ResNet.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ResNet-50 Architecture with Data Parallelism"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The network in this notebook is an implementation of the ResNet-50 [1] architecture on the CelebA face dataset [2] to train a gender classifier.  \n",
    "\n",
    "\n",
    "References\n",
    "    \n",
    "- [1] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778). ([CVPR Link](https://www.cv-foundation.org/openaccess/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html))\n",
    "\n",
    "- [2] Zhang, K., Tan, L., Li, Z., & Qiao, Y. (2016). Gender and smile classification using deep convolutional neural networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (pp. 34-38).\n",
    "\n",
    "The ResNet-50 architecture is similar to the ResNet-34 architecture shown below (from [1]):\n",
    "\n",
    "\n",
    "![](images/resnets/resnet34/resnet34-arch.png)\n",
    "\n",
    "However, in ResNet-50, the skip connection uses a bottleneck (from [1]):\n",
    "\n",
    "\n",
    "![](images/resnets/resnet50/resnet50-arch-1.png)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following figure illustrates residual blocks with skip connections such that the input passed via the shortcut matches the dimensions of the main path's output, which allows the network to learn identity functions.\n",
    "\n",
    "![](images/resnets/resnet-ex-1-1.png)\n",
    "\n",
    "\n",
    "The ResNet-34 architecture actually uses residual blocks with skip connections such that the input passed via the shortcut matches is resized to dimensions of the main path's output. Such a residual block is illustrated below:\n",
    "\n",
    "![](images/resnets/resnet-ex-1-2.png)\n",
    "\n",
    "The ResNet-50 uses a bottleneck as shown below:\n",
    "\n",
    "![](images/resnets/resnet-ex-1-3.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['',\n",
       " '/anaconda/envs/py36/lib/python36.zip',\n",
       " '/anaconda/envs/py36/lib/python3.6',\n",
       " '/anaconda/envs/py36/lib/python3.6/lib-dynload',\n",
       " '/anaconda/envs/py36/lib/python3.6/site-packages',\n",
       " '/anaconda/envs/py36/lib/python3.6/site-packages/IPython/extensions',\n",
       " '/data/home/zhangruochi/.ipython',\n",
       " '/Users/ZRC']"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import sys\n",
    "sys.path.append(\"/Users/ZRC\")\n",
    "sys.path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from collections import OrderedDict\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "\n",
    "import torch\n",
    "import torch.nn.functional as F\n",
    "\n",
    "from torch.utils.data import DataLoader\n",
    "from torch.utils.data import RandomSampler\n",
    "from torch.utils.data import Subset\n",
    "\n",
    "\n",
    "from torchvision import datasets\n",
    "from torchvision import transforms\n",
    "\n",
    "from torchsummary import summary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from coke.visualization.image import show_batch"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Model Setting"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Hyperparameters\n",
    "\n",
    "BATCH_SIZE = 64\n",
    "NUM_EPOCHS = 10\n",
    "LEARNING_RATE = 0.0001\n",
    "RANDOM_SEED = 7\n",
    "\n",
    "# Architecture\n",
    "NUM_CLASSES = 10\n",
    "GRAYSCALE = True\n",
    "\n",
    "# other\n",
    "# torch.cuda.empty_cache()\n",
    "DEVICE = torch.device(\"cuda: 0\" if torch.cuda.is_available() else \"cpu\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_transforms = {\"train\": transforms.Compose([\n",
    "                            transforms.Resize((32,32)),\n",
    "                            transforms.ToTensor()]),\n",
    "                    \"test\": transforms.Compose([\n",
    "                            transforms.Resize((32,32)),\n",
    "                            transforms.ToTensor()])\n",
    "                   }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_indices = torch.arange(0, 59000)\n",
    "valid_indices = torch.arange(59000, 60000)\n",
    "\n",
    "\n",
    "\n",
    "train_and_valida_dataset = datasets.MNIST(root = \"data\",\n",
    "                                train = True,\n",
    "                                transform = data_transforms[\"train\"],\n",
    "                                download=True)\n",
    "\n",
    "test_dataset = datasets.MNIST(root = \"data\",\n",
    "                                train = False,\n",
    "                                transform = data_transforms[\"test\"],\n",
    "                                download=False)\n",
    "\n",
    "train_dataset = Subset(train_and_valida_dataset, train_indices)\n",
    "valid_dataset = Subset(train_and_valida_dataset, valid_indices)\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "train_dataloader = DataLoader(dataset = train_dataset,\n",
    "                             batch_size=BATCH_SIZE,\n",
    "                             shuffle=True,\n",
    "                             num_workers=4)\n",
    "\n",
    "valid_dataloader = DataLoader(dataset = valid_dataset,\n",
    "                             batch_size=BATCH_SIZE,\n",
    "                             shuffle=False,\n",
    "                             num_workers=4)\n",
    "\n",
    "test_dataloader = DataLoader(dataset = test_dataset,\n",
    "                             batch_size=BATCH_SIZE,\n",
    "                             shuffle=False,\n",
    "                             num_workers=4)\n",
    "\n",
    "data_loader = {\"train\": train_dataloader, \n",
    "               \"val\": valid_dataloader,\n",
    "               \"test\": test_dataloader}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_samples,labels = next(iter(train_dataloader))\n",
    "print(batch_samples.size(),labels.size())\n",
    "show_batch(batch_samples.squeeze(), labels.numpy(), (4,4))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([64, 1, 32, 32]) torch.Size([64])\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAusAAAI4CAYAAADTU6e8AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAgAElEQVR4nOzdeXzU1b0//td7JntCSEIgQAiEJSwuBZVFBXFFqbu1tdpFbfHS2vbWrmp7219vv9/ee+23vba9t7WtO+5Stcp1V8S9isgiOwRIJCwJAUJC9pk5vz8Yz/uMNyEZMsl8Zub1fDz68MVklhM7bz7HOe85R4wxICIiIiIi7/HFewBERERERNQ1TtaJiIiIiDyKk3UiIiIiIo/iZJ2IiIiIyKM4WSciIiIi8ihO1omIiIiIPKpPk3URmS8im0WkUkRujdWgiJIZ64YoOqwZouixbpKHHOs+6yLiB7AFwDwANQA+AHCNMWZD7IZHlFxYN0TRYc0QRY91k1zS+vDYmQAqjTHbAUBEHgNwGYBu3wgZkmmykNuHl6S+asLBemPM0HiPI4VFVTesmfhjzcQdrzUJiHUTd7zWJJij1UxfJuulAHY6f64BMOvTdxKRhQAWAkAWcjBLzu3DS1JfvWqeqI73GFJcj3XDmvEW1kzc8VqTgFg3ccdrTYI5Ws30+xdMjTF3GmOmG2OmpyOzv1+OKOGxZoiix7ohig5rJnH0ZbK+C0CZ8+dR4duIqHusG6LosGaIose6SSJ9max/AKBCRMaKSAaAqwEsic2wiJIW64YoOqwZouixbpLIMfesG2MCIvIdAC8B8AO41xizPmYjI0pCrBui6LBmiKLHukkuffmCKYwxzwN4PkZjIUoJrBui6LBmiKLHukkePMGUiIiIiMijOFknIiIiIvKoPrXBpAJJ039FHWdNjfhZ9YXpNmfV63/3lL3caLNZsa4fR0dEREREyYyfrBMREREReRQn60REREREHsU2mB5Ipp7qVX1ResTPHrjsDpv/s+YCm2u3jrM5b0U/Do6IiOhoRGz0ZWdrHpyv9/F1/bldqOGQ5ubm2I+NiHqFn6wTEREREXkUJ+tERERERB7FNpgu+HJybO6cMcnmuadFHv41Jq3F5h0Hh9hc0Bjsx9ERDTCf30bxu7mb/9Z3ltRlTGnEj0J5WTZ35mfY3DZEW8xC+hLwBYzNwUxdzs+pC9icufuwPqCySp+nra3r8RElCX++08oyfKiNwaJcmwM5WltNo7XmDpdqPXX3sV3p6616l7dW9WGkRNQX/GSdiIiIiMijOFknIiIiIvIotsGEubu+hE4Yb3P1wpDNd414IeIx9UFdXmxbWWRzzpZdNgdAlCDcdpd0/avBVzbS5o6yQs35Xf/1EUrT5fW9p0vEz2SEtqYcX1pj889HvWJzgU+X3ncFB9s8NaPe5pt3XmLzex9OtLnioQod9/rtOibuZEFe4dSZL0OvIZKnrStm5NCIh4RytH3F+LSmGkbr7i71U51dX8Zra1hp0T6bvzPqXZvn51bbvKFjkM2b2kfY/JeGy2wueavL34aIBgA/WSciIiIi8ihO1omIiIiIPCq122DcwyLGj7F5yxfybH7vjN/anC7ONhUALvpggc1lL+sye2BHNYg8y33fOzsf+Qq05SRQqrsbbb5al+d/fuFTNl+fX9fl0weNto4FELkzUlOow+ZO4+z00s1QS/za+pLujHtR+as2fzhC8zU537B54t3azoblzk5OIe7WRAPMaX3xFzp1NqnM5v0nai1WXLs54uFfd3pQhvj1WjPUr/VU7NNWmRC0BqudHZU+aNPr3EOHTrT5r0/roX5jntNdzkZ8tMZ5TqIk4FxHInY3y9D6kWzdtQxpXU+TzWGtw1Crs/NYP11f+Mk6EREREZFHcbJORERERORRKd0G48vTdpe603XZ/49X3Gtzuuh/z/ygZl7E44ffqUsl/jUbbOZyIXmOuww/RHcuqvz+BJuHn7LX5h+M+7vNc7JqbR7sc5YHoc/ptr7UBXUZ/b6GUyKGcdfbZ9qcV+XsONPe9bBDulkGDld02rzjkrtsnpGpr339DN3t4uWXzrB50Cp9LdPONhgaWL5cbXHZc/Vkmx+/+Tc2Z4m2q7gtLQDgd1vX4LZjZju3633+ckjbXe54UHdOGnNXZHvNJ8a1fmSzaddiDAVZK5Rc3OtfcJzudFZzju6IdOkX37b5+qJ/dPk8Fy7+kc0T79V2zeDGrTEZ56fxk3UiIiIiIo/iZJ2IiIiIyKNSug2mY6YeptIyv8nmOVmHbG5yvtn75uv67XkAmOgeftTaCiKvcg9fCVTo0t8vrlxs8+lZuovRyDQ9JCxTdDeYjwN62Mr/HJ5i86Idp+rzP1ts89AP9f4AcNw+bamB8w16E+y6eUyydByHZpXa/PQ52sJ2UY7W6wWD1tr8zBBtuRnsHHoWbO+m54aon4jTxuK2do1Py+7i3oBfIj9Hc9vMnmvR9/5DtafZ/ME63f1o+Jv6+PLXttkcqN8fxaiJ+p8/P9/m3defYLM4HVjD71tts+nQHZD8Q/Va0zxNd1YCgD2n6/Q2d+oBm08apvO28TnLbZ6eo4fozcjUa0qhX69/rtyKBpsDhdrmJl3dOQb4yToRERERkUf1OFkXkXtFpE5E1jm3FYnIKyKyNfzPwqM9B1GqYd0QRYc1QxQ91k1q6E0bzP0A/gjgAee2WwEsNcbcJiK3hv98S+yHF3tpZaNsrjpbv3H/15Pus3lnQJccv7L26zaPeypyST+01zkUxjnghQgeqxvJ06W8+s/okt2Z2dr6MipNl9fPXn+ZzTtXa9tM9j5d5MvdrXUyuFpbS9K36LJ7sG5fxDgCvagTcQ6h6Dxzqs0dX9OlzJMzdecaH/T3aQxpu4svoM9pWJ+J4H54qGb6Kq1U66b2s7o7S+llVTa77S4tzoFhP9+rbWUA8OwLs2wu2qDv5ew63SFpcl2jzbJLr02B/Vo3lJTuRwLXzYFLjrO58BJtUTFGrzXVBdNsDgzS93/WZG1FmVOqLZAAcP0gbWs5MbPG5kE+rZnXmrUV+t5a3T1szWC9Lv64SK9nrrZ27WeTUP9fX3r8ZN0Y8yaAT1f7ZQAWhfMiAJfHeFxECY11QxQd1gxR9Fg3qeFYv2BaYozZE857AZR0d0cRWQhgIQBkOZ+AEaWgXtUNa4bI4rWGKHq81iSZPu8GY4wxItLtGoAx5k4AdwJAvhTFZS1anJ0gaufrN4ZPP8+2eGF2li6NLD7sfKv4WT0sybduTcTzhtraQHQsjlY3/VEzxtl5pXCLLrdfsuoGmxubdGeKYc9rzUxcpQc+4IB+S940aVtYqEUPQurrMSq+8eU2fzxPW9VePEFb1Ur8Or67Dmm93rVtts35NdoH4+4gQIkpEa41rsaZ+r4MXaoffP7n2CdsbgnpJXhZm+6K8fKjkW0w45/eY3Oo2mkV6NT3NQ/jo64M9LWmN/wFg22uPVv/nr573LM2Z4nOyd4dWWHz0DTduc9tb7lzn+7+BQD/vnq+/qFK20DTG7W9JkO7aCJ2adp7kdbiDwr1kKOPA3qd83+kbaNptc7OgOgfx7obTK2IjACA8D/rerg/EbFuiKLFmiGKHusmyRzrZH0JgOvC+ToAz8RmOERJjXVDFB3WDFH0WDdJpsc2GBF5FMBZAIpFpAbALwDcBmCxiCwAUA3gqv4cZF/5Cwts3j9TFyl+NuIFm//Rrksav1hxqc0TX9fdLIJtPEyFesdrdeO2qWS8u97mwZl60FdJnd7HV6Xfhg8e1NYXhPra5OJwDorBLB1H5aW6ZHnpvPdtHp+uNXowqGN9cZ9zkMYSbVvLWb/T5gDbYDzPazVzLPwT9WCiXefo7b+f8j82T0zPsnl/SA/Te6dJd6bI2h/ZkdAyUQ9/kQma/W1ajxk1B20ObqvSB3MnpKTm6bpx/o73D9bWkp03HG/zP81cavO4NL3WrOkYbvOWZs0P7Zlpc+M+vSYMWR45nS1fr9eItA0bbA426Gv4K8bpmK7Q15hRrNe/xpC2kN647Ys2j3xHbw/VRu561h96nKwbY67p5kfnxngsREmDdUMUHdYMUfRYN6mBJ5gSEREREXlUn3eDSQSd43R5o6RUlwpL/Prr/3L3XJtHPaS3BzdX9vPoiAaAsxTu7mKU+cIHehfn7jFrdvH5I/4ofv1zYI62r2z7kt7+H3Mfs/nS3Fqbl7bqbjW3brze5gNbimye/IrzrfydulMA0UDYP2uYzaefstHmC3KcVjLoe939tKw8S3ddyvxiLVxNnV1fqptbdVek9OUjbB79lFPvzuF9bjscUX+TNN1ipXnOJJsXfO15m782WOvkN/Wn2fzwe5qH/kNrZuQqncOVVm22Odiku8QAiLjmRVzPnGvSwRlar9M/p4cq/aD4LZsfatSWnf0PjtYxrXVeewDqip+sExERERF5FCfrREREREQelRJtMAem6MlcpxRvsXlzp/63yrs79FvBE1c6O2H089iIkoKztOjP091cQhVlEXcL5ukhR803a2vAa8ctsrnAp3X5bIsu7d/63pU2V1y70mZtgum/AymIeqN1qO5+MT5Xd4hIc1pfWo3uTHTAOcloWpZed4aNb4x43rqA7qQxxK+HkZ2cudvm34072+bX/DNsLl02SJ9ouS71E/U3X7bufLTz8zqbOid3k82HnB3G3NaXKf+t14fgem056evhX2nDdDel/Sdovd5U8qrNWwO6y8zvlurhSpOe0IM0/1fbTT/jJ+tERERERB7FyToRERERkUelRBtMZ54udQxJb7b5sYOzbB70rrbKBGv7eDKvcxCAu/uFpDn/utP1W9KSoRmdupAfatdDmIx7qAsPuaB4cdpd3Pe2r2CwzY1ztaUs/zt6MBEAnDtUlzO/mL/G5tFpuux4f6N+Q///vvA5myffud9mtqeRFw3aqe/M9+rH2lxVoId7vdc2xubfbppnc8Mup9VlVeQuSoWbdAenhgnaWlDwFd3x6IGKR20+c4HW2Y9HfsnmSev0Ohdq1QOZeE2hfpGuc56ZFTtsHurTeU5DSD8zTj/gvO/rddeXvvJlac3s/pweXHb+vA9tHpOmtfvLuuk2j31Gx2rieDAmP1knIiIiIvIoTtaJiIiIiDwqJdpgXEHnv092NA+xOaeur98xVr48XdIPfkaXXGpn6hJk4yRdWrnoFG0HeGHTcTYPfUkPvBiy7GObA7t0BwCigZQ2utTm5iklNh+cpK1c3//mEzZfllsV8fgcn94vDTnoytrmUTbnb9V6DW7eHv2AiQbQoCWrbfat0wNUbhjzfZvTWvXv/hFr9D09PKgHeqGzM+J5TUAfM2S503721kibz/jmj2x+5vO/s/m/L7rf5h8d+LrN5b/V606oWdtDifoiot23UNsjp+Rpa1aW6N/rX1qj78nSN/R9H6zT3ZT6ate3TrZ53pffs/mXJe/avKhRD216dfFMm0e9sdxmtw4HGj9ZJyIiIiLyKE7WiYiIiIg8ipN1IiIiIiKPSrmeddeMAj0x7oHxE23O6+rOPWj/rJ4Y9/E1ugXQN056y+YpWdqTOMyvp1+NSdMttG4oftPmKzq+bXPBxkJ9MfasUz/zTdPvTlRfVGBz2Tn63YnrS5+xeWianrh4ZlaDzXm+rvvSj+ZzhStsfvbME2wOfXSiju/t1SDyGuNstxuq1OtL1sfO39nONonBlpboX8Tpmw1V6daoObv1tN9dzomn5Wm6BV7nRL3WuFuvEsWKCTob69bpdrsPP3emzYvyzrB55Bt69+yV+h2OYB+3EzWnT7V56IW6xemPhuoc69XW4Tb/9i09qXTK07p9dzCOfeoufrJORERERORRnKwTEREREXlUSrTB5O3RbRm3HS62ecHod2xefPpJNstJx9tsVm+wOa1ct+ICgOqrdBu70vO1PeDe8udtrurU17t51ZU2d+zV9oAHL/6zzTMzdeknc5Bzgmk6lyypf/mHDrV5y1d1Gf3r85bafM1gPfFtbHp3DWNZ3dwO/KmhzOaPDusWjecXrLf5nOy9Nt9+ymKbf/jdL9g8rlZPSQ1u03YDhHi2KXmD6ezoMseSf5Ru3dg8Sq9zx2Vo60t1wGlF26PbAUe0KxDFitvmdVi3BJ3wYL3ex+d8TlyrtwcPaAtlrzinxaeNKYv40car9Tp0+5jXbC7yaw3cvl1PEB77N62f0Hadz3kFP1knIiIiIvIoTtaJiIiIiDwqJdpgBq/RZZYP1+ry+dYReoLp/5myxOabfnSNzcUvzLL54CRdcgGAiy/Uk7Bucr5hnOUszfxkyxU2Z74zyOb2qbos2mn0/4YnD+uJX7JO7++v32MzFy+pXxRq60vRZP0W/7UF2voyKk1bXzqNvhP/0a5tWte/rSfS+fZlRLzEoB36+UDWAV12fOXUz9j83XNfsvnGgq02f/t4rbEnJ56vz1Ol3/Q3bIOhZOfTWms8WXeAKTtB28dK/Nk2v9+muzkVbtBrUzxPY6QU4fx9HNy49Sh37DtzOPIU3tOn64mpZ2Y5u7s4862du4tsnrJcxxfsp7a1vujxk3URKRORZSKyQUTWi8hN4duLROQVEdka/mdhT89FlApYM0TRY90QRYc1kzp60wYTAPBDY8xxAE4F8G0ROQ7ArQCWGmMqACwN/5mIWDNEx4J1QxQd1kyK6LENxhizB8CecG4SkY0ASgFcBuCs8N0WAXgdwC39Mso+Cm2rsrl8if4H5ncHabvLn097yOZnZt9h88/LLrX5qqJtEc/7lcFrbR7iHP5yMKQHTxxXWGvz26fn2vyTz7xq86tNuvvMI2/MtnnCUj0ww+zRZRzytoStmTRdXt+3S9uxfll8gc1NAf0m/fo6PVCibbPef9Jjh2yWXXpoCwCEGvXwJNPhtIJln6rPO0t3WfIVaM35RNtmjC+yJY0SXzzrxl+o1wXJ11Yvc1Dfy0HnvRtPaWW6A8ye2VoHvy7XHS9qg3oNerxOW8aGrHF+B+4Gk/AS9lrTF04bWFqJ7mBWe9HYiLt9c+gim9NFP5f+1b7pNuev0euZV+q7O1H1rItIOYCTALwPoCT8RgGAvQBKunnMQgALASAL0Z9mSJTIWDNE0WPdEEWHNZPcer0bjIjkAXgSwPeMMRH/CWKMMQC6PBvWGHOnMWa6MWZ6OjK7ugtRUmLNEEWPdUMUHdZM8uvVJ+siko4jb4SHjTFPhW+uFZERxpg9IjICgGf7NNxvvWe8+IHNZb4ZNv+s6HKbX5v6oM1/G687U/xv+o17v7PMUujT2+8oW2Zz5fBXbP7zvrNsfuXFk22e/NgBm0OVVZrb9YAk8r5ErBk5dNjmUS9qW8DKD3WnFncHl7INeoBFaLPWVegou0y47QZmynibD0zV551fqO1l7nL+sv2TdBy1ejuMPpYSW7zqJjReW6/qT9JduLL3a8vJoPW6QxL266FDoUNNNptAp97HdDk/ijjIRTJ0tyTfIG2/kczIiVP7RG05qzlVD3u5+KzlNl+eq/X4ZLP+PqvemWjzhI1rdNxsg0kKiXit6Qu/06bWMGeMzV+46dWI+52ZrfX654YTbH76qTk2lz+bOLvs9WY3GAFwD4CNxpjbnR8tAXBdOF8H4JnYD48o8bBmiKLHuiGKDmsmdfTmk/XZAL4KYK2IrA7f9lMAtwFYLCILAFQDuKp/hkiUcFgzRNFj3RBFhzWTInqzG8zbALrbeuHc2A5nYOVUasvJ7rXDbD5woi7jl/j1m8eHQpEb5Tc4q+9Bo/+Kmky6zavbdJnm7u2600twSbHN4x7Upclgq7u8380yKnmaJ2vGWXr3uUvs6fpeDR3UZfScp3UXo5xuDhrqdtnQ+ba+LzfyS0stp06wufoyHdOP5j5v83k5+tpPNOn91/yjwuaKBueQixDrJBnEs26axjk7wFyqy+dZOfr38ZaVo2weXKm7UBRs1etC+qE2myXY9fvSpOmCdkeRtrQ0jtZa7MiP/Ncw4qKPbX50/BM2fyZDH7+jU3cPu2vnZTaXvqnXs1CL3ocSnyevNf3BuaagVFvC9lysbWcLC9a4j0CWaD395eV5Nk96Ug/JDFbuiOUo+1Wvv2BKREREREQDi5N1IiIiIiKPimqf9WRzaKq2opxweqXNJX5tE3BbX36zT79FDABv7dXdLFo7dMmlsU6XVIe+o/+Ki1/UA16CtVtt5l4W1N8iWl8mlNvYOkZ3vkhr0caWzE27bQ7U7tPHui0xbmtNtu6A5CvSHV+ap+puGgBQ8yVdkr/vVD204owsvX1th36G8Ju1unxZcbfT+rIl8oAyor7ozNH38gWjNtn84+L3bM6brDXkczoP7jpUZvPWVt3OujWkO724cv26s9eJOXpo2JcH6fvb3V0MAILOjkch6LXmYFDbWm7Y+iWbA3/QVoGs53THGKJE5C/UQ/cOTNPry7/MfNrmHF96xGPeb9c/F67XejXVu/pjiP2On6wTEREREXkUJ+tERERERB6V0m0w/g79tv6hDl3Grw3qMuXNOy+1ueGmyCX9oirdUB/OQTAjg7pkaTq0jSbYEbmbDNFAEWdXlvpTdBnxd7/4k81NIa2B7z+4wOaxD+vtwW1VNqeNcg6SOVtbAern6Y4YT875Y8Q4xqRpG02eT9sKtgX0MVe+/R2bJ/5Gd+MIOoeEEcVS8VPrbX6n/lSbn/naiTY/eso9Nk9xdlG6Nt/ZUcLJoV40OKaLs8sF/N3e77DRa9K6Dq2bhav+yebCR7T9ctBSPViMbZaU6MwI3a2v7jSdt30lX9vIMiWyDeb763W3yqIten1J1B2R+Mk6EREREZFHcbJORERERORRKd0GM2itfvt+2zt64EWVc0BGTVOBzYNr9dAYAAgccP7czcExRF4gebk275+my4gV6dpmMtin7+Gffelxmx8/Z4bNexom2VxWoO//m0c8rM+ZoXV1fHrkjhh+0SX89R362hcvvUkff6+2lJnNTosBa4z6SbCpyeacNzbaPGab7qryvTJtz9pzur6vh8/W3SWmFmnO8UXX9tgY0Hazl147OeJn2bW6m0XRJj0IZsy2g3qn2hobIw7XI0pA4uxg1nCizsP+5VzdAcZtfTkUinzP+54aYnP6hs02BxP0sEl+sk5ERERE5FGcrBMREREReVRKt8EEd+luLuMe12XNWzd+0+asel1yDO3THQMAAIbfs6fEYJr1G/D5W/W/0TucJUF3SfGKXK2NM8Y+YXNzSB87yKfv/1Fp2joGZHU7jt8fLLf5v5ZeYPP4v2ud+VdtsTnUrrtgEPUbpw5CTksMNmrOqNRrxLhK3Rms/XVdbl+do7tWOOcm9Y5zOanYVhfxI2nWJf7Q/gM2B9vaQJQsJE2npO1n6U5MjZ/XOvz8IG2NPBzSIrtw3Zcjnqt41SGbgwcPIdHxk3UiIiIiIo/iZJ2IiIiIyKNSug3GOEvswQ269D5oQ9f3Z9MLJSrTosvoQ1c127xg69U2zxhSbfO1he/ZPDHdbXFRnUZ3Z3F3dvnFzkts/nBrecRjClZqK8HEd3RpUrZU2Zyoh1ZQcjOdurtLYHuVzX43x+i1uO8RpSL/KG0v23uaXit+cvzTXd0djzSNs7ljcUnEz2Sn7gCTDDuJ8ZN1IiIiIiKP4mSdiIiIiMijUroNhihVhNq05SutcrfNB+8eb/OzxaNtfnTcbH1svh5SFME5W8J/SP8qGfqh3j75o8iDxLD9Y31e5+AWk6AHVRARUWy0lxfbHJyk7Zrn5FTZfMjpR15SO9XmoW/ujXiu0KHG2A8wjvjJOhERERGRR3GyTkRERETkUWyDIUoFzrfhg/v22Zz/qJOdu0d+r74PLxuj5yEiouTWXqgH85UW77e5yKc7wzzbMtTmjSvH2Dxxz9qI5zKBbto3E1SPn6yLSJaILBeRNSKyXkR+Gb59rIi8LyKVIvK4iGT09FxEqYJ1QxQd1gxRdFgzqaM3bTDtAM4xxkwFMA3AfBE5FcCvAfzOGDMBwEEAC/pvmEQJh3VDFB3WDFF0WDMposc2GHNkm4bD4T+mh/9nAJwD4Evh2xcB+FcAf479EIkSD+uGKDqsGaLoJFvNZDRq68r2am13ubngDJufX3+CzZPvqLU52NrWz6OLr159wVRE/CKyGkAdgFcAbAPQYIz55N9sDYDSbh67UERWiMiKTrR3dReipHSsdcOaoVTFaw1RdFgzqaFXk3VjTNAYMw3AKAAzAUzu7QsYY+40xkw3xkxPR+YxDpMo8Rxr3bBmKFXxWkMUHdZMaohqNxhjTIOILANwGoACEUkL/9fbKAC7+mOARImOdUMUHdYMUXSSoWbSluqJehOX6u1bnftUQO8TROrozW4wQ0WkIJyzAcwDsBHAMgCfD9/tOgDP9NcgiRIN64YoOqwZouiwZlJHbz5ZHwFgkYj4cWRyv9gY86yIbADwmIj8CsAqAPf04ziJEg3rhig6rBmi6LBmUoQc+TLxAL2YyD4AzQDqB+xFvaEY3vmdxxhjhvZ8N/KCcM1Uw1vvoYHgpd+XNZNgeK3xBNZNAuG1xhO6rZkBnawDgIisMMZMH9AXjbNU/J0ptlLtPZRqvy/FXiq+h1Lxd6bYSrX3UKL8vr3aDYaIiIiIiAYeJ+tERERERB4Vj8n6nXF4zXhLxd+ZYivV3kOp9vtS7KXieygVf2eKrVR7DyXE7zvgPetERERERNQ7bIMhIiIiIvIoTtaJiIiIiDxqQCfrIjJfRDaLSKWI3DqQrz0QRKRMRJaJyAYRWS8iN4VvLxKRV0Rka/ifhfEeKyWGZK8ZgHVDsZfsdcOaoVhL9poBErtuBqxnPXzC1hYcOQ63BsAHAK4xxmwYkAEMABEZAWCEMWaliAwC8CGAywFcD+CAMea2cBEUGmNuieNQKQGkQs0ArBuKrVSoG9YMxVIq1AyQ2HUzkJ+szwRQaYzZbozpAPAYgMsG8PX7nTFmjzFmZTg3AdgIoBRHfs9F4bstwpE3B1FPkr5mANYNxVzS1w1rhmIs6WsGSOy6GcjJeimAnc6fa8K3JSURKQdwEoD3AZQYY/aEf7QXQEmchkWJJaVqBmDdUEykVN2wZigGUqpmgMSrG37BtB+ISB6AJwF8zxjT6P7MHOk74mD5+DAAACAASURBVH6ZRJ/CuiGKDmuGKHqJWDcDOVnfBaDM+fOo8G1JRUTSceRN8LAx5qnwzbXhXqlPeqbq4jU+SigpUTMA64ZiKiXqhjVDMZQSNQMkbt0M5GT9AwAVIjJWRDIAXA1gyQC+fr8TEQFwD4CNxpjbnR8tAXBdOF8H4JmBHhslpKSvGYB1QzGX9HXDmqEYS/qaARK7bgb0BFMRuRDA7wH4AdxrjPm3AXvxASAicwC8BWAtgFD45p/iSE/UYgCjAVQDuMoYcyAug6SEkuw1A7BuKPaSvW5YMxRryV4zQGLXzYBO1omIiIiIqPf4BVMiIiIiIo/iZJ2IiIiIyKM4WSciIiIi8ihO1omIiIiIPIqTdSIiIiIij+JkvZdEpEBEnhCRTSKyUUROi/eYiLxKRLJEZLmIrBGR9SLyy3iPiSgRiEiViKwVkdUisiLe4yHyMhGZFK6VT/7XKCLfi/e4Yo1bN/aSiCwC8JYx5u7woQE5xpiGeI+LyIvCh0/kGmMOh0+MexvATcaY9+I8NCJPE5EqANONMfXxHgtRIhERP46cvDrLGFMd7/HEUlq8B5AIRGQwgLkArgcAY0wHgI54jonIy8yRTwEOh/+YHv4fPxkgIqL+ci6Abck2UQfYBtNbYwHsA3CfiKwSkbtFJDfegyLyMhHxi8hqAHUAXjHGvB/vMRElAAPgZRH5UEQWxnswRAnkagCPxnsQ/YGT9d5JA3AygD8bY04C0Azg1vgOicjbjDFBY8w0AKMAzBSRE+I9JqIEMMcYczKAzwL4tojMjfeAiLwu3J58KYC/xXss/YGT9d6pAVDjfDL4BI5M3omoB+HvdiwDMD/eYyHyOmPMrvA/6wD8HcDM+I6IKCF8FsBKY0xtvAfSHzhZ7wVjzF4AO0VkUvimcwFsiOOQiDxNRIaKSEE4ZwOYB2BTfEdF5G0ikisigz7JAM4HsC6+oyJKCNcgSVtgAH7BNBr/DODh8FLLdgBfi/N4iLxsBIBF4W/n+wAsNsY8G+cxEXldCYC/H9lMCWkAHjHGvBjfIRF5W/g/bOcB+Ea8x9JfuHUjEREREZFHsQ2GiIiIiMijOFknIiIiIvIoTtaJiIiIiDyKk3UiIiIiIo/iZJ2IiIiIyKM4WSciIiIi8ihO1omIiIiIPIqTdSIiIiIij+JknYiIiIjIozhZJyIiIiLyKE7WiYiIiIg8qk+TdRGZLyKbRaRSRG6N1aCIkhnrhig6rBmi6LFukocYY47tgSJ+AFsAzANQA+ADANcYYzbEbnhEyYV1QxQd1gxR9Fg3ySWtD4+dCaDSGLMdAETkMQCXAej2jZAhmSYLuX14SeqrJhysN8YMjfc4UlhUdcOaiT/WTNzxWpOAWDdxx2tNgjlazfRlsl4KYKfz5xoAs472gCzkYpac24eXpL561TxRHe8xpLio6oY1E3+smbjjtSYBsW7ijteaBHO0munLZL1XRGQhgIUAkIWc/n45ooTHmiGKHuuGKDqsmcTRly+Y7gJQ5vx5VPi2CMaYO40x040x09OR2YeXI0oKPdYNa4YoAq81RNHjtSaJ9GWy/gGAChEZKyIZAK4GsCQ2wyJKWqwbouiwZoiix7pJIsfcBmOMCYjIdwC8BMAP4F5jzPqYjYwoCbFuiKLDmiGKHusmufSpZ90Y8zyA52M0FqKUwLohig5rhih6rJvkwRNMiYiIiIg8ipN1IiIiIiKP6vetGxOSiMaMDJt9BYMj7tY8s9zm3bP9NgeGdeqdnANipU3vk1utuXRZk95p1UZ9aCAQ1bCJiIiIUp2/eIjzB2d+Nm6EzQ0TdbvK5lKd97UVh2yWoN5esEmfcthbdTaHqmtsNu3txz7oo+An60REREREHsXJOhERERGRR7ENJsyXlWWzjB9j8965RTY3jot8zPATa21+cOLfbJ6RKehKXbDF5sebTrD5D5POs7ni7uNt9q/ZanOoRR9LRERElNKclmV/QUHEj2qvnGhz01i9PTBS21Ry8hr1B07LckGGtjIX5bTaXDJbW5aXV0yxecKDOpUOrt/cy8FHh5+sExERERF5FCfrREREREQeldJtMJKZabM5scLmys/n2XzjRS/ZvLBgQ8TjMyXd5hD028P/01Jo865OzbOzK/V5C7TF5ZLz1tl8XvD7Nk++o1zHul4fazo7uvp1iDzPP0XrTBp0STFYf8Bmvr+JiKgr7rzNN260zfUziyPuN+U63VnvhpI3bV60b7bNb7yvbceF67WlJrcuaPP+4bqrzIxvvGPznEuX2HzP5kttLq7KtTnU3Hy0XyUq/GSdiIiIiMijOFknIiIiIvKolGuDkXQ95EimjLd561Xa+vLMF263eUK6/iuq7HS+LgxgW+cgmze360b7f16mu7tk79bN+B85c6/ND015wObRadk2vzX/dzZf8tHNNo+s1V1pAnv0eYg8x6fv+bSRwyN+tPVaXaocvEXzsGVal4Gqj/txcERx4NSEv1AP1zNlWh+BwZmIlq9VD87zN7XpD+obbAzW1+vtJvIaRpQI3NYXmazb8m39su4A84V570Q85urC5Tb/quYim2v+qK2YFY+91+Nr5+ZqW8uTZdpCc9c1f7b58BhtoRlWqGNiGwwRERERUQrgZJ2IiIiIyKNSrg3GXzbS5s1fybf5iSt/b/OYNF3SWNehy4bXrboh4rl87+hy5oh3Dts8eYN+CznUqkuTreum2fzzn1xs86Ixr9lc4teWmIYTdIlz+JvOhv9sgyGvcQ+nGKrfnq+8cXTE3X50+TM2P75rus2Nh7SNLKc3bTBOW4GkR/nXWFC/6W8CgaPckejYSZq+L/0lw2yuna+H7s351gc2/37ECptbQt3viJQu+t5f0qy7jf1q44X62i9OsHn4Y/oeDx482KuxE3mJu+vLlq/qvKvyS9qK0mmCEY/5Wd2pNtf9h7bODHqu59YXl3t9yTjukM3D/driklet88RAza6onr+3+Mk6EREREZFHcbJORERERORRKdcGs2+uLrfPPG2TzSX+Tptv3qO7uWz58XE2j15bFfFcpnWH5g5dtgw6y+zut++zd+uyyTubdJnSX/66c389XOn4yTttbh2q7Tsp938aeV7a6FE2b7lR89NX3x5xP3d3pf/37GU2T1y/z+bIxUzly8rSP0waa2PjRG1nM35BT/K3Od/Q/2Btj/cnOiZTJ9m4YWGOzX8/X1sux6br3/c7OvWd/4f6syKeqiOkdTMhu87m8/L0oL7XT7nP5j+N1ZbLx3LOtXnEf77b6+ETecWhE3Q3vN9e9pDNbuvLO23pEY954cHTbS77SFsro258dNrZvjJB29ZK/AP7WTc/WSciIiIi8ihO1omIiIiIPCrlOiqCevYKalv1UKOLVi+wedBd+m3jnOXr9LEtLX16bQnokicCulwfdFpfQtC2mQyfLti09Ly6TzSg/EN0afLA7FKb/+XyJ22e6BxCBkTuZBEa1m7zvtlDbT58je6cUXL6bpuPK9RdkEZmrrJ5bKa20KSL1sy7TXr4xcfOrhlbntPbS3VVk6jP/Mdr68vGb2rry6Jz77J5U4cehHTFi1+wecTr+tlZfmVTxPNKUK8LlbmTbb5v9nybT7pCr1W/Ln3e5g1Xauvnwad0J5pAdY2+QKi75jOi+Mvep23KP1l9hc23hHRiVPR0TsRjyt7W1pdglDvoude2fZdOtPniQc/a/PO9c23Ore3/XcX4yToRERERkUf1OFkXkXtFpE5E1jm3FYnIKyKyNfzPwqM9B1GqYd0QRYc1QxQ91k1q6E0bzP0A/gjgAee2WwEsNcbcJiK3hv98S+yHF3sly2ptbqvS5fYhLbqMkf6R7hLT19YXVzBHWwJyi2P3vORJ9yOJ6uYT/mI98Gj/Z3V5sOJGPQjs83m6/Jguzg4un/Kb056w+aNpZTaPz9QaPTW72ua9wVybf18zz+ZlHTqOHTXaTpP/YabNeXt0mb+sssFmpzGN4u9+JGDN+HJ0+X3HF7Q+FszSw+5ebjzR5r89N8fmyYv1vSg79DCVYFNkG4xxdhUT50Cw0TW6S9jbpbpz2eArX7V5+uAqHUfOrKP8JpSg7kcC1k00MtbpdaDsd7rbmIS0LvybNkY8JtCoB1VG3eY1TOtYrqy3eZQzY37hNT3Ur2Jbz7uZ9VWPn6wbY94EcOBTN18GYFE4LwJweYzHRZTQWDdE0WHNEEWPdZMajvULpiXGmD3hvBdASXd3FJGFABYCQBZyursbUSroVd2wZogsXmuIosdrTZLp824wxhgjIuYoP78TwJ0AkC9F3d5voAS3brc53ckR94nh66WN0eX9nafpMv51E1/q8bGrNpbbPHmfLulw6T7xHa1uvFYz/kJtd2w4T3dSkWt06e/3Zc/ZnOfT9/nRZPn0ILFMZxeXJ2p1efFfd+jBSek7tY2s+COtAn+H/iuqqG3T2yurbA7u1w+eQibu/0rpGHjqWuO0orSfPsXmsWdW2XwokG3zMy+eavOER/S9GNywRZ+zt+9LZ0nfNGs7pb+96y3D0sW5v3BbsVSTSNea7gTr99ssTo64Tx9fwz9UWyjrZhfb/NvJf7X51Rb9b54R7zh1VbMH/e1Yd4OpFZERABD+Z10P9yci1g1RtFgzRNFj3SSZY52sLwFwXThfB+CZ2AyHKKmxboiiw5ohih7rJsn02AYjIo8COAtAsYjUAPgFgNsALBaRBQCqAVzVn4NMZO3jdGmlfZa2snxt8EfOvXS5tNPo0kr+pnSbffWHbGYbjPclet2Ic5hRx9SxNu/9rLauPDXlYZuznMOOXmzRXVh+vUMPbQGAzqDeb3el1kbOTr09v1rf4ZM26vte9u60OVjb8wdFPOYlsXi+ZpwWkrQSfe9u+ZLefkvJGpt//e5nbZ70lO7uEly/uU/DcA9sqb9Yd0KaNLPK5najbWUbm3XHGN9hbZvhdSQ5eL5uPMyXpbuVtZ6sB4aZy7TVZmqGzttOffcrNo/fqPcJNjf31xCtHifrxphruvnRuTEeC1HSYN0QRYc1QxQ91k1q4AmmREREREQe1efdYKgLzk4Bh0u1nWBuuS5/Fvq09SUE/RL2srZ8m3N360KlGYBlFqJP+AoG27z7DF0q/M4pL9o8LVPbXeqD+v5c26aHVlRvHB75vB3aMpBTr58V5OzVGsjbqTu6SPVum4MN2hJDNNAkTdsSm2aOtvlXs5+0+bWDujPM8KV6eTUr7OGSx8RfMc7mfXN0R4ohX9UDyO4fr4eMbejU68vzG4+3edLeDfqkho0wlNpkvLa+7DxP63vR8Y/ZfGfDVJuHPOVsb1mnBzUNBH6yTkRERETkUZysExERERF5FNtgYsXdKaBcD0I6cLzefnXxezYHnL0qaoPtNn/3vRtsnrRSd7xgCwANJMnT5b62Cm1LuSpfdzE6GNT/1q/s1FaZk7OrbH760j90+xp7A4Nsbgjp6/1qw4U2F9w3yebcV9fbHGJbGA0wSdfLZf2JmoenNdj8+nsn2DxpjXP4UXfPmabP43N2ecGQgoj7bb96iM2fv/Rtm783RK8pbmvlC43O0v3r2q4WatNaJkpF7iF/u8/Vuvry+W/YvLNTb7//yXk2lz+ruz0NxA4wLn6yTkRERETkUZysExERERF5FNtgYsQ/SJf0q67WQyi+eYnunjE3Sw+UORzqtPmZJufb+v+uSyuhqpqYj5OoNwI79Jvug5eX2vzvU86zuSSj0eYHXptrc6azy0vbxMhld1+a7vqSkak1cOOUt2x+6eS7bP5a/hdtDu6bYLO8q8uRRAPBdOpBQ8XrNO/q1GV1FGtLY8ewPJsz9jj3cVpfUKS7Lu367DCbM+fti3jtv0z+q83TM/Vgo2zR1pddQb196R49LGnIGq1TrT6iFOLs0NdwgbZWjryiyuZL81fZfO3Kr9k87gHdkSzQGr82Mn6yTkRERETkUZysExERERF5FNtgYqR9eoXNBXP32rywwDmEArrp/poOXSL9r+d094uKGt3xwnRq2wxRvAy/e6XNVY/qri1VkmvzxNa1+gCji+3ubhf/i093SvrvW7QGxnxeWwAuH77a5v86+zKby97txcCJYsgEdU+XvA37bd7SNsLmJXPusPnHpVfa3NKp7S7fKl9mc75Pl9XHpB20+U/7zo547X/+6zdtrrh4q96v/O82/3L3fJuDj2tLDdZ80NWvQ5QyZOpkm5u+qG1hL1c8ZfNt9TNszntaD6cMVG3SJwp1t69T/+Mn60REREREHsXJOhERERGRR3GyTkRERETkUcnbs+5s1ZM2rFhvz87q4s5H4fTfhurqI34UnKp96k0/1D6oh6Y8qC/nbK21qVO39bp509U2T/yTbtEYOMyTGclbIk497KcTENObtH/9QFC/z3FGTqXNd8zQnl5fltYxT2WkAeH0q4aq9e/sl/4wx+ZHps22OXe0XhPOLtM+85aQnii6uFb7ZFet0K1Jxz2l1woAwKkapxd8bLPfOTn7je36+Emv77I5ENBtJolSRdoo3XJ4w7f0u1Z/PfF+mxc16hzub89pHY9/frPNwTj2qbv4yToRERERkUdxsk5ERERE5FEJ2QbjLoFj8jgbd84vsLm9QNtX8qbo8nlZQUPPz++c89YW1H9FmzedEHG/jCG6/P6bic/ZPDbNWaJ3nuuV5in6vK8NtTlQzX3oKLVlHtQ6Wds8yuYzs7fbPCRXT2hEum6DinanZcDwjEbqf8Z5zw19fpvm5XpSaWexLr2vHHqyzcszT7E5Z5+2qEyqOaDP/6nTq5tv0K3n5uTpEv1Pdl1gc8Fr2nIZ3KXbBxMNJF+ubukbnKZtJpVX6bzIZIRsTmvSlmWfHmqNQTs0D3ujVp9zq14T4LSBpY3ULVQBYMPPtA3mV3OftHl5y3ibH3hWt0id8JC2OQf3ay16BT9ZJyIiIiLyKE7WiYiIiIg8ytttMO6OLqN1SWPfWZoPX9xk81cmLrV5TKYuaZyRXW3zCL8uFXb7stCllXajy5TLRudH3G+Qc/rccenuLi5d7zgzOXOPzR2zdNy7bz7d5mErdHk1Y78u+8sO/XZ/sFF3GSBKVG47W9sQrblJObqEvz2gJz9WVWvr2KQ2rSW2vlA8BWvr9A9Odj8Jy3NP8vXrdc1tpwk59zHTj4t4jYuP1xOC3WuNuwPMhDWH9fE8/ZoGkH+o/t3ccK62mZjr9DTq75S9afPyQ+U2F2XoPGdCttZPZauewvvSWdpCnLVe50vZ+/Tv/v2nRO7a8ttzHrO5KaTXmvtfPsvmise1LTq4UXds8qIeP1kXkTIRWSYiG0RkvYjcFL69SEReEZGt4X8W9vRcRKmANUMUPdYNUXRYM6mjN20wAQA/NMYchyO7vX5bRI4DcCuApcaYCgBLw38mItYM0bFg3RBFhzWTInpsgzHG7AGwJ5ybRGQjgFIAlwE4K3y3RQBeB3BLTAdXoksrdedo68vYr2+x+Y4xS2we5MuweUW7LjU+0fgZmw8E9JvKJenaTnJ6ji6BTMvQfy3pos9zfnb3Bxb5oO017g4wrtOydMnl4Rn32PzW8RNtvuc0XeI5fEDHWvzO8TYPW7bb5sAObfEhb4hnzbh8ObobhW9Ikc0h55vuoZYWxEvrOSfanHeaLpfOz9X6fvTQSTYXfqg7wHCZP/l4pW76g3EPJurmkCLJ1mtI5RdyIn72y6L3bG5x2r581foY38e6K403jnGh/hbPmvEXaIti45m6K1/mAm1R/N2ExTZfs+IGm2X1IJsDufp+fqFEt4M5cYLuiPR/Zug8r+0UvQ6816gtN98e9lrE+KZk6GfRdx/S8RnnI+qDJ+jv0HnqaTa3DtO2zHTtWMbw97TVzPeRHtg3ENfRqHrWRaQcwEkA3gdQEn6jAMBeACXdPGYhgIUAkIWcru5ClLRYM0TRY90QRYc1k9x6vRuMiOQBeBLA94wxEd9wNMYYoOuPk40xdxpjphtjpqcjs6u7ECUl1gxR9Fg3RNFhzSS/Xn2yLiLpOPJGeNgY81T45loRGWGM2SMiIwDUdf8Mx6azXP9jcP+Z+q35d8e+bHO70TaV/69uhs2LV+vBE9mV+ib0O6vnhyt0yaXlNG2hmVakh04cTbvRx6/s0G8bP3NQD8BoDOgy5RmD9XlnZ1fZfGHeeptDFfrfT5tb9PdfVjfN5iFr8nQQzsEB5B3xqpkIFWNs/PgC/X5RyXL9ln3G+p02hxoO2WwCzukUvdltxTmcQpzdLiQt8q+Y9rl6sNje63U3pT9NfsrmNqPP9US1tsEMW67XIO7/kpw8UTcDydnxzFekh/pdP+/1iLtNydAL12/qdbm+wLlURexKQykjbjXjtCnvPVX/zn5r0sM2/3DnJTaX3qHtK+nL19gcatb2Yre1ZveVuiPSSzdom8kDY3RXmQWD3cO/uv+Pjfm5G50/aNx4xkibT8jVtpuvDa6yeVmrzre+X7DA5nE7tV4Hog2mN7vBCIB7AGw0xtzu/GgJgOvC+ToAz8R+eESJhzVDFD3WDVF0WDOpozefrM8G8FUAa0Vkdfi2nwK4DcBiEVkAoBrAVf0zRKKEw5ohih7rhig6rJkU0ZvdYN4GnFOCIp0b2+FEahumyxpjS3W5wz206EBIv1n/8l26k8qUl51DUxwdo7QdoL1Inz/Tp8v+ftEFh5DR79VXdmorDgC81KzLNP+98mybRzytLTWZDTq+N8/RXWlKpuvv0xHUpdCWV7VFIf9jfe2K9z62OVCjBySR98SzZlyNk3RJ8dprX7L5LyfOtbngLT1UpaBSl9rTD2qLigT1fSiBkOa2rndkaR+jO8905Ef+FTPqZt116S+jntXXdtb4btml65TB54bYbFYt7/L1KDl4pW4Gkj9fl9gPnqY7np2dF/lBaJZoHT384SybJ613DkLqjwGSp8WzZkJ5On8KDtJrxM6g3v7eSt3pbtKKDfrYblpfMKzYxkCO/lp+0Xf3oVCrzS+1DLe5sk0zAIScfy05Pr1WFadpO+WcfN15rCGoX7D9zX7dqawpqC3OwSynyny9/spnTAzsqxERERERUa9xsk5ERERE5FFR7bM+0ERX3BEMOa0p3Sz4BZxtQuvO0iWRA3N0CeTG6a/bfE2+fiO5xK+7trQbfeGagLa+fPkj/SYwAOQs0m8DT3pDN8gP1u/vcnzlS7u82TlOCRiMyi7v0/UxGkTdy9mtrSyPbp9u8z/O+qPNg8/Rlq1Fjbp7zD8O6WETh5ydjmpb9DCL3bt0yRIdWp+/OudJm6/Ki9yEwG1hawxp/pOzg9L7z+oSZPlTTl2FeNQLJRdTrq0vx39/rc0nZUT+jd8U0jbNrGqtWX9drc28RtCAcuZncHbw8jvzM5Opd/IN0RZkydT3cMN5FTbvv0J3Vfn1yQ/YfF52vc13H5pi86O3fdbmwsUrI4Zn2nXullaqhyI1f0ZrrjNPr1uDKp2WslW6Q59rLP5h80DXGz9ZJyIiIiLyKE7WiYiIiIg8ytNtMBkNuvRXVV/Q5X1K/PrN4+e++/+6vE+Wc2BLjujOK+mij+10dn15uy3X5m+8fqPNU26POBgMZru20QTbI3eKIYo3//v67fvhPxlr8+zrfmTzrZf83ebP5elOLV8ZVGVzyFnvDDpLnKHjTZe3D/Zp20y6U28AsLRV/3zjB9fbPGSJ9rCNfUNfO1C3D0TJRNK1BaC9WJsgv1ysS+yZEnlp/nq17pA0fLleFwMf14AoHnxN2rKS3pBv87QMfe++O/93Nj81R9tXOp3DLE/PedHmijR9bx8I6XXnis3X6GN/qy3Oha9p64s5yhwssFt3B8yq05aaLJ/ODU2ntxvJ+Mk6EREREZFHcbJORERERORRnm6DSd+oBwEVvqyb639nwhyb/1j6ts0jnB1d3F0nuts95oFG/Vbwr968xOYxznkUx23SZfjgzt0RjzedXR8KQ+QF7vsztGWHzRP/pMuXj7x0kc13l6TbXDdDn8dX4hyQ5HO3AOjmdd2dm+qyIn424m2txQkbDuoP9uy0MXDIaTczPOqFkouvotzm6ou05mZmap21f+ptv/bZyTaXb3RqhfVBcRJy5kMj39SdwS6ccanNL03Rg++uy9c2S9d2p/vkmq1fsHn3s7o72aiXdIe9tI836Rh6237s1Emiztv4yToRERERkUdxsk5ERERE5FGeboMJHmiweeir2hKzufYEm2eXTDvm589s1CX9yducpffN2jIQaGsDUaJzl/4C1bqMnr5XDy0qzNTdkQpX6zfuQ7l6u3P2RbfE2VnJ13wgchw79Vv5wcOHnR9wOZ9Sw+EK3dns2/Nettnv7Fr2m/0nRjxm+Pu63B/cG3nQGFE8uLuv5CzfbnP7v5XbPGPMjehJmtPzlbNXd4Mp21Jtc6Bm17EOM2nwk3UiIiIiIo/iZJ2IiIiIyKM83QaDkC6nu8sgmW6O1UvF6HmIEom7lBlxsNeGxi7u3cvndHKw23sRpQ5/8RCbGyboZffawWttPuRchO5/+ayIx0/apte8AA/gI48J1ju7tbymuagPz+ntI4oGHj9ZJyIiIiLyKE7WiYiIiIg8ytttMERERImuSHeAaR2mjWJZ4rf5/fZcm8teiWwCCNVH7qpERKmFn6wTEREREXkUJ+tERERERB7FNhgiIqJ+JO16KFletR5+dMG6L9m8e0+hzcd9VBPx+EBLSz+Ojoi8rsdP1kUkS0SWi8gaEVkvIr8M3z5WRN4XkUoReVxEMvp/uESJgXVDFB3WDFF0WDOpozdtMO0AzjHGTAUwDcB8ETkVwK8B/M4YMwHAQQAL+m+YRAmHdUMUHdYMUXRYMymixzYYY4wBcDj8x/Tw/wyAcwB8soa3CMC/Avhz7IdIlHhYN0TRSeaaCVTvtHnYHZpxh8aJ7v37f0iUBJK5ZihSr75gKiJ+EVkNoA7AKwC2AWgwxnzyd0oNgNL+GSJRYmLdEEWHNUMUHdZMaujVZN0Y8jOkIQAAIABJREFUEzTGTAMwCsBMAJN7+wIislBEVojIik7wmGRKHcdaN6wZSlW81hBFhzWTGqLautEY0wBgGYDTABSIyCdtNKMA7OrmMXcaY6YbY6anI7NPgyVKRNHWDWuGUh2vNUTRYc0kt97sBjNURArCORvAPAAbceRN8fnw3a4D8Ex/DZIo0bBuiKLDmiGKDmsmdfRmn/URABaJiB9HJveLjTHPisgGAI+JyK8ArAJwTz+OkyjRsG6IosOaIYoOayZFyJEvEw/Qi4nsA9AMoH7AXtQbiuGd33mMMWZovAdBvROumWp46z00ELz0+7JmEgyvNZ7AukkgvNZ4Qrc1M6CTdQAQkRXGmOkD+qJxloq/M8VWqr2HUu33pdhLxfdQKv7OFFup9h5KlN83qi+YEhERERHRwOFknYiIiIjIo+IxWb8zDq8Zb6n4O1Nspdp7KNV+X4q9VHwPpeLvTLGVau+hhPh9B7xnnYiIiIiIeodtMEREREREHsXJOhERERGRRw3oZF1E5ovIZhGpFJFbB/K1B4KIlInIMhHZICLrReSm8O1FIvKKiGwN/7Mw3mOlxJDsNQOwbij2kr1uWDMUa8leM0Bi182A9ayHT9jagiPH4dYA+ADANcaYDQMygAEgIiMAjDDGrBSRQQA+BHA5gOsBHDDG3BYugkJjzC1xHColgFSoGYB1Q7GVCnXDmqFYSoWaARK7bgbyk/WZACqNMduNMR0AHgNw2QC+fr8zxuwxxqwM5yYAGwGU4sjvuSh8t0U48uYg6knS1wzAuqGYS/q6Yc1QjCV9zQCJXTcDOVkvBbDT+XNN+LakJCLlAE4C8D6AEmPMnvCP9gIoidOwKLGkVM0ArBuKiZSqG9YMxUBK1QyQeHXDL5j2AxHJA/AkgO8ZYxrdn5kjfUfcL5PoU1g3RNFhzRBFLxHrZiAn67sAlDl/HhW+LamISDqOvAkeNsY8Fb65Ntwr9UnPVF28xkcJJSVqBmDdUEylRN2wZiiGUqJmgMStm4GcrH8AoEJExopIBoCrASwZwNfvdyIiAO4BsNEYc7vzoyUArgvn6wA8M9Bjo4SU9DUDsG4o5pK+blgzFGNJXzNAYtfNgJ5gKiIXAvg9AD+Ae40x/zZgLz4ARGQOgLcArAUQCt/8UxzpiVoMYDSAagBXGWMOxGWQlFCSvWYA1g3FXrLXDWuGYi3ZawZI7LoZ0Mk6ERERERH1Hr9gSkRERETkUZysExERERF5FCfrREREREQexck6EREREZFHcbJORERERORRnKz3goiUicgyEdkgIutF5KZ4j4nI60SkSkTWishqEVkR7/EQJQIRuUlE1oWvNd+L93iIvCxV5mfcurEXwidajTDGrBSRQQA+BHC5MWZDnIdG5FkiUgVgujGmPt5jIUoEInICgMcAzATQAeBFAN80xlTGdWBEHpUq8zN+st4Lxpg9xpiV4dwEYCOA0viOioiIkswUAO8bY1qMMQEAbwD4XJzHRORZqTI/42Q9SiJSDuAkHDnxioi6ZwC8LCIfisjCeA+GKAGsA3CGiAwRkRwAFwIoi/OYiBJCMs/P0uI9gEQiInkAngTwPWNMY7zHQ+Rxc4wxu0RkGIBXRGSTMebNeA+KyKuMMRtF5NcAXgbQDGA1gGB8R0Xkfck+P+Mn670kIuk48kZ42BjzVLzHQ+R1xphd4X/WAfg7jvThEtFRGGPuMcacYoyZC+AggC3xHhORl6XC/IyT9V4QEQFwD4CNxpjb4z0eIq8Tkdzwl30gIrkAzseRJX4iOorwShREZDSO9Ks/Et8REXlXqszPuBtML4jIHABvAVgLIBS++afGmOfjNyoi7xKRcTjyaTpwpN3uEWPMv8VxSEQJQUTeAjAEQCeAHxhjlsZ5SESelSrzM07WiYiIiIg8im0wREREREQexck6EREREZFHcbJORERERORRnKwTEREREXkUJ+tERERERB7FyToRERERkUdxsk5ERERE5FGcrBMREREReRQn60REREREHsXJOhERERGRR3GyTkRERETkUZysExERERF5VJ8m6yIyX0Q2i0iliNwaq0ERJTPWDVF0WDNE0WPdJA8xxhzbA0X8ALYAmAegBsAHAK4xxmyI3fCIkgvrhig6rBmi6LFukktaHx47E0ClMWY7AIjIYwAuA9DtGyFDMk0WcvvwktRXTThYb4wZGu9xpLCo6oY1E3+smbjjtSYBsW7ijteaBHO0munLZL0UwE7nzzUAZn36TiKyEMBCAMhCDmbJuX14SeqrV80T1fEeQ4rrsW5YM97Cmok7XmsSEOsm7nitSTBHq5l+/4KpMeZOY8x0Y8z0dGT298sRJTzWDFH0WDdE0WHNJI6+TNZ3AShz/jwqfBsRdY91QxQd1gxR9Fg3SaQvk/UPAFSIyFgRyQBwNYAlsRkWUdJi3RBFhzVDFD3WTRI55p51Y0xARL4D4CUAfgD3GmPWx2xkceTL1S9ZdJw62eb6f26JuF9mesDmQbfn25y29MN+HB0lsmStG0nTv0r8pSNs3jt/VMT9Dp4Ysnlw2SGb2zrS9blWD7J59EuNNpsV62IzWEooyVozRP2JdZNc+vIFUxhjngfwfIzGQpQSWDdE0WHNEEWPdZM8eIIpEREREZFH9emT9WTiLyy0+fDcCpuDN9bb/MrxD0Y85m9N2iLzWP6FNvNfKiUrt0UMFWNs3Ht6gc2HZrXZfOFxKyMef16BrsJOSq/r8jUWT5xu8yNZZ9pcviL68RIRUWpLG6vXquYpw2xuLe56tpbe4rRrfrjX5kB1jd4pFIzhCHvGT9aJiIiIiDyKk3UiIiIiIo9K6Y4Nf/EQmxvPnGBz01d1B4o3T3jY5hzJinj83Vtn2zy0rr0/hkgUd2mlI20+OGe0zXvO1qXCS6Z/YPP3hy6z+Y3WcRHP9X83abtYc6sewnHt5OU2LyjU/Pqp2pLmnzDW5mDljt7/AkRJyl8w2ObWWVorh8bp7kpFm/Ta5F8W2ZZGlKz8FXrtqblwuM2dZ+j87jMjdnf52NoW3ZFs6wq9/k24T6fMoR16OKzp7OjbYHuBn6wTEREREXkUJ+tERERERB6Vcm0w7uEtbdPKba7/oh549OYpd9ucJRk2v9OmS4sAEHinyOb0Xfot4QCIEpt/iL63az+r36QffLWeVv3QuKdt3hfUQ8H+s+5cm19aenLE8459RutsaIdWyt03aUvZZXNX21yWd9DmXaN1mT+NbTCUoiRdr0ktsyfavP+GZpu/OF7bXR56+mybx2iHGlHiELHRl52tN4/SA/jaywoiHlI9V9ssTz5vo83/Uqrbzk90aqnT6O4uh02nzavH6vP+uO6fbC59pMnmYG3XO5vFEj9ZJyIiIiLyKE7WiYiIiIg8KjXaYHx+jeVlNu+cp0sgD06/0+YhPl1maQzpAS/fWLEg4mnHP11rc6Dq49iMlcgD2k7SnVciWl8mPmLzfQ2n2Hz3G2fZPPINfZ4Jz66KeN5Qm9aTzzmIDId0OX9fUA9e2tWsS5CZNQ02D+xxFETe4R9WbHP154zNr56s17C33F2YjLYQECUKt93LP1wPMmo+UVtfdp+hU9jJp0e2Rt5f9pzNE9L1ulMb1M+o/96s7Z4NznVnfIbO7U7P0naXSVdutrnxTd1hBvv2a+6nw5L4yToRERERkUdxsk5ERERE5FEp0QbjL9Kl9KprdAnluxfrMsmMTF0qDEGXFrcGdAeYkffpt4sBwOzhjhSUnA5M0ff6rPx9Nv9Hne4s8erfZ9g8+fE9NrsHFumxSf9b52fKbR41Ub9NPyFdD63Y06C7zJRtWdfzwImSXPtEXX4fU1Zv8+g0bd8MGv0cTo5WhEQe5SsfZfOOq/Q9/+0v/4/N3xxc3e3jA87VZ1mrtrv8YPUXbM5aqocfGZ/OAcd+cavNc8e/aPMtpS/YfFP5P9s8aJMemBlq1l2ZYomfrBMREREReRQn60REREREHpUSbTCH54y3efTZumyyYPBW5176r+LDdr31mhe/bfOUFdsinjfY3AKiZDTyUf3W+/ZVk2yWkLaIla/TtpTg4eiX/pqHa6tNeV7DUe5JRJ+oulh3yfh1+Wtd3mdbm+6ekb/ddHkfIq9xD63cebm2vvxlwR02T8/ocB7R/RT27TZtTblx6bU2T3hED+NLX7PB5oYLptg8dbDugOaDtsd8JkN3Fgxk6+2S3v9TaX6yTkRERETkUZysExERERF5VNK2wXSeP93m1ht0if2/xj5lc6boTi8fduhG9gtWXW/zlN/rZvfBA59aqu+nze+J4s19r6etbNUfhPQb9sF2p1/MRL/UfuA4/azg2sItNnc6TxUK8UAXooavnmbz/Lkrbb4gR3dR+rBdr2ePr9cDyya/pUv62gBA5G0BPaMoovUlU3Ta2mr09m/tPD/i8R89eoLNk984qD+odA6wzNCWsubhej1aWLjceQ19vZ/VzrE5f7teF0PH0AYaLX6yTkRERETkUT1O1kXkXhGpE5F1zm1FIvKKiGwN/7PwaM9BlGpYN0TRYc0QRY91kxp60wZzP4A/AnjAue1WAEuNMbeJyK3hP98S++Edu11n6JLgD8e/bfOUDP3vkz1B3c3ltp1X2Fz4SJ7NwcpN+qRse6Heux8JWDeW814PtcRu16O08tE2Byv0eWdk60FK9zfMsjnjfT20gpLe/Ujkmukr0ZYvf0FBxI/q5moDyxVFK2z+/9u78/Aoy3N/4N97JjshgUCAkBDCvgiIigjigrgUbC22WqtHK7a21lO7WOupnvac9vT8enrscll72tpWq5W21p2KOyoFxQ1BEJAdWUwgIYGwBAIhM/P8/mB87mdoQmaSycw7M9/PdfXqN8ksT7zeO+/LPPf7PPmi0/i/2KltAP2f1ZWWgjW74jpM8pyHkCZ1Y5zVxoo+1Dx91bU2twZ0RZb9jdorUzlPvw8A5St15b/gbt3YzwRabQ5NGmnzkamHbO7nL7B5b0jbXea/P9HmMfVOi3Sg+xvMOvxk3RjzOoDGE749G8DccJ4L4PI4j4sopbFuiGLDmiGKHesmM3T2BtP+xpiP9xevA9C/vQeKyE0AbgKAPBS09zCiTBBV3bBmiCyea4hix3NNmunyajDGGCMi7S4FYYy5D8B9AFAkJXHfncFdRD84Re/+HXHudpsvLtCVJrKcA/KXDefZ/OG8ETaXv6YbwgTj2Pri79vH5pYJVTYfGKZTmfvG6n+i3h/otGi/1+p0TFudu5nZmpOSTlY33V0zyVTzmQqbPzdG29MKRKcRH1mnKzkNW6AfGOk6NJSJkn2u6W6+/Hyba740JuJn109+zebxOQdt/tCZfX9vU5XNY99xVoBJwBQ9eVdKnWuM/pXv87Ze8xxp6GuzuzFfv4O6Goy8tybipQKtx9AWXw9tnamdpDX349Mf1vdzVplZcqTM5gELnUvmPSdOZnSvzq4Gs1tEygAg/P/1HTyeiFg3RLFizRDFjnWTZjp7sf4MgDnhPAfA/PgMhyitsW6IYsOaIYod6ybNdNgGIyKPAJgOoK+I1AD4IYC7ADwuIjcC2AHgqu4c5D/x6V2//v79bN78VW0bebjyOZsrsnSq4+nDepf9/MWTbR4135k23KN3+UZLsrWVRU4ZbvOe04tt3q83HiNv1AGbz61Ya/MtpYtsvufsi2x+9YxTbB76VInN2a+tstlwutMzPFk3HtDzEzq1eWPJWza/c3Swzf4NOk1pNmhtUHrLyJpxzmW+fjrVX/XprREPu6HXUpsLnE1hbts50+Y+7+gKaME6fpCaKdKqbpzN9YJbdIWwHCe3+9ST/MyXl2dzwFkBpmWyrgBzRQ/dOGlnUK+lHq7T1clK3tHzV+CgPjcROrxYN8Zc086PLozzWIjSBuuGKDasGaLYsW4yA3cwJSIiIiLyqC6vBpMMvh66osvhCeU2P3fur2wema3THtsDuvnKHctvtLnqOb3jN7DdWWElSu5KNK3njrd52+X6/S9N17aWb5SstDlP2v5PnwXdzOL3FUtsPlK+0OaJR261edRaXWEmULc76rETdRe3LjB+VMTPrh+s9VDptKfdVD3N5j4f6ApHbO2idOYvLrJ5/+SBNn+r7JGIx5X4tV3mraO6Udi7i3XVmBGvOK2cLS1xHSdRKjPjtDV562f0Guu28c/b3GL0XPPWkUE2r31Nnzt092p90QSvxMdP1omIiIiIPIoX60REREREHpWabTDO1OHuyXoHfB+/3g/sF/13yJ/2TbW5ZIFOvfsXvd21cYwcanPt13Ta8elJv7X5FGeVmBajU5kbW3UK5ZmDE22+tni5ze4qNoU+bevJrWqy+dgonTr1sQ2GkqWdFZrW/WvkrnhT8nWVi/datHZr3tZ2tuFv6mPYBEPpxl05LDRMj/t9V+nqEjPyd0U8p1D0XPC/2y61uextPY90ppWTKF1lDdBNW7dcrq1jt12iKwXeXLzD5g3ONdnPN11s87DffGhzsFlbqhONn6wTEREREXkUL9aJiIiIiDwqddpgnGn2UIlOafSc3GBzNnRTpKAJ2Tx/m67U0q9GV4CJleTmRny96Uu6OdFPJujd+27ry8HQUZuXHNVNL+54/wqbq/5Hp1923a+bNv1n/1dt7ufXdoJhfXXTpj1lVTbrfxWi6LjHtC9fW63grD7RrqAet+JsOrF3eqXNj13864injMrW1x33mrMy04IjNnNVI0o7oucm/wBtE6s+T1s6V011a8WpRQCHjLZZNryqrTOV726xObFrUxB5jC/ynFVz9TCbvzT7FZu/XNx2m+WTB86w2bygq+wFd2+K4yA7j5+sExERERF5FC/WiYiIiIg8KmXaYNyNkA6O1laRBRPutrnIWTHFbT85urnY5twabZuJZtrQ50zvN1x7WsTP/uNTT9k8u8cem1uMrkrz/boZNr8993Sbh/z5A30hpxWhNdRx+8EH23UFmFHrD9gcauvBRCdwW18OXKHHtP/6epsn9tENVnyiR1bAWdFoSY2uhhRYpTU5ZeYam0dlR67n0ui0zmRt1JrOqanR14ridyBKJf5+pTY3zNANV0Z/dmNUz79x22U291uprZzB+oa2Hk6UGZzWl6yqQRE/GnGFtq9cW6wbUmZBzzsvHdH80Nvn2DxmnrMCTHxG2mX8ZJ2IiIiIyKN4sU5ERERE5FG8WCciIiIi8qiU6VnHMO1H2nWxdhG5fequBc26vFWf1dpDHtpe3eFbSZb+Z5Eh+r4zv/FGxONm99huc2NQ+3ovXfllmwse1l7eskWbbQ4eOmxzVp/eNvfK1t73bGe5L5e/TnuOZceHbT6GKGK5uD66zGjjzBE2X/3vL9l8doEen80hPcaOOX3qffx63N5SusjmugmFNo/K1vsoCpydFwFg0qNft3n40/tsDu6sO9lvQpTSAkPLbG44v9XmJyvnO4+KrBXXxvkjba5cp7suBpz7o4i8wN9L7xFsumC0zYcG6nnEtPMxcWGtc223yrm/cLMut+jr0cPmlmljbO7/o8glFv9j4Iv6M7/Wltun/o2X5tg85reN+n4Nujy2V/CTdSIiIiIij+LFOhERERGRR6VMG4xv/yGb83f0OskjT06c1gB3AtFdotGM1zaBTd/WqZsHSt6OeK3qoP7n++7Wz9mc9bS2tRQv1KW5QgcO2uwfM9zm9f+qv89/937C5p4+3Qn1P+sn6ms6sz0hp52GyOUr0Om+5jN1mcWLbn/T5i8Xb7B52vIv6pMX6TFcUK8tXg2na/18ddbLNt/WW1to4CyNdaISXdURsmOXzaHWzu8sTORF4uxk3TBRa+K2s16wuTJL28f2BZttnrb0pojXqlys547gbi7XSN7im6DtLpuv03PHJy9cZnN57j50pK5FW2g2H9Kdfj+sm2BzoFZrqXSktg3/tFzrCgD6Oq0vC93Wl+dvsHnkQ002Bzdqqw1CXlmwUfGTdSIiIiIij+LFOhERERGRR6VMG0zIuTu3zzq9s35Da4vNI51pxzPzdNWXveN06r7kPWeXq/U6de8r7WtzzTk9bf7rlF/b7E6rAMCnV1+jX8zrY2O/f+hujCakzTZm0libq8/V6c+7LnnY5nE5OtYnDw2w+e/zdXetIW/r1E8w6L3pGvIGKdNpxNqztdR/WPq+zWuP6fGW/ay2Y/Wb7+yseExXr2juP04fL7Efe3tnHbG5cKe2guW8s97mUHMziFLdsfPH29xykbaxfL5onc2tRtsvdwS05bLHC3oOAgDfNq3HIFvGyAPkjFNs3vwverz+4JPayvv5nrU2L2jWFpff10y3uSRX/95fVfquzf/RT1ff21+lrZgrWnQH99E5u20+8frMBz231Qd0fNmH2lllr0TPf8ZpLzbuikvO9ZyvUFelCQ3VMfnrdTW0QLVeC3ZVh5+si8ggEVkkIutEZK2IfCv8/RIReUVENof/v3dHr0WUCVgzRLFj3RDFhjWTOaJpgwkA+I4xZiyAKQBuEZGxAO4EsNAYMwLAwvDXRMSaIeoM1g1RbFgzGaLDNhhjTC2A2nBuEpH1AMoBzAYwPfywuQAWA7ijW0YJIHTkqM2FW/UO3m9u+bzN80Y/ZnNVlt79O/2C1Ta/3qp3FQ9+SacxWp3pjabT9L3O0L1hsOmE6Ucz32l9eXaLft+5k/jQ+bqyTM3FOv1y/hm6LManCvTu/nsadWrpD4tn2DzyWZ1aCW5wNkLiphie45WaaS3Xab2+Z+p0YYvRtpbfNVxgc581uuISeuuU5Z5p/W32XaAbR3yih07n73PqZ12r1tXU3MhWmb9Nvd/ma2u+afPwPZX6oA+clWVivSu/nY3EWCfe55W6iZeGiXryuGHUYpuLnVW+Vh/T4/uLq26wufzdyJUzuOoXtSWZNVN/ZpHN/37p322+slBX+VreopeY337+epv7v6Ovs7tI/2b/eHapzX885S82n5Kt7WKVWfudUTgXaCcxIXenzWdeoC2Xb1XoKmnZ1brxWEGdjkkCTuuLczo6VKGPaRmo59Tyl7TFuUci22BcIlIF4DQASwH0Dx8oAFAHoH87TyPKWKwZotixbohiw5pJb1FfrItIIYCnANxqjDno/swc78Bv86MrEblJRJaLyPJWtLT1EKK0xJohih3rhig2rJn0F9VqMCKSjeMHwsPGmHnhb+8WkTJjTK2IlAGob+u5xpj7ANwHAEVS0vm5aGc6XHZq28i+J3VFifm36kovswt1NZhflL+qj/mc3lX/X5Wftjl3m06zXDhqlb4t9C7kza06RQMAeY36s8DIcpv3jtO7kpsv0taC35/2qM1n52krz4vNumrHg09fZPPIeVpzsl5bX4wHF+ynSF6omUCeri4xspfWTND5u729SVu56ifpHfPN5TqVd8Z03TjpP8uft7nF6OvfWv0Jm1ft1jvjF03SthcAmJyrLTKzZiy3ecGxSTYPqDzD5oJqrROp0xWhUKTjax2gLTvBfB1TyK/TlP5jWqu5H2mLQXCzsxEGJZ0X6qYr/L30WGwu12PurAL9+33UBGx+5qAe96W/1tZNs0HPQQBgAgEQtSVZNXNkgP59vbG4zub6oLaEPL3/LJtH3ets5uW0JTaN1+uqxqPaItZq9LPkkHPO2hfSNuUlR3RlwCVN2sYCAGf31HbK8iz9m/8D5xzWv1Lfo9FZWW/+IV31rPpoic0Nx/S8M7KH/iddvm+wzXWvDUF3iGY1GAHwAID1xpi7nR89A2BOOM8BMD/+wyNKPawZotixbohiw5rJHNF8sj4NwBcArBGRjxdo/h6AuwA8LiI3AtgB4KruGSJRymHNEMWOdUMUG9ZMhohmNZg3ALSzxAIujO9wohPcqytSlD33kc3/dcoVNh+Y8ZLN1zibUFzbU6cuZlzwfzbXnKutKyV+nWbxQdtjxufoVA8A1F6uq8MUF+tmL7eM0PeeUaCrxOwO6nvctedMmx9+d4rNYx7V3y20ZbvmFvaTpQqv1Exuox4zS7YOszlvkJb9LZX/sLnxmzrFN6NA20N6ik7ArWvVevjupitt9v1BpzJ7OyvD/HWkrm4EAF8s1jvxf1m21Oa11y2x+Y8zz7X52RUTbe61WpcKPlyu79FnotZ0eaGumlScrXW8bp/eX7X/ac192QbjGV6pm5j5tPWqeapOxVeO0w1hJuXouaLWmW5fWDvK5h4L37OZaxdRNJJaM1EcpK1Oq+SxgdoitmuankcmfeoDmx8Y+KLNA/363LXH9M0eajzf5ucWaRvZwCXadgYAz5+t11ihgXouGFCq54gppdttHlugq9iEnBacHc3aBrNtv+a33tBzW8U/tE2tZJm238SzYTmm1WCIiIiIiChxeLFORERERORRUa0G4znOaiiBGl3sfuRterfx/d/4pM01X9Dp82/0ecPmns70pbv5kQ/aruJyN1oCgA8v/JPN+4LNNu8P6XTM4wdPs/kPr+smR8P/pi0Ko9/TDZKCR3W6hqgr/DVaDz2W6uYPjefosXd2nnOHPjSHnJnVPx4Yb/PvFutqRcMe06l93xJtafEVaJ38qerSiDHlfVVXCvhUoa7MNMBZueUH/RdrnqUZs9ChJqcFp9X5HW45fLXNhw6yyYDiJ6u/toB99CX9+z1vxCM2+yXb5oXN2irTsExbsnqALVmUOsRZoOhASNuA3U2/vtF3sc2P/Z+u7DW76H2bh2TpdVir8/nxAmeVvB9+cJm+/iParjnsCWd3pRMMbeeWWn9fXQFtXcVom1f11pZLn7N6WHatbsLUv1E3F+x70Nmc0rkm7a61+vjJOhERERGRR/FinYiIiIjIo1KzDaYdplWn5cv/oJtKrFqi0/gzZ5xtc+XM7TbfM/QJm4dltd0Gc6LmkL7fJatusDn7r3rHcO83a2wefWiTzaEmnRIKccML6gbBhj02l7+sGx6dN/Df9DFl2hIjPm0PCR3UafvyhdpOMnqx3ukeOqCbdrmNJaFmbQkb+IcVEWN6+s3Ty+V7AAAgAElEQVQLbL73vNk2N43UGvAVtiIWoVb9zKHiWf2Tlteg9Zm9W8faa6e2nUWuH0AUu/pZugnKFSO0zXJottbQ/MN9bf7pAt2Mb+Rdq23msUipZOBb2vI19bSv2Lxm6p9trnSupW4t0b+72aKtMj6nXfGSNdquGPiLtsEMfl2vo4L1usJeZxoa3dUE0agtLlm+thfVCTitlTBOlZrEtlPyk3UiIiIiIo/ixToRERERkUelVRuMK3T4sM2yWttPBlf30sc8r6vEfL34FptNO9MhJ1O655B+UasbvwQOOt9P4hQKZR7jtFeFturmYSN/o3fum1ydqofocS8Bvafd7NOpwqDT+hLNMRw6YXUjWaNtNBXVukmGFOh0qXFWB4iKO476vfrtY9oG47aamWB33a9PmehwudZNZa4efz7ns7BNR8tsLtil33fPU0SpJHu5/i0v+52ucDSy+ms251R0fHy31OrqYRUL9W95yRt63RbYpxsZuSuvdIp7vjDOec7jfWj8ZJ2IiIiIyKN4sU5ERERE5FFp2wbjMi264kWgThe1h5PdxpfYm2C6byF8oniIqIHqmpM8MnHjCO6uT9o4iOIlr0Gn1fe06qpLjzTphkcPvHWezSOWsPWFUp+7ol3Ou9qyMqpWW76CRXkdvo6/SdsssUvPCcF9+7o4wvTCT9aJiIiIiDyKF+tERERERB6VEW0wRERE3WHAEt1k5UmZYbNxVlcavlo3CvOv1k1dPL4ABVFU3JYYrGtq/4FtYAtxdPjJOhERERGRR/FinYiIiIjIo9gGQ0RE1EnBtRttLl3b8ePZ+kJEseIn60REREREHsWLdSIiIiIij+LFOhERERGRR/FinYiIiIjIozq8WBeRPBF5V0RWichaEflR+PtDRGSpiGwRkcdEJKf7h0uUGlg3RLFhzRDFhjWTOaL5ZL0FwAxjzKkAJgKYKSJTAPwUwC+NMcMB7ANwY/cNkyjlsG6IYsOaIYoNayZDdHixbo47FP4yO/w/A2AGgCfD358L4PJuGSFRCmLdEMWGNUMUG9ZM5oiqZ11E/CLyPoB6AK8A+BDAfmNMIPyQGgDl7Tz3JhFZLiLLW9ESjzETpYTO1g1rhjIVzzVEsWHNZIaoLtaNMUFjzEQAFQAmAxgd7RsYY+4zxkwyxkzKRm4nh0mUejpbN6wZylQ81xDFhjWTGWJaDcYYsx/AIgBTAfQSkY93QK0AsDPOYyNKC6wbotiwZohiw5pJb9GsBlMqIr3COR/AxQDW4/hBcWX4YXMAzO+uQRKlGtYNUWxYM0SxYc1kjqyOH4IyAHNFxI/jF/ePG2OeE5F1AB4VkR8DWAnggY5eaOQZQ/HK8ie6NGDqGhFJ9hAyRVzqhjWTfKyZhOG5Jo2wbhKCNZNGTlYzYoxJ5EAaABwGsCdhb+oNfeGd33mwMaY02YOg6IRrZge8dQwlgpd+X9ZMiuG5xhNYNymE5xpPaLdmEnqxDgAistwYMymhb5pkmfg7U3xl2jGUab8vxV8mHkOZ+DtTfGXaMZQqv29MN5gSEREREVHi8GKdiIiIiMijknGxfl8S3jPZMvF3pvjKtGMo035fir9MPIYy8Xem+Mq0Yyglft+E96wTEREREVF02AZDRERERORRCb1YF5GZIrJRRLaIyJ2JfO9EEJFBIrJIRNaJyFoR+Vb4+yUi8oqIbA7/f+9kj5VSQ7rXDMC6ofhL97phzVC8pXvNAKldNwlrgwkv2r8Jx3fYqgGwDMA1xph1CRlAAohIGYAyY8wKEekJ4D0AlwO4AUCjMeaucBH0NsbckcShUgrIhJoBWDcUX5lQN6wZiqdMqBkgtesmkZ+sTwawxRiz1RhzDMCjAGYn8P27nTGm1hizIpybcHzb33Ic/z3nhh82F8cPDqKOpH3NAKwbiru0rxvWDMVZ2tcMkNp1k8iL9XIA1c7XNeHvpSURqQJwGoClAPobY2rDP6oD0D9Jw6LUklE1A7BuKC4yqm5YMxQHGVUzQOrVDW8w7QYiUgjgKQC3GmMOuj8zx/uOuAQP0QlYN0SxYc0QxS4V6yaRF+s7AQxyvq4Ify+tiEg2jh8EDxtj5oW/vTvcK/Vxz1R9ssZHKSUjagZg3VBcZUTdsGYojjKiZoDUrZtEXqwvAzBCRIaISA6AqwE8k8D373YiIgAeALDeGHO386NnAMwJ5zkA5id6bJSS0r5mANYNxV3a1w1rhuIs7WsGSO26SeimSCJyKYB7APgBPGiM+Z+EvXkCiMg5AJYAWAMgFP7293C8J+pxAJUAdgC4yhjTmJRBUkpJ95oBWDcUf+leN6wZird0rxkgteuGO5gSEREREXkUbzAlIiIiIvIoXqwTEREREXkUL9aJiIiIiDyKF+tERERERB7Fi3UiIiIiIo/ixXqURKSXiDwpIhtEZL2ITE32mIi8SkQGicgiEVknImtF5FvJHhOR14nIgyJSLyIfJHssRKkgU841XLoxSiIyF8ASY8wfw5sGFBhj9id7XEReFN4FrswYs0JEegJ4D8Dlxph1SR4akWeJyHkADgH4szFmXLLHQ+R1mXKu4SfrURCRYgDn4fjOVzDGHOOFOlH7jDG1xpgV4dwEYD2A8uSOisjbjDGvA/DUZixEXpYp5xperEdnCIAGAH8SkZUi8kcR6ZHsQRGlAhGpAnAaju8SR0REFHfpfK7hxXp0sgCcDuB3xpjTABwGcGdyh0TkfSJSCOApALcaYw4mezxERJR+0v1cw4v16NQAqDHGfPyvtSdx/OKdiNohItk4/sfzYWPMvGSPh4iI0k8mnGt4sR4FY0wdgGoRGRX+1oUA0urmBaJ4EhHB8Xs81htj7k72eIiIKP1kyrmGq8FESUQmAvgjgBwAWwF80RizL7mjIvImETkHwBIAawCEwt/+njHmheSNisjbROQRANMB9AWwG8APjTEPJHVQRB6WKecaXqwTEREREXkU22CIiIiIiDyKF+tERERERB7Fi3UiIiIiIo/ixToRERERkUfxYp2IiIiIyKN4sU5ERERE5FG8WCciIiIi8iherBMREREReRQv1omIiIiIPIoX60REREREHsWLdSIiIiIij+LFOhERERGRR3XpYl1EZorIRhHZIiJ3xmtQROmMdUMUG9YMUexYN+lDjDGde6KIH8AmABcDqAGwDMA1xph18RseUXph3RDFhjVDFDvWTXrJ6sJzJwPYYozZCgAi8iiA2QDaPRByJNfkoUcX3pK6qgn79hhjSpM9jgwWU92wZpKPNZN0PNekINZN0vFck2JOVjNduVgvB1DtfF0D4KwTHyQiNwG4CQDyUICz5MIuvCV11avmyR3JHkOG67BuWDPewppJOp5rUhDrJul4rkkxJ6uZbr/B1BhznzFmkjFmUjZyu/vtiFIea4YodqwbotiwZlJHVy7WdwIY5HxdEf4eEbWPdUMUG9YMUexYN2mkKxfrywCMEJEhIpID4GoAz8RnWERpi3VDFBvWDFHsWDdppNM968aYgIh8HcACAH4ADxpj1sZtZERpiHVDFBvWDFHsWDfppSs3mMIY8wKAF+I0FqKMwLohig1rhih2rJv0wR1MiYiIiIg8ihfrREREREQexYt1IiIiIiKP4sU6EREREZFH8WKdiIiIiMijurQaDBEREf0zydLTq39Af5uPVZXa3NI3R3OR3+amwRLxWqadj9Vy9msuqA/ZXLirRR+zqdbmwO4GfUIoeJLRE3mIT2sjq19fmw9Mq7K5cazffQaM+6XR6NZMUbXWQM+N+oPguk3Oc50nJxE/WSciIiIi8iherBMREREReRTbYMJ8eXk2S0WZzUeG9bH58IDsyOcEdHqk9ypnbmXrRzaGmpvjOUwiT5LcXJt9wwbb3DS6d8TjjhXq5wPu1H72YWcKv/qIzf7NNTYH9zbGZaxE3SWrotzmI2P1PFJ/utPuMkHPCVX9tUVlakm1zf/db1nE6+ZK5LnnYy836/cf23OWzUu2D9XnrtA88PV+Nst7G2w2rcfafH0iL/AXFdrcNEXPL32/ud3ml4c9G/GcAl8O2vJSs56r7t15gc0fLtA6GbRJ68ortcFP1omIiIiIPIoX60REREREHpXZbTCid9xLpU5ffvRZvXP/9Nkf2HzvoJcjnr4rqHcSz5x/m82VC06xObfhqM2+loDm+n02Bxs1mxa9i5/Ic5y78n09Cmw2o3RqcsMXdMry+5c8HfH0zxZutbnIp61nLx3R17rtvatsLn1ihD7+5fU2Bw8ejHnoRN1t98xKm4uv3mnz3GFP2jwhR2uoxbTavD+k54eGf1qoRc8LjSE9bVdlH7b5xwNfsjmvXD+H23GWvt8VE262eeTPdNofG7fZyHMQeU6ptiPXTdbjedHwF50HRV7ObmrV2vA7y8FMzdM2y4uHv2Dz532X2Nz8dz2fBddv7tyY44yfrBMREREReRQv1omIiIiIPCrj2mDcjSp8xUU27z9dN6qY9tmVNt9b/qbNIUTekT8kS79+afbdNm+9tERfN6jT+8/tPdXmFc+Ptbny5SYd3/sbbfbKXchEH/MX9rD56GRtUdl2tbaUbZl1r80hRG4o4dbQEaPH9yX5+rg3z/6dzTeVXW5zY/Mom3NfXK4v6pFNK4iCl2lL45OjHrU5V/S8UxPQafhlLdp++XTD6VG9x7IdOkXfs1Bf64z+unLSlKIPbb6haJfNb5z3a5vP3X27zaP+T1eJCezQVWmIPMFpWW5vg7AT3bTxWufpeo64snyFzZ8t1NbK6wa8Y/NtN19n88jvOC1iAW1VSzR+sk5ERERE5FG8WCciIiIi8qiMa4PBqTqVvuUKbYP5zKy3bf5e6dvOE3IRjcFZOU4+5PxE86weeufyti8vsPlnl860ef9XqmwObd5uM1tiyAvMEJ223z5b/3ws/cTdzqPy233+0hZtg1l3tMrm0/O323xGjj7/uxW6wsX113/R5mGvaXtZxMZjbImhJGrapJuAPTtKV4ZpCukx/YvXZ9k85Eld9iVvpU63n8yIVl1RCT5tD9hZXmHzT6+bYPMN1/9W30P087msQbpahsmL7jxHlAyt/Xra3Gvc3qie0/jSQJsrnttt87PF023+xc2fsHnJxffYfPV5b9m8/MzTbPYtW2tzolti+Mk6EREREZFH8WKdiIiIiMijMqINxnfqGJs3fFWnz38x/WGbz87TO+YLpP1p/Hbfw/l3jw/S5mMKnCnIMTkhm/9fxbM23/zra2yWH+iKMb7letcyN62gZDE+PYZNtrac9HY2ODpk9Pi8fP3VEc8//LBOTfas1tauP3xTp+SXT/qbzeNydNOYO07VTcnu+dfP2lxx7/s2R7TEECXYqHv1PPLA63qMSkhrZew6nZIP1e+xOXhYa6AzfFXaotba+592Vfrnx/vYMkapwX9EzwN79xae5JHOc5zOYTmo7ci+xv0259ZoW/SuoLaCjc7XOl5cebbNRSt0QyZ4rQ1GRB4UkXoR+cD5XomIvCIim8P/3/tkr0GUaVg3RLFhzRDFjnWTGaJpg3kIwMwTvncngIXGmBEAFoa/JiL1EFg3RLF4CKwZolg9BNZN2uuwDcYY87qIVJ3w7dkApofzXACLAdwRx3F1nbOI/sabdNWXX13wF5svytfpkOxOtL7sDuqGFD+rv8DmkLNq/5Uly2yelqdTOW7bTEWWTr/8fvgjNn/6vO/aXLWzv82B7R/FPFZKrJStmw4cHKV35U+fqHfGu61fLUZbvGrf1rYXABj2eq3NwZ2aD3xhnM1+p10sH7rK0uxC3ejlwA26SswDU6faPPh72oIT3LjlZL8KeUw61Exg2w6bC5wWl4jHdLHdJeK1Zpxh89ardIr+9vNebOvhaHJqM3exnhfRuLuNR1MqSIe66YivWXtafA09T/LIjplmvW4b9A/Nn6/4V5sfm6Eb8x24Sjet7P3uAJuD1Tv1NRPQEtPZG0z7G2M+PtPWAeh/sgcTEQDWDVGsWDNEsWPdpJkurwZjjDEA2r1TRURuEpHlIrK8Fbwxkgg4ed2wZoj+Gc81RLHjuSY9dHY1mN0iUmaMqRWRMgD17T3QGHMfgPsAoEhK4n/7udPu4svVdpL9n5lo8+fOfsfmWQU6pQE4d/a244jR6Zfv1Z0b8bNn39XF8vsu09cS57d8edTpNo+cut3me4c+YXOZX1twqrJ0tZqcKY02t75eoq/PNphUFVXddHvNdEFLsdbbWUW6OUvIORc0OStfFFZHPt8c1PrzVQ3Sx/XSVVyCzlS9+7rFzoozN/faYHPBGD3JPF00o+NfglKJd841MQrFsd3FPc8Fp+t5Z+t1+v3/nabnlE/30LaWZS16brrm5dtsHrOwQce6/0D8xkpekPLnmggBXd3Id6zt1fZOxvTsYXPLBD3v7DxfN+k7a/RGm4dm6XXfD8c/Z/P95Z+xOatWa8zLbTDPAJgTznMAzI/PcIjSGuuGKDasGaLYsW7STDRLNz4C4G0Ao0SkRkRuBHAXgItFZDOAi8JfE1EY64YoNqwZotixbjJDNKvBXNPOjy6M81g6RXJ0tYjDMyfYnH2DTlFc02up84y2f+UWoyu1rDim0+23O5u6HFtQGvGcMa8404hb225NKR2odw9X7x5i8z9uHmrztT1r0ZY5w7V9528jZtncd43exR88eLDN51Jyeb1uYuHvVWzz0RKdghyaozOrbv1saO1rc6/NkX2Q7p34h88Zpq/Vewc64r7H0had1vzZq5+yecyeOpsTu2UFdVU61Uyn+LRdxT9scMSPGif3s3nvJ4/a/NNJT9k8u4euPrPKaRX46urrbB7za213CW3eZnMipvGpe2R83bTjwAhtp2zprddhZoK2Yn5//NM2X9tTz2f79KlY3Vxps79F23GO3wqQOF2+wZSIiIiIiLoHL9aJiIiIiDyqs6vBJJVk6bD9Zbp8aP2/6BT746N0c6Ex2XrHb3vWteoU5LfXXmVzyc90uj171QcRzwk2OSvLtDMl4m6SUfp+H5sf3DHN5mvHPdnmcy8r1Pf7/VCnDaafvg7YBkPdrVSPt+ZBOl1+Zq5Oqbc6h//mFp1yzPnghPYWp3b3jtU8s/f2Nt+62VmN6bUjOo5b39b2tNH/p9OX7kYVRMkk2dqi6cvX1kq4uajQxqNDdMWv7edGnrPOu2S1zXcOWGBzZZauJPZei3729k2nfbPoz07b5Fq3JZQofX3uPG0jPq1Az0Pn5+sSZcU+rdHFR7WWfrD5Spv3vanns6ptumJM8JiemxKBn6wTEREREXkUL9aJiIiIiDwqJdtgfIXamrLvrIE23zlhns2D/CG0xV1RYldQ7+y9q/pym3P+ptORviU6laKP7pzcj3STo11LdNwrR+pYT8vRfz+5U5zHqnRVjUCpTmvKli4OiqgjOc6UfK4eq4U+3YTsUKid3e8k8vOAY6cPt3nghTodeX2v95xH6cZgG1v1T9T3P9AaHfN9bX0J7HRWUwp1tUqJusDZvEjG6mpH+0frikpH+mhNHK7Q/rHR03R1lnmD/x7xssOztQ6ynPo4ZLTu/mu7Tt3nPKjnsIJ5bH2hFOfUlel4L0sAwE/6rbDZ3dxyY6u2vjxycLTNv35bF88Z+6NdNhft0dcJHtWVmBKNn6wTEREREXkUL9aJiIiIiDwqddpgnGkQ9NNNV+o+qdMbn+qh04hFPm0hcbmrvnx3s94xf/jRMptLHnm7S0NtT2DrdpurHtN/J1079is2Lz/vdzYXirYZ3D9trs23rfiqzQO6Z6hEMfE59Zkn2mqGfiURj9v6OW2puadyoc0VWboqxr5gs833119sc8E8bSUIVK/r2oCJuoEvX887G2/UdsV5l/3K5vE5Ha9OBuR2/BAAuwLaRrNpzSCbR85fbnNit24hig/J1RoI9NHzQ6Bf7KuwuBtd3vjODTZX/FlrceRLy/Q9Yn6H7sdP1omIiIiIPIoX60REREREHpUybTC+Qp0GOThe22DumfoXmwtEpzRC7Uz+/W73DJsPPaGtL30f1jt+Ez1t6L5fq9HVNkKiPwkaX9tPIOpmckRXnJDmXjbvC+md8e7mEucXbLb5md+cGvFaj1b91eZx2XogHwrpcX/LR5+0ecv9erd+yePJq1GiaBy+ZJzNYyd8ZPMpOd1zqh2UpeeF3kN1tbHQ5FNslrdWdct7E8WdT9uUmz490ea8r+qKX/OHP+48IQfRuH3952we+Ig+J3fh+zZ7/ZzCT9aJiIiIiDyKF+tERERERB7Fi3UiIiIiIo9KmZ51qRhg8+6z9N8Y5+fvtTlb2u5f+k7tFJtXPDre5ooXd9gcaGlnB8Z4cvqxTL6OtUeBvrcfgrZUt/bRx7R4vbuK0knQ2SG05xbdefcfzRU2X1G4x+ah2XrvyG+Huv2FQH+/HvfZovVwxZZP21z70BCbS5/dqONIRI0SdUHhBw02r92s9fHmIK2J+mBPm3/54UU219X2trnkncjlHX3OWnJTv6bLMv687C2bS3sctvloL13mNLpFIImSQ7L1nLD3C2fY3P/67Tb/asgTNldkxX5Etwb0XFPYovdHmdbYl4FMFn6yTkRERETkUbxYJyIiIiLyqJRpgzHZOo0RKAraXNBO64vrlefOtHno0zX6Orvq4jS66Ph6FNh8aJhOU/5knO5OWuBre3e7n62+xObK9UfbfAxRdzBO+0mP3TqFuGCfLlN3ReFim7OgtVqR1fZOwgDwmc2fsrnur1U2939JdyIO7G0EUaoIVe+yecRD+jf+397WXad9zga/Rdv1b/moQ0f0MdVaAwAQHKrtZ/Ut2kbjcz5vu2zAapt/c+ZgmytfiHr4RAnXep62Jh+97IDNP696ymb3U+Uf1Ov13N836NLAb51zb8Tr9m5nF/tUxU/WiYiIiIg8ihfrREREREQelTJtMO3xtbN6iqvndl09JbD9o5M8snv5+ujd/o1jtFVgZn6zzX5nF9aHm3QFmB6Le9icvXaTzdoQRNQ9fHl5Nhun3Pa1aFuXW4d+0c8AjoQiW7b+5cPLbY5sfdG6DNQmtj2NKF7cljHf0g9s7rvSadd0dusNHdX6MM5qYRhaGfG6m+dord3d790239sPZ5ULfgxHHubrodcz1Rdqbdw++kWb3d15f75XW1+ee+Jsm/tv0CugA2dHrpLX26mB8yu22PxOv0k2FyN1dFjSIjJIRBaJyDoRWSsi3wp/v0REXhGRzeH/793RaxFlAtYMUexYN0SxYc1kjmj+/R0A8B1jzFgAUwDcIiJjAdwJYKExZgSAheGviYg1Q9QZrBui2LBmMkSHbTDGmFoAteHcJCLrAZQDmA1gevhhcwEsBnBHt4wyhfl76z9o952ld/QPumhHWw+P8PqBkTYXb9NdMYJ79rb1cPKItKgZZ0q+5ZxTbK6brtOOc0o/QFtajT7mjaORE401j+qGR2XPbbU5ULe782OltJAWdeMwgUCbOYI47WOF2hqw81NlEQ/7rwt1c7FZBftsrg1q2839H06zueyt1NnshTovVWvG16+vzb1O1Q31Lu2hLb5LW/Tc8ZeVurHl6Kfq9YWclrKXDo+JeI/rivS1vtJnic0vjNOWmj6DdOOyQLWuFOhFMfWsi0gVgNMALAXQP3ygAEAdgP7tPOcmADcBQB4K2noIUdpizRDFjnVDFBvWTHqL+jYUESkE8BSAW40xB92fGWMMANPW84wx9xljJhljJmVz42PKIKwZotixbohiw5pJf1F9si4i2Th+IDxsjJkX/vZuESkzxtSKSBmA+vZfofuE2j4GIx/j/JaSpV+0OzXZRe57tI6rsrlulu6G8fyIJ3R8TpEcCunGGMvqdEWAvk3OThrkeV6umWiEpk2wefsXtMb+MO0hmy/I15UsdDISaDF6rC48ODbidfu9o+cRtr7QiZJVN+6KR1KhLSjHKrSNMXuf/m02azdrjvU84ra+9Cmx+eD04TaPunJjxFOmF2y3Ocv5BPSv+8+wOfAPbS3IWfBWbGOilJWK55pQsbZ8De9Va3OZX4/tX+3Rc1DxMr1GCm760GZ/Xz3mf/HWJyLeY8qF+riJOXpNljW8yeZjVaU2+zzeBhPNajAC4AEA640xdzs/egbAnHCeA2B+/IdHlHpYM0SxY90QxYY1kzmi+WR9GoAvAFgjIu+Hv/c9AHcBeFxEbgSwA8BV3TNEopTDmiGKHeuGKDasmQwRzWowbwDt7jx0YXyH0z5p1RUm/Id1QqDZ6F3vBZKDtjSXOdOO/fvZHNi5K55DtHxDtH1lx4X5Nn/7zOdtLvTptI7bNvDbxok2y3M6RZq9dZvN3dO8Q/HilZqJhq9Apx1loN6DtOMbepTNO/OPNg/J1oaXA86d+MU+bSNwFfgjV6Uwuc7GL+4mMCFu75Xpklk3MlhXhfjoM3qOKLlAp+i3rx5g84j/1r/rpkmn1V0RtVWgj0evIhv3TdaaG/Q1ba15sErPFQCQL/paS47qafv+N863edQSbTHruDmU0kEqnWtite2wsynkbuf8YPToDu0/YPPABc75BMCGc7WdbWJOg80iTnWk0OZhKTRUIiIiIqLMwot1IiIiIiKPimmd9aRqaLSxZI3eAbzycr2reGqebhDhc/4dMvwi3Xxl78Yqm3vO0xukO7UyjDON7y/Wqc0P5+jU5o8+94jNVxTq4v9u68vyFp3ifO5/LrC5/4J1Ngec6R6iLnGOWzN2qM0bv66tWb8//c82u60vi47o3fM7W3WljJuKt9ucLfr6E/KrI956cV/duKUgX1tnQocPRz18ong7OF6n3Ed/UjdT+Xnl0zZ/yXetzb4+euxLTra+kLPSy7Hxg23eN0Jra98pOg3/nYu13eXmYnejvMiWzoMhXXnpK099x+YxD+j0fnDjFhClmpDTxeOu7pfn12uy1gK9nvMX6bWW9NBrp10zI1spR+fUOl+lzqVue/jJOhERERGRR/FinYiIiIjIo1JmbiC4V9tg+tZIG44AAAt1SURBVC7T/J11V9r86sSHbC5yVqf4y7C/2zzt6i/bnLf3VJuzFr4X85j8w6tsbv6NTsEsGfMLm3s743BXrnFXfXn+x9r6Uvz8GpuDzc0xj4moI5KtZd94Sk+bfzDlKZvdDY9uqr7E5jcXj7O5tZ+2ct38ifv0+87d+u8drop47/w6PaZDPL7J40r9Wis/Hf6kzb97dIbNLUF9jE+0ZWxWH13a+tx8bXHJc1plCsVpoTnJ6Xjq21+1ueoZ3ZwpuHlbWw8n8jQ5pi0uNU29bN4T1GP7t5Uv2vzInbpS0t2zLrI5cExbLhdM/1XEewzJctosnfaaYy1ac76W1FmFjJ+sExERERF5FC/WiYiIiIg8KmXaYNyF8M02XWGi5GejbL795zpd/8OBL9lc5tcNKf4w8S823/z162zOrZqquUnfa+84nbLsO3l3xJAG9NAVWn5QvsDmPj59vzdb9N9Dd2zU9zv2d914o//LuupLROuL4dYW1L2M88/1Ap+7mpIe9w1HC21u7a3Tl2eO1Cl4d5qx2Vnp6JGVkyPeb8z+vTYHeXyTR+Q26nG9vl5X88odqlPm47J1yvwnA3WKvr2J9F4+Pb26mxq5dgb17/13q2favHTtsIjHDX9Yx5e1UlerCXEzMUpBZutHNuf/eLTNn7j9Kza/eoZuxndNka50dPbUD20OGT1PuW0vQOQ57Dt1eh4qXqTXZ741q/W1oh9+UvCTdSIiIiIij+LFOhERERGRR6VOG4wjdETvGM5epVMiS5+eYPMrc3Ta5PM9t9t8Wo7+++T3p/7V5leH6ioXh4K6gcX4Am25mZavrwMA2TrLghJnyvOBg7rRzF3/uMzmIfN0KrNknbYQcMMjSqiQtp8U7NFp9N2teld+ALqB1+2DtMXrowElNk/I3akv6fy7f1dQ79Dv/a672gWAA4c6OWii7pO7Rqflix/XFpQz/Nq6+KdT59o8IceZSnem211HnNW/fn9AN0j61Wpd/St7lbaYlb6v7WOjd51QJ1t0fNxAjFJd6KiuNuZftt7mkvv1Gu6sPd+y+bapr9j8tV7RrYD0v3vH2vzKU9oGM/h13QwzmEK1xE/WiYiIiIg8ihfrREREREQelZJtMO4qKcFDOo1ROb/B5p8WfcbmF8/VBfVvd1ZtmZxrnKybEbncKc4Q8iN+9pM9421+a4+2vux4o9Lmkc816RNW6nRPIKAtMUSJZILa+lK4ps7mX704y+bSyx6z+YpCbYmZlqcrIq0+prVxxeZP27zl9Sqbh76uq78AQOjgwU6Omqj7BBv03NFroa4Lkbe3yuYvnnqrza09nCe33QUDcRZqya/Xc83gjdoCkLNlu82BWq0trvJCmcK06Cpk+a9vsHlkQ5XNc5deavO9A9spuBMUVjs1946ew0Lbq9t6uOfxk3UiIiIiIo/ixToRERERkUelZhuMy5kuDK7XdpfhD+lUZvWGETZ/4RTNgb56931nFK3Jsblwp45j6MpaHdOW6O5cJkoYp2YC23WViRF/1o1bftR0jc3/XqmrWrj8jbrSS7/3dMpx6CsbbQ427ot8EjdCIo8L7tHWrexXNZe9Gv/3YjMkkQo1OW3D72prcp939dt9OvG66dBUxk/WiYiIiIg8ihfrREREREQelfptMO0IbtLNknq5ORHvnYD3IIq30Gq9E79ydedfh8c/ERFR/HT4ybqI5InIuyKySkTWisiPwt8fIiJLRWSLiDwmIjkdvRZRpmDdEMWGNUMUG9ZM5oimDaYFwAxjzKkAJgKYKSJTAPwUwC+NMcMB7ANwY/cNkyjlsG6IYsOaIYoNayZDdHixbo47FP4yO/w/A2AGgCfD358L4PJuGSFRCmLdEMWGNUMUG9ZM5ojqBlMR8YvI+wDqAbwC4EMA+40xH688VQOgvJ3n3iQiy0VkeSta2noIUVrqbN2wZihT8VxDFBvWTGaI6mLdGBM0xkwEUAFgMoDR0b6BMeY+Y8wkY8ykbOR2cphEqaezdcOaoUzFcw1RbFgzmSGmpRuNMfsBLAIwFUAvEfl4NZkKADvjPDaitMC6IYoNa4YoNqyZ9BbNajClItIrnPMBXAxgPY4fFFeGHzYHwPzuGiRRqmHdEMWGNUMUG9ZM5ohmnfUyAHNFxI/jF/ePG2OeE5F1AB4VkR8DWAnggW4cJ1GqYd0QxYY1QxQb1kyGEGNM4t5MpAHAYQB7Evam3tAX3vmdBxtjSpM9CIpOuGZ2wFvHUCJ46fdlzaQYnms8gXWTQniu8YR2ayahF+sAICLLjTGTEvqmSZaJvzPFV6YdQ5n2+1L8ZeIxlIm/M8VXph1DqfL7xnSDKRERERERJQ4v1omIiIiIPCoZF+v3JeE9ky0Tf2eKr0w7hjLt96X4y8RjKBN/Z4qvTDuGUuL3TXjPOhERERERRYdtMEREREREHsWLdSIiIiIij0roxbqIzBSRjSKyRUTuTOR7J4KIDBKRRSKyTkTWisi3wt8vEZFXRGRz+P97J3uslBrSvWYA1g3FX7rXDWuG4i3dawZI7bpJWM96eIetTTi+HW4NgGUArjHGrEvIABJARMoAlBljVohITwDvAbgcwA0AGo0xd4WLoLcx5o4kDpVSQCbUDMC6ofjKhLphzVA8ZULNAKldN4n8ZH0ygC3GmK3GmGMAHgUwO4Hv3+2MMbXGmBXh3ARgPYByHP8954YfNhfHDw6ijqR9zQCsG4q7tK8b1gzFWdrXDJDadZPIi/VyANXO1zXh76UlEakCcBqApQD6G2Nqwz+qA9A/ScOi1JJRNQOwbiguMqpuWDMUBxlVM0Dq1Q1vMO0GIlII4CkAtxpjDro/M8f7jrheJtEJWDdEsWHNEMUuFesmkRfrOwEMcr6uCH8vrYhINo4fBA8bY+aFv7073Cv1cc9UfbLGRyklI2oGYN1QXGVE3bBmKI4yomaA1K2bRF6sLwMwQkSGiEgOgKsBPJPA9+92IiIAHgCw3hhzt/OjZwDMCec5AOYnemyUktK+ZgDWDcVd2tcNa4biLO1rBkjtuknoDqYicimAewD4ATxojPmfhL15AojIOQCWAFgDIBT+9vdwvCfqcQCVAHYAuMoY05iUQVJKSfeaAVg3FH/pXjesGYq3dK8ZILXrJqEX60REREREFD3eYEpERERE5FG8WCciIiIi8iherBMREREReRQv1omIiIiIPIoX60REREREHsWL9SiJyEwR2SgiW0TkzmSPh8jrROTbIrJWRD4QkUdEJC/ZYyLyMhHJE5F3RWRVuHZ+lOwxEXldJpxreLEeBRHxA/gtgFkAxgK4RkTGJndURN4lIuUAvglgkjFmHI6v3Xt1ckdF5HktAGYYY04FMBHATBGZkuQxEXlWppxreLEenckAthhjthpjjgF4FMDsJI+JyOuyAOSLSBaAAgC7kjweIk8zxx0Kf5kd/h83QyE6ubQ/1/BiPTrlAKqdr2vC3yOiNhhjdgL4BYCPANQCOGCMeTm5oyLyPhHxi8j7AOoBvGKMWZrsMRF5Vaaca3ixTkRxJyK9cXz2aQiAgQB6iMh1yR0VkfcZY4LGmIkAKgBMFpFxyR4TkVdlyrmGF+vR2QlgkPN1Rfh7RNS2iwBsM8Y0GGNaAcwDcHaSx0SUMowx+wEsAjAz2WMh8rCMONfwYj06ywCMEJEhIpKD4zcvPJPkMRF52UcApohIgYgIgAsBrE/ymIg8TURKRaRXOOcDuBjAhuSOisjTMuJck5XsAaQCY0xARL4OYAGO32n8oDFmbZKHReRZxpilIvIkgBUAAgBWArgvuaMi8rwyAHPDK5D5ADxujHkuyWMi8qxMOdeIMbzRnIiIiIjIi9gGQ0RERETkUbxYJyIiIiLyKF6sExERERF5FC/WiYiIiIg8ihfrREREREQexYt1IiIiIiKP4sU6EREREZFH/X/jBWYG27+ZQwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 864x576 with 16 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "batch_samples,labels = next(iter(valid_dataloader))\n",
    "print(batch_samples.shape,labels.shape)\n",
    "show_batch(batch_samples.squeeze(), labels.numpy(), (4,4))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([64, 1, 32, 32]) torch.Size([64])\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAusAAAI4CAYAAADTU6e8AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAgAElEQVR4nOzdeXic1Xn///vWaLclW7ZlWZZXvBsDBgw2u9mCAyGQlEIISUgKcdMkDSRpGpLm903aJg3pt78sDSkNCQRIKIRsQNh3zI43jPcVG9mW5d2WLMnSzJzvH1bO/YwrIY01o3lm5v26rl58PJrlKJ3bz/Gce85R55wAAAAACJ+CTA8AAAAAQNeYrAMAAAAhxWQdAAAACCkm6wAAAEBIMVkHAAAAQorJOgAAABBSfZqsq+o8VV2rqhtU9ZZUDQrIZdQNkBxqBkgedZM79Fj3WVfViIisE5GLRWSriCwUkWudc6tSNzwgt1A3QHKoGSB51E1uKezDY08XkQ3OuU0iIqr6gIhcISLdvhGKtcSVyoA+vCT6qkn27XbOVWd6HHksqbqhZjKPmsk4rjVZiLrJOK41Web9aqYvk/U6EakP/HmriMw++k6qOl9E5ouIlEq5zNYL+/CS6Ktn3e+3ZHoMea7HuqFmwoWayTiuNVmIusk4rjVZ5v1qJu1fMHXO3eGcm+Wcm1UkJel+OSDrUTNA8qgbIDnUTPboy2R9m4iMDvx5VOdtALpH3QDJoWaA5FE3OaQvk/WFIjJJVcerarGIfExEHknNsICcRd0AyaFmgORRNznkmHvWnXNRVf2iiDwlIhERucs5tzJlIwNyEHUDJIeaAZJH3eSWvnzBVJxzj4vI4ykaC5AXqBsgOdQMkDzqJndwgikAAAAQUkzWAQAAgJDqUxsMgNynJx/vc9PEgT53lNu/9Q+NVJ9bJrV3+1yVy4t9HrQp6nPFsh0+Rze/d+yDBQAgx/DJOgAAABBSTNYBAACAkKINBsD/Er3gVJ/f/Yj9NTFpup2pUV3W7PP5VWt8vmGQtbQc7buzpvr8u00n+9zx25E+D9652+d4S0sywwZyXmTwIJ/3z5vm8+FKa0Ub/to+n+MrrDaBrKT23o4Mr/b5wLnjfd43JZLwkOKDlke8fMCeavVGn+NtbakcZVrxyToAAAAQUkzWAQAAgJCiDQaAiIi4M07yecuNMZ9/Mftun+eWdvjc6mzXl8DGLnJfU53PIwr3J7zGzUPe9vkjlUt9vnzbzT4PXj7GHsASPpAgPn6Uzx2f2OvzlKpdPm8+MMXnihX9My4gXQpKSnxun27v/9ovWUvLo+MeSXjMy23DfP7GwE/7PG5/jc/xd7ekcphpxSfrAAAAQEgxWQcAAABCijYYIN8EvllfMNAOOVr/93b7b+bc6fOMYmt9efVwqc+PHZjj89Pv2S4vLSuqfG6vtceKiNx+7q99Pre0yefpx9tBSHtnjPWZJXwgUfMEq9nRlZt83tJkdTdgW/bscgF0qcB2dymotdaVxtPsGvTI+McDD0iczn6w3K4vX51hu4q1jxpiz0sbDAAAAIC+YrIOAAAAhFR2tsEEl/ED3xLWUssSd10+1LXbDhYualtYuOD94zEBcpVGbHmxY9Ykn7888xmfTyy2Gni0xZYgv/nItT5P+VmDzyN222FJ8ea1PkeG2NK8iMh377vM57umWUtMeaHV5a4iFQABgZaAA+Mtn1Juuy0tXWEHxEx+5a3+GReQQlpU7HNkmLWrNF5kh+Z96lNPJf28BZG4z7GSIp8LS62lJmE+GMhhwSfrAAAAQEgxWQcAAABCKivbYCLDbLP7nVdO9Lngyt0+728q87mj1ZY9Kt+2VpnhS1p9Lq7f43N0s+1MAeScQBtMS40tO1YUWD280Fbp8/954OM+T/lFvc/RbdYGI86WGQvKy33e/DnbJUZE5DvH3efzqIjV5ZItdhDSxDXN9rTv82sA+aLghMk+j5hnNdgas/qtWh4RIJu1XRw4mO/D1g551elv+HzjoOWBR/RuCnvm2Hd9funy430eMu5kn4cts+uOLAy+RjjwyToAAAAQUkzWAQAAgJDKyjaY+KjhPp/3uTd9vnHoK3YfsSWUDmf/Jtl2ziCf1x8e4fOSg7YMv3TH9NQNthvRqC1ZHm60toEJv7NdMQoXrfM5fuhQ2seE/BDcEWnws+t9vvPQRwN3snjcmh0+xxosd7drkhbb0vxxF7+b8LPTSmzXmBK19308FtgBJhoXAObQuAqfTxtsuy09ttGW9McsaxEgm9VfZPOi33zgv3weXWjv7RItlmR9q/YJnz/xodd8XnTBcT7/98LzfJ7aau2b8RVrkn69dOCTdQAAACCkepysq+pdqrpTVVcEbhuiqs+o6vrO/1a933MA+Ya6AZJDzQDJo27yQ2/aYO4WkdtE5N7AbbeIyHPOuVtV9ZbOP3899cPrWmTvQZ///NRsnx+dMMPn9j222X1BZYfPx9XajjEfqFnl8xdHPGcvYN0xsuZwrc8nBpbwRUSKtOvl+mDbzZ64LfW3Odv9oi5ywOd1HdbW861m23lj0iarL9pgss7dErK68Zz1uMR22y5I5c+3dXmfeDcHiQVFqqt9brjaDlr6/qg7E+5XE7ElzOdabWem0vWBet2+1cbX9W+A3HS3hLVmMqx+nuVrynb5HN8w0OfC9Rt8pm7yyt2SxXWz+2/P8Pmycxf5fHpJ4BokJZKshpjtbrYnbo+fVGRzr1MGL/N5xrm2y9KXy67xebydA5hRPX6y7pxbICJ7j7r5ChG5pzPfIyJXpnhcQFajboDkUDNA8qib/HCsXzCtcc79ZZPlHSJS090dVXW+iMwXESmV8u7uBuSDXtUNNQN4XGuA5HGtyTF93g3GOedUtduzS5xzd4jIHSIilTokJWecxBttGXDib+zwo/bhtiRYtK/J59gAW3pvq7a2lvtHjvL5jrEf8Dk61Jb6y96z1pW2iYE2ARHRSNe/jgvubLHfXrtwuC3LvHa2fdN5VOF2e+2ywHNG+P5vrnq/uklHzfRGsq1WkWFDfd5/wQSfT79+qc+zS/YlPGZ33FrHvrzsMz7XLLRWtfieoz8kAjJzrelXateNwprhCT+66JSVPo8rtutfyT57THxfYq0BIuG81kSm2GGWB861edU1Q97s6u7danM2V/unhrkJP3t64xSfO/Zbm2V5tV3nPjfVdhCcP9jayP5h5jM+33/JpT6XPP+Oz67D2kP7w7HOBhtVtVZEpPO/O1M3JCBnUTdAcqgZIHnUTY451sn6IyJyfWe+XkQeTs1wgJxG3QDJoWaA5FE3OabHNhhVvV9E5orIMFXdKiLfFpFbReRBVb1BRLaIyNXpHOTR4m2BdpRVdnBQZFXgPoH7B5pSpCyQy4usRaV22BCf3RA7OEnqG+z246xtRkTERVS6ooFDXVyRLZW898FKn1vOshWntw/bkuewJfacrqm5y+dH+IWxblJuuLXBNM6xm58d+ZLPJVoafIT8ePcJPpc+ZfVQtnSjz7FudpxBbsuLmnkfWmgtly0zxyT8bO7gP/rc4eyyHQlu4ETd5KVQ102gtSsyxHa323Sd7R72seOtFeW4wOFHwR1gmuI2j1rRbgeE3bd7rs9v3X9SwkuPXn7Y56L91vrSMnqAzz+9xh5/1VnWanZq6Waff3SjPU9Ngb1G+es294wdsB0KgzuppVKPk3XnXHcb11yY4rEAOYO6AZJDzQDJo27yA99gBAAAAEKqz7vBZLPgt3mjDTvsB8Ec9Paqrm8/WqC9pvWSmT5XnmHf8TgUt38nfXP5R3we+6IdCBPdZ5v3A2FQOH6sz9vnDvN52snv+hwJLH0+0xpsPBO575lzfJ78qu36Emvk+0/Ib1pkl+NdM4sSflZXaDu9/LLxXJ/Ld3Z9MB8QChr4PHiEtb7Mu2yhzzcPfcPnigJrfdkVs/aTe/ef6vMvFp/t8+CFdv+6+1cnvHRs/36fXaA1ZWDgsMmy8dN8vv+EE32eP9jmeo+efrvPVxbO97lkj7Wq6TJriXGHbdypxCfrAAAAQEgxWQcAAABCKq/bYNIlUjfC560X2r+HXphxt88vt1o7QeWD9u3m2A7bmF/isfQMEEiCFtpfE40XjfT5rz73vM9fqFpi94nZ0vzfLfi7hOeaeq8tTcbXbhQgrxVELAZ2y2g7oTXhbkVqO728ucCW7icttfYxrhYIM1doc6ExJdYCWa7W8hXc9eXOfaf7fP/D5/k87SdrfY4FDtDr7fs/Fjg8bMTrtuPeT6fZ93GvmmeHH40ptFbON0+72+fzp9zk89B1tutZjDYYAAAAIL8wWQcAAABCijaYFAm2CuyaW+fz7NPX+Fwe2CXjxf1Tfa5c32RPFGMxE+ESGVHj84FJdvt1gxb5XKC2nP9k82SfJ9151EEtG97zkUNckO8ig+xgsL3njvb5l3PuSLjfwtbjfK5eajtbxN+tT+PogP71x2a7wNz3mLW+TLrDrhvRQOtLX0U2bPN56EK7bv32TDv86OaqddKVWEngD4Xpn0rzyToAAAAQUkzWAQAAgJCiDSZVTpriY8uHD/r8szGP+fztxrk+r/vX430uW2Y7adAagLDZ/Enbuehrlz/k86jAt+SXttsOMN9fcJnP01YmLiHGWlrSMUQgK2nlQJ8bZ9vtJ5ccSrjfbdsn+ly6p8Pn4MF+QLYo0lggWwvlf/zpCp8nfM92ZIn2w3XDWZdyt+MLKv9rOzyzfYNdIyMv7kn94IRP1gEAAIDQYrIOAAAAhBSTdQAAACCk6FlPkd0zbQuus0a97fPyjnKf/7zMtgOa9rqdVBqjTx0h4wJbV0Vm24lvVw5c7/O6Dvu3/k1rPuHz1J/ZVqTx5sTeWwDGldvJh8Mn7fa5VBMvzUsXWc/6lG3WE8tGvwizgjJ7f7/7UTuhd1KJ9Xt3OHsXR9qscTx+KP3XjthE22Z77yybh11VscLnDlcWyDbWA0/W+jxq/Waf0zWb45N1AAAAIKSYrAMAAAAhRRtMH2iJHWG1b4adKnf+IDu19MdbL/Z55JO2BVBsr7UWAGEQqbRWro0ftPatL01+3ueqAlvWfK1tkM8711b7PGiVnWzKVqRAooJSq6G2Oqu5z463bX5faStNeMzIBXZ9ce9tT+PogNTR4mKfjzt3s+Wi4CmkRWkfR/DaJqNG+Lj5Yts69e/PfNLnIQU27rjYtsQbopYHNFh2Tc0pG2t3+GQdAAAACCkm6wAAAEBI0QbTB60X244ZJ5zyrs8Fassjb799nM/TXtnic9TZsiYQBgcumebz8efZbkVXDFzt89J2a/36l9V2UumYJ+1b8rS+AN0rGGnL8A1nWj1dMsBq7h/rL094zMCNB3zuj10ygJSI2OfBMwdv9blC+3f+03z+VJ+3fsB2nPmrM173+dpKOzH1sLNxLzlc4fONz/+Nz1PX2En18da21A22Gz1+sq6qo1X1BVVdpaorVfWmztuHqOozqrq+879VPT0XkA+oGSB51A2QHGomf/SmDSYqIl91zk0XkTki8gVVnS4it4jIc865SSLyXOefAVAzwLGgboDkUDN5osc2GOdcg4g0dOYmVV0tInUicoWIzO282z0i8qKIfD0to8ykAtvBJTJ1QsKPGq477PM/1dmOGT+utx1gRrxmSy7RBjsIALkrW2um8cp2n/991FM+10ZsZ5hvbpvrc9Hvh/hc+uJSn60JDOi9bK2bZEWHB3amOMEOEAvuQLFsx8iEx4w9fFiAo+VLzRyLlo/O9nnbFR0+f2f2Iz5fNTC4s5LVX2PMroX/s+sMnyfdbc8jazb56Drs/umSVM+6qo4TkZNF5E0Rqel8o4iI7BCRmm4eM19E5ouIlEp5V3cBchY1AySPugGSQ83ktl7vBqOqA0XkDyJys3PuYPBnzjknIl1+Y8A5d4dzbpZzblaRlHR1FyAnUTNA8qgbIDnUTO7r1SfrqlokR94I9znn/th5c6Oq1jrnGlS1VkR2pmuQ/U6tdSVSZQe/bLpmWMLdbjvtF3a/wA4wG18b6/PE1+0b0OyRkT+ysWaunr7Y51klscBPrBXsrXp7b09YsM3nYOtL8NCX9xPcNcbFYl3eRyP22loY+OuqwD5n0GI7VCM+aYzPsdKe/3or3NdqY1hju3Gwo01mZGPdJKt9kC23nznaDtA77Ow911ZfkfAYbUuYfwFettdMJDDfig60f1MUjgv8XT7U6uHAFMuxYnvs0YZdb7vv/d+6V3w+rdRaXyLa9T9Q1nfY93FfefYEn49723aMibelfweYoN7sBqMicqeIrHbO/TDwo0dE5PrOfL2IPJz64QHZh5oBkkfdAMmhZvJHbz5ZP0tEPikiy1X17c7bvikit4rIg6p6g4hsEZGr0zNEIOtQM0DyqBsgOdRMnujNbjCviEh3aw0XpnY44RAZZN/Wbzpvks9fu/qPCfc7oXifzx9fe53PNW/Zkn50S306hogQy+WaGTvU3vO7zx3tc3FTbdLPVbbTdrgo3NdiPwgcGBatsi89tdYE2msC/+u2D7QFwr3zrK1l6GDbaaM7u1dW+zzl9lH2ups29/hYpFYu142W2HJ7c521bX291nZdagu874e+nfg/gzvYnMbRIVuFvmYC7+l1zcN9brONxKQg0OAx7ATr1nn3Ovv7uHWiXSv+NPcnPh9f3P0UtkithTLmrFEzLmVd3v9A3Npa7t1pO/qN+5YdnJTJnc56/QVTAAAAAP2LyToAAAAQUknts57TAocfRaeN8/nD//Ksz5+oTGxp+X2z7Yyx/091Pte+utbnrve4AMKnJW67VMQCy5eFgUXWx6fagRLy/b693j/umOXzU1um+Rx4ablw7Cqff1T7ZlLP3+rsoIo2Z5UYXMq8ve40n//QMNfn2h9uTuq1gPdTMM5axvZNtzf45KIBPr/bYa0uBdGjdtpzHDWG7ONarbVk7Z9n+rzkRjv0qyay2+cXT3zAHnyixeD1KNj10+F6N8MK3i+461JToK7u3W/Xo8VPT/d5jLzWq9dINz5ZBwAAAEKKyToAAAAQUrTBdIpUDvR5zyTbgeJLVXZoRYkWJTzmX35vuyFNeLbR59ievekYIpBWjyw/yecrBy/x+ZzS9BwQ9M81trz4reGvdHmfIrXPE+Ji9XfYdfjc0U2LwLcbz/P50VV2sEXBLmv3Kdltz1+7pH8PuUD+ODTJtr8YdeIOn/fFbBekH+063+ehLya2XEbZDQZZKN5mu7iMetLmRU9fdbzPs0uf8bk6kp5TVN+NWhvM4812Lbj9Odv1ZfKvrcbGb1nvc1hamflkHQAAAAgpJusAAABASOV1G0xk8CCf91xu3/698Muv+twSt+X26S/OT3j8pD/ZoSvxLVvtB+6ob/IDWWDqd22Z8vPXfs7ntomB9pBujt8oLLZWmZmjtvl817jHEu5XpsVd5p/tm+LzfRvtW/nNB+0AC3fI/rqqfsN2b6rcYkutGrWWmMJm2w1marMd5qQdgbaedqvvePMhywL0TfAgpP0T7L37j2Nf9nlT1G5/9NVTfZ560HZBEhGReFgW44EkBN63bu27Pr992yk+n3PODJ+/cubTPs8fvCFlw/jMyk/5rL8d5vO0BXatijUEWpk70tP62Rd8sg4AAACEFJN1AAAAIKTyug0mOn2cz7susqX0rw17w+fgpvlVL5UmPL5g83s+xw4fFiCbxTbYMuX4+6yVK1Y1oKu7J3AR+3f/7spxPl9Q9+XEO3bTRlO6z5ZLR+yyWioI1JV22M4ZWh9Ysty/PzAQGzetLMiouL0XNdAZua5thM+3v2s7Fk26397f8RbLQC5wgb/Lhz610eeq5dU+3/nOZT7fft5Bn780/QWfLx5gh07+W8M8n194y3aYERHRuF1shi0J5Je3+xwNzOHCjk/WAQAAgJBisg4AAACEVN61wRTWjfR583m2vP/V0/7sc1XEDkXa32Eb5ZftPWphPbCTBJBLops2H/Njg0eHVfVxHK6bDISdi9r1YcRrtqT/aNu5PpfvsmuKLl4aeGz4dqMAUiXWuNP+EMh1O6xFrG1lnc93TP6wz7cNtpaWivesfqasCLRDiojGAm1o+wM792XpoZV8sg4AAACEFJN1AAAAIKTyrg3m8CRbZomfYksjn6m0bycfdrbMsr7DFvIjhxPbYByHHwEAuhK4PrhFK3wetqibu6d7PEDIRRt2+FwYyNXP9/zYXN/9i0/WAQAAgJBisg4AAACEVN61wbRW214VY4fawSrxwCLKsy1DfP7yG9f4PHmr7QwjIiId7AYDAACA9Onxk3VVLVXVt1R1maquVNV/7rx9vKq+qaobVPW3qlqc/uEC2YG6AZJDzQDJoWbyR2/aYA6LyAXOuZNEZKaIzFPVOSLyAxH5kXNuoojsE5Eb0jdMIOtQN0ByqBkgOdRMnuixDcYd2fLkL/0fRZ3/50TkAhH5eOft94jId0Tk9tQPMbWKD8Z8Xvee7Qzz1YHn+/z0shk+T/vGJp9jR2+mz24w6Eau1Q2QbtQMkBxqJn/06gumqhpR1bdFZKeIPCMiG0Vkv3PuL8esbRWRum4eO19VF6nqog45nIoxA1nhWOuGmkG+4loDJIeayQ+9mqw752LOuZkiMkpETheRqb19AefcHc65Wc65WUVScozDBLLPsdYNNYN8xbUGSA41kx+S2g3GObdfVV8QkTNEZLCqFnb+622UiGxLxwBTrfgpO5Fi0lN2++bAfSbLQp9jAvRNLtQN0J+oGSA51Exu681uMNWqOrgzl4nIxSKyWkReEJGrOu92vYg8nK5BAtmGugGSQ80AyaFm8kdvPlmvFZF7VDUiRyb3DzrnHlXVVSLygKp+V0SWisidaRwnkG2oGyA51AyQHGomT6jrxx1NVHWXiBwSkd399qLhMEzC8zuPdc5VZ3oQ6J3Omtki4XoP9Ycw/b7UTJbhWhMK1E0W4VoTCt3WTL9O1kVEVHWRc25Wv75ohuXj74zUyrf3UL79vki9fHwP5ePvjNTKt/dQtvy+vdoNBgAAAED/Y7IOAAAAhFQmJut3ZOA1My0ff2ekVr69h/Lt90Xq5eN7KB9/Z6RWvr2HsuL37feedQAAAAC9QxsMAAAAEFJM1gEAAICQ6tfJuqrOU9W1qrpBVW/pz9fuD6o6WlVfUNVVqrpSVW/qvH2Iqj6jqus7/1uV6bEiO+R6zYhQN0i9XK8bagaplus1I5LdddNvPeudJ2ytkyPH4W4VkYUicq1zblW/DKAfqGqtiNQ655aoaoWILBaRK0Xk0yKy1zl3a2cRVDnnvp7BoSIL5EPNiFA3SK18qBtqBqmUDzUjkt1105+frJ8uIhucc5ucc+0i8oCIXNGPr592zrkG59ySztwkIqtFpE6O/J73dN7tHjny5gB6kvM1I0LdIOVyvm6oGaRYzteMSHbXTX9O1utEpD7w562dt+UkVR0nIieLyJsiUuOca+j80Q4RqcnQsJBd8qpmRKgbpERe1Q01gxTIq5oRyb664QumaaCqA0XkDyJys3PuYPBn7kjfEftlAkehboDkUDNA8rKxbvpzsr5NREYH/jyq87acoqpFcuRNcJ9z7o+dNzd29kr9pWdqZ6bGh6ySFzUjQt0gpfKibqgZpFBe1IxI9tZNf07WF4rIJFUdr6rFIvIxEXmkH18/7VRVReROEVntnPth4EePiMj1nfl6EXm4v8eGrJTzNSNC3SDlcr5uqBmkWM7XjEh2102/nmCqqpeKyI9FJCIidznnvtdvL94PVPVsEXlZRJaLSLzz5m/KkZ6oB0VkjIhsEZGrnXN7MzJIZJVcrxkR6gapl+t1Q80g1XK9ZkSyu276dbIOAAAAoPf4gikAAAAQUkzWAQAAgJBisg4AAACEFJN1AAAAIKSYrAMAAAAhxWS9F1R1iqq+Hfi/g6p6c6bHBYSVqo5W1RdUdZWqrlTVmzI9JiDsVPUuVd2pqisyPRYgW6jqPFVdq6obVPWWTI8nHdi6MUmqGpEjJ3vNds5tyfR4gDDqPAWu1jm3RFUrRGSxiFzpnFuV4aEBoaWq54pIs4jc65ybkenxAGHXOSdbJyIXi8hWOXLA07W5dq3hk/XkXSgiG5moA91zzjU455Z05iYRWS0idZkdFRBuzrkFIhKqw1iAkDtdRDY45zY559pF5AERuSLDY0o5JuvJ+5iI3J/pQQDZQlXHicjJcuSUOAAAUqVOROoDf94qOfjBEJP1JKhqsYh8WER+l+mxANlAVQeKyB9E5Gbn3MFMjwcAgGzDZD05HxSRJc65xkwPBAg7VS2SIxP1+5xzf8z0eAAAOWebiIwO/HlU5205hcl6cq4VWmCAHqmqisidIrLaOffDTI8HAJCTForIJFUd39n98DEReSTDY0o5Juu9pKoD5Mi3jfmEEOjZWSLySRG5ILDl6aWZHhQQZqp6v4i8LiJTVHWrqt6Q6TEBYeaci4rIF0XkKTmykcGDzrmVmR1V6rF1IwAAABBSfLIOAAAAhBSTdQAAACCkmKwDAAAAIcVkHQAAAAgpJusAAABASDFZBwAAAEKKyToAAAAQUkzWAQAAgJBisg4AAACEFJN1AAAAIKSYrAMAAAAh1afJuqrOU9W1qrpBVW9J1aCAXEbdAMmhZoDkUTe5Q51zx/ZA1YiIrBORi0Vkq4gsFJFrnXOrUjc8ILdQN0ByqBkgedRNbinsw2NPF5ENzrlNIiKq+oCIXCEi3b4RirXElcqAPrwk+qpJ9u12zlVnehx5LKm6oWYyj5rJOK41WYi6yTiuNVnm/WqmL5P1OhGpD/x5q4jMfr8HlMoAma0X9uEl0VfPut9vyfQY8lxSdUPNZB41k3Fca7IQdZNxXGuyzPvVTF8m672iqvNFZL6ISKmUp/vlgKxHzQDJo26A5FAz2aMvXzDdJiKjA38e1XlbAufcHc65Wc65WUVS0oeXA3JCj3VDzQAJuNYAyeNak0P6MllfKCKTVHW8qhaLyMdE5JHUDAvIWdQNkBxqBkgedZNDjrkNxjkXVdUvishTIhIRkbuccytTNjIgB1E3QHKoGSB51E1u6VPPunPucRF5PEVjAfICdQMkh5oBkkfd5A5OMAUAAABCisk6AAAAEFJM1gEAAICQYrIOAAAAhGgI0FIAACAASURBVBSTdQAAACCkmKwDAAAAIcVkHQAAAAgpJusAAABASPXpUKRsVFBR4bObMtbn2IAiu89LS9Pz2uXllqsG+xzff8ByS4s9wLm0jAMIg8IRNT63nDTaZ1eoPg9Y2ehzdEu9PZjaQAhpSYn9YfpEHzuGlPpcsqPZZ7d5q8/xQ4fSO7ijaFGxz5GaahvHgYOWm22s1ByQOXyyDgAAAIQUk3UAAAAgpPKuDUYm2HL72r+xtpTioW0+j3spPS+tdSN83nWWtQBU1Lf7XLRguc+uw24HcoEW2l85ey8Y7/OEL67xeUzZXp8fu/tsn2t/tt1nF42ma4hAUoLv6fipU31e/zfWWjlxvLVzbXnOrkHjfm2tL2lrg1FrKysItOl0nDHd5+2nWJtO7avW+qJL1/rsDh9Oz/iArgTet5FBlXZ7oGXLlVjtuVUbEh6ea9cIPlkHAAAAQorJOgAAABBSedcGc2CqLaeMnmBLk3uay7u6e0pFh9tr773Y2m5ay215cfSSAT7H9tEGg9wSqR7mc9s1+32+ddSjPtdEynx+5dIJPusvbPeKXFviRJYJLtGPGulz8/9p8vn5aff6/ODBk32+u9DaYKL1thtMumihtePIFGs9u/SnL/o8u9xaCP5OvujzqI12zYrt2pWeAQKdgi1lBVVVPh887zjLn7DdiiYO2e3z4Wvs2iIiEt1h87tc2MmIT9YBAACAkGKyDgAAAIRUXrTBBJdW9k+yf598pu4dn+9eOyft4yjcZwceFdQP9fn8i1b4vLHSdoyRwOEUEo+ldWxAf2iaPcbn80Yt8bk2Ym1ov26yGtix2PL41s3pHRzwfgKtL8EDvdb8q/1d/qvJv/L5R7vO93nBPaf5PO7Ot32Op3yQ/1tktLXprPrCQJ/vGWQ7j81+7Ms+T312n8+x3dZmAKRb7KwTfF57nc3bfjD3AZ8vKLNdwTZFrTXy20M/mfBcutt2FcuFnfX4ZB0AAAAIKSbrAAAAQEjlRRuMzLSDKiKn2A4UB6KBHWAWD0r7MFxRxOeOwdbW8vEhb/j8zxXX+awRu7+jDQZZqKC0NOHPWz9qu7h8v+otnyNqnxs81Gg7Z4x4M/C+z4Fv9CN7Bf8+PjTTdnSZP/NFn2cV23L79YvsfTz9MVu6j7ZYO2S6RAbb9WzvbGsl+81F/+XzN7Zd4vPYPwdqa1NghxpqDmkWqbZDjjZcYYd2/eR8203pgsBBeSVq15Rxhbar3pqv2k56IiJTbpvks66w3Y6y9XAvPlkHAAAAQqrHybqq3qWqO1V1ReC2Iar6jKqu7/xv1fs9B5BvqBsgOdQMkDzqJj/0pg3mbhG5TUTuDdx2i4g855y7VVVv6fzz11M/vGNXeNw4n1fdaO0un5/0nM8/X3auz5MetWWWdH1D/3C1jePk6e/6PCRiSzkSUUFOuFuysG5SoiDQvjVtQsKP/mrGUp9PCrQMxJx9q3/ltlqfJ72x2e6TyjEijO6WkNVMcCexSJ29L7edZ7efMWC9z1etv9Lnumft7/JYvbXBpEtk6BCf986b7PPEL6zxeXqRXWtefG2Gz1NX7/A52tycriEiPe6WkNVNMvZ+wK4RJ59m7SrB1pcytetDXKw1a6DagV+3n/ObhOf98diLfH7vuVN9HvP4AZ/dMquNsO+41+Mn6865BSKy96ibrxCRezrzPSJypQDwqBsgOdQMkDzqJj8c6xdMa5xzDZ15h4jUdHdHVZ0vIvNFREqlvLu7AfmgV3VDzQAe1xogeVxrckyfd4NxzjlV7fYr4865O0TkDhGRSh3Sb18t33q5HQRx3ZyXfX66cbrP1Y/aN4/d6pVpGYeW2Gs019lSzq2jH/c5EljWCR68gdz1fnWTqZpJlYJiW5psOCdxl6W/r1zt88AC+1b/gkAnmG4p8znWuDMNI0Q2ysS1JlIz3Od3P2U7wHzigy/6vD9mk5wtT4/zedyb7/kc7YdDWVydjXXnbPv17xz5hM9fCewAM/7PNqb4jkCdsQNMTgnjtSYyzA4Sazzfdgj73sgXfN4Vs9uXtQ/zud1Zm+VHBtiCwoVlibsszZ70O5//bfBZPj+31w7ArNlgtRtvaur9L5ABx7obTKOq1oqIdP6XKyrQM+oGSA41AySPuskxxzpZf0REru/M14vIw6kZDpDTqBsgOdQMkDzqJsf02AajqveLyFwRGaaqW0Xk2yJyq4g8qKo3iMgWEbk6nYN8X4GdJwpOsG/AD/ygfbt91gDbeeXBJ872edILm3xO1zJlwcRxPu8+xVaZZhRZ/tXBaT5ri23Y71y69qVBuoW+btJIi63dq3lO4tLkuMJ9PncEdoD5t80f9rl6Kcvw+SiMNeMGDfR50JmNPn916BKf//89p/hctc52lIjtSM+HmcEdanT6RJ/fu8x2gzn/9Hd83thhLQfLf2k7wFS/bjszxdsCfWjIKmGsm57EJtb5fOqUzT4PLbDrxec2fMznLa9aC1rMOovlobNtN5evjnwq4TWmFNln0V8ZZq3QT11ih2S2bLW5V9nTy3wO48FJPU7WnXPXdvOjC1M8FiBnUDdAcqgZIHnUTX7gBFMAAAAgpPq8G0ymFZTamsh7l9ohXd8+7j6fX22a5PPgwB740R22rJnaMdkuF7tOt6XJS8+0pdPVHXb//3jdvqE/bY+15rhYuDfpB/4iuDQvdbZL2NXTliTcryZirV0tzlrPNiy1Zc4prwZ20UjlIIFkBXZGicXts61YYAevIYWHfN471doyC1tP8rl0l7WZRHbuT3oYsWGVPreNsB0stl5gr3f1Ba/4/MmqN3z+/+qtxWzoL1/3mSZLZMqhOtvxa0a57ejy671n+LzzD2N8Hvvfb/pcONx2htm4wlpafn1z4m4uX69+0efaQmtn+/b0x3z+xpnX+TzxZZu3xULYBsMn6wAAAEBIMVkHAAAAQio722ACBwdpuS2nDDh7l88zS7b7/PUX7IvQU9456HO69pzQsaN83nOqLTZ+afjzPj/cdKLPE35j94kfsPFxOAWyRcEgW6bfPduWKT8/9FcJ9xsWseXIF1vts4IB9Zaj27YLEAod1oi1Z+9gnw/ErUXx84Ntt7Ehn/qtz29+5Difn1h3vM/Fy63lq7dap1obzflT7GCxH9c85/O0Yquh3xy0137nZWsDHS+7k35tINXahth7tbLQ3tt/fNdax4auC7SiBOot2L489FFrpVzzmcRDWtvsMiSxwM562zusXbqoyeaSrj3QmxxCfLIOAAAAhBSTdQAAACCksrINRiP2DXgZaksaNx73qs8Vai0kw1+1X9MtXZnewYnIntOrfZ5z8pou7/PH+pk+V75gO2bQ+IJspAMH+Lxvut1eHGhZE0lcjvzKymt8HrI6PYeSAX3hGq1tZPT9tq7+qxNP9/mqQYt9vrjcdjK6eqAdivSfIxf6fPg8W25vifdu6f1QoG6GFNhhYuUFthvayvZWn3+4yrbYnniXHRDI/mIIgz2n2jtx9oCNPh8cbTuyPD9ljs81LwYOAgseClY1yOdP172U8BqDC7r+LHrhwbE+l++wGVe8LXw7wATxyToAAAAQUkzWAQAAgJDKyjYYUfs3hiuzJcGzymw55ab3rvB5wPb+XWJvG2pL/1MG2jeXN3fYbgI73rPDkirFxg1kI1duy5eV0/b4XKKJnwdsi7X43LFgqM9lS9b7zFI9wiLe3Oxz6XPv+Pzm1dbr9fSJ5/jcONv+7v+bD7zg8/wqa3X8732n+nzn4rN6NY4RzxT5POOm5T7/YrS1fr7ROt7n6CrbnSm2qetWTCBTal6x68KC2VN8vmDQKp//NPU0n4eca7vE7D7BWr9i5x3w+eyy+oTX2Bs49atArcXlWyOf8HneubZTUvXL1h4TW2+HU4YFn6wDAAAAIcVkHQAAAAgpJusAAABASGVlz7qLWVdr5MAhn+/bN9vnaRW2XdXTw60vqaS83Od4i/XP9lWkyraQPDjDeuT/boht2fU/B+0Uu4q11oMIZKNItW1RuvPswLZ2J/zI54FakvCYFXYgpAzYYU2FsT170zBCoI8Cp0i7w9b3Gt+42edBjXZy9uC3rFd8wf2zfH6h7AyfI622XeO0A71737tyq6OmqOW3Dttzffe1D9nzPmDPG4vzLRCEy9DXbH722qfstN3PDX3N57sv+7nP6y8aYY8ttO+RjCzc5/OPdp2b8BrP3GdbP1ZfutXn/5xopwx/aIZ9D+XlS6xHfjg96wAAAAB6i8k6AAAAEFJZ2QYjgdPcXJMtibyww9pdvjv5IZ9f+rTdvr3GTg6tWtvz6XEuYltxNY2KJPysbZj9rK3Glho/fdrLPg8tKPO5JWbLl4UtnFWK7KYDAu/t4VYLxxfZdqqRo7Zu/OVO2+aubFegJ4alemQRF7X3bmy/bR8nwbzFYrAKgn/z9/Zdv/0fzvT5Q1W2XePXN1zlc93jdn2Kr3u3l88M9L/Ytgafm35/is/f/LS1cv1sjG2xeG6ptc2s67DW5+tXfcpnvcfaMkVERr9qpwlvHDTG5+drp/r80So7ffjRWSf6PLJupM/Rbdvf71fpN3yyDgAAAIQUk3UAAAAgpLK0DSbwDf2WVp9bH7eTsH5eMdfnz419yefFnxzn88oDtT2+VIHaa9UWHU742egy+ybyqQM2+3xmwklatvvMmkM1Ple/fUiAbBarHuRz+/TWLu+zO5b4Pn/5FdsRaXJ9YMeKFI8NyDZaaJfj+OwZCT+bcLmdcj23fJ3P/7HlEp+nrrUWnHhH/57aDSQjuLPSiCdsvrS8zE4Gvuji4T6fMcLaut5oHOdz8V12EnzFk3ayr4hINLDb35in7LTsH429yOf/e+bvfA7uDPPqh2wnp2E/z5I2GFUdraovqOoqVV2pqjd13j5EVZ9R1fWd/63q6bmAfEDNAMmjboDkUDP5ozdtMFER+apzbrqIzBGRL6jqdBG5RUSec85NEpHnOv8MgJoBjgV1AySHmskTPbbBOOcaRKShMzep6moRqRORK0Rkbufd7hGRF0Xk62kZ5fuIt9lySt1D9u3fTS2Tff7GHNt0f+ToPT4PKmnr8fkPx+x/olUrxyT87K3A2n39qfYP1+l19k3nusAGMtsODfa5aPEan9kXJreEvWb6QktsR6MDEwf4fNNM++Z+PPCOXtZuh8SIiIx4I/Bu37pDgL/I5brpDS22XZTqLypP+Nnto57xeX2H7XpRttkeo7s4WCzf5ELNROvtwKJRf7Tbmzfajiyv1llbStlu2w2w4rFlPscPJ7YpB9ulC95a6fOwyXb40e8m2fPeXGs19tQF03we/uv0HKSZrKR61lV1nIicLCJvikhN5xtFRGSHiNR085j5IjJfRKRUyru6C5CzqBkgedQNkBxqJrf1ejcYVR0oIn8QkZudcweDP3POOenmA2Ln3B3OuVnOuVlFUtLVXYCcRM0AyaNugORQM7mvV5+sq2qRHHkj3Oec+8tCRaOq1jrnGlS1VkR2pmuQ7ytwmEpwOWXonZZrnrP2lfbR9q3glgrL3SmI2nt88vOLE34WPBhj+Tfs0IonPm7fbj5+yFqfC9WWbwoGWgtBbB/f3M81oa6ZPohUD/N5/2T7t/7nB9u39VudHTZ2z87zEx5fsd6uI7GDCdcUIGfrpje0uMjnIWcktoiNLGzy+WvL7SCkmkVWa7HGnPyfBT3IpZoJzuFKg7mb+8e7uf1owbna0KX7fX5rqR2YeWiE7Rr4oYkrfF52uh2kWfiK7RgTfM7+0JvdYFRE7hSR1c65HwZ+9IiIXN+ZrxeRh1M/PCD7UDNA8qgbIDnUTP7ozSfrZ4nIJ0Vkuaq+3XnbN0XkVhF5UFVvkCMHK1+dniECWYeaAZJH3QDJoWbyRG92g3lFRLSbH1+Y2uGkR3Sz7RJTEMjJdmi9364tJfvtpw3tg7q8z6ASOzimeYS1E8i+fV3cG9kqF2qmO7FaO4SibXTX7Vu7YrY8+MaLxyf8bNJeaxHr7RIm8kMu1013tMh2c3Gj7JC+r014NOF+pYHD+VpetWvH8GWbfO7fRXmEQT7WTF/F37Gd+OqeO93nf5t6mc//ctxDPv/xulN9nrbarn+xnbvsSV369/Tr9RdMAQAAAPQvJusAAABASCW1zzq659RWoiKBBf6I2r+HdrUO9Ll49fr+GRiQQrFy+yujuMLaYA47W4R/rW2szxN/k3hQS2xHVmxKAPSLyIjhPq//hB2sd3ZpY8L9NkWtabNys11fog2BXWO0m26IfliiB7JGoB4qF9qOM5ufs+tW9SRrWX72Az/y+YaHvuxz2QK7T7zJdmtKFz5ZBwAAAEKKyToAAAAQUrTBpIgGllZigX8DxZwtWTrX3Ze2gezQNNqW4y+ZYIeENQcOQlpwYLLPuvdAwuNdtEMAHBEbbjuHferSF3yuKChOuN93Nl/hc3mj1VBBuR0RX1A12B4QaImJbt2WkrECuSa6bbvPo56znV7mTft7nxeff5vPLZ+3A5VKd9hhm7rMDr9M12FJfLIOAAAAhBSTdQAAACCkaINJkXiRLTuWFNgySFRiPh+ORew+/TMsoO8K7H3bVmX/vj+zYoPP7YE2sH3ttjTv2o86OImdKQATaFcZFGnt9m4tHdYWs+dUu3q0XH6iz5HaFp/L3rCdx0b8mDYYoEuB61Fkve0MM/ypKT6/fIYdQvar4+/1+W+O/4rPQzdV+BxL0yGXfLIOAAAAhBSTdQAAACCkaINJkQMz7Bv651as8flfd53i88GXa3yulI39MzAghYqbbNlwYfN4n08qsaX21bvsfT46xhI80Ff/MvEhnxeNPM7n1/ZZXrpwos/Dl3bfUgPgf4sdOOjz0FfsuvVPP/+0z0986d993jPPaqxq9Sh7ooW0wQAAAAB5hck6AAAAEFK0waTBd9Z82Oe2BfZN4jGP7/U5LkCWCBzsNWSZLRX++ZnZPm87xw5kaVtnB724ji1pHhyQvSK7rZ5+9Pw8nwsufDzhfosOjvP5xZW2U0XlctslZsJC2w0m8o61WXKtAXohbjv3xbY1+Dz6IZsm/9d1Z/r8+RMX+Pw/J17i8/C1lfY8B62++4pP1gEAAICQYrIOAAAAhBRtMCky8lk7OKYgWuXz8IXv+Ryt3ypA1gkcHKGrbXn9uN9N8HnDBluaH7XVDgVz7bZLEoBE8R07fZ58jy2f377t8oT7le+wGpy20Nop3bv19lxthy0HlvQBJMdFA9ew7Y0+P/zg2T5/9hPWqtY8xg43qxk2xJ6INhgAAAAg9zFZBwAAAEKKNpgUGfjgG13eHu3yViA7xdva7A+LV/o4dHHX93dd3wxAjqqnRSt8rFvU/WNocAH6T7zVDj8ac9tyn39abbs3Dd4WuNLF07P/Uo+frKtqqaq+parLVHWlqv5z5+3jVfVNVd2gqr9V1eKengvIF9QNkBxqBkgONZM/etMGc1hELnDOnSQiM0VknqrOEZEfiMiPnHMTRWSfiNyQvmECWYe6AZJDzQDJoWbyRI9tMM45JyLNnX8s6vw/JyIXiMjHO2+/R0S+IyK3p36IQPahboDkUDNAcqiZfhDYDS3e1OTzxK/0b+tzr75gqqoRVX1bRHaKyDMislFE9jvn/jKurSJSl54hAtmJugGSQ80AyaFm8kOvJuvOuZhzbqaIjBKR00Vkam9fQFXnq+oiVV3UIYd7fgCQI461bqgZ5CuuNUByqJn8kNTWjc65/SLygoicISKDVfUvbTSjRGRbN4+5wzk3yzk3q0hK+jRYIBslWzfUDPId1xogOdRMbuvNbjDVqjq4M5eJyMUislqOvCmu6rzb9SLycLoGCWQb6gZIDjUDJIeayR+92We9VkTuUdWIHJncP+ice1RVV4nIA6r6XRFZKiJ3pnGcQLahboDkUDNAcqiZPKHO9d+xJaq6S0QOicjufnvRcBgm4fmdxzrnqjM9CPROZ81skXC9h/pDmH5faibLcK0JBeomi3CtCYVua6ZfJ+siIqq6yDk3q19fNMPy8XdGauXbeyjffl+kXj6+h/Lxd0Zq5dt7KFt+36S+YAoAAACg/zBZBwAAAEIqE5P1OzLwmpmWj78zUivf3kP59vsi9fLxPZSPvzNSK9/eQ1nx+/Z7zzoAAACA3qENBgAAAAgpJusAAABASPXrZF1V56nqWlXdoKq39Odr9wdVHa2qL6jqKlVdqao3dd4+RFWfUdX1nf+tyvRYkR1yvWZEqBukXq7XDTWDVMv1mhHJ7rrpt571zhO21smR43C3ishCEbnWObeqXwbQD1S1VkRqnXNLVLVCRBaLyJUi8mkR2eucu7WzCKqcc1/P4FCRBfKhZkSoG6RWPtQNNYNUyoeaEcnuuunPT9ZPF5ENzrlNzrl2EXlARK7ox9dPO+dcg3NuSWduEpHVIlInR37Pezrvdo8ceXMAPcn5mhGhbpByOV831AxSLOdrRiS766Y/J+t1IlIf+PPWzttykqqOE5GTReRNEalxzjV0/miHiNRkaFjILnlVMyLUDVIir+qGmkEK5FXNiGRf3fAF0zRQ1YEi8gcRudk5dzD4M3ek74j9MoGjUDdAcqgZIHnZWDf9OVnfJiKjA38e1XlbTlHVIjnyJrjPOffHzpsbO3ul/tIztTNT40NWyYuaEaFukFJ5UTfUDFIoL2pGJHvrpj8n6wtFZJKqjlfVYhH5mIg80o+vn3aqqiJyp4isds79MPCjR0Tk+s58vYg83N9jQ1bK+ZoRoW6QcjlfN9QMUizna0Yku+umX08wVdVLReTHIhIRkbucc9/rtxfvB6p6toi8LCLLRSTeefM35UhP1IMiMkZEtojI1c65vRkZJLJKrteMCHWD1Mv1uqFmkGq5XjMi2V03/TpZBwAAANB7fMEUAAAACCkm6wAAAEBIMVkHAAAAQorJOgAAABBSTNYBAACAkGKyngRVjajqUlV9NNNjAcJOVe9S1Z2quiLTYwGyCdcaoPdU9SZVXaGqK1X15kyPJx2YrCfnJhFZnelBAFnibhGZl+lBAFmIaw3QC6o6Q0Q+KyKni8hJIvIhVZ2Y2VGlHpP1XlLVUSJymYj8MtNjAbKBc26BiITqYAkg7LjWAEmZJiJvOudanHNREXlJRD6a4TGlHJP13vuxiPyj2KlXAACkGtcaoPdWiMg5qjpUVctF5FIRGZ3hMaUck/VeUNUPichO59ziTI8FAJCbuNYAyXHOrRaRH4jI0yLypIi8LSKxjA4qDZis985ZIvJhVd0sIg+IyAWq+pvMDgkAkGO41gBJcs7d6Zw71Tl3rojsE5F1mR5TqqlzLtNjyCqqOldE/sE596FMjwUIO1UdJyKPOudmZHgoQFbhWgP0jqoOd87tVNUxcuQT9jnOuf2ZHlcq8ck6gLRQ1ftF5HURmaKqW1X1hkyPCQCQc/6gqqtE5M8i8oVcm6iL8Mk6AAAAEFp8sg4AAACEFJN1AAAAIKSYrAMAAAAhxWQdAAAACCkm6wAAAEBIMVkHAAAAQorJOgAAABBSTNYBAACAkGKyDgAAAIQUk3UAAAAgpJisAwAAACHFZB0AAAAIqT5N1lV1nqquVdUNqnpLqgYF5DLqBkgONQMkj7rJHeqcO7YHqkZEZJ2IXCwiW0VkoYhc65xblbrhAbmFugGSQ80AyaNuckthHx57uohscM5tEhFR1QdE5AoR6faNUKwlrlQG9OEl0VdNsm+3c6460+PIY0nVDTWTedRMxnGtyULUTcZxrcky71czfZms14lIfeDPW0Vk9tF3UtX5IjJfRKRUymW2XtiHl0RfPet+vyXTY8hzPdYNNRMu1EzGca3JQtRNxnGtyTLvVzNp/4Kpc+4O59ws59ysIilJ98sBWY+aAZJH3QDJoWayR18m69tEZHTgz6M6bwPQPeoGSA41AySPuskhfZmsLxSRSao6XlWLReRjIvJIaoYF5CzqBkgONQMkj7rJIcfcs+6ci6rqF0XkKRGJiMhdzrmVKRsZkIOoGyA51AyQPOomt/TlC6binHtcRB5P0ViAvEDdAMmhZoDkUTe5gxNMAQAAgJDq0yfrAAAAQJ8VRHyMnj/T5y3zin2ODYj5XLnWprAjf74k4anibW3pGGHG8Mk6AAAAEFJM1gEAAICQog0mDbTIlmwKKgfa7eXlXT/AOR/j+/b7HD1lsj02Gve5cE+zPXbnbh9j+w8c03gBAOkTqaryWcvLfI4F/v6OjBju8+EJlkVEDo61A2tK91sbQFGz5cKmdsv1u3yONuw41mED6RdofdFTp/u86Wr7LPn3H/iJzxGx+dJHyj7vc12xzbtERIQ2GAAAAAD9gck6AAAAEFK0wfRBYd1In2Mjh/rcPMbaXZpG2RJPu62EJrIOFxlYP8bnvRfaMo6L2/MUNNoS6chX7HXLHn6rdwMHAKScFtolVadO9HnnGfaXf+sw9bm8cazPh0ba7e6kpoTn/cjERT6/uWecPW+TtVk277XrTvmG43westZeo/LtRp+jm+vtBeLWTgP0p8jEcT6v/yt7P396zkt2e3uNz7e9e77PwxZY64trtzawXMQn6wAAAEBIMVkHAAAAQoo2mL8IfCM5MnCA3T6i2sdodUXCQ+rn2LJj66ktPv/1tNd9/mTVGz5PK+56N5gOZ0uQT7faa59UbDsF1ERsB4EitbGOH3Kjz5Mf7vLpAQD9oOA4azlZ95nBPn/rsj/4fEn5Jp+bnLW+dDj77KyiILEtpcM2wJAbhrzW5WuX2FNJ5GLLfz5ku4r9+DdX+jzujkM+x/bstQfQEoN+tPNca+u95EI72Gh8yU6fv/ns1T5P/qXthjdgqc21At3EOYlP1gEAAICQYrIOAAAAhBRtMJ0igyp9bjnDvsXfcP1hnx+fc1vCY8YUdt2akqibg5AC4oEFnPNK9wd+Yv/vOew6uswSC6x9AmEQPOSiIPH9qSV2uIsWBf76iQTqpyPqY/yw1Z8L3M5SPUIj8H7ffLXtc9uv/QAAIABJREFUWvHFeY/7/MkKO5io1dn7fnuHfV72wL7ZPq9tsucREdnVYu2R0VigvtT6Yy4audbnb1bbzmCfqbRdX6be8F92nzV/63Plc7aTBofroT8dHmzXiKHF1uLy0v6pPg97y+rELV3ZPwMLGT5ZBwAAAEKKyToAAAAQUrTBdNJya2nZO7XI5zfO/InPlQWJLS0RPfZ/6+yL2e4xv22a5PNDO2b63BHvrrXGDFxf1ON9gHQrKA/UxiTbEaN9WGLNbDvPDrGoOs2+7T++0najeH2d1cOIp+39XbV4l8+xdRv7NmAgRQqm2/u16mxrd7mmYoXPDYGurfkbbWeLA7fbIXgVm6wFILIn8VCkqoP77A+xwJMFrkELLjzD8o3WyvnSCb/3+ZTiwEF7n7V6ijaMs99n8Rq7T6ANDUiHuF0SpESt1XHRjtE+175z0OfAxkh5hU/WAQAAgJBisg4AAACEVM62wUSGDvF59+VTfA6cQSHDFgaWFvfYLiwaWGXcH7edWjZFA7tRiMjgAvsG/a/22RLkY1uO9/nAZjsYo3KD/dto0GZ7ruKDtrtL4Z5Wn0tczws+Y/baARvR97kfkAqRGjvAovmMcT7Xz7P7XHGaHWxxQvnWhMdPKLbWl9GFtrRZHqjLxpHW+rLmzBE+r2gd5fPT22yngL1rhvo8+b/t+WMbt9iTsnsM0iRWaTscja+0NphhgYPsbtlxms/b/zTO55FP284W8VZrUYl2HPW3eS/ev5XPWvtKU9zq48LPfdTnx6c/6PO90+71+SNz/tHn0e8Ns3Fs3dbj6wLJKhxhux0NOMvascaW2EGQLWts7qSrl/pMGwwAAACAUOlxsq6qd6nqTlVdEbhtiKo+o6rrO/9bld5hAtmFugGSQ80AyaNu8kNv2mDuFpHbROTewG23iMhzzrlbVfWWzj9/PfXD6wW19fOCMlt2XP+P1vpy8QW2hPLE4hN9rlpjy5eywdpgRj5nO1NcWWDLg+WN1hIjItI2xP6tM2CHLVNWN9g36Gv32XPpXjtsIh44eMIF2mviUZpZcsTdEua66UHhWPsm/t6z6nzecZ69zy84aZXP3xr2ps+nldh7e/tRh3bdG2gXe3uvtbVcPXKRz3PK3vW5OtAq89Wh9honl1uLy9I6233mke3n+Dz6D1ZL0c3vCULvbsnCmokX265dZRFrabyvqdbnPz9jBx5NesRaw6IpPIAoeJhRxUvrfd4z0K6FS79jl/w5JZbbTredaKKvWAup0AaTDe6WLKub7VdN8PmTY5/0uVQDLcGtdu2It1mLWCoF23E6Jli9tg+yVkwNTPvKNtl8Lr7F6rg/dk3q8ZN159wCEdl71M1XiMg9nfkeEbkyxeMCshp1AySHmgGSR93kh2P9gmmNc66hM+8QkZru7qiq80VkvohIqZR3dzcgH/SqbqgZwONaAySPa02O6fNuMM45p6rdfkHXOXeHiNwhIlKpQ1LzRd5A60ukosLn+vkzfJ5/2VM+7+6w+5RvsV+5qMG+uR8NLLPomg0+1zVU+hzbG9g9RkSqBtnPgt/kDy6JsAcFuvJ+dZOWmumFpmvm+LzjHHvZmSfaAUTfHrnA57GFVg/PHprm8y93nOvziqdsCV5EZGC9PW/pfquO7330Up9rh1tLWsMOa7X86dn3+Ty92Gq3KHCQxt6PDfD5nS0n2es2NPrMQS/ZKSPXml7Ydl6pz58YZO0nDzWe7PPQZTac/mjJiu2xD1qHvG11+pPtF/t88rgnfP7IlHd8frPadq4JNIoiS4XxWtN6rh36dcGA1T7/pPEinyu2pG44hy+z93Tjqdbi0jbWdvQbVmMtl9UDrC0sFrcGlPWNtlNS5Uun+DziyUBr25b6FIz4fzvW3WAaVbVWRKTzvzt7uD8A6gZIFjUDJI+6yTHHOll/RESu78zXi8jDqRkOkNOoGyA51AyQPOomx/TYBqOq94vIXBEZpqpbReTbInKriDyoqjeIyBYRuTqdgxQRkQL7xn2k2g5B2X/+cT5fft0rPl9RYct68/78FZ8nP2NLHbGt27t8qeDuLMHlxKPFUvhNfuSW0NSNDcjH4K5J7WdY+4r7tB1OcdfkP/k8pchqZkm7LQPeuOYTPu9/yQ4vqqi3r8+PfXh5wjDizba8KIFDv0pmnulzw/5qn2tet3F/Y/BHfH5t1q/sOWWPz/9Ra206Z9TZMmVFcbG9LG0woRS6mvnfA/QxeOje4DnWYnVCiS2Hf3+HnRQ2Zmvm3nMFB1t8XrRkos8tY23njVMH2A5Mr5bZzjW0wYRf6OumkxbZ38EnjrS5V0ngFMoX37br0bQ37YCk7tqJtdCmsJG62oSfNXzQdhvTy+wa8c9TrEX61BLb7Wh/3MZXH7X6PqXEWi7LJ9vfAZ8bfYXPOxttHjqg0a6jqdzFpsfJunPu2m5+dGHKRgHkGOoGSA41AySPuskPnGAKAAAAhFSfd4PpL5HAzivNc8b5XHajLafcMOQ1n7+2xZbMRz8Z+Fbxcvu2vuMAIuSJgpLAgvbkcT7Wf9aWwn8/zXZbGVtoNfOzvaf6/IvXbaeX0U/YkuCYJ5f4HFz6SzxGTBLbccptq7CC9sB9Kq0uDxxn4z7cZLtuDCywfLytXsqrbfaKRc2Buo+xLxP6RgttF4mW2bbsfebwxT5v7rA2MbfRdiYqrrfl9v6+6rjATmUVm6ydtM312+YfyFPBNhU5cZKP4wes9PmlFrt98HK7f2y1zdWCz1NQZTuEdUy1VpctF1p7p4jIl695yOfLB6zzeX10oM8/3T3X58fXT7fXbrBrU9lY27nmhyc+6PP/Gf2ozx+fYq3WFW8M8jm+I3VtMHyyDgAAAIQUk3UAAAAgpLKmDUZGDvdx60W2lL562h98vr/Jlib3/mCcz+UvrPA5zk4QyEM60Jbkd86yZbp7T7/N52lFtsy/st0W63/x6nk+T/wfa5sparRdYtzEcT4HPwHQo5ba4+XWs3KozpYtna3Oy6TRtrvGD86z+h5VGGwgsN/nsLMxzV/6aZ9Hrm21122j7tE3WmotWVsus2vQ96oW+fyDejvca/hia8nqj4OQgLDRQPvlux+2687XKm2XsH/f/EGfK7Z13SQWbH05MNfmefuuOeTza7N/lvCYQ87q794DdkDZz5ec43PNk3Y9mvisHf4X22U7ukQG27i/cpdtqvO7U37hc/ugwHWuPLEdJ1X4ZB0AAAAIKSbrAAAAQEhlTRtMe7Ute4+ZYsvkHc52eXhk50k+l+60b+G64E4Qgd0ohG/DI18E3usa2KJlf9y+9d7h7FvvMbE6qaqzw782/vVgn0tq7bFThttp1vHAY5vaE49VmTvclj+vHbzQ5yKx8UUCJTqm0L65H3PWprMvZge9PNNqh2EMerDCnnPFantsnN1g0DdabO+/q894y+dI4L277kVbop+wJHM7wABhoBHrb4xOsr+zqyPWvrJlj7W4jNofqJTAXK19xmifiz5r879VM2zHl31H/RV/y9YP+bzmrsBhS4/ZAWDRBjvwqLsrhIvZBbNlu12Pds20658GH5ymeSWfrAMAAAAhxWQdAAAACKmsaYNpGWHf2v2vSb/1OXg4yn+O+5PP53zuZp9HPjnT58Gvb/U5utWWKYFcFj9gO7cMf97e99+56nKffzzV6mpGse2w8sIpd9vznGxLfAXBlrKAjsC38Es1kvCzosCfC6Xnb83HAs+1rsNa265feb3Pg75vLXKDltlhG7Hm5h6fH+irJ5pO9Ll6ma2HR7ds7eruQP4ItMGMHGrtlAO058awyPBqnxtOsXne/ZPv9/lAoKVz7uIbEh5f8wObM1YvW+ZztLVVUuGBPXN8HrrCrouxrQ0pef6j8ck6AAAAEFJM1gEAAICQYrIOAAAAhFTW9KxHS60/9oTAFlpBtRHbSudPF9hpVivPHunzrasusQe8PtbH0t3Wc9ReYa9Vsc16EAduTOyB1cCWkK2jbcu44r3tPheuq/c5tmdvl+MG0s1FrUcwts22qxryD+N8/uaov/V5ywftr4aTZ23wefyAPT6vb7JThbc12SlvA4rt/f/BWushFxG5YfDbPg+LDJCePNdqPY+ffeVGnyf+t9VeZNk6n+PBfkS2ZkVfFdj7L3gKcHmB1VBH4PhdjQXec2wXCnjDy21r4JLA150SvvoUyK0zx/hceJ5ddwoCW6VetuITPtf+a+Jnz7pyrc/xtjY5VhoYoCuzmm6OWk984WFrnndR+75XKvHJOgDg/7V352FyV3W+xz/f7vSSpLOvnZCQfWGRJYEAAsoqqCPKqICKDKKMDjowjndkeOZexqvXi8/M6MwdlzsMIHEukInsKltAVsGQsGYjJGQPWchK1k5317l/dHG+v2bSpCtdXfWrqvfreXj4dHUtp3jqy+90/b6/cwAAKcVkHQAAAEipkmmD6b3BTy18de1HYv7l6Gdjrjb/2+P4urpE9lMoRx//y5j/MHlCzNtb/BRnQ7WfMtlwwHdsXL/PsyRlgp8emVj3Vsxr9gyM+fWVY2Me9PTkmAf+8gUBxRCavU2ldbG3kNQu89N6k9eNiXnb77xdbEutf56rE6f++jcllqxr8JaY2dec0O61P/OhZBuM37474zX3w3dOjvmehz4c86T7E21orydaX5qaBHSH6gH+Wd507hExf6yP75z4+91HFXRMOevjx7Zdx3ut1CdO7+/PJFpL6R5DN6iyg3+w2nUrJvL+gT49nTbMl0GtTWzB3ZqYg1W91X6p1NautL70SEyNjxge47dm/D7m+Tv9uFi9PzFwdjAFAAAAKguTdQAAACClSqYNptfCt2Ne/PNjYh53zsSYbzv99pgHVu+N+cgeflri2Npke4yv1JJPyVP6i0f7uf6/Gn5JzE0bT4q57uF53TIOIBcdtcfULD70Y6sHDIh51ye83evCUUva3W9gB18P3LjptJgfvtd3hht/7zs+pjd8VRpWekEhWK23hu1t9FPuR/bwVYdqqg69G2OhVfXx1cl2H+07Qf75tCdj7lPl7+2Hr18Y85Fr/dgJFEu/N3z1mMdfOjrmb398TsxfGD0/5v+46OPtHj/k94mV+Db46k3JldE6Uj1qZMxL/sKPbT/r83rMP336vJgnr/PdWRObqubVIb9ZN7NRZvakmS02s0Vmdm329oFmNsfMlmX/PeBQzwVUAmoGyB11A+SGmqkcnWmDaZH01yGEoySdIukaMztK0vWSngghTJT0RPZnANQMcDioGyA31EyFOGQbTAhhg6QN2bzLzJZIGinpIkkfzd5tpqSnJH23W0YpqeXtDTEPvHuH59f8ityr1/umLplaP03eY/SemIf28xUlThq8OuavDnou5qm1vrnS4Wioqo/5ZO+60bfG+inI//GRS2Me+3CXXg4pk5aa6W7VgwfFvPMsX1mpz5XrY/7rQXPbPaZvVc+YVzZ7Ld73R28Lm5JsfVmyLD+DRepVSt10l+q+fWNuPm58zOvP8u/kvtzvlZhf2O/37/tQgz/PW76ZDNs6pVsqaya5kVg/b8caXu9zuNp2OyF1YLG3PY6721ufLx7i87yfnDA75hFfWdHu4UvHeg2Mudc/38nnrervqz0dONbnkqvP8onbTeffGfM/bD435lGP+BzT1nibdnfJ6QJTMxsj6QRJcyUNy35QJGmjpGF5HRlQBqgZIHfUDZAbaqa8dXqybmYNku6RdF0I4d3k70IIQR2szmpmV5vZfDOb3yzWQ0bloGaA3FE3QG6omfLXqdVgzKxGbR+EO0II92Zv3mRmjSGEDWbWKGnzwR4bQrhZ0s2S1NcGHv4SDonVHzJ7E1erv+arTYx97eAPrT7aV6doHuwbGz3y4caYe3zOr+H90TDfuKUp+GZMj+71UyaS9PhOv0L5mN5+6v+C3r6SxugefvplRM12H8cQf16Un1TUTDeo6u0brOybPi7m3V/048MDk2bFnGx7kaSWxIn1G9/2q/eHPu+nRTPLVuZnsCg5aaubkPHjQmIBGO1NPHsmFHEF5EQ7QWbS6JhXf8JbMS/7qLd47kpsIvONl78Y89gXfePA1m1+nEL6pa1mrMo/Y6Gnt5PU2MGbqgb19Tbl5obBMVc3+6ottc8vivnIXb4C4De+cXnM5x/j95GkxVO8zXLfSG/H6Wn++M0n+3xwz/l+/+uOfjTm1/d6Xf3hjhNjHvmit9O0vtvu76Nu0ZnVYEzSrZKWhBB+nPjVg5KuyOYrJD2Q/+EBpYeaAXJH3QC5oWYqR2e+Wf+wpMslLTCz975yvkHSTZJmm9lVklZL+nz3DBEoOdQMkDvqBsgNNVMhOrMazHOSOrp095z8Dqd7tC7yq9uTpxJGZE6I+b5jj4v5h0NfjnlXxjeK+dc17d/u5t+Mivl30/xq5UEz/HTK6AY/PdIc/CpptXTiamiUpHKomY7Y6BExrzvb//fxh2m3xDy42ltlmkP7U5937fLNJl5NXOE/6pk1Mbd0YtMKlJ9U1s0+3+Cu30r/LL/RPPhg91aoTgw/sSqGMl1bV6Wql69QVjXEV2HK9Pc2y7Xn+uouX7zwqZivGvBizN9Z+6mYG//dWxTC6sRKGmw4VjLSWDPtNh1avynG+xb4fOtPzvBW4z89wvO/T7sg5rErfYUxbdoSoy3zY8XUv/XP8MufOL7dOAbv8c9xs3fBaO2V3s785bOejvkz/Xze97PNZ8X8x7t83CNu9rG27i3s5mFFbLYDAAAA8EGYrAMAAAAp1anVYMpV7SrffKX6Db/it+kMP42TPKU/Z+pv2j3+xsG+Gsyn+vpmE9PqamNOtgG8dcBbCOo2VfR/epSoPRN81+qRx/smF0MTddIafAWNNS2JJTQk/eQXn4151N2rYm5Z3/2bSgC5Sq7y0Odp36DrmV1TYh5f7wtt7B7urS99RwyPOez21sjDceB43+Bl5UV+fDlhuq9I8Y+N98Z8Vk9/vZt3+HFq0/f9eeqe9OXTMs3e7gnkS+tuX+ll6ve8leXb//K5mG879lcxj/uS19LfTPnTmPs8NSnm+u3e3tJa46+1ZXr7VrMzTngj5u82+uouE2p87vXcfl816UuvXhlz/18mNnN64PmY/chWeHyzDgAAAKQUk3UAAAAgpSq6F6N1s7fBDH3ZW1S+uvpjMd859skOH3/DYL8yuEfiP2Wy9eWe3b5qwE1z/iTmKT97y8eRy6CBAkuuRLF5mn/On516R+Je3gazO/hOeBe/8rV2zzXiiW0xt7y9QUCp+3SDt8fsveahmB/73FExb9s3pEuvcfmYOTFf0Ns3AhxW7fW4K+Ptm7fsnBrzj5/x49nUuT7W1hY25kM3S6yC1LJ6XczDv+sb6l3+Q28/ueOE22J+/vSfx7z3w4nWl8RiRcnFl+qt/aI4vcxb0urMW8du3Oyruzzw69NjHnOnj691vc/P0rI2Et+sAwAAACnFZB0AAABIqYpugwkH/Ar4hnmrY37r3/xK/7/8S7+q/opBf2j3+GNq/bTLosTV9P9z7SdjXvqgX8U89dd+mqXlna2HO2yg2yVbX1Z9xzeb+M4lvuLEgKqeMW9v9Q0ifrb9xJiH/pNfbS9JWuEblLH5CkpJSGyQ9NAdp8W84/NeB98Y8lTMF49fGHNzFz/qyXaXOvPX29Lqqy19f9O5MT91v9fgUXeuj7llp69uQ/2hoBItMZnlPt8a+YOJMX/po9+Oedcx3k45Y9LKmL823Dcyuu71S2IOoeONJsN83whp+Fx/3jGLfTOwlk3eFt3VTcy6A9+sAwAAACnFZB0AAABIqYpug0meBmzZ7Av2D3rUb39tu185/Lnz/dSiJI2evCnmDS82xjz8j34KZfTCROvLqjVdHDBQGOFo3zylfpqv4HJZH/8MVyeusJ/X5KcZ77rr7JhHv7qg3fNm9rXfJAkoFZn9fvp89D3eWrJg3XEx3/DnvmnYz8bNjnl8TUOXXvu6DdNjfmKNt1a2zu8f89BXfHWXMW/6SkstK73lAEiDkGgbtte8NXLUpmExt/7eP9ubB42N+X8M9M9/47rE8eQDurpq3l4bcybR7tKyf//B7p5KfLMOAAAApBSTdQAAACClKrsNJilx9W/rps0x93zYWwAmvT253UP2jRga8/g3vCUmsyrR+pI43QOUirXn9on5qvGPxNyryltfXmzy0+7XvfqlmMfcvTHm1j2+SowkVqBA6Upu8JJoLem3xY8R77b4Rkh/MuFvYm6t69pL93srE/PQDV539Ut9HC3rvDUnfWtZAAcXWnwzr+RnWIlck7h/MndWy6Hvknp8sw4AAACkFJN1AAAAIKVogzmE5CkavbSo3e/qX/LMaUeUk/1H+1X2Fzb45i57M/6/jF9tOSPm/nf7ahety/z+QLnL7NoVc++753ouwGuXw+l9AIfGN+sAAABASjFZBwAAAFKKNhgA/0XY4au+rG3xDY/ebvVVKX73yodinvr4sphpCQMAIH/4Zh0AAABIqUNO1s2s3sxeNLPXzGyRmX0ve/tYM5trZsvN7D/NEnuPAxWOugFyQ80AuaFmKkdn2mCaJJ0dQthtZjWSnjOzhyV9W9JPQgizzOz/SrpK0i+6caxAKSnpujnicd+86Nphl8acyVjMox/03Lpla2EGhnJW0jUDFAE1UyEO+c16aLM7+2NN9p8g6WxJd2dvnynp090yQqAEUTdAbqgZIDfUTOXoVM+6mVWb2auSNkuaI+ktSTtCCO8t87pO0sgOHnu1mc03s/nNasrHmIGScLh1Q82gUnGsAXJDzVSGTq0GE0JolXS8mfWXdJ+kKZ19gRDCzZJulqS+NjAc4u5A2TjcuklDzfS8/8WYR99fjBGgEnGsAXJDzVSGnFaDCSHskPSkpFMl9Tez9yb7R0han+exAWWBugFyQ80AuaFmyltnVoMZkv2LTWbWU9J5kpao7UPx2ezdrpD0QHcNEig11A2QG2oGyA01Uzk60wbTKGmmmVWrbXI/O4TwWzNbLGmWmf1A0iuSbj3UE02aNk5z5v+6SwNG15jZoe+EfMhL3VAzxUfNFAzHmjJC3RQENVNGPqhmLITCtSmZ2TuS9kjaUrAXTYfBSs97PjKEMKTYg0DnZGtmtdL1GSqENL1faqbEcKxJBeqmhHCsSYUOa6agk3VJMrP5IYTpBX3RIqvE94z8qrTPUKW9X+RfJX6GKvE9I78q7TNUKu83pwtMAQAAABQOk3UAAAAgpYoxWb+5CK9ZbJX4npFflfYZqrT3i/yrxM9QJb5n5FelfYZK4v0WvGcdAAAAQOfQBgMAAACkVEEn62Z2gZktNbPlZnZ9IV+7EMxslJk9aWaLzWyRmV2bvX2gmc0xs2XZfw8o9lhRGsq9ZiTqBvlX7nVDzSDfyr1mpNKum4K1wWQX7X9TbTtsrZM0T9JlIYTFBRlAAZhZo6TGEMLLZtZH0kuSPi3pzyRtCyHclC2CASGE7xZxqCgBlVAzEnWD/KqEuqFmkE+VUDNSaddNIb9ZP1nS8hDCihDCAUmzJF1UwNfvdiGEDSGEl7N5l9q2/R2ptvc5M3u3mWr7cACHUvY1I1E3yLuyrxtqBnlW9jUjlXbdFHKyPlLS2sTP67K3lSUzGyPpBElzJQ0LIWzI/mqjpGFFGhZKS0XVjETdIC8qqm6oGeRBRdWMVHp1wwWm3cDMGiTdI+m6EMK7yd+Ftr4jluAB3oe6AXJDzQC5K8W6KeRkfb2kUYmfj8jeVlbMrEZtH4I7Qgj3Zm/elO2Veq9nanOxxoeSUhE1I1E3yKuKqBtqBnlUETUjlW7dFHKyPk/SRDMba2a1ki6V9GABX7/bmZlJulXSkhDCjxO/elDSFdl8haQHCj02lKSyrxmJukHelX3dUDPIs7KvGam066agmyKZ2ccl/bOkakm3hRD+V8FevADM7HRJz0paICmTvfkGtfVEzZY0WtJqSZ8PIWwryiBRUsq9ZiTqBvlX7nVDzSDfyr1mpNKuG3YwBQAAAFKKC0wBAACAlGKyDgAAAKQUk3UAAAAgpZisAwAAACnFZB0AAABIKSbrnWRmq8xsgZm9ambziz0eIO3M7FozW2hmi8zsumKPBygFZnaBmS01s+Vmdn2xxwOknZn1N7O7zewNM1tiZqcWe0z5xtKNnWRmqyRNDyFsKfZYgLQzs2MkzZJ0sqQDkh6R9PUQwvKiDgxIMTOrlvSmpPMkrVPbZjWXhRAWF3VgQIqZ2UxJz4YQbslu6tQrhLCj2OPKJ75ZB9AdpkqaG0LYG0JokfS0pIuLPCYg7U6WtDyEsCKEcEBtf/BeVOQxAallZv0knam2nUkVQjhQbhN1icl6LoKkx8zsJTO7utiDAVJuoaQzzGyQmfWS9HFJo4o8JiDtRkpam/h5XfY2AAc3VtI7kn5pZq+Y2S1m1rvYg8o3Juudd3oI4URJF0q6xszOLPaAgLQKISyR9CNJj6mtBeZVSa1FHRQAoNz0kHSipF+EEE6QtEdS2V3rwWS9k0II67P/3izpPrWdrgTQgRDCrSGEaSGEMyVtV1svLoCOrVf7M1BHZG8DcHDrJK0LIczN/ny32ibvZYXJeieYWW8z6/NelnS+2k7zA+iAmQ3N/nu02vrV7yzuiIDUmydpopmNzV4od6mkB4s8JiC1QggbJa01s8nZm86RVHYXZPco9gBKxDBJ95mZ1Pbf7M4QwiPFHRKQeveY2SBJzZKuKceLfoB8CiG0mNk3JT0qqVrSbSGERUUeFpB235J0R/YP3BWSrizyePKOpRsBAACAlKINBgAAAEgpJusAAABASjFZBwAAAFKKyToAAACQUkzWAQAAgJRisg4AAACkFJN1AAAAIKWYrAMAAAApxWQdAAAASCkm6wAAAEBKMVkHAAAAUorJOgAAAJBSXZqsm9kFZrbUzJab2fX5GhRQzqgbIDfUDJA76qZ8WAjh8B5oVi3pTUnnSVonaZ6ky0IIi/M3PKC8UDdAbqgZIHfUTXnYujQOAAAeLklEQVTp0YXHnixpeQhhhSSZ2SxJF0nq8INQa3WhXr278JLoql3aviWEMKTY46hgOdUNNVN81EzRcawpQdRN0XGsKTEfVDNdmayPlLQ28fM6STPefyczu1rS1ZJUr16aYed04SXRVY+Hu1cXewwV7pB1Q82kCzVTdBxrShB1U3Qca0rMB9VMt19gGkK4OYQwPYQwvUZ13f1yQMmjZoDcUTdAbqiZ0tGVyfp6SaMSPx+RvQ1Ax6gbIDfUDJA76qaMdGWyPk/SRDMba2a1ki6V9GB+hgWULeoGyA01A+SOuikjh92zHkJoMbNvSnpUUrWk20IIi/I2MqAMUTdAbqgZIHfUTXnpygWmCiE8JOmhPI0FqAjUDZAbagbIHXVTPtjBFAAAAEgpJusAAABASjFZBwAAAFKKyToAAACQUkzWAQAAgJTq0mowAADg8FT37xfzntMnx/zO8e0PzXXbPDc+tiHm1uUru29wQEpU9eoVs40aEfOeSQNj3jmmfc00DfAcqg/+vD32ea7fGmLutbk15j5zV8fcsmlz4kn9/oXAN+sAAABASjFZBwAAAFKKNhgAh6fKzy1W9az3PNhPTbY0Dmj3kHfH+enM1lqLuW5nJub6bQdirtmyN+bwlp+OzOzff7ijBtJj6OAYN3yxKea5p/9zu7v9Zs/omP913+diHrxpS8yZXbu6Y4RA3lVPGBtz68CGmA8M9ONIS0//Lnl/f8/bj/H2k1Ef8pawm8Y+0u41zuq5O+YlB/z48vs9U2O+f/1xMa9bNtTHscKnxg2L/JilzYnvt4O3yhQC36wDAAAAKcVkHQAAAEgp2mAAfDDzdpXqPn1iDqP9qvzdk3xVi22TvT2m5Xg/FSlJv57hp/ePrqmN+d93jor5/62ZEfPKhX5qcsKsGn+i+Qs7PXwgrTJ9/LT/F4+aF3Ovqpp29/tyX293+fsT/ZT+4Hkj/U4L3+iGEQL5t+rSxphbj/VjxOlHLon56Ia3Y26o9rbHafWrYt7a2jvm/aF9zfxix8SYf/rqWTH3e8ZrbthzvszSxEVzDzrWwja7dIxv1gEAAICUYrIOAAAApBRtMO9JnOq3Wj89n1yMX7XtT7N0qKUlxrDPT99k9vvV/gp+KjP5euHAgcR9CrvoPnAwVT17xrz7rCkxb/3ynph/eeJPYz7OP87/RXPiI90UvE6u7LfK87GeHx3v7TXX1V8e88T5hxw2kHqtDV4sNw5ZnPhNx8eahlHvxtw03NsAaugMQ4m45HNPxZz83O/N+Pzn4b2+UtJPV3sby48WfSrmhtX+fXPd9vbzpcG/fTPm8VteOeg40tLi0hl8sw4AAACkFJN1AAAAIKVog8mqHuQbuew8268irvnqxpj/aeLsdo/pX+WnbFrlbTT/tOncmJ+774SYj/y1L+Bvzd4CsONkX1Wj3+N+6qZ1p5/uVKaUTtignOz7yNEx77zSN155ctotMTdY8rS9qSP/sHVazJ/o+2rMHbXOnN3Tr9b/zKm+WgZn/FGpah7uH3Pt0y/FTNMkSsXMV06N+WNnLoh5f/CVWq6//4sxT/zfvtLRpAO+MpJafV4U3tc23JpsKS4DfLMOAAAApBSTdQAAACClKq8Npso3bKkef2TMaz89POaLv/R0zF/u7wvlj+7hq2JIUo01xNyaWN3l74fP8fwZv/9jE72dYMiwnTFfPuaRmG8f+PGYh92TaInZsvVg7wboFj3Gem0s+4T/Tf9/j7k75n5VB+9d2Zm4ov/Up77Z7nejZvn/cu4648yYv/7JR2O+ZsDSmGvM67WhR2I1pQ9YLQMoZ1XJBcOay+tUPyrDpJ/55/ZvR14c8w/G3xdzywBvFbY6P9a0bt/ezaNLp0N+s25mt5nZZjNbmLhtoJnNMbNl2X8P6N5hAqWFugFyQ80AuaNuKkNn2mBul3TB+267XtITIYSJkp7I/gzA3S7qBsjF7aJmgFzdLuqm7B2yDSaE8IyZjXnfzRdJ+mg2z5T0lKTv5nFc3abHKF95ZeUl3vpy9WUPxXxJH19rYlmLt7qc++jX2j1X1V4/Ra9MIvb2K5St2f8esj7NMf9g8v0xj6nZEfMvzdtgULpKsW6SrS+rLxkZ8ydP8VVYZtT7CkXbElfi/3KHr/Jyx13nxDz+Kd84SZKqX/cNMCas9Ne4Y+XHYp590Ykx33X07TGf38dXDfjVz78e85TvLok5s8tXq0FpKcWa6aqaLXtjPv11bwd45Jg7292vocpXydhymrcHDFpwTMxhPmskVaKSrJvXvcV347N+7Hi+0Vfi6z3IayMM6OuP3bipe8eWUod7gemwEMJ76xBulDQsT+MByhl1A+SGmgFyR92UmS6vBhPaFrfscIlXM7vazOab2fxmNXV0N6CifFDdUDPAf8WxBsgdx5rycLirwWwys8YQwgYza5S0uaM7hhBulnSzJPW1gUXZt6HHcP+jcs3nRsV8/mdejPmiBj+F+I/v+CoVDz4xI+bJv97d7nmr9iQ+3IkF+ZsH9Y5542m9Yh7/iVUxD6325/rGssv89vl+Gj/sbt9CgJLXqbopVs3smTo05oFn+QZe3xjsqyPVJzY/enS/t5H9+3Mfifmo21fE3LLpnXavkUlu7rVkWYzDNvtqR2/XTo7558PPiPnvhj4f80/P/1XMNyz7Sswj71wec+umDv+3hNJRUseaXGXq/RB85jBv50qugvR+gxt9JbGmwf1i7mBfMVSmVB9rQpPPnXpt8pfd2OSf58Z+3nK5Z3xjzL0zEzr3Ihv8Lbcm51IlusHk4X6z/qCkK7L5CkkP5Gc4QFmjboDcUDNA7qibMtOZpRvvkvSCpMlmts7MrpJ0k6TzzGyZpHOzPwPIom6A3FAzQO6om8rQmdVgLuvgV+d0cHvq7DhjTMyjEq0oXxj4Qsw3vu2rsLz0oF9hP/HX3g7QumJNu+dt7eB0Ss2EsTHvG+pX8f/30b+JeX2rn+555+EjYh658OWYM/v3H/T5kX6lWDdbjvUWlz8/wj+Hk2r8M/xms38m/2WFv5Vx93gttGzY2LkXTLSOJTf9anx6W8z3HeltaJd/1uv19HrfGGPkp1b5U84Z6M9PG0xJKcWa6arWBm9e+eGw1xO/YdMvdE4p1k1VL28P3uXTJZ3YsDrmY3uvi/lHV/hqYft3dG7J+H4LvK1z8AI/btUu9uctpVbJLl9gCgAAAKB7MFkHAAAAUupwV4NJPZvurSwbLzoQ842j5sR8z46TYn75fr//mDu93aVlrZ8y+eAXtBj3TB0S8/Bj/DTLuB6+mcV/bPPXa3w2sQJMi98H6G5V9d7isu/YfTEfW7825kxi1a+Hd/vndvuzvhrMqCd8pZauyrz+hj/v416j/3qGn9X9ycgnYv7IYF9V5sl6b5sBAKRD8ljz7ieOjfnEM5Z6rvPjTlPwFZG+d9yDMa8+MDjm6sSx6ZOJTfMk6d9O9JXEfrPMj1s1r/tqMsNe9NUB61/yVcySm+ulZU7GN+sAAABASjFZBwAAAFKqbNtgln7Frzb+xSkzYz6pzjeU+Iu5J8Q8efb6mA+n9aXHyBExrzvb/wb6PxMeinleYsH/B148MeZJL/rmTCWxkwdKW5WfXrRR/rmdMXZVzBNqfEOK7Rn/nM9e45/b4XO7f8e7+o2+mcXjC6b6LxJtMHVVzTFnahLvLfE+S3UjDAAoWR0ca0Zd562LN4z0OdItW0+P+YGlH/KnWdUz5r4r/elba/zYdPv57Vsgxw3wVcVuOP6RmGecsirmvzznkpi3zpwS84Alvmll1aK3Ys7s81bR5GpmhcA36wAAAEBKMVkHAAAAUqps22CGHOmbpoyr8dMhOzJ+6iKzpS7mlpW+GH+HkqfVJVX3bYh51eVHxvzls5+K+fyefhr/62s/EvP42em4whiVx6r9c7zzBN844tJBz8Q8pNpr45ad4/z+f/T7D5y/KOZCN5lUJb5nmFLnG5fd0+grDjT09la45NX9QClravH67dOSKeJIgA+WPNbsH+sb1p3WZ3HMP1zvG1Ku+ddJMY+f9cfcXuzn7X88MMRX5fv5J/405u+f4xskfevEJ2M++3/+Oua/X/MnMb97o7fH1Mz1lcoK3RLDN+sAAABASjFZBwAAAFKqbNtg8ibR+lI9YUy7X638wrCYv/LZR2O+pv+SmDe1+oZM8zf6AvwjF/ri/6xTgUKy2pqYt031v9eH9jh4q8hPXjk35gkP+31ad757sLsXRI15XZ5e7ys87Rjvt/ed56sv0QaDclHzcP+Ya59+KWZWEkPahGaf/9Q++XrMz5890u/U6q1cffe+4o/t4mu3btkS88C7/Fg1+F5v8bz/w+fF/C+f9ePct2b8PuZeP/OWnVn/zVt2ej6TaIkpwPGFb9YBAACAlGKyDgAAAKQUk3UAAAAgpcq2Z/2dt72vb+2UvjHPqPOlFP/6vN/FPPN3pxz0eaqrvJ9qyoDN7X53zWB//Iy6rTHXme+29fu9vmvX3jd8TK1b3/zgNwAUQEsv7wysN98JNLk0YtVqXw7RFr0WcyjwDm5JyZ71pEzy/2hVdtD7AKWsytuA2/UEA2mW/Ky2btn6AffM1wv68Sk0+W7brYnc8xnvR5+6yudqd5z5sZi//e3ZMc/4wbyYH73ttJhH3r8m5pa167oy6g7xzToAAACQUkzWAQAAgJQq2zaYCb/yHUK/2f8LMd8ybWbMX+izNObzjvFleDrS/31/2nz42WtiHj3Ud0m9Yay3x8zeeFLMI/6QWKQxw4KNKA6r8bLvMW53zP2rfEe2TOLv+Or93k6S2bu3m0fXOa3B29My8mysXwcA6ITMHm+L1tIVMQ5r9vnj9478fMxf+9RjMZ/0JW8JfXbwcTGPvcd3ts+8fuh5ZWfxzToAAACQUkzWAQAAgJQq2zaY6hcWxDy81/ExX7nMW1dqp/iuVo39PO9r9h0eN2zxXRBrVviqGJJUlfhTZ8iR3k7QHPw/6+LVjTFPXeSryfhJFqDAzD+4g/r4acA6S7Zmpfvv+JbEvr87M77KQI9kl04LrWYAgE5ItCZn1r4d84Q7amP+jy2+SsyFlz8f8yWffjrm3607M+ZBvmlrlx3yiGxmo8zsSTNbbGaLzOza7O0DzWyOmS3L/ntA/oYFlC5qBsgddQPkhpqpHJ35+qxF0l+HEI6SdIqka8zsKEnXS3oihDBR0hPZnwFQM8DhoG6A3FAzFeKQbTAhhA2SNmTzLjNbImmkpIskfTR7t5mSnpL03W4Z5WEILd5oUvvo/JgnLhoZc9OEYTHvH+gL4lc3+5ISEzckzqu/6q01krRm1pSYbxj5UMwLmvw1alfXxZzZ9E6nx4/SVao1k0bN/b317MjRW2LeH7y+r1vzyZj7L/eNnTLv7urm0SGfKrFu7ICvZPTEPt/o68z69psdJTcB2znJbx86bkzMLStW5X18SLdKrJlCaLeJ0mLfwPKI9b7B5gOnHBvzrJNuiXn2yI/EPCiPY8qpZ93Mxkg6QdJcScOyHxRJ2ihpWAePuVrS1ZJUr16HO06gJFEzQO6oGyA31Ex56/RVZGbWIOkeSdeFEN5N/i607Tt+0BWOQwg3hxCmhxCm16juYHcByhI1A+SOugFyQ82Uv059s25mNWr7INwRQrg3e/MmM2sMIWwws0ZJmzt+hvRoWbc+5upE7uhvymC+IUxVQ0O731055YWYJ9f4acobVp0c88DFXiPtFuBHWUtzzSRbxN5ePiTmFRMGxjyqx46Y94/w+1dPGh9z65tvddcQo31DfGWma8c8GfPKZv+eYcmdU2NunLcs5tZdtMGUmjTXTXeo3udtW7O3+nHj1BFPt7tfsg2m37FbY947yeu3ljaYilRpNVNUiZXUmpt8+rymxa/ftWZ1i86sBmOSbpW0JITw48SvHpR0RTZfIemB/A8PKD3UDJA76gbIDTVTOTrzzfqHJV0uaYGZvZq97QZJN0mabWZXSVot6fMdPB6oNNQMkDvqBsgNNVMhOrMazHOSrINfn5Pf4aSP9fDT8AdOntTud0N6LI15SeKq/qUvjol50h+91YaNkCpD2msms8dXOJp0q2/m9Q+TLoh5zKS7Yv7aaX5K/s6VPvxR/+Kf7cz+/XkbX/Ugb8fZ0+in/4f32Bnzb3cdF/OIB1bH3LJ1W97GgcJKe93ki9X4Jist/bxPeESdf76rraP/DIAr1ZrpMTxxvWu1/z8+s8NroKhtw4n6q+7TJ+bdH5kY85XHPRPz0+/6yoC93z7o5QFdlu5tCgEAAIAKxmQdAAAASKmc1lmvGMkVYPr6CjBrrmptd7eT6v30+482fCzmIS/5aZCWVWu6Y4TA4cv45zi8sijmpttPifnfvn1mzH837KmY75wx3R871VeGUeJ5DkuVnwrd/jFvNzvqkiUx1ycus799wakxT9q9wp8n075GgbSpHjk85rUf7hnzjUMWJ+5VI6CcWJ23fK35sh87MolZ6BG/97ZM/fH1PL64z+mSrc1Wnfi+OtGOU9XXW1/2He2bXB642tssp/daGfN/+7erYh79sK+Sls/WZ75ZBwAAAFKKyToAAACQUrTBHERVL98iqen4sTH/00l3tbvfhBr/z/fcUr9KeOLqvQJKzYBX/RTf756dFvMxF/iqL8cP97z8aL8Cvt8rXXvtqmO99WXjGb6y0o8afSOkO7d568vYXyQ3G9vXtRcHCmjPUb4SxlmfeamIIwG6V7L1ZeefnhDzdVfeG/M9G0+Medcbo2LuaKPKw1E90Dct2nWmz9V2jvXWl31D/Zgy9VRvcfnLEf8Z8/pmf55rZ30l5vG/fSfmlk3ds/8U36wDAAAAKcVkHQAAAEgp2mAOwhJtMFuO9dM4x9a2P73RI3GiJjT53z3WnFhtozsGCHSDzFu+utGEu/xz/w/Dzo/5pul++nLm1+pjXtv3tJiH/vz5Dl+jx7gxMe8+aqg//rNeM/942qyYVx0YHPNvnvaVaCa97CsFZFp8lRggjXqMGR3zmpN9NYq7hj2RuFeDOjJ+9tdjHndfU8w1r70ZM+sgIc1aa3xFlpE122O+dfzsmO/+/tEx3/JlP6bseatfzJk6n1XVD/eNk/bt9OORJF34oYUxT+3lK7QMq/HWs0HVvvrMroyvzPTC7gkx/81LF/tj/9PvM+EVbwlt3bDJXziwKRIAAABQUZisAwAAAClFG8xBWF1tzMkrhOut/f2e3e//+fq/5qc2q1a8HTOnJlEqQpOfXq9a6KcNR902Nebv7Lg05otnzI+58c92xvzYh06Oeegf/Gp7Sdp2oa/ccvo43/Dobwe/GHOrvND+7rWLYj7yEd9iIrOXFZdQOkKtHx+ae/sxZUfGvy87f+4XYq592E/7S9Lk57b6c63wjfZa9+/P6ziBfArN/v/sIc/4vOiv/p9vItTnJF9JZWhvb0uZPnxtzDMmPx1zsnVleZOvrLRgl29eJElbmnrHfP/O42JescIfU7vJ53A1e/y402+Fr0g2do0fa6oW+cZlLbsTGzh1U+tLEt+sAwAAACnFZB0AAABIKdpgDiL09dMnw6dtjLne2v9t8/2Vn4x50GI/Hdm6fbuAUpZsM6l7blHME/dOjvmxVafEvOcob6E547g3Yt5/jJ/+l6RBzYlNMg741fvfef2zMbcs7Bvz8Ll+GrV+3vKYaS9DSXnHNxwbPcc3Vvn8O9+JecjLXkO1T7ffLKm1+UA3Dg7oJhn/P3XLSl9tbNyvvOVk7x+HxLyjYWDMG/v5fOuF4R/y50xMw2q9+1LV+9u3otS96z/X7PG2lsnrfAWZ6m3v+gOavMZat/ocLiRqz5+l8PhmHQAAAEgpJusAAABAStEGcxCh1v+zTB3gVyTXvK8NZsVq39Rlyg5vGyjmqRIg3zKJFSeq5vkKLqOW+YoVLRNGxPzSR4+Jee+Y9hsWJVdNqt3tpymHrfNTjfVLVvrzbvKVAlozNL+gNCVbI2se81WURjx28PuzmR7KWcuKVTHXJnPiPr0TeVAeXztZWy0d3it9+GYdAAAASCkm6wAAAEBK0QZzELbHT/s/vnRKzK8N/kO7+1Xv8P98dqCUTqgAhyd5ZXzrO96iYol8xAtdew0qCQAAd8hv1s2s3sxeNLPXzGyRmX0ve/tYM5trZsvN7D/NrPZQzwVUCuoGyA01A+SGmqkcnWmDaZJ0dgjhOEnHS7rAzE6R9CNJPwkhTJC0XdJVH/AcQKWhboDcUDNAbqiZCnHINpgQQpC0O/tjTfafIOlsSV/I3j5T0t9L+kX+h1gE232l/YFPDov5q/VXtLvbkPmJH7bu6O5RoYRUZN0AXUDNALmhZipHpy4wNbNqM3tV0mZJcyS9JWlHCOG99tJ1kkZ28NirzWy+mc1vVtPB7gKUpcOtG2oGlYpjDZAbaqYydGqyHkJoDSEcL+kISSdLmnKIhyQfe3MIYXoIYXqN6g79AKBMHG7dUDOoVBxrgNxQM5Uhp9VgQgg7zOxJSadK6m9mPbJ/vR0haX13DLAYWrdsjXngbS8k8gc8pjsHhJJWKXUD5As1A+SGmilvnVkNZoiZ9c/mnpLOk7RE0pOSPpu92xWSHuiuQQKlhroBckPNALmhZipHZ75Zb5Q008yq1Ta5nx1C+K2ZLZY0y8x+IOkVSbd24ziBUkPdALmhZoDcUDMVwtouJi7Qi5m9I2mPpC0Fe9F0GKz0vOcjQwhDij0IdE62ZlYrXZ+hQkjT+6VmSgzHmlSgbkoIx5pU6LBmCjpZlyQzmx9CmF7QFy2ySnzPyK9K+wxV2vtF/lXiZ6gS3zPyq9I+Q6Xyfju1GgwAAACAwmOyDgAAAKRUMSbrNxfhNYutEt8z8qvSPkOV9n6Rf5X4GarE94z8qrTPUEm834L3rAMAAADoHNpgAAAAgJRisg4AAACkVEEn62Z2gZktNbPlZnZ9IV+7EMxslJk9aWaLzWyRmV2bvX2gmc0xs2XZfw8o9lhRGsq9ZiTqBvlX7nVDzSDfyr1mpNKum4L1rGd32HpTbdvhrpM0T9JlIYTFBRlAAZhZo6TGEMLLZtZH0kuSPi3pzyRtCyHclC2CASGE7xZxqCgBlVAzEnWD/KqEuqFmkE+VUDNSaddNIb9ZP1nS8hDCihDCAUmzJF1UwNfvdiGEDSGEl7N5l6Qlkkaq7X3OzN5tpto+HMChlH3NSNQN8q7s64aaQZ6Vfc1IpV03hZysj5S0NvHzuuxtZcnMxkg6QdJcScNCCBuyv9ooaViRhoXSUlE1I1E3yIuKqhtqBnlQUTUjlV7dcIFpNzCzBkn3SLouhPBu8nehre+I9TKB96FugNxQM0DuSrFuCjlZXy9pVOLnI7K3lRUzq1Hbh+COEMK92Zs3ZXul3uuZ2lys8aGkVETNSNQN8qoi6oaaQR5VRM1IpVs3hZysz5M00czGmlmtpEslPVjA1+92ZmaSbpW0JITw48SvHpR0RTZfIemBQo8NJansa0aibpB3ZV831AzyrOxrRirtuinoDqZm9nFJ/yypWtJtIYT/VbAXLwAzO13Ss5IWSMpkb75BbT1RsyWNlrRa0udDCNuKMkiUlHKvGYm6Qf6Ve91QM8i3cq8ZqbTrpqCTdQAAAACdxwWmAAAAQEoxWQcAAABSisk6AAAAkFJM1gEAAICUYrIOAAAApBST9U4ys2vNbKGZLTKz64o9HiDtzOwCM1tqZsvN7PpijwdIOzO7zcw2m9nCYo8FKBVmtsrMFpjZq2Y2v9jj6Q5M1jvBzI6R9DVJJ0s6TtInzWxCcUcFpJeZVUv6maQLJR0l6TIzO6q4owJS73ZJFxR7EEAJOiuEcHwIYXqxB9IdmKx3zlRJc0MIe0MILZKelnRxkccEpNnJkpaHEFaEEA5ImiXpoiKPCUi1EMIzklK1GQuA4mOy3jkLJZ1hZoPMrJekj0saVeQxAWk2UtLaxM/rsrcBAJBPQdJjZvaSmV1d7MF0hx7FHkApCCEsMbMfSXpM0h5Jr0pqLe6oAAAAKt7pIYT1ZjZU0hwzeyN7lqps8M16J4UQbg0hTAshnClpu6Q3iz0mIMXWq/3ZpyOytwEAkDchhPXZf2+WdJ/a2jDLCpP1Tsr+xSYzG622fvU7izsiINXmSZpoZmPNrFbSpZIeLPKYAABlxMx6m1mf97Kk89XWulxWaIPpvHvMbJCkZknXhBB2FHtAQFqFEFrM7JuSHpVULem2EMKiIg8LSDUzu0vSRyUNNrN1km4MIdxa3FEBqTZM0n1mJrXNae8MITxS3CHln4UQij0GAAAAAAdBGwwAAACQUkzWAQAAgJRisg4AAACkFJN1AAAAIKWYrAMAAAApxWQdAAAASCkm6wAAAEBK/X+i3BBDhqQO5AAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 864x576 with 16 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "batch_samples,labels = next(iter(test_dataloader))\n",
    "print(batch_samples.shape,labels.shape)\n",
    "show_batch(batch_samples.squeeze(), labels.numpy(), (4,4))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def conv3x3(inplanes, outplanes, stride = 1):\n",
    "    return torch.nn.Conv2d(inplanes, \n",
    "                           outplanes, \n",
    "                           stride=stride,\n",
    "                           kernel_size=3, \n",
    "                           padding = 1,\n",
    "                           bias=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "class BasicBlock(torch.nn.Module):\n",
    "    expansion = 1\n",
    "    def __init__(self, inplanes, \n",
    "                 outplanes, \n",
    "                 stride = 1, \n",
    "                 downsample = None):\n",
    "        \n",
    "        super(BasicBlock, self).__init__()\n",
    "        \n",
    "        self.seq = torch.nn.Sequential(\n",
    "            conv3x3(inplanes,outplanes,stride),\n",
    "            torch.nn.BatchNorm2d(outplanes),\n",
    "            torch.nn.ReLU(inplace = True),\n",
    "            \n",
    "            \n",
    "            conv3x3(outplanes,outplanes),\n",
    "            torch.nn.BatchNorm2d(outplanes),      \n",
    "        )\n",
    "                \n",
    "        self.downsample = downsample\n",
    "        self.stride = stride\n",
    "        \n",
    "        \n",
    "    def forward(self, x):\n",
    "        residule = x\n",
    "        \n",
    "        out = self.seq(x)\n",
    "        \n",
    "        if self.downsample is not None:\n",
    "            residule = self.downsample(x)\n",
    "        \n",
    "        out += residule\n",
    "        return F.relu(out)    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "----------------------------------------------------------------\n",
      "        Layer (type)               Output Shape         Param #\n",
      "================================================================\n",
      "            Conv2d-1           [-1, 64, 32, 32]          36,864\n",
      "       BatchNorm2d-2           [-1, 64, 32, 32]             128\n",
      "              ReLU-3           [-1, 64, 32, 32]               0\n",
      "            Conv2d-4           [-1, 64, 32, 32]          36,864\n",
      "       BatchNorm2d-5           [-1, 64, 32, 32]             128\n",
      "================================================================\n",
      "Total params: 73,984\n",
      "Trainable params: 73,984\n",
      "Non-trainable params: 0\n",
      "----------------------------------------------------------------\n",
      "Input size (MB): 0.25\n",
      "Forward/backward pass size (MB): 2.50\n",
      "Params size (MB): 0.28\n",
      "Estimated Total Size (MB): 3.03\n",
      "----------------------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "def test_BasicBlock():\n",
    "    model = BasicBlock(64,64).to(DEVICE)\n",
    "    summary(model, (64,32,32))\n",
    "test_BasicBlock()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Bottleneck(torch.nn.Module):\n",
    "    expansion = 4\n",
    "    def __init__(self, inplanes, outplanes, stride=1, downsample=None):\n",
    "        super(Bottleneck, self).__init__()        \n",
    "        self.seq = torch.nn.Sequential(\n",
    "            # [1x1]\n",
    "            torch.nn.Conv2d(inplanes, \n",
    "                            outplanes, \n",
    "                            kernel_size=1, \n",
    "                            bias=False),\n",
    "            torch.nn.BatchNorm2d(outplanes),\n",
    "            torch.nn.ReLU(inplace = True),\n",
    "            \n",
    "            # [3x3]\n",
    "            torch.nn.Conv2d(outplanes, \n",
    "                            outplanes, \n",
    "                            kernel_size=3, \n",
    "                            stride=stride,\n",
    "                            padding=1, \n",
    "                            bias=False),\n",
    "            torch.nn.BatchNorm2d(outplanes),\n",
    "            torch.nn.ReLU(inplace = True),\n",
    "            \n",
    "            # [1x1]\n",
    "            torch.nn.Conv2d(outplanes, \n",
    "                                   outplanes * self.expansion, \n",
    "                                   kernel_size=1, \n",
    "                                   bias=False),\n",
    "            torch.nn.BatchNorm2d(outplanes * self.expansion))\n",
    "        \n",
    "        self.downsample = downsample\n",
    "        self.stride = stride\n",
    "        \n",
    "    def forward(self, x):\n",
    "        residual = x\n",
    "        \n",
    "        out = self.seq(x)\n",
    "        if self.downsample is not None:\n",
    "            residual = self.downsample(x)\n",
    "        \n",
    "        out += residual\n",
    "        return F.relu(out)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "----------------------------------------------------------------\n",
      "        Layer (type)               Output Shape         Param #\n",
      "================================================================\n",
      "            Conv2d-1           [-1, 64, 28, 28]          16,384\n",
      "       BatchNorm2d-2           [-1, 64, 28, 28]             128\n",
      "            Conv2d-3           [-1, 64, 28, 28]          36,864\n",
      "       BatchNorm2d-4           [-1, 64, 28, 28]             128\n",
      "            Conv2d-5          [-1, 256, 28, 28]          16,384\n",
      "       BatchNorm2d-6          [-1, 256, 28, 28]             512\n",
      "================================================================\n",
      "Total params: 70,400\n",
      "Trainable params: 70,400\n",
      "Non-trainable params: 0\n",
      "----------------------------------------------------------------\n",
      "Input size (MB): 0.77\n",
      "Forward/backward pass size (MB): 4.59\n",
      "Params size (MB): 0.27\n",
      "Estimated Total Size (MB): 5.63\n",
      "----------------------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "def test_Bottleneck():\n",
    "    model = Bottleneck(256,64).to(DEVICE)\n",
    "    summary(model, (256,28,28))\n",
    "test_Bottleneck()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResNetZrc(torch.nn.Module):\n",
    "\n",
    "    def __init__(self, block, layers, grasacale = False, num_classes=1000):\n",
    "        super(ResNetZrc, self).__init__()\n",
    "        \n",
    "        self.inplanes = 64\n",
    "        \n",
    "        if grasacale:\n",
    "            init_channels = 1\n",
    "        else:\n",
    "            init_channels = 3\n",
    "        \n",
    "        \n",
    "        self.pre = torch.nn.Sequential(OrderedDict([\n",
    "            (\"conv0\", torch.nn.Conv2d(init_channels, \n",
    "                                      self.inplanes, \n",
    "                                      kernel_size=7, \n",
    "                                      stride=2, \n",
    "                                      padding=3,\n",
    "                                      bias=False)),\n",
    "            (\"bn0\", torch.nn.BatchNorm2d(self.inplanes)),\n",
    "            (\"relu0\", torch.nn.ReLU(inplace=True)),\n",
    "            (\"pool0\", torch.nn.MaxPool2d(kernel_size = 3, stride=2, padding=1))\n",
    "            \n",
    "        ]))\n",
    "        \n",
    "        \n",
    "        self.middle_layers = torch.nn.Sequential(\n",
    "            self._make_layers(block, 64,  layers[0]),\n",
    "            self._make_layers(block, 128, layers[1], stride=2),\n",
    "            self._make_layers(block, 256, layers[2], stride=2),\n",
    "            self._make_layers(block, 512, layers[3], stride=2)\n",
    "        )\n",
    "        \n",
    "        self.avgpool = torch.nn.AdaptiveAvgPool2d(1)\n",
    "        self.flatten = torch.nn.Flatten()\n",
    "        self.fc = torch.nn.Linear(512 * block.expansion, num_classes)\n",
    "        \n",
    "        \n",
    "        for name, layer in self.named_modules():\n",
    "            if isinstance(layer, torch.nn.Conv2d):\n",
    "                n = layer.kernel_size[0] * layer.kernel_size[1] * layer.out_channels\n",
    "                layer.weight.data.normal_(0, (2. / n)**.5)\n",
    "            elif isinstance(layer, torch.nn.BatchNorm2d):\n",
    "                layer.weight.data.fill_(1)\n",
    "                layer.bias.data.zero_()\n",
    "            \n",
    "    \n",
    "    def _make_layers(self, block, planes, multiplier, stride = 1):\n",
    "        downsample = None\n",
    "        if stride != 1 or self.inplanes != planes * block.expansion:\n",
    "            downsample = torch.nn.Sequential(\n",
    "                torch.nn.Conv2d(self.inplanes, planes * block.expansion,\n",
    "                          kernel_size=1, stride=stride, bias=False),\n",
    "                torch.nn.BatchNorm2d(planes * block.expansion),\n",
    "            )\n",
    "        layers = []\n",
    "        # 第一个 block  stride = 2 时， 需要 downsample \n",
    "        layers.append(block(self.inplanes, planes, stride, downsample))\n",
    "        self.inplanes = planes * block.expansion\n",
    "        \n",
    "        # 后面的  self.inplanes == planes * expansion\n",
    "        for i in range(1, multiplier):\n",
    "            layers.append(block(self.inplanes, planes))\n",
    "        \n",
    "        return torch.nn.Sequential(*layers)    \n",
    "    \n",
    "    def forward(self, x):\n",
    "        x = self.pre(x)\n",
    "        x = self.middle_layers(x)\n",
    "        x = self.avgpool(x)\n",
    "        x = self.flatten(x).squeeze()\n",
    "        logits = self.fc(x)\n",
    "        probas = F.softmax(logits, dim=1)\n",
    "        return logits, probas\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "----------------------------------------------------------------\n",
      "        Layer (type)               Output Shape         Param #\n",
      "================================================================\n",
      "            Conv2d-1           [-1, 64, 14, 14]           9,408\n",
      "       BatchNorm2d-2           [-1, 64, 14, 14]             128\n",
      "              ReLU-3           [-1, 64, 14, 14]               0\n",
      "         MaxPool2d-4             [-1, 64, 7, 7]               0\n",
      "            Conv2d-5             [-1, 64, 7, 7]          36,864\n",
      "       BatchNorm2d-6             [-1, 64, 7, 7]             128\n",
      "              ReLU-7             [-1, 64, 7, 7]               0\n",
      "            Conv2d-8             [-1, 64, 7, 7]          36,864\n",
      "       BatchNorm2d-9             [-1, 64, 7, 7]             128\n",
      "       BasicBlock-10             [-1, 64, 7, 7]               0\n",
      "           Conv2d-11             [-1, 64, 7, 7]          36,864\n",
      "      BatchNorm2d-12             [-1, 64, 7, 7]             128\n",
      "             ReLU-13             [-1, 64, 7, 7]               0\n",
      "           Conv2d-14             [-1, 64, 7, 7]          36,864\n",
      "      BatchNorm2d-15             [-1, 64, 7, 7]             128\n",
      "       BasicBlock-16             [-1, 64, 7, 7]               0\n",
      "           Conv2d-17             [-1, 64, 7, 7]          36,864\n",
      "      BatchNorm2d-18             [-1, 64, 7, 7]             128\n",
      "             ReLU-19             [-1, 64, 7, 7]               0\n",
      "           Conv2d-20             [-1, 64, 7, 7]          36,864\n",
      "      BatchNorm2d-21             [-1, 64, 7, 7]             128\n",
      "       BasicBlock-22             [-1, 64, 7, 7]               0\n",
      "           Conv2d-23            [-1, 128, 4, 4]          73,728\n",
      "      BatchNorm2d-24            [-1, 128, 4, 4]             256\n",
      "             ReLU-25            [-1, 128, 4, 4]               0\n",
      "           Conv2d-26            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-27            [-1, 128, 4, 4]             256\n",
      "           Conv2d-28            [-1, 128, 4, 4]           8,192\n",
      "      BatchNorm2d-29            [-1, 128, 4, 4]             256\n",
      "       BasicBlock-30            [-1, 128, 4, 4]               0\n",
      "           Conv2d-31            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-32            [-1, 128, 4, 4]             256\n",
      "             ReLU-33            [-1, 128, 4, 4]               0\n",
      "           Conv2d-34            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-35            [-1, 128, 4, 4]             256\n",
      "       BasicBlock-36            [-1, 128, 4, 4]               0\n",
      "           Conv2d-37            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-38            [-1, 128, 4, 4]             256\n",
      "             ReLU-39            [-1, 128, 4, 4]               0\n",
      "           Conv2d-40            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-41            [-1, 128, 4, 4]             256\n",
      "       BasicBlock-42            [-1, 128, 4, 4]               0\n",
      "           Conv2d-43            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-44            [-1, 128, 4, 4]             256\n",
      "             ReLU-45            [-1, 128, 4, 4]               0\n",
      "           Conv2d-46            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-47            [-1, 128, 4, 4]             256\n",
      "       BasicBlock-48            [-1, 128, 4, 4]               0\n",
      "           Conv2d-49            [-1, 256, 2, 2]         294,912\n",
      "      BatchNorm2d-50            [-1, 256, 2, 2]             512\n",
      "             ReLU-51            [-1, 256, 2, 2]               0\n",
      "           Conv2d-52            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-53            [-1, 256, 2, 2]             512\n",
      "           Conv2d-54            [-1, 256, 2, 2]          32,768\n",
      "      BatchNorm2d-55            [-1, 256, 2, 2]             512\n",
      "       BasicBlock-56            [-1, 256, 2, 2]               0\n",
      "           Conv2d-57            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-58            [-1, 256, 2, 2]             512\n",
      "             ReLU-59            [-1, 256, 2, 2]               0\n",
      "           Conv2d-60            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-61            [-1, 256, 2, 2]             512\n",
      "       BasicBlock-62            [-1, 256, 2, 2]               0\n",
      "           Conv2d-63            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-64            [-1, 256, 2, 2]             512\n",
      "             ReLU-65            [-1, 256, 2, 2]               0\n",
      "           Conv2d-66            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-67            [-1, 256, 2, 2]             512\n",
      "       BasicBlock-68            [-1, 256, 2, 2]               0\n",
      "           Conv2d-69            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-70            [-1, 256, 2, 2]             512\n",
      "             ReLU-71            [-1, 256, 2, 2]               0\n",
      "           Conv2d-72            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-73            [-1, 256, 2, 2]             512\n",
      "       BasicBlock-74            [-1, 256, 2, 2]               0\n",
      "           Conv2d-75            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-76            [-1, 256, 2, 2]             512\n",
      "             ReLU-77            [-1, 256, 2, 2]               0\n",
      "           Conv2d-78            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-79            [-1, 256, 2, 2]             512\n",
      "       BasicBlock-80            [-1, 256, 2, 2]               0\n",
      "           Conv2d-81            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-82            [-1, 256, 2, 2]             512\n",
      "             ReLU-83            [-1, 256, 2, 2]               0\n",
      "           Conv2d-84            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-85            [-1, 256, 2, 2]             512\n",
      "       BasicBlock-86            [-1, 256, 2, 2]               0\n",
      "           Conv2d-87            [-1, 512, 1, 1]       1,179,648\n",
      "      BatchNorm2d-88            [-1, 512, 1, 1]           1,024\n",
      "             ReLU-89            [-1, 512, 1, 1]               0\n",
      "           Conv2d-90            [-1, 512, 1, 1]       2,359,296\n",
      "      BatchNorm2d-91            [-1, 512, 1, 1]           1,024\n",
      "           Conv2d-92            [-1, 512, 1, 1]         131,072\n",
      "      BatchNorm2d-93            [-1, 512, 1, 1]           1,024\n",
      "       BasicBlock-94            [-1, 512, 1, 1]               0\n",
      "           Conv2d-95            [-1, 512, 1, 1]       2,359,296\n",
      "      BatchNorm2d-96            [-1, 512, 1, 1]           1,024\n",
      "             ReLU-97            [-1, 512, 1, 1]               0\n",
      "           Conv2d-98            [-1, 512, 1, 1]       2,359,296\n",
      "      BatchNorm2d-99            [-1, 512, 1, 1]           1,024\n",
      "      BasicBlock-100            [-1, 512, 1, 1]               0\n",
      "          Conv2d-101            [-1, 512, 1, 1]       2,359,296\n",
      "     BatchNorm2d-102            [-1, 512, 1, 1]           1,024\n",
      "            ReLU-103            [-1, 512, 1, 1]               0\n",
      "          Conv2d-104            [-1, 512, 1, 1]       2,359,296\n",
      "     BatchNorm2d-105            [-1, 512, 1, 1]           1,024\n",
      "      BasicBlock-106            [-1, 512, 1, 1]               0\n",
      "AdaptiveAvgPool2d-107            [-1, 512, 1, 1]               0\n",
      "         Flatten-108                  [-1, 512]               0\n",
      "          Linear-109                   [-1, 10]           5,130\n",
      "================================================================\n",
      "Total params: 21,289,802\n",
      "Trainable params: 21,289,802\n",
      "Non-trainable params: 0\n",
      "----------------------------------------------------------------\n",
      "Input size (MB): 0.01\n",
      "Forward/backward pass size (MB): 1.53\n",
      "Params size (MB): 81.21\n",
      "Estimated Total Size (MB): 82.75\n",
      "----------------------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "def test_ResNet34():\n",
    "    model = ResNetZrc(BasicBlock, [3,4,6,3], num_classes = NUM_CLASSES).to(DEVICE)\n",
    "    summary(model, (3,28,28))\n",
    "test_ResNet34()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "----------------------------------------------------------------\n",
      "        Layer (type)               Output Shape         Param #\n",
      "================================================================\n",
      "            Conv2d-1           [-1, 64, 14, 14]           9,408\n",
      "       BatchNorm2d-2           [-1, 64, 14, 14]             128\n",
      "              ReLU-3           [-1, 64, 14, 14]               0\n",
      "         MaxPool2d-4             [-1, 64, 7, 7]               0\n",
      "            Conv2d-5             [-1, 64, 7, 7]           4,096\n",
      "       BatchNorm2d-6             [-1, 64, 7, 7]             128\n",
      "            Conv2d-7             [-1, 64, 7, 7]          36,864\n",
      "       BatchNorm2d-8             [-1, 64, 7, 7]             128\n",
      "            Conv2d-9            [-1, 256, 7, 7]          16,384\n",
      "      BatchNorm2d-10            [-1, 256, 7, 7]             512\n",
      "           Conv2d-11            [-1, 256, 7, 7]          16,384\n",
      "      BatchNorm2d-12            [-1, 256, 7, 7]             512\n",
      "       Bottleneck-13            [-1, 256, 7, 7]               0\n",
      "           Conv2d-14             [-1, 64, 7, 7]          16,384\n",
      "      BatchNorm2d-15             [-1, 64, 7, 7]             128\n",
      "           Conv2d-16             [-1, 64, 7, 7]          36,864\n",
      "      BatchNorm2d-17             [-1, 64, 7, 7]             128\n",
      "           Conv2d-18            [-1, 256, 7, 7]          16,384\n",
      "      BatchNorm2d-19            [-1, 256, 7, 7]             512\n",
      "       Bottleneck-20            [-1, 256, 7, 7]               0\n",
      "           Conv2d-21             [-1, 64, 7, 7]          16,384\n",
      "      BatchNorm2d-22             [-1, 64, 7, 7]             128\n",
      "           Conv2d-23             [-1, 64, 7, 7]          36,864\n",
      "      BatchNorm2d-24             [-1, 64, 7, 7]             128\n",
      "           Conv2d-25            [-1, 256, 7, 7]          16,384\n",
      "      BatchNorm2d-26            [-1, 256, 7, 7]             512\n",
      "       Bottleneck-27            [-1, 256, 7, 7]               0\n",
      "           Conv2d-28            [-1, 128, 7, 7]          32,768\n",
      "      BatchNorm2d-29            [-1, 128, 7, 7]             256\n",
      "           Conv2d-30            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-31            [-1, 128, 4, 4]             256\n",
      "           Conv2d-32            [-1, 512, 4, 4]          65,536\n",
      "      BatchNorm2d-33            [-1, 512, 4, 4]           1,024\n",
      "           Conv2d-34            [-1, 512, 4, 4]         131,072\n",
      "      BatchNorm2d-35            [-1, 512, 4, 4]           1,024\n",
      "       Bottleneck-36            [-1, 512, 4, 4]               0\n",
      "           Conv2d-37            [-1, 128, 4, 4]          65,536\n",
      "      BatchNorm2d-38            [-1, 128, 4, 4]             256\n",
      "           Conv2d-39            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-40            [-1, 128, 4, 4]             256\n",
      "           Conv2d-41            [-1, 512, 4, 4]          65,536\n",
      "      BatchNorm2d-42            [-1, 512, 4, 4]           1,024\n",
      "       Bottleneck-43            [-1, 512, 4, 4]               0\n",
      "           Conv2d-44            [-1, 128, 4, 4]          65,536\n",
      "      BatchNorm2d-45            [-1, 128, 4, 4]             256\n",
      "           Conv2d-46            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-47            [-1, 128, 4, 4]             256\n",
      "           Conv2d-48            [-1, 512, 4, 4]          65,536\n",
      "      BatchNorm2d-49            [-1, 512, 4, 4]           1,024\n",
      "       Bottleneck-50            [-1, 512, 4, 4]               0\n",
      "           Conv2d-51            [-1, 128, 4, 4]          65,536\n",
      "      BatchNorm2d-52            [-1, 128, 4, 4]             256\n",
      "           Conv2d-53            [-1, 128, 4, 4]         147,456\n",
      "      BatchNorm2d-54            [-1, 128, 4, 4]             256\n",
      "           Conv2d-55            [-1, 512, 4, 4]          65,536\n",
      "      BatchNorm2d-56            [-1, 512, 4, 4]           1,024\n",
      "       Bottleneck-57            [-1, 512, 4, 4]               0\n",
      "           Conv2d-58            [-1, 256, 4, 4]         131,072\n",
      "      BatchNorm2d-59            [-1, 256, 4, 4]             512\n",
      "           Conv2d-60            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-61            [-1, 256, 2, 2]             512\n",
      "           Conv2d-62           [-1, 1024, 2, 2]         262,144\n",
      "      BatchNorm2d-63           [-1, 1024, 2, 2]           2,048\n",
      "           Conv2d-64           [-1, 1024, 2, 2]         524,288\n",
      "      BatchNorm2d-65           [-1, 1024, 2, 2]           2,048\n",
      "       Bottleneck-66           [-1, 1024, 2, 2]               0\n",
      "           Conv2d-67            [-1, 256, 2, 2]         262,144\n",
      "      BatchNorm2d-68            [-1, 256, 2, 2]             512\n",
      "           Conv2d-69            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-70            [-1, 256, 2, 2]             512\n",
      "           Conv2d-71           [-1, 1024, 2, 2]         262,144\n",
      "      BatchNorm2d-72           [-1, 1024, 2, 2]           2,048\n",
      "       Bottleneck-73           [-1, 1024, 2, 2]               0\n",
      "           Conv2d-74            [-1, 256, 2, 2]         262,144\n",
      "      BatchNorm2d-75            [-1, 256, 2, 2]             512\n",
      "           Conv2d-76            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-77            [-1, 256, 2, 2]             512\n",
      "           Conv2d-78           [-1, 1024, 2, 2]         262,144\n",
      "      BatchNorm2d-79           [-1, 1024, 2, 2]           2,048\n",
      "       Bottleneck-80           [-1, 1024, 2, 2]               0\n",
      "           Conv2d-81            [-1, 256, 2, 2]         262,144\n",
      "      BatchNorm2d-82            [-1, 256, 2, 2]             512\n",
      "           Conv2d-83            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-84            [-1, 256, 2, 2]             512\n",
      "           Conv2d-85           [-1, 1024, 2, 2]         262,144\n",
      "      BatchNorm2d-86           [-1, 1024, 2, 2]           2,048\n",
      "       Bottleneck-87           [-1, 1024, 2, 2]               0\n",
      "           Conv2d-88            [-1, 256, 2, 2]         262,144\n",
      "      BatchNorm2d-89            [-1, 256, 2, 2]             512\n",
      "           Conv2d-90            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-91            [-1, 256, 2, 2]             512\n",
      "           Conv2d-92           [-1, 1024, 2, 2]         262,144\n",
      "      BatchNorm2d-93           [-1, 1024, 2, 2]           2,048\n",
      "       Bottleneck-94           [-1, 1024, 2, 2]               0\n",
      "           Conv2d-95            [-1, 256, 2, 2]         262,144\n",
      "      BatchNorm2d-96            [-1, 256, 2, 2]             512\n",
      "           Conv2d-97            [-1, 256, 2, 2]         589,824\n",
      "      BatchNorm2d-98            [-1, 256, 2, 2]             512\n",
      "           Conv2d-99           [-1, 1024, 2, 2]         262,144\n",
      "     BatchNorm2d-100           [-1, 1024, 2, 2]           2,048\n",
      "      Bottleneck-101           [-1, 1024, 2, 2]               0\n",
      "          Conv2d-102            [-1, 512, 2, 2]         524,288\n",
      "     BatchNorm2d-103            [-1, 512, 2, 2]           1,024\n",
      "          Conv2d-104            [-1, 512, 1, 1]       2,359,296\n",
      "     BatchNorm2d-105            [-1, 512, 1, 1]           1,024\n",
      "          Conv2d-106           [-1, 2048, 1, 1]       1,048,576\n",
      "     BatchNorm2d-107           [-1, 2048, 1, 1]           4,096\n",
      "          Conv2d-108           [-1, 2048, 1, 1]       2,097,152\n",
      "     BatchNorm2d-109           [-1, 2048, 1, 1]           4,096\n",
      "      Bottleneck-110           [-1, 2048, 1, 1]               0\n",
      "          Conv2d-111            [-1, 512, 1, 1]       1,048,576\n",
      "     BatchNorm2d-112            [-1, 512, 1, 1]           1,024\n",
      "          Conv2d-113            [-1, 512, 1, 1]       2,359,296\n",
      "     BatchNorm2d-114            [-1, 512, 1, 1]           1,024\n",
      "          Conv2d-115           [-1, 2048, 1, 1]       1,048,576\n",
      "     BatchNorm2d-116           [-1, 2048, 1, 1]           4,096\n",
      "      Bottleneck-117           [-1, 2048, 1, 1]               0\n",
      "          Conv2d-118            [-1, 512, 1, 1]       1,048,576\n",
      "     BatchNorm2d-119            [-1, 512, 1, 1]           1,024\n",
      "          Conv2d-120            [-1, 512, 1, 1]       2,359,296\n",
      "     BatchNorm2d-121            [-1, 512, 1, 1]           1,024\n",
      "          Conv2d-122           [-1, 2048, 1, 1]       1,048,576\n",
      "     BatchNorm2d-123           [-1, 2048, 1, 1]           4,096\n",
      "      Bottleneck-124           [-1, 2048, 1, 1]               0\n",
      "AdaptiveAvgPool2d-125           [-1, 2048, 1, 1]               0\n",
      "         Flatten-126                 [-1, 2048]               0\n",
      "          Linear-127                   [-1, 10]          20,490\n",
      "================================================================\n",
      "Total params: 23,528,522\n",
      "Trainable params: 23,528,522\n",
      "Non-trainable params: 0\n",
      "----------------------------------------------------------------\n",
      "Input size (MB): 0.01\n",
      "Forward/backward pass size (MB): 3.97\n",
      "Params size (MB): 89.75\n",
      "Estimated Total Size (MB): 93.74\n",
      "----------------------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "def test_ResNet50():\n",
    "    model = ResNetZrc(Bottleneck, [3,4,6,3], num_classes = NUM_CLASSES).to(DEVICE)\n",
    "    summary(model, (3,28,28))\n",
    "test_ResNet50()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "def resnet50():\n",
    "    return ResNetZrc(Bottleneck, [3,4,6,3], NUM_CLASSES)\n",
    "\n",
    "def resnet34():\n",
    "    return ResNetZrc(BasicBlock, [3,4,6,3], NUM_CLASSES)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = resnet50()\n",
    "model.to(DEVICE)\n",
    "\n",
    "#### DATA PARALLEL START ####\n",
    "if torch.cuda.device_count() > 1:\n",
    "    print(\"Using\", torch.cuda.device_count(), \"GPUs\")\n",
    "    model = torch.nn.DataParallel(model)\n",
    "#### DATA PARALLEL END ####\n",
    "\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=LEARNING_RATE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "def compute_accuracy(model, data_loader, device):\n",
    "    model.eval()\n",
    "    correct_pred, num_examples = 0, 0\n",
    "    for i, (features, targets) in enumerate(data_loader):\n",
    "            \n",
    "        features = features.to(device)\n",
    "        targets = targets.to(device)\n",
    "\n",
    "        logits, probas = model(features)\n",
    "        _, predicted_labels = torch.max(probas, 1)\n",
    "        num_examples += targets.size(0)\n",
    "        correct_pred += (predicted_labels == targets).sum()\n",
    "    return correct_pred.float()/num_examples * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train_model(model, data_loader, optimizer, num_epochs,batch_size, device, metric_func, random_seed = 7):\n",
    "    # Manual seed for deterministic data loader\n",
    "    torch.manual_seed(random_seed)\n",
    "    \n",
    "    loss_list = []\n",
    "    train_acc_list, valid_acc_list = [], []\n",
    "    \n",
    "    for epoch in range(num_epochs):\n",
    "        # set training mode\n",
    "        model.train() \n",
    "        for batch_idx, (features, targets) in enumerate(data_loader[\"train\"]):\n",
    "            features = features.to(device)\n",
    "            targets = targets.to(device)\n",
    "\n",
    "\n",
    "            ## forward pass\n",
    "            logits, probas = model(features)\n",
    "            loss = F.cross_entropy(logits,targets)\n",
    "\n",
    "            # backward pass\n",
    "            # clear the gradients of all tensors being optimized\n",
    "            optimizer.zero_grad()\n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "\n",
    "            ### Login\n",
    "            loss_list.append(loss.item())\n",
    "            if not batch_idx % 50:\n",
    "                print ('Epoch: {0:03d}/{1:03d} | Batch {2:03d}/{3:03d} | Loss: {4:.2f}'.format(\n",
    "                    epoch+1, num_epochs, batch_idx, \n",
    "                         len(train_dataset)//batch_size, loss))\n",
    "\n",
    "        with torch.set_grad_enabled(False):\n",
    "            train_acc = metric_func(model, data_loader[\"train\"], device)\n",
    "            valid_acc = metric_func(model, data_loader[\"val\"], device)\n",
    "            \n",
    "            print('Epoch: {0:03d}/{1:03d} training accuracy: {2:.2f}'.format(\n",
    "                  epoch+1, num_epochs, train_acc))\n",
    "            \n",
    "            print('Epoch: {0:03d}/{1:03d} validation accuracy: {2:.2f}'.format(\n",
    "                  epoch+1, num_epochs, valid_acc))\n",
    "            \n",
    "            train_acc_list.append(train_acc)\n",
    "            valid_acc_list.append(valid_acc)\n",
    "            \n",
    "    return loss_list, train_acc_list, valid_acc_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 001/010 | Batch 000/921 | Loss: 7.48\n",
      "Epoch: 001/010 | Batch 050/921 | Loss: 0.37\n",
      "Epoch: 001/010 | Batch 100/921 | Loss: 0.09\n",
      "Epoch: 001/010 | Batch 150/921 | Loss: 0.18\n",
      "Epoch: 001/010 | Batch 200/921 | Loss: 0.15\n",
      "Epoch: 001/010 | Batch 250/921 | Loss: 0.05\n",
      "Epoch: 001/010 | Batch 300/921 | Loss: 0.16\n",
      "Epoch: 001/010 | Batch 350/921 | Loss: 0.03\n",
      "Epoch: 001/010 | Batch 400/921 | Loss: 0.08\n",
      "Epoch: 001/010 | Batch 450/921 | Loss: 0.04\n",
      "Epoch: 001/010 | Batch 500/921 | Loss: 0.14\n",
      "Epoch: 001/010 | Batch 550/921 | Loss: 0.05\n",
      "Epoch: 001/010 | Batch 600/921 | Loss: 0.06\n",
      "Epoch: 001/010 | Batch 650/921 | Loss: 0.10\n",
      "Epoch: 001/010 | Batch 700/921 | Loss: 0.14\n",
      "Epoch: 001/010 | Batch 750/921 | Loss: 0.14\n",
      "Epoch: 001/010 | Batch 800/921 | Loss: 0.08\n",
      "Epoch: 001/010 | Batch 850/921 | Loss: 0.11\n",
      "Epoch: 001/010 | Batch 900/921 | Loss: 0.03\n",
      "Epoch: 001/010 training accuracy: 98.72\n",
      "Epoch: 001/010 validation accuracy: 98.80\n",
      "Epoch: 002/010 | Batch 000/921 | Loss: 0.10\n",
      "Epoch: 002/010 | Batch 050/921 | Loss: 0.03\n",
      "Epoch: 002/010 | Batch 100/921 | Loss: 0.04\n",
      "Epoch: 002/010 | Batch 150/921 | Loss: 0.04\n",
      "Epoch: 002/010 | Batch 200/921 | Loss: 0.03\n",
      "Epoch: 002/010 | Batch 250/921 | Loss: 0.03\n",
      "Epoch: 002/010 | Batch 300/921 | Loss: 0.04\n",
      "Epoch: 002/010 | Batch 350/921 | Loss: 0.08\n",
      "Epoch: 002/010 | Batch 400/921 | Loss: 0.10\n",
      "Epoch: 002/010 | Batch 450/921 | Loss: 0.09\n",
      "Epoch: 002/010 | Batch 500/921 | Loss: 0.06\n",
      "Epoch: 002/010 | Batch 550/921 | Loss: 0.07\n",
      "Epoch: 002/010 | Batch 600/921 | Loss: 0.01\n",
      "Epoch: 002/010 | Batch 650/921 | Loss: 0.11\n",
      "Epoch: 002/010 | Batch 700/921 | Loss: 0.02\n",
      "Epoch: 002/010 | Batch 750/921 | Loss: 0.19\n",
      "Epoch: 002/010 | Batch 800/921 | Loss: 0.01\n",
      "Epoch: 002/010 | Batch 850/921 | Loss: 0.02\n",
      "Epoch: 002/010 | Batch 900/921 | Loss: 0.01\n",
      "Epoch: 002/010 training accuracy: 98.88\n",
      "Epoch: 002/010 validation accuracy: 97.90\n",
      "Epoch: 003/010 | Batch 000/921 | Loss: 0.05\n",
      "Epoch: 003/010 | Batch 050/921 | Loss: 0.18\n",
      "Epoch: 003/010 | Batch 100/921 | Loss: 0.01\n",
      "Epoch: 003/010 | Batch 150/921 | Loss: 0.00\n",
      "Epoch: 003/010 | Batch 200/921 | Loss: 0.02\n",
      "Epoch: 003/010 | Batch 250/921 | Loss: 0.02\n",
      "Epoch: 003/010 | Batch 300/921 | Loss: 0.11\n",
      "Epoch: 003/010 | Batch 350/921 | Loss: 0.08\n",
      "Epoch: 003/010 | Batch 400/921 | Loss: 0.05\n",
      "Epoch: 003/010 | Batch 450/921 | Loss: 0.02\n",
      "Epoch: 003/010 | Batch 500/921 | Loss: 0.02\n",
      "Epoch: 003/010 | Batch 550/921 | Loss: 0.01\n",
      "Epoch: 003/010 | Batch 600/921 | Loss: 0.03\n",
      "Epoch: 003/010 | Batch 650/921 | Loss: 0.11\n",
      "Epoch: 003/010 | Batch 700/921 | Loss: 0.00\n",
      "Epoch: 003/010 | Batch 750/921 | Loss: 0.01\n",
      "Epoch: 003/010 | Batch 800/921 | Loss: 0.04\n",
      "Epoch: 003/010 | Batch 850/921 | Loss: 0.03\n",
      "Epoch: 003/010 | Batch 900/921 | Loss: 0.00\n",
      "Epoch: 003/010 training accuracy: 99.54\n",
      "Epoch: 003/010 validation accuracy: 98.40\n",
      "Epoch: 004/010 | Batch 000/921 | Loss: 0.00\n",
      "Epoch: 004/010 | Batch 050/921 | Loss: 0.01\n",
      "Epoch: 004/010 | Batch 100/921 | Loss: 0.15\n",
      "Epoch: 004/010 | Batch 150/921 | Loss: 0.00\n",
      "Epoch: 004/010 | Batch 200/921 | Loss: 0.04\n",
      "Epoch: 004/010 | Batch 250/921 | Loss: 0.01\n",
      "Epoch: 004/010 | Batch 300/921 | Loss: 0.00\n",
      "Epoch: 004/010 | Batch 350/921 | Loss: 0.02\n",
      "Epoch: 004/010 | Batch 400/921 | Loss: 0.05\n",
      "Epoch: 004/010 | Batch 450/921 | Loss: 0.01\n",
      "Epoch: 004/010 | Batch 500/921 | Loss: 0.00\n",
      "Epoch: 004/010 | Batch 550/921 | Loss: 0.00\n",
      "Epoch: 004/010 | Batch 600/921 | Loss: 0.12\n",
      "Epoch: 004/010 | Batch 650/921 | Loss: 0.11\n",
      "Epoch: 004/010 | Batch 700/921 | Loss: 0.00\n",
      "Epoch: 004/010 | Batch 750/921 | Loss: 0.03\n",
      "Epoch: 004/010 | Batch 800/921 | Loss: 0.02\n",
      "Epoch: 004/010 | Batch 850/921 | Loss: 0.02\n",
      "Epoch: 004/010 | Batch 900/921 | Loss: 0.01\n",
      "Epoch: 004/010 training accuracy: 99.40\n",
      "Epoch: 004/010 validation accuracy: 98.80\n",
      "Epoch: 005/010 | Batch 000/921 | Loss: 0.01\n",
      "Epoch: 005/010 | Batch 050/921 | Loss: 0.01\n",
      "Epoch: 005/010 | Batch 100/921 | Loss: 0.00\n",
      "Epoch: 005/010 | Batch 150/921 | Loss: 0.03\n",
      "Epoch: 005/010 | Batch 200/921 | Loss: 0.01\n",
      "Epoch: 005/010 | Batch 250/921 | Loss: 0.02\n",
      "Epoch: 005/010 | Batch 300/921 | Loss: 0.00\n",
      "Epoch: 005/010 | Batch 350/921 | Loss: 0.01\n",
      "Epoch: 005/010 | Batch 400/921 | Loss: 0.03\n",
      "Epoch: 005/010 | Batch 450/921 | Loss: 0.06\n",
      "Epoch: 005/010 | Batch 500/921 | Loss: 0.11\n",
      "Epoch: 005/010 | Batch 550/921 | Loss: 0.04\n",
      "Epoch: 005/010 | Batch 600/921 | Loss: 0.02\n",
      "Epoch: 005/010 | Batch 650/921 | Loss: 0.03\n",
      "Epoch: 005/010 | Batch 700/921 | Loss: 0.09\n",
      "Epoch: 005/010 | Batch 750/921 | Loss: 0.01\n",
      "Epoch: 005/010 | Batch 800/921 | Loss: 0.00\n",
      "Epoch: 005/010 | Batch 850/921 | Loss: 0.01\n",
      "Epoch: 005/010 | Batch 900/921 | Loss: 0.00\n",
      "Epoch: 005/010 training accuracy: 99.37\n",
      "Epoch: 005/010 validation accuracy: 99.10\n",
      "Epoch: 006/010 | Batch 000/921 | Loss: 0.07\n",
      "Epoch: 006/010 | Batch 050/921 | Loss: 0.04\n",
      "Epoch: 006/010 | Batch 100/921 | Loss: 0.01\n",
      "Epoch: 006/010 | Batch 150/921 | Loss: 0.00\n",
      "Epoch: 006/010 | Batch 200/921 | Loss: 0.05\n",
      "Epoch: 006/010 | Batch 250/921 | Loss: 0.00\n",
      "Epoch: 006/010 | Batch 300/921 | Loss: 0.00\n",
      "Epoch: 006/010 | Batch 350/921 | Loss: 0.00\n",
      "Epoch: 006/010 | Batch 400/921 | Loss: 0.00\n",
      "Epoch: 006/010 | Batch 450/921 | Loss: 0.01\n",
      "Epoch: 006/010 | Batch 500/921 | Loss: 0.00\n",
      "Epoch: 006/010 | Batch 550/921 | Loss: 0.03\n",
      "Epoch: 006/010 | Batch 600/921 | Loss: 0.01\n",
      "Epoch: 006/010 | Batch 650/921 | Loss: 0.06\n",
      "Epoch: 006/010 | Batch 700/921 | Loss: 0.00\n",
      "Epoch: 006/010 | Batch 750/921 | Loss: 0.01\n",
      "Epoch: 006/010 | Batch 800/921 | Loss: 0.00\n",
      "Epoch: 006/010 | Batch 850/921 | Loss: 0.00\n",
      "Epoch: 006/010 | Batch 900/921 | Loss: 0.05\n",
      "Epoch: 006/010 training accuracy: 99.64\n",
      "Epoch: 006/010 validation accuracy: 98.50\n",
      "Epoch: 007/010 | Batch 000/921 | Loss: 0.01\n",
      "Epoch: 007/010 | Batch 050/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 100/921 | Loss: 0.08\n",
      "Epoch: 007/010 | Batch 150/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 200/921 | Loss: 0.03\n",
      "Epoch: 007/010 | Batch 250/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 300/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 350/921 | Loss: 0.03\n",
      "Epoch: 007/010 | Batch 400/921 | Loss: 0.05\n",
      "Epoch: 007/010 | Batch 450/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 500/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 550/921 | Loss: 0.03\n",
      "Epoch: 007/010 | Batch 600/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 650/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 700/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 750/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 800/921 | Loss: 0.07\n",
      "Epoch: 007/010 | Batch 850/921 | Loss: 0.00\n",
      "Epoch: 007/010 | Batch 900/921 | Loss: 0.00\n",
      "Epoch: 007/010 training accuracy: 99.65\n",
      "Epoch: 007/010 validation accuracy: 98.80\n",
      "Epoch: 008/010 | Batch 000/921 | Loss: 0.01\n",
      "Epoch: 008/010 | Batch 050/921 | Loss: 0.04\n",
      "Epoch: 008/010 | Batch 100/921 | Loss: 0.00\n",
      "Epoch: 008/010 | Batch 150/921 | Loss: 0.00\n",
      "Epoch: 008/010 | Batch 200/921 | Loss: 0.04\n",
      "Epoch: 008/010 | Batch 250/921 | Loss: 0.01\n",
      "Epoch: 008/010 | Batch 300/921 | Loss: 0.01\n",
      "Epoch: 008/010 | Batch 350/921 | Loss: 0.08\n",
      "Epoch: 008/010 | Batch 400/921 | Loss: 0.01\n",
      "Epoch: 008/010 | Batch 450/921 | Loss: 0.05\n",
      "Epoch: 008/010 | Batch 500/921 | Loss: 0.06\n",
      "Epoch: 008/010 | Batch 550/921 | Loss: 0.02\n",
      "Epoch: 008/010 | Batch 600/921 | Loss: 0.02\n",
      "Epoch: 008/010 | Batch 650/921 | Loss: 0.01\n",
      "Epoch: 008/010 | Batch 700/921 | Loss: 0.03\n",
      "Epoch: 008/010 | Batch 750/921 | Loss: 0.01\n",
      "Epoch: 008/010 | Batch 800/921 | Loss: 0.00\n",
      "Epoch: 008/010 | Batch 850/921 | Loss: 0.10\n",
      "Epoch: 008/010 | Batch 900/921 | Loss: 0.06\n",
      "Epoch: 008/010 training accuracy: 99.76\n",
      "Epoch: 008/010 validation accuracy: 99.40\n",
      "Epoch: 009/010 | Batch 000/921 | Loss: 0.02\n",
      "Epoch: 009/010 | Batch 050/921 | Loss: 0.01\n",
      "Epoch: 009/010 | Batch 100/921 | Loss: 0.00\n",
      "Epoch: 009/010 | Batch 150/921 | Loss: 0.03\n",
      "Epoch: 009/010 | Batch 200/921 | Loss: 0.00\n",
      "Epoch: 009/010 | Batch 250/921 | Loss: 0.05\n",
      "Epoch: 009/010 | Batch 300/921 | Loss: 0.24\n",
      "Epoch: 009/010 | Batch 350/921 | Loss: 0.01\n",
      "Epoch: 009/010 | Batch 400/921 | Loss: 0.03\n",
      "Epoch: 009/010 | Batch 450/921 | Loss: 0.00\n",
      "Epoch: 009/010 | Batch 500/921 | Loss: 0.05\n",
      "Epoch: 009/010 | Batch 550/921 | Loss: 0.01\n",
      "Epoch: 009/010 | Batch 600/921 | Loss: 0.00\n",
      "Epoch: 009/010 | Batch 650/921 | Loss: 0.00\n",
      "Epoch: 009/010 | Batch 700/921 | Loss: 0.00\n",
      "Epoch: 009/010 | Batch 750/921 | Loss: 0.07\n",
      "Epoch: 009/010 | Batch 800/921 | Loss: 0.04\n",
      "Epoch: 009/010 | Batch 850/921 | Loss: 0.04\n",
      "Epoch: 009/010 | Batch 900/921 | Loss: 0.00\n",
      "Epoch: 009/010 training accuracy: 99.70\n",
      "Epoch: 009/010 validation accuracy: 98.20\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 010/010 | Batch 000/921 | Loss: 0.01\n",
      "Epoch: 010/010 | Batch 050/921 | Loss: 0.01\n",
      "Epoch: 010/010 | Batch 100/921 | Loss: 0.00\n",
      "Epoch: 010/010 | Batch 150/921 | Loss: 0.05\n",
      "Epoch: 010/010 | Batch 200/921 | Loss: 0.00\n",
      "Epoch: 010/010 | Batch 250/921 | Loss: 0.00\n",
      "Epoch: 010/010 | Batch 300/921 | Loss: 0.00\n",
      "Epoch: 010/010 | Batch 350/921 | Loss: 0.00\n",
      "Epoch: 010/010 | Batch 400/921 | Loss: 0.00\n",
      "Epoch: 010/010 | Batch 450/921 | Loss: 0.01\n",
      "Epoch: 010/010 | Batch 500/921 | Loss: 0.04\n",
      "Epoch: 010/010 | Batch 550/921 | Loss: 0.01\n",
      "Epoch: 010/010 | Batch 600/921 | Loss: 0.02\n",
      "Epoch: 010/010 | Batch 650/921 | Loss: 0.00\n",
      "Epoch: 010/010 | Batch 700/921 | Loss: 0.00\n",
      "Epoch: 010/010 | Batch 750/921 | Loss: 0.00\n",
      "Epoch: 010/010 | Batch 800/921 | Loss: 0.01\n",
      "Epoch: 010/010 | Batch 850/921 | Loss: 0.01\n",
      "Epoch: 010/010 | Batch 900/921 | Loss: 0.02\n",
      "Epoch: 010/010 training accuracy: 99.67\n",
      "Epoch: 010/010 validation accuracy: 99.10\n"
     ]
    }
   ],
   "source": [
    "loss_list, train_acc_list, valid_acc_list = train_model(\n",
    "                    model, \n",
    "                    data_loader, \n",
    "                    optimizer, \n",
    "                    NUM_EPOCHS, \n",
    "                    device = DEVICE, \n",
    "                    batch_size = BATCH_SIZE,\n",
    "                    metric_func = compute_accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "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": [
    "plt.plot(loss_list, label='Minibatch cost')\n",
    "plt.plot(np.convolve(loss_list, \n",
    "                     np.ones(200,)/200, mode='valid'), \n",
    "         label='Running average')\n",
    "\n",
    "plt.ylabel('Cross Entropy')\n",
    "plt.xlabel('Iteration')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "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": [
    "plt.plot(np.arange(1, NUM_EPOCHS+1), train_acc_list, label='Training')\n",
    "plt.plot(np.arange(1, NUM_EPOCHS+1), valid_acc_list, label='Validation')\n",
    "\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Accuracy')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation ACC: 99.10%\n",
      "Test ACC: 98.94%\n"
     ]
    }
   ],
   "source": [
    "with torch.set_grad_enabled(False):\n",
    "    test_acc = compute_accuracy(model=model,\n",
    "                           data_loader=data_loader[\"test\"],\n",
    "                           device=DEVICE)\n",
    "    \n",
    "    valid_acc = compute_accuracy(model=model,\n",
    "                            data_loader=data_loader[\"val\"],\n",
    "                            device=DEVICE)\n",
    "    \n",
    "\n",
    "print(f'Validation ACC: {valid_acc:.2f}%')\n",
    "print(f'Test ACC: {test_acc:.2f}%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.6 - AzureML",
   "language": "python",
   "name": "python3-azureml"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}