part2/forest-houses.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "3801244f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Note: you may need to restart the kernel to use updated packages.\n"
     ]
    }
   ],
   "source": [
    "pip install --quiet --upgrade torch torchvision pytorch-lightning wandb matplotlib pillow testresources"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "374aca6d",
   "metadata": {},
   "source": [
    "## 1 - forest-vs-houses - categorical"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d158059b",
   "metadata": {},
   "source": [
    "### load data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "cd08fc16",
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL import Image\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "70162f48",
   "metadata": {},
   "source": [
    "on cocalc the data is in ``/data``\n",
    "\n",
    "data sources (to use at home):\n",
    "- train/test: https://paste.madflex.de/YtmeRXrY#houses-vs-forest.zip (500mb)\n",
    "- test2: https://paste.madflex.de/pyHrvXyG#test2_f-vs-h.zip (100mb)\n",
    "- test3: https://paste.madflex.de/La6ZMHvf#test3_f-vs-h.zip (90mb)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "a54133ed",
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_DIR = \"/data/houses-vs-forest\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "7b6d11c3",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "G0021068.JPG  G0021546.JPG  G0042882.JPG  G0077563.JPG\tG0109966.JPG\r\n",
      "G0021170.JPG  G0042712.JPG  G0054872.JPG  G0077705.JPG\r\n",
      "G0021184.JPG  G0042762.JPG  G0056373.JPG  G0088211.JPG\r\n",
      "G0021264.JPG  G0042813.JPG  G0067018.JPG  G0088454.JPG\r\n",
      "G0021422.JPG  G0042836.JPG  G0067028.JPG  G0109938.JPG\r\n"
     ]
    }
   ],
   "source": [
    "!ls /data/houses-vs-forest/test3/houses"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7f67c81d",
   "metadata": {},
   "source": [
    "show example image (commented because this increases the notebook size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "7bc8b4e8",
   "metadata": {},
   "outputs": [],
   "source": [
    "# img = Image.open(f\"{DATA_DIR}/test/houses/G0023671.JPG\")\n",
    "# plt.imshow(img)\n",
    "# plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "147aa0ec",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torchvision"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "b00a2e42",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Global seed set to 42\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "42"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pytorch_lightning as lightning\n",
    "lightning.seed_everything(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "91b6fe62",
   "metadata": {},
   "outputs": [],
   "source": [
    "transforms = torchvision.transforms.Compose([\n",
    "    torchvision.transforms.Resize(100),\n",
    "    torchvision.transforms.ToTensor()\n",
    "])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e7cec04e",
   "metadata": {},
   "source": [
    "#### training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "17492cab",
   "metadata": {},
   "outputs": [],
   "source": [
    "image_train = torchvision.datasets.ImageFolder(f\"{DATA_DIR}/train\", transform=transforms)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "176f3a16",
   "metadata": {},
   "outputs": [],
   "source": [
    "image_count = 107"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "ba55b27e",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_ds, validation_ds = torch.utils.data.random_split(image_train,[(image_count-15)*2, 15*2])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "645e0eef",
   "metadata": {},
   "source": [
    "#### test data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "0d4e25fd",
   "metadata": {},
   "outputs": [],
   "source": [
    "# images from the same tour/day\n",
    "test_ds = torchvision.datasets.ImageFolder(f\"{DATA_DIR}/test\", transform=transforms)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "499d0556",
   "metadata": {},
   "outputs": [],
   "source": [
    "# images from *another* tour/day\n",
    "test2_ds = torchvision.datasets.ImageFolder(f\"{DATA_DIR}/test2\", transform=transforms)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "882e77f8",
   "metadata": {},
   "outputs": [],
   "source": [
    "# images from another tour/day and another bike\n",
    "test3_ds = torchvision.datasets.ImageFolder(f\"{DATA_DIR}/test3\", transform=transforms)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "a5d456f9",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_test_ds = test_ds + test2_ds + test3_ds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "ca246b8e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "108"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(all_test_ds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "97c5ff47",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['forest', 'houses']"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "image_train.classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "743a8e4e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_test_ds[0][1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "f55e3f08",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUMAAAEICAYAAADFrJaoAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAAsTAAALEwEAmpwYAAEAAElEQVR4nOz9S6xty5omBn1/RIwx5pxr7X3OvZlVWVVZtuyG6YAEdIyEOyUQEgILd8ACI8tIhVICIWEBwmVaNEAyHcANBErJSAYhChAgI9kSQqBq0EGmAAvhMiVjV+HKqryZ95yz91rzMR4R8dP4HxFjrrX3OffezH1uJiuO9llrzTke8fz/738TM+OtvbW39tb+/72FH7sDb+2tvbW39uvQ3ojhW3trb+2t4Y0YvrW39tbeGoA3YvjW3tpbe2sA3ojhW3trb+2tAXgjhm/trb21twbgjRi+tbf21t4agDdi+Nb+BDQi+ikR/W+J6EJEf5uI/okfu09v7U9fSz92B97aW/sB7X8AYAXwWwD+fQD+ZSL615j5//Wj9uqt/alq9BaB8tZ+nRsRPQD4DsC/h5n/pn72PwXwe8z8V37Uzr21P1XtTUx+a7/u7d8FIBsh1PavAfh3/0j9eWt/StsbMXxrv+7tEcDT3WcfAbz7Efry1v4Utzdi+NZ+3dsZwPu7z94DeP4R+vLW/hS3N2L41n7d298EkIjoH+o++/cCeDOevLU/0vZmQHlrv/aNiP4qAAbwn4dYk/8VAP/BN2vyW/ujbG/I8K39SWj/RQBHAH8A4H8O4L/wRgjf2h91e0OGb+2tvbW3hjdk+Nbe2lt7awDeiOFbe2tv7a0B+BWJIRH9R4no/01E/yYRvUUDvLW39tb+xLZfWmdIRBHi9vAfAfB3APyrAP4zzPyv/9F17629tbf21r5M+1USNfzDAP5NZv63AHd/+McAfJIYHk8P/P7rn4JCAIEAAgBGzgV52+QiZhARQkogALVWMANEBKL2rForai0AAwxGoIBhHEGBwMwAAzFGhBAQYkSM0e9lriilopaCZZnBpSKmJP0i8n8hRukiBRABIUQQEVKKCESoXMH1jpkQoZSCdVnAOpb+XmYGM6OWgpxzdyOjlAoASCnpe5L0QQduP2MMGIbRn01kYyVxQAFQK6PWChAQiKCTrX2UvygQgq/D7opX22vf37PSzz2jv/b73vWnpfH978zotwyDwZX9OluXGIL/DQBVQYvsx5ezd//J5yAOf+IPBuRc6L6uDNmvuo9k/wI5ZzAztm1DrQUE2u1tEMn5BlBK8Q4SgKBn0lqtFSVnsBxk6Yc+p+Ss51/+wZ8qlwYixGHQT9nvlZ86u0ofoH28np+wzLdXt9+vQgx/G8C/0/39dwD8B+4vIqLfAfA7APD4/mv8p/7yP43j8YiUknSfGN998w2+/cOfy8BrQRoGvPvqa1AImG8zaq1CGELQQTOW2w3X81kmsxRMhwN+++//+zGOoxAZZjw8vMPhcMD7r77C+6+/0kmuyNuG56cn3K4X/H/+jX8Dt+sVP/mNP4PD6UGISgyYjgecHh4QYkIaJqSU8PDwiGEc8NOffoVpHDDfFmzbJhPOrISJcH5+xt/6t/9t5JwxpIAQAo7HI4ZhQM4ZpRScP37Eh2+/9UXMueB6PgOA9OV4xNc//SlODycMw4CUBhARYgx4eHzAX/yLv62fB8QY8NOvH3CYRt28wLKuuN1WxBgwjkk3iOyBEAICAYdpkP4RIdhm/QSVCjo20o0FvH7g7siuX7O/lrttjd0Gf+15L1vrA32GrP4qBPeXk5faQWTooUYjZBXCpNacUZQAMgPzumHNBTEQUggYh4TH00EYsg5iywWVGSEQAgUnLiASXdc9gdR3t/lvf1dmVKfI5H2uzPjw8YrrvGKeN8zLim3bcLvNCCFgGAfkXPDhuw+Y5xl/8LPfx/VyQRoGpBiRS0HJBSFEpDSilILL+YLKFSEwKBDevXuHaTqASIDH5XLGhw/fObiptWJbMnLO+PDNH2K53bDOZ+R1EeYdgp5jYJwm/PTP/BZijEI7UFFLBdeKUln+lYJt2xBCwjBM+D/9y3/1k6v3x57Ci5l/F8DvAsBv/rnf5nm+gUjQj2wW4QLT8SCDKAUhRt/uD4+PICLnENu2oWwZzECICRRYFmoYMKQBwzBiHEaACDFGlFqRS8aWN+dCOW8oZUMpWbggZGJPDyfcrldcr2clwAPSwEhxRCDCw8MR4zCAS8Uyb8g5y+RzFU6mEz/Ps6A7MErJyNuG6/miY5AtuS2rcERmcC0opaDkDSEEjOOA6TCBS8Fyu4FLBY+s6DbJRlaqwywI8DZvyFm4J4gQKeDd41FRKaGUiuttRWXGkBJCIGxZ0UcgBBKy0s6UIRH5fIh2MI0Lf5rUvP6tffr5e39Y+/TTf20aN1TSa6JUGNr1NRIh0l6BTywMSD5jmXtu19Ddz8+1qmi0shCIyoxcqvRDF7yyEOqn5yuez1dcLldcrzcsy4Lr5YIQAw6HAyozbrcZ27phvl6x3G5CxIYBISaMhwNiCEhpUMlECG9lkeK2rSDnqxD1QAIY4oCCjDVvQK1QtuFAZ1s3rMuMEARVmtRWSsH1cgERYV1mkRSVS8Q0IqYB27ridr0JwU6fJ3e/CjH8PQB/X/f3X9TPPt2YFRllEMGJA7Mc9EqCO0IQkZaIMAwDQgjYchZIvm3OcYW7yK6JMSrCIcQUQRQACHSvSjCMWJUi6KyW4lw8pqgEumJbV+RpE8IcImQzEoYUkVJUrlpQS+1gvIxtXVfknJWwkBP4eZ6xbZuiK6CWrISsoJaMWoqMj2STxBBQa0XeMmIQMZ+IEAWG9pOqoktRiqDEcEwYks0jo1ZCqRW1MmKoAAJqZZRaERDAKm4Tt+cC8HEkNNGD7lHI97ROwPOfPfG6R42/DFn7oyaEv8zzeqzliKz7hajhauq+FERO7QbWtQAUicvPewJot+xwevfD0GmtSgiZfQ8I42zvrUwolbGuGfO84nabcb1eMc83XM5nARYq8q6rAIG8bch5E1WOq6OCEi0CUfAzlQu5eqjW6sSwluJqKRez71qtjKrEG4Cos5R45yzqtXWeUfT8gAgjpB+1yBkKFFTU/jTm/1WI4b8K4B8ion8QQgT/0wC+NwMxV8Y8Lwi0ohRDVBlFkVstFcNIOBwOiCm5eBgUSUJFXWZGzQXDOODx8QEhRizLjJw3vP/6JxjH0RclxYiShRAu84xlvuHbP/yZ6/WGQbhIHERXBxY9Ti0VNBLGaUBKEdu67gioENSKZVmwLDNKKch5w7auuF4uem0FFI2llDAOA9IQcT2fcf74ESVnLPMMrgLvmSrWZUEMQbY/Aet6wLiK2H56eFAGIfrCIoodnPMNAFR0jsrqAZAcolIrWPY/ahVEvuWCUhkhVNfjEEx/JaJsUNSYQkSIvRD8w/FfTxzur0V3zf1vf6KbqhQIcL0yQZCdETrTbRMYsbs1kCDB4BOnesXKCBQ7Qgl/bt9sXnOpKLXitmSsW3F0WCtj2zJAhBQb1qzMuN5uuFyveHp6wtPHj0IMn58QY8TxQaS0nAtKVQa/rqIHjAHL7YbldgMge6Z1jJxy1yoEL29ydkotKJuIxbfbBQCQhgExRByOJ4SQHDiQP4b9vDjxrwzOFRVC8GqM4BhBYIyDSGm38zOq6TBfab80MWTmTET/JQD/ewARwP/4+0KkjFNt66ocS/RbtWYwF9UZViROcqiHAaSGh6g8Mm8rchaiJPqwEe++En3g5fkZedtErBuGZkAJAaVUIYjrivl2w4dvv8W2bQACYoyIKbqxwnSAXGUzpxQRUxCxuiox5GakmOcZl8tFvqvyjmVewLUiBDRjSAg4HCfRa+ociBiwOkcMoSJvK9YQUE3nVMW4Mk2j9qUZVWxziTK74sAM5sGV2k66uDbdFQNBRaXAQGACmQ7pTqZLUXSKzKav1S9+ANG6J3bfp9T/nOj3Od3gL9M+3ZdPffP977c+sqE4IlA3n/5Z9zlzFaLphgdGABC6EVcGlKcCMP3tvleG6BsQFRS4lYp5zbjeVgCkesuKbc2urgqB/J5lWbEsC263Ky6XM+brFeenj4hpQFWxWkTqim1bkfOGoY47FZTtZaKAQQFGjAnQs1VZiHHeRCIqJStQWeSsDAMoBIzTASEkLLcrtnUDlNBxLUoz2qSYuoi5AFybpAUWMFQK1vkmovsn2q+kM2TmfwUSNP+DGxFwOBxFL6jwPW8r1nUR/dq6Aib2ESGlAIoBh5MaXVhQZNlE1BURW1BjCBFgFqTFVRBfjIgpIaQBeVsVvYk4W0vB8fQeaRxBQcVGbvqTUgqWZcXz0xkhBKSoVuYglucUE0KISvSEWJRalPOtKgLIBio5i+4ub5hjwHwTJBdDxDRNqLWoMUbem0uRfqs+dJwmhJh0E254fj6r6BxAAHLJ4FoFBWbhtCVnPWBwMR4AhiEhRHluhIhIQa3g25ZVPRFBekhiINc7maHlT3L75Y0jP3zkjbA3NmDMLSjEyQoIclFdeWcggDNC1TOTPMvm/3M9MWa9bRVLLrjdVlxvsxoUKrJKI4B4IVAQdRQghrdaq4i4QxLGG4LvBbixTe4hVZuIAaSiFtOdr6Ia0O+n4xEhJUzTQfZdiCjDKEy6ilGTVDoJMYEIGMYRMSY8vP8KaRyxzrMQNAZq3RAUCBhiDIGQNyGuKWdwKSg5Y1vF6p3zdqdi2rcvWwOFZRKPxxPGaWrWtNsVFAjbuopxBPCDG+MIAuF0OuF4OiJvi0zKIsSTmRFiAvRnLQXzfMO6LhinyRFmGoVzGTHcVEQeDwdMh5PoF1SfUhkohZFLAS8LykfRwHARXeAwHRBjwun0oISMxRWCGQUErkIMS+c6I7oUwqw7VfQvogieDgfpmy5WzgUhZkTt+zCOGKcJMSUUZizrhqenZ8QYMY6j6yYBVn1MwJA25HEEIByaqxwCIsJ0GBFjUJ2rEPEKwrJsuN5uSCniiAkxBoABjoRSE8xu88dJDX9sQvvL48LuWtV/9QYp03Pbs4Toifoi54Kci6BwAKjNlaTU6veZyqIZWl72ykgvM7BuBfOacbktOF9m2febAIHr5awAIoDU2yHGiGVZnDAPw4BtMJez4NIugYAohr4YRGIThKf6yG3D7XJFLRl5W8T4+NVXGMYR0zRhGBIwCP61Wd+2DNLzmxXkjFPQ8y0i+vnjR+RcXKLkys3lRnXttYohsuZN+rSt2JZZbQaKJj/RvigxJBI/IzMchCii7DhNYK4IFFBywTCOOJ5OSGlw7jBNIl4ejydsjytKLhinK5L524WA0+mEWgvWdUZWbrCtUQgeQy27K8q2odYMgMSdIZJbdlMUF5rTwwmP798hxog0DGK5Oj8L+iJRDpdSsK6LoMYkuhzmiKxEpDVGyRsy2BXkpizmWlGLoLi8rSAKCAFK6AQRJhX5TezlWjDfZkGaW27WYAIyiVhbhoRaKkIMSMMgBDoLUo1bkH7mIoRSuaVtQrHDCP6ozKAquslcKigGhN4a2p3H7yMYnxOVaff7L47AXmu/HAJ8eV/rt+n+Pt8/m0NW622p5tjSnue/K+HMpYiOMAUBA5XF8GXrgSZi2lNMv2s6bHtvVgS4rBnLsrk7V1a9+bauWBZRzZi7SowJKVVsKnEVNXQww9VI5v/a/P+aOCzShOjnCYxtWVBks+q5EB18VcMlBfMbZncZOhwP4NpUPvMseniQgImYoqLRilrElpC3TZ9f+0kVg1HeunHUz4FCAF+cGBKGcRBRcgVODxOmaUJKEdM0qk8hMB0mfP2Tn4rxRHV+Dw8POllFCA8ClnnFOE0gkok6PZzAteL3/+7fwfV6UUjMOD08oj6KDiFvGctyQ86r+kMFDEPEtswoteJ4OOHdu3d4eHzA+6+/kkWICcsyY7lesBUh2iEE5G3FtlacTkdM0xElBEQCSqcTNf30sixu+QLQdJk5Iy83lJyx3q5ikQuEcUw4HU84nk6gkNxSRwDKlt0nMQ0JMQRMo6gE7EAMKSENA8ZpxMPjI2qtWLVfDCDliEBRdZ+iaO/Ri7kwlMLqG1fBMGOKHOc9AXu9vUZY/iS0X7XfrNbbXCu2urdixl7hR0CuFesmapQRQGHxRXSrfef2VJjVLsZNf6dqna2IXngrFblUPF9n3OYV87wibxnrsmKeZyyL6bilT+KyJUjQ9um6ra62GcYRKSUcJnGtWedFCKKpqZKckRACYgiYY0BeF9QakZPsSVRWhr8hpw0xEWIKqBoAEWPEV6r7X5YF25bVAyMjRvEQGacRh9MReQ0ABOmJuM8q6UCJrFipt9X6mb+XEAJfWkxW5bG4nhTUQ2n6ELE0wDZNIEKIwY0ggKEpuHFAnIeD6wxMiWpcqyhUHoZRjCVsClYWBKbPICJs24pt3TAkQWPSXTWmqJW7dO4vMQasy6oGGxGhxaAiiwBdGGn9SrASIOFYtdPvlVrABO3Lim1bENeINAgHL7koQauuCC45o+pY+rHXKgfD5ts+A16iutrNm7k95CyiEpvSXP3TzFXJ9Eb3VuD7EdMPwme/PIl8zSL9fW/s3/ZDeteLt/I3v7jgk6i361RvWPFbue3bov+CIjLDhM1kyohokVgEIYyVxY903gSxLWtGLgXLuoqhIjf9uonFLYABathg9dODSiBdH9VqSyGAagWD3T+3P4+m/mkuc03CMClIHKAXDBBpztQJAHx/BgoqGU0qRRW0aC7Bw3K2M0rZVDcZvQ9EIj2VbHu+W5nPLPiXRYaQybmeL8jbhmEYkYZBvUtFJ1FKBtcCUHDEE5S7CPGRhSUAKUbRswSZhHmeUXLxxZ+vN2zLrLrvICJxCgBXDCmJaDAMCDHi/PyE8/Oz+zbmPKFkWYSlFizzDbfrFbVWfP0Tiep4+nDF8/MzUKXPQfvM0OGoro4ZLo6INVr0RCWLfmObr05EQwx4fvoo+tNScDhecXp8j9PDoyq+bwgqQgNNt7qtiyuymRlJXXmOx6NYyolQix0G4bTmnmObtKpyXaIIgkc7hBCwjQmBCDmIaoH+NFhSfsF2R5q8fYoQin6vmVBqlYMaVfFqTCiXglUNhrKGwDgKhTUB2yJWkq5LE42BwoQtZ3x8vmLLGZfLTUTinFFKxaZ+gSaZMRjrdOwYJKlOWSNcwoA1JcQQsTErAYtq2NBQPXNt0fEHImzbhnVdsS2LemoY6gTWZUGtjMvTM27xitPjO5weRB+Y0gDUivl6BWmky5QG/PSnP0GtLNFit6sCF6CWDfP5owCKvAEE1GxucRXEQM6LiuNRQmHhHOST7csXkTc9GSt3YQYZB2F1AanNlxB65gySm29frWJWlzjj3IkMrDoQWQQ3+a8LagwARzdeAL0bTQWXKj4M6seUs1hoxSIlRMc4E6uxhWtFzQWlbGrt20+6WQYbTldUSwEc2F0daq1gmQJ3Uch5w7YldempCGDUEEBMziktosWGxHXvuLqp36MYWeBuRiHoPIaqjrji82VhUaUUgAMoylF253UV/yt14p6M1MfXDR+/ThTzFzKC4IfrHLn/rbvJRFsFLHIF7+3S3O9Bndui80wgFBZyaC51hdi9MLiKs0lhYN0ylnUT8XJZO70fIxdBhxY2l2LCOIo7TM4Ftr4tzp6wC6BWy0kzzrDvP/naN7syf3b9tqhuxGuCIIMwH0FDooXU+GjGv0G8JET0ZXf9aXPMSieKP0Ms9KF1x+wSRBDPP+Men17HL0oMJbpjAYWIYQh6mAuIA8DiIV41LE3cYzQyJYn/U+WK+XrDusyYbzcs64wtL1i3BSkmnN69A1HA8eEdhvGAvK5ifV5n5A+rRHbEhtYAwjpv4BoQSFxchnHEMCTkvOHjhw/iXjPfAAYOxyMk9GdBNn3KMKKEiFAStmXG+fzkZnwiCWEKIaioIT5PRAPCJBz+ennG9fwB4go0IIQoaO82YxwlJnRdFgzjgnEccThMusDNrQCsDuBcNayOXFwxPanuDw1jEp/FnAvGUVQIWZXi4hqUMOYBFCOmcZA5U0a0bkJwpzEi3SUT+NPS7NB/Kr/d9xHKnlAEIkwxIpMQHkN5REISDeWVUlGCXBNDxJaFu23FVEnCvEsRt5c1F4lXrmJ8WZYV33z3QXSD66bRRXLvtm0oOePd4yOO7x5wnEY8Ph6Rc8Hz+YK8ZVwvZ2HARWKg87qiFHXFUp2h7Dk5o1wKhumwS7xgc5JSwunhQXwG4yAS3/EkEpE+l1TU3coiFmIlhiklxPg1UkqoRZlv3oBaOkdrRgwRrOfMAIcZgkBBAAtLroOqc2fE8lPtCxNDQT0i+zfkBgjthgevGwKUA02K0iykrhRDMwW1ij6LB9U1BsleE0OQiJIYRC+3raqHVARaTVfD7lclOkjyflUWjrouqyRbOB1V1K2KnDR0UKX8bSPkIgrbbd3E2psSoIYNaeS6ypSiuq8YNw3uJlMgXv5FnVIljjr5IXPnUX2u9Le5GdjW5CoI0FBwjAnrg4xn2LIixeLEtPmL2Say5zCqxpkWquAqGw6a/IF+AYr4x00875/Pdz/vv/2hCPCHNI/f6XRcRNz1yZB7C7Tzb3iPwAG1RLuExGAEkBLWdVMxOGcsy4p5XkT1ogStlOISjlmACaKyScGcoQNqIZQq6qWsZ851f2jGNGaT2Br6M7UQmyWHxdvCLM8ttFYYas0ZtZITVbFwt7PNtXjgRK3o3mkSVtNREhRya59qBULUuTa03a/N91hRvigxLDnj6cNHPL7/CsMYscxXrIu4x4zDhG3LGIYJMY1tcpS65030hbWKuwiFoBtDFLsxDXh4eNQwPFm4bbmh5g1P332Dj999Aw13QQgRw3jEdDyJlepwQLoOGkq34XK+4HA64eHxiLwJMY0p4t17qVv+8z/8OZZ5weF4xOnh5ETt43cRl6ePEsc8jorG2MVtZkYJWzP+qPO1pTCqQcT4GkR/tK2LLLwqmo2gQy3ALf0TOxEXX03R1xiydvcjihjGAe+/eieWw3VxgmtGFq4VJUlapqTO2SkmlKK+lIcBIPF3tPYnGRl+TnK6N5z07VPGGnN4ER2hEJPKog4xhG1RQ6UIsUsp4nSY9B6oO07VfS3+iCL6iiGNQLhcZ1yuN6zrqjrCjOv1CpCGsoaAZb45Kqyl4PwM8ZUl8lR3WxF3msvzM5Z5Vl2wuKxYWj2CELt1WVRvvaDkTbLPAKojFKPfuiwYhhGHhwfV88+NudaquvAMmmdxmUkD0jgCqOqbu+GbP/xDxBgxTQeEmNz4ab6PZRiQhkl18kV13SsqSwwyokihgIIj9QXe1u2zrO+LEkPzHXp8/16ssauY8adJEEatEk3ivoiex6yhF6BxKuMOgURJOo0jpsMBaZBciIfjAeM4gStjuV1l4nJGjAMYAXEYdcP1YXsSHjdNB414gcZIJ4zTKCJpzliXBcfTSRyixyQpva4XFRkIw5B0MyyubBY9iSqcA4FKQNaYbNNT1o4Dl1yQwyaGFVWcr8sBADyu2xBi0hC9ZZ6xzrPqVRgxJQxjFuSrESyGAs1A5HOrz2MA6yYbZ9vMLYFQY8RYoh5mtSTem6Z/QHvNAPHr3HodHwBXxrfP9Yg59RTE0odTCsISvTbn6pZ5s/CmFJU4NkssWAwMlm1my8WJ4bJIMoXbbcbz89mjPkKMOBwmiagqgrLMt2+dZ6CK318w7w0LVVslvt6invK2Ia8rSHOB+j4xzwfdO4Cch21ZJMHJuur5FLVX3ixblPzM27aTQqbjEcM0AmC3C+RNsjeBCWmoHk8sSDaqG08EB5ZUYVSwbWIwqbF2unSo7jEBKN8rA3xxP8OYkipuN53UijrItkopYhgnUfBqyp2oaXtSjG69AoDDdMD7919JBIcSwa++/gppGDDfbli3DTEkTIcjTg+PeHz3Fdb5iuv6BI5V9YcSFE9gd+gM6nOVS8btdgOBMB0mxBB8A0wHyTV3PB4wTQPWZcH1/IxlmXE8iZUukBD/FoOsjqo1u94DYAlbIoAoIqRRNoES1MoWTrTaBEpCiRgxDhMoBsQ4qZJejuh6vTYn7rxhnA6it4xBXZUI67oi3G5iOYyjJItYVs0DVzAOI6ZxRC0Vt+ssoXspIcaIIUnkAbPp2Bsq/XVv90Tth7ZP3bPTHXZGBVKnX9MPrp6QRLMblSIIkBtDY0CJlDguj6Pk+1xWFYW3jC0XbJswrqePz3h6esI8zzg/nwVlxYhYCq7nC4iAj999h3m+SZBBaR4CYkuQFHWPX30NqGvWMAyiqglK7plRtw1lXT0FFkF0gmCgbBuWWjHfrljm2d25xMG6uO5ckhbbfDRmLIYh8TMkBBxOJ/GHXSTV3LwsCNsmevxpgjlupyFjmI6ISSS2qpmwLAWeGV1lCNXP3vfpcr64n2GICZV1knLRZIyyTWMU4hVjwJDE8hWjuKsYcgMAMDBNkzpNRxwORxwOIx7fvUMMAbfrVbhaCBjHAw7HEx4eHwEuuDyJxThERWdGDDXSxHQctVQs84JhGERXSOQ6mFEdnKdJjC3XyxlPH5/AteBwOLilzFyAALVql4y6rahl881SVX9EFBCScDzT94qeTlQElWUxUxrEkDIdvA+koYBmUCk5Y9tWyQGnrkfRfDZJwh5DEOPOwPAEFg0hwlHqvCwALPFFwOPDUccj4h4FkvRfX3Qj/XLNiNd9Xz9nEHl5Lb343vTcr7Wioqgl2xC1SXFR2tChPSxEcf8ah4RSGZe8ChHM4kh9mxcsy4qn52c8fRRiaCm2TqcjWPdCrRWXp4+4Xi+ODk3vJnszYzocNHmxJPZIManaoJm+zQ8WzMDxJAw5JpeQTEpal1ldg4L725r6xVRbYqSuyLWFqVo8c9C9bPHxtYhPLaAGnGEQX99chBiOE2oViU2YucRakykMRcHkQESkmF8jYhhjwrv3XwGscYw654fjCT/9zd90XZqk479hXcWtNITgStaUhEMVc69BxDAIasl5QyGCJW2NMWIYx2bNBZq+kcVH7+nDd1hmOfDmBmFOqH2aoKqJZUUPJ2Z/c3/JWxZFNCryJiLBfLv6IomYInGRpQoH87hKkvjkmAYMw6SlBpQ5pChO5aEZd8ySbCFLKSWxetpGVyJb8iZ61nV0y7cQXailjd2RPOumJRLfzaA6IypGrAER41OzUrMYscQgJJExr2Zc7tpLMYV2hOhLENTXiFtvzvj8va/3sKEQ7FCJfWcuNoDMQVU3lpgEhZUQQFRgIXdAy4hkDIrVeNDr3/qU+DlXXJ439SWUjOeLqqEYrKU2dLwZKOuCdVnx/PEjYkpuaTXUlpcF2yJZadZlxjgdMB4OovqZr80QCrjV2Z2vIcSQCDieTjCH7loZKWVXPUmI3yjx9UEz9hAQo+0m6XGI4iNsORBjikjjKJ4nOrmH0wNyHrHcLshbK7kBZs0RQJ/dm8AXJoYpJfz0p7+BD999g3VdNL9ewOO7d/hzf+EvCJdZF8y3Kz5++EZ1GbMiQpmc04Ol2hJuRxgwTSNiiqIgZXbH7KBxz6KDFEuaJaK0NELf/OEfYBhGfP2Tn2I6nJx7sE6iKL1l492uVzAz3r1/j5QSzs/PmOebiA4xIncpvM4fvxPLntZpURLkIUnSCBQD0jAipaFl8+lcBQQtBGcUljxhXRYVWxMiN2uliQt5E+fXlMTlx2I5SRFnrwPKGqWQkjhzh0DOlUtntebBRBwNMysSohfDJ1LP/4D2y4quf5Tt+1xl5Jof1ktDeT1QNKM8q/GzaGr6YZBw0Jwz7LzCc05WzzBTcmleA2b5tzjyWt3F5Pl6AQEYD+ICNt+uyNsmesAYReVEljS4IG8VtfxcUNnxiBAiNk0wUrYNVVHf7XbB4XjE4XgEg7FczxoTLASmV5WaeF1yRkwJj2p0nG+ze4IwiWfGdDi6xMcs0Vgg9rRxQAAoSC5Nkn0Wk0hw4+EoiG8BKFacVPe5zDes6+aAhZlRNCNO+GPMdP1Lt2GcQEEiImplj4YoQBc6JBtrWzctkJNAQQiFWWlLKe2womXKNeRjG0gU1INwIxWHHx7eiSVLC0GZGBtg+d3E2bOWgvl2VefrDQBpujFRWBfNZEOglul6W8X1h6sTQnMwd8WuesbHlDBMB0e8IcbdPeZOYKhg2zbEKNEzppsiDWYX8dbGLrPCasGrzKAYBYlqZvCSMxY11lheO2ZV4ivBNl1LUZ3Tpn5sMQBFkevnLLK/Du1Trja/yN3fR7Rtv7bnGxIXNM/UnKqzqltcPNataiIj0YbbLMmPJdGCuNHkKol/t3VRa+/qVtyiOnhDZmY4YNachTGpykk+m44nl3q4SqJkDox1vsm+3jb5lzeUbUWOEet8cwtx3jbP/0lBAAZXBlOFlbrgWrHpnqvqJ0gh6F4TAse1SEmOIuogDzU1JaoaIcsuGzzUP5a8T0FTkVEwRAmYS447k3c2h9fal/UzBKOi4PTuEQAhryJiTuOEGAgrK2VfFkEilXFbBI0dDkfJvqIm81qKmvGTKW1kQ1pUiDqcSmjagMPpAbVWDMMZx4dH/Nk//9uIMeF2vbov37LMGIcRNAwAS/aYvAnKE7FVCGcp2d0PLPqFAM0K/OzGoT7usuTifUIISOOIYTpgGCacHh4RgqXjsrRawLLcRKRWHdC2bcD1inGacDgeEWJwsdXEZNf5AUAQQ9DlckZMAwoLgU4xYBoH3K5SiuBwPOB4PDQXG/OjBNQYJDpHgHGbbxguA4CjKtTpy8q5P2LriSjdfW7MzhWo+s8d/QFwFUvqskjOwGkatZiRqCy2LWNZM8Ky4bqI4eR8viLngnmZkbNUc8zbhuvljPl6we16xeX5o6pGsmY3elR98hHDWJUQStjbMAyYDhvSIOqTy9NHdXkR0fL88QPm2wV5XVC21cdYy4aoHhLPHz+glIwhicFxmI5Iw+jxykUNeCGIxwhpPLT5NZp3RggS9HC9yJlZbuoadHxQZ24h7M9PF6zLiukwYTqIreDh3SOWecb56QNqKZhOR/VvHFQNVlByAyDEodMnvt6+cD5D4YxSMU7iBcVjXLK6rPpv2zZRxnZhOOZuYvo3zzpt/nZKCJjN+RguBor7QvR/MSYcjycEVb56Kqva3B3MCpW3FfMsqcylFKmgOUlltEkIEZGjKhuniTAmOsYUETlK2F4hWHEbL3ATdE7ctUaNGMpNLVzRYQS3vgItDMr0chYLbS4O5o9V1Qk3W8jflsEHduMUqc5V/A5VX6rIsM+Htw1JMooQUGoCgiQMsFT3zePmXoP2qb9+XHTZZ6ne96Wj9J8wkvSfmjXVRUjWewHXeXl6NBWF7frKkkYNpQJZGNJNxctlWZBLwXwT5CagYUbOq0sCFkduzvvmntJ/Jio2kQwMqaHuQ08t+3TeNteBUtBMNNwSoNQYQRzcION6OZaM2hY5Q0HSyMH6F0RtZVnrLf9gyZKOToIgBE3L9mm+tiY1QiUxN0TqPdC++lppBv0A+H7+VPvCESiMbc04HI5iLn8c1aF3w8/+3t/Dcrt5soRxmkBpcCOGKIk3LPMVhIplviFvC3Ie5aCDsS3qPZ/l5zLPYlneMlIckGJCTAMOxxN+88/+FtIw4EMasGoFrZwzxD22YFsL1lVqOnz49lswgGE8IsSE08NRXIS0SJWJuACL1UsNGFKgRrjiV199jWk64Ha9YJlvWpI0aQKFCqLqNVjmq6b0UvFnGEdQip6eCID7BQYATOTuE7VWUIpIkAqBplKorHkTN1GaW/B+rRWHoziqT9MBh8OEm9aycAskkR7CTdxqVInPDOTjhEGt/ilJzo1B+9kEll9ir/zA6/44wCi/+tP+33pm6hGhRHIA85YlzZsWRNIAHYlgmkakWDCPK7atYFkz1lXyGKY0gOeMec5Ytw3X6xU5Z021pYlJa8Xzxw+Yb1KoaZlnWG7NEBPiJIw+qlhM4W4WuTo4iDGBK3vBpsuzJAcx44OkupP0WMyMZI78YCwqmqYhgWJCXsXHMOr5AgiFLOenGEDTMHp0WIgJl/NHXM9ncBExvJaKmkW3V7YVXBMiBDRMB5GE5JzO7rdYNqEH4JbtmqFzyZL1fttm5DyLKgydauKV9qPoDC0rrejvErZtEd/AZcG6LohBrMCueYZaXqshlE39lwQhbpvkJsxd2U6gWU1bxEf0+MVhHJoeUXVuHg5VWyB4VoLHMI5VlXiVxpVIFr/3g+Q2WNffmM7SOXWzQ3ofDRnuUg8R3H/NjCDwvmqq/06p3uY4AlwbalRkva2rJNLVhJdVjUstLVrXB10Cs2hu64Z1WLFtGtM8RClKRSzpnYi6EpcNKd5tgLvP7tEY3f39WuNXXWVe2qs//d3nn/7ynt1n/mLefV5Vd1grhBjxvqdiCAiSIEPzD4oLmSYktbT8q1hyrdAYVNJYlwWLFmLKedOkrK0a3X36LdOZMBhSSdMc5t25RyQ2laJqt1d8zyjqytuqYnBx6QSw84Lm+gYDZiqxuFTTMlJzFWTItYBLq5kcTPfvuh5J8hBiVGkFzXk7b3I/GLW0/I82DwCDctBuvrai+/ZFiWEIEafHh2awyBsIFdfzM7779ltP0hBi1GwpEl7GtUrafCJwyViXiOU243KR4tIokh374d174ZApIMRJdA8hYRsGbMoBb5cTRk2HX1lKDa6rKqK3DRb5YVZZCgHvf/KbytVEL2FpuKzW8aY5Bj2crRSkNIIIGKYJQXVGUkDHNpy8f1CFchqS6D/VOluDhFURkfoKqrFF67oAalBZFtHVuPe+1F4Bq+M3Q8fEWG435HXFhyjPsapm798/AgDyuuK8rrjNM263K7gyhknCCufbVQ+FZMJhjUFlME4PR4QSELJs9lH1PYM6+UZL+4VPI7l+q37/tpX22vPu73ntGa/14XUj0GtmE/m7giWWnpu/oBEd/1vF4K1UP5AiYYiu7dtvP+J2W9xfVSJKbtg0Q/O2rnj68AElZ4yjeAPcLmdcz2ePL44aoGBWUy4FeVwRavSaP8Y4e2UmQ/wE5+u1IywSV19rBrgxQkBCSq+XMywBguUAhVqnK7OG1k0OKHwWWYxGIELRjPBW5mI3T+uGECsOFDzYImgtFnEIl2TJ67KgrosY3jupEURAkMgyK/DGFLDMkhqvdyF6rX35TNdDUiOROiZn1mpys8r0ogO0PH0mHhAkfCiDQVUsuXlbQQAu52cM44RxOmIYLVefWJC5MlAyOG8S0WIWW8B1ELUU1xfWUkGQRKp5y0jjgOlw0NyHEu2xLjNK4fYM9YC3Mbp1SxW6QY0P5pZiXBPKAYms0FLvC8WePdj8DK1+ckOQVRNDSEWwnW+kzbn2UTi4PHedZ5ScVWSJ6rpRpcC4HkITpwaWrCMW7L+uC2KKWNcJ67Zh00wngRlUuxDJSqAETUYqpQJsbF4x7s4VpxGkXyx5wq+ia/w88aT95+ZG0lmN+6KCQhip0/u2gu2mBzb9WwhSynNZN0c027q5UY7VaXldF5Rtc32j6XrFCBE96Yen01eUB8CJIdcqHbVa4qqz7AuHuXXbJAnDs2Tjs6w7skqWm9P8B1kHZwWtBLEZeDSjZr3TyzfvCpdyLIKkkx5sjGIEiihb03+yIdtOP0+BFBSZuiCq7jJ+Vq3yxa3JpWSvqXq7nLGt5qAs3GldZk9lH2OQuq4hCAHKGcM0YlCrsimBr5czhnXBoDVD3qefus9cioT1+ozL80fkbZWoi3EUEWUreH56wny74eHhHR7fvdeqe8Ida82I4YDHx5+ILkNzpz08DgAY1xiwzrEZImBEpzoRytsGbHAH6nEcMY6TFMvRTXi5PCPcpBgPQfzDailKSCOm40HKJqaIYRKfxNPp1GJPaxXLWbXEtll1KFqbQotGBd0MWxYUvG0bQgw4Pz/h44cHGCkQpNySPRCaf5tlGSmaCafkiutt0RosGjerUS/TOCClgCkljENCJHGo/ZwusRd9f1Ei90NE4k+hwtd+fupZRhjFbUZW3Qi8JEAInlsTVJBrq1XMzMjr5saReZ7x/PSEvG2Yrzecz8/yDhIEtFzVwZlF98Y5C8HbsmStSckJ3rYKOChZCOy2LkKsVJc8Ho6Yjg8Aa0KIbcU6CzLctll14M0zARBPAooiVVn/a9lQANzojHVZMB0fMIxHxDggUEAuG7Z1dimp5RdlbAtQS1Bpb/A5DbEAJHpxLhUbxKVIgIkQfzlnErlmOkIjyFTUCGXRXW69Tjg8vHMj4uf21I9gTa5AIBADt+u1FVuvzUBS64h6PIFo8BKGpWQ106ubh1qIxcVmRi1ZkjHooAGJdeYgCGqZbwBYiysliZHMkg17vl3x9dc/xcPDoxTCXlh9oCS113R4AAJhWW4AKqZJqoIVE0mV07bURgVQTmguC0QEiqQO1klilqvEAi/zqjoZ8WP0eOZByoWmUWvBqI/kMIw4HI9qfdNccrWVJagltxjN0NWnUIfuZZbkEaXKRp1vN1yvF9dJmrXZuXynQyyl6boAib1dFkEurRxDM/RUTmrlrIAWk2JyuCE/7ijUa7pA30I/cKu9Jj5/TqRuhJDu/n7lDur6aPow9YmzOSb9V1lqYbNKPKVWrKumhts2rNuG89MzLucL1mXGcrto8pAkpR9WSUCwBU2XpehPsg3VFg9cq2crslo7pjIR74uie/kE870zg1reVo/vBZrIL4aJoBZeHT9bQoWKdZ4R4oZhOiEkDWXVvWIWZ0u6UE4PCCUiEyApvFgTKMicW310QJlItpA98W4wnaFJLCbxW8IH8eWEEDwjkDVq/WVx0F6X2yu7oLUvjAxFV7DoRM3zTRbuTr9Q1aGZwVLj12OSLRA+gAKDSGIeAw3qELy6Tk8sZ8KB59sFl/MThmHA4XQCAE1DVNw3alkXhFvE7XIWz/28anJMsS4HknhRVotc5upO10lLdq6LxIl6AsxOCS3xoRmW5VrcVDboVlARTIjIoKmRpqP4Vj48PuL48IAUk1cfM/RoSR+yukLULJX/uGostybNjZp9hEC+gdKoKBvAphEFMRBq1phqQ7ZKIC3WOmhol4TzZT18jcBZJp1aT5imwYlerQGECKsP/Jo2jl/5/VPX2J76oRblTxpDvue7Tz2H7z5kaBidGUJK1VDHonkJxX3mNs9Ylw3XyxXXyxXLKntN9s+zqztyzrg9PwlTvQmzWm4Sd8+6Z2pRdGWMpVassyBCMVA0o+JNPReCejJIqN2Ckldh7LUFO5RNPnNXMojlWYyK4j5DMSIkSe4AZqxKuEvZkNcFVvmROWCbb7r/pbsitURXi1hWncpFwYyk96IQJVIlBlBpYbKGeEOI2mcRwUOIoEGSrdSSkYYJp4f3Ako695vX2hdHhqKIvYglcl0c0get32txjHlbIdvruNMfmMWTqxHGCEuwmtdNOVeGuRFs24r5dsXl/BHHh0ccHx/BAObb2rJZgLGu4kZwuTzhdjnby1C5QIghIajT6XydNcWR+EaNpwmnh4fOQlZ1uOybsdSCAmiVuyToSwlQSAlNeNTktDHi9Pgo1e0eHnFUIg6wpiRT15cqadstuw1bPRYjhNV8DgZglFew5suz2G0GSyabQGA1bpnYt62rEEMNcQRBHbDNj1OiUoBWvsDCIi1CRiEqqlo9Ewew1Hp4sUXuCeKLLXT3+w8lhHY9373lUyLx5wjiJ+9hcREshbHlik0TrW6583CojNttwe0243K54nK5ePbnRYmhJT+opWBbmi4dzF7xzfZRHcxR385FxazJGUpe1dgBR/zLvCANg8byZmzr7OF3rEEBVQ1xeVub3pokm04lwqp6R6uLImgQ2JYZi4IAcAsXpEpYlyvCFj3wYDo+YDycDGxKPgIlmDkvAFjzjZrOPiKHrHOsNdBrEYkJjLIVZfoiRVmI6eH4gPdf/wTz7Yb5dnt1z1n74siwlupOy+ACQGqPVA1rG6dJuMwii25Gj5y3Fp4GoIXcibWOCBqhwjJokNyjFl9WIrkuC4CWDixoKFHJBRtWd9kZhtGLxS/zjE0NIlwrLuezlBRQHcu2rripmBtjglgL2DdFrXb4ROxw1xFVqtecEYMkYTW3H9+EUIKkxhY39GgqsKybQohPfzDNrUH+zhCxScTtgCGQx2lbxItnEO4MBYbyGCKaVHe+Fis8g8VAxEAu5rgrupllkSqB5k5CEId7MLDdKcld+a0iaO9+wx115LvrtWefFa1t79nPRhD339Er136umSGMWQggM2NZBQku6yYJWXPBqnpdc3gXN7LVs8lISJ1kfhGCUMTBuRTXvVlvbE+zGzqablikooq8LWrUEwLnLlma/AHMyFoL3MRKIYTFncB7hl414sqYm6QB09T6VQu1d0acpqpiF7VjGsCxZdAWw4bEHhNJCv+ShSnkVYih+SxCDTbBXGaUWJvLT6CAGgio5nkhiVzEaBJclVQt+84n2pclhlrIfBxG0CTpswIB6zyrr2DA6fSAbd1w/vid6Fjy1gVYK9LwHa11Oeab6MSSeLY/P33E9Xp1UW9dFwASznY9X1AL8PiVlhtQDiuuNVIvpZQNh+MJ795/BaKI5+ezWHwhIvjTh++wrgvGSfR/t9tVCDCzu+0Azb+PLLQN7HoVQMtvqn9VGkacHlUfeDh6kk1m4Ha74nLuMpVAmIDoRVXvolE3RhBZrcPmsxVKUfE0IU2TpG5SFw9Jr0QttvPO2mdiCMBaBmHBnBJSkuzgXPf6UekHafab7GoQrlKOoUQpYn+P+EVMV+shmTFC3w9dc/3uxd56Zb/1xO1T/+6vuyeI9++4f3XV1PRbFl3q9bZgWTKutxsuV/EHnOdZszELUbTYdonxXbHcRG99PT/jepZUcObQv2hcfNOnKeMrMr8WO2zuMswF6+2m+0qjiNzvTury1JIRokWOiHhcqlRHrCotWaEl5oKcbW9Z5UdVcdSKyptklC9qyFMmbRbtoARJ8mkmlzAqS3bqNIgXSM2roNRtw7peAbDG6YtkJURO/lW3AUjdcCYC1wgmqbkSYlI1hYjRUoyKkZ2xvN6+sGtNb3q3dFPmYpPBUapiMRtXkQ0U1GxOJPosCyvbO4V2iEIX3wrX5E0Dx5kQwipxlMWyP2upQvVOjyGhdGFx1mdA0RagoUWk1bs6Mz+ziwctxZISB0vBZbo7RV+evTdFj2Sx8DuCWin1ebtwv2p1W9RwEcylAzB1gxFj04tKxTXWkMRWPtFKirKWReBqYU9KvOWRAFs2lYK8StGuvAXkVQ7UbJEJyvUJIkpb4twYCFMeYJm/SS0RrodUnWWgRkwbEbRrGFCUCdwRJ/o+gkYvCKHp+uxZr+kOcfdZLxbbTyPalrI/q/P0poaSmguWdVZxcFZd9lWLixmSa+FlbL+ba5maT80QAwaK9SGQj9oIkiA09TeN5Ojf3F4k83RRw0lpRNCzy5t002bCfRV1bVmd9ksRv9PKXU4AJYYgArGIxxYaCsCdvENI/nwboyHKphu0Gi7mJmQueFCia5ISg1NLnGyO1/bf92H9L0sMQ8A4DWqg2BwRLvOMZb5KivNhhGWmAIBlvoLZDkLAkCTRo+niJIwnQ4pIiy/eOI4YxkHqOiw33K4XSYCZRtHfgLCuM1IaME0HQXPMACpIRXEiKRI/TgHH8SBinyaM/eonv4FAJEHzeVXiElDyinXJvuGN6AOQGtExSoaaYcSyCIqKmmZsGCecHh8k0Pz+oKs6wIhprYS1WPaOCcHiRjWeuJbiG9NEs5gk7VEkYNR61ayuCaeHB/zZP/fncD2f8fTdd4r+pP9JSyMYRy3bhhVA3TastxtKzapf2lz5fzo9Ig0D3n31NQ7Ho+h/mZG3g45JvAlgojKJ5Z+C5q3rRGATg2IkTKOg5SGa7hifVYh/H2H73mv405/3BJFBKJWdAK6qQli3FcsiIaHruuJyfhbp5PmjuNLcboqs5eBv8+zicc3iI2g6Q2EuaqUnuOdAiAkRVidWiKWlwLfSnFL6YdDoFo1Dvl6UgV2V4Jie18ZlxLCXDqBnbFTjzaaqmtU5g/n92bOEYUeUPPl8AXBXOgAYDxNAjDiIy9q2ECoYtW4SLDBfEeIAL3ZVM6ywfK4yf9s2gysr0JAXyYwxuG5gzm6B/lT7wuF4LYTM3AJYLaKmmyq0OfcAOr1MZYCaFanX7AiHMr2JiQTkKKZqkoIY1eiikRFkRdTZDiMhpdmTSOxqwFLANMmCStJeRs4RtUaH8IX6NGJt4Q3lWEhgH5YXU8J0OGAYJ48uMUW2c1FzFVDu7jVV+vf5wX3p1MpK4IMpvdWaWzoO6mJYrb4eDT2xcmVGVr1pDQE5ruqIPWu5yQsASRiQ8oBhnAAC5ln86YigTvfkFvSouRpF/GJUdS6GMhIzDKQqPqfMoq5wJAz4Br8PRGu7bv/zszv0Xqf0CYLYUKH2VYmFJWVtiVlbgtZtk2glSfMmtW1yziZro5Stc2TW5BqqG/R8kWREjzWMcj86d4sBlBE36antG9a1Lnr+9vsNaG5i8jjuxqxnYjcXe0d/Q2OisK4QK3pFUCTJgNqD2MfL6kdZQ3BdjwUQiHP4ilranjZJzNCihaLWklFDVFTckKUnpfi1ca2pFcvt6hYxyVrR4ozzJkHgghSS/wTE8AIWjrIsN6Qk4WsFUCfMiBgHxDSKrmwTJf+mhhhmMdw8vnuPh3fv8fjuHSgEfPeNJGj46qs/I/VLasVqzqsmqoBxOIz4rb/wFxBCwDc/+33M16v6/5Hv0W0VqyyzFGISHG8FmyRt0fFwwOH4gJoFWZ0eBZVZpmsw43o9SxD6JiFLFm0DmCgnmyUU8ZOUyITND5GpHmp3mIik7vM4TlLeIEZwKagQA831+SyhWZoiqrjSXAj/7fyMdVlgsdGtwtvmukJWgr3NV6RhQMkbDrcTuEq6tcPhgKeHB8Qklc9iCFJfJlr9maYkz5p0NgRCChKSxXxEjBGLMrPjNGDwuNyX++1e7P2F9yvvf/a/mwAmPnMyH6VK2c5Vq8St86zx9lKCYb5cRRK63ZC3DbfLRfwIs4TBLfNVDCaatKByQ2wgYdCBo9KKCkZVqVVcSbg2HR0xY6uqD7wx4rrcyRtCxKsaF038lmQGHYFTHa1vZQqgkh1QIETwZkSzi6cHNIcj/NwCjMokxFFBDQUJSCCKGKcDYoiYr+edcTBeznqtRJPUkkWyqFVcdmrx7D0EcSavLI7u6w24RNFdj4dpFz9937541hqh8Lk7uM1XzkKQQgiqPzSkB+UUan0tBay6R4PkADxvG8GQRfVQOwCePSMNA2LSDaQbQQ6kFmRSZAf9jkgMLcfTCTFGfFRCHUMA1Ghj91RdEIsfZvUdNeRlaZaiitZDGjAdj2I9pgiPXkFL+266VZ1EAJoRhU2311Bsrwyzftn7o6Ywa8+BbtTiaKWFJ2p4k1qkDclYq6VocR9xz5B3wEXaqn5zIUYs84x5vOmhJnWkJ83co648gRA5ogZBg9uWNaEEIYeAsVZs4wBPBsyEylazA3eY8E5r+AMo4j0ivCeETRfbXaP3Mbfynx6pUyzUrTgybEXQ7v+tWtt787kXf7vaNF4mhrJUi+wlgF0j8wpo4zLkJSJ2qyO0kx5w99OkIjAY4RVVAcuzXoie/OJXR7FVUCKD3DFaVAJZahKpW0zTjzJQ2dOJxWT6e0aTZGy/NmRY1Oot+swMq8xngQifal+UGOac8eGbn/tEb6o4rqUViPENUDMYktWGQsThMHgWaGZBYIIwVXcSAt5//TUOx5Nw1FpQvym43a7uwT5OB7x7/xXGadLkCloIOyas2wq6XvD08Vt8+83PMAyjxDofJPHp8XQQnZUSQbFCiy7xdj1rEaoNBMIwjHj31VcgiJOsuRuUvImOhivmWRy7t23BMt/UeVSIyPPzMzZ3gq2SDHYYXcUgPl9WUL75MoIldRRSBG/N4GKlFaXkZMV8k+I9VivZikTlbcWqCRlyFh3LRVH15fxRIw6ixqWa+Jc1Q7HqXEl8M1NKiMMgrhhRnGvHeca6LmI5X6Ws5fVyQYgR7969k/BJjR2/XW+4Xi6esGAaR8y/8VOMw4CHxxPGYcBhGnb7q1eRv/bzFxGXXz64OVPLZypqshy8eV6wrhmXyxXnywXPH58kzM6Q4jxjuYpP4XK9utQhSGbDtszYlhu29erv6wkdK2IrOesZ0Ip7JEzKgIMg1KyZlST6qJaCwqUZKLgRf8vCbQEIPlP+anL0meKoz7BEIWK0AIvlH4r6FEn47WB19ykZRAJyXA1QRDIkZa6mAjM3tsoa/pcDxmnC8fQg5UiZEVbCOsv7ahFcW/IKK7TGLMSzTBNyLsjL7bPZrr+4mDxfr861zGfK3ET0KoHcqiOE6YxGDWNTbutcpRQnatMklfC2bXELV88VYkoYxwkxDU4IzPpsafyX+Yb5dhFRd5B6rmlIGFKUzCuuXxNkQ4Ewgzz9EUBau1Yz3KglbNP6yVmdzG3RSrXQN6mpbMH5EhFiaMsMJ6LHC2oFN0XxDiGTWeXavHufVQ+Tq5WuFPQx327ap81DAU1/VFjmTwqFL0icwCE2YqiJPeWe4jpZ1gidlAYhgGpQokAYSnGd6abZSUaNnwYIERIRs2gt3tv1hmWScrB5mjBOo3sl2Dh7EfBzBPHlptwjGX5xVTNniS5V9XAqbVTVTUvM9oZ1WbEuK5ZlwXyb1S9TUvQXKwlhaoiqyXNLS0klAQMvugAzaJgaxPpLtbpnhASCaMEpVg8McUkWwuvW2kYMyTIg7azI+71jP0VlFQTZAZIAZUdcSOfLZ8y/sSQSEjkWnIGb9EbEYHVds/eZMcYkSDKpgityjKiayarvQa1FM9mYaJ87wptfWd/WvngR+eUmaetDCAiqRK8SVCiEL0pSRsncERCCWIssLjMEEWcBgEjy65XcW4rUylzFNykNlmI8iAcCi/X5qtlCxnFSy/GM+apwPEY8PL7Db/7WX8DD4zscj6KrOj89gWvF7XrBOt/EOToFz6uYUsR0+BrDKDHVVZ1PS94kvE9R1I0kGcK6LhjzQTpOkskEDE24yZjUj7HVfghaHCo4McjLDMv67e4ZekBke5EeJvFvi6UgxMH1gcZszKBiZUnN1cJEt3W5iVM1s8ba9uKUWfYlTjuWLPVs1hVrkFrY2yCpwKz+Nfl7RbWxHA96Thmg0V00tpxxu12x5dXrY59ORw0N07yB8XV3GtMZfk53qPxEmIMSJ8uwfm+plgw0wozneRGVjYrIt3nBtmWcn5/x8cMHnJ+ecH76qG5IokM8P4kVeZlVWrDn6u/cERN3ekYj+EATbW2qbH8JMdQs0FwBkjKfrEZKV/3YuElGLwZELXBfwwuCaO40YvEWgCKi7t5AZ0zC71P1ihk6uIovpCQfSa4aK0X8FI1QS/TL3Iw9zEiqSlvnGy4fv1PfzKvaHJbmPM4a1NDNV8vMbYDrVzCgENHfB+B/AuC3dLS/y8z/PBH9FMD/AsA/AOBvAfjHmfm7zz2La8W23DT10KDIh3ab1kKKaslSXEYRXNbDP05RRSmro9z0IOZF5sQwSK5Aq71susRSMuZZXGKOp0fENODpwxnL7aJ1JBJOpwf8xm/8GUzHo2TdJsLtcmkRBOsiyTspwULg4jjg8d2jphpq2UQsLXrW7LyOqKrVGhEOri5cGuxf1eAxasD+qsi2Iahai9ZhNst51lTojQPavFaWVGk1VQwhIqi4gtCQI6MZjcw9QxyGOwdeBjh2HL+2aAhDxrLxJFIlblaydEWk4Fl2ZG+Jo3yMScIBtbaz5Ltk7Y/4L6accB7Pqov7DY2U0H3FeJHsQXevkBYCwPRZoghmFHUuDjDPBHR3kPpfVqzbhsv1JnNuTtebEMnr+YLz0xMuz0+4amjdtkqlwtv5uXNbMh0WeTYkeY1WirPBdSBfBYAdmuVaUdiQoRkHuH3n5RuadlW5oEtofg7V32knniujIGWoHnnS6Rs9hRd28l032ewp76ysrF1fy4YMdsRq+9hyftpn4IpN/TRL3rBq6J5JfvtIqc7GzlXPoZ2ET7cfggwzgP8qM//fiOgdgL9ORP8HAP85AP9HZv7niOivAPgrAP6Zzz9KxbmSUUndONi4IGsVPPEhLKoIDpqbsNSKqmUPY01e4yOmhHGaRBzbFoQluPe7OTRbbZFSC663K4iiIDFNNUXEmK9nnJ8+onLFdJACN0ZOCOLAfT2fIfWFJYEEQRS2ctjFdWS+albezdI0XUQ/CXFg3lgs6PbsEMStxxO2lsYgas7IRGpxlywy5m8G6FGvLVrAM9jYxtfDLBENwLouGokSwTG5eE0kaHRhidYpGrNqCXi5VmRVO4RQQbW5MTH3HFc3NEs8askZOayYr5KZaJ0mbJoPcVD/yDSMiEPCu3ePGIaEZbHojHVnZOBacbtcwKXidr0hhojrVVx3pnHAOA4+lr05hVrX2H40VYyoCzSbzJbBAGIJLhZGzdhSWb6fZ4kr/vBBUN71enP0l3PGH/7+7+Pp4wfcrhdcL2dF6ZK9RSJPJPa2z2TeHPT37in+G5vHQgChwMwaRuBExLT6HncH3oyPYBAawTT02xtN9u81Y5EQkb7OiTrOuGHRiJ3lKJVnGsPjbk82Mdj2KNXQjHX2XkXoFm5Y8oZMhFA1ICBvHmrYiLz1VXI32qc1b8jrLH+8zjG9fS8xZOa/B+Dv6e/PRPQ3APw2gH8MwF/Sy/5FAH8N30sMAeKCmjcRV3tfNkgg+DCOgiABgOQzCbwWi6oVn5Zi8BK1MR1OSOOIdZZwm6hpvzzVv16bc8H5+Yw0jDg+vEMIijCJcT4/4dtvfo73X32Fh8d3UnZAN4To/iqePsoBmKYB45TaYd0WbOtNQ6UqrOhULQXn87O4mBxOiJptO+etEfNAGIeoaBAoBE+BZXrNbRMls5VY9FZaDV2ZLgKIdeMbLWw+ZvV2lSS1QRPf6rwTgEFruCyzOcTf/IAww8PuiLL2XcbYtmHjxiIyia6MAOQt43Y5YxhHTAexnE+Hgxi1xgnDOGL9yU8wTpO6SZGOq6Jsrc50VV+98/kMEGGcBhRm8CM03h07S+q9fblHhVytbKcmVKiSaJVZ0swTEQ4Qb4GqKcvWdcPlcsPlfME3f/gN5tsN3/785xLKOYu64sM3P8fl/CxxxpoyqsUMi/vTqnkGWdG2JRtxXWCj2l1T1NfMxE5QbL2NMX1SGHQE2ODmyzT/Jl21tWfTkWbrnRC2NI4aHhdAVes861pR24AdkJVw3BB6AqYZkna6bjWKaN9ylqQtoWTUoBEpm4V+iq8lQZyN4P2WecmbuZYFpDTeT+qu/UI6QyL6BwD8+wH8XwD8lhJKAPh9iBj92j2/A+B3gJaEofdrch04NVgP04nooODOty1MzlELA4MbVwpALWOK1ZJ1JS9r0HmSVFiBgMvzR33vhhgJaRBR1PLSlZIx367YlkWzgayoRQooZU3hNRv6U51FCAElpGYNhip2Czo3IlF0r8uM548ftQ5yK3ptTtUFTXktuQwHfZ6EJcUYUYvpENVZXN0o7DwJAqyo1MTX3m6w3G64Ds+qh9k6g0gTm9qakWa0FhWFqzhsM6Od11oKMsi/swMXY0QpWftePH3VtkwwGGdZmOf5pplz4Pe26n6aHqtIHLY4lQMuJuq+MphgEqasa/X7ih5kSV4K1CAibNYEFMu6YF03PD8/47vvPuByPuO7b36Oeb7h43ffYttWrLOUdT0/f9Ti7eIpYGfP8k3K3s3d32I82PmGuvqkScQWD+z7x9engqtlrw6d+cKMYO6cs7tPk0TtkGFPFF+jpraXzfBo5SV8n3Mj5g4qHf0q66WCviSAfS+GN1I+0Bl5uPXb5xLsSLpnCP34KovPpIj3m7vMYffuffvBxJCIHgH8rwH808z81CuXmZnJ8PCLCeTfBfC7APD4+I5beh/AzEAmzVtpTxNvCRZ7WBGHEaHzkys5Y61F3DROjwghYJ0XbMuKYdTyo7ebblLNo8aSGQPDgMM0otaKn/3df0czhWwYhwHHwxGnhwcM4+i+cusyY51nfPfNz5C3TXIskiR7zVrEe9tWCXnbRJmNuxyMjiJVn8dFjCuXp4/4u//fv6XZakYtsj24gYSIMEySHftwOuH0cJJIkJyxBfJa0qxuNeui4kVQXaqqIcRRe0WlgGWRECknELVgvp4xz5LmSDZ7E8NMldEj+RAaSvUD5TorIWjmTA/dtK00qtV0iTg9vMN0OODy09/QYkCCMNd1FSI0S60bIsnYzcyYlxlpHDEvK0KKmLYJU5bxJwo7cShQ05kC0MS0EOtvLmoUkfm83WYnuOIiVJFiwscPH3F+ljo9f/Czn+H8/ITf/73fw7rMbhSZbxdRZ2haOnBnLc1acMyMWXLmXd8lc0h6INhdm9yqzyot6NSSnxoxSJu7FVFjhoAmdYD5Kgoxq0WlhhZsAvMDZDb9HLdKjJ2YXXYibnuvIbjaiflNLwwHL8aIQmegcuJMrSDaPSV2VVpQ8b6weot06NVcjYyZVImVFknG4vCbmP5a+0HEkIgGCCH8nzHz/0Y//hkR/Xlm/ntE9OcB/MEPeRa4T4sPmBhh3zXFpx1ETUQAaJbrKKl9VBdTghoMakuakCV4wq2szJKfTzamKPSt3KFl0SAQYqS2OUpzGamaqcVicMESF2wVyrIqfEnvEd2OOY2rcaJ2YogiNX9Pl703qE6QQ0BMAIJY10FNU2h7pd8yDfXo5mHb5J0+RedUEm5WEBvRypphxg5nX/dX0YUjEXT6ndeWVzk0Qfts0Ky3ODY3jrxJHsXL+VlpaPQ+iY+e1AAhrwGSvGRBLrlDeFWND92+tXkBvJ5xUfHY7vHf1TVG9mZbsxgjLucznp+fcH5+xuX8jOtFyr2u6+L929YVpfTuMU0f7noym7dAjm4aMfRV7Cz0nVXZ81LawreB9mKordV+zTqkfDdB1N3qawd47Livub+33eDXd8iN/B5uL8H9fY0Q9vum/9u7qHvQE7pYVJht9u7e/p/3r2Pqv7IBhYRS/QsA/gYz/3e7r/53AP4pAP+c/vyXvu9Z0nHxWRPRKnbFjpoljIlQYZtAZniYJsQYcDgecTw94HY94/x0Qdik/kiICeN4EPRRxIq4zTfkdRYdpTpkZg2en9VRlWuVqIkUpYRjFb87I6RSm+IsLjHXZxXja4PqXSyqKcvFqHMAxaC6OPIiTDVvYkDSBS4aShRqkLqxFJDT5r6KwzBqwQqJI3UHdd18LhZkJaqqVK65oiLfzz6YJaMw0JyxmYvoPVUxLU7dEjTvMaRK6FNKnmKrkZr+EKtbFFe3HHrIooryAMQCnsXBdplv+Fvr35QQy+mAmEbVDQ+a325VvfGKmjdcnj4CzDidxOXpcJgwrhOGlBAVrRrdsP/lrWraqIJSpFpbVkffnCVc8Pnjs2Z/XhUVCkP75ud/qO4yH/Hh22+xLDPOlmzhctYoHE2wUDatVFg8/6Zld2Fn9uRIy3V2O+JmVuBO5XBHJzwEUQGFbIe7+GAjFoDEIyvHlL8lWQYFY1fs/weUiTqIMqZufxmzbc7NnxI+m4qkKUF7nuq0lsXvtffjtdC5qgTQqlVaboPdGO0c9KDDAADx99FBAD8MGf4jAP5JAP9PIvp/6Gf/DQgR/F8S0V8G8LcB/OM/4FnaYVa4jN3iWCO0UoNWz8QtbaabYqvFwAAFRAYwygSYnrCUTVGj1Z0VZCBK/c3TakkihaCHVQ4zZUIIm4rBkojAnmUWMxsP0PRltukbErODecf5GgSGLWpFlXUzJFF7dNH+GbruF96Iq1lI7RC2Tezyjh/OoEXOa2F3fhVxJHT9UpSyQx22+Wr3e3unzw+JJT6AYHVL2bUpcnRLzqihYr5dEWLAWCrSkCVxBcMtilDfx1JMZ6iFf+r9XECsq/1UMzsKzFlTkJWCXIQZbZtEicwaMzzPN8kErmLb5Xx2RDjPV43W2dSqmTsJpLyC6lxjp8Sx7fUeyVQN+9ytOwuaeWFLAbrxybf9VQ1lKnkjqPi833+uTXDBjHfPvH+WPuoFAeJdfz5HGHeqNX/PPUJEZwTbfYcKKq8jSLz2mY+mzeOne/fDrMn/Z3yarv6Hv+/+Fx2jphuxFPlpGIFxdF2SIKYuPRAYSX395hghae+vyOviyUgJ6u4SCB+/+QbLfMNyffZMwdUOQykYwxEPD1K/eZ5vgg5jQBoG1JIx5w1xTdg0+ebt8iwWKbA6p8qhj1qGUIwVwa2ZUOszM4O2zTdii7W26JEWeUOAlj6AEm7xt6JN5kuc0NmvqWpsEh/EzQ+o+BkWPQCqj/E3WFIBIYY5s5ivfbElJpaYsCksMB8tE61F7wRkrJ0juDhSW1ie+CQKkxL/RQKxWVXlHnHJkCgVghb90WS2ppuKGms+HSZxut1m5LJhWWekZcCWs6+rIy8W1Yqd56LZdtZVROrr7YplWV2vlLeWbv/3/vbfEqnhdkOpxfMynp+ecb2csdyuuF7OWrhMrMfrctWwxM2NG+7470hNyGE1o6F61ZgKgzvCQwyXNgB06cxsa/REqnMh6pcR++e62B8aA7ObTNq055k+z95lIipDkKIZ5u5euOtbT3AaCmzEmJUpex7SfoidpZtx7yrVcLWAxqZfZM10VD0KhyAZ5m1e+e5FL9sXTuHVmhya5pQsTtKts7W06A2AO6icPaSploKAlsrcXFIsrG5bF8/2YkSgVlEqW+lOU64bPK+FVenPEEfPxRMUxNjqtAAAXOmr2VZs06nYI0REA+Q7htSUxC8Xp3HPpm8qSrCy6qZs/iwBQG+JNFTiYBANJfmTFfXtggZMVNNIoKq6FiPq7cByUxEwu9LbDSzc0jkJ9iPA9F0UwFR1I8eGJgGgMJiDovKIOqhLRDIH9gxeG4L1ynAdijJkCEUCJoXUKslWJdRRcg2aqLVpXsHb9Yrnpw+SbuwmOf5ilIpvt4sYl5Z51prZ5QUqtL1sh3ZHkBxBtT7uiFoPrVyafYlg9gjKNsz3S4Au3pJgyB044vtr93vY19z6Sy1pak+ienH3U31uIXYdXu6IpQ4H7ZeG6DoSDcOU/cBJ8V+A7DGurVZS15nPTdOXJ4YMwDI0V9KQpDsmY1yKlXOB4TqrZb55BhVL0DqM4rKz3G5gMLZlkRCodRGHS8t8XTUV+LpguZ5BIWBbxZBS8wQOwd1xSt5E77gJCgBYUw2RI5EYdZnYuJi4/7ASMiEaewucsUKKAYGimvxlIU15L2qvICitEjhXlBKc6JFmvRFr982LS9kBZZawOK8/Z5BDPyGIRRU1e79V/lF0CHd87RXV8ixFN0rkuZIyGBOp2yE3omsbPQQhnjLWskMYxtBKziCQJvltoj+BMKRBi6cnpBhQcsYyz7hdb0gxoUwWpik1txlQpFxxvV7dR3GeZ2zqvzjfhAien5/ws98XC7E4nEvW8xAkya8h703DHzfN7m1O6VaRkU1HZkPzg9+Wfy9m2mTZ+dcyo3q4DT2+QGN6VpyZGEPHnrB8hu55hwjwOja2FywBg/fTQAA3nOaqIbTr9m9puka7ViKLeq/UDhXr8L4nuUx7l1PFrhcEkUJgTkavKRleb18eGXK3WB0k9q9tcY0f6GZpCRwl5ZSlSjfrK8BSXpTZy1pa4SKL3KiKRiVqZAaRGC0YGh9aJHO2IUOL9922De7MrP5cgjx00hlgrqAYPJdcVYX1LtQKuuAqTovf5D6lEuywkCmECySUzCqXWbqxwZXK5rJTiiHDCmJNkgndoAwliP6yxvF3m+9O76drQWRjZRf5Rb1DEjbTL7CvpR2QprckHdcu6kAPHxGJkcjdNfowMrG0W0RRCKGL7JDECETwOHRQkr2iiHBZF6yL1CO53aQ2ybIsuF3P+PDdtzg/f8TTx+9ENaKZmwOJ72fJLe7bkizkbVGVQW7j6dFgNxsMY5Z7JNgIoRKHPX1o+jkncr6LXpltOEXpdXOkpOflxftnfSrPn4+l3zodYnzZjx4Jtnfs+2Sg6O5N3MTwpufuUCC3Gfa+MTrC2j0bhhZ7ivnrhAyJQDF510zEETGyKnEJuqkHcA2oWdLcl1yw0SrZoAcrOC2bcFslvEmqaUF92CSQf1s3VxQTWfaXgnVVYqgw2lLT70Ufc+5dQRSQYgI0LZI5VytbRh/gbn6SQtibMr2pBooTlQpGzsGJjaCKCOoU+ARo0Zvq6gLTt0qZScvmm11vZe8htLFTt99JxZBduFRHIGupL651JsUAzA/NJ7c929RSzaABmDXRCOSnlODmh2lFyO1fKRmX8xNiSnj3/itwlVrWy7J4veHD8YjHR0kFdnp4ACDZb0ouOD+dMS+z1CleFizLjGWecb084+njd7hdniX6ZhNXGWbTz1rVt+rMx7MnK1MxRM3oDDfUUKD73Nk66JyTz8MdfuuQlP3+Ain1FGJ39x51OdEhcr/eXsb0EM4d87N+SnRWpbpDuXcU2Gi7r7lISOTnANYHiKqJGY3ZdYSyqRCwu8f2hh41mC6zZz0vZ4junvH96PCLI0OLIgGAEC1pqUZNaA0HcXVJ4BqwxQQgu8+fEKI2yOqK+4qSR8CSpyZJBLptki23ckXSFPu1SgYXQZXqm7gtrruU4O4mdua8qfh70OSl4pLiyu0eZQEwXSIzI7Dq9NhccZpeTZAeo2DrjDACH7zuyI77VhStwWsWc69OVzQvZDWHdlHgG4Hdc1yAqCEVT07rZUib3q/5gLZeyMSTejRb0SYnu68SPEOSZuDole093za3nKLjqUWquG3bisvzR4QYcHn+ic6f5Ei8Xi4YxhGn0wPmeZHiVjrmbZGY4efnZ0WFN6+lPd9uuJyf8PzxOynOpKGIJW/qyyIMqqrKoJTcMsHUFv1gJ9jFSTQCVDXpAgEeD9wW9BVCCDnG4f7avu0UZo0gulCoEkZ/vQg0tFunO8sF7gmGRYVYZMi+64p3XZLbbVWYE3VDnA0o2N+1dn64tg/92XZtz2V7Qrg/c15GtLtfRqViv1HSz7QvWxAKFsYF2CGQRWflFkWArUVOsBhRdo6r+jdbTC4zSpaEAVIyQPR+XKTwe4xJPP6LVYVLjZipvm7n9P3KOweNfhmGwcOiYJlF2HLBGWQnjzsG7mM/e2U/RC8T7D0E0uSoDEbggJzFAdmiASSRRfS+yaFsoVC74+F7249Ic1O5O4DVxDzdbE40e5GOLP7VgGHbqGyxXYbWXw2Ibwe/15m5uI09kirZyglIFuxtlSS4FAjPH77Fukh5zWGYMB6PGKejFF5nSfybtLjQ9SyZhs7PZ2zbiptWpLtexF3mdpXynFa204iw6Sn9ALoLT1NXgFs8bDVC2BNEUyuoSsBp504evkd2+0+I6MX3tsZo23bXqNsIPaHaueu4zG5LvGdeTcXxSh+U2Pbv3fehN2jxi/v393D3e3v+i77qXILN8LJHpBWia7VW1T/WxiazXu9me9++sJgMd8Ng1pCfSopKSlMWE6EGy4ahol8BwOY7OLiRo1ILW8ubZNWwQ0WaFJY3IUqW2l/qkYxqAGgH04mLikFFM7xMWmfYKtEV9Tf08oU6wyZiu3hAcGfyvj4Juo3J5jflm4/UUCN6yVqtXKmGKA5wYiH91RRNkH+GmgkCblzE5ea76UeEzN2lQ4L6fXA1hJ9gCODUUC+CR7Ag9HKNoI99ySB9q4pHfahhO3Dydvt7W1fc6IyYkvp6LrienwAwfs4VaRiQBlmX48N7HE6PeHz/NW63GQ+PjzrnFd99841GlgjzuF0v6mD9LT5+9y3W5YrL+YMm4J19naSzrfCQ9L1Tdxhjs7nVfru7zCs/mZv9lcHNEXq/KgLafErbPL6qWtBrej2l69t2xPJOeumub8y6nb/+b1svj5ahfr++bDZXQP2ELrKpE/o8iC/F5Zbq7T5ixdflE3NjUk67hqFE5NU+A1+aGDJglaoY6i5jDtH3/k9VF9jFSslo2wwptlh6kCmAg6bbaiy46Vw6eC46sbKD4eIXxg11obnbeMU9J5jq1sFd1mEdn9WokD+UNbv3/14YkepmJqs2YsiVgADvSyDJzQEIYTQfwxbn3YnfnfhC3RzAuWojljrwOwSJNlabHt8/7IxGbhVq2/Q1DeWjv63T37wmyvg7vUvcxqjzbclxAUluUTytmUT4FE1zP0wTAMbD4wNKKTg/P0nNbJ2f6+WMZbnhcn7GfDuLQ/26OiNzJGfz5dPTUIkRwPvW5r5Hf3fX9HPeXUcd8bqjka/OF3dr+Oq7XkjkHVKz7zvk1BNK1xtX0+m9OhQnUEa82uf2zj5uun1m79gz5/0Z3fe77u53a/0r82LXtPFY3zqa8In2hTNdFyy3s/+d89qSKKDpGYQpW0YMtewyA4WQ0yroTJERiFBRQIEReQCoKWd9o3XqkVZSsGjyV01DpGFuZjwxvV+0ou+aUxEs/e71i4DqeRDBZvQwaA/ex1KR+SoWDz8KLo41J12qAYVWIXYaelUAbItm792EMJgDe6+Il7F2GYt9w6ojb5f9h1svEag5+bp/lotadmjtubaxLMGuD7AdeNdLNULb9oI8o+mM9t9btT9P9MviQcBgnDcJpQyq7jg/f0RIIx7PP8Wyrnh4fIRl+/7mD/4A27b6HJ2fvsPtesY6Xz3jtGQ/skwxzQBh6KwdWSUe/lvf+O6/fSPXk3KTRvpDj+a3d099euQpX+8Zx06cVULRE1STPGzp2P5gE+8b4m8ouDohE6PHPbLszlTnJG5qj0+JxtbMPa1/lvJWNKba1GdCEJte8KUuGyq6G6rs50XBTq2vrExrX9yA4tERZIef2+T1k62c2D/jqoSuuE7PZq+hN3FtkSi/4J+9wiadu1toGoA9x28r07HglwfgtYPuG74TJ7gTqxwFGsfibrwuKhrSE09612Gzhd4Vf7+5qwBtUzWk2ZpttB0yVLj6Yut2iLCpl+7e0c+BncBuSO3xZB3q5r8/MP08MsAtkYFVTGRu0kD1A2MEPSBUSOmG2wVEkpqtlorb9aw59mSO5tu1c8hvPoJ+eJhfUMGeqNwftG7FfRz7A+cLtz+492vzifnfv+dur93DLXs7i5ztH1uf+yfo393m785EG6NJAvfI8ZUh+HfGaO9F29eulWH0xN3G2V+zf+/+ITYXOgZL7Lrrf3e8PtO+cKlQCb8KsaXgNyTi02E+a744YhwwZ82yRWRHf+JeMYwTQoyiB1RLsljyMmrNKl7JQ3uRWbhFRq17nQu4U/qyKWab8rW5zgghJZi6jBVhmA6u524tTxzBOLGKI7lNAIElvRcRaiQE9W+sRkRhWVea/qPXE7a6E1DfPbgBJoSIphiA94WJEDg0IuoE2vwCqXtXk51t/Fb7hIISc02j5YeGAELnjE5kigSbYkFjRuSIBbigif4yh40hAIxAVXWjUuPi+vQB2yLV9z7+/OdgMNblJvo9FYG39abZeTJqNmfx7sA5cuuYlW2F2sXM69hehNR9Cnm88rH5wZEtYrf/7Jb733z/vmJY8bIMHco2gK6wyck1KTNu2XyMIci9Xhe7mrGoi63uF+6VsfXMA4444f1vDH/f+rRffv/uVV1E1ysAxK6zc2Z/G1HuXEFfbV+YGEI4fMVuMen+IrTBGzJ0FNklSnDOEtRKrQgxJisLEHfQev8mPbhs20MmMbzar0aIgCYKeEd9s/WHpEWemP7FuSXQjEiQTXwfRu7c2/wXu8PXz4d1ISg1NLLVNkV78l606A5Y9957rt8jwJ6Yy3dkZwyWyaS7+uV86xwaASZ9X6+HJKPoPUqxebx7ciNk6vqyrWBAdY1ytVVJrFWq0eW8ud7aPRTuTogTx3uUAu7m5x69dPvDviDXMexRWNdMV2jahPt1+MRtdw/ZQfVX7m0r7HvOe9U6uF97vrv25TX96jYHZzjR3d3runnsPn2tz59qryHyuyt8v9grGrL93sd/aTGZWxhWv/qkHMOsrTat3E+ScqqSgbVxSKnvkYHIIEs2qgc/xKDV5RI4DQix1eywrNK7KmQM18/BfP7AWvcELnI72eL94jNXUDUxqRFOwKzJPbVgHz6UU7sfmiKuUiwoq7+tE8MYsHq6lr4pBgl1ZM3gLBZfs8bR7ty4fyEA9z+zjcNNlQDs3Ydgs2AuUE7iSOd2TwideJM5fofuvayGGnPp6ZCnfV9r2+RoBhuuQCUGkVjSq3L/GlspBGY1ilhuSmeke2OTEVz5u/qbfBiEhgxpTwh269LD7u6HTqJdLV8FHYsaGCozUPu+9A/YMxbu9s+9uGmJOHY96eRFo48C3IxZGyFpTFeH2jHmO0ZgjLYTRXf5M9EydO8lspdzJ/1unhPed1Kmq4P1OWKgd+5uM7P/aRLl5xChtS8fm6x6sB7H7hwOOpM+7FNfiBbra36CQtCK+ABXCeUCeLcAYgi5N8/vEYn1Rap/NbHTCTiR+hh35Mk2V6eoc0KDZqwAmkJav3xFR7SjUvqDXYndzwc6jt30OZaUFSAO3Zy9jjL2BGGPLAxa2gFnZkekPcftQ+iNebkivA3MCaYxICN8NmfyLmoP7rrRdUh/7b6ntgYinleANbuOz5USP01P1ovFjjXvxgq93yNs7Czypw9Wey7dD2N3+Q4BGyq0zM8A1KehWytuDyN9U+OFvla7eVOiRncE1PppUsxu7P1+8Wl9OdD+s0amSdUV3N3XnS+iuz38+jMNwZGBo/vr9Ke9h8BitHwFMPg4sN+3n6OJX9i1hpvjbyeyVSMyOxh8/9M4hx1iOYi1ZgmtCxEMQigZawAoREnxr+jHcUsnnkkweuOqQkc1agOWVIDRp3fpiYm5+PQ6nPv+t4XusrkYsaB2qRBbO3h6T5UkUC0d+r6vzEDJMh99LDHBfOLk8Ho6qBB8o5kLEndz0y0Tmq5Otpo5mLso/tretkte28jcnVlmMJtu2CzokltR0tHbgdd50LHb+LIqDyP6MEZ9bpVi5e7nqMkoPKdlhzjtma/7CLJKArxb2yY63h9ANELZrVXbDtytTydJAOAi9/WuWr6PbOsRiUXamXRjQhbhsuuHNvcjvQMYcD2aRjRZcEBtc2RqI9LaCb2/YUdfAEjtmcrke9xarc3Fa6deQvubdb09pRdLfH+zjvfggO+eb+cPvoY75g9WuvFpYmztyyNDPeDULWrHUO0Tu9p/9tzDJoCZXHSRKngRHCK2hSQxg2Z+JnQ+g94RdoIGiHhJXawxWR9Zs24zYMGdrrfzJKbyPttAPg7qxmAOu/3IuBO6mXdhTyY2efxqJ5o5qujmwyMjiFC6yazc4oxjMFeJ6ByzIxv7deJGKPqN7yIXevTx2hMapTfiv2N2/vPeT7MxKBs37nLsmYuTHVAiEgOwEat+Q3WlKz3cq8ecPTLybjUG1iN27u7Rzu7n7G7M9xxGPvKjDbCIouz/rI7IPX4hT8PfiG3fl4aqdnd1RLzU4mK/GS+FAJmYLNmBLHwQ+roaAojhSZCNuNpzmw/hS39EtjXx2eMX/bLTXnWvEUt2I5fAdPPdo9TXjCzAS2J4D0o+1744MdwRpL7ZJvSDZ1Sh+8ftADaxBIIW3R+tSuaYGjwzdTM2cHvHnW6P1WId1B3HCIGjSYIgBSNQ1m0jhvLHiyE1AibIr2K/MDaeFyI7GqEDCQry8+D0pEXeO+cjNPSjb3A06PobuZB9vvcnt+mE5BD4oWmz1dDY/UD7tfPmbE9/a1E4L07wLg1YS4jbngKgtrXhKok+fc2KRT3IwW9W/bYH/JDpO4wIGjOzzwxtNjHPNudLCcaJKAXR094TBh13z1TaHrpjFN23TaRvaKk/H3Zvv88aSuwJZSNEDWGxMxDb727oC62mkACWPQED2NO7OaEE3AXM+AXu+tK3FjnT4uap38s+d68TMlmzCrDYBz7VjIF/Hzb88uF4Plv2YX9s7Zeei+zv7y/xA8HidF3KBqp6cChI6J47bCoP8g2yz9IhzyI1OKjIZI7ZoYljMM54h3LsAL1oPU0nkmS0+rnTHE9wfcflagU0JVgTz9qD7x1xZd92oqDOVytt0Lg7AAQmNO3a3TrcbUyLxd19b2eeuvtsbN3zerHOkOBON9xd37skydyQlHRw9UaHZtgkBJnDWqukYTNRjQBzLjZUf69H3REk/bx0/QKpW3k3+czGcLrrfOy1LWgP5+/nuCMuhuruWamJj9IqehzexrHrKoBmgOir1vlca8xuKXtxttR92YhIaAYe7/NLZs3MQCBECmr8ECNlcXGWXr33ThQEqOXffBGC2N0aQsvODWghMMCzKL2IYmHaLcPnCOKPlunaDpEHs38PjHXFPBma6jmzwv5ahRjUoD/7nHh2eduS906hvW6Cmuyy7/Q9kPFO7Q8XCEpY9wQLkBSrtkFpd0nPAHqRy8bWPu/FOyj60+FJpl/dQwHkOR9fbO5+HDAe9JIJGUp6bYmaqOwT0ne9/WJqDfcw3BOlxqRqh1gt7Ln9Lf83B3YhDaQJGSWPoekQ5epWvtI4D1t3HHn2XZd10zFb3XanbdaJpmLY6fYcLVf4CXzRf19Bn2cRlXuG2lKrod3eT+grSLJDfby/pj2bPRKmR/+744G+H/ZOu6fNVc8o7QyDVHdYX5Ic2v3S1n6Xs8jXo38ndvO8u74jfi2SqSX/6BEz3/18rX15MbnbgC89z/WaHk7Jt4JstO4BdZ+33xilZhffAlGLVOke5ojDHk57rrUTLcwJNMWm/6OGChoy4FcIpZyql7qO4AeRuQAM1NCJCT43Nj4JmrcN1ogS+zMaKmjqAymizYgUkUISXaHnrrsbs/4zMYmIkOxamxMSiLQjkp9gYE4gewah2bwRqu7jvaoDunFbAgRbHInA2emZapWYdReHRVQOISKGpL6oWle3E4+dl+zPE2zHeXZnpWMxGANpy2z6x9oZmNogO3RCQqRdv9h2xQ7FCRFoKd6qMshAhL242O3Le4Kmu9nqVpuO2PavldbsUVbjU9Yje76Q8sBNfRD0u2YbUaZOwQt8GZ00P90QLH8n699ht+bs+ndYcJlLN7LdGmJ2R/d+JuycBJME9zSlR/4yQgbfGV/u24+CDBua6DlA45QM05nvRc8dDaT2JCjHs+u5VsmWfOemY0TEnuYH/b5/LxCD9e01z/m2u2zziGhFDRXYG4m6TchOZDSa9MWC3vehfd5xSuOcXX92fbxHeGizaoejOwrdb9xQez9P3Uluv7a+GNp/yYPbWvcMyJFtd8c9KvH+kLg3xRQQOGhsckAaRqRh6BT8QF7zzgDlKtU70ELdh1JbzMbMLRVbR7C53dU9oWdH/T7tx6gICrp+frftw/1U9ajL5qtdw9gTQyjyVqKlc8oq2eykCG7X2w92Qg4Zc23Pa8SJduOwwZr+vl8xezDpWOVP6taAHA239bD9b+oe8jG0d3dAwTfMHch55dzyKzvrtfaF8xmKyFGLcEDnFtwsfXIdtV3bML9+yz5RPcbv9SsZlt2FYenkQ7AaIur8S8FRnvftlUPMaPVcmzMod/+6Q9yJIrsU+7sH2sZs4rujWUOIwTgv7QmoT4vzOq29y5YuYTfZkqdZmAOpm5ARNuHaRWOo0Z8KgJvhxoCNl3d15KOX9vJ5P9zuGoJNhyGWliDAbjBRnnspWgm96QZTEqX+eJBUauP0iJRGjNOEYRxR64rKM7ZlweV5RcnAxiK2NZtK0z0KwVMvAjQxa93gdZOtvEDVw2hGILfMGlFRwtMQkPFCVjUpC4oidAhF9mB13a0y9v4/3Qtc9glNGzFsxjfW/hfPBAVfQGeajF2dciN0pCoAqfMNR8GylHsQYN2PUfcC11b5j9t6yr3VRip7Mlg9IkWrMOOMoer7vdMBA+pSwzmD0DFrYbTe4t37lbrr2mfaj4YM95yt+8lQyL0TJPa/M+vh6hCJNiOuu5OMO0TH3OWSY7vx/i3+7HvdzP1YWoau9v/+/h3CudvMvX7wc0sl+o+OCN0fjPsH3E3tXkRiowZ3fTSkdHdv3+++Uw4buo/6gffEsNPjMF5GKnB/PbUbQxAmJNFEQpzSOIFCxDQdkNKEYZowDgNKJZSaAS6ISahqKQEU2A+Z1ROXvqgY7JmGRKyNtaqLh9TVloLrjWl4VpueaXNzMyETb7DfDQzfKG1ieI8Md8/s2o7h3p0dQ3a42xP9g/ut/lqTrlC3x9pzmw9w64MAiaZCuEeIvf6uf4fNMwVWn1K7+27c3Zl6YQfe7ZPXTg2/+uv3+Rr+OMTwbkHbRrCfjev558Yha1UdQujQ1/7ZJtr0YoURROd2tqAdJ+tTh8NEDLAiC4KFtdH+DChhapEiIgbJgaptB8H0FlamUqx97QBVQ4KKAioap2toTLjeuokYWNRRzWJ9fbhoRMz+rkXQnvis6btqd7Ed5F5fpafInpUVAYQQpMIfugNk832vm+n9BGvzfXQ0T0CKcLFUEtlK1NAwRBwPUhVvOpwQ04Dp9DViGhHCARSihiAG1HpFLoR1SuAgtU/WZW3O9QxJpqsp/RlyMC2lWdCEwsMaO0dmQtbcibVWZD3gCeZXamM0cbrjALyXNniXtoqAIE7UnaoSe9czbllYaktW8lL1AwB2Lpp7jPG8YIjLsrSj08MpceZOzWTIN5qkoutrekcJidPn2HmAGrLAvk8195ACHHlzjANCjCCuCNyti0kj1jeG63BFReKzqJPVEVkiTbUXdjkABPG3MX2q6JW1H8ea3HMYbR0Y0O86kfCVU36/KeRKO5Sv6Xbuu9CIhN3bozTrn23m11LZ9ziw13kaKu25uf18KebYe4zu7J4qhAdmdCF/rynb93NgaNjmpx2WXue3J3R3Y6XXZk3Hw2KUMWfkwNzy194RxD0Sb9/ZGF1U136w8rZAcGI4DAHDGDEdBsQ04HA8IKYRh9MjYpoAGgCK0AAJScYQRjAqhmlESBlCJKo6vQM1VCeGwtyC+5ZamdfEzcVExEpSl45uj/Rpd7oD6oyyYyL3yGq3f/bbUKfbpAh7Vi8BvHyOXdsjQ/Rrj3tU1GUUMmLYvW8H9v2+jqjrvZUZVpsS1O0Bu5IIXcgMVCkJkN1VbeH0a+r63n7uNDj+3P0kmMuNI9Ku7/uRf7p98aw1pTQr3D1sdV8++av92B0ye9jnXgTsHrNDV43byQbYJ7VsSQY6Qh1Ep2LKftM9eop42GYzfU8/5urE2xe5GxbB3D8IZKmw9FD5PpL9A6CiFPMJ4z0RhelTtI/M7iiNVedA++YJXHuuDO7m33mwps0Sv7TKzYoaGAiVtc6MOXPrz8q76a/qGO3Iwr4hSVFGgTAdE4YUMB4ShjFhGCKGISGlAdN4REwTDsefIMQBcXgAhSjhXx0xCfGIKY0Yhg3D8IhaVszzB9SyIZcFkjFJdbWsSkS1VoMhuuEKDEnE5ZxVH6slFwKJrk0Q+d26UmNhMndO+WVfkFinjej4/EPpqjEL3+DGmHQ/gNEyAzWCKHpF+VnZdM+CUqu+7F463DNnea+B+WrnsmeK1JghWKsGUtPz9mDFVARtGGH3rlIVLcIkpeq1jNr52hN0X+MOaPRdAyz91z47FCny9/F/jmbgR8la00UYvMBvxla73ncL+YIjNvjYPaHjBtR/gkYQ29u6ZzYK2nv579GioS3lwLZo3fP7sVof95v/5fUidbAWD797Tjcl9u7aoRZ/jj1L+2rhTa22iW0OejHrLpKDpCZNN71m9Su1EcKqWXW4smb2lj4EH7P1pJ+3PeFyphTkMA+DiMSH44DpkJCGhGEYkOKIIR0R0wHT4R1CGMBhANDHwcpOCmFEigFAwZBGlLKCaEUpC7ZSNJ+dQReZj1oIpQC1WCgcEAEwy+d2wGMQ3aEZ4tp6tnH1+5qMUKKTelRvWtH2nRUxamxM94ytWZfEoS0M+TUGUo1wMrcoEOrOx70U1fax/dyfC1vG3V60vuo56g0Shsp6GMHdmfIzUGWvQZmHicdNchHJwxFrv2d6kHPXmrRlYzD/2u6cvZzJXfuyyBB4EcjtDNSWn/bX78SuNtKmqG6XdHq1/Qu4E/P0YZ9lEu68Wm3x5WefpTuEu76ywAPbEDvVsv4qlknJSyI6SN06il5bB2QTU4BaQZvOznSNTpBUXgu68ZuY0wi/WdmCpS1zYmVexfvZ0ilSZKjvUet3LUUII8RHMqWIGEZ5RtytnKdXMmIeYJl/AmIKoEAYRtENnk4DUooYxkH0P2lEGg5I6YDD9B4hDAhxgtW67gkNKwIyf0xxtzkixgFcvxaiuAGlLmDe4EkiAKEkUdAsBxGluaguNUsVxn5MPsLa69fuvu02bhOZlVBQ67OFddZ+b+ut3AMBgiREJkJ2nXk7E1V1Da2gmhGsfTJaI3wvhtPp/Wxf36t/9kYbCYErnb5bJAD5Ts6bdpzaPnLka2eXhB6UUnYSi6A5Ri2N+HYDa8/ZDeHu4He6712O0k/Twi9fEMohrGlEqR/YHenvCKEJVwAa8bjXCbxGC+UBjWBiz8Xlw36yem4nG81qDMN/9nVOsLu+62D3ewuRokCgIoTL9ByvNgJChJVMARRVF6+ypwRS5+yeacoGNeKnnDfo+I3IwzKzwDl1B2h3XNt1fKXKv+7zcRhEVON90gdSXZt9GHTdUoqYjgNiCjgeE1IKOBxHLTgeAQTEeESIJ4zDCYfjT3UigtQA1NyE93WIuTIKKigkpGECVCwuZQXzgpAlNrzPTk1RolxqBVhtIHVjcIGUcd1XXtet0jHKvi4wN7Tj13K/vRpyYs3m7b6QOuk9KmsIURKJNGNL2a23r42+N3qcLnlfe4mi22KO0GNXu8f6zfYfc/Ndt36RMH8mZXDBggPUk+Me2FjGbIuUDIayWy0gil3Yq5BXwGSdnfTTxnbfmq8i+d/UGY5+bZDha824m0By/9R1Ba01FGWo63UR1SasPcC5EDeLVePcYUdEdwYOjQ32z1XcC6Fx7d3m/fQgXYxoQ7mLLW4j74asiVoLt+QuSlAkw01DSMzc8vvpczzMq7BRwbbHDBlSKy9qCAFo1kadpGbN7Amj6k6lBEEAUpRYYj0gKQXEKG4xMaqLTBIkOE4DQiAM6i4TaABRRIwHEXeHE4bhASEOgn4Z7j/XzzbZ2G2tlPhDkXqIIlLHcFDkVyBefVXG7khF6W1lVeIx0hTksyJoUQJfjODsXYW4WwtHg4qAmgjZ9NRsmy7IWohbJ+32UVDxvHdLinqwK3cIUaUfc65wf1hmt6hKiVZFb92+CxqiqccAdvZ2ETbcmGObdD07YDACTBWL7hmmFmKgRcPY+lWgkvgXi56SUXTyAvkJ362z/UYEd7/bO/S8RIxG0J04/9ogQzQI3zfngt2Ee6y904/7bB09B1b09ULGtl9NsasHxftCnSinC9AddNkIFcy2AWUhmyj8kj/1BhL9YPd50yvhrv9mBWtP7JPJmlsGqfhMZsY1dwoA4M44pCCvViCXuhPRbEal3l4nSmhY235D6dgrOyIFw4lgCIRcMkARFEbEGDANQsSHISBEwjQljFNESlGJoOgHAYLQNwKFCUQJ4/g1huEB03TCdHxEKepGVIsUcGJJ17bj9h3zEidz0gxEQIwTQhiQt0cwJ0iCjgKvoatILwRFiAAQRDSejgE4BHAmqcy4Mba1gAjIRcT1bStKCPo1A0yc2yEU7W6xWtlKD/uQP5FEmjQRQ3SxmpmBJE7kzOpG4vvy/pRbVp+qEo2K/lzAtteoIUKJ64Y6L7f9X2uvLiLfOaSEXAw31RGm7z2/39QMLVVXz1CNCRkBBoDYibc9QLL5JAMD6FBxP/Ju/5puX+bg18m1xggbtT960awDc2jITQmFweXOH+q+CbFwluyHxAgek03UK+nC/dpgO0/e2RU26glxE5uoIdBXB92I2D3xdvgA7N7BSui4YufC4a4vBo6Eoirh487NhRuAok5E78Zq/RJCYAkUXroItlGjEWvSeQIjpoCYIuIQkcaAFMUdJoWAlAQVDmPEMEjJ1aA+gcxCDCkkEALScEIIA4bxiCFNoJBQtHSBRRfsJIE7puKW9X4+oCISAyEOCFxBZQBh0AFVMDEIkmkmBrXYBpNUSNYfGhmjk04BqBzFMwLCqHKuLaqG26EnP4TsnMj3/51kYwRSCNnLPWLERupot2Sr9+t0/4cTL2r32J70fJB3qKkRZFt7+7zT7fmr7OC2vezGtmroUL57lRhCUKYZPF4eJEN2TYw3yNDQ6x2KdMQOMFkC2P119+2Lh+OFENDrqQiSPJULq2i3R27Ujdzn+248PrFqlDBa0yCyEISqiEG4Udzp/XxSSRCCi5/oUu87MXsZ69pomd7n+g3tA/eFqXa99/97pt8SBaWQZc+BirRiKBHLs6Wzb4kNWLm/jEOsL5EIUUOsTNYuubjjc2VGihFIhjSh82Ur1sRAG68ZQEAJw5RweBgxDBGnhwEpRjwcElIUwigIMSKl5CeSEZBzBFHCML5DjAMeHn+CYRA/whAT8paxLgtyKVjXrXt3j6KqIxB0hAS6ZggB4zAAYAzlBAoDatnESswzwAyiCqsmGyKBA0vuSFb/TgRH5WVklCxGsOEQUAtjWTJKZlyvG3IuKEXRVCkomltRhm7GIyGKRIydOpJ6FEONgPVgAeZPp+7NrHkdFSH20lL/fztnpI7ttt8DEYZhaHPJVRlnQ1GGDJmbIS4G83vtkJkTJLvWdMylXQd2B/N7FVdMhHEatG9qrNNM4NTNR18fBeqWY/vfzpMxJHBzMZN5S6+CKGtfXmdIdz/111aP6M7/8I5b7Qe+b4YyewL6qT64tVUnureyvbBUScfae+4WEmhI5dM6CQZ/sj+020hsP7mquqBDj8bu+r4aQkR/IFRZbcSDuvljeHxwrVWy5lTaXVu1TGhDwLoZo4SohUgIEUhjwjiJESSlgBQtekSvIUKrXS2mW0JEiCNCGJCGA2IcEdOEkAZArcWlVmQlKI0Qt0w6JgLu3Fy4raXPi64RhYgYWZAoJYCjoFOCzBXLwPvDZ/E1ZliRKwKq+gvWILq+ECqGregcWY69sN/LDJhXgLi+NMv/fi90fbZ9bD0j42cdY0KTTpphqN0jyNcOQy92UkNjSnQtO1Av2mvglfazJQfe8fUONdrBs9/J8nECYG49sf3XN0OK96qmJtG8/O611t5nLxNpq98Tr7UfTAyJKAL4vwL4PWb+R4noHwTwVwH8BoC/DuCfZOb1ex4iqfm7BW3f9X80CPiCOH6mcWVffMJ+xVzBrWVFY4p3aE+XyTdf25A9evSfzOruEKR+RJ9sv3v1DsWgjYXN+566TaTflcpeFiiQOmTbfSSiRuEuJT8zSHVkzchRUTnvRKEUxRCRqxCZnItbh2OsauhIIk5Ao0wU1aRRrLnHw4BpTBjGiHGUeRwPSfVbQvyGpASTpFRrrRHMCaABFA5Iw4TT6SukNOL48A4xJieY83zDukr427ZtiqL2IpKVT7gXn4OJyiBUdddQeRfDMKKmhG07oRagZEF8qNn1rqZ5MhVK0H9cqaGNQRB0ncSotR1FxzYdEnKu2LaMXCrWuWBdsqLEtqtlUYMQRPUW4FLc8T6QpMZKalmVORTCDAB5Ky35rO9T2QNF3YDEKquMC8GZq3JPdyInkuqRALCtK0olUJGUcp6YIwQlgU1WavMFXx/L92EIt+gHUSm2e7dqUox1zdg0pNQigvK2wY9hx3jddacUbNsmhqUkflxmOTdk6l4WvCee0Pf8USHD/zKAvwHgvf793wHw32Pmv0pE/yMAfxnA//AHP427X8h48v0FjQjdE6T+90ZgLJVXa26A6LmQcS2gq2yHhoyYfZFf1V9wuzYE3Qg7fdDrA97rd1yr5fc511TORq741cwePgd7EcU+7JFhn5sQoBY6i05no/6Kggj02ijjs+gXR4R6gIcxYhgTxjFgmiJiEp0gwdC2IXOCucOAxGWGaEAII0KckIYjUhqQ0ogQohM1IdJ5n4Ukxt3c9kxpHwHkK6QTKb+QnCxEJoSQHB0Skl6kJWRtXXmPfsj2p6Nr/Sl0EbUyhiFq4SRGUD9FrhElV2y1V/1027DnmNyhKSWIZiho6E+vxX4vOyADt0s6dCfPbogqUNCIGhF57XzJkhGowvWSLkF182o5DLsuCMHU+ZG47ur7xnweZRySFLOWKkk0mDv3JXZth/186epWhJlwp+sEWt7dbhPsdZN7A+xr7QcRQyL6iwD+4wD+2wD+KyQ9/A8B+Cf0kn8RwH8TP5gY8ouOfYqG3CPDvkJXn7JH+7kjqr3RQia3hVXZxjOdU68j9ENtYU2dO44sMHkMrRx3UomUdpvdxik4y3QmSmwg6M9Vc4AXaspFUmvZphZ6pihQ0V/uQ5gYWmFw71PmKdGhuqBRxliy+CsWlj4UddMgdbgVFxgRc4dDQoyEcYiIkTBESYyQUkAI+7MpBocA5gHgAA4TmBKm4SSGkfGA6fCIGBOGYQKIsK4bmFds24ZaK9Z1dYRjB6n3gQMEEdoe6PfHziugbQJ5RgwAEsZxApixoULC86IkAYYlAhYjCgJUf1cbYfI9CU9YYXtiGAJiAoZhAFfGNEZsh4Jtq1iWjFpl3rlCQxubJbdwQVHXGtlbQZGbWnntzUrp+I4gCnIE4hCdoO/mQ/+OlgEoBPXphEgHqp+ztHrmUWB7S44QqdVXk2xw60LQ6CJH8dwQoKWl69xNFRyrR4HzctmHvqftJGvfLMGJXSuuQgFWitcc6W3fVxXDjYz3VvpPtR+KDP/7AP7rAN7p378B4AMzZ/377wD47dduJKLfAfA7QMuVpsP3genSdTd9vjP9Aveffe5eF6V8YzekabpGE2Mbh1YvJl/5Tx+6vj/UcWnDgM6pur4YAtO3dXJGceTo8cbK+TzM7k4vJs/uib9dA+/7zi/NP4eqFuTVki6L1BJMGEdzj5GsIBEqAvfRJXYAPbxC0CBDLLchHTGMDxinAw7HBxCJOC7Eb0MpxYmgEbq9pfOloeuFdXkPue5WX1QjRKKmiTGhxIQQEioKuCZdHyM6Ki10CIjQHWZVI/RzGIMwRQ7NbUUIT1HEa0yLUSrt8myaX50gQhPPW1oxa2xSFDVhxt5PuibwLbFXLZgaI0X1BY0RzBJ/3fZq29umChIJAu29pPMkU3A3/yYRmHaxkyz0xFm8dCB4TkVf1/K6GOvW6d237UyaNNLOV2e1hiVN4n7SXm3fSwyJ6B8F8AfM/NeJ6C993/WvDOR3AfwuABwPB7bSnj1n84Xof94hrD77hUNfYP9PF8JCyF4itV2/djC8uUI09w/xkAcKFyUUQH9Im0jbiEwT2wRVZk2gWmpRcaA7Vixc1Qmz7mQ5HFGRg4hw9h6Wle73QhPvVDR1sQdKF8gQgVxzOh2gkiOIDDEETOOA42ESw8ggDsdxkDlMqgd0NGAHNgTEkCCZSMRxOg0PCDFhOjwipQnT4YhpOmjfBNnN84JaqyNCI4LWXiOCOWclLHmn/+lFWhf/TWrQdTbdU4wJNVXEMiJm0amiVhBtgPofEsQZW8TlPSMjEmv8Mot63BCQR/eowScEuO40xoBSGGvK6pso+zOqRwOrz2Mbs6474Cg5xqBETLwKJHFxc+0BrKzDnikQTJKRyJ8hJRjdqqwJWkPAFAaAbS6V2jJUbaFeDuYgHjsdKsh9E+H7WFRHxt8J8HRgKOZkLaJzz1BaiKp2H3Z+Vceq1wbX+YfO+g1n0qZS0vzEvl//KJDhPwLgP0FE/zEAB4jO8J8H8DURJUWHfxHA733/o14JCeo2sYOjDoHtiDn3gudL52cbeU8Mg+tE2uqQBbrbIeoWxLl1CL445ja6c9BWTuS+ZYQdlLfU7lUzUZda0DQ61HQxd8xK0KlVGpPD4kYRDYnaVT7Vd/fImLr5238vm2WcxDl6GAUBxiAK+3FIOB4nQYZDE8/bHuJOPJIxiII/QexrI4iSRI6kEcfTe0zTAcM4YhjGZrSpBcuyoJaiKoG6m4a9waShXSOYUtmNfQ57jN7rQ3snOWN8IUT9NyDEUcYUBvXpTLqQGwDelW5w0suSFzJvIhSlIal6wbJbt70ZosztkMQnMSgRizGo2kaIVykRu7KvlrfcRFKooYCgOSQFllXQjikYk75vltk6xYhBDQ9VYgE9AkWer8QTjfCUXAEUmEO+TGfcPX/bsKu2Z1ZjX09YpAt2QIj08KlULNlwuntsPS2XqA2UQlMj9FJcCxoAUMmjtOxd1PDUq+17iSEz/7MA/ll94F8C8F9j5v8sEf2vAPwnIRblfwrAv/R9z7pvL9ZNJ0Wgbb/BIYfSL3qln+iguf40D3bmntgGWD783nhR70o3SolRTU9eBRlKhi0CsxTkNnGzia0aoSHrgo5u3Ms2UDqzW6BeDwM0vy6/TQmuiT0NFqr4r4OR5yWQGl8MVVim6OmQJCROkyQIyhOxJUbh4oPe0+fDZJD40RXAJpERxSgSEtLwgBgTDsf3iGnAOE4IMaFWdnF427JbillzMjqPQCOEbe1lkgztsU6SqwN04gIbQjAEuHfytgkmmyuDkxQBGmQ4dYAqCwG2dd5LMKzidKCOIHBDHhSU0JjjPFSsNFE9AGlgX29mRhoDGCaq29zGXWmENERdc8kuZPfKWNuY+33kaFYnt1ZWR3FT3TBiiHcnyiJXRM9W9J62T3u0bteTG1AA7BIN6/I5MRxSRIoBxs89HR73IaCE7gUKZHpjSPBKj2ZgChQBljW3FGuGnotGYN3Hst+3X8XP8J8B8FeJ6L8F4P8O4F/4vhtMrJQBfup76AbsAuDtvk/dg/1XrlD3uE5WA4FCfWqTKA/pD6A6fuvBqbWg1GJYTpBakWdw3zfAxVlDsDu0AzsUfT87VOtDJSdxsjmqH1zqqGWwHIJo4kRDteJGk5ImSR1VGa9O0JP6BYpRJPjBMeQeIzRmWA6arIgcwI2BXAUdVA4gGgA6IMQBh9NXSGnA6eEdYjJ3GSDnjJxXJ4KlFGyreWHJnAbb4LZ2duBkYmWDl+Ljl++4HXpdg1qrE8L+n62To0kKIIrCMMKgSCJDBJ0CIEOU8m3fGYHgSggkztykjsFRs6DD6oJUc+WgRi40EW3UAJhSRGweERCDiHWlyN5njpKJSDfPMCQVAQsyKqi0vW7rOug4LdxPwvYasJBImW5MHQH1fcyN6BjxLBYMoEmGQ7J4fTEIBdMRkK5Yp7cwNyAxNlIrv6rMtlSRFopG8Tjj92VmVU+h2wPGXNq/FIKfG1NXlVJRsjDgnTfJJ9ovRAyZ+a8B+Gv6+78F4B/+Re639prVtUc591DfFNL347k/OO16fabrGTuS2S22v/quOWHor6P+8oZmGgm1i9i/vutsS65g9yo1tAN/f6+JACb6VusbwbgGLNGEFS8yh2cndilgHJtlMgTCOEYnmCFYhIVy3WCuPBona0WLagBDE6EiSGaYMCClAeN0QkoDDscTYkwSO0yGqlmJYUHJxV1m2rzY/LWB93pCAPt1eLU194l7ZFlrbQiCDD3uRfHGJaPgfTIrZ/G9YAsuCKedWAP9bvEOAIcoxexJUHQl8Rft9Vu2Vdz6CkKpzY2qKkoMPieKsCzBbIzwkhGqEwxBfApRzKfVCLkxWCG4/burdqbF4rOrIiwO3RQRPl9K70yFYEl/e0nLlqCX8MR74l6ENmbYrnX1oyde9ul2dY8zUUOI1PaNLJUhdXIU2e+t19qXD8dzEU8am4CrRKu3hu5/2rlpZMcG7M8yQtNtVvSTb4eqVtRiC9ttbGr3eiU956LwFTaCGNB0RTtZFneHVxeaSVAFmB0Jgagddu+/3taNzz42sZI020xQJf04jRhSQhpEDzhOEYdpQEoB05iaAQWG9toUiehrIV36HkkhorV4gyQmYAIgqfaHYcI4HTFOEx7fvZO0/IcjCOR6QBOJc85KCMWHsBtgm7K75r5hzB5yZRf3+7nNl4rAnQWamVFyAQdGSnEnjjg6VB2iZPpOguJIXG0smYNlxw7BnH0ZoxIF1jx94uJBGuYZkHOReGViAEWIDlVX+QDd+BQVRk00y7YmxjUh1moiFgaGhEARKRoOkmelKFm4t5zBXEGhOtFizeZt6iBhzJYdHdjW7CF57O+V/pmFO/YZnlicw8S7oSh6FIQPNr9DsWADTY8Pw8mdoBcMKGgId6u1IgRUJfpu/Q3ZW1afPl5a0opVAwsBiMlh66t7zdqPVhCKOjETaLjw3q+w//6eEPYj6wmZveO179uhMmMAt++pu9cB4R6OG7qUe/RzbiN54fypz2hoz8SI/Xz0/XQw6hzWCEAAa0H7EFQPqH5/45g0OarEeUrafEGKUmKz49LUwKXNJxF5sSYZo3FTRU1qJIlR9IPDdMA0HTGMI1IaXaENNP1eVVeZPv9hGye1QX6m7VB8P6M6T7TTJfciXtVrmotRe2W/pv06t5RUoCoEBDIvdv8OKdqv6GQbZkc0DYkSTO9FbWMZVpF1UUIgmaCBEDSA4C5hrvdfeuYMzNfTeLftHds/FlVSGxFu89vt+W5P+L2+7fUaTWRh7+JuM/s8tAXc9V4eIXPEJbT1qe08tjlux6Sdnx4Rti9dO0po62q/q87YVF2faj8CMewmH7phzNJr6Ah36JG7jRP2UNkI6EtCCNyzAVZOLrqXfZCRWSX7Q0uwSm1RJrEX30ws83t9dADMt2m3v7RL1EpX3vXdCMR9VpAQlPilgEMYEDUtVgiSMEFqCCctokSu8zPXjmQ1ol38V+ttV2AKAAoDpspjSKxYiCMCRYzTCTElHE8PmKZJfAang4TwpahitRiRTC+4LIsYSrrDRt3APkUH+4N6T0R38xSMmHFDFJ3Ddq2SIEKImKRi87U2sZKbMYVCBDFQOIrOj+V7CVmEHljTWxZH/IJ0ZA/XkjXzjzASI3QERnbxVdEeBa2lLO8pqEAtQhdNZT6ornYrXSGrRm7cKBIaYY+RENjC8NSRmoIg1dL02gAj79N7uhjcKg6KMzSzuokpspUpE6u3MU3Wea1Qox/Xzo+2rS1XdmNkSxLsPMQZhnsKGJP18RCSSje1SKLbquUqhjRovzW7j4YTEhEiRZ/719qXz2eIRjB8UXdF14F7bEWqpNiTCPmdDA1gf7jML89QGb/2+J6FuX7ReJ0eEK3k5dZHtMtI2TvBut+cZxon7GAe9r+amH4HYht3079jCK4kt4Sp0yguHVENHeMYpbZwhFiENWFCsPeYcG9Ixli7ikoVUGQIWAgdUQSFQZIcpBHDMGAYRozjhGGQ8p0UOkSlmY5b2q2mH3zJn9qgP7VBXyLCfnbp9W8cCd7f28059ujQJl2klQ4R+7/+MMrbTMfpel8lEsYoJfxQbjV3D1OVtL3QiBsBXmNF5lLHqc+uhSxyGuZcX5k0/NQARnum72Lb1j2K7abGz6Lt525uXqzHXWxvQ2HkKp8dYPB5uSOGO1Ta8iXeH4K2VmJBYRXNjbjtYPndI4yQthV9fUx9++L5DK2gjsyRHcZ9XQeCJTrFHRUD7KOGrNrEQ91LmEVxL8WNXjGWdMhyr3OUibfSkTF0OhLnhOqoSt3Cal0TQqszUmtWjqWe80QN8RJ5FmPTD4HZnVGTirdRlb9SJS5iHCIOB8kQHQcZhtcSGZI63UrarzYOPYTc6W2UXYu1kZAzoWRCqeqoGhKG4YiYEk6P7zCkAaeHRwyDEERBW02ZXrIQvUWjSObb7G40ltK9IQj4hu1VIj2zse920gMasrWyBezMSNejE5PMijwMg7vYGBKRNZTqewBjUz2XxFATiAYgBBCL/6Eg0AzTJRoaA4Aa1Nk3GaKXZKoxaNwvAGZlWml0t4+eQdcs+rEUg/SJG7E1Rj+kpHtF7sslSkKIZcNWJeltznKqgs5lVI+MorWia+fNYPSDFcVGkjA8c0ezTNqe2YibRNEQoToBxSgJWLjPRNOAAYzgMXvAgDMVbue5+de2ImPwtQyidslFj68wjECGIxvTkknVfz0TeC1ZZ9e+sAGFvL8OBDslW/Mds5/9vXYNOn1Rd8ErzOUF5EI7ki8s0Oin0lJPdWjOfhJ1G6Znsa9MtKGUflRee7chQ8ebKn70FmHLGJ1SxDgmHKYkqoIorkJmQJHCTJZwoMvc0zULhfJKdUz6T4wj4rkvluIQB8SYHAVO04RhGJ2wyPCafs7QYP9vjwpZGcodd24T25CyT9+9hLBfLyjBkG3VKqE1y3n7d48E7RpXuXQM0lMvIwJWHkAcEf3w1tqkjhogemMSCy5BfgYr5KX7PmgurGrinjITEUoqmk+kEv1urRwPW9SGEoScjRGgZYPqxyIr76CBOyTZ5rGhPC8IVRryuj+H/k+JlcyMfc8vjiLbJqj9w3j/YJiej9zy39QhOi+VwSTMhu/v241391RHxntc+7J9eTFZ/Y2auNYNy6kENF15Xw+htXtCaIczqANpCBqh0KMOfXcMXQ67T0wgSAUhq2cMO3P2f9UvGvespaU530H/prx3pqwpxpI6NnOX9GE8iEPqw8OIlKITyZiiiscBaVTrmW7aFJPqQwQJOgIqzfE0F5ZEpKp3qjWgZcGJioQSUkqYkhRsPx4fkFLC6eGEGKOjQXG70ISaXHdZZpZl8WQLTgi7/ze1AmnafokV7leg6nPZkES/LB1CtLnpiZyoEqS8gP3z2F/LoqwRSRabawhSyiOoTtHCC2vW91XdCxFmxU9JrdeWLstc7chGXCWMU5EXd/9VsMeDi4gpBLRyRq7Fs8mASNxuGLBEHYAQ4gh1N5sIlcWxPW+me7O56wwHAUBpLjttRZSYKOo2ryeb4xhJwi2ZwUh6NgzaWzbr2lKQ9aIUGr2rOgcuJbirU5MIdnSSbbYArhtykMgf0wmThcb2uggI3SiKAG1dOx3CZ9uPYk02eV+I2t6qbE3Wn1v1sVeIpvkfOp6jzqBhXIbIiacRC2k9m3Ko1t7eQEeD+/51U+aaCNETwt4diNkIaxubxVCamAwW/d5xSkhDxOkkmaMt7C5ESZYQQjMgGUGMOiZ3aQgaqcDVxaOcRazbttocexmu74spaYJViRpJw4DD8YiUEsbxdTQoYy8qDovBRIiiONGaCCwlBWxW73SxRN162BqzFw/a1atBswr24vMOCWqs6muI0IxShmLavfcJERQdkyJEBDDEwKHJ9nWPWc5AaiyVupGyZWpupF6XGu2Yy+4SYggvc8AILuLanjRyIWCMEYLMVUrAMAZ1tK4a+tmYkDMgCkA1hNjWJBhzsn5XWyc7T6Z6CN4X83etxfZCRa3Z81f2VhMdYVtDQ6B+FumV58oA/DwRgMIt5hwS0x9Ul05yuAHAS+lK3LIie58PW4XX25etm8ysMZpAg7c6YZ1Sded2QQ2Z+dW+SOYi0Cy7AJBSAoPd3UNyp0nUthFQsvC1jhAanPYfyp2MoEpaLdO9dDHDYHS11+V+5U5JXWHS0CrHEZFaey3KY8SQIh7fH5BSkEJK1J7fyyakqGEY0u6Qb1n8xGKIoKB59LaKUoB1kz6XMkAcdRMQAsZRcgpOhyOmgxC/YZwQY8Q4jpCU9eZALbOS9T29KLxteZd0wXFHt352kG12BW1RYzLKNMp99moXn9rcKi/1g2VGjCb6thhdrlVSlbmrjfRO+i+FproVd0Kg3EJ0YYhwUyfLFcwirlkZiaD1NTRWR9KCqb40kEUTKwm07OU+JkFuKojLHtN+8W5jOVbS+SsgYoyjWH0DCWpcTWeuRK5YJEqMGFS+qbVLYKHPJNXfG3owScjWwvT6Rbn8roys9qn6maJ2lhhOOC05RS/aOXPV55TaiCGA/Xu4OuOzqCgGg3S+CJ2vqdIKVlRdLPXYJ9oXR4a1MNAhHEAFT/oEQQR2E9e71bS/sfs+KrcQREPIWwFz7lCbOpDuninNFNc73ZX+ZvVZyCRsdCJHt7HtXiKo6wvheBw19E36kGJUhXnE8ThgGBK++uqoQfxyjREGyzVneyjEgGEYhEirkjtnqRcSFDVsGyNvggLXzbizuMtQlOwy43TCOE04PTzgdDo5EnyB2NCc0A0B2j8LezKiaDi6EcKGrXcoycWyJkgLKjAR3NamHch7cduddTvm2Qgheb/v61PL4Szyr4uG2a23SS0atsdkIi+DuaBa9jo2AbARQjCj5oqarWaIOHxbeol+u3CDlUIQiYAKJwhW7z5oxhh0hIfZRHZSJieMv+bi2bUBzbVZmu5NdJ5C9DRtrjzP9v4rTQwp8H3ZCCD7uRAHaUPDwRMtCwhSPWkwIHInhelzmjtV2ycmffn6+4FT0Zs7acGkJKcPtj2a0etT7csiQ9igBY4bOrsXk+mVTb17CLVfjFsAztAQEDXeXjN7dCeh1qpo6s7nSAmY+UH5y1SMAASJy3mQxSoefwrXsZgKIETVA2oewGmKgg5lhEIIo4TLTaN8J1Y5ct8pJ34kmUos+0gI6vfITSwoFSiFUPTAlgLkDBHv1Doe4gAKAWmYEELC8SgRJGIcGV6Kw4D7pYlluGLLm4fVWTRJS7La1sDmj9EslGSK8O7M9WyIQIB6GwjRaestB/hun+i6CS3pdam6qzRMDDV09JKcYdr2N9ErhiJGpmr+gEEyzejhlr1g5COCIYQHgKRwdDRloqgeWB+HSTSuYJT9b9u8FhS2OtYykSGm3RzVqkXB6oZaxEG7bEoklSjEpAk2TDVQxO9OrOAKCDpiA1Y9cMcYbJ64HQPvh5wDcgd9Uiu8u2YxHEm3ezopr1aJbKIGHABuzyXFwG6VubcbNEToRiM94+J209BKrVUjovDHmqjhl2qM6sRGlNE9AtFDAxV3HH7tn9B/ZFXe7HkAYVBiWksBVdMZNuU/9ED2xNCSR/ZcyUANcbPyAhCDRK3YNMTMLjZ3GUtwEGPA8ShuMcfjKC4zShAknVJCUM4uoq8iCOXaJgZaIXFPP0UEsCBIC3MrmVAKIWcL1JcDKYRwFBFpmhBjcpH48UEcqCXF1tAOQIdKVw3TWpe16QVr9agSmXdjOL2lHL5ByfR4ljSBm2hk6MIZjmbjadEw9pjOH037GVyCMMTY3DJYyxnUKuVAzZBifXN3G3V/KSR5+agChQkegaMxy1Ljhlqca82olZA1CUCMpHo8EuhSlYkyuanOcsYFN+IFR7h6KQLybs6mNMhe9gzmhsIlO7Zk0q4admbSCJTgKkJS5/FtrcAqqNEksVIFiWdVccgeM32lH4G2phCCI+smPwOJKqGgNmNitzeINKmsqS4UBQYO/nTb6yz6BGWY3AhxB4yIyMNSa5fPjtB0iaZQsCgoQOf9M+1HMKCo4hpwbm5IgfrvqQ1cuInqSrrnWOPuF6KGQP0PfQ7IfrZFkXPWILpxnP55bpDpR0Gq43Hu1gwGIYhOMKaWOMG6219jBhFXBN/9JwSWW3ooFUMqiyhVWcThUliIYFUEwIoEgyZUSBNCjJgOB4SYcDhMSDGpz2C3+dHEkqy+g6JXq+5A3ZyKew7VG7Ko+/2VNfc7upV0/Y7qsnyt2lL3Yi7ZExie2p1cp9SJU9RutAO1c5e675Ojq7v3kY3BDqwSSWKhYKwhiLZ3tMSrknX4dFjkEfU67oa+5B/5PZbAwfsOqGO2bD6G7FvLRGNRHcNAoMCSrTwYIiNwBKrS4VzZrUnMIuJWgd+uu3Mjjq+Z/L/3mGCG6OQI6l7V6YyBZjVWYu/Jbzm0M7/fGLp3Za/fS+2+usy773ZGIJPyvCdo++n+IHftiydqkJx5zcprHL8hi2bdI+OaQKc3gIpGsiF7EdgniYVr+UthRKhVxWvVtIwIWlKG+wMj+f0ai2yRmKKb1ELnqkhPGgt8Oo2SMWZKYnEk1a2oc2v0SBFJ/hlVBAa7vVL+UasQJjo1CfdaM1AqcJsFIXgiBS2+JMRuVMuw6AOPJ0GE03Ro6a1C0680f0GrRVKR86ZxxrUjNsYkurWl5uxr/bV1pWDxzYR28m15TGyVn+I4zBLaFRqK6OfdWGflqo71ypw096QRLdkOKj51eQ3v/RftYFlJBVG0a7ZrfZc4sNvYGWQRJiBwJQg9qqihgFGVyYmnohF7yXlB7uZj7K9Y9ncGiIOnaiMyqcSyWMs0VpafTBVbybjOq6hripyX6SCqmcd3onZJKqnYuuRsCJOBKESqQsIxs+psG7MOus+FSDFajR5Va2otHaCoHtnLQoSAYOdNVhVbyahbRkoJkfs6L0bc9G8iAGWnsjJmVbllMt8lXG47Wa9Dt18bIPlU+1Fcawwp2KZ2kcoI0WvUu8PrbmHtDqdd0jf2g9fBT/2oN9bc6yx7QXyfg7ARAlk0IXKyKJJRxHwCUwpaelPdYmJQkVjE4xSDG3heoM79cB2llCqZhysDOYuFt1az9KoVNUhChTSMnmF6HEexFKte0BOfGsEwlOVWYguna+n4e4teQ9vdeuoHLtJQv2Bt3bgvnNHN670z8F6fi26/vLzX5qhHhq7DDWH/rP7O+/1Dd+jPka4SMCsG0v0zYi7bqyFcf0eHbq2f5nJD6szdEteK0k12bDsPTYwmoEClECGIIUpiDs0TJgxZJRJL92+aKPFnlQw4ku6fJBEtiaEPUB1xNUQXukXeI362X3VtJJ661WF24qNo2aQ1S0RRmT1fY6218X5ZiLYHSPcH2gv7Cpgv15b1fHd/7/bOa3tI2hcnhiGYBVA4DRiqa2l+VT2htK4bWmzGEoakWTJhQ/5vugUjhHJIWrC/6bm8jCbpllZxwl1llDhE9Qe0TS6HXbQmgyZJCF3onP0+jFJV7mDV5cZBDCnjiHFsbjG1VpS8CZHgAtt0gUSUMR2mEEDGvBRxEM7iWgASnVJI4gozjgekYcR0mHBUX8HD4SB6StULmkOvqQRKac7T67qpPlRrk+TiRMMIoaE948iWVMO8AExRzS56FUFaRFrFr6EjuQ8vjDGid226MtMt9q25fhi1rU1XWAooRjfa3B+ansAD7IyqVgIrupQTLEyMAoGLpbkigKOes4RAFRyU0Ec7fKKDDronZGwMoKLWTTdSAiNg2wryJum+jNBZKJ8wVZFtGVLzOtcMRMYwBYSUME7qwE7C4KzmMpGJ7BXgihAZgz4zjeKovazihjOOCVwZ25o1M7SgeVYESSTjqMyA+jiSntrkkkBETUmQm6aEK2pgMgu0nz+uyLkRPKKAMCghhZ0zghewVzRsZ4Zyx6RUj+L7rReh9WHMohP+DC38cXSGr33WGNCnsKxc4wHsgHP/PVvZt91Blk8gxLLBZ7V5+rFiRpfwtBFWdAQRpCFwMXi0A/5/7f1drG3blh6Efa33MeZce597XdflQKWwUVwRFshYIVgWskWEECbCthAWkoUMCAwY1UsEDooUXPgBkPJAFBTiSMRQggCKLAwYgy1HgYAxrwU2IGNsCgwmdlllqgz33nKds9eaY/Te8tDa11rrc629zwll1j5X2WNr7bXWXHOO0X9b/9rf13zTGKFC95Q6L7G5ea6sI8McCRPpFa2itNnyQ+EZEmYfnFOcbUTsRJceVd+2fcflcokvQ6rOcMO6EWUkDBGmYf79iJCNjIF9NlfvR2DwufOYTRQRVhH63Vfed330sl7KvBJ9UBXPmNJsl77wvNr+lzUTrq9EhLQIVs9wFNVZ1mlZl1WzUUAbc5S9bKnCK++Z5Ktom9qKpfKZ0snaIHK1w7q3zb97G5xlJm2h6gKyobva3zuDvlM7qH2eLIMAqp2+boAgAcmSB27gmYgsmHi/ch3IMs70CLf+gnYm1Dm89wQu5R5pKUQ07B4DCpwg4wOCEPgo3uQMVAWyyhyQiCGgddlvxvhhPTKCAY37sZ+ML4q/RTwUVtUloLTaCQhAnEedaiKRiDYLr+CkiADX646+dXzz+97g7dsLrvuOy75ZvN8cFlJz2eyUbswd3kJIMrwELmzFBe7BMBZHZafbBY8TOIflEKt2KCRo9S/XN+i94/rmDbZ9w5uHt7hertj2Dftld8/1ttjQuDjJZswAaiMkNW/x7elYTQg+iCErJEfTMydjDu07whaU4S72/pRjGp5zllrgppkFFa+rmEjU5mMRoB7qNFuplTJNgzgjggGOYlI15Ziwb7V9abMSVKsl+7rvOzL3d+IcN8wx0iPaTfgkYrb1GxJFTOSImI3SVNRu9kQVY8xW4Jgnpk7cjhtu4wCr0VlNkc1lVwukb8HmfrAdinFqINzWBNvewdrccwJDvGaINqBb1tJ5HlCYvW+MgeOLmwlEf/Zl39ClIztTw8/8d6bKOro1U00Lp9xxWJlYTEWb6hoaHUzugXbBOljbWVtQnOXhlnWsV6GXa+/eHHZ/fSQ+w9wwhMBxaoAn613DRdyg/Rx9KBIhLpuem01e+hTifffOnMqhZ6FALQaaNOIWO7jjzZsLrpcdD5cLzjFwjNOE4b5bvQvQQZDGYikbOY7OqRa3qKRcHzhH83hBweHM0xaW4WQKrvoyY2TbN1yvF1yvlkXSt76kIHKREiVURJjCQV01GoWIYjUTPJ9LRUVr9fsqCNdxr/NEJEHZQzRz/9lqt3rpnlWVT/uhqa3sSz6/Ys2XbojyHkd+ocRZY/IA9oqMw5EUvRIibs9jX2SpZaMUiKIgS7+/y9U/RO2P4TFzpu75ivcsIbgJgmFR9ngNlpwx1PLgSy68iqAPtWp7TpbKWiVkYlJoMEffPIyIJqs618uhJeWbnzBEcoxgOMeJ8wTadHQc2DfvEfLAvd4NLYRejRTwVpZFwqbI3UvvkwJ2vT4y1DqAdUPAikjrOiiLTAwUkC+ZhtLiM6puYBWqH2vYB3/if62XzaoKbL5Q3GNFg7DZ4yxc5lvf+gzX645vfd9bvHlzjWySdtrnRACM005VyftLBBSbTcsyOKwQzpzA7WYxmOc8fQMAczaocwu23rFtO1pn2pzZA/u24c2bB0OD+yXUYuZ+1pAhoi4TuqfbC89SwL06Sgr6ZvpiqD1EWsxqyKyhqFks5kHlfE0XtP7BQFMUOkRldX6XaAHVuOc9MuRzaXvctgyo5vu5DtKDmZuU6Xo0EYgIzlnqbwMWEA4jgCWZKA+Y6XZEK+8K0NtM+68iQ2D8yTA7n2Dvgt52zKE45zRKf59DRcMcis9/9obbccPTOJzMwW2h1w7sVhHu6Tx83gSAlUMVTOg2AZmBClWaoaypHkKlGMeMMKVNGrRbeNCYgBHYCC4PO6BcCwzbKUjQDyGoZUnV44a7b46BE05uUvay2SRrqp5ESQU7QOwZovTCq7Nj+JrT8Wyt8Fe2464e27PrI3mT8yRJgeibdaypeFr+p4h8hhoBuPTLU4YT4z+vCIPIVLyQT3ndT02qfrGgG7BfNlwuG77xjSseHi54+/YBb64XX/z2/LN5O5hfS2LKQJ/W13GeOG6He4SNp+7x0Rcm05KUOaUkcd3Rt4sVX3ow5mkThh0PDyYMK0FBoiAU9TjDKt5HuVWWTwomyb+siMlzjBe7nP29ZoNE0HAs0jIYUM9Xvzu77zQeCr7qWa62YHUEZfM3/FBL22EVftne+nsJY4lTOz3kdPSpayg8MLS0XKRHWE3jWnD1dwyrxyz0uDK4ppmTZogCgyQDRpyrsBClx8cDj483HPPE0OnlHoCtA5BmKudptbk32bxf7kDq06mv4H5KwcAsWgjDiXxPCIw5exMwC1G7qdYAPP6wLAiFH25pcuABOQpYFMCFIFVam//U6rIuUMytIln+NLFezF8hcqngyvbyCgIqqH/p+mg1UOx7/p4nRAu2mnL2xvcsOF7vMZd3Lqq2aslNzXuRNebhYbd4wN3iAZnzaVkdZwjD3hvefvbgucTmEbY9MyNtKDzU8SifYKmexBRExzkwp2WOGGOJeGWw3Z4pVnyp9x2tG6nq5Xq1WMEHixVcVeLVTGBtUvfsJao+jtO9eafHD07Pcba6JTpLveBYPIoPriTkAs24NFt9WRs3s4V4zxBmDS5gXr43hbiNJYppI5G+5D561q78ev73+9ICtHUyl7xqJPHFUu6T6XaZTcH30rPJHR2lLMQ26DlPV01JPKCBXqYqMNTXiI1975tpF5geOmPhM+cYZh90J8yhJwBBy+SMcA+mZ97b7f6f1sVTZO0Q2OCxlTAThjSBOtU+42DjYCsCMe19le4NMTFE5VRst61D55bRCfcaHHLX1hfXzMw8bMsZl0vWz7X3rS1eH00YLjB2UX39dF4t2eBE1rKR7O/iKAHAIuHxiscxSRlkW4TAm7c7LpcNn33jwZmPLcRlnAPH7YDCnrltDd/4hjkrHh52O5nFA6GZo6zTmXfVnT2I9C2mryUSm+YdnsBxWgnOcwgUFkohraFvF7RGYWhxghfPI3544225XiOmDBwP7kE4ddd5xsEwVT2g2tXimTnVc1aq/hzjRAE8cevoA3W5mn20LUI50eh0h8bzz+ks812FbkG0tMvdvydq8pa5rQgibLXVZCF5jwz6r+2HFVhviaRVsT5BBXOS2qu2yFnLneXIhLQ6+a4fijBSCqhi35x0wyYCCgtkVlXcnk53MjWLFhCYwPGYQgrD6c6SqAMzrQ5KYx0fsZjCyTFUOmus4U3r+KupuQ0QGFVXw7QARRtx+zYL/yHH1AVaax6Gw1rXPmT3yFy2zdP7UE6zFJwhwBMt2SGkvKfBxuGOuAjCriJUCtD/wPXqRA33ZTFDEPpoyroP7A9+mvmRvNwPQAi6JQbRB0uAMA4TrfQuePOwY9s3vHWUd71YsDRvPHspOgOSJJAr0DZQa8WDSfYVFxjNc3BJd84T/mQu8TDSVVXBINlqaxA0OMsDWrMaxNu+Y3NbIAOoSYV+fyrSt04hPdyOZm20NpBqa1lkLmzSw3o/eerlNCVssnK3uPmVt5zRDh4YY7D6ed5axLMqiCZ9gS/2Yy2qEV/NpruAlWzTHWFDddqtHm9xotfpBdw9dFokYhQj+0ljOZavKtiJinioUvgwTMkEgwkgqtewQlAYMG7JGmmg6BvQJpGYo08KHeeKHOfE0IlxeE67l9oTVUAHVM25MubEOUkrlk5FccOmlbItBdGkwSiaOLYSKAvw6A7mYwPwcMYYvzaBvj/nAQj116c6HacSZgOAjiZg6LmaWfywgUiwice9J4I4N9FSwZgfQIeviwzVwjm4iiPxPtaTLjYA+5axWO+7YgP5IaSeGm8DkoLQQlssAPpb3/oMl8uGb33rLfZ9QyPTR00eD+jvQtHtKjpc/XTqfQpE9YXKUBpVxdOjobDbjXWDLVbQUuhskasn1be+L/ai1o1Y9XJ9g4c3DwvZ6uWyh+CpQeWQdAiMMXCOYYHUwyi2VI2TUFUDLWkJMQmaMtSTVFfPMhetJNpikHRFW8GoM2ifsrooHKMFKbgQIlnncyRoqhyXRj1EFeYpzU2MUCFpnkgPaBWKCBRL5MpMnuboRnsHnEWIh7GqIVnadM2GOAFRJzg1UlHFwDzPYP6e8wREsV2aOzKsOSeMc/GybbhcjNR3niTrMIHXRw+ErTqdlVVwDKLHgXGzDbDvfig407Zlldjhex5ePW9v4VABADHjY9ixaVLIcgoz54zCr5ljIzItWTpCTeXWbiFm1gZHeE44zL0JpCYHmCa1uyOF58+cExjmST+OIw4qFPsifC4yaN85AXjifEAI8nr1glC5gSvlk8TfeT3zNFOqB8T+UO8kFroRlDZse/IHspbIvlturjnF8vQLy0XASRqDPYiVy2FBBmy4HURpCqi2pqDRTCXU+fIgFC6mJlvYzIatbwUJ9tjk4QUNb2wMHKKIj7OHpLCbccKmDU4CheQZmyoT54bBRcIN0VL1ZLnMnFMKLSLPvPdi0ylz/czB5SQIIvXviXrjFrK2lQ6k+7Gqa2rBl4FIUy2jMH7J2VLn03U75IohhuLrrd48TBjnabZPRjIMBUQUAxOju/32TC0np6Xs6gIgpLyHjjzABFA8FHYA9Z55wxxSQbIFMS2Pf+tzGjpVQVNZ2qNOIRY1SVwDYw3kWJbcLPCDeybnQHr2s69cl9F0VS+pyvUnKEvUO5Hjr34v15yfaznvuV6dqGHrHWfUnk1DO0knuWCHs1NOz9Vkr1q1R6VIWQa++f0e3lxwuW64XCws5uGy47O3xiZthZUEm9tFWLWr9c3DIwjdTV0YY2CcT5hTsW/iSGTtHC0VqjDPnnpowhRM7R63JQm7GqwMZ78Y5G9Gm365Xs1TfDU75uV6cdLVVgznLSaaJ6GNV6KxKNfpnmIjeSA5J9UT5OjxxK19unNkcUx6czouVGFBQZUH3qKKl9Q0XkEg7RUHEZukGecE6oFjZgWGZWxeI5ePb56Bs3v2TfWsc1yaOHvCCweZBQMDs1k6mrSOpjC7bTP0NohG/Xu7J6BQjbICQIM0Rds6MIBzWH7546Mh84eHCzYGIOvEvKTWMTxEqPn+MOYjgVYWa2mQMS3HV9UPWbVAbQDDU+32PbOh9ssWWpJCMb3+83b1+jFHMyHnWSS9T/SdJCY2l4enbOrNcqkHJgZmrIk5nEfSoa8qogjWGBNzKHrfnDyimRY1J85xgjWFYn0AGL5IrCqkiSzWaA5N0vP2qOZbTCYPFJun3ta1d3+9MjIs1be0CptUXaodiIiCp4PkbVIQ+QYSzdNtcwKEfe+OBDuul4b90mNh9PZ8E5tQ4WoXsK5CCh2sKMyPT55EeQgnIrRYQZO10wk/lfaR5hyFvYM1G2og9bZv/vP2zAGwhhEkegpVaqbnehFK9chWXWFS7RbRMQpy9wEXR4bB7cdNuEwKkFIDZcIKOvRhphksBaHEZ+5Ab1GFi32QwrCtjpIV1dX7lLu+0P94zp2w5DzzdY12+o1cICG0GL61zDvEapRMYJx+kPstp6+VCA0pOBPR3wYREzTljIm9pU7dL9RESjeYLFCHhHsQ92sDMA1IxM0TAqvMNzG7zXvfxASnOmhtgDTN+PbafmiYBYg6OWwC+8ictFvbvufckfHmHuDVu8dcitsyi+mo7s8PXa9P4dU7ztMMxsa3JzEBtaDSZGS6q28O3lMgFlXRGFgQRZMeLhYuc33YsO/NMkSulrfZnAD2dEOvx41aJTMQXpsHd2s9HA5mF2qWLN62ZJxuDfM4cbuZCs3A4nEaa/JxuhDUDobbwgXf1jq6h8u0ZjV+W+94uD5YvrMLxcU25+oDE+EBXYTEcRxRmIneYiKpOg/LvPi906TvhxaSLJVeQ6qeRhG2IRwkinRAuDBdhC7UYeCM+4ObiM/z/jUR5OFS8qPdxsy6LESxbNNL9ZFDlY8sHCM0rSFQVPFpUhjn4bRsI5xQEQUwjebfSEY0hJJSJYuCvdbvqbqmUmqzORmKx3cKwWmay96tlOi0HOPLtcXQmcnHvm8N0ClO7mCV9C7dgqT3zYKvj5tpAN1DY+CawrYBDw9Zy1lVcZy2354eD6jeoKN5USgBVKwe9N6Cn1FVLMh7E+AKbJi4HcNKlvJodKGoioyBcfvidtlcUPuSgOJUcyx+XuptK9Ke2/cOYcmMSfq+VI2r88ULzGCUPdEgUJlec/r9IvGj2AzzRGexHH3WyNjo1Y4F+AlJVRMg3xqdI72bEOy9O52+2Qm3rfvxW21svpixCmHbPIm46AllG+gt44k/SyqdBVHfUW2hTJj0EIZ929C3LYQe6xIbw41nIfS+9t+bXWuHUCACzKBwJupBp8iKcmSxK9TXMhyJiJAygyYAOiJW1LWiVHtZly+izLoYFoRS1kR0k+pf+apmFTb0pdCZxdbm7b5H09kvzff7JlKGAfnPcVBPXdANEVz0VDkXdr8ZCM3UZniFPUC9SqFi32PEQYbttKMRLTstvjsNxlDAy5Y2b8eA5fcOJ55tojBCWXtOapXmmba6OpbiNl1FNWdgrm1oFjEzz69npwDhtc0IEW+/ZLyfORUlxqv5faajY87HVHOQnB5/a3vJwJJ6THCgV9/+MV9ZhH1ZYbFjeS5/QBACH6Vuckp8VjIzlc6jlxiL5f1yU0kEFJNWv0Wh9Y7rw252wIcdlhfco4Jc7828uDeLztc57DPbQ7DICIDHx0ecw4JVlQJtpO2NwCaEoAqebpYw/3Q78fR0GlpAg8WfdQcODVAvwtSap8vt6BuZpi2AujUWebJc08XD5mNnthDaXtY6KVSHa6EmkiBwg2YIzqpixlLiGZNnTYh/BucSm7MAVAiq8jy2uS5RtrUTqbmQv1etiehe8ibHRcnTUjPYd3M07U5gYUQFmRtNwzuFXjp2nu8RHnZBYMHiUSxDyTaUT+SBQKHL44N1bzQIYi/7Ad3MLqhMzyxxjBZ2ddhavuxhWqIwgQAXaU7WYUKPgfxzAuJ1lOdGj6wdiqbpuFPxsmF6qA0gOMc0jY3eYBhJLVrHPjePr32IOVJVDJyYmHiSA4ecjpAVc5u4bUa/9vg4Yz/baHgc8bR9YUkBA+c48XS7Ra1vc1iaUJ1Q9NGNTo8qe9SksfomY5bDShVba0bhFvtjWmG4F5YTr9cPunY0JyD3namR3JTc+OFtovrUqEK5IXVLxPfmjQlDVqDbe6pyrTU8PR045wkWCGraA3nt+w4BIvZOfNZ0pu1NY6Fi2QQMaL4dE8c5XZAwfMPvA9dvmhnk+2YGflJtMYvEvN7kObw39K5oiYzUAPOoywJ19W71Zr+AiGizLCopZ2AVhHy/t0NSrWTMYFUx/YVnJ7H4TUT4nYiHGRz37UwhvrZhHZVYGy3rywAS97B4UP/wIgizffcbhM6zOHRKTGYATmZrLNe9+LefxZldbAyMNks9KM/4HVMo8/sYI2ru5MGViKt1eoQtINpYvdyb2wXaKJhhKHB6JtaEO7J6zIOALEGpvyompp7YFIAzUjM4nNyEXZsJzBOQ7msTXi60Kc5heeFjImpDi5exMLVZoKdCT0OGZFmP6fH1lKCgOEEYS+gHI52EPKzDeVb4adXry7zven2ihuk2FUIQINQOQ8K0idnru2djXC7d0RNZpMVqDG9WdL31hsvuqpJBO/NgTQEwrZwizAO7b1lOcDB5Ui3EhRuMCzKyE9SENqC4HQOQiaebZZGw/ojFpzUXBNap1gwRXh8erEbx9eo8gxkyU/kGAYQwrDZUVUtlIxFBTLDSUZOhNNzwzCQhFRYXWfcP0+s+o6xZUV2rCguEfTTsuzPrXfC+Lzkr4gbP0Nf6y5yr4KvXLMJRYJEGUyc6OkQVc3TMNjBawxaqeqrRHCcAofraL/A5SsEMkTiUeu8Rr0lNxii7ivrvtkMVVx1whglDoOjBVmNrUaGYw4LspxcWa80cdwLB8LCb7nbQ1o1gQZ34Vcc0b/FMKrzWMr+3NcHDg3H5347hJhObe8tUUcg8IHpz++lpSQUXI4X1CEvcjhPjMHKJp6fDtJ0He97T4w2WC33YXBA1c53CeRcnPO+6HIrMy56mKxu9LQDp+MY3L5jD6vqY7dM0r96BrYvLB+6FWHSx/umt1qlmc++WYtt7M3tl73jPCgXwEYKu1TdosX/adwbX8wD305CZFtfLjm3vuFw79kt3dRnY9w1v317RGpOEFPMYdkIF8QCMGh0CoIdToqon5h5pS1unLzr/FcwRvTl77+1QHCdPYDeS02blUN5CCDou1wdcLkavdXHbYIs4yC3NBmW6BGXyp4ZNxcYlY7Ps7yOS7tPjWeINNe9KlZV1LjAFM+p78MlIISZZTa46NKiqxxi/SNel670gSx/jaYREVRoLEAHlRa1dWXGa950/z7C3RXiUDdQylvFsD3gXcWZLX3PWn4YhSV66hAWVqmyAQmVCp0DmhJTSBr1ZtTqOqQlBr409exkY2+DjNDTVd0Y8qGd5WGDWGAfGMVyQOmp0Uoi9WbjY9bIBaKFy8nA8naFIW9o+ebXNTU+e2HhOQA8rQXu72WGofp/jZk66x6cnPyi4b11gL9PJsgUS+1wMnwBTwxZqMbMXDz2i3XsAOtCb1XHhejLVel1hPOROPwB0KnRTbNpjXrdienrpemVkmDYrALmYGYjZrIaI2b4GBMDluoVTZNs7rtcN+8VKbDYXClZXGLEIaVWotiKqTnRoPj7ewqoT//mmC6eN1Fqz6RjTaZtiTLev1JgBH3Cq6CRUeLgaq8xlt7KcURFPijMAvt0VselDLb4ryERUloJpFkFF5MBNSW9tC0FKVFgFxn2AcbYpbXmVCTuEkp/2ISzU+sFAb171/ovzpcpAfzTV3Poaf00V2H6mUJ+jBUM330dTQMamPluSxX5Y1qUzmKvuGHOEWcJPqGjQnG5P9jVgY2oZKM/6pZLv5YmEBi9sUlpn95oK3G4Wf6NiNu95DEeGsAJQnncMr49tnzPhaUzWfvg105hULe95nKc3Qfzgt+8DFos5ThPu6JkVNObp7fc1d5o3Wlxlbd0Yf2qaszCjKbQdU6QZKD7nxDxsf5kjFNCr0ZZhKs4+oozBmCxD4QHlBRTQlm/afzpV7Zm2rz8kCIGvKAxF5FsA/jkAv8xn7O8D8OMA/hUAvxjAfwPgb1PVb3/oPuoN68JMAdosGmab2PcND9dLCEMghSHtgtersUrT1iEiHkJAzEOUp5HJwcVl2qDloH7x7jFsbQK4s6WHvY0b10Jr3KgvPPGMbTpPdmYamANFesfFheDbt2+9Dsl1KcpUq7UtY6RAqrUzVIBKsVWdDMMT4e/r6lYmFgrczZ0LW40100RQKayqIDRBQdX0PMdiR7OFl44MAOHAUepnfiDRKfQcO75A2yZsW9pF1QeIB4ahEkNSInYAxriqBaLPJl5e01HBM13c7J0QQ8e0pXZYqJM0EwKRU417E4W6Umirj5x8lmnB/FA+00NdlILahJZJj5n35v2nhbxMTExx2qtpMYSYhujN2+vFwNxAlvZkf2q38XsagnGaen7eXBiapMV5qGeGuODxvHmA5SJYp0bBsrrHOXG7WahRE5gpy1Mys8fWfWpdQ4cFUat97hwTx9NAbw3XyxUiRktmDtXmXJ9JcsIQtkCdLdGeSPJYFgyUKaYlWuKl66siw98O4N9S1d8gIhcAbwH8IwD+gKr+EyLyWwH8VgD/8JfdKCqyLRtVsIkFSV+v1qQxTb9vXvt1c4eJpc8lW4zATk0BbSe05YhFxG8bIi2HwlJnDjKRC1GgqxZM0qcBGiDakagF4S9G8LTxsRmd1h7hMhkmU0lXn4eR5JjwxhkwPe/+Vjaih31E4aiZqj2QsXtmu8l81Bpq4m/NH0JVpgSisEGqm0tbsk00hdCzpzqDnkndeKDlQ+ksoYBoyfzD+6EI/hBo2QJ6/lsz+5W9lgfHs+DrilyKuYbtifcWVdzIHKr9Nedl9YhzMNY1Atdk4iXXI6WEhUiDCTi4NgJnqIZCm6m1XZoDQAnTUtgCiYacNs5kWUpXgTPZzGaF3xWwAvOIzADS6TfPAGICAyRTP7P/xekFouOW8+kbpa5tp5CAwJw/6IBcnDDW67/YAcgv07zoWIw96Q4hHq1cUxkpkfO5ePffLwu/XBiKyPcB+OsA/D1+8xuAm4j8egB/vb/tXwLw7+NLhKGIJYlTvWGnI1PkuuGzt+ZZbZs5IsYwGq2Hy8UmxlGMhcscHsBqvIOXCwUnnS1WLvM4Jo7D1AbIwFDFu6fD0NY4oVCcl4GtdafVUkjbjD/OiROIag2R+MnUDLpsLW2D27Zj33e8ffsWvXe8efMQtPxRkOl+c8KW64yFkwixqn31ohAk6cJJZ4/PNtVgq4HiJQd8/IKcwNOmVtGCYtPLhVVtb2YvSGFWvfCARkU9Mmf3zTax1n8Ly0+i4d1zse9X7ZwTN9xvwvpZV4UUgWRUWwg0IokMzM416R3LsY0DQBz5WuiOsULPZV4mHYJI1d/mzgPV1fUSpgFCoToAOHKU6YeTeZy12z2P2wGaPwBF2+3Q7Q+2V1zNwe028O7z09ThZqami/SoQAlBxCY26dgBQAXHk4eHTyNSkHGEPR8NXuTMODu7p0qOaUH8w8OWch4aJsxmDm0RM2j8lW7n1eH7Hxb2Ao9kuEzobiFB47B7kkZy2yQEJGChRDsk2I/CjquwVD5VI7sttv+pq930pb3E66sgwx8C8NMA/gUR+asA/GEAvwXAD6jqT/p7/iyAH3jpwyLywwB+GDBVlOE0oZ55itC2d6/3auk/zEGUZoucNoA4C5SnTtqlLMqfJyNDLOyIFIjnR2JRlfjTnGp05ITgBv6RFl8EShFHTlRLojKdI0KiwG0rNYrv0sSWazmwVg9yhHNoqsGpDuuzn8N8LaSfr89xnj2xsae6oxwXEHhVuy4/mgjA3wkicAG90In2VTXBZSADSWGofGZdqC8Lu/r8sq5C+FYb4bOvOTN2dR0KEMGEt7eMwfrMNZg7v0pcZXyINuQ6vQW9xHvrV74TeD7nCkAm799gZUoFFl6S80KTz5y+S3yd6lRUMyVBiQwGVgPBTuNtYC4v0VmqnfZDhMaJREhT68zTzkOhWOD9OQzelvzdg8SNIg4QR3/brlEqY84S/YCJpm6n5jptTmYrJSLCnxnr6cPA8CsJww3ALwfwD6jqj4nIb4epxPk8VZXnQVf8248C+FEA+OyzB+2bJb8DXkZza7g87Pjss6vlDF+aOxwutqnmDm4jAIBTAVl/zXMnMnOBiaB3y+cdUzCehi9OU3OGh9xsWxq7DVkpdAxYfvCG1jZ0p9RiSIk4dXrrlton3YTu9XrF9eEB+2YFmbZtw5u3bzyoO/OKX7x8sVcV7J5rcCnadA63e9KRUdVkFyA60TZrH5AGZNq9jgOxUQT0U/qice9gegDpLMqMDKp84oQWmQBvfyfSsagzqn12kk/OXIlJtG/W/hMFsS60bXSG+eYE83htw9GLrC7wx+RBaXZFQXITCseUqluQbvD3QiO2ruUY9/M8QOElYCJBC6qq+EwgagNzEKD3zdcjAJBQ177Ok8zQRUsArMYwgLPvEFzRZKKJIfJ9M9vkOA9MEaAZczQjG4jiLFTHcoofLt00nGFux9Y2E4ZNAVFby5cenmr4AapTMQ4AIhbgjobrwx6pmTz05iiZXUUgNrUMLPU87LD7TUUXWyOXq62rh2nphX/+Zw6oelKD2BwOONJ0g7G0ZBSKQ8XXvtk6X4p0WK+vIgx/AsBPqOqP+e+/GyYM/1sR+UFV/UkR+UEAP/UV7pXISCRyibvHAnX3RsVJI81rh/gJ7wu98u/lAHDhJRGCHdzT4DvqCZEsL+KGawWDf4VUIXh+etNpQy9wD8fEtm1ekGlfahVTNV02SAhzlHbFb+9FhREIXoTfM+hfUJWU+wHIoGjrCkTayyclETBhXaCwZSJ5m6Uf/JsgSVKzX1jMahWh83Ma86tRoChOe//OKIvnCK7e2uml3GYYtFBTLWav3D/lsjlSSCpakVmssTvkKveDUNoRaNj77TKumCFIjJbsKgv2JhIjqhFDnXOYDdiWdSvv4eHp/DX+GMsdt0aQhai7erm1hqHAbC6wWHjeBQy3Ag8iLb/Tk7ttFvs7XIWGipdFNS0wlkUMTGoLdOD4u4t5RyDdhBmpzloDOlFnW23B/Dn3TKLc+PnnKgxV9c+KyJ8Wkb9cVX8cwK8G8Mf86zcB+Cf8++/9snuJWNL1ftldAAKtA5eHjn5p2LcNb65XE5DNmaK99OJ5GxjzxHmcVs3tnDg9HQwwNXqXBkjHOT3WcDLfkuwdVgIT0KhzDBjqmWpsImgbIBsUHSPdx94DD03ZOnrrwTr95o2Rr+7b5ulTzdPCZHEYVVWzTsuMRezI0JERbVSjpIJZSEQKSsbHqW+IOaq6F9LCBaivRKFQ11T1/f0huhyVmWeSwjBbzTg1KwHsTgumCPKZoTbfrSn+L0YFZpu+x33NbiTo4uVQoy4w++Mxdz4+TZsjH0/vVLgjRXDiNJOHKowMY8OcPQ47hoGkfQ6WeyuldstdOJFVQOQhaWNqw6T+3hEppkT+AvqS6UAybQVosYZbB9B44MPi8KAhnAHBeU6McWC/7NjbFUMV53gylLTZmKoC6rbvqRa7OMb0dDbzuL65XoCt4WE3B+MXj4fFIarxbrddAGrjcNubZ4f0zYDMdr1a38g/MgVtdnRt6Jdua+dwATXcQVM0H1sFZo4SQdYW7/YdjtivF8V52BiPaTnLc9v8XhR2vqQYTjNtrze3+fJQ/dD1Vb3J/wCA3+me5P8awN8Ls1L+qyLymwH8fwD8bV/lRvQM960ZaWTTGNzePfyjnJ2Aeq7wNAHohc7nsBKbRD80tJI4wVgqnCgzksJtgwo8o0IVrdUcX/FuWYnGWMUJeExwOHK1LBIv1OT1YO89xvXiyR8IKV6/R3aukjJEpqYGFiFTUYSRwEjcN29fRlJfFk42HmQ0ztxant4IlLf0Js8I/z5jgadZo9oe41zxMTW1jEG5vIdpACKZUrVtJKsIYGHCvc1gvanxkoAHyAtNKsliw9zc2jYqcmy/xOGRCPUerTNMRxVB5ssxsLnyviphtsR48WVZiBsa4OYe1k+Gj0+tcUkPvSlKZiNXvz89qbHOvA1zKOapFrytQG8a0Q9b2yye8RyGCr3AvTBsMdZOOhCbP0e8MLw54QxVN7cZqqv5HZbFY8jWBHO1bSsPgzDL8Mv3D1Mtu4SS1gQY/rmKOrmPbazXwHy+50Pi8CsJQ1X9TwD8ihf+9Ku/yud5SQP2h4b9wVTjfTNuwct+wXWzIksyE+qrTtyeDowx8fjuEcd5YpwHziBO0IDPcyputwMiI059en2nk2padbkLiPctH9I9guKQHt0S/CciZor2MxKrXq9Gwf/m4Q0ulx1XJ1+luv8h28RUJEJzfWH64mWK1XEk0UJVjRVu34JkbFzE+0mcrug1y8bHXgSXSFwXX7gtfubrYXdjetOkcHLBMav3mX2wpdiWbnuwLhBtG2PiOFgDJVGrpcva7617KhHlo5DoV8pHqP6kYOJz2Nfqubf7eiRC2L+8/ZqoPXDeIuWzt82jBuz95r2f44DqtNrCUGDOEEa86GyZrq4JVU2GwCqPG7KeT2iz+2zSXb32tTKteuGcA8fxCJ0T0ht6t6JlJgydvNcPNNGGDjXy27ahieA8mq+5GwCLxLjIhuH7QjbxvF5bJ9RMpiqeDg98liMOkhgzNRslSbaHSU/0i/kKVE7M01IESTTMeZBpNkAF3Gzm9kB4WmNkg7Z096hXKXR6rtuNMaHUFgRd09Hzob35unyGIth2p9raBPtmKtC+bV5tjr1NVGRqsdXxOI4j2Jv9juFhnmCgNuPArEbFnMwiAVQ7Ltc9cZU2X9QuEAINWfF3noJGuZU2wt2DpykEL5fdw3mY55qqMeJZiNeYM5ybmq+rq2un14rJ95vNKzf+vLMTcXzT1rMKQghwH9oTgtDetAhDqpY8tbmGpm/6FBz+ewieglYr2SoQ/cs2yWIyCDWdQs//D3tfAdDlDfbZJfg2nxsIvfb57gbVhEGVeXnd54FClt57kVHmJgWnaMHui+C2ee/NBBXH1Q43cVKSFkXng/hMATDGD153WE9zqnjsqNVHtvsxb5cxemiCuQG9XdDbFRANc8sYB0SAh7cXtC7Bxj5FIyU2NSZrC8O4TpwRp1u1jqEwSgQRZ46RmFuZzeLMp/F9qofZ6TQJ2trEvveYa64p2+fqAfSCzZEtzStjlHvpxJy5lknyQh7M912vLAyBy9VyjRtPQjSICkRZh9WJCCAhGE5HSgBVJwobpGrN4xYRP2oqhACscTuYhucoj4iHleyYUqVwwdc3v7/ZAS/7jtabFWz30JnWuRjz+daGunM1vL4sE5obke/QsJcxvEcdZYyZjDAhcKiyxacR48Y7rspyOhGq+kBBGC29c9BUBBYqI9bFz9rQtfQA4unFwaXrs/iaQqEjnRE1LCQDcZ97A+2ZRaWSgvL4Oj3dofqmwKwqfGRwlNG0A3SWDZb5yLQLrs4qb580R4iTZ3vMPWAhXAJgm1vxZAt02Nyq5yNPhdNqIVzn0xc3kWXrXmBKYIXiNZ2NEHdCNq7tielu3KmmFkf2jqvonTY3MX5EQldTgy0Sg+wyjw5UuP67O1IaBJujXaq12+4EvE2gF8U4PLVuFs//aUWntt2dJZsjus3Q3aSqDlPNBeo5y6aWW+jcQLVJMuHAVOdE+S9drysMm+D6sAHTBKClhZtApNrLEAxG/A/aCF0wNEbZ++ZsreNyuQKApxAVgQL/ckF5DsV8vJlguyRbcjD9K3UXLyjkmSPX6xW9sWB7i8LtFkfItMK0W9TSoYC187yrm3x/VSRBochFYrbPTNdKx0j9JF8iqvOfy6bn2BC9zokFucEF3b19UkE7mH+PR6/ciEthn9Kx8MjieX6oQg0p3OUw23ophAtCoVdHLHOuEX30v8haDIrhRZW8o6JHqsAAgqKqpj3ScWLCfSZ/I8qhYD+twp/rUNXDZvK+F3VVELbuGE8HMT+equI4vIg8v1ydN0EISFf03Y5w8hYyDbI50UOo5DBESe5DaRpOTFa/k2YHyOB60IJuZZotfg4c58TTuwPHOdD3hrZ1XHpzflBB44EOhTRg222viMfTz2PYl06c05l4TnN2bZvnXPu0bnvDgFjG4shDuUGcGswjEOS5MFzWfwENL12vKgybCHZfcFCjDNp6d3YOwIpj+8TqzC9U9CFekyE3tjr6OD2B3aqRChCBzvD7eqhAy01DVUSai+W+eZGmTKF7eLiC9E2tWW0SS/VrBQ3RTyghyHLk11AYoCBCpeGearJ7kUvVuyauthMHukBK9OeqXBE29MoR9ZhAtkR7BhNZNtCMZwC8Hcefws83cKiNL4SUxKbJftnL3q/pXvtRCA+AO3RKldT+rgXJmrauUE1TRGsEkHnyhyPDpUo9qFDUvjgIZRWm9VqEOg2NFEb1XmUQQnA823bej/LKdI+8SOk3KPQty6RBAVFMVmZ0D3UDTN0dituTCbzuhAmbj2nz/oWN1G/AkClpcCemq6CKrLSnyEXKg93HqncPit5sXTGxYGst6OE4H8PHS+eBCUFzLdDsgFxPA+i+G0XQdk8d5Z4fgl27R1fkHlMfQ3FLAlw+dC8pO5jlNIs28wFp+Oo2w4fLBXqat+166dh6UgNJc649VQ9CnVFVq7UtIHNzfgTSjisspOTpNjxmbHMVgel0Gidkc7VrBqOvbYRQhS/X9BBfLJj04eGhIAgJjzEFLXNjeZl6X9GRBtq4tyOOmSl14zwjoFpVsUchKHhQaQYLMxm/BxEFchHDHAb7Zbdxh33uOPz5HqkP75OO4dXG4ILHw0nU0vzUT94yky4QMrYLsJCgqqpG6NKonv0UFBVV+mkYiJPef4aw2DhTNV5Zd9TVJ7N1WTvP83RPdBGIkiVEJeYvWW0WdB5qtUM1aV7bXm3cRAAx0pEU+/7//aajNF4EJ2IsxNVVCsImHoYjagJOXRg6ddWcxhiNNjFuE7fDUvou7nS4vLkGH6IIIKxVrBmrezZbU9veAzXCBSJm87zi0gV1tN3VWKO6MUc3iIGDrWPbBFs5JKbCwmlgPJwCxYaGDh/z3qBz4HTTWOuWntuvPkfu2d61R62iOdQrUps9dLgw5ELvvVkZBRmYp623EYc4Pni9ekEo4f8S+kOoXwBP//QmUwhNNdU6r6JyxcFFp4C5nrjZAkABCEOwC0GWBTUiyGS/ttS6zXOO26J2VUGY4QzrSIfKfPc6bWK0Qy0brwjLavQv+M+eD0AdkabDpnwOggWvVFXVIFagPDoGXpys8uDqnOC97z+m/l/0CVXNqrdOtFLWMXjEK4jkqREUVTxQXSI8mgYEsKygsDHmM+JQ4ZxhfT578Gy6yjzw+70DioiQqFmXLwrIdQyz0xrPtcDgfK7Aaa1EobNjiuCgxiC8Z46pBJpyYV+88iJ+CE6PW9QJK7tqghdMINMchzRv2MXMn22zw+jNwwVjm1HdUVrGROb8e36Tf+eRqmKpflZcILOGZCpaP9FbdxRp40fnpAV332sh3mZfQ1G5scFkhlfsG5O5Vi9fr8xnaIZQ9dOcdSrO88RxnqAaDCBiA1vfDRUEgarz6DnzbRWEfWs2gG2H85AjTmzH0RNu03AShv1yQWsd1+sFvW/JRO2OGVOPM4Ca6pU9054d6WcULhCLkFdE6IDLKAjIEjKCSTtUaJ+srEznjhlFhBGJe886cmMDKYAYeEzEVL3R4XBhzN2wgWm9R8aA3cuRUDPXEwVmql5ZxIvtDhUeGv2ibhrrVly15SElRZhTiNOc0ZwbT4v65OMfNXRcYjYfH+Pdy5jEag+ktz//FmDN+0zvPdUq9QMQrhb2BT1bpwbEa39YKFKi56lp+0WMIZlauh9KiL/RDMCjunuUxNbNyLZt3U0xN9wwgj2a49Bax7bZum3b7nnCExBF657pcU7oMX0PHcAUHOJgYOtuPzfn03QVl8+QJrg4wetls9fbZ28BNJyHOZvOccNxPmFiYujpjNeuWaiTdwBWkL5bdtY5J27TWKTePZ4QETwdit4aLq2hS8O1d2Onb5bldZ4TT7fTTGJn2rAV6jWxBaIe4t4UbQPOY+D2dHsOTsr1ykzXps5m/qYCLCkIIhg4Miy/+8epkponqXgX3UYovvnuIUt2P3agcx32QIS935MrtECEz9DA824tiIJIYPHGxqYzuJUqcyhncYq3l55H21oI//s26N3v2ZZ79Zye14pQsl/FGSQVxaVqXIDoMsIVDcbL0U63S5U/iLfTyEHXeavoM1HK2geiwzmmj68UgVfaXRDhy1dpPw+QBRPl+5Y5k5f+ttwy7su3h406lJp8trANfgDGAMDHyBbHuh45FqjrJR1oqhONhA5VMANgzc4AC0o1WWPdxHyzu3GoIxik1FKgLKBdLEyIsaecR/WUu7w39zjH3EyTVsaBYUStENDYfHcxUNS7HcDugEeMsPiB2sw+aYNqBcLaVqLIX7heVRjOqfjiZ294ejowhuLtZw+4PlwgYrREcxr1jw2UCQxWYOseNHueJ25PJ8S9vcsGZQk3KBhvaMHvZiNszeshb1tQbF3cU0z26QiZcTSxCCSuUV03T+UQjI17pwJzkSYFVP49AnE9or+VRQ0YnVJvRvHETcLFmrnKpoSY0dgQgZyJDAGzyTYkSuIm7d1Zt30xUfBEGphyUUugJ8u0sFAR9mP43HHH0GZFNWZMIigJ0k94wv22tTuVlegqyzeoGgko0bzVkGm43Q7zRAoJZkvWQStC/IWLZoMcS0Q/aP+bY0DH6TV9j7vP55yjCJzlkONzytp51gZl9ADDssSEmNjcnU6fJXAUZ9Y6iJOVNA+jETScpwKiOM4njHlAmsVBbtqwOfbctg21pXNMYAqUB4wXqYIjdQHCbqwcH1VYJIgXauoKwYbjHDgeFRDzLtsEmEJsdP5qAaueW90UaFPQiM4Pe+7WOnZ0iDbbv33D1ne0bmvmPO1ewx10CidR6YC0gTZCamMMY8j/cz/1My+uA+CVhaEqLFTmYOYHbSW+CUG7GTM1Mo+RpJXqzgpxUspQre3PACTNHy47svZsS88XyVe3UqO4pYH9nok691IKPfteDe258ZllUm1HAIrKWlhm/Gquty1lAJTqHoURsVXaHa2BvFVuvlnqdCgK5VIvrCxa4FtBT/EdiWayTSvazDFI2FqFTxWIHEPV9W92YGRb6r2ZBgbAKcgkxmQ9rIi66he79jKqR2lWItAUhMvbSp/XW1WBl58T8bHT+vflkYj5uUNIcyhUrFAU19NSpfFuTnhYKeiYgbGST2ft8TrgjeaOu5jKaLprKOpB1ynXtXxnWz07KYbeD3V4ELm4PbI8ybaFkznM8CAAlAN2myC0EC+9a+2jXXT6/uQ45Pwh0LLvp8510n2N4b3XKwtDo4/ve0cH0B15Hcdp2SUeh2eLwSW9qzjnOTBk4BgDY3ooiNPun56B1zqN5A7fXR3pXoVuv1zw8PDGyFc/ewuSADC7hILCvI1SAmLXjWC/Z0bFnDO9xZMboRihFRjjLHGGc9moFDd8oca31Qmk3cw+keoGkScj7lPw5KYUrE4H2hRjXoZ7TaOOTF01GhuMG0DLJqb3m4I5qvtFbB4FXDTJhTMM4Qi8wtrIOjPl6SHE7w4r3tPmbUs0SBUaPCglEvjT6WK9TMRWsm/cJkoSW3rbmzCzpwqwTI9TzxJa1f1yBMRBpndjWDtrbbgdGQJjdlgb3xVAuPf00oEJK04Go/KydXlCMdGbujNFXfOw/GQTetbWIGs9jmiXBWU3jM7aOSTTMJsiy+sSlUZp0qEAD+kpdsj5GrxsG7RLCNQ5BuY5ICrYxULBNqbbDbddwswWvQu209agTmCcRkc3JjVAwOqWqpdcGB7G09AFwPaBAxEfo1SokpIn06fGmHh6cmbfaTmcZHg3z5UHzc7clLZOc0NUZEjbAb83R4KWQne1wkyXC1hr1zYIA3AlyQNQUU0ih9oXqshBrlqCo+tnWDuE9UxqsDaNMBKdKegLiMyR1u4ot/wkXwHMsv3iJYEstshsYyKxatvKzA2iFQ11rvgEkfGgz5HUVAaLE8XcIy3OWWoAREZ+9xB8OUdpw6XQrx7+5Sp2RfY3Ea976e/uY8hSTZvw/OCXbMbV1hd2WbvjM3W/LqKwI1PglAOCa0AV6XypAfhx+CbbkLW/2Z5hPv9tuLNlApiAM0SxbApaOsRWmi9EDnITE+xW9c+D0okmfZ7OOTDOid7MLcmEhWUtlCm3Q8ocmNOFu6jAIry9GiDgLk6JtUf2HWgtiaeuOcLznAuMlQENrQgAmF2D53NTrldPxzNG680328TtdsPT7cDtNmIxiAAbk7SNWwq07bRunGaqPVlGGpGgEyt4jV8iwuv1iuv1in3f8XB9MOdIxJrVkItsa1XrUn2xxTs1bUtZp0Rj7CtiO8/TCvs83UrRdar33cdF7p5d1Cbcve4bazoSS9sjEZtdYSd0lGxG5Z5BuFW98P8y+NdPfU07YBUCHJu0iwJ5ACFKiOrkoqyCqggvf65CI6BWeYjBMwuIaEP9RYxjxBn64wEk642/OCcdKB64HcJR40MpXD2AV3PeAYvZ3HwBGyvSiDjSDLFBTlYgQJpFRvzN5sN25fSsiVAtKUMM0np8IaWlLRqdLANhX+dBFmin8IJ6UbNcGw8PZife9h0Pl4vpTGKH+O04Le1vUhXlYJrnm0LJKPOebN37umSoCjGIrWmxLJOrjRXXhN1b4czOMVgS85iHVZQF8P3ZzRxpNm8fNzMBKLbNqc/UnjdGs+wUsbVjtGUmoMeHa8h/BKIGL+zUWsPT7XQV+cRxDq/PYQigdxpegVgIChdyDWOIq5uZiB+2LReGm5flfHh4wJs3byy17rI/O+nT4+hCSQEyDedZkgKO4TCG9Fa0U94NEk1YwPPhqYatqLMlnY0ffkEQcpNwo66Ok7mEtrARpr6iFNGRoj6mEOADlDBBDIHQcRWV1noKnkBCgSbX89YC5l2F1Nzs1bMbRBK1AuCcQNEYBOqeUG4a+7wVvYer2gUp+hhMhRnj4UQFkBhr8LvPUhXUFIg8HYjeKtmDwrSWMc5AZhHKVE402/urbTXHnM9TX9pFmE6+thK/BoKPNWHvG5hZeH74PT2cxQSl4nphWMqOy8MVxol3hkAZQzEmkTJt1+WZ6kkNTxZGdPqMNPfaAgC6+UO6IDQ/0/Do2eZag2kW5XBordtcethV37bwVNvRQaHqq08nzmFt6d1Ly/j9b2COt/PaNAvk1pnr433X64fWzIHb7QmA4DimF7b2nMwmFh/kFcJskazKhxG12uBkbunmaThZq3jb+qISXy57MGjnaS6BnIDc2KEm2RN9MlOfyZojiSZDUIUqMzyQdHgKmn22Mrn4kESNilo4YQUbCjgTBzWu3GR1ipMwIJ0MXhCqNdT4vHofhivZ+K7ItHnOW2Z7GHO4TqP0NwHDexXPtuasRc4rgDVYmk6yFDw8lFpLQX6P2kt312sRuqtd1MbaMjoCfpVDkH/nbdMEksJocXLMVFcNEW0QF0LQkm1TBWF7qRNrh9IWzHZ6Uz3cxuqWMHeXApLr1z31rdtB0mxRnSp4PCbaprhMdRRplSYvavbip5sVpmdaHllds+40cIVllYhT47WuMTfKA7WJFeXivhJDauOwdd6dV7cxHLSRTAVWKU8Eu+uzjPfUmSQWgDvqlNvC5rJvJqD3fcPWG6ZaSqsJQ3fOtvsFs16v60ABMOeJ220iabUEx2mhL1ac3SubTaR0KBA6WEfQnLShuTA0nsHeOz775me47Jurxls6RcomqQID4KBi2Tz2uy7fgbQTpoBj20yoGIy3hHiyVVOwtBBK8YBQ9ymcq5PMcQ3IUfdsTOO1O4QqzNNs2HY7aaNKno9FciFOYESqv333G299W+xuqmLB4U0yrc/HZ0zSpxH1+mZyh1SOP8nptEQVpDBszcwkwT8nuQYQyKqgMJpLQHtvMtAADFWBz9U9ki8Gg9IXY4g+MaeheWk9GHtYUtYC4T0Mqyu0NYxxhBq5mBDy9tlmQX16gX1Im6HLgPSY242MD9THEa7y3vwG2qggQaA4p+K8Kfo2cR3DtbMdooqHLo7KH3HoGUIm22wkDB3AvhkRrNy8NKmnuUZQsHeSArRRS5vA7TBE3x1RMqOkec0dABYSI5Y8IQLcjtPthfDYRd+LU834qQpliqYfnPtmIWhjDuM9FTijDcBqlu+7Xt2bfJ72ZYjAUN6cpnpY3WMXfFzL4rhNixoMs/cZv2DHvpcsErJOe1H4mpFRN1HawBDfVRNZ3Ye9EB3SfkZkRa8x1IJFjZlkFoHpC96dkLRXEUnRK0hEQ6dJtEpqM2TRSO+920TTLupT+LgkuQ+FqTbPRSAhizwJnTx8FgUPERXHtfYl1E7u3ftwIYbRJLLgGm2CyDsO/kVnplkYdjgeyz1TmK2ovXiN71T6qsbSPhf7uokZ+MvnlwNVPGIBcOGrgGeZRJGyZf44/neHBoiuqyOG7V73r7gtEY6sRXOOWPIg20RTge23MU29BBoOD9lhKQMRK7VAYUiBBZhTJc4hXedVBPZmqKVCtpYT6vtDxAvDw4R4CEMBZoylR444006sH8p7NwsYqC97Wc02OD2+Nlh6xIkkYMKQHfsQNn/1OMPHp4HbzQShK1E4huI4bYP0PiOmUCBeTMm7oYAxATfPGGENkrfGLuMV6a7X3YtfF5UY5XRdNpH9R2HASa8qjsFsvp9pWhmkqx5ac54nbrdbqFj2ARcGkSJXGWzSS2x9NSHmHAWgMLTPry2v9qi6obUs1AjEjQNFl7bdbsYankiZ6ArYLyzVak80R9Bqv2pBEuHBus0QQk2jA8ze2PsWweYiGYfJ/lvlOiIGQ4aXy25ZB8YxtYbOvGd9TRcu4t+5oauArhdtvwACDWcOeg/EbyBNSnstPU8n7Yxmx2zk7OeE+aEH6DL++T2G6RnyX73X9pmu1CpsDLlmeFCaDMrFIrAwlDmA43bi3dPA1gVj8OAw+10TsYqKuo7TnIrzyFRZBZIhvcFSFDlevQO9B9o1Wkab78uuUCNhT4YkmGNkxOe5tJxitimaArLBqqN6ii6GQEZLk5kaP+L01MMmamBoz3rnJwYOOXOwX7heP+h6ADRaB7mrUKK7CgUeCUBuf8YNSqDC7ml09BpHGt0zlRjLLlgX3WpWVQBrCMMKiu6dB9W2dF8rg+22x7MtKQheHKPSY43G0oZZkeBz4zyQqOp5HZYMIeHnTnIrOnrkZ3l/+FjU5wUaKH2rKi7AymsK1mW+76r18V6tr6l0fNeE53rlKliEWY5xcQvl39Q0Ch4EABiNlW+pueHxjOfZMPeXHWwJaYm+KxpPr/W65pboAz73xeXgq4GIrHaP95G0i9J+axR3Ep7bps7m4/eZwxwvdnkdIWUGEoW+P5qqeoyhZIqcSMybNHF7UYwQIGrCddqBqDNZp6ePvSJnL4S8rnuScx5mfBcPDSZcVWmq4rq6O2GIKD8gCIGPIAxvN6P5J0lrEwBeU0GcXTqdDDXmzIRhbzu2bcflcsH1wT3EHi6z72sWSVyhPtVBIZx2wekW3fMcVnVNzRkiTbA5uQI3d5B0enL+1Dzp2doIQHZ00XrzdMJZmpX2S3BRS6ofzDig2i13G4GCVRVhh+xeQiHVTNx9DrjdzBb2+HTDeY5Q+bbecdm3ct+sUVFps6o3PPut6GJq3ObEuY+PTzg4lmfsWIrQmBobgw4rwzAxx4nZzIPZWoNKB4RECymMQ7W+CyYH4KQX4imFUg6g6lQRzDFwHLc45Ozzu29YF+Ycw1Bjc/6M2svshTph9UiaQoY/q5jT3r8xXn7R9sTaV0pOtsJsao7UAUgXXK+2h4aHn0lTtAlsbaLphI7hjO8aBKrM4tq8dIDxhhqpajKwl7XbxI2SDW03bs9slc+rB7JDFXrad0YE3G4j9oK4n8C0dLX515nS39Gn6sQJc4rsV0MpenAd2fvMGauxP7NAV6Wge/l6dQfKom74qdKKjWU9Vev5TFTYHRWSWGELREgj6vq5u+c7YkibBrGhgHYbpgQqwFx2m6SaB6yZVsdNE+zQy7GG2Ey58fkHfda+BRvdob9EZentrY6de1KJfM569/AguyOA758iCyN0FfCm/r08rgxaSCElpSN+Mge04RxUlLt0OvqVtjUbxLDpSn1Wjmt6kHNM+DkKlxxPKeNYPLhsQxl/Wfqtdw3mIEggs4yvM3SUXavPRv5NX5h7pC2xLpXodxn9+MlRV/cUNNPOhUXxDHhMz94Ypk4yPW+5k+TAcmy1vKM1D/i7i8zgwR0Hno+BQCDdxo3rt65hPpT7c97N0f26o+CEgnE1RevyPe1gJuaXHugPXK/OZ7h1BcSZmtsO6ZuXCbVjkEqHOhY2e1oz46wY5f7lesHDwxWfvf3MEeEe6AuAq+PZc+HkohjH/T8KxzFKrVvwtDQVbbjQOzyAmu+tbNAUHmNqTKt4+EMyKqdaca/eUhtRPg/0Ziahg3kt23JKh4Okt7wRngtS9jpfd0FKdOejfh55Gz6nCoO0ddmXiDjjMcJ5dDssm+h2HDjPYTUsuofjOGMzWB/k/vIXzZanmLM5KrbMpNY6Lte1HCudZPfjGeMjZC2j4yjTIc025ejOBTbR30IfD/iGynxvIDccCzMNZcGmdRyVtsmC/inkyU2Y85XCmfdpgRCtzoiZEGYKMoEJQbE9RrOTjdmGoR06nnDcbhZ2MhTavCh7E/Tr5rRfFlGwHJACQJrVuG4N+/WCrPkjOOfEOIal086J1hu23dB19/dtLiDPm1W7PM9KeuFC12NHz9NClKh+k0NI1MNuADRvY79QE7BDc3h84xyeluhzMCbLlL606Ox69QyUMDDzhUgroq1QXEgRITJn2AtM71sQsDJsZttYFNyvuw7HGfNsIMomd+M+XyPiMC/2LAwqCbsXanxN1VJBwzaF1Ypk4pkvga2ijvOEjBAYANrSCQNkHGDtDdFM2MkKGnmGXOPvWEh22b76+dVbSweQIfLKKjfcfEBPZVMG7FRCi7uTekGSFEaASKY3yrSi6+C4Ll91TJ8fNvxM7W99vSIzpSqqCLtxPUhqw6ttWBpgnFL1K3pV8tafYTF/j8Z7iaarUH32Wa4t8PAmX2QGutPiqmiYYMre3ToQQetSquMVcR8I1w/13oxToDgVMZOAZIyZKNE/b7KU7OCJ/jLkaT2cuO5FxCm/UPpqkIH7qvnczDoP4uzt3sfpY69foim/OrmrtM0CVD2P0dBrgwcZme1OxH/3lLrW8fBgAdTXhwdcrxfs+4Z9316QJijCwQVbxMkVfcNXHBHDdCG22J0c9tvAclMWh4lnWvCQbIy98wcYCjGbHO0jz+1/hTnbFc5xmuClqmHV/xiOYMJgY9w9CU6jzRNjMB0v+xGtUizCdfG237msRYw/jiogP2t9cxPBUIw2Q3ukADFtyTbNnIp5HADcdlPUKLab6nkidc6UAhhoDXjoEvVygYwf1IJe2U4KRX8xTsPFNE907BfbQPIckm+YSSezjjj31XzBMCALtp8Gc2y7RsB5c1LXMNX4WFdFNQ8w+IFAsUSBL/kZp+yy+yuMJotrDoCejooOyzDRE2Sw6ZsxYfeLCVC3xhkpgiqGVSyxlrm9e+uczwGZhginKm5PJ45jDU6HDKjA7iqABLmuRYtsW4NeukVhHC6CrTw2Nh7wPgh0tuaOkbIKcr5VTe13KngrMqUkuVCsusPz69WJGpgPjNahaC6tpcQn0QNqAUatdYsd9BrF14crrlcjWWCR9xefE4hMXFggYH1FTM/tckzf4clsGx9qqgcFZOUkzIBiDz/gqY5czKGWIhd7qnlSNgRCdQUyY6W7epBeN272wDNpKylMMfcCcUGLFOBlzMJWFWOxhidVh0n0Z7JdYgJmJieftOaI82RLbCyZkUIhphnorepZaR4IrqroXUEDRPbHfjP5qiEIicoM0K7orPy39ruMuTLdr6jyMzZdmgk4HzZPrajrvn4xYy3b0nYvO6sj+njl3MnyWhXW1YZGG50/GCKbI+gyL5N0+wodGodj87g9YbxfN0Q7nTdxKkIgQic6eil74UJqqrFZn6YWHzcThmGzlxnPV7V5H6LQyTk3lV73Hg4QagrSEKm4eWjn/giHFjFiYBuiSUQBNYu7HLEHq330pevVkaFqw4ho8hkCqvUN4GknzbzKvePN2zfYtw1v3rzBftmMf9A9tC+rWX4p4GVOEtmUN0f1OZ8E2gfDVkIiSqfnJ8pjTGF61hI58fRWnWB9qLnYdXwKQ7VjDejaMkRoENPnQoVQ2qhQQG7ZwKX93JS0qQEpQKoqvAxinMI8tBK1pmqVBwSFbhw4rfmCdiSkCjCsY5a5boZ2gSqMrVOpOlGwqPfLhKq2UnyL40kFqqjLvOPpEDOzbZDjFJ9ZBSYBGCAer8i/MC4yvaREk0bRJiEkFfk3Cjs+pzlNl8lqx6o+poY4sw9Sd3y0K2eNamCDo1M0Nx8zH3567KWaNnXZLXOkWd1ktAmIQsXINYZODNpFPZ99kx7rRqfiOE9PNaUWIGiw2uPznDin+nzDkKh4GYNmJSaaiCFFYWytBoCACPpmWs/tdkQwdTIleZxj93ROLx16HrNEX5h9fd+b0X7t0+dqtX/fX6+ODKf2UAEoDBkz6MEBkEbyVReC+4Y3bx+C8VnuV8R7+peCB9QuFhvMaqCW9Ubq+cVRq2QGA00EywLLpsrPNwCpRil7JomMIzMm3pELvvWOBgTjNpHinBMnRvZd8lSdXis3kKYwVQzJQ3eHhO+HTaBeBS3pvvjZ8/SQhanZHqJpZXiLqV4yXEipoClNETleS9hPDDezJQoKEiIJF4bKXOaZSIsoGYWMgdJM4RUHzewAzewEE8ou8Hu/mwHek/+Vw8oPmt7SIQRYloZpAhlsn5qAIxOGJJX0wIpws1B9Ik5Tf33tyP2cea6wwm1yG0QUfQNUB8bTgbNkj+xbx9s3b3zCBiYGTn2CYmK6VrAIQzR0D2WhoJtj4undzZxb7gCStrkG4Db1MTFOdQRomtbsDdKNvbqL2Se7h9aZxqURRtf6BlXggLPvuDoOINRgbbYejtOQ73Gs4TOXqwlDoGGPg+YMR9RL1+vnJrsaw/ADm6hKp2V0/tfLFX3r4SiJsAm8LPueOwZsMi3PVzwAtW42Xd5b1UhjxRg4TzuZqA4n23DxSotjewjC2KRp4DXlDhGnR+Rrz+LCd9nmM0U7WmRm0BjMvlckw7Y4CBHh90oIcb/NubmY4aPltbwvkJt7CSMKFagiUVcr9YVnNKuPwfNCFe7RdSGKAOhxryrTBCmcc740YgjvnSmc/VAZXdVKoQJfCxmOxPHm+3N8EYNbbYStW7U6lBrXWp5clpbfZ03pi5uLHRo27y1KncYMRwYN4rPS2BfGALJUhjG/DCcGkSZxsFJDmtPsmarDvMqOukiVJwPQKNQ+0Jrg6KchzylResDSJr0kb9shrWNgwOyUE+QepEhn8bETA7Olqj+neX6nmhcZYnG9IoLt0tG2hmMMLwdiMHiCe5LI0PaMthY0e2Monm7GFB4s3x+ShHh11hpgzOYCwYSg7QgP3uw9UuzefvYW29bx5s3Vg6hXVYZXXVdZg6MsuMrLVkRpCsX6c+ZqngdT6zI1z5wgZbNEk1bBCllTjkSMJcTsifxyRwqQGxkpxIBilPc2Rb/Kvam6VMTLxqXaSKGR7eE+pNDh70RCZEBJoUh0Si9dClCySVvBeXdESNpnWuueFaGwkJoZ5L30DpsHURxNlPFAyaip5gI/BChwExFzHFYUTLss11GuE6S67OwmGTp1bxKwcTW72AYZAzcctqEXMlEfR2G62OoYqWiWB6cdhoI2Sd3lkCFMBXy2OPmFIa4xFMdxc0+xqfXmUKr2XsTamtMCmqcaC/aAqcss/ymHAKcFvJ9jGojYzN52kc1UZxV0NPTt6uDlApENJw7c9DBtaj6h9BLnacBhjqxV03s3nsRhY/70eGQe+tZwebig9YbH2w1yHhiHhfCYqj6gKhinOxR7h4jxHoypuB3AeBrYNuByMYTKmkbvu17dZshNijjJU1XJFLv8qp5duzTQgv8Wd45XS39D5dIUCKsHcbqQTkLMOYYzU3vAyIvCJh0N6gbl1PmqF1A9R9M2ey0VYIwovtmBZcOI2D0yvkziu/qbuGVSLbxHu6ai0t5CVZICFHA0RIdI/ax72an+x9PuUGn5peRUr02Zjij5yXJM4G7mSjZJjc1Moo3Vj5hXXeMvHZaGgut4aWSHpDljRX/8PAlFnl3iAkxbrBGizkCpjn7ux4TtEFjoVhyKJEsoB5hgTbkLZ50ItE07PJAHQx5iRMoMEzJzx8IVx0e5xiZqAnmMYUKxqxOTCQ6xlcCl3kgZg+n2QMHuESJt7pAoU7pqDTaSrnEU7YJZZwrOi8+Nj3OQHfHjWufJ72OhkmazJeON2hyaV/n5NPJ6XWEoMHWYKWq+8KNwe99wuV6NdOFhLzF0GpPLbWMqFX+w29OAfy8MFSgciTXOKplbbrdbUPKT+nw4BRD4xLD16Cr4PKaOExrOFgUis8a9zClkTSiZ/cwptkLNc5OAaiAoepPB189kewaAJFP2pTYBU4foXYZX2HPSBdWwSQ7IUvHMvjMUyBYcY8tE05g+R4kZBMKcQFTITJ7zHDjO0/p6z0ATKqrZkrNUay+2zvivCGM/UCFYBDTfCizz0ZeNlgrtpJQogpDjHEjTYwcXROrmj75tkDahB22ZPHA39A4MnOYVrw3T1ALQTFha/RQjAOGBJUhnWsyut6kzprBJxNuLV4Q8z+bza+8lce6YCjlNddx3H0OOldt5+2jQYcQHuAFTFE+dkYrOQNP8kJYDTSf6Lmgb0HeByO5jfPV9ccOcE8fTk613n6sxJkjjlVRtka5gdY7odBNjs9ZpxBiAEYO4Acraz0BzdT/+qTg8VKd3QIbGun/f9eoOFIgE2WdlB2GoDFPrqurIi0Z/CsKolcxTzb4hZthPiFxKyNeLCpUoaqKm8NR7lF8SbYAnMRCnb9kwNUylIouINUNudJWX08jCa/4e6FNVXiBjriiwYmNRneTn4vN0MBlGuQN7y+PCOaGJYsKWmw/Oc+J+nIuOeG/ns/FaQ3leVGlEYkzpIFuGJTSB0s8XVaO6KtZ+V7slQKM94yzr7e6YdEqbV69+bV9twZ1g932BIlRJHqv3H2a3fO7rnxNVwQUM0T2cgxELX2hFntT2a26ETk0QrxLPYPhN2ODLmHGmYw34vbKCIp9T9AT/8Iz1WxrxwpVzJGBlxaTw8tc4dv73DwDD11eTW+vou8UJsn7vftndUWIFmxbGFZ4WnNymVkgaHxynRZWszBu2sBRr3uKM3MV6csQipL0oAq+NayhiJqfb10DyhkRUzYX+ysjiCxD5HsB540SeeVq9K9EEnp6BFqgjEbG5t1vLJmeMVSLyFm2iZ5vhOXMqjmOtM9J7D45DIDMOeACpIrJjQsTUnO05fR81R1Td82h7zLOtjxQSjI1LYX2vNvM5eXhwlwnEPeMWkGylAjRChkQMjVIQZiF7Q2PgIeFjp16aNp+n0VceokncwVIBG6DAaC098TGXutCtCZqVtYQA2tIr7SS4c1okQ/M56nu3EB1QFXRigulV8hRQaZZuJxZPyHx0L2BtMX172tqghh7HBKQ1q2QHI1427aCOu+IcChkDfT/R1c0FYp9n9MF5nAj7MzrI89nV43m72SVjvcKzZADgANowUwAA4JyY53BkbTrzvrugd1p/EZjXerc107hbxAK9vz42Q4EjQnOWtG75jt1thffUU4tKVGR64q37E1OW10yI8uhMOi474f00K8KvBrhWxBXorfzOE2m54kTO9qzEsv6qJOqrtqmXj626gRCL5qVxWT9V050Q1Oqx+V5EXy+nseEOobMPdWMsfwNj6JbhXe5XGXCqDXnpVfz3wvOrnaIIwkDjy+FX37m2X++eIYKSseQXc5ejL9Ux4xEAbqKx3HQ+MxMImFZYdBv7ned0rANTcVs8qZIkpMbiqzvbqIgD2VAb4jsdTwtiB9Ef21FSQP1p0lwQqnh5UfGDj59RxD+13HHugYUxRtf+sl0OJuNQqW0HZFm/sT+i285jVZCoK242H5Lp+gEa/kJ4k0XkHwLw9/td/1MAfy+AHwTwuwD8AgB/GMDfpaq3D94Hgr7tuFw9bGbz3OLeg5MwQlBiv5sgiSp2dwirLpBQCz1zgItMqFopPYMoqnC1qyUBQQqhKiznnUrKZPiCkoCs+SEJ/dP7WVmfMy7Q+upxZDrJFZptoOqt7FNj78uecPMDEgmm8EyC2Ub6f29fZpQgckaX9LjIg56lfkq+hx7mrBhn9926QDYjZ906CuNQW00h8VWFIPtu476onMsS4AmUbWpOVU80aWE4dwfQ3XM5b5C0Zat7x+HIJteatUVhi8HMN0YfheMWRAOWhGPUclHnRy2YhWObKt3dweJsLCFMFNBmsXLqCO4ICixrz7ZfDNWfN8h059kUyNY842Sgy0Dz+TB06sHWw2M4WV9oN+Q+54QOG7d9m77exRArzFuhOvD0NHJcKJgptAqy14DD7jyZGmmPt4PRCC36LgLse7PcaWnY3LZMNHwOK1VgKawWJ0lzGsPGjBhXnEv1ZfAAfAVhKCK/EMA/COCXquo7EflXAfxGAL8OwD+lqr9LRP4ZAL8ZwO/4kpuh9Y6+9aDf2ogQyUwdxon4UCArO93yKFjsbtDnwpCIxiXjM/WYtg6sv1fnQb3ukaM5QCTCMHy8DAHfbbS8F4Vfvj9BiLBrvln8swzzQJ7KmYuMmPy4CzNAGhluKpKknbZYrASl7zyh/fepRWA2oE00VGfC9ODjDJrt3rYoWhRj0xA1UV4QhJxdYrwCDMpcsdE0gYj/nHZLLbTxdVyrPS/+QtQkGcuYOdoVxSUCzQNMoN1ea93X3GkB9xTOdkB1NBWIDCg81zMWJhZUzHVe94CqQhuD1y3UZk7z62RTBa0ZF6WI5SAHgrPV4FR5w0pvdksXHCpwCwLGLH31LA8TKMORFhGW79FmczrGiTFZ01i8jSsKi7Gnk5NrTenpzthBBV3CrqF1BSI0qoUQVeXe8FjFqUFywrU3y5pmid/3XV9VTd4AvBGRA8BbAD8J4G8A8Hf43/8lAP8YvkQYinj1KkeBFIh0njxDVHfq5aoiarjQGQRMkLQsep4OPlBzcKE65ZbbeU6ny7KsE8+GqCImNrtGHBw3T/NkZkEKtjmHr/e0T92rpESGVY0OgV9U0wm3n4kTNcTfLBbMNiiJFxCblaQOLESWiAioB0J61rOQFengrQuKJtNDZwQqCBUokWHd/A1NgMvesHfB2ZshD22YmuUYAJZ7RVQ2TIp86x/HxH6zDWIRAIJt6049liouVa914RUBSNUMebgxddEEqqcRlosZLFTj/EixexVVW9WrIYbdtqC66A2fZ3MWiLzMvQYAyDjD1tWFUFZmPE/z6m6d8sUE8hJcrEYFZ4ERE10GZs9BUldHx+nkGAQLmwlTAYDWjHTBZBL23bOIPMLgOC3GkTWAEkGDIxWs1kbLxsFQDFXoTDJlskQpxBhvxGIUy+i5UKSn2c4W1pe34G4bY6sS2NCcdYp7/33XlwpDVf0zIvJPAvhTAN4B+H/D1OLvqCrLMv8EgF/40udF5IcB/DAAq1VS4wg3o+zPzZun9Hpa5mAsqMmOhjQ2cxFJsqEE9tNZFqkjCNUQgpZbOuN9TA3kgPtYsE+JbvzJXVpZAEgVC0QfzLBBLPTcttV2aCpIvse+WcyhOONZokLaZlpDqL0Rq8fvcLQKQ1FcrLVolaGNzLbh+MA3vTaF0C7kYxFCQnPxi6uOIoK9Cy67hU6MLjiH4JyrrZJ5xp1U9SmqUDcA26jIEJY5m38uVludrjL+WK/3bIjorq5vCm1D794bc1tRDou46zIuS28qKl+0AR6CdBD6WoBGyI0S7c/CuNSIpF1YN3cw+MFyeo59E8UU6weZra0vliEyzszooJGm9RbpisMR7bY3r7ro4VZ94jwVB6YjNs3BRPFUqwtFco0K0vEEF4RKKyRcGIqHl1WTAqx/vlxU4ONTMHwBVkADxvBiWO+Xhl9FTf75AH49gB8C8B0A/xqAX/Nln+Olqj8K4EcB4Pu//xdolO9ksR3VwrlWXSMp/thJv1+ctvCBC88/U5fqArMPxQ3VUQ/JF1K11rLIFEbBxJbFDvHBJXISL77OhY+y+B2ptR7OgqqSISbVhdadaaA22+chhYiQE8UXdVGT7c+6jpUL260TbfuiWTZyHjQo/eA97ZDw1SdsXD4nRltgJz0mnm4TYzTLLZ0KRXd+O3VWkzLLdxkvL10tzCU5Xpwz45UpB4r3Zclp9U2Uf9YiKFfhJ1KWjSrum5W/v7+9VSBWcozoh4tIajumQbwgze+eu7bF76PAHKcfFrrMD2C2QFX1TJWJw+NU72tdU4saXqahTedyaHC102IAIcA8LA7X5td2ClwIdSf8nYfZSFnQfrpt0ggbOqYOjHFaP0Sh00qETrX3sCCYoAdJimlBzlBDB1bnnrK+tJ7jov4+pvW+7/oqavLfCOBPqupP24TJ7wHw1wL4lohsjg5/EYA/82U3kiZ48/Bg8FokQwIAM/iWdCqfqfyR3101u1c/FK4OSm74uiACcyhDR868J4XgSPXGJI6NZFV/+X5rUwplAClI/YFBv9WyHEGGsyRyozrLJucm8p9d9SQxAdFPE09dmylE2E8bI9pgHC1uPYU5UsBq7vocbXajZZ/jfUWg1hlwHBA2qEdHgdxoravPvSHMxXtcnBT3V9rpkrKNaqoqnQzpndYyR0SeSs2DObjlTgiUTuGEjKZaBGHiuTlp4njW3GW8npdiRWgIxhNW+ngXgkW1nEvMDu27Q9L/V6jn72bbE0A4+nOnxhwT43jydhn34WX3dFFXwYfb4KRNK6nRBXLdIGoqsQKY7ijSyCYBLK22Yb9smENxO53p5nYGczkg2MTqnbPes4ha8fepePfO6iXvu9nkHy4WRnSexn0IKFpz58rFemq10QElZ6Qi9rFxPgp633/OwvBPAfiVIvIWpib/agB/CMAfBPAbYB7l3wTg937ZjQQSHISxhZS5mRr1RuonfDrjFX3PdyAXjIjZEyLNbKozaiQDTaV0t9M0vcpcmOms0QVFNFdBi8yw1lIOC2F9VY9X1deEuT9B+Hn7YS0l6chBaHs026rFWlr825RCRBojGzd1YV5RV46rPdPHA+xjOhFoLCe3H8fE05FNrSpxEqqgkzHGoaaqZRtssklOy9cWDWwiVAAAOIdJREFUoRutXc0M/srCI5lzUISJPgtkWX5bP6vl23MkWBqZ66Z8X0oEFDNAIEzJtmldOBriN/qefcynzmVc8t48HO1+gkyctJsLLJ5vA8C8+IkGHc1jLyWeafdbRyzaS0QuEocPi7yr82P3LuibmLOjGR1r371QWTd7sK1DM530vkWmlqpV7LMCVkCHZdY0R6Rbs6yTscgyibENfC18vZooNCD1BxSPr2Qz/DER+d0A/iMAJ4D/GKb2/j8B/C4R+T/4a//8l91LmqDv3VPHfPLBRcUIf1sYtHulyriiPL2bdkFGtwf3mR+rc5wuBK3ugtkKnXrJ1ZdznM+e0T2I2GwNSVPVvBxpgMKyiBlPBvGKcq3W60A4aETEDLtF/aX315LNZxjj01ljNVBay1oQFvOtzl03QKO892BRJcZkWp55+0L91bRzpc2RmRX2/ThPOH21o1ILkGZNGHuvdZ1ButEO5RpVSAmkFhFYxh1NBmnHXMJffPwyHKfE9vmMc83wIjqoNkR7nfNDQbgihWUTLRCM2T1UJbl+0vkU9lawT06yIMOAQFB4ec57kD9kdIL4AWUEp5I1o3TVOkTg1GmIfjDcx4beD0YBLl2t9rAKFBvOE1A1QTTOOh4C6RSauefo1NrccXietmb+/M/ecLuR+Ubx2Tcu+MY3L6Ymd/Eg8s0pvYytfAxTn7d9x75fPT3VDuRxmrrcm3my926mnWsXXLcGzIbRmIRhh+j09TXoPCoZOzPGbZSg+PdLw6/kTVbVfxTAP3r38n8N4K/5Kp9/dr84LcXIM0V833ABvx/KAv65cnqnOpAnQTgBpnmRAhFyMaouCx/159rW+9/L58JSJ2mlfNb2u+esPz9HMvEeNwfYYU1jcCp24qdhfZohbCpNRLW4ex7cxkrVbf18JXUlGgCAqR1zSjzb12A+l6duoBQNNMejK9VLjuyq8iI2NQJmV5TlPc/fY9xLo4i+ol8pDXMcZjoZRO/ev7633CX7y/ekfhN/zbGb3t+KCPM+Lz1PUILH4+9576WrHKLSuhDunoZGV4wNZxXSVMcb9t1iDS67MQuRToulG9innDtX1WceBPYmalV2mI3gLdFoG79MlR9R8MyEN9mA2Ies8gfxNnl8pfkJJkQF0/s6AxlanKla91wrTFvozwkZ/oW86ulGO1wudG6me5heW1/VGjOWpiBEnLhRUOm0+sDneeI8zxAMChOQdNjUz3CxKDUZzcXHiT/PE5NZNB7TR3r7ZDoGoBJCmHWNqaK3xlgs8xQChSrMhTdtjVELOlQZNY4274uqwt3MULXYsYqqp8UmWG8F/rwaKyge+oNAodtGDkmnSDoZWH1GH4xFGYHiKm0/FLhcdkM4jlLmNETexD2R4oHgTp01FU5SysDdgvwlN3E4k8Kjr8tyqSumOZok1T7r6vbeoE1D4PIZ92r6st54ZzGDvWkyE6S/Ma/vfa7yjHnXOVwQ5hqnyqyqbl7wLBKSsrowMcHQrKa3CyVxgUeauiiQ2BiMzWwYbyPE5xpeHxvobx/MLvewozXBu8cn3I4DxzEwjwHpTMOzWsdmUtnCdikNWeRNT9xuw8Js0NAg2KUBE+jOoDBPBabg6d0jxnyE6kQXm6frpcc8CgDxkgRDJh7PA+9uB949mWYXNHGulquTUly3ZgH+vdlYHgM6LHRneNjP+65XJ2pIRJVpU+/DgevpvL4rTkGXWnnirh68MScyhCSdEhmInM8gclqeJblJ4qyr6FBfQIO8rRA9UCVKBKV3z7ff53uFAL3ItnNWKrCi3Zc28hRmpkgOF0/quvGrp5uG/EQ5eZAZSauu0xGnRe18toebVoSxmPQIw9GdPLtVGcJnf6tjch9+peUnuVsz0V8eihri6NmV41db88LXneAMJP+eL7tv7gGolHXrGRp1DO/ZrQvCul92ibjtuk8+41Npz+ah13vD9eEBvTd3gghUD4yRQreuW5q34pmS42jrxMJhVATdgSrjVvPvlrHEtda7BGCAqgXOFyZ37uPBda+OfskbN5sJWyUm50Hy1a/Xp/0v0LraBWMjLgb1qtq4R/I+b7S8L0IVfOCO88Q4j8iOqPYtXrGPydbRLFjTZIBXvvOo1nOQqIEozIURN3TsDTfnqhrlkBhj3bq5EnWxTSkfM52oyUqkcJwmNG+nLc7jSFtfHccx7CQkR6ONtzMIb1oEPELocSM34RgQMVJNMpp7y3dt5r4XQQeFx0oZZqUIJrZtR992NFVsm1FV0SbGTJjgsqsbTn3c9B6xEdF5sr9iGb/YuAVhZRomb+Go2TWKlzzDqyBEzLetMwvIe+b0aAKZ6f1OJxrjVyeM9CHjBHVmnW+jSSuHVi3yxH5HnKKvdyNhQ2sjhBdc8EGAeXo4izJeFT7mHQ8Pb3C5XPAX/8D/FNeHKz7//Gdxuz3i29/+Dr7znZ/BGKfZ2YGYV4nSE75pXDDZ/pJYA41a3lQcjwPzVLx7Z8SrTgiGh4cNbz+7oDeLSQVM+9CpeHw6cc6JtnVIb7idto+H+nhJ2omVEeHtNKbsqaxuYHu4Ae3yfI7r9cpqci7UvHzA4vSs76+CME/P55jh/jkaExKCsCzceLLIXVusPdUTSKHkByYIEBeBHKhL49f8zrauba52JC58/iLxnpf6VgSun5hQHiz1GUSOGo4ThEDW5d5y98A0X7yAbmiJ8sLn9l5mVDyXJoEgmgm1QIb1UIt5l2h3CJlFnWfkARHY8/F5Pl4a5g6iwtLzkHgfMqzX8aSZJYrO84tTHEOZ82t95xpv8cb6Hn5nMyhHFqwbbbxDw6qATDSvcFeGNb5z7da5NkTW0fuO6/UNHh7ehEnn88/foXeyzLgZw5cR0/JC4GZTklVJjAuRWskcXlbWv+yDxv60bcZz6RYVuJPbM8am1252IOVzlamaarZDNUg4VU2ATkvZg5Zkjme8Buv1+nyGQGQNyN2sVdWG855FxxHvwXMFyAbI4wSP2wHWNOYqMJTBRB/EJuONeYLX8p291RVeGEoWFS2FBFoiEQr3bd88jpCbiKpSxszB+923WiAqUbTZmlbVxP0r8Z7nKmOOi4aALYbssnsDdTha5cFTx5+2z3uhwfdGUfD696rSAWZ/arYBeKKnoLWfj2M4UrYPamljPN8lhciaelYdLZlfrotQZomCGqZT+7HeJ/uiarm752n2t0y+4hvcrVsZncUIy6YfPsZMnSKkOUK0/O4MgobUtEvra0RKOEmxcoApMJWZOdVZQZshwIMwetzt9+N4gs6B/+6//3O4XC4h4KcCve84jonjcHvvOCECi/trwEY7uFpkhfk6rMW9mY2VjsBt69CmeAOj4Z8qUDTsO2D1WARDDeafw6j7ucY1sqLE7IG2MC2ERwTSdK2hPhy0umC2VEdDkh/AUB9HGKY3C7hvnd4twGd2Qx7uwk9ycWiwx0ym3tCggUxVsmfUB3IDFDop5CatG6v+jQwnxpdn77X8YWsTNyr5+pixEouSCIcLtWFJSQRMBaa3bc5R+gNvFxFzHhK1nUsffXP40+J9FBbk4NMIzVmFYcRm3s1lCOlAXhzPRJhU29qLY7y2N9IhRSBo2UZkGl7TVDVXm23Z8KVvKxhMkwDbGWskkkRW5BzrZtIDP6MS2723W5Fxe+WR/oxiK4UG4hkMqEeqsXFvb1sc3iCJ0ApFeaCZuuwt8TZUModQcbwV4zygc+Lzz38WT087dnd6qZpTQ9FwDoZvWXiMbgJ4WYbWAHFhaA1xswelsCNDeoovF2fI8SBwQ4PrfI4YZ+unJRjZGu52wsCSodRLAUjwBQy1eaLxUEVpEflSTeJVhaEIwl1eVdQ4h5VpS5LzvPTATaK6LgJ11HK7HZjDKObnHB5nxI1stPw9vJF0DOSzTDhViRl4CuUdgepILmCyTTBnLtB7lb9SblmKp6M2F+pzulNEKjqmDe65gKPXdVPTLVg0K2yqLpwEjiZCACUCT7OBHyaucuc6rvRcmbZFFMiSCdUOrCjoSDxvum9oamoLveDU2dpdv9jXl5Eup6SorFOdPUVcAyhLx8cpphNJbca1E38raldp3ottEEgIr6od5OfrF4rQXoEzApHXg5YxnFVwAnNwTdzZTqkmazCtBUhojQLVEGM+z1CZ9aN71sc7tPaE/bahNcHT4xOO44bjeMLT7RZ2eKipok0ndDaz6ZZ8eahY0frD+z+sbV1zHiyP3g86YfrgxFkO0tYE+2bZJMO1Ovh4GB2Zpeq1C/Ojs6TtUN+n/sDp++DeoXR/vToy7L2kVK2HG1h3lVcKrXpM+zcXJOpIcIyB47gV5pUZRJXqKXa9NWf2LV5SSGxybqSctnri18va1Vv30A0UG1a/s+3kKcz+KsqCVi4QLTFpXNxlsxYTAhGQiDgvHTwdKduCrlDtGLBA13g+1jZo2ZCqioHhm9ZeSzYfH3PYZ5lvfF+dz+5jD5nuLcxAarfjAIG250vCpghC8XZVUGyCKu13RCMpz/KUDRSMFFz1ymnIbJWqLt9ZBdhACHqM59oFjUNsIddVrqd1oFjJrgrd6s2PVyXvo88HG3M660xMp5OfNBfeogCrKPFm0gLBHY+PAIDNwcpxWqLCcR64PR0+PtMPUUNyJhTt/k0YqA1zCB2ucRR/ZWgCTZzT0hx9t9OKTjEGlYQj+2ZoUXS1e0OAvXvyw4Xqse/R09aDLwsg1mYixPddHyG0JhdPGrTVjZ/TKZ0yWp+zu6jMMZ/qSeV3wq9+lc2+7IM4ctf2VdQQXl5/IKPbqeJCqqoHNKS9LW2d9uzgv4uYrGVUkOzB/J2eTg0b5WIfDDNDMWIXdJaIpKKOIjSUxu0kqeCmHNNiHaFZ5Ck9tDmuoTYHuk1Bp449dDKrpGSpqM11CChuegqvEFp3aDhe8X7FHPiNKRz9PeuEUphbJyJUKYRuEoWk4w5hWkj5wxS3HNNYI0o7pngebl1ciwEIAGNbZ5QD4HiEOcFn6x4tS72L9ztMEjBvK5Gief79IAKXnrod06zogGWVqALYmxVQag17g3l7v7HbXA8rwr7tDVvk1sMdI7MMuWJKrZtsMipdX4pNNVTlnF+G39jvw8MjJwSqLNGgXlytGRIeJZXQn91KYHiMaTM6uXtvQ70+is2weomhRG62eE6vMrfvllRNHjkr88dFwwVrTpPTU+xChStqCmcoB2w9oYmQAnP5D+aFPf01X6B7d7IEKZuoCMQugLRUKexO/q6tLGa3Fz2DHRLPrmi1pmRF8XVNCjHb1+umrdkl6WQwAWClTXOMagqc5ZDmcT4KBxw9uaCQu7OlcrFOzo0Ll8FauS1DMQI7ex9as+BrCsSKyykQQMSYHy5CgmNEsRNvAme9OmHSnluHv+Zv81NrNki1v1Xb9xgzVTmhECMSq2cfDz0fd53JzuwZO8ZEvfk8ei0ah0w6JWyNeW8Tc1aXG+FVVQ/daZ7jy7UyJ5z9RbymCnB7IrMQsGnH9dpwuXZIU6hcMcbA7ckEz/VKDk973u3JnCzNDwI7DH0fu+oy6AzhMwawbRxHj10tXJaqTimmghmhbMPuqzBtYAJ62Fxvm68j2Iahii6NReY6Lpftg9T/HyXoOu0LWLMWkIusqgOLShd1HmyxJcHCusEg4rWIkVoB6sahMNPYBFXteb4n/PlzWrbJXb+oUnGvEuHRKBgntxhZRQ3zqGj5PtyCP4cKotygBc3x9YKomFETY1VuxoyNlxhVkivQ20E02iiESpiyj60JyYy1a/5eIodQ/YVorggddXOHIzZDfNPnsMXYhE2wjBHufs4X7wbw7qoCr67HOJzvkPIsJoHzNDMC86QBc29yTASML2yp3SDTDBR5+OtdWwLlVtHpbaMDIuaJ+ijnAGJFP3zd2VJ1jSMOiulo0UrQQoBeBL6C6rZ5dNtpnnALCRUPuPdDH37/iegj+8MDscqd2HL+2oQ/ozFXvmD6RQsRlxNY1jeHaQ4b24kMDl/OQl/fguno9/0L4/WRoWa3p+cxphRrblPUVcjdLV5VO9mGp5mFMPCvJg3qPH+2gUwdMWGUQcQAIMKoXW7oVNnquNFhIC5kY5M0QacxWhDxT6wr0mSzcAOhDcdrARf0yA25IDhvC6Xk4qTQiYgdX9L/AHjlPghT4KYz2uSp3GJz5MYKpF1yaTkPUEVHN3gRm0cp31Ld9b83MZWGlQFjQwfyVPP0hUC0tpmA9jozLWu29N4Kei17Iq5yikmmWL6kEDFXdbKedRWmvo6Gp29RRT6O0wkFhqUTNgmG9stlpYUyIZiop5pMGO96MkyJa59AoAhB7n2LVsjDYLSOOQfO48RxHmZ/E0OVHPeLj+txMmvIUuTGKTicV/A4bSy3jWjRBvU4LblgeIC/tIl9N4fLdtmsXQ54z9uJ4eo1D7x0sHn0xM61zbPUhPpxKJ5uJ7beLS8asKhwX4NVnk1XqVmNUmC2SlVPLAAwmyFC1roWdZ7UYbRfhwienk7MkWN8f30UZMhrSfSOl/OUJhJ6SRiqDzpPCBMGcRzkRhAgGITL4q+Khnm1ilBeWkJhVTevxim9INjye7z3pR3pr9ewkTo2gZLiDwiEx1Mf5Xe+14SIUhZGU+tmJDLjt/uDMtBt7Suej2/u4CoIrGPkQFyF4d343wlk4G74vc9sS4zVnRC09rroK+9737DXia+Okux/NR2Y0DqOA8dBRnTLK8/88hTcoBgOFboF0q6PXroZp1h14bBtd6FCoHaRtWQA5ihr0M7ROZU56DkHFLmLqQp3UyvwrBiTT0SrYd7juiFil/Kat5Fmj7q9l7ky4O+omluUnb1b/IGrbYTrWShlEa8ORnFmb062H/ofUBlenajh8IR/fyUWOu2AjFhXN6bebxom2odKQ6GgiWISKfg3Me/fksoXC5QDSGcJ4u9GvuAJ9opI22rOeKI6MYctr1MZbrISxBpibDByzViKixSKPvBEJOGEU/r3ljWObSySqTsn2u7cNyu6xe4bbdmTd8nDb/ruZAjO9cG6FMtcUXAacr4X9AJGBnCoTf0X5La2utikC5Ow0/H+y3oIgbmujZTJ5lFsKtCOO0LcdYGHSLk738zrafRQqhooL1VlG2NDgCe++93v4vHpEU9PB263I4Vaa+iBDK0G+PVyQe8Nbx6u2Pfd0xgbmkwM5FxNrnfXLMxJl2sZDfE6EdWyTj1ioKtvXXXnVCoRYNnS3h3dep1hbYpTMnQHeJ7UsG0bejPnhJkEFGgnpFnRK0g6A20MgHGY7ZP23NYF225tOD1dlBpT27xuy9aw72Z3hE60bjWcAbXayHNinia8mlDlTnHIMbx0VvgpJhTJuMPjGJh6gCQhH5CFHwcZRs1itiwO9hKOEAv8+am9oi/lzopAWF7P1KDYcHnfxceXR1uqJkVlTNymLwqIF9X6+MoFL/H5iviyyzy5s3uJBrHcOf9GjYthfGxnVdMY1iQ+FLwlu2mfk/UJ9wiiZIFU22+gBGQ8F+1mAI3kFUVWdLqc9XnD+5WrGnUx7q/nAvG5bbM+t445x4uoksLw6emGd+8ecbsZMkxU1tBpO1azeQmsQNW+7y6Ecv3kOK3B2OtAZz+qxmCHkiznuwJAEzRtxXTg6KkMY2BBsUMmnU3M6S3jE43we6jNi8IdEaoWA+uLx4K6TTDTucPDpzWaNmhH9C+4JgxPI+4IKjCIEz+rQmS4k4iOpRmmjxyyghAlf889zrxw2p+fDfWz69WR4e124L5JUv5LFdD+lvYtM1REJsScWUiG3mg3pPXuaoRsd+la7c6bRN62RHREj6QRn7FQioCbXinuPL1AT8e2pb1QpGHb/OT1dgkFtSo4yRp66zISGbaRJrpog8ComnqzeMan213oyxweyDpxTit0NcbwWsISedbBWedImgipjo2NKY3OJlyNKHNayddAoB5mQQ93FL4iTVkdb46D5BiAAvH5FUhcktafm41B5nF37rgYTZPwyrEVagnmZWQVPB4MUyeO88B//+1v4/HdI7793e/i8fEphQgRSai/zQz40nAMy+k9zoHLvuHhesHD9QIA6Ft3e60xtjSmJPbNMLQSHfsaUg2hoY74CYzuNzMdTawLAs06xOqkphRwvQOXi9lsLwQQPi/nebqz5DTChybYNuCcguMQzNPsfBBgd3TXXUhdLh297RinOSl4sNqXMeGcflr3ztAk9/c0RVPBtgmuD5utOZg54rgNnIOnlQbJ75zNGG9yywCYoWqL5+lb3Webn94UuKQ8eel6daIGOkaiTbIKw/peoJxXat7jLCVZnSwZmgPAi8HY3bj5gWKTiIekah5B13F6ayCo+yVY7UqAqSWzbMRUubQ4JyjVvQ+aQaZRhxaproXtS0pkVCAY+ETX9pSfeVjMNNRrS2RAhBoqb6C05ytlRhzcaksj4gn0XIVQoCHa9IjGAIYapZnjbko4hqU9oRzJ8686Jy9fklizLDoi0tAukOrp4+MjPv/iHd69e8Tj45OjvZ456XEPw/lWZgIYs0Ng3Im9NVz2PU0zDcBAaXuaDujtjP4vSNE+80wxQKzQZ/MW3uppzkN436URmd293+WNTKC1CUSGSMM8EPN1eoZrFwWmQDqMi7BZgLT4PLCtNsWOLieRetmbLWFGc95EVUOtc9K26UBIkSYmAoPSDceggS6bAColy00E3UN03ne9uprMhVnEg/3sGzSQuwsQdXjPXsVn5nSKqpxwnvwWV+R1U8tGpI0vXgACmTDcxULdNNSlqnrnQl5/j/Zqqn7i0B0nqZ5qv5JoQWDJ/LVvbP95em5ynN9pF02BbJ9M0lO+k4IOSRYhRpt0Rspj7rz8bNUnaNLPgyio6k1XM1TSqgpcbGIu8ESmEYDWoVdxQXI3H0gB/czMUS4erC/9XYqAKLhwXXR5pxCCX7x7h8d3j/jOd76LL774Au/ePUUGjs3Jhn3bIJFuKOZgaMasLMMOofOwtTTOE5fLjjdvHhyZeXrZMO2gu41VULNVVonXeofADiVDcloUFSlfGp8uxDjLfpoTOJzSKhMIzK7Yt24xhzKgmKHG9mmaiE7BOD3wedq87d1LmA7F6BlkHYcP3O49gXO4Oedi3uS+dWx7C32nNafucqDUBNg3O4DmkOAWsPhI0zqmphN2DGt33+i0Y72VclC21YR1f32UoGu78gTJqcTS+Nwn5TRAvjGMz34xR7K17gG1jOK3B9Htn55SfpDeJ8lnqE1k3UEV8eTvnnUydREmEp2iIEnUQ8RELVk0wZUghSFzQZXqHtEw0+DinskYHS2gaiGWQ00Ep/AcViXdUz1hA/O4rMyjq8Z05nvNMG/hSnVA61xOR1DVnpubmDnb9CCGyl/GUu8/F9k66+KmKhtTznsUdfAZFPX7nOfAu3ePePfuHT7//HN88fkXdhi55jDGxGV3+6AfzobYxOm83Hs7J4bHF053cl0frmhhikgh3xyq2WvtbmyRa1IEcGcaB5bzroYAUG1qnM+0PGiMwZiASoN0MjDZ/PZubbO83uneautTF3Mvzsl+ASIKbEbnJWNiNqtux3K04uMCZ5+xpAlmbrUsJwqCITPJcN02gQVSS7MiUGNmxT6xSouqpn6rk0ioWmZL39xNXQXLPUnuC9frO1CKegbYxOUkU35oSoZyggGMrNdin2OQq6G7JoKt94jnAyyLJZhyPdr0HgnxR5/GQiahccostFPeXjLVmC3IvIzsAuKUE5SS8n56S5CAUqzQDrYgz9agY1j7h9lkTDjxPUk5VtVKFVnEj6KkF8bJ4zxH6Ng6c7Z7MRekYLdc0hZUXq1bX0N9BUKwEr1EzKA4ui0qP9vQeRJwdPzHSpPGUBIiUBvaHIP6Ofv7ndAzaLX+Hu1Ieqp3X3yBd+8e7WBrDa3DiTskDoM5p9fy9XmezrcnJvjRN0A9nfFm4S6Xdzv61nG97LECBOn4aCIhuIysIw9c7gcefIycuAc4Jnfyb4rC/FKiL2wJTy/Mzq+Mae2wolLgPArQtoauit7N5ml2ZOA8jT1G3f43hmJ0RL/GZGwlUwMVx+HCbJvoZ+gdKbdcgOb6zvupH0oWkjPj3go14a6KtrUoyQBkLDMtFR+6Pg7tf3C+pZ1u8cAqym5JNZLojlkREBMqrKfBDW31O1qc6hMedR8ZBOLqhwmeshV9cSbxAZ8JYDHW5wJNR8S2bQhHvy/MECiKQDsmOAB47Qr477uHoqQQJl+cBfzaF00C7ugp/It5X0edM4dRFV6s255rL1pNCxYH4sGiVD/ihoWSXRR6mm2m91w+FPp2Xy5Ujfnme9KW6IdNmV2Ow719EC4QK/1XRiTk50hBRk9y1QpWAZmQIerOHAOff/45Hh+fIvyFam3UfJlGDdfRgyBjkM6e5oHWIGi4ncNqc6vN677v2DeabVLL4Zg0mhW8vfl/rqfq0KKQ8JPP55f7qBaGn1QFoGoUWHMqTs95d8Jx20tCpK7xz/aXV7TrEwOK83AhNB3MiIW/9KYYPefN1FoBmY3GhLNcA30z1nOSR3BUzOkF3ztczy40vU+GxM3UwygKc5qYIGxdAjSpKIbOJaX1fdcrC8NyclXIjyoMVzCrc2JQYAnpuCgM7Yw1DrgMSvXDPFVvdyjch4XUkId4emshCJktkXyG5X66qt3ThVc4RZYFbZMZdlE+i8nzkkIzVZ30iC1kshSoBWUBnHgjixBBJPt796yNpt3FAcC2pbeUworqc7VNMgDa84jX1pafK2HDaiuNz4gA8KwDagBAsGHbIdeKIJS1nQXR51iWNaM59rK+LT5WD2EKO2aZsC1Ek3EmqDunknAQ8JhQdtT6Ph2BCcaceHy6Yari4byUFL1EPCHIocEG7jKtoPkClUJrQkGCtT/2VtM6fUwDVZoEM+7MzLLxlCBE7GkcIB4L6LF+qla+c6pAXCCJ3wscMhVYfKq3S0iqy8JW9voZ/LhWIkA6D7E6r/Y5tcK6YVryaNWIw+SML041KWharO0ful6/BormiR4exUVYr5L7vKtZEhx/sakQhaw32VxAai5cwMNwzhCiohq5pRnAbd/bvkWRHKM9VwBeSc8XDlXFLEhvC1IgGH3LlKA7AReHgf/GjRDr2t/DE5HMyCR9VQ8t4A0oHBUWXB3o1NXKTbYFIcjgfXkDps9loXu7L4fWN4XOrNER6MFURxMU7Zkwqh50qnik/TLUJYDneNMsYU1KwRd9JGJvGV7jHVhGtzqzqHksjNb10An1zchBznFaVbjzDOE01YJ0pSEKD1kwdHolo6/eTkNuNle9CW7nwPmzX+DqoTZb786gzja7INQJoLntOXPPo34N62T7nOfKynYMn2emyJksF0CbCS4Xhk0E3dPwpDHZwD8zR+6vshYFVsi9AZDdHUFegL735jXGSTrRMAbDrqz9224IcN8NFaoKng4DDaLTIkAajVSuHbgjSF11n9pwTqcl8zC4zavijZF24fBg+29RjEo+rCi/emhNrEJ/4d6LZuu2hEOAHfJ3zaJmq3oxJm5SqmAmNCt7TR6scrdpUjivGxrlswXl3d2zCnJuQNsbvJkjmSoYizqkDl/iBIuyAHnf94WTxEOJoJBq14K43GJJZJdqTAq4kMrlmbWgUTKwJDrLcS9i6W682cQMXXpff+oBUgVhCufa91Q31/F4SQ169pKjDc5HxK4SQRMhBblunS+N74tDTblOc1HQ8TX84DS12Q7cHFtnSCeib2tjc01pzOv9wTPdSZHj7ugeJBRxAe1oM73NDt3CVJPRAzxj4mdfQCxJapEAJtSaxDsRecX8vIjbSDm/Bb3lYzzW1+nGjPytiAkN++A63+oOJiJ3jlWtMilJ6PthYPixbIY2dERWAolOcVMDdmqwzneUCPRwFzv5jOG5X7YllIYhMaHOKnn3JOvoMl/XQ1U2IjpwgY0Y1LALuUA4z+JZvZugpG1fNzRXoGh6BaswUlHIAJoqtPnpVnZxqLyt26Yrqij8dG5uNwEFlFYkTkFm4RrNxyBRDfw95izaesfUhnFuZqucDGVJmimi6zRue+pV8dzw1E5bp71IogO7Vwpe61M2qQpL/v7SwZCCuArEdePUy+zGAIZ6Xe2RaOw8oz60MgzJ59CyE62vrTVs2wbAyFAtGN8PpdYstFBhbM+3E3/+Z7/AZe/Yv++bbpd29OSeZBFLq1NNu7hy7T4LI9IkNj6ZN831yKB3E9TDtaI5G6Y7ILpOLzVhvWt1mBXBEs80UiNUsTHfL7lP4OiVNUuY8RYOTl9z5HSWlgWlqBUAZl66PZIRiIecCf/zVIzTntd48Hq6a3BFK9cg97tJyC6Cbd99z9zlnN5dH9Vm+Ax1EUWF7i9lZ7xAv+NvebYxFuRWUCFyS6i6sA2BhLIZ8zn3P92jwdykL3zi7hS/v1+qkinYKOQl2l3QG9VHRWyaaMVLp57k64m3Cvom8qpCwxvdWgOcjIB2tYralv5o2t7uT2+GU7Cr8ZdF2NX3r2NXUdmXX3dpbM9AtK59FqLWjJ9UP/RqplOkdjWJA4BrTJCoXNeHJfhy5G9cnXxeD42hCv40rVDNe/la1yiRVm1FWStl9sO9pFpKkq5jJSLupBDEFikN8WbHGA5nYaWvxh5JezUfoLEeadcXvzGbOb1NWavHu0WGI+Va9UYR2HLEFEngUhY7CaxWg9vz6yPYDJNXjAGt9BQadbgtuMtuwabDXfIpJOF8ahan1VuzgNEiyOhFRiyogqNdQDJ1jwWYWrd7QRRjDrfTtVAv60azyXTyhLJhmas6J9uqEeKTKpfX6Js0EcjC2GEISTEL7Tvvb3Y1Bu7WqaW67Seme/mY2NI6HTvpiQYE0joY8hJCXpJqi1RV4d1mVgDSO5/CQ3GeNwCCbbsgU/GMW9HGtLkBvXqQ62GISCUjorD6zX5IuInEUlm9nk3wLOY46d3v939b8JUqjuP0qnwWvnQcJ87jCPvZ5XJF33ZcLhdcr1fMOXE7Tqg27PtOyeCb1e+vRNKwcCtVvHu84egdb98emApc9j0IBUyuGJo0BnGimOUIx/0lzVI/55Qob0H6OApbaippP50ASD5i7yH+ZTSFSENTYAwxtDfpHLFBFAE2N51qUIoBt9PW/EbJomQ4ZxqlpflNH2vxVWIRH9aO4YSt3W3T5zlxnDM1Dl+rE4pxs9C204Xg6bRlFqvrB03xd33oel1hGEgtPcKWD+wCxPNb4XDYRrqqzfU/idM6bF7gXNF24+8NaM13OEoAUjD4W+0tWa0rIUztSNoBJdqZAtsfEEKGJx8N00rUAEcA/st0SUpuuZpXvdj3nl1M1wp8nW1WtvcFjFHG8v70j7fInQpbxrd6mklvZeEyCFRCVKich6WmzfrA6t01dEK71HsEW73FXduJAu+1iZd+JzKMg3QmMpy+Bnrv2LYNl8sF53nidjtiHoWIJm7qx7DSgWPbfZymRJ+npevtlBgUpHcazt1E3WkpXEMk0+VCWz9fQ7dMMBrguFeyEmVxngpiD3RYD5uC3rSsCSddiAJUnqqI2Ks8OzzWcmmCxHqCAuJ5zJEoody3Gutqunu58ihGW2Nt5Bh8SCB+lBoo3DhjjEKuAACJtk63MZ1jxKKrql0EaiiCCj9P/3Wm2wujwMOcDoCpCj15itnqtFACOmIQtkpdFhiC7LV+RRrVpJ1xxM+xkCT7AEdYFiqUC9HuUQgOfN2T4p2xZdPHqwd9Pm15ubEtZQkYkQXRov90v7TWMcbAu3eW3VPDTmwu2PoGj3NdMlPq0BNBEsV1IkNhONQL68ODy3MsW1nUFIAM+ygeyLtF/rLjLGUG549I8HacOEeN57RAd6jicr3i5/2878Pbt2/wzW9+E4+PjzEn55H1kyXaNq1t6nbiZpqEtoYB4It3jziOE/u+WT3hu8OF4xhoHWmH5VK+l2Xmoe+YTaCHRw9Q8HiIVOuc5wZB1hSB5v15X9r8zgGc50qvR9QZMa5bx64AZEIw0HvDfvExd/sluniGmDtd3NvORAJVtap66oWt4PZMMadKE+BEHri9CXQ6RRiAY+TBxSgRkjfMafdq24eh4et6k/07F9JKtLAiMDoZqiCgOsG3caHQmLpws2m5nSQHm03kGvoCwEMimM9Ywhtwj4T8A8X2xHsFgpUsgck+RvZCnIwA2tpWnvZh75zT0WXWEqYqFujUrUsZGO4LKRQ2ifvlCTtCZYrsHeHmIzob0VcFPaIF0XIsdP2dJoXlPUiiV8ZM2t8k5zXu64eCvuxdRUHSaus9EfCdRFyFIcD0wioQeTDPQSS42gsVwLbteHjzBm/fvsU3vvEZRMx8cJ5WltYOJzIi0R5X0uuiXXaAPd1OzGHPrdP/0oHCK4iM33eJWGxm+HtCLViEYou1aotvjFSXKesYyGwHXZpQKkLn3dXNVW4JgE5F3wSs6U7HaGVpU/W6LBAjcIh4XSL1fD48q6k1AKdrktI8TMZrIznYUPVcbtfIuGeIintFuy9cr+5AGY6QKHBIy8NNCW7aeRqKqe9FqoA8qe7DROxvcBWFdjcijLKLKFiSIyzQChGVBYW+jC7sDFq/gmJKWsSFsWiS38HvYz9bkHHaLXniUkUxtdn59SJP2p/lsIziiUiRqgODvld1zz7NU5M0YyIlnS/U2xQcNp5uwzMdyvJQt25C0tkvSFN1HocHESe6pWqU9s80sFscGE+Iip69DdTTaljGh1ZZqEWCcIyk8SDGkQeExRlmhg/HDH6o7fuOh+sVl8sF+7bh4XrFz/vmN/B0u+E73/mZ0EyCnBeuHreiOgKesZO28sfbEXnoW++AeExoMaVII7muBUO/t/cC0NudYUcVbaqxV7vGs7kGllUojZbuONWRma8fBXQWjzJSnMQs+SG3ORLu3TJbaHsWMfue+oeZHRWa1dY8h3nGjHN+eP8OIy7uI0EGCFCgGKf3k1EbDJ1yp1eo+h+4Xj3OkOE0izAsgo4LOcI0kJvnGWrjSUpEGP/ZZTIn4++oYsabpUxsID9Xmz3YOu7lEwu4sLk/9EGhksJdqSIxB7lsZto6WzmtAndRffZwlvOcrqLSMC7YcFfZDavaPSfRctkQvkkybKPH4ypS5pIPlBAoOwOfmasMKPzMiJN9nGc6gJhXTAE/ky6/rgveYPUec1PoevjEfwVl310v2QufP5PIcJQAes8DVzPgNzHb3uVywWW/uN1Q8dlnn6H1jm9/52dgVGkNVnrThUibHhHA7B0znagqbrcb5DRuz6134z3cJMY8tFHOqUjU/1QQEJTDn7PTBJjpRbZDZNWWYkjcqUKKO8brzmGZIRp3zjUV57pvphD8njkkGwlDLJSmNRibNQwUqCoY/cO1BTHqLpyrnq7KY9D3SAf6MAfMnMU8FKw1NAUoVGr8cjMHXLn3+65XdqCUYjhawmc8N5Zu0OotrarDTB0xJjuQHxC246hD21qomjwlKSR6N8nIIkl2gk8Mh+nV/kfbSIa3ZApWVV+qzXIN/Hz5StSaCw6AJaojUWWoDDwppbQP8NN29biqmwXGpDqWaKl7IruprXAa+ErDlQI6+QztShRfBK/mYRKTE/ep+DBFWt6D6HNG+2ccHvl7I/2SAJkpI/EYIRZ+73hr2QspTMIsoDRl5GHBMgWNeeqSbd/3HecYTl6xonTel2sN4geAF3eCC6CnJ0vy7a1h37fs03Ioaax1VaDHCV6PAC1fru1Iz30BoKuTFgxLFZxMv7uzFfYm0K5RRqI6ifgupmaOYc6wxrTR1tLO16lFufh2s4xSS2GfppOojER5acLweRCJMqT7Rv5HAy4HhJmEy1edaQKOOUtHX7henenaVISyacR8ShsAdY47e98ItUAg0DFjkQBFIEA8NqksooIUNdQAH/hqgAVi4U+viDY9+JbpeL0neUJuQidwWAQAJ57IlsnxBZWVRR5iR1AcCW7rKIXb4W0M5OXjZPU1Mo1NR4ZVmAARaBdXsQsDeOlHJzrtvE+FuV7jZWoEHlM49N68hkoS49LORKlNnJNTptG/cDDZ7oClkDFgm7pAjt10D/vU6fWi8zDI+UBQaWUfskt2+Ja16MJlqAaj0WLHnorWO7Z9dwq0ZMyR1nB9eMBUC9YfzZ1jflKJiAdg0+HUgG71RVQ4tgNfvHvE05Ng3zc8vLn4xyUFr+YYxNdi99LaoZCRDJkS2UMYmFNhuG1tYs7TDqA5fL5MtWRZ0KFZ69jmhLnpaU9u7vRkQca2kSwlm0VCjU5bnr3o+w6m9bjdlUNoxKxEezBasGYB1vu1YUwnhZgmtCtFKSm+4pwUhCms1Ll/8Xp1B0pdcOL2J9FmdvDWjNcssjRcbDijcFWjTFNKVbMKGG7K+Ke+mVwlIpOuAMtCoxOFp0uiHy3fuQFjl0VB+Hqv+tlEilpU80R1KzJMpMLOxMYXG5/qiAiBEOMBbx8KksrVaW3j2AdojpFj36t6rVVl9afx4BhFcAs074l7FTbRopa+xY+ah4sJhYoy2ZZ7BJrqMDealnbW94bavGgJOS/1i+834b8t42xdTHMID0yu62YuW8CFSJM06XBOmzRMcaEyDSAcxxnoimMHpDB5eRvn5JnJaYMAuFyu3v6+aDjq9t1xnni6PWGOEyfgURI3QNWyUgTumKKZJQ9sWxurpjXdo9/iQMz2cz7BsKs7LgDeM+bW55DrrOJflDXbPEurbwIZwHlf/gHeZkm2KX1pCMv1+mrycUS6XEDnbYNij5Odkwb1IkYqoTIyJoyIsbmBuZ6oHNTmAnW4+juGnUJNxGLDil3pOI7Va41CF6WeZMUi2kQiGImGxBPVnSYrA7A9wNxtoL0lMrN71IUDfx5A7d6ABMNxuRCIrJI2S2RGn1orJyOco9APhXjQHUqqoTvVJqNzFAFhGGJCcB7D0ryUcWGa7S0bUOqCVrWDzZFX5AcrSozYepiYUDRkat7dFqFNYXd2tFQPrenfQ8WXVSAGEhzD7IS0G3qIkrSGfbcgaws8R6w/8yZ37PuG6/UC1Ymnpyccx4HLpaFL9/6YE5AMGX3QM9rRpOF4OjDHicd3T7hsGy6XHdeHC1iwiYKwjn899nPdCKR3XC4X9L7hG9/8pqU79h1NerHZ2pi/e/eIb3/7uziOG7742e9inAfO2wGdE/um2JoJvE2dWYYHVjO2Gga+j2nojKh5QNF83XZZUzSPYiwRSkxwfvy9JQURsHRFOmfiw3DhtltQeOvejtHRxgxHj4XqSZA5EHmuwnW9Pg7TdYjoVJ1UFaIpyFYNoCKzilBWlMB717vHTwV1xV0WxFdv8dLvHxpGopDVwcNH6NKXL73V8/djRXcvhQbV9bJ4XKmev+9UXNRIwTr4+uLHEs3Vtq6ezhRmiagcir+nIc9RH/Dl414/+6XveeG9Wpt0d4t7dfz+qoL4Qw8tWBXM7w28TdRYYjWf3eKF17NJCb8Tzbp5p2+JDovPqvfTVGnpYJbQ/QAs6F5c4ArMdihADe5edqTWttnrgdh98Wce05dfUhtS7knhHlEmL05BrsmqNb33WV9lEf2FukTkpwF8DuDPvdpDf27X/wTfO20Fvrfa+73UVuB7q73fS20FXre9/zNV/Yte+sOrCkMAEJE/pKq/4lUf+j/w+l5qK/C91d7vpbYC31vt/V5qK/D1ae+XlQX4dH26Pl2frv+/uD4Jw0/Xp+vT9enCxxGGP/oRnvk/9PpeaivwvdXe76W2At9b7f1eaivwNWnvq9sMP12frk/Xp+vreH1Skz9dn65P16cLn4Thp+vT9en6dAF4RWEoIr9GRH5cRP6EiPzW13ruV71E5C8VkT8oIn9MRP4zEfkt/vr3i8i/IyL/pX//+R+7rbxEpIvIfywiv99//yER+TEf439FRC4fu428RORbIvK7ReQ/F5E/LiK/6us6tiLyD/ka+KMi8i+LyMPXaWxF5P8uIj8lIn+0vPbiWIpd/1dv9x8RkV/+NWjr/8nXwR8RkX9DRL5V/vYj3tYfF5G/6TXb+irCUCzU/Z8G8GsB/FIAf7uI/NLXePb/D9cJ4H+nqr8UwK8E8L/xNv5WAH9AVX8JgD/gv39drt8C4I+X3/+PAP4pVf3LAHwbwG/+KK16+frtAP4tVf0rAPxVsHZ/7cZWRH4hgH8QwK9Q1V8Gq1f+G/H1Gtt/EcCvuXvtfWP5awH8Ev/6YQC/45XayOtfxPO2/jsAfpmq/i8A/BcAfgQAfL/9RgB/pX/m/+ay43WuZ6wY/yN8AfhVAP7t8vuPAPiR13j2z6HNvxfA/xrAjwP4QX/tBwH8+Mdum7flF8EW/d8A4PfDkqT+HIDtpTH/yG39PgB/Eu6wK69/7cYWwC8E8KcBfD8sXfX3A/ibvm5jC+AXA/ijXzaWAP5ZAH/7S+/7WG29+9vfCuB3+s+LXADwbwP4Va/VztdSk7nAeP2Ev/a1vETkFwP4qwH8GIAfUNWf9D/9WQA/8LHadXf9XwD874Eoo/YLAHxHVVmU4+s0xj8E4KcB/Auu1v9zIvIZvoZjq6p/BsA/CeBPAfhJAN8F8Ifx9R1bXu8by6/73vv7APy//OeP2tZPDpS7S0S+AeBfB/C/VdWfqX9TO64+eiySiPzNAH5KVf/wx27LV7w2AL8cwO9Q1b8alp++qMRfo7H9+QB+PUyA/yUAPsNzNe9rfX1dxvLLLhH5bTDz1O/82G0BXk8Y/hkAf2n5/Rf5a1+rS0R2mCD8nar6e/zl/1ZEftD//oMAfupjta9cfy2Av0VE/hsAvwumKv92AN8SETIRfZ3G+CcA/ISq/pj//rthwvHrOLZ/I4A/qao/raoHgN8DG++v69jyet9Yfi33noj8PQD+ZgB/pwtv4CO39bWE4X8I4Je4R+4CM5L+vld69le6xDiB/nkAf1xV/8/lT78PwG/yn38TzJb4US9V/RFV/UWq+othY/nvqerfCeAPAvgN/ravRVsBQFX/LIA/LSJ/ub/0qwH8MXwNxxamHv9KEXnra4Jt/VqObbneN5a/D8Df7V7lXwngu0Wd/iiXiPwamInnb1HVL8qffh+A3ygiVxH5IZjT5z94tYa9ohH118E8R/8VgN/2mgbcr9i+/xVMtfgjAP4T//p1MFvcHwDwXwL4dwF8/8du6127/3oAv99//p/DFs+fAPCvAbh+7PaVdv4vAfwhH99/E8DP/7qOLYB/HMB/DuCPAvh/ALh+ncYWwL8Ms2ceMNT9m983ljDH2j/t++4/hXnJP3Zb/wTMNsh99s+U9/82b+uPA/i1r9nWT+l4n65P16fr04VPDpRP16fr0/XpAvBJGH66Pl2frk8XgE/C8NP16fp0fboAfBKGn65P16fr0wXgkzD8dH26Pl2fLgCfhOGn69P16fp0AfgkDD9dn65P16cLAPD/BTHFvuq194oTAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "number = 3\n",
    "plt.imshow(all_test_ds[number][0].permute(1, 2, 0))\n",
    "plt.title('%i' % all_test_ds[number][1])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "edafda55",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mmfa\u001b[0m (use `wandb login --relogin` to force relogin)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import wandb\n",
    "wandb.login()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "64017f95",
   "metadata": {},
   "source": [
    "### Model (categorical)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "d4eb4908",
   "metadata": {},
   "outputs": [],
   "source": [
    "from torchmetrics import Accuracy"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d04ade5f",
   "metadata": {},
   "source": [
    "Some background about CNNs:\n",
    "- https://stanford.edu/~shervine/teaching/cs-230/cheatsheet-convolutional-neural-networks\n",
    "- https://deepai.org/machine-learning-glossary-and-terms/convolutional-neural-network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "4a3295e7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([3, 100, 133])"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_ds[0][0].size()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "eec23c55",
   "metadata": {},
   "outputs": [],
   "source": [
    "class ForestHousesCategoricalModel(lightning.LightningModule):\n",
    "    def __init__(self, learning_rate=1e-3):\n",
    "        super().__init__()\n",
    "        self.save_hyperparameters()\n",
    "\n",
    "        self.learning_rate = learning_rate\n",
    "        self.batch_size = 16\n",
    "\n",
    "        channels, width, height = train_ds[0][0].size()\n",
    "\n",
    "        self.model = torch.nn.Sequential(\n",
    "            torch.nn.Conv2d(in_channels=3, out_channels=20, kernel_size=3, stride=1, padding=0),\n",
    "            torch.nn.ReLU(),\n",
    "            torch.nn.MaxPool2d(4),\n",
    "            torch.nn.Dropout(0.5),\n",
    "            torch.nn.Flatten(),\n",
    "            #  500x + channels=20, maxpool=2 -> 1653360\n",
    "            #  100x + channels=20, maxpool=2 -> 63700\n",
    "            #  100x + channels=50, maxpool=2 -> 159250\n",
    "            #  100x + channels=50, maxpool=4 -> 38400\n",
    "            #  100x + channels=20, maxpool=4 -> 15360\n",
    "            torch.nn.Linear(15360, 2),\n",
    "        )\n",
    "        self.loss = torch.nn.CrossEntropyLoss()\n",
    "\n",
    "        acc = Accuracy()\n",
    "        self.train_acc = acc.clone()\n",
    "        self.valid_acc = acc.clone()\n",
    "\n",
    "    def forward(self, x):\n",
    "        return self.model(x)\n",
    "\n",
    "    def step(self, batch, batch_idx, name):\n",
    "        x, y = batch\n",
    "        y_hat = self.forward(x)\n",
    "        loss = self.loss(y_hat, y)\n",
    "        self.log(f\"{name}/loss\", loss)\n",
    "\n",
    "        logits = self(x)\n",
    "        preds = torch.argmax(logits, dim=1)\n",
    "        \n",
    "        if name == \"train\":\n",
    "            self.train_acc(preds, y)\n",
    "            self.log(f\"{name}/acc\", self.train_acc)\n",
    "        else:\n",
    "            self.valid_acc(preds, y)\n",
    "            self.log(f\"{name}/acc\", self.valid_acc)\n",
    "\n",
    "        return {\"loss\": loss}\n",
    "    \n",
    "    def training_step(self, batch, batch_nb):\n",
    "        return self.step(batch, batch_nb, name=\"train\")\n",
    "\n",
    "    def validation_step(self, batch, batch_idx):\n",
    "        return self.step(batch, batch_idx, name=\"val\")\n",
    "    \n",
    "    def test_step(self, batch, batch_idx):\n",
    "        return self.step(batch, batch_idx, name=\"test\")\n",
    "\n",
    "    def configure_optimizers(self):\n",
    "        return torch.optim.Adam(self.parameters(), lr=self.learning_rate)\n",
    "\n",
    "    def train_dataloader(self):\n",
    "        return torch.utils.data.DataLoader(train_ds, batch_size=self.batch_size)\n",
    "\n",
    "    def val_dataloader(self):\n",
    "        return torch.utils.data.DataLoader(validation_ds, batch_size=self.batch_size)\n",
    "\n",
    "    def test_dataloader(self):\n",
    "        return torch.utils.data.DataLoader(all_test_ds, batch_size=self.batch_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "85f07047",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ForestHousesCategoricalModel(\n",
       "  (model): Sequential(\n",
       "    (0): Conv2d(3, 20, kernel_size=(3, 3), stride=(1, 1))\n",
       "    (1): ReLU()\n",
       "    (2): MaxPool2d(kernel_size=4, stride=4, padding=0, dilation=1, ceil_mode=False)\n",
       "    (3): Dropout(p=0.5, inplace=False)\n",
       "    (4): Flatten(start_dim=1, end_dim=-1)\n",
       "    (5): Linear(in_features=15360, out_features=2, bias=True)\n",
       "  )\n",
       "  (loss): CrossEntropyLoss()\n",
       "  (train_acc): Accuracy()\n",
       "  (valid_acc): Accuracy()\n",
       ")"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_categorical = ForestHousesCategoricalModel()\n",
    "model_categorical"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "d11e913e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "                    Syncing run <strong><a href=\"https://wandb.ai/mfa/workshop-p2--FvsH--cat/runs/7sothvxb\" target=\"_blank\">prime-rain-6</a></strong> to <a href=\"https://wandb.ai/mfa/workshop-p2--FvsH--cat\" target=\"_blank\">Weights & Biases</a> (<a href=\"https://docs.wandb.com/integrations/jupyter.html\" target=\"_blank\">docs</a>).<br/>\n",
       "\n",
       "                "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[34m\u001b[1mwandb\u001b[0m: logging graph, to disable use `wandb.watch(log_graph=False)`\n",
      "GPU available: True, used: False\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "/home/mfa/.virtualenvs/machine-learning-workshop/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py:1584: UserWarning: GPU available but not used. Set the gpus flag in your trainer `Trainer(gpus=1)` or script `--gpus=1`.\n",
      "  rank_zero_warn(\n",
      "\n",
      "  | Name      | Type             | Params\n",
      "-----------------------------------------------\n",
      "0 | model     | Sequential       | 31.3 K\n",
      "1 | loss      | CrossEntropyLoss | 0     \n",
      "2 | train_acc | Accuracy         | 0     \n",
      "3 | valid_acc | Accuracy         | 0     \n",
      "-----------------------------------------------\n",
      "31.3 K    Trainable params\n",
      "0         Non-trainable params\n",
      "31.3 K    Total params\n",
      "0.125     Total estimated model params size (MB)\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation sanity check: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/mfa/.virtualenvs/machine-learning-workshop/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:132: UserWarning: The dataloader, val_dataloader 0, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 8 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.\n",
      "  rank_zero_warn(\n",
      "Global seed set to 42\n",
      "/home/mfa/.virtualenvs/machine-learning-workshop/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:132: UserWarning: The dataloader, train_dataloader, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 8 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.\n",
      "  rank_zero_warn(\n",
      "/home/mfa/.virtualenvs/machine-learning-workshop/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:432: UserWarning: The number of training samples (12) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n",
      "  rank_zero_warn(\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "c3a070543e30485dba14f0e3efebc3de",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Training: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "wandb_logger = lightning.loggers.WandbLogger(project=\"workshop-p2--FvsH--cat\")\n",
    "wandb_logger.watch(model_categorical, log=\"all\")\n",
    "trainer_categorical = lightning.Trainer(gpus=0, max_epochs=10, logger=wandb_logger)\n",
    "trainer_categorical.fit(model_categorical)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "b252c684",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Restoring states from the checkpoint path at /home/mfa/machine-learning-workshop/part2/workshop-p2--FvsH--cat/7sothvxb/checkpoints/epoch=9-step=119.ckpt\n",
      "Loaded model weights from checkpoint at /home/mfa/machine-learning-workshop/part2/workshop-p2--FvsH--cat/7sothvxb/checkpoints/epoch=9-step=119.ckpt\n",
      "/home/mfa/.virtualenvs/machine-learning-workshop/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:132: UserWarning: The dataloader, test_dataloader 0, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 8 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.\n",
      "  rank_zero_warn(\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8993a912ffa14d41a3daea932679c0e6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Testing: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--------------------------------------------------------------------------------\n",
      "DATALOADER:0 TEST RESULTS\n",
      "{'test/acc': 0.7870370149612427, 'test/loss': 0.5997459292411804}\n",
      "--------------------------------------------------------------------------------\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[{'test/loss': 0.5997459292411804, 'test/acc': 0.7870370149612427}]"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trainer_categorical.test(ckpt_path='best')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "f9a7169c",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<br/>Waiting for W&B process to finish, PID 623770... <strong style=\"color:green\">(success).</strong>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(Label(value=' 0.00MB of 0.00MB uploaded (0.00MB deduped)\\r'), FloatProgress(value=1.0, max=1.0)…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<style>\n",
       "    table.wandb td:nth-child(1) { padding: 0 10px; text-align: right }\n",
       "    .wandb-row { display: flex; flex-direction: row; flex-wrap: wrap; width: 100% }\n",
       "    .wandb-col { display: flex; flex-direction: column; flex-basis: 100%; flex: 1; padding: 10px; }\n",
       "    </style>\n",
       "<div class=\"wandb-row\"><div class=\"wandb-col\">\n",
       "<h3>Run history:</h3><br/><table class=\"wandb\"><tr><td>epoch</td><td>▁▂▂▃▄▄▅▅▆▇▇▇█</td></tr><tr><td>test/acc</td><td>▁</td></tr><tr><td>test/loss</td><td>▁</td></tr><tr><td>train/acc</td><td>▁█</td></tr><tr><td>train/loss</td><td>█▁</td></tr><tr><td>trainer/global_step</td><td>▁▂▃▃▃▄▅▆▆▇▇██</td></tr><tr><td>val/acc</td><td>▁▇████████</td></tr><tr><td>val/loss</td><td>█▅▃▂▂▁▁▁▁▁</td></tr></table><br/></div><div class=\"wandb-col\">\n",
       "<h3>Run summary:</h3><br/><table class=\"wandb\"><tr><td>epoch</td><td>10</td></tr><tr><td>test/acc</td><td>0.78704</td></tr><tr><td>test/loss</td><td>0.59975</td></tr><tr><td>train/acc</td><td>1.0</td></tr><tr><td>train/loss</td><td>0.03891</td></tr><tr><td>trainer/global_step</td><td>120</td></tr><tr><td>val/acc</td><td>1.0</td></tr><tr><td>val/loss</td><td>0.03131</td></tr></table>\n",
       "</div></div>\n",
       "Synced 5 W&B file(s), 1 media file(s), 0 artifact file(s) and 0 other file(s)\n",
       "<br/>Synced <strong style=\"color:#cdcd00\">prime-rain-6</strong>: <a href=\"https://wandb.ai/mfa/workshop-p2--FvsH--cat/runs/7sothvxb\" target=\"_blank\">https://wandb.ai/mfa/workshop-p2--FvsH--cat/runs/7sothvxb</a><br/>\n",
       "Find logs at: <code>./wandb/run-20220227_180300-7sothvxb/logs</code><br/>\n"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "wandb.finish()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "c274d7f2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# alternative to reshape without knowing the dimensions\n",
    "img = test_ds[0][0][None, :]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "474486eb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([1, 3, 100, 133])"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "img.size()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "32b283b5",
   "metadata": {},
   "outputs": [],
   "source": [
    "import termcolor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "feb5cccb",
   "metadata": {},
   "outputs": [],
   "source": [
    "def test_result(idx, image, model):\n",
    "    img = image[0][None, :]\n",
    "    r = model(img)\n",
    "    is_correct = bool(image[1] == torch.argmax(r))\n",
    "    is_correct = termcolor.colored(is_correct, \"green\" if is_correct else \"red\")\n",
    "    return \" \".join([str(i) for i in [idx, image[1], test_ds.classes[image[1]], is_correct, torch.argmax(r), r]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "389e9b43",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.2663, -0.3796]], grad_fn=<AddmmBackward0>)\n",
      "1 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.8222, -1.9089]], grad_fn=<AddmmBackward0>)\n",
      "2 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 2.1652, -2.1790]], grad_fn=<AddmmBackward0>)\n",
      "3 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.8663, -1.8476]], grad_fn=<AddmmBackward0>)\n",
      "4 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 2.3496, -2.5498]], grad_fn=<AddmmBackward0>)\n",
      "5 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.3001, -1.2972]], grad_fn=<AddmmBackward0>)\n",
      "6 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 2.8249, -2.9902]], grad_fn=<AddmmBackward0>)\n",
      "7 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.7455, -1.9884]], grad_fn=<AddmmBackward0>)\n",
      "8 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 2.0554, -2.0904]], grad_fn=<AddmmBackward0>)\n",
      "9 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.9758, -2.1721]], grad_fn=<AddmmBackward0>)\n",
      "10 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.1121,  1.3902]], grad_fn=<AddmmBackward0>)\n",
      "11 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.6234,  1.5733]], grad_fn=<AddmmBackward0>)\n",
      "12 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-3.6904,  3.7501]], grad_fn=<AddmmBackward0>)\n",
      "13 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.4420,  2.4142]], grad_fn=<AddmmBackward0>)\n",
      "14 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-0.9640,  0.9489]], grad_fn=<AddmmBackward0>)\n",
      "15 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.8978,  1.7828]], grad_fn=<AddmmBackward0>)\n",
      "16 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.3503,  2.5697]], grad_fn=<AddmmBackward0>)\n",
      "17 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-3.7533,  3.7792]], grad_fn=<AddmmBackward0>)\n",
      "18 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.1455,  1.1023]], grad_fn=<AddmmBackward0>)\n",
      "19 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.7593,  2.8759]], grad_fn=<AddmmBackward0>)\n"
     ]
    }
   ],
   "source": [
    "for idx, item in enumerate(test_ds):\n",
    "    print(test_result(idx, item, trainer_categorical.model))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "49abcd09",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.2319, -1.2127]], grad_fn=<AddmmBackward0>)\n",
      "1 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.7231, -1.8192]], grad_fn=<AddmmBackward0>)\n",
      "2 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-1.0754,  1.0351]], grad_fn=<AddmmBackward0>)\n",
      "3 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.4455,  0.2473]], grad_fn=<AddmmBackward0>)\n",
      "4 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.8709, -1.8220]], grad_fn=<AddmmBackward0>)\n",
      "5 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 2.5683, -2.4219]], grad_fn=<AddmmBackward0>)\n",
      "6 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.8763, -0.9510]], grad_fn=<AddmmBackward0>)\n",
      "7 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.9395, -0.9447]], grad_fn=<AddmmBackward0>)\n",
      "8 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.9938,  0.9235]], grad_fn=<AddmmBackward0>)\n",
      "9 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.2479, -0.2415]], grad_fn=<AddmmBackward0>)\n",
      "10 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.3621, -0.3231]], grad_fn=<AddmmBackward0>)\n",
      "11 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.7188, -1.6429]], grad_fn=<AddmmBackward0>)\n",
      "12 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.6542, -0.7275]], grad_fn=<AddmmBackward0>)\n",
      "13 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.5492, -0.5776]], grad_fn=<AddmmBackward0>)\n",
      "14 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.2044, -0.3863]], grad_fn=<AddmmBackward0>)\n",
      "15 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.0198,  0.1568]], grad_fn=<AddmmBackward0>)\n",
      "16 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.8103, -1.7572]], grad_fn=<AddmmBackward0>)\n",
      "17 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.9672, -0.9509]], grad_fn=<AddmmBackward0>)\n",
      "18 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.8451, -1.7750]], grad_fn=<AddmmBackward0>)\n",
      "19 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-1.0344,  0.9713]], grad_fn=<AddmmBackward0>)\n",
      "20 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.7359,  0.7806]], grad_fn=<AddmmBackward0>)\n",
      "21 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.1594, -0.4181]], grad_fn=<AddmmBackward0>)\n",
      "22 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.8076, -0.7799]], grad_fn=<AddmmBackward0>)\n",
      "23 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.4382,  2.3358]], grad_fn=<AddmmBackward0>)\n",
      "24 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.1265,  1.9598]], grad_fn=<AddmmBackward0>)\n",
      "25 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-0.5887,  0.6569]], grad_fn=<AddmmBackward0>)\n",
      "26 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.1689,  2.2924]], grad_fn=<AddmmBackward0>)\n",
      "27 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.6103,  1.5488]], grad_fn=<AddmmBackward0>)\n",
      "28 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-0.5114,  0.7269]], grad_fn=<AddmmBackward0>)\n",
      "29 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-0.7165,  0.7681]], grad_fn=<AddmmBackward0>)\n",
      "30 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.8903,  2.8769]], grad_fn=<AddmmBackward0>)\n",
      "31 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.9984,  1.9811]], grad_fn=<AddmmBackward0>)\n",
      "32 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.6472,  2.6753]], grad_fn=<AddmmBackward0>)\n",
      "33 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.8618,  1.9145]], grad_fn=<AddmmBackward0>)\n",
      "34 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.7614,  1.6876]], grad_fn=<AddmmBackward0>)\n",
      "35 1 houses \u001b[31mFalse\u001b[0m tensor(0) tensor([[ 0.3362, -0.4820]], grad_fn=<AddmmBackward0>)\n",
      "36 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.1386,  1.2229]], grad_fn=<AddmmBackward0>)\n",
      "37 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.2627,  2.2110]], grad_fn=<AddmmBackward0>)\n",
      "38 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-0.8972,  0.8766]], grad_fn=<AddmmBackward0>)\n",
      "39 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.5791,  2.4686]], grad_fn=<AddmmBackward0>)\n",
      "40 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-3.5351,  3.3726]], grad_fn=<AddmmBackward0>)\n",
      "41 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-0.0372,  0.0318]], grad_fn=<AddmmBackward0>)\n",
      "42 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.2020,  1.1322]], grad_fn=<AddmmBackward0>)\n",
      "43 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.9614,  2.0385]], grad_fn=<AddmmBackward0>)\n",
      "44 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.3330,  1.3175]], grad_fn=<AddmmBackward0>)\n",
      "45 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.2684,  1.1167]], grad_fn=<AddmmBackward0>)\n"
     ]
    }
   ],
   "source": [
    "for idx, item in enumerate(test2_ds):\n",
    "    print(test_result(idx, item, trainer_categorical.model))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "ed2abf90",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.6518,  0.5386]], grad_fn=<AddmmBackward0>)\n",
      "1 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-1.1508,  1.0810]], grad_fn=<AddmmBackward0>)\n",
      "2 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-2.6155,  2.6391]], grad_fn=<AddmmBackward0>)\n",
      "3 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.8300,  0.5967]], grad_fn=<AddmmBackward0>)\n",
      "4 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.6251,  0.7260]], grad_fn=<AddmmBackward0>)\n",
      "5 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.9106,  0.8477]], grad_fn=<AddmmBackward0>)\n",
      "6 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.9305,  0.9121]], grad_fn=<AddmmBackward0>)\n",
      "7 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-1.1129,  0.8192]], grad_fn=<AddmmBackward0>)\n",
      "8 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-1.8555,  1.7542]], grad_fn=<AddmmBackward0>)\n",
      "9 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.3702, -0.3716]], grad_fn=<AddmmBackward0>)\n",
      "10 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-1.5137,  1.4858]], grad_fn=<AddmmBackward0>)\n",
      "11 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-2.0860,  2.0245]], grad_fn=<AddmmBackward0>)\n",
      "12 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-2.4765,  2.4381]], grad_fn=<AddmmBackward0>)\n",
      "13 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-1.7980,  1.6739]], grad_fn=<AddmmBackward0>)\n",
      "14 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[-0.0125, -0.0567]], grad_fn=<AddmmBackward0>)\n",
      "15 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.0674, -1.1798]], grad_fn=<AddmmBackward0>)\n",
      "16 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 1.6550, -1.7195]], grad_fn=<AddmmBackward0>)\n",
      "17 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.8816,  0.7382]], grad_fn=<AddmmBackward0>)\n",
      "18 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.4735,  0.6595]], grad_fn=<AddmmBackward0>)\n",
      "19 0 forest \u001b[31mFalse\u001b[0m tensor(1) tensor([[-0.8483,  0.9272]], grad_fn=<AddmmBackward0>)\n",
      "20 0 forest \u001b[32mTrue\u001b[0m tensor(0) tensor([[ 0.0746, -0.2587]], grad_fn=<AddmmBackward0>)\n",
      "21 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.0054,  2.0523]], grad_fn=<AddmmBackward0>)\n",
      "22 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.5663,  1.4584]], grad_fn=<AddmmBackward0>)\n",
      "23 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.8199,  2.0113]], grad_fn=<AddmmBackward0>)\n",
      "24 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-0.6935,  0.5970]], grad_fn=<AddmmBackward0>)\n",
      "25 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-0.8130,  0.9157]], grad_fn=<AddmmBackward0>)\n",
      "26 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.4681,  1.3279]], grad_fn=<AddmmBackward0>)\n",
      "27 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.5741,  2.4286]], grad_fn=<AddmmBackward0>)\n",
      "28 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.8249,  1.7309]], grad_fn=<AddmmBackward0>)\n",
      "29 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.4208,  1.4221]], grad_fn=<AddmmBackward0>)\n",
      "30 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.2483,  1.0872]], grad_fn=<AddmmBackward0>)\n",
      "31 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.6643,  1.4038]], grad_fn=<AddmmBackward0>)\n",
      "32 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.5304,  1.4990]], grad_fn=<AddmmBackward0>)\n",
      "33 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-3.1284,  2.9009]], grad_fn=<AddmmBackward0>)\n",
      "34 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-0.3797,  0.2793]], grad_fn=<AddmmBackward0>)\n",
      "35 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-0.3391,  0.1844]], grad_fn=<AddmmBackward0>)\n",
      "36 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-3.2703,  2.9346]], grad_fn=<AddmmBackward0>)\n",
      "37 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.2135,  2.2176]], grad_fn=<AddmmBackward0>)\n",
      "38 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.5700,  1.5912]], grad_fn=<AddmmBackward0>)\n",
      "39 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.8817,  1.8887]], grad_fn=<AddmmBackward0>)\n",
      "40 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-1.2511,  1.3170]], grad_fn=<AddmmBackward0>)\n",
      "41 1 houses \u001b[32mTrue\u001b[0m tensor(1) tensor([[-2.4702,  2.5375]], grad_fn=<AddmmBackward0>)\n"
     ]
    }
   ],
   "source": [
    "for idx, item in enumerate(test3_ds):\n",
    "    print(test_result(idx, item, trainer_categorical.model))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d94c38d",
   "metadata": {},
   "source": [
    "## 2 - forest-vs-houses -  binary\n",
    "\n",
    "the same model, but changed to a binary model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "edbf5f13",
   "metadata": {},
   "outputs": [],
   "source": [
    "class ForestHousesBinaryModel(lightning.LightningModule):\n",
    "    def __init__(self, learning_rate=1e-3):\n",
    "        super().__init__()\n",
    "        self.save_hyperparameters()\n",
    "\n",
    "        self.learning_rate = learning_rate\n",
    "        self.batch_size = 16\n",
    "\n",
    "        channels, width, height = train_ds[0][0].size()\n",
    "\n",
    "        self.model = torch.nn.Sequential(\n",
    "            torch.nn.Conv2d(in_channels=3, out_channels=20, kernel_size=3, stride=1, padding=0),\n",
    "            torch.nn.ReLU(),\n",
    "            torch.nn.MaxPool2d(4),\n",
    "            torch.nn.Dropout(0.5),\n",
    "            torch.nn.Flatten(),\n",
    "            #  500x + channels=20, maxpool=2 -> 1653360\n",
    "            #  100x + channels=20, maxpool=2 -> 63700\n",
    "            #  100x + channels=50, maxpool=2 -> 159250\n",
    "            #  100x + channels=50, maxpool=4 -> 38400\n",
    "            #  100x + channels=20, maxpool=4 -> 15360\n",
    "            torch.nn.Linear(15360, 1),  # <-- 1 (because binary)\n",
    "            torch.nn.Sigmoid(), # sideeffect: bring output values in correct range (0..1)          \n",
    "        )\n",
    "\n",
    "        # different loss for binary\n",
    "        self.loss = torch.nn.BCELoss()\n",
    "\n",
    "        # alternative, works with values outside of range (0..1) but also results in values outside of range\n",
    "        # self.loss = torch.nn.BCEWithLogitsLoss()  \n",
    "        \n",
    "        acc = Accuracy()\n",
    "        self.train_acc = acc.clone()\n",
    "        self.valid_acc = acc.clone()\n",
    "\n",
    "    def forward(self, x):\n",
    "        return self.model(x)\n",
    "\n",
    "    def step(self, batch, batch_idx, name):\n",
    "        x, y = batch\n",
    "        y_hat = self.forward(x)\n",
    "        \n",
    "        # transform y\n",
    "        # [1, 2, 3] -> [[1], [2], [3]]\n",
    "        y.unsqueeze_(1)\n",
    "        \n",
    "        # cast dtype:int to dtype:float for y (target)\n",
    "        loss = self.loss(y_hat, y.float())\n",
    "        \n",
    "        self.log(f\"{name}/loss\", loss)\n",
    "\n",
    "        logits = self(x)\n",
    "        # no argmax needed -> logits are result\n",
    "        \n",
    "        # target stays here as dtype:int\n",
    "        if name == \"train\":\n",
    "            self.train_acc(logits, y)\n",
    "            self.log(f\"{name}/acc\", self.train_acc)\n",
    "        else:\n",
    "            self.valid_acc(logits, y)\n",
    "            self.log(f\"{name}/acc\", self.valid_acc)\n",
    "\n",
    "        return {\"loss\": loss}\n",
    "    \n",
    "    def training_step(self, batch, batch_nb):\n",
    "        return self.step(batch, batch_nb, name=\"train\")\n",
    "\n",
    "    def validation_step(self, batch, batch_idx):\n",
    "        return self.step(batch, batch_idx, name=\"val\")\n",
    "    \n",
    "    def test_step(self, batch, batch_idx):\n",
    "        return self.step(batch, batch_idx, name=\"test\")\n",
    "\n",
    "    def configure_optimizers(self):\n",
    "        return torch.optim.Adam(self.parameters(), lr=self.learning_rate)\n",
    "\n",
    "    def train_dataloader(self):\n",
    "        return torch.utils.data.DataLoader(train_ds, batch_size=self.batch_size)\n",
    "\n",
    "    def val_dataloader(self):\n",
    "        return torch.utils.data.DataLoader(validation_ds, batch_size=self.batch_size)\n",
    "\n",
    "    def test_dataloader(self):\n",
    "        return torch.utils.data.DataLoader(all_test_ds, batch_size=self.batch_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "027cdbe9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ForestHousesBinaryModel(\n",
       "  (model): Sequential(\n",
       "    (0): Conv2d(3, 20, kernel_size=(3, 3), stride=(1, 1))\n",
       "    (1): ReLU()\n",
       "    (2): MaxPool2d(kernel_size=4, stride=4, padding=0, dilation=1, ceil_mode=False)\n",
       "    (3): Dropout(p=0.5, inplace=False)\n",
       "    (4): Flatten(start_dim=1, end_dim=-1)\n",
       "    (5): Linear(in_features=15360, out_features=1, bias=True)\n",
       "    (6): Sigmoid()\n",
       "  )\n",
       "  (loss): BCELoss()\n",
       "  (train_acc): Accuracy()\n",
       "  (valid_acc): Accuracy()\n",
       ")"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_binary = ForestHousesBinaryModel()\n",
    "model_binary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "f3c1b375",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "                    Syncing run <strong><a href=\"https://wandb.ai/mfa/workshop-p2--FvsH--bin/runs/8zp4lao3\" target=\"_blank\">lunar-snowball-6</a></strong> to <a href=\"https://wandb.ai/mfa/workshop-p2--FvsH--bin\" target=\"_blank\">Weights & Biases</a> (<a href=\"https://docs.wandb.com/integrations/jupyter.html\" target=\"_blank\">docs</a>).<br/>\n",
       "\n",
       "                "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[34m\u001b[1mwandb\u001b[0m: logging graph, to disable use `wandb.watch(log_graph=False)`\n",
      "GPU available: True, used: False\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "/home/mfa/.virtualenvs/machine-learning-workshop/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py:1584: UserWarning: GPU available but not used. Set the gpus flag in your trainer `Trainer(gpus=1)` or script `--gpus=1`.\n",
      "  rank_zero_warn(\n",
      "\n",
      "  | Name      | Type       | Params\n",
      "-----------------------------------------\n",
      "0 | model     | Sequential | 15.9 K\n",
      "1 | loss      | BCELoss    | 0     \n",
      "2 | train_acc | Accuracy   | 0     \n",
      "3 | valid_acc | Accuracy   | 0     \n",
      "-----------------------------------------\n",
      "15.9 K    Trainable params\n",
      "0         Non-trainable params\n",
      "15.9 K    Total params\n",
      "0.064     Total estimated model params size (MB)\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation sanity check: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/mfa/.virtualenvs/machine-learning-workshop/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:132: UserWarning: The dataloader, val_dataloader 0, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 8 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.\n",
      "  rank_zero_warn(\n",
      "Global seed set to 42\n",
      "/home/mfa/.virtualenvs/machine-learning-workshop/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:132: UserWarning: The dataloader, train_dataloader, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 8 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.\n",
      "  rank_zero_warn(\n",
      "/home/mfa/.virtualenvs/machine-learning-workshop/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:432: UserWarning: The number of training samples (12) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n",
      "  rank_zero_warn(\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "05e3b14ba02e416a8516c8ff6ff3b46a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Training: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validating: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "wandb_logger = lightning.loggers.WandbLogger(project=\"workshop-p2--FvsH--bin\")\n",
    "wandb_logger.watch(model_binary, log=\"all\")\n",
    "trainer_binary = lightning.Trainer(gpus=0, max_epochs=10, logger=wandb_logger)\n",
    "trainer_binary.fit(model_binary)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "2805011d",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Restoring states from the checkpoint path at /home/mfa/machine-learning-workshop/part2/workshop-p2--FvsH--bin/8zp4lao3/checkpoints/epoch=9-step=119.ckpt\n",
      "Loaded model weights from checkpoint at /home/mfa/machine-learning-workshop/part2/workshop-p2--FvsH--bin/8zp4lao3/checkpoints/epoch=9-step=119.ckpt\n",
      "/home/mfa/.virtualenvs/machine-learning-workshop/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:132: UserWarning: The dataloader, test_dataloader 0, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 8 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.\n",
      "  rank_zero_warn(\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b65f56a02bee44568a940fbc29565830",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Testing: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--------------------------------------------------------------------------------\n",
      "DATALOADER:0 TEST RESULTS\n",
      "{'test/acc': 0.7777777910232544, 'test/loss': 0.5638598203659058}\n",
      "--------------------------------------------------------------------------------\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[{'test/loss': 0.5638598203659058, 'test/acc': 0.7777777910232544}]"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trainer_binary.test(ckpt_path='best')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "afb7cc22",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<br/>Waiting for W&B process to finish, PID 623855... <strong style=\"color:green\">(success).</strong>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(Label(value=' 0.00MB of 0.00MB uploaded (0.00MB deduped)\\r'), FloatProgress(value=1.0, max=1.0)…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<style>\n",
       "    table.wandb td:nth-child(1) { padding: 0 10px; text-align: right }\n",
       "    .wandb-row { display: flex; flex-direction: row; flex-wrap: wrap; width: 100% }\n",
       "    .wandb-col { display: flex; flex-direction: column; flex-basis: 100%; flex: 1; padding: 10px; }\n",
       "    </style>\n",
       "<div class=\"wandb-row\"><div class=\"wandb-col\">\n",
       "<h3>Run history:</h3><br/><table class=\"wandb\"><tr><td>epoch</td><td>▁▂▂▃▄▄▅▅▆▇▇▇█</td></tr><tr><td>test/acc</td><td>▁</td></tr><tr><td>test/loss</td><td>▁</td></tr><tr><td>train/acc</td><td>▁█</td></tr><tr><td>train/loss</td><td>█▁</td></tr><tr><td>trainer/global_step</td><td>▁▂▃▃▃▄▅▆▆▇▇██</td></tr><tr><td>val/acc</td><td>▁▃▄▄▆▇████</td></tr><tr><td>val/loss</td><td>█▆▄▃▃▂▂▁▁▁</td></tr></table><br/></div><div class=\"wandb-col\">\n",
       "<h3>Run summary:</h3><br/><table class=\"wandb\"><tr><td>epoch</td><td>10</td></tr><tr><td>test/acc</td><td>0.77778</td></tr><tr><td>test/loss</td><td>0.56386</td></tr><tr><td>train/acc</td><td>1.0</td></tr><tr><td>train/loss</td><td>0.07574</td></tr><tr><td>trainer/global_step</td><td>120</td></tr><tr><td>val/acc</td><td>1.0</td></tr><tr><td>val/loss</td><td>0.07903</td></tr></table>\n",
       "</div></div>\n",
       "Synced 5 W&B file(s), 1 media file(s), 0 artifact file(s) and 0 other file(s)\n",
       "<br/>Synced <strong style=\"color:#cdcd00\">lunar-snowball-6</strong>: <a href=\"https://wandb.ai/mfa/workshop-p2--FvsH--bin/runs/8zp4lao3\" target=\"_blank\">https://wandb.ai/mfa/workshop-p2--FvsH--bin/runs/8zp4lao3</a><br/>\n",
       "Find logs at: <code>./wandb/run-20220227_180849-8zp4lao3/logs</code><br/>\n"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "wandb.finish()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "1f421708",
   "metadata": {},
   "outputs": [],
   "source": [
    "def test_result(idx, image, model):\n",
    "    img = image[0][None, :]\n",
    "    r = model(img)\n",
    "    is_correct = (r[0] > 0.5) == (image[1] == 1)\n",
    "    is_correct = termcolor.colored(is_correct, \"green\" if is_correct else \"red\")\n",
    "    return \" \".join([str(i) for i in [idx, image[1], test_ds.classes[image[1]], is_correct, r]])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3c6a92eb",
   "metadata": {},
   "source": [
    "Accuracy: https://torchmetrics.readthedocs.io/en/stable/references/modules.html#id3\n",
    "default threshold=0.5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "a4c09777",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.7163]], grad_fn=<SigmoidBackward0>)\n",
      "1 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.1081]], grad_fn=<SigmoidBackward0>)\n",
      "2 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0369]], grad_fn=<SigmoidBackward0>)\n",
      "3 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0263]], grad_fn=<SigmoidBackward0>)\n",
      "4 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0828]], grad_fn=<SigmoidBackward0>)\n",
      "5 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.2214]], grad_fn=<SigmoidBackward0>)\n",
      "6 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0097]], grad_fn=<SigmoidBackward0>)\n",
      "7 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0339]], grad_fn=<SigmoidBackward0>)\n",
      "8 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0413]], grad_fn=<SigmoidBackward0>)\n",
      "9 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0652]], grad_fn=<SigmoidBackward0>)\n",
      "10 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.8933]], grad_fn=<SigmoidBackward0>)\n",
      "11 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.8701]], grad_fn=<SigmoidBackward0>)\n",
      "12 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9957]], grad_fn=<SigmoidBackward0>)\n",
      "13 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9925]], grad_fn=<SigmoidBackward0>)\n",
      "14 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9138]], grad_fn=<SigmoidBackward0>)\n",
      "15 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9826]], grad_fn=<SigmoidBackward0>)\n",
      "16 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9815]], grad_fn=<SigmoidBackward0>)\n",
      "17 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9988]], grad_fn=<SigmoidBackward0>)\n",
      "18 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9791]], grad_fn=<SigmoidBackward0>)\n",
      "19 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9780]], grad_fn=<SigmoidBackward0>)\n",
      "20 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.1195]], grad_fn=<SigmoidBackward0>)\n",
      "21 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0959]], grad_fn=<SigmoidBackward0>)\n",
      "22 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.7844]], grad_fn=<SigmoidBackward0>)\n",
      "23 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.7698]], grad_fn=<SigmoidBackward0>)\n",
      "24 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.1082]], grad_fn=<SigmoidBackward0>)\n",
      "25 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0056]], grad_fn=<SigmoidBackward0>)\n",
      "26 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.2212]], grad_fn=<SigmoidBackward0>)\n",
      "27 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.2195]], grad_fn=<SigmoidBackward0>)\n",
      "28 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.8134]], grad_fn=<SigmoidBackward0>)\n",
      "29 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.4822]], grad_fn=<SigmoidBackward0>)\n",
      "30 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.4788]], grad_fn=<SigmoidBackward0>)\n",
      "31 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0566]], grad_fn=<SigmoidBackward0>)\n",
      "32 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0997]], grad_fn=<SigmoidBackward0>)\n",
      "33 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.1710]], grad_fn=<SigmoidBackward0>)\n",
      "34 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.1538]], grad_fn=<SigmoidBackward0>)\n",
      "35 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.3417]], grad_fn=<SigmoidBackward0>)\n",
      "36 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.1002]], grad_fn=<SigmoidBackward0>)\n",
      "37 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.1167]], grad_fn=<SigmoidBackward0>)\n",
      "38 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0842]], grad_fn=<SigmoidBackward0>)\n",
      "39 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.7488]], grad_fn=<SigmoidBackward0>)\n",
      "40 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.3141]], grad_fn=<SigmoidBackward0>)\n",
      "41 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.2847]], grad_fn=<SigmoidBackward0>)\n",
      "42 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.0709]], grad_fn=<SigmoidBackward0>)\n",
      "43 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9891]], grad_fn=<SigmoidBackward0>)\n",
      "44 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9767]], grad_fn=<SigmoidBackward0>)\n",
      "45 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.7941]], grad_fn=<SigmoidBackward0>)\n",
      "46 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9689]], grad_fn=<SigmoidBackward0>)\n",
      "47 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9792]], grad_fn=<SigmoidBackward0>)\n",
      "48 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9861]], grad_fn=<SigmoidBackward0>)\n",
      "49 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9595]], grad_fn=<SigmoidBackward0>)\n",
      "50 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9988]], grad_fn=<SigmoidBackward0>)\n",
      "51 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9893]], grad_fn=<SigmoidBackward0>)\n",
      "52 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9983]], grad_fn=<SigmoidBackward0>)\n",
      "53 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9854]], grad_fn=<SigmoidBackward0>)\n",
      "54 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9815]], grad_fn=<SigmoidBackward0>)\n",
      "55 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.8023]], grad_fn=<SigmoidBackward0>)\n",
      "56 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9001]], grad_fn=<SigmoidBackward0>)\n",
      "57 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9903]], grad_fn=<SigmoidBackward0>)\n",
      "58 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9371]], grad_fn=<SigmoidBackward0>)\n",
      "59 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9811]], grad_fn=<SigmoidBackward0>)\n",
      "60 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9881]], grad_fn=<SigmoidBackward0>)\n",
      "61 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.6254]], grad_fn=<SigmoidBackward0>)\n",
      "62 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9472]], grad_fn=<SigmoidBackward0>)\n",
      "63 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9733]], grad_fn=<SigmoidBackward0>)\n",
      "64 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.6653]], grad_fn=<SigmoidBackward0>)\n",
      "65 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9038]], grad_fn=<SigmoidBackward0>)\n",
      "66 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.9347]], grad_fn=<SigmoidBackward0>)\n",
      "67 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.9429]], grad_fn=<SigmoidBackward0>)\n",
      "68 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.9606]], grad_fn=<SigmoidBackward0>)\n",
      "69 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.8757]], grad_fn=<SigmoidBackward0>)\n",
      "70 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.7221]], grad_fn=<SigmoidBackward0>)\n",
      "71 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.7553]], grad_fn=<SigmoidBackward0>)\n",
      "72 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.8529]], grad_fn=<SigmoidBackward0>)\n",
      "73 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.7902]], grad_fn=<SigmoidBackward0>)\n",
      "74 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.9739]], grad_fn=<SigmoidBackward0>)\n",
      "75 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.3769]], grad_fn=<SigmoidBackward0>)\n",
      "76 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.9456]], grad_fn=<SigmoidBackward0>)\n",
      "77 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.9794]], grad_fn=<SigmoidBackward0>)\n",
      "78 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.9871]], grad_fn=<SigmoidBackward0>)\n",
      "79 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.9712]], grad_fn=<SigmoidBackward0>)\n",
      "80 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.6518]], grad_fn=<SigmoidBackward0>)\n",
      "81 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.3858]], grad_fn=<SigmoidBackward0>)\n",
      "82 0 forest \u001b[32mtensor([True])\u001b[0m tensor([[0.4167]], grad_fn=<SigmoidBackward0>)\n",
      "83 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.7221]], grad_fn=<SigmoidBackward0>)\n",
      "84 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.7607]], grad_fn=<SigmoidBackward0>)\n",
      "85 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.8564]], grad_fn=<SigmoidBackward0>)\n",
      "86 0 forest \u001b[31mtensor([False])\u001b[0m tensor([[0.5878]], grad_fn=<SigmoidBackward0>)\n",
      "87 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9850]], grad_fn=<SigmoidBackward0>)\n",
      "88 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9761]], grad_fn=<SigmoidBackward0>)\n",
      "89 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9730]], grad_fn=<SigmoidBackward0>)\n",
      "90 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9094]], grad_fn=<SigmoidBackward0>)\n",
      "91 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9496]], grad_fn=<SigmoidBackward0>)\n",
      "92 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9340]], grad_fn=<SigmoidBackward0>)\n",
      "93 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9886]], grad_fn=<SigmoidBackward0>)\n",
      "94 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9731]], grad_fn=<SigmoidBackward0>)\n",
      "95 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9216]], grad_fn=<SigmoidBackward0>)\n",
      "96 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.8428]], grad_fn=<SigmoidBackward0>)\n",
      "97 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9880]], grad_fn=<SigmoidBackward0>)\n",
      "98 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9781]], grad_fn=<SigmoidBackward0>)\n",
      "99 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9986]], grad_fn=<SigmoidBackward0>)\n",
      "100 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9599]], grad_fn=<SigmoidBackward0>)\n",
      "101 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9315]], grad_fn=<SigmoidBackward0>)\n",
      "102 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9909]], grad_fn=<SigmoidBackward0>)\n",
      "103 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9940]], grad_fn=<SigmoidBackward0>)\n",
      "104 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9713]], grad_fn=<SigmoidBackward0>)\n",
      "105 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9891]], grad_fn=<SigmoidBackward0>)\n",
      "106 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9522]], grad_fn=<SigmoidBackward0>)\n",
      "107 1 houses \u001b[32mtensor([True])\u001b[0m tensor([[0.9910]], grad_fn=<SigmoidBackward0>)\n"
     ]
    }
   ],
   "source": [
    "for idx, item in enumerate(all_test_ds):\n",
    "    print(test_result(idx, item, trainer_binary.model))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "71d8038d",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}