHCx8iIiIiIoq9Jl/cIKyZdvToCztQPzwAACAASURBVFR8+flTTE5OZqaKN+3WzVc/+fr/mm2efvoOFbfk4NHGyMjINI/16zdKxYMnDFZx2ODWd9frYU9vvvK0yeFwyvg4UF5pHvMXuPAXNwhbJCOswZbiLex7npmVreL2bdqouN7ZgYw7KypUvOLdpSbnZB1z0tLsIMpJk65U8TlXnmtyJk85Q8WzX52v4lceeM5s88or96qYi8a0jFatdCO4vzBA2DlFdkjzuM8f7DjuzGEmp0tX3cBdsWK3yfH5Ndq2bVeTM/Vy/TVcOna0yfEbwx945Q0V//2W2802TTVglZpX797DzWMzLtXHsWsvmm5y2rTWC7lUezUCAHsr9TnDayt0Tcx5ZI7Z5vnn/6biujr7uhddfZ2KRw8baHL83xXPvbNQf56/2uPsunXvqbi5F5XhmRAREREREcUeL3yIiIiIiCj2eOFDRERERESxd8I9Pn5PT2FhB5MzcJTuWynOyzM5h2r1sLuFGzaqeNmyt+02h2zvQyrLyysyj519wSUqHjvU3jPpW7Fa/78pK9t0YjtGKaW+XvcRlK7dYHL8e3g7F+qBjTltcsw2/s/qyRwwSS0jrOcyJ6dAxZnewDm//gBgX1WViktLV5qcpuh/CetJ8vscs7JsbQ8ere+XHzPhdJMzvLse3Fo+Rv8MrX1PD2AFgLTZen/q6tjj09SyMnXP2YxzbzQ5X/iJfmxoSYmKuxTZ363H48z+/c1jX/vlj1T8i//3HRX37m37gi7+4hUqPu/MM0xOm9b6677j4WdNzp9/+ksVr127SMUcsBsd/rG4Vy97jOo/eoCK/d53wB4Pd+zbZ3LeWrVaxS/940UVL1+s6wgAamt1LWV6P5cAkFek+0Fbh+zf7gO61379Yt2TvmrVu2abGm/AanPjOz5ERERERBR7vPAhIiIiIqLY44UPERERERHF3gn3+LRunavis8/+hMn55PWz9DbePeUA8PcXX1Xx//z7D1W8erW9LzDV16fPzNTrrQ8cMMbk3PAlfS/wkK56vf/Nu+3MgHeefkfF5eU7j3cXKQX5vRKrVy8wObv271dxt7ZtVTz5vHFmm3/cqu+LL92y6nh3kSKifbsS89jYSdNU3LNdOxX7M6IAYHt5uYr3NdExx5934s9nAYB+g/W98GkhM6ou/8z5Kh7eo4fJ8e9H75Cfr+I2Rbb3lJpW6AymyR9X8Q9++zWT438/w+bb+Q7V6B7GymrbR9DWm2Hlz/UBgCvPmajiiu/rHp9Ro4eYbYZ5/WTvb7J9uI/d87yKH7n7DpNje+lS+5znVBVW1wUF7VU8aNB4FV/9nU+bbSYPHqTi2pDZaCu2bVPxHbf9y+QsevMtFc+f/4yKDx+2M3r8/vyRI6eZnMuv0I919847AODe5/S5/MJX9fnqltDzjpNb13zHh4iIiIiIYo8XPkREREREFHu88CEiIiIiotjjhQ8REREREcXeCS9uUFlZoeKVK20jtj8gzznbyLTsrWUq9hczSPVhi2FD9WbMuEHF/3bLTSZnUJcuKt5/SA+Ruu3395ttnnxEN0HWhTTAUXyEDZQM+xlKVpybax7LaJXVZPtE0RD2Pc8r1LWRlaF/Dazcvt1s89qzc1Vctqu0CfYO6NpVD4y89rtfMDkfG6OHk+Zm2a+pMEcff7NCFtDxf2bKvAVC9u/Vg/eo6RUVdTSP+cM+T+vWzeQ0tJhBVcgAxL898YKKd24qMzmXXPYxFfu/jwEgP1sPcvz4heeo+Kk5unkbAP75h0dVPOflp0zO+nXvq7i6psrkUDSEDfvs1KmXikecOUHFZw3Uw0oBoJ232EbYcNJ5S5ar+JWnHjE527froaG1tQ0PCM3L00OAe4Xsn7+YQdh5yKoFevGCjRuXqrjygF4opyXwHR8iIiIiIoo9XvgQEREREVHs8cKHiIiIiIhi74R7fPx7b7t06WtyCrLt/Y8+5/Ux+EMcU03r1nrY3YwZ15scfxCbP9QMAA7X1an4gVfeUPGDf/uj2Wb37q2N3k8ioo+iptYOtzu4/6CKj7ev0B/0N3DgWBWPPX2g2aZjQYGKq0P2L60xAy1rdZ/o0ve8e9FXrTbbhPXXUeOJ6L+tnnHGeSbn/DNHqzhsiKjP7y2Yt3atyfnTD3+u4m3b15mclx58XMU/+9stJmfyQD1Ucu9B/bPwpx/8wmyzdOkcFbMPN1784fRh539TPn62is8+c6SK/QHKgB1Mft9jL5mcVx54TsUrV9oes/r6OvNYMn//AWDMGP2zOeaCMSanokr3ob244D2T8+g9f1bxjh2636imRvextwS+40NERERERLHHCx8iIiIiIoo9XvgQEREREVHs8cKHiIiIiIhi74QXN0hL09dO/YcPNjlhTVypTTfKlpTYQU6XXfM5FV9302UmZ2hJiYrrQ4Y9PTh3nor/fsvtKt6yRTfgEhGdbE21+ExhYQcVX/T5y1U8uGtXs0269ztm4y47iDI9rVjF+SEN8mt36u3+8evfq3jNmoVmm1RfZCfVpafrU4yBo4aYHH/xisb4oFQP0L31u3eYnDVr9fczrKl64aIXVfyHH/QzOf1u/65+HW9xjQMH9pptuJhBfPgLsgBAmzb6eHPWJWeZnEnj9eDlXu3bq3j3ATsw+fXlK1T86oMvmJwVy/Uw6YYWMgDsIiO5uYUmZ9zMcSoePbi/ydnnLW6wcelGk7PLG25dEzJcuKXxHR8iIiIiIoo9XvgQEREREVHs8cKHiIiIiIhi74R7fHxp6XaQXGOGy4l3H7d/T+Lx8u8xDpORkani3r2Hqfiz3/um2ea6i2eoOL8RQ1o37dplHvvnLXereN68J1XcmPs3iYiaSmarVuaxgvb6nvDi4i4mZ/fuLSoOG/5ZUKDvcx89TA8sbczwykFdbB+Qzx9WCgBLvb6QvXu2qbimRt+/TicuP7+tigeMsQNqG/M99wcn3nn7wyp+/vm/mW0aMyjRz1m40A6M3Fb+RRV3LtQ/CyPPmGa2qazcp+LWrXNNTkXFbhXv3bsjZA9tXzCdXK2880MAaN9eD6MfOdr2tvfuoPsZszP167y7Xg/2BICFL+m+tEWLbD1WVNjzyIb4vfh5eUUmZ+gI3cveo107k7N2h67Ryn2VJqe21u/pSb0a5js+REREREQUe7zwISIiIiKi2OOFDxERERERxR4vfIiIiIiIKPaafHGD/XvtUKaDNbrRNK91a5MzeIJuDuvRQw86KyvbbLbxFy7o0qWvyRk3WS9CkFdomwxzC/JUPPXCM1U8pk9vs012ZpZ5zOcPWbvjtn+ZnLlzn1AxB5+R77BpFgT2VNqmwmQSsqCI3+BIp6b6Ot1s6g9W7texo9nmxk/PUnHHnjbnqTseV3Fp6UqTM+S08Sr2G8XD+IvjhI0U9Yeavjhvkcn57bd+rOLtOzaomMfeppefrxukRw61A0L9AbVVIQMP73zseRU//A89sPTQIXvecTz27bPDcd/9QA8Rv+ECvZjB93/1VbPN++v1z0ufzp1MzmuzF6j419/5hsnZvXvr0XeWTooePYeax6ZddIWKz+hlzxH938HLtujFX/73u3pYPQC8+64eWOoPAz1e/qIygwdPMDmDuugFa3Kz7Dmuf96xf89+kxOFoc88EyIiIiIiotjjhQ8REREREcUeL3yIiIiIiCj2TrjHx78v+sWn7jM5D54/RsXXnvsxk3PV2RNVnHPXbSpevXC12aZiV4WKP3ndTJPTr5O+tzYro+EvuZWX45wdwLSzQn/u8pCeC7+n587f/8TkHDxYYR4jSlYWcp/va8/OVfGUQYNU3CE/32wzYoT+uQvrwWjM0D+KjrD+sIP7D6q4tk4PSQ7rwezk9eKcf+Zok+P3kG1ZvcXk9BjSQ8XFebbn0uf3IIX1gCzcsFHFi19fYnI2blyq4sOH7ZBTalp+/e2tPGhyag/rc4i5a9aanD//9JcqDjt2NQV/qCgArHxnhYrrztPH0aElJWabIV31kN2wnssu3s/U7IfONTkvvniXiqPQPxF9+nvVvn03k1EyQH/PM0L6Z/1zxHmLl6v4vfdeNtuU7dxoHmsKOdltVFzSp6fJKcpt+Fi8evUmFZeu3WBywgZXpxq+40NERERERLHHCx8iIiIiIoo9XvgQEREREVHsNfkcny1bVpnHHrr1fhWPGtLf5Azvoe/9vnL8OBWXDdFzfgBg9fbtKj69e3eT488IaAz/nuPSvXtNzn/98E96X5Z+YHI++OANFbOfh45HWC9C5T49t8LvQ+tSVGS2ufRrl6l4zpxHTM4Ob7YJRVvY/LM5Lz+p4heWTFfx9NNOM9v4vZHd2rY1Of58k7DeSL/XoTEzXMoq9KyIh56ZbXL+9vP/UfH27etNDo+/J59ff0/c9azJWT1+g86542GTs3atP5fJ1lZTCJvltOJd3Rvm926E/Sz4de33qQFAx4ICFQ8YMcTkvPyyfp26utTvn4g+/b3auXOTydiwRB9fXlq61OQsfVf3ob3ywHMq3rxZ944BQH19nXmsORyusXW+afcuFVdU2X7fx//0kIoXLnzB5EShD43v+BARERERUezxwoeIiIiIiGKPFz5ERERERBR7vPAhIiIiIqLYa/LFDcKas15/TQ/yvPn/tTE5n7/5BhV3LS5WcfeQBsLRvXureHt5ud2fkKZC38pt21TsD4cMG8T35CN3qHjPnu0mp7kaMIn279WLGxys0Qsg5GRmmm0+mKMX4Cgv39n0O0Yppbqmyjy2du17Kv7eNV9R8fNTZ5lt2ndrr+Ip5483OQM6d1ZxYwbirdupa/Cl2e+YnLlP6uPx7Nn3m5zdu7d5j/DYmwr8+vvn335tctLuSlfxvn1lJudkNX2HNWYvW/aWiv3zhRLvXAUA/FepCxnqWFmtF/KorLBD0Knl7QoZIL7o7bdVfKjSLgSwaa0exLtyxTwVn6yaBoD9B/QCXSs/eM/kPPXcmyquOmB/dyxfro/F+/fvaYK9O/n4jg8REREREcUeL3yIiIiIiCj2eOFDRERERESxJ2FD5v7vSZFmuVE6M7O1eaxDez18tKCwg4rHTZ5htslpk6PiNYtXmhz/PkoXcq/t8hX6vsWdOzeqOGyAZNigs6hzzknDWamtuWo21ZSUDFTxV39ys4oHDetrtvntN/WQx1deudfkRGH4WLI41CyQWnWbnm5bPzMydM9Yhw49TM7gwRNUXNKnZ4Ofa+5rz6t469Y1JqeyUvdu1tTY++mjJg51m0o125yyMrNVfP1XfqTiG750hdkmr7U+x3nlHdtTsfDFhSp+7rF/mJzSkIHwLYU1m/w6aV5s/9f459ap9LvV338AaJOnh567kD7J/fv3eo+k9iHgaDXLd3yIiIiIiCj2eOFDRERERESxxwsfIiIiIiKKPV74EBERERFR7LXI4gbHI6zh1lcfsnBBY6RS01lLYvNilOhvVVFRRxXn5uSbLXaWbVIxm8RTR9TqNux4XFCgh5zmZNtB1b4ybzhgWE3G8fgch7qNWs02lcJCfazt0rmPySkq7qTiFd4CSoAdIJ3qCyaxZuOuMd/eaP3v4+IGRERERER0yuKFDxERERERxR4vfIiIiIiIKPYi0+NDzY/38FLUxKFmAdbtqSYOdcuaPRb/2xv9/1WsWYoa9vgQEREREdEpixc+REREREQUe7zwISIiIiKi2OOFDxERERERxV7DU0GJiIiIqJHYQ0+UqviODxERERERxR4vfIiIiIiIKPZ44UNERERERLF3zAGmREREREREccB3fIiIiIiIKPZ44UNERERERLHHCx8iIiIiIoo9XvgQEREREVHs8cKHiIiIiIhijxc+REREREQUe7zwISIiIiKi2OOFT4SIyI9E5J6W3g+ixmLNUhSxbilqWLMUNS1Vs7zwSSIis0XkxmZ8/Q0iMrW5Xp9OPaxZiiLWLUUNa5aihjUbrkUvfEQkoyU//0cVtf2lphe1Goja/lLziFodRG1/qelFrQaitr/U9KJWA1Hb36Zy0i98givEb4vIYgCVItJdRB4WkTIRWS8iX03KTReR74rIWhHZLyLviki34LkJIjJfRPYF/52QtN1sEfmJiLwZbPeCiLQLnmstIveIyG4RKQ+27SgiPwMwEcDvROSAiPwuyHci8iURWQ1gtYj0DB7L8D7fjUnxZ0VkefC5l4nISBG5G0B3AE8Gr/+tIHeciLwV7Mv7IjIl6XV6ichrweu8CKBdc3xP6NhYs6zZKGLdsm6jhjXLmo0a1mwEa9Y5d1I/AGwA8B6AbgByAbwL4AcAMgH0BrAOwIwg95sAlgAYAEAADAPQFkAxgL0APg0gA8AngrhtsN1sAGsB9AeQHcT/HTx3E4AnAeQASAcwCkB+0nY3evvrALwYfM5sAD2DxzKScv5vOwBXANgCYHSwz30B9Ej62qcmbdcVwG4A5yNxETotiNsHz78N4DcAsgBMArAfwD0n+3t2qn+wZlmzUfxg3bJuo/bBmmXNRu2DNRu9mm2pW91udc5tBjA0+B9ys3Ouxjm3DsCfAVwV5N0I4HvOuZUu4X3n3G4AFwBY7Zy72zl32Dl3H4AVAGYmfY47nXOrnHNVAB4AMDx4vBaJQuvrnKtzzr3rnKtoYH9vcc7tCV6rITcC+IVzbn6wz2uccxuPkvspAM84555xztU7514EsADA+SLSHYlC+75zrto59zoSxU0tgzWbwJqNFtZtAus2OlizCazZ6GDNJkSiZlvq/r7NwX97AOgiIuVJz6UDeCP4dzckrnJ9XQD4/+M3InG1ecT2pH8fBJAX/Pvu4HXvF5FCAPcA+E/nXG0j9rcxjrbPYXoAuEJEkou7FYBXkfga9zrnKpOe2xi8Pp18rNkE1my0sG4TWLfRwZpNYM1GB2s2IRI121Lv+Ljgv5sBrHfOFSZ9tHHOnZ/0fJ+Q7bci8T84WXck3o479id2rtY592Pn3GAAEwBcCOAab7+Otr8AcOSblpP0WKekfx9tn8NefzOAu72vP9c5998AtgEoEpHcpPzuR3ldan6s2Q9zWbPRwbr9MJd1Gw2s2Q9zWbPRwJr9MDfla7all7N+B0BF0BiWHTR+DRWR0cHzfwHwExHpJwmni0hbAM8A6C8iV4tIhoh8HMBgAE819AlF5GwROU1E0gFUIPE2YV3w9A4k7sk8KudcGRLF+Klgf6+HLoq/APiGiIwK9rmviBwpaP/17wEwU0RmBK/VWkSmiEhJ8FbiAgA/FpFMETkL+m1PahmsWdZsFLFuWbdRw5plzUYNazYKNetaphEsuRmqC4D7kHgbby+AuUeeR+Itwu8BWI9EE9R8ACXBc2ch0US2L/jvWUmvORtJDV0ArgUwJ/j3JwCsROIqdweAWxE0dQEYD2BVsB+3ug8bwfp6X8N5wT6VA/g1gNe8z/f54HMcAPABgBHB4xcB2BRs943gsbHB9nsAlAF4GkD34LneSLxFegCJZrTfgc2LrFnWLD9Yt6zbGH6wZlmzUftgzUavZiXYGSIiIiIiothq6VvdiIiIiIiImh0vfIiIiIiIKPZ44UNERERERLHHCx8iIiIiIoo9XvgQEREREVHsZRzrSRHhkm+nEOectPQ+nKh41mzYtyWGX2bo13lsztVHvmaBuNYtHQ2PtRQ1rFmKmqPVLN/xISIiIiKi2OOFDxERERERxR4vfIiIiIiIKPaO2eNDRC1B35aakdHKZNTVHVaxc/XNukcnR2Nuv478beZERETUQviODxERERERxR4vfIiIiIiIKPZ44UNERERERLHHCx8iIiIiIoo9Lm4AwG+YbtUqU8XZ2W3MFq1b56o4MzPbvqzXcF5dXWVSKg/u83IOqriurs6+biyHV56aMjIyzWN9+45U8cRpM03OBwvmq3jhwhdUHFZr8aibOHwN0ZeWlm4ey8nJV3FxcWeTU1V1QMX79u1UcU1Ndchn4/ecos4uypKern+GWrVqbXIyW2WpOC1dn7IdOlRptvHPIeKx8A1Fx7EXZ0oP+d1RV6/Pcw8frg153ab7PcB3fIiIiIiIKPZ44UNERERERLHHCx8iIiIiIoq9U67HR8Re6+Xnt1XxoIHjVDxq0iSzTY/BPVRc2KHA5Byu1fct7t+z3+Qsf3u5ihe8/YqK169/32xz8KB+Hd7DG11+rxgAnHvZx1X89a992uQ8/abuA9r8xWUqLt2yugn2jijBP24WFXU0OePGzVLxlCvOMTllm8tU/MBf/6TiTZt0HQNAfX1YnyNRS9E9DH6vDmCP6+3bdzc5ffuOUHFJr94mp9vAEv26uboPaNUCe5xf8PbLKl679j2TY3uD2EcXb80z+FvEvm52dp6Ku3btr+IunfuYbfaW71Dxxo1LTc6+fbu8R46/ZvmODxERERERxR4vfIiIiIiIKPZ44UNERERERLHHCx8iIiIiIoq92C9u4A+I7NTJNhBOPudSFU/7zDQVTzptsNmmMMcbYJoRNpTJebFdhGDb9DNVPGeRXljhkd89aLaZN+8pFe/bV2ZyuOBBqvIbY+2PYH47vVBGcZ5dAGH0aQN0TtsuKt6ydY3Zxjk2sNLx8Yc69+s7yuRc/OXLVDxj9AiT89KixSrO+1eRisMGo3JxAzp5bLO2P4DRX7igc0iz9ohRemGPUdPtz8u48aeruF+nTiYnN0sPMPWbyfdcpgcCA8Czb+lziL/99DaTs2jRSyo+fLjG5FA0hC3Y5R+v09IaPtWvq9NDQ8POF9LS9OfKyysyOSNH6vPnyZfouPcwew6+afkmFT/4xztNzuLFs1V8IjXLd3yIiIiIiCj2eOFDRERERESxxwsfIiIiIiKKvVj1+GRlZpvH+vTVgx6nX3qFyfnkdTNVPLhrVxW3ztT3S4YJux+y3nssLWTYU4732p2m6J6f/Hb5Zpua7x5S8dtvP25yDh2y9/5SKtA1UVNzyGTsKytXsd8rBgBdivS9te3alZgcouMRds94QUEHFU+edZ7JuWDcGSpu3aqVydmxQQ+qqzpYoWL281Dzsb9/MzN1D02bNsUmp3PnvioeNvIsFZ8+ZZjZxu/f6dGunclJ9/olDhyyvwt2H9DDytu01uc4BTk5ZpuLJ49X8cZlG03O2rWLVLxnzzaTQ6nA1qw/MDesz6avd97r9xLv3Kl7agBg165SFYcdi/1hpOMnnW9yLv68Pp8e36+figuy7Xn6lr66T27FvBUmZ8WKuSpmjw8REREREdEx8MKHiIiIiIhijxc+REREREQUe7zwISIiIiKi2Iv04gb+YgbDhp9jcmZdd5WKL75wisnxB4dlZjT8v6W2Tjd+HaqtNTmV1dV6m8OHTY7fAJzXurWKpw89zWyz9QY9cNVvVASAzZt1cxgHmqamsAa90pW6yfBgjc3Jz9Z10rFLNxWHDUblkDpqDH/4HQD0768XLphx8WST0zYvT8VrduwwOX7TalnZZhVzyC41FX+RjpycNianpEQPgh40aILJGTxeDzAf/THdOD60xC4s4w8rX751q8lZMEcP8/1gzgcmZ+9uPZy8jze4+oYv2cWa/MWZhp411OTk/F4PyebiBqlCL2bgDwwFgNzcQhUPH2bPe8+aOVXF29bp7+/ShXbRBF9xcWfz2LTLL1HxBZeebXKGePXnn9OGnV8X5+qhwNl5rU1O2DnN8eI7PkREREREFHu88CEiIiIiotjjhQ8REREREcVeCvf46HsQc3PtIM/TTpui4mu/+wWTM2viWBV3yLev4w8S89WE9Ob4w8Y279ljchauWqvirWu2mJy8Qn3f8bSJ+l76Xu3bm20mjR+u4mf6jTI5W7euUTH7O1JTXZ2tre3bN6h438GDJsfvp+g9rLeKMx6yfRqsAQrj90IUF3cxOZNmzlDxqF69GnzdRes3mMeWvT9fxYeqK1XMHh86HmFDd7Oz9TGyW7dBJmfiNG/Y4qzxJmdkfz1c0R/AuG3fPrPN88/MUfGcJ14xOf5Axt27G+6zKS3Vw1LPuuQsk+P3HOVm236JrCw7RJJant/Tk5VlB9T26DFExdM+caHNGdJTxQcr9DlEZqb9/o8dr4eRjps5zuRMnzRaxYUhA3QrqqrMY8nChu76/fD79x4wOYcP2z7648V3fIiIiIiIKPZ44UNERERERLHHCx8iIiIiIoq9FOrxOXZPz4QJev1wALj6259R8UXjRpucfO9+XBG7frnfw1PtxTsrKsw276zV/TuLXrKzdBa/uVDF27evMzk5OXo9/dpq3Ydx/WXnmW26t2ur4z59TE7m2/q+XvZ3pKb6ejtfyZ9tsvuAvd91QGe9xn7HHh1UnJlp7+s+dMi+DpE/H6FXr9NNzsRzda9kblaWydnh9Tq89fhbJmfD+iUqtvdts8eHPrqMjFbmsXbtdK/L6AnTTM6MT+vHxvfta3L8c4Z53u/+F+59yWzzwuP3q3iL13MLALW1uk847HeB/3Xl5urzhfYFtme53uuTKy/fb3KqquxjdLLZc1G/1oqLO5mc0Wd9TMUDRw0wOYe9OZOZ2brnd+KF9mdh/DTdX35G794mJzM9XcUflJaanDWb9cyqUQP0z5Q/1wcAyvbrety9Y6fJqatjjw8REREREVGj8cKHiIiIiIhijxc+REREREQUe7zwISIiIiKi2GuhxQ1sU1dDixnccPNNZpvzh+tBnmENt3Vew+C+ykqTU+oNH127UzdWrXh3pdnmgzeW6nixbeTdtk03NFZV2eZyf5DU/Of6q/jCaWeabToW6AbHdiV2yGlGhh1gSanINnPv3btDxRt2lpmcCf36qbhX324q9of3AUBFxa7j2UGKHX8hGX08mTBDN88CwBnewFL/uAoAr69coeL5r802ORX7d6s4rKGbqCFpabrJuk2bYpMzYMAYFZ93o10oaPLAgSqud7Ye/cUMnvzrMyp+4cl/mm38hYzCBlX7whZeysnR50VnTtdN6UO6djXb+MPVFzy/wORUVOw2j9HJ5Q8rBYDsbD3Qrv15cAAAGZlJREFUvnfv4SZn7Pm6rkuKi0zOOu+cYcTZ+nUGhdRNt2L9MxR2jH9l2TIVP/y/j5qc4i76dcYO0ue0/sILALB6+3YVb9myyuQ05meosfiODxERERERxR4vfIiIiIiIKPZ44UNERERERLF3knp89L2rWVnZJuP0089W8af+8zoVXzjC3uuYm6UHIe337m0FgCWb9TDIue8sMTnb1m5T8c5NusendL0dPLp+/WIV79ixweTU1Oj9cSH3Dztv2FhttR7SVBNyP2SGN0Qqv60dYuYPJaToqKwsV/GebQ3fj92zfTsVFxZ2MDk7dmz0HuGwyFORf2955856APIZM0aZbQpyclS8a78dgPj6Q2+oeM2ahSantra60fv50dj+iIax/qNDf3/93tgBA/SAXQCYddNlKp46ZIjJycnUvbBLSjebnJcfek3Frz73kIr9Xl4AqAv5ve3zfw7D+jLPPf96FZ9/1VQVhw2D9Psw3n75RZPDAaYtQdew36cGAPn5ejj90DFnmJwh/XW/Zbs29vyvY77u28zP1j8vrVvZgb8HqvX56rMLFpmcf/3qPhWH9eJcNl6fu3cp0j1IldX2d8Daxfoce+dO/1ylcT9TjcV3fIiIiIiIKPZ44UNERERERLHHCx8iIiIiIoo9XvgQEREREVHsnZQO+IwM3UjVp88Ik3PlVz6j4lmjdVOXv5ABYBczeHTeOybn7cf0YNH1y20joj/My28u9wdKAkB5uX7s0CE7GLUxw/nqvaFMaWm6Ac5vvgxTXWWbxfxFEyg6/AbwTcttw60/XMxvcGzfvrvZZtUqPciONXJqysrSCxWcPnySisf31cNxATtcceGGDSZnweu6CfzAgb0mxx4TP3oNhjUFt2qlh1f7gx8BICNd/x6qPLhPxdXVB802TTk0j46fvxBAp066wXvyrHPNNudN1IMei/Ps4gF7Duih4nPeft/kzHv5VRVv26YbscN+z/v7658DAUCHDj1VfPb0K0zOlV/Wg9yHd9fH9aVbtphtHrr1ERWvXm0HmDZlozg1jn8MDVvkq0MH/f3tP6q/yenqLRYQVtdZGfrU3j9f2LLXHpuffV2fPz/8+3tNzpIlr+v9628XXzht3GAV+wtwLN60yWyz8IV3VVxevtPkNOViNHzHh4iIiIiIYo8XPkREREREFHu88CEiIiIiothr8h4fEXst1b59NxVPv9Tey3rR1AkqLvQG5oUNJ31igb4v8KFfP2hyli9/W8WHQgZ31dXr+13rvfjgwQqzTVWVvjc4vJ/HvyfRDtnz77cfMGagitu1aWO2qa7VQ053b9llcg4frgnZH4oCv/7KNpeZnJrDuvcgN0v3OHTuqu+BB+x953V1DfegUdTZY05RUUcVDzt7mIrbhxxzDnnHnPfeXmpy/KHOtbVhx6CPfp+2/zslrH+na1fdlzRsxBST06VPZxUvm6+HWS9ePNts4w/S8382qemFnUPk5uqBjMOG6YHn51082WzTrbi4wc+1tVz3865fvN7kHPTOGfzejFatbB+uf87To8dQkzNx5jkqnnXx2Sanc2Ghilfv0L3Fd91mz3lefE73Zhw4UG5yOLz35PN7fPLz25mcwafpQbxjzrBDd9t6x+ewYaS13vnB+jJ9DnH3P5402zz7rwf0Nt7xHABaZ+Wq+KzzZpic8f30sfiw10/27lI79HTJEj38uvkGXSfwHR8iIiIiIoo9XvgQEREREVHs8cKHiIiIiIhijxc+REREREQUe02+uEHr1rnmsWHDdBPfzE9MNTklxW1V7Ddvz16+zGzzwC//peK5c58wOWFD9Hzp6fp/w+HDtV5sm3QbM9jOb9L0Pw8ADBo8XsVjp45Ssd+0DgDLt25V8Yr3lpicsGF8FA3+Qhml69eZnGrv58MfdNttoG6uBWz9cThj/KWn22GfnTv3UfHwMXrgXFZIs6zfHDv/eTssuqLCX2TlxBcyAIA2bXST+vDh55icC669TMWTJ400OV0K9eC/2Uv175Q/fFsPsgaAPXv0sbamhosbND3d9B027LNbiV70Z+qn9DnE6d3twOZMb4ijf04B2Ibzrv26mpwz9utFB/oP1EMbe59mF5Lp2KuTioePGGBy+nXUOTUhQ0XfWLlSxU/8QZ/j+AsZAMDu3XqoqXNcxKZl+ANL9UJWvXvrRWUAYPKVepGOQV26mBz/d31YXa8r0wNAH3jgBRU/9Nc7zDalW+yiAz7/XH7yrAkmpzhXXwNs3KV/L7zz7Hyzzdata1QcvlhY0+E7PkREREREFHu88CEiIiIiotjjhQ8REREREcVeE/T46PsY27a198hOvGSKis/o1dvkpHvDFT8oLVXxU3c9b7aZN08PYSov32Fy/IFzYfeQ19RUqdg5fW964+43tPezp6Xp++uLijqZnBlXXari8X37qtgf/gQA85YsV/GaNQtNjt+nRNHh19/evbauD1brAV/52XqoXpe+9t7gzEydU1NjhwJTvGSk236J3n1PV/GgrrpW6p09lq3xBieuWDHX5IQPLG3Ise+DB4ChQyeq+OpvXG9yZp41RsX+AGwAyPD6nUb07qnidu3sz4z/s0hNz++zyc0tNDkjx+regkljh6vYP/4Bto5rQ3oaO+TrYbhXzLL9YxXTdB+u318ZNvDXHyoZVkc79u1T8Rtz3zc5D9z2dxUvWTJbxQcP2oHs7OlJDf7A8HbtSlR85rkfM9tMGztCxQUhx7E673zU76EBgAcffFHFD//lbyreXLrCvq7389GmTVuTc9YFurfuzH79TY5/zvrSvEUqfvW5h8w2VVV+HTfvcZfv+BARERERUezxwoeIiIiIiGKPFz5ERERERBR7J9zj48+J6NFjsMkZPUHfU57XurXJqajSfTbPv6zvIX/jRd3PAwB7925XceP6Whq+/9Xej9vw/YZhvUMFBe1VPPnsK03ORRfrGQGF3hro63bq9dgB4LUHXlfxzp0bTQ7v840yXW8VFSHzRSoPqLhLkZ5R0qmn7SfLytL3wTdmxhVFW1jPTK+hPVVclJun4tqQuRBrVuljTHm5PS4dTz+M//ujS5c+JueCz1yu4zNHm5yCHF3bLuSY7d8bv/+Q7pMrL9ezigDOujoZ/Lk9Xbr0NTkTLtbzQvp17KjiNK9PCACqanTP2d5KO9vO78XpWFBgcgq8/qEdFbo3x59xBQCbtuieuE3L7O/oBS/pWVgrV8wzOX4vRk1NtZfBHrTUYOsvJ0f3jw0bNkXF51+mz/0AoHvbdg1+pq179e/tx597w+Q8dtfdKl695l0Vh50r+z+HAweMMTnTZp2l4jYhvXXLtug5Us/f+ZyKt22zcwnrQnrZmxPf8SEiIiIiotjjhQ8REREREcUeL3yIiIiIiCj2eOFDRERERESxd8KLG2RkZKq4z8ChJqd/J9to7dvgNQi+86xu/NtSusps4w/MO/5hc8eznW5my87OMxljx85U8TX/8QmTM6SrHvjqN2S++JZuSgOAt994SsUcRBlvVVUHzGNlFXZwXbLOxXYIYHZ2vvfI1hPZLUpJ+rjUOuS41G1QdxVnZehfAwcO2ePJ+g82qPhQSE0ez/4VFnZQ8bSZV5ktLr5gsoqLcu2CDdW1ehGCmpBm2UxvIYUlK9aqeNvWNWYbDjBtag03gY8cY5u+p40bqeLsTH3eUV1rm7UXbdQLCqwptce7Uf30Yhpdi4tNTq1XSwvXrlfxq/e+arZZ/O5bKi4tXWly9nkLhNTU+gsXcJGiqPAXBgCAPn30kN0LP3eRis/o3ctsk+4NPd213/6ef/I1vfDXQ3/8u8lZvvxtFfuLGfhDgwG7GNf0qy4xOcO791DxoZCfu6ef1YstvLtAL25w6FCl2eZkL9LBd3yIiIiIiCj2eOFDRERERESxxwsfIiIiIiKKvRPu8WnVKkvF7UrsAKZ8b8jR4ZD7r1dv18NIN65fpuJD1fa+wKa7/9re76ieDbkfMj+/rYonTrzc5Nx082dVfM5gO9y13vsaZi9fruKHbr3HbOP3O/E+4Hg7eHCfeWz7PvtYsrZ5bcxjBQX6Z7O0NKzuU72n4dg/q6SF9R4WtNdDGv3hj+UH7aDH3Vv0EN36Rh1z7PcqM1P/vhgyRA/Eu+zGC802PdrpY23NYfv7o7Ja90dktbK/2kq9wX8LXtD9k2W7Ss02PLY2rbQ0+7fWzp11n83ZV4cNdmxrHku2ZscO89gTD72s4p0bbE7FdN1Dce5ZdjiuP3A9w+uJ275ZD2wEgHXr3ldx2LBoO0Qy1Y+9dER6uq6BkpIBJueCq3VP96yJY1Wcm6XrCgAqqqpU/Nibdqjtg7fq4aRLFs82OX7/u38Om5lpB4+OH3+ximfOmmxy/IGlc9fYvshXHnxexTvLNqk4FY6pfMeHiIiIiIhijxc+REREREQUe7zwISIiIiKi2OOFDxERERERxd4JL26QJvraqaBdgcnJ8AbHhS1KcLBSN3X5Qxvr621DlN+wdbyLHfiv06qVHo7Wrl2J2Wbi5MtUfO1/XG1yJg3QDW9hiyS8s1YP0bv3l/9S8aJFL5ltqjmw9JRy2GtUBIC9O3SzrL9IRn62bZxs00Y3CIfVY+oPbEz1/Ut9YQ3myWoOHzaPHfYGhIY1qPqvm5aWbnI6dOip4kmzpqt4aIk91rbyGomrauzQvNat9ADBsMF6z7+sB/+9+dLTKg5rQKem5Q8rBYCJ0/Sg7+mjhpucTG9BAb8J/LUFi802rz2hByeWeU3WAFBepheJad/VLs50Zv/+Kh7WQw8AnjtsoNnm/7d3b8FVVlcAx1fMPSchJEIChPtFRKwoWoU6tmJFsePYTp2iPtipnelMZzr1oU999qk6fehUZzq2VR+0WhDEAooUCCYqatCQQCQCQiAlYELuycn1JOlDpzOutbYkmkBOvvx/byuzNrmc/X3n25y99qqq0gcrxOPtLgdTQ0qKv18WFMxR8d0bH3Y5D2/eqOKiGXru9/T7hrX7ampUvOPZbS6nsnKfikONby17b16wwM/ZTb98QMWrSua5nN4B/SxycJ8/fOHo0TIV24MWkgGf+AAAAACIPBY+AAAAACKPhQ8AAACAyBt3jY+YOoHe7l6XYhuWZqb5b5sT042RbLNF2zBKRGR4WP+7oZoF+7W01HSXE8udqWK7/3HTI5vdmM2PbVLx6pISlzNk6pKO/sfvMX75ue0qLt3/qoo7u1rdGOocppehYV9z0WH2ptvanMw0P8/tNRWq9RgamvzmYpg4iYTfX93Z0qlie58qiMXcmIWrdF1DQcFclzMyckHFsRxf77lmzd0qXv/DW1VcGPjetkY0lpnpcrr69PtOee3nLmfvK7tUfKZO14XY9xOMn63zWrVqvcu5/3FdCzF35kyXY+dodb1+L614q8KNOXXqsIoHArWx1dWlKi7Z4d/HF/5K3zeXFc1W8cZH7nZjLp7W10Jp6asup6PjkoqHAo3dea+fDPqZMdQEev26H6v4sd/+1OWsmqdrZGzt5IHjn7kxrz39TxUfOvSmy4nHL//eL+Kfe2MxfU3d/r373Zh7bte1dRmBZ4jDZ87on293ucvpMs+syVg3zCc+AAAAACKPhQ8AAACAyGPhAwAAACDyWPgAAAAAiLxxH25gi2eb6ptcTnefLiqckZ3tcm5aqItn123coOLm5vNujP1aqMlpbm6BiufNW+Zybr3r+yq+8yd3qnjD6hvcmFl5eSoeDDT9K/tcF9hu/+sul7Pz9b+puKXlgslIvsIwXF2hotfmhmadY+a+bfgnIpKbq4vNQw0mh4b8PMZUou8XPT1dLuP8CX3fjG/QDfBChxtsflQXww72+wahn777oYqL5y5wORt/oQvZ1y1fruLMdF9Qa/UH7rUHPjuu4h1/esPlfPKJbmjZ1xcf9XthfGIxfc+5Y8M9Lsc2CE0NHLrS1KkLut8r+1TFlR+/68Z0d+umoaF724ULuoF46e4dLqdoYZGKZz2s5/BtS5a6MS2/flDFbW3+uejYMV0Y3t7e6HISCX2d+Wccng8mmj30Z/78lS5n48914+XvLvVzwKo8e1bFrz/nDy4oK9uq4s7OZpczlsMC7GFgc+fqn+/W+/ShMiIiJYWFKh5I+Hv8RxXHVHzunD+gYWjIj0s2fOIDAAAAIPJY+AAAAACIPBY+AAAAACJv3DU+g4N6f3htVaXLsc3GbH2MiMjCa69V8eNPPKTiFWtXuDEX6y6quLezx+UsMI33rrtukcu5cf58Fc+eMUPF1wQao3b26oZ5bx+pcjlbnrHNqPy+89bWL1U8MkIDSWjDgQamXa26dsPW+IRqJfJn6/32oYa/iJYe0+xOROTw/kMqPvLAOhXftdLvabfN+J78zaMu5+wj96o4OzAHlxUXqzg/J8flWF2mRvTAZ35f+bY/blNxeflWl9PRYffLUx9xpc2cqV/vlbf7uZVvan57B3zT3UOnvlDxR7t1PVl9va7xEvHPJqHaCPt+e/78CZfzzqu67qegWDeDfPAH+voREbn3lpv09/n9Ey7n7ed17dCRI6Uux9b8xuOj1y1RBzQ+6WkZKl60aLXLWbFysYpDdWmnm3TN1raX96j44L9fd2Pa2vTzYKhu3b6eKSn+e+fk6GfYZctuUfGqG32te5r5Hc63trqcmnJd49PaetHlTIU6YT7xAQAAABB5LHwAAAAARB4LHwAAAACRNwF9fPSZ3adPH3E5pTveU/H1Zr+4iK/xsXU3K8zecBGRhNn/OBzYw5tp+pmkpfreJXZ/ZsL0TTnxpd53KSKyc4/+nXa95PeUV1UdUHGopwb7bzGa0J7ZloYWFQ+aOWvnvYjIjFm2xof/94i6AVPnICJSVaVrCbb8Rfd4yHryZ26M7bNWZOogRUSK80evIQvVS35Ve4+v09xTXa3id154x+W8//52FYf2nnOvvfqysnRPqPQMX/cV79dz9EJ7u8vZ+8o+FVdW7lVxd3ebGxOuj7h8Tk9Pp8upMf12tj2ra0AysnQsIvLA+ttUbGt+RESyfpel4uy/+3q32qOfqLih4aSKQ/N8YEDXxIXbvnAtfJ0R87fp6vJz61ilrgUbDLxHf7DnYxXv3b5FxU1NZ92YUM8+T99Dbd8hEZHs7FwVFxbNVnFelp57IiJdfbpuveKL0y7nRI2+F/f1drscX0uXfHONJx8AAAAAkcfCBwAAAEDksfABAAAAEHksfAAAAABE3rgPN7ANwELFdjv/8ZKKM7J9MeDmxzapeMWcOSrOyfBjssbQgNEeVBBqjmYbNe3/4FMVl20pc2M+/PBfKm5uPu9ybAM14NsINd7r7NTNGLtNk8fMXF3cKCISy9eFxqmp4778keRCDZFbWhpUvHv7Cyq+eNbfy+586C4VL795ucuxBx7Yww5EwkW1X/X+yZPua+Vb9f33g4M7XU5He5OKx1LYjiuvo+OSiive/tjl2NfqdJUvqj7wlj68orHxnIrDReGjF1XbW2voIJnePl3AfbxWN0/d9XyhGzM8rP/hNTf46+Vac48uWVHici7W69/TXruZGbr5q4hIIqGfccZWMI//s3Ogrq7a5bz2Z30IS0GBP3zr3DndaLmh4ZSKBwf9s+hY2ENj0lL9gSHp6fo+m3KNHnPkmL/P1pyoU3HZVv/caxv82mtDJPy8kmz4xAcAAABA5LHwAQAAABB5LHwAAAAARF7K5fbjpaSkTMhmvbQ0XZ8zp3ixy7ll7X0qvuNHd6g4tKe8IKZrFvoGB11O45e6FuLEYb+3sbaiRsVHqw+quOlSvRszMGDrd5J/X+NoRkZGRi+aSnITNWeTSajR6M1r7lHxUy8+o+K1ixe7MX94WtfavfjcUy4nHu/4Fj/h5InCnBWZ7Hmr/4SZmb5uID9fN8ArDtzD7T73BYv9PXvJTbpZ6qySWSqu/ajWjako1w1Xz5ypcjm2geW3rfm4WqIwb8cyZ9PTM1VcVLTQ5eTmFqg4HvdNRBsbdf2Br4+YqNc20HTXNIi0v5O9NkREliz5jooXL13tcmxtU3PTBZdTX39cxZea9LNIX3/cjbF/m/Az3jf/e02XOTuWBqFpabquJvQ3Hh7W96Cx1VqN5XXRP19qaqrLsNfU/PkrVVxYOHfU71JXd8x9zTZd9c/BIlPhPssnPgAAAAAij4UPAAAAgMhj4QMAAAAg8lj4AAAAAIi8q3K4QeBfdl/JyNAFg7Y4Kxab6cbk5OSpOJHwhxv09OhCya6uVpfT36+bUV25wsnkNn2KF6ca/7Lk5enrY926h1R8/VpdXCsicmDXGyo+efKwy7HN75KP/luMjAxP+Tkrkvzz1h6wYYt7//c1fYhNdrZvomsLwfPydPNH2/BSRKStrVHFvb1dLsfe+0ONW5PJ9LnXjl4obhsyhprP+ueUybxcvnnxe0ag0aido4lAQ8vEkJ7X9m8zUQcXjMX0mbNTjX9Z7PyzB3KE5qy9hw4O+oML7IENyY7DDQAAAABMWyx8AAAAAEQeCx8AAAAAkTdJNT5IRuzhncr0S2f3zYsk2z75iRGFOSsS1XnrXxo7L+1e8+Sv75gYUZi30Zyz+DrMWUw11PgAAAAAmLZY+AAAAACIPBY+AAAAACKPhQ8AAACAyEub7B8AuPrGUqM51Wog9c97uUNLgKvDz0E7LYeGbEM85i0A4MrhEx8AAAAAkcfCBwAAAEDksfABAAAAEHmXbWAKAAAAAFHAJz4AAAAAIo+FDwAAAIDIY+EDAAAAIPJY+AAAAACIPBY+AAAAACKPhQ8AAACAyPsvdGpzWTIjtxgAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 1080x360 with 10 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "np.random.seed(100)\n",
    "random_idx = np.random.randint(len(X_test),size=5)\n",
    "plot_five_reconstruction(X_test,y_test,random_idx,encoder,decoder)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "def find_idx_from_label_dict(search_str):\n",
    "    for i,val in lebel_dict.items():\n",
    "        if val == search_str:\n",
    "            return i\n",
    "def letter_digit_gen(input_str):\n",
    "    cls_idx = find_idx_from_label_dict(input_str)\n",
    "    latent = np.random.randn(latent_dim)\n",
    "    latent = np.expand_dims(np.random.randn(latent_dim),0)\n",
    "    generated = decoder.predict([latent,np.expand_dims(to_categorical(cls_idx,n_label),0)])\n",
    "    generated = (0.5*generated) + 0.5\n",
    "    generated = generated.reshape(img_rows,img_cols)\n",
    "    generated = np.transpose(generated)\n",
    "    plt.imshow(generated,cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "letter_digit_gen('V')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "# def digit_image(digit, latent=None):\n",
    "#     if latent is None:\n",
    "#         latent = np.random.randn(latent_dim)\n",
    "#     return from_latent(digit, latent)\n",
    "\n",
    "# def from_latent(num, vec):\n",
    "#     vec = np.array(vec)\n",
    "#     im = decoder.predict([vec.reshape(1, latent_dim), to_categorical([num], n_label)]).reshape(28, 28)\n",
    "#     return im"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Interpolation on the test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "# X_test_encoded = encoder.predict(X_test[:10000], batch_size=1)[2][:,4:6]\n",
    "# latent_space_df = pd.DataFrame(X_test_encoded,columns=['vector1','vector2'])\n",
    "# latent_space_df['label'] = np.argmax(y_test[:10000],axis=1)\n",
    "# latent_space_df.head()\n",
    "\n",
    "# sns.set_context('paper', font_scale=1.5)\n",
    "# sns.set_style('white')\n",
    "# plt.figure()\n",
    "# sns.lmplot('vector1','vector2',data=latent_space_df,hue='label',fit_reg=False,scatter_kws={'alpha':0.6,\"s\": 40},\n",
    "#             height=8,palette=sns.color_palette(\"hls\",latent_space_df['label'].nunique()))\n",
    "# plt.title('Latent space by class')\n",
    "# plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "# # not possible with label as input\n",
    "# def interpolation(img1,img2,step,encoder,decoder,plot=True,saving_folder_name=None):\n",
    "#     img1_encoded_z = encoder.predict(np.expand_dims(img1,0),batch_size=1)[2]\n",
    "# #     print('1: {}'.format(img1_encoded_z))   \n",
    "#     img2_encoded_z = encoder.predict(np.expand_dims(img2,0),batch_size=1)[2]\n",
    "# #     print('2: {}'.format(img2_encoded_z))\n",
    "#     step_size = (img2_encoded_z-img1_encoded_z)/step\n",
    "# #     print('step: {}'.format(step_size))\n",
    "\n",
    "#     step_list = [img1_encoded_z]\n",
    "#     for _ in range(step):\n",
    "#         img1_encoded_z = img1_encoded_z + step_size\n",
    "#         step_list.append(img1_encoded_z)\n",
    "#     if plot:\n",
    "#         os.makedirs(saving_folder_name,exist_ok=True)\n",
    "#         fig,axes = plt.subplots(int(np.ceil((step+1)/5)),5, figsize=(5*4, int(np.ceil((step+1)/5))*4))\n",
    "#         for i,(ax,vector) in enumerate(zip(axes.flatten(),step_list)):\n",
    "#             predict = decoder.predict(vector).reshape(img_rows,img_cols)\n",
    "#             ax.imshow(predict,cmap='bone')\n",
    "#             ax.axis('off')\n",
    "            \n",
    "#             predict = decoder.predict([encoded_z,np.expand_dims(label_set[idx],0)])\n",
    "#             predict = (0.5*predict) + 0.5 # image back to 0-1 from -1 - 1\n",
    "            \n",
    "            \n",
    "            \n",
    "            \n",
    "#             fig_indi,ax_indi = plt.subplots(figsize=(5,5))\n",
    "#             ax_indi.imshow(predict,cmap='bone')\n",
    "#             ax_indi.axis('off')\n",
    "#             plt.title('step {}/{}'.format(i,step),fontsize=24,fontweight='bold')\n",
    "#             fig_indi.savefig(os.path.join(saving_folder_name,'step_{}.png'.format(i)))\n",
    "#             plt.close()\n",
    "#             if i == 0:\n",
    "#                 ax.set_title('From Image 1\\n0%',fontsize=24,fontweight='bold')\n",
    "#             elif i == step:\n",
    "#                 ax.set_title('To Image 2\\n100%',fontsize=24,fontweight='bold')\n",
    "#             else:\n",
    "#                 ax.set_title('step {}/{}\\n{:.0f}%'.format(i,step,i*100/step),fontsize=24,fontweight='bold')\n",
    "#         plt.subplots_adjust(hspace=0.3,left=0.1, wspace=0.02)\n",
    "#         if (step+1)%5 != 0:\n",
    "#             for i in axes.flatten()[(step+1)-len(axes.flatten()):]:\n",
    "#                 i.set_visible(False)\n",
    "#         fig.savefig(os.path.join(saving_folder_name,'summary.png'))\n",
    "#     return step_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "# np.random.seed(150)\n",
    "# saving_folder_name='interpolation_images_EMNIST'\n",
    "# random_idx_2 = np.random.randint(len(X_test),size=2)\n",
    "# print(random_idx_2)\n",
    "# _ = interpolation(X_test[random_idx_2[0]],X_test[random_idx_2[1]],10,encoder,decoder,plot=True,saving_folder_name=saving_folder_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}