metrics.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 169,
   "metadata": {},
   "outputs": [],
   "source": [
    "import gensim\n",
    "from gensim.models.phrases import Phraser\n",
    "\n",
    "import pymystem3\n",
    "import pandas as pd\n",
    "import scipy.spatial as spatial\n",
    "from sklearn.model_selection import train_test_split\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "oc = pd.read_csv('tmp/opencorpora.csv.gz', compression='gzip')\n",
    "criminal_code = pd.read_csv('tmp/vectors/criminal_code.csv.gz', compression='gzip')\n",
    "criminal_court_orders = pd.read_csv('tmp/vectors/criminal_court_orders.csv.gz', compression='gzip')\n",
    "civil_code = pd.read_csv('tmp/vectors/civil_code.csv.gz', compression='gzip')\n",
    "civil_court_orders = pd.read_csv('tmp/vectors/civil_court_orders.csv.gz', compression='gzip')\n",
    "\n",
    "oc['source'] = 'opencorpora'\n",
    "criminal_code['source'] = 'criminal_code'\n",
    "criminal_court_orders['source'] = 'criminal_court_orders'\n",
    "civil_code['source'] = 'civil_code'\n",
    "civil_court_orders['source'] = 'civil_court_orders'\n",
    "\n",
    "criminal_code['name'] = criminal_code.apply(lambda x: \"Ст.{} {}\".format(x['article_number'], x['article_name']), axis=1)\n",
    "civil_code['name'] = civil_code.apply(lambda x: \"Ст.{} {}\".format(x['article_number'], x['article_name']), axis=1)\n",
    "\n",
    "civil_court_orders['name'] = [\"Гражд. дело {}\".format(i) for i in range(civil_court_orders.shape[0])]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "tmp = pd.concat([\n",
    "    oc.reset_index()[['index', 'title', 'lemmas', 'source']].rename(columns={'title':'name'}), \n",
    "    criminal_code.reset_index()[['index', 'name', 'lemmas', 'source']],\n",
    "    criminal_court_orders.reset_index()[['index', 'title', 'lemmas',  'source']].rename(columns={'title':'name'}),\n",
    "    civil_code.reset_index()[['index', 'name', 'lemmas', 'source']],\n",
    "    civil_court_orders.reset_index()[['index', 'name', 'lemmas', 'source']],\n",
    "]).reset_index(drop=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 54.2 s, sys: 2.56 s, total: 56.7 s\n",
      "Wall time: 56.8 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "phraser = gensim.models.phrases.Phraser(gensim.models.phrases.Phrases(sentences=tmp['lemmas'].str.split()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "criminal_court_orders_train, criminal_court_orders_test = train_test_split(criminal_court_orders, test_size=0.33, random_state=42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "tmp = pd.concat([\n",
    "    oc.reset_index()[['index', 'title', 'lemmas', 'source']].rename(columns={'title':'name'}), \n",
    "    criminal_code.reset_index()[['index', 'name', 'lemmas', 'source']],\n",
    "    criminal_court_orders_train.reset_index()[['index', 'title', 'lemmas',  'source']].rename(columns={'title':'name'}),\n",
    "    civil_code.reset_index()[['index', 'name', 'lemmas', 'source']],\n",
    "    civil_court_orders.reset_index()[['index', 'name', 'lemmas', 'source']],\n",
    "]).reset_index(drop=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "gensim.utils.save_as_line_sentence(tmp['lemmas'].map(lambda x: phraser[x.split()]), 'tmp/metrics/phrased.txt')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = gensim.models.doc2vec.Doc2Vec(\n",
    "    documents=gensim.models.doc2vec.TaggedLineDocument('tmp/metrics/phrased.txt'), \n",
    "    epochs=500, vector_size=300, workers=4\n",
    ")\n",
    "\n",
    "tmp['vectors'] = model.docvecs.vectors_docs.tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save('tmp/metrics/all_texts.model')\n",
    "tmp.to_csv('tmp/metrics/all.csv.gz', index=False, compression='gzip')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Labels for train and test dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "tmp['label'] = tmp['name'].map(lambda x: x.split()[0].replace(\"Ст.\", \"\"))\n",
    "criminal_court_orders_test['labels'] = criminal_court_orders_test['labels'].str.split()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Guess embeddings for test samples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [],
   "source": [
    "def embed_text(lemmas):\n",
    "    return model.infer_vector(phraser[lemmas.split()])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tqdm import tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 12%|█▏        | 695/5594 [06:42<33:34,  2.43it/s]"
     ]
    }
   ],
   "source": [
    "vectors = []\n",
    "for vtext in tqdm(criminal_court_orders_test['lemmas'].tolist()):\n",
    "    vectors.append(embed_text(vtext))\n",
    "criminal_court_orders_test['vectors'] = vectors"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Creating KDTree from train vectors"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "804"
      ]
     },
     "execution_count": 110,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(criminal_court_orders_test.iloc[:804].index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {},
   "outputs": [],
   "source": [
    "test = criminal_court_orders_test.iloc[:804]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 184,
   "metadata": {},
   "outputs": [],
   "source": [
    "def find_within_radius(embedding, radius):\n",
    "    neighbours = point_tree.query_ball_point(embedding, radius)\n",
    "    return neighbours if neighbours else None"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 185,
   "metadata": {},
   "outputs": [],
   "source": [
    "code = tmp[tmp['source'] == 'criminal_code'].reset_index(drop=True).copy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 186,
   "metadata": {},
   "outputs": [],
   "source": [
    "point_tree = spatial.cKDTree(code['vectors'].tolist())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 190,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "radius: 0\tacc: 0.00000\tn_size: 0.00000\n",
      "radius: 5\tacc: 0.00000\tn_size: 0.00000\n",
      "radius: 10\tacc: 0.00000\tn_size: 0.00000\n",
      "radius: 15\tacc: 0.00000\tn_size: 0.00000\n",
      "radius: 20\tacc: 0.00000\tn_size: 0.00000\n",
      "radius: 25\tacc: 0.00000\tn_size: 0.00001\n",
      "radius: 30\tacc: 0.01990\tn_size: 0.00521\n",
      "radius: 35\tacc: 0.13184\tn_size: 0.06264\n",
      "radius: 40\tacc: 0.37065\tn_size: 0.24145\n",
      "radius: 45\tacc: 0.65299\tn_size: 0.52406\n",
      "radius: 50\tacc: 0.87065\tn_size: 0.77765\n",
      "radius: 55\tacc: 0.96642\tn_size: 0.92025\n",
      "radius: 60\tacc: 0.98756\tn_size: 0.97217\n",
      "radius: 65\tacc: 0.99378\tn_size: 0.99002\n",
      "radius: 70\tacc: 0.99627\tn_size: 0.99522\n",
      "radius: 75\tacc: 0.99751\tn_size: 0.99719\n",
      "radius: 80\tacc: 0.99876\tn_size: 0.99848\n",
      "radius: 85\tacc: 0.99876\tn_size: 0.99874\n",
      "radius: 90\tacc: 0.99876\tn_size: 0.99876\n",
      "radius: 95\tacc: 1.00000\tn_size: 0.99975\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA20AAAG5CAYAAAAdwtzaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4wLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvqOYd8AAAIABJREFUeJzs3Xl8VOXd///XlUlCVsISdkKCCoRdCGEVCq64gYgKuOO+oL27aNtbW229+/21aq0LVIpLkSICFlt3RbQRBQIBlDVAWEMwbGFLSCbLzPX7Y4YQEEjInCST8H4+HvPIzDlnPuczC4Q31znXMdZaREREREREJDiF1HUDIiIiIiIicnoKbSIiIiIiIkFMoU1ERERERCSIKbSJiIiIiIgEMYU2ERERERGRIKbQJiIiIiIiEsQU2kREzlHGmOnGmP9zqNadxphvz7A+zRhzj//+LcaY+U7sV2qPMWadMWZ4XfchInIuUmgTEalHjDHbjTFFxpiCCrfJdd3X2bDWvm2tvbyu+xCfqoZ3a213a21aLbQkIiInCa3rBkRE5Kxda61dUNdNyImMMQYw1lpvXffiJGNMqLW2rK77EBE5l2mkTUSkgfAforjIGPNXY8whY8xWY8xg//Kdxpi9xpg7TnpavDHmC2NMvjHma2NMYoV6yf51B4wxG40xN1VY19wY84Ex5ogxZhlw/km9XGaM2WCMOewfCTQn9flthcfWGPOAMSbL3/cUfwDCGOMyxvzFGLPfGLPNGDPJv/0p/9PRGPNrY8wW/+tZb4wZc9L6e40xmRXW9/UvTzDGvGeM2WeMyTs2emmMedoYM7PC85Mq7t9/2OcfjTGLgELgPGPMxAr72GqMuf+kHkYbY773v3dbjDEjjTE3GmNWnLTdz40x75/mdaYZY/7PGLPYP9r6of8zedtfN8MYk1TZZ2mMuQ+4BXj8WB3/8u3GmF8ZY1YDR40xof5ll1b4XP63wnu9whiTcKpeRUQkcAptIiINywBgNdAcmAXMBlKBC4BbgcnGmJgK298CPAPEA98DbwMYY6KBL/w1WgLjgb8ZY7r5nzcFcANtgLv8N/zPjQfeA570190CDKmk72v8ffYCbgKu8C+/F7gSuBDoC1xXSZ0twFAgDvg9MNMY08bf143A08DtQGNgFJBnjHEBHwE7gCSgHb73rapuA+4DYv019vpfT2NgIvDXCuGwPzADeAxoAgwDtgMfAB2NMV1PqjvjDPsd79+mHb7QvAT4B9AMyASe8u/ztJ+ltXYavs/8WWttjLX22gr1JwBXA01OMdL2c//6q/yv8y58oVVERGqAQpuISP3zH/+I1LHbvRXWbbPW/sNa6wHmAAnAH6y1xdba+UAJvgB3zMfW2oXW2mLgCWCQf8TkGmC7v1aZtfY7YB5woz/kjAV+Z609aq1dC7xVoeZVwDpr7b+staXAi8DuSl7Tn6y1h6y12cB/8YU08AW4l6y1Odbag8CfzlTEWvuutfYHa63XWjsHyAL6+1ffgy+cZFifzdbaHf71bYHH/K/Hba097aQqpzDdWrvO/z6VWms/ttZu8e/ja2A+viAJcDfwprX2C3+Pu6y1G/zv/xx8wRpjTHd8AfKjM+z3H/79HAY+BbZYaxf4A9a7QB//dqf9LCt5XS9ba3daa4tOse4e4Elr7Ub/61xlrc2rpJ6IiFSTQpuISP1znbW2SYXbaxXW7alwvwjAWnvysoojbTuP3bHWFgAH8AWYRGBAxXCIb1SuNdAC3znROyvU2VHhftuT6tqTtj2ViqGusEKPJ9SqrI4x5nb/oYfHeu6Bb7QPfAF2yymelgDsCOC8rRN6MsZcaYxJ9x+KeAhfiK2sB/AF35v9h4beBsz1h7nTOflzPd3nfKbPssqv6yRneh0iIuIwTUQiInJuKz8PyX/YZDPgB3z/YP/aWnvZyU/wj7SV+Z+7wb+4Q4VNck+qayo+Pku5QPtT9XuKvhKB14BLgCXWWo8x5nuOn0+3k5POvauwvIM59YQbR4GoCo9PFXRshR4a4RvFuh1431pbaoz5TxV6wFqbbowpwTcqd7P/5oTTfpYn91/F5cdqng+sDaQxERGpGo20iYic264yxlxkjAnHd25burV2J77D8jobY24zxoT5b6nGmK7+Qy/fA542xkT5z3OrOMHJx0B3Y8z1/gk7HqXyUZ3TmQv81BjTzhjTBPjVGbaNxhc09gEYYybiG2k75nXgl8aYFONzgT/oLcMXDv9kjIk2xkQYY46dg/c9MMwY08EYEwf8ppJ+w4FG/h7KjDFXAhUvb/AGMNEYc4kxJsT/upIrrJ8BTAZKz/IQzTM57WfpX78HOO8sa74OPGOM6eR/L3sZY5o71K+IiJxEoU1EpP750Jx4nbZ/B1BrFr4JKw4AKfjPqbLW5uMLG+PxjbztBv6ML5AATMJ3+N1uYDq+CTDwP3c/vvOl/gTkAZ2ARdXs7zV854StBr4DPsE3yuc5eUNr7XrgL/gm5NgD9Ky4X2vtu8Af/a85H/gP0MwfQq/Fd65fNpADjPM/5wt855qtBlZw5nPMjr1vj+ILmwfxjZZ9UGH9MvyTkwCHga/xHb54zD/xBc2ZOKQKn+UbQDf/oZP/qWLZF/C9xvnAEX+NSKd6FhGRExnfqQYiIiLBzz9yNdVam1jpxvWQMSYS3+yTfa21WXXdj4iIBAeNtImISNAyxkQaY67yXyesHb5RwUBGFoPdg0CGApuIiFSkkTYREQlaxpgofIcQJuObEfFj4KfW2iN12lgNMMZsxzdhyXX+aflFREQAhTYREREREZGgpsMjRUREREREglidXactPj7eJiUl1dXuT+vo0aNER0fXdRtyjtL3T+qavoNSl/T9k7qk75/UhRUrVuy31raobLs6C21JSUksX768rnZ/WmlpaQwfPryu25BzlL5/Utf0HZS6pO+f1CV9/6QuGGN2VGU7HR4pIiIiIiISxBTaREREREREgphCm4iIiIiISBCrs3PaTqW0tJScnBzcbned9RAXF0dmZmad7V9qR0REBO3btycsLKyuWxEREREROaOgCm05OTnExsaSlJSEMaZOesjPzyc2NrZO9i21w1pLXl4eOTk5dOzYsa7bERERERE5o6A6PNLtdtO8efM6C2xybjDG0Lx58zod0RURERERqaqgCm2AApvUCn3PRERERKS+CLrQJiIiIiIiIscptImIiIiIiAQxhbYGxFqL1+ut6zZERERERMRBCm0nmTBhAikpKXTv3p1p06YBMHXqVB577LHybaZPn86kSZMAeOaZZ+jSpQsXXXQREyZM4Pnnn/9RzQ8//JABAwbQp08fLr30Uvbs2QNAQUEBEydOpGfPnvTq1Yt58+YB8Nlnn9G3b1969+7NJZdcAsDTTz99Qu0ePXqwfft2tm/fTpcuXbj99tvp0aMHO3fu5MEHH6Rfv350796dp556qvw5GRkZDB48mN69e9O/f3/y8/MZNmwY33//ffk2F110EatWrXLq7RQRERERkQAF1ZT/Ff3+w3Ws/+GIozW7tW3MU9d2P+M2U6ZMITExkaKiIlJTUxk7dixjx45l0KBBPPfccwDMmTOHJ554goyMDObNm8eqVasoLS2lb9++pKSk/KjmRRddRHp6OsYYXn/9dZ599ln+8pe/8MwzzxAXF8eaNWsAOHjwIPv27ePee+9l4cKFdOzYkQMHDlT6urKysnjrrbcYOHAgAH/84x9p1qwZHo+HSy65hNWrV5OcnMy4ceOYM2cOqampHDlyhMjISO6++26mT5/Oiy++yKZNm3C73fTu3fts31oREREREakhlY60GWPeNMbsNcasPc16Y4x52Riz2Riz2hjT1/k2a8/UqVPp3bs3AwcOZOfOnWRlZdGiRQvOO+880tPTycvLY8OGDQwZMoRFixYxevRoIiIiiI2N5dprrz1lzZycHK644gp69uzJc889x7p16wBYsGABDz/8cPl2TZs2JT09nWHDhpVfP6xZs2aV9pyYmFge2ADmzp1L37596dOnD+vWrWP9+vVs3LiRNm3akJqaCkDjxo0JDQ3lxhtv5KOPPqK0tJQ333yTO++8s7pvnYiIiIiI1ICqjLRNByYDM06z/kqgk/82AHjV/zMglY2I1YS0tDTS0tJYsmQJUVFRDB8+vPxaXuPHj2fu3LkkJyczZsyYs5oy/pFHHuHnP/85o0aNIi0tjaeffvqsewsNDT3hfLWK1xiLjo4uv79t2zaef/55MjIyaNq0KXfeeecZr0cWFRXFZZddxvvvv8/cuXNZsWLFWfcmIiIiIiI1p9KRNmvtQuBMx+iNBmZYn3SgiTGmjVMN1qbDhw/TpEkToqKi2LBhA+np6eXrxowZw/vvv88777zD+PHjARgyZAgffvghbrebgoICPvroo9PWbdeuHQBvvfVW+fLLLruMKVOmlD8+ePAgAwcOZOHChWzbtg2g/PDIpKQkVq5cCcDKlSvL15/syJEjREdHExcXx549e/j0008B6NKlC7m5uWRkZACQn59PWVkZAPfccw+PPvooqampNG3a9CzfNRERERGRmmOtxeO1lJR5cZd6KCwpI99dyuGjJRwscLP/yFH2HjxCbt4Bdu3dT87uvezYlcu2nTls2b6Dzdu2kb1zZ12/jIA4cU5bO6Diu5DjX5brQO1aNXLkSCZPnkzXrl3p0qXLCYccNm3alK5du7J+/Xr69+8PQGpqKqNGjaJXr160atWKnj17EhcX96O6Tz/9NDfeeCNNmzbl4osvLg9cTz75JA8//DA9evTA5XLx1FNPcf311zNt2jSuv/56vF4vLVu25IsvvmDs2LHMmDGD7t27M2DAADp37nzK19C7d2/69OlDcnIyCQkJDBkyBIDw8HDmzJnDI488QlFREZGRkSxYsICYmBhSUlJo3LgxEydOdPotFRERkSBjrcVrweO1eK3v5vFavF7w+B97vRaPf7n1b+vxL/da8Hi8eL0evGUlWG8Z1lOG11OK9ZRhPaVYTyleTxl4/Y/92+D14PV6sF6vb5nXA8ce2zLwerFeT/ly7PH71nrBejFeD9Z6MOXrvRjrAa8XrMd33//Y4D1+33ow1r+t9fruU2FZSQmL018AwGBP+Hni/WPrTrXM/9OeYtnxT+CM9Y09036ClQUsxloMXv9PT4XHXgyWkAr3j20Xgte33L8spHyZ/zHHHx/bLgxLBF5CTNXfl42hyfDk0pp7C2qYsbbyF2uMSQI+stb2OMW6j4A/WWu/9T/+EviVtXb5Kba9D7gPoFWrVimzZ88+YX1cXBwXXHDB2b8KB3k8HlwuV5W3LygoICYmhsLCQq688kpeeuklLrzwwhrs0Hm5ublcddVVrFixgpCQc2dC0c2bN3P48OG6buMEx75PInVF30GpS8H8/fN4LWVeKLPg8UKZtb6fXvBYKPNaPPbYff/25due+LjisvLtK9Q49tjrKSPM6ybUW0y4dRPmcRPmdRNu3TSyvmWNvG4aUUyEdeOyZYTgIRQPLus5fh8vIfbYfd/PY8tDzfHlYf5lYZT9aF0o3grP8xBuPHX9kZyRxxq85lgc8N08/n/ye33vSHkkKN/OGn+y8sWr49HJFycqsuXbnBi7Tlx3umWcUO+EWuWLT18rWB2LZeB77zkWuYx/uSmPZWBC/K/XgP/zwRx/7Ft3fDswvvsmxL+N/7G/lu/x8W0qPj528zRqSsR5Q+ruDTqNESNGrLDW9qtsOydG2nYBCRUet/cv+xFr7TRgGkC/fv3s8OHDT1ifmZlJbGysAy1VX35+/ln1cP/997N+/Xrcbjd33HEHQ4cOrcHunDdjxgyeeOIJXnjhhVOOEjZkERER9OnTp67bOEFaWhon/7kQqU36Dkpdcvr7Z62lsMRDQXEZ+e4yjhaXld8vKD75cSlHiz3l933rjz92l57uOqiWcMqIpJho3EQa388oU0wU7hOWRVJMtHETRbHvduy+KSam/L5vuyjchFNW5dfqxeAJCcNrXP5bqP/mKv9p/fetCcUb4sKaMGxIBF4Thg1xYU0oNiQU/PdLQnw3a1zgCoMQ/3oTCq5QCKlwc4VhQo4vN64wCAkjxOWrZ1y+ZSEhLkyICxMSQogrFBPiIiQklBCXC+NyERLiIsT4tg8JdeEK8W/jcuEKcZXXw7ggJMT/0+X/h/nx+y5jqPp/wfvo7z8JZk6Etg+AScaY2fgmIDlsra13h0ZW16xZs+q6hYDcfvvt3H777XXdhoiISLniMg8Fbn9oKi6lwB+yym/u09yv+NhdRkFJGVU4oIgwlyE2IozoRi5iGoUR2yiU86KKSYr6gUS7i7ZlO2lZkkOk5wjhXjdhnkLCvEWElhUR6ikixFY9XFnjwoZHQ3g0hEVjGkVjwpr4HodHQXgMhEX5Hx/bzr88PKr8eSdvHxIWSchZTJImIvVLpaHNGPMOMByIN8bkAE8BYQDW2qnAJ8BVwGagENCJUSIiInJa1lp2HSoiMzefzNwjbNh9hI2789l76CjFX3xKied0o1rHGQMx4aHERIQS0+j4z9aNI054/KP7/sexjcKIDvUS684h/OBW2L8O8rJgv/9WVGEONlc4NDsfGsdDeKtqhapj2xpX+FnNQC0iAlUIbdbaCZWst8DDZ9pGREREzk1FJR427fGFM98tn8zdR8h3Hx+dSmweRXLrWBIjiul8XiKx/pAV7Q9ZFR8fux8Z5iIkpIrhp/CAP4xtgpwKwezgNvBWGCWLaQXNO0G3URDf2Xc/vhM06eA77E5EpI44cXikiIiInOOsteQedrNhty+YrfeHtO37j+L1H6IYHe4iuU1jRvVuS9c2jenapjHJrWOJbuT754jvnKLk6jXgKYOD2/2jZZuOB7O8LCjMO77dsVGzlsm+cNa8kz+gnQ+RTQJ7E0REaohCm4iIiJwVd6mHrD0FvpGz3Uf8hzjmc6iwtHybhGaRdG3dmGt7taVrm1i6tmlMQtOoqo+OnU7hAcjbfHzkLG+z7+eBbeA9vn+iW/pGyZKv8YWy+GOjZokaNRORekehLUgNHjyYxYsXn3GbpKQkli9fTnx8/AnLn376aWJiYvjlL3/peF/Dhw/n+eefp1+/SmcmrbG6U6dOJSoqShOoiIjUMGste/OLy0fNNvjPQdu6/yge//BZZJiLLq1jubJHG7r5w1mX1rHERoRVf8eeMji04/hI2f5NsN8fzgr3H98uJMw3Qhbf2R/Ojo2aXaBRMxFpUBTaglRlga0ueDzBcU2WBx54oK5bEBFpcIrLPGzeW1A+Ocix0bMDR0vKt2nXJJKubRozskfr8sMbE5s5MHp2eBd8N5Pua7+CtY/Dga0njppFxfuD2VUnnWuW6JtiXkSkgdPfdCeZMGECubm5uN1ufvrTn3LfffcxdepUtmzZwnPPPQfA9OnTWb58OZMnT+aZZ55h5syZtGjRgoSEBFJSUn40wnXnnXfSuHFjli9fzu7du3n22We54YYbAHjuueeYO3cuxcXFjBkzht///vcAxMTEUFBQgNfrZdKkSXz11VckJCQQFhbGXXfdVf78V155hQ8//JDS0lLeffddkpN95wKsWrWKQYMGsX//fh5//HHuvfderLU8/vjjfPrppxhjePLJJxk3bhxpaWk8//zzfPTRRwBMmjSJfv36ceedd5KUlMS4ceP44osvePzxxwF49913eeihhzh06BBvvPEGQ4cOxe128+CDD7J8+XJCQ0N54YUXGDFixGmXFxUVMXHiRFatWkVycjJFRUWn/Dx+/etf88EHHxAaGsrll1/O888/Xz6SePPNN3PVVVeVb7tmzRq2bt1KVFQUDzzwANnZ2QC8+OKLDBkSfBdTFBGpK3vz3WTm5rOhwuQgW/YVUOYfPYsIC6FLq1gu79aq/Lyz5DaNiYsMYPTsVHJXweLJsO49sF6iIttChwt94exYMGt+AUQ1c3a/IiL1TPCGtk9/DbvXOFuzdU+48k9n3GTKlCkkJiZSVFREamoqY8eOZezYsQwaNKg8tM2ZM4cnnniCjIwM5s2bx6pVqygtLaVv376kpKScsm5ubi7ffvstGzZsYNSoUdxwww3Mnz+frKwsli1bhrWWUaNGsXDhQoYNG1b+vPfee4/t27ezfv169u7dS9euXbnrrrvK18fHx7Ny5Ur+9re/8fzzz/P6668DsHr1atLT0zl69Ch9+vTh6quvZsmSJXz//fesWrWK/fv3k5qaesK+Tqd58+asXLkS8B2aWFZWxrJly/jkk0/4/e9/z4IFC5gyZQrGGNasWcOGDRu4/PLL2bRp02mXv/rqq0RFRZGZmcnq1avp27fvj/abl5fHv//9bzZs2IAxhkOHDp2wvm3btnz//ffln9vXX39NYmIiN998Mz/72c+46KKLyM7O5oorriAzM7PS1yki0lAVlXiY+vUWVmYfJDP3CPsLjo+etY2LILlNYy7t1rJ89CypeTSuQEfPTsfrhc1fwOJXYPs3vinx+98HA+4nY9V2XdxYROQUgje01ZGpU6fyySefALBz506ysrIYOHAg5513Hunp6XTq1IkNGzYwZMgQXnrpJUaPHk1ERAQRERFce+21p6173XXXERISQrdu3dizZw8A8+fPZ/78+fTp0weAgoICsrKyTghS3377LTfeeCMhISG0bt2aESNGnFD3+uuvByAlJYX33nuvfPno0aOJjIwkMjKSESNGsGzZMr799lsmTJiAy+WiVatW/OQnPyEjI4PGjRuf8T0ZN27cafe5ffv28j4feeQRAJKTk0lMTGTTpk2nXb5w4UIeffRRAHr16kWvXr1+tN+4uDgiIiK4++67ueaaa7jmmmtO2d+iRYt47bXX+PbbbwFYsGAB69evL19/5MgRCgoKiImJOePrFBFpiHYfdnPvjOWs/eEwPdrGcXFyS5JbN/YHtFiaRIXXTiOlRbB6DiyZ4js3LbYtXPYH6HtHhfPPttdOLyIi9UzwhrZKRsRqQlpaGmlpaSxZsoSoqCiGDx+O2+0GYPz48cydO5fk5GTGjBlz1hfGbNSoUfl936XtfD9/85vfcP/991e752N1XS4XZWXHrzVzcn9n6jc0NBSv9/iFTI+95mOio6OrtE+nhYaGsmzZMr788kv+9a9/MXnyZL766qsTtsnNzeXuu+/mgw8+KA9lXq+X9PR0IiIiaqw3EZH64Lvsg9z3zxUUlXh4/fZ+XNK1Ve03UbAPMl733Qr3Q+tecP1r0H0MuBw+3FJEpIEKqesGgsnhw4dp0qQJUVFRbNiwgfT09PJ1Y8aM4f333+edd95h/PjxAAwZMoQPP/wQt9tNQUFB+TlhVXXFFVfw5ptvUlBQAMCuXbvYu3fvCdsMGTKEefPm4fV62bNnD2lpaVWq/f777+N2u8nLyyMtLY3U1FSGDh3KnDlz8Hg87Nu3j4ULF9K/f38SExNZv349xcXFHDp0iC+//PKsXgfA0KFDefvttwHYtGkT2dnZdOnS5bTLhw0bxqxZswBYu3Ytq1ev/lHNgoICDh8+zFVXXcVf//pXVq1adcL60tJSbrzxRv785z/TuXPn8uWXX345r7zySvnjY4dQioicS/7z3S7GTUsnMszFew8Nrv3Atm8jfPAo/LU7fP0naN8P7vgI7l8IvW5SYBMROQvBO9JWB0aOHMnkyZPp2rUrXbp0YeDAgeXrmjZtSteuXVm/fj39+/cHIDU1lVGjRtGrVy9atWpFz549iYuLq/L+Lr/8cjIzMxk0aBDgm3xk5syZtGzZsnybsWPH8uWXX9KtWzcSEhLo27dvlfbRq1cvRowYwf79+/ntb39L27ZtGTNmDEuWLKF3794YY3j22Wdp3bo1ADfddBM9evSgY8eO5Ydrno2HHnqIBx98kJ49exIaGsr06dNp1KjRaZc/+OCDTJw4ka5du9K1a9dTnguYn5/P6NGjcbvdWGt54YUXTli/ePFili9fzlNPPcVTTz0FwCeffMLLL7/Mww8/TK9evSgrK2PYsGFMnTr1rF+TiEh95PVanpu/kVfTtjCgYzNevTWFZtG1dAiktbBtISyZDFnzITQCLpwAAx+GFp0rf76IiJySOXaoXm3r16+fXb58+QnLMjMz6dq1a530c0x+fj6xsbFV3v7YuVKFhYUMGzaMadOmnXJSjUAc20deXh79+/dn0aJF5WFLqi8Yvm8nS0tL00n4Uqf0HazfCorL+J/Z37Mgcw83D+jA09d2Jzy0Fg6q8ZTC2vdgySu+ScSi4n2Ti6TeDdHxlT/fT98/qUv6/kldMMassNZWeqFijbQF6L777mP9+vW43W7uuOMOxwMbwDXXXMOhQ4coKSnht7/9rQKbiIj8yM4Dhdzz1nI27yvgD6O7c9vAxLM+//qsFR2CFdNh6d8h/weI7wLXvgy9xkGYzisWEXGKQluAjp2XVZOqeh6biIicm5ZuzePBt1dS5vHy1sT+XNSp6qNb1XJwO6S/Civ/CaVHoeNPYNTLcP4lEKLT5UVEnBZ0oc1aW/P/MyjnvLo6LFhExGmzl2Xz2/fXktAsijfuSKVjfHTlT6qunRm+QyAzPwQTAj1ugEEPQ5sfX7ZFREScE1ShLSIigry8PJo3b67gJjXGWkteXp4uCSAi9VqZx8sfP8nkH4u2M6xzC16Z0Ie4yBqYkdHrgQ0fweLJkLMMIuJgyE9956w1buv8/kRE5EeCKrS1b9+enJwc9u3bV2c9uN1u/WP+HBAREUH79u3rug0RkWo5XFTKpFkr+SZrP3df1JHfXJlMqMvhwxKLC+D7tyH9b77DIZsmwZXPwoW3QKMYZ/clIiJnFFShLSwsjI4dO9ZpD2lpadWa8l5ERKQ2bN1XwD1vLWfnwUKeHduLm1ITnN3BkR9g2TRY/ia4D0P7/nDZHyD5GghxObsvERGpkqAKbSIiInJ632Tt4+G3VxLqCmHWvQNJTWrmXPHda3yHQK6dB9YDXa+FQZMgob9z+xARkWpRaBMREQly1lreWrydZz7OpFPLGF67vR8JzaICL+z1wuYFvothb/sawqJ911Yb8AA0q9sjX0RE5DiFNhERkSBWUublqQ/W8c6ybC7r1ooXx11IdKMAf32XumH1HFgyBfZvhNi2cOnvIeVOiGziSN8iIuIchTZxuqXRAAAgAElEQVQREZEgdeBoCQ/OXMHSbQd4eMT5/OKyLoSEBDC78tH9kPEGZLwGR/dB654wZhp0HwOh4c41LiIijlJoExERCUIbd+dzz4wM9hwp5qXxFzL6wnbVL1ZWAvOfgJUzoMwNna6AwZMgaSjoEjsiIkFPoU1ERCTILFi/h5/O/o7oRqHMvX8QFyYEeMjip4/Din9An9tg8CPQooszjYqISK1QaBMREQkS1lr+vnArf/5sAz3bxTHttn60jgvw2qEZb/gC20U/g0ufdqJNERGpZQptIiIiQcBd6uF/31vDe9/t4ppebXjuht5Ehgd4XbQdi32jbJ0uh4t/60yjIiJS6xTaRERE6tjefDf3zVjB9zsP8YvLOjPp4gswgZ5rdmgnzLkNmnaEsa/rwtgiIvWYQpuIiEgdWrvrMPfOWM6hwlKm3prCyB6tAy9aUgizbwZPCUx4ByLiAq8pIiJ1RqFNRESkjny8OpdfvPs9zaMb8a8HB9G9rQPhylr4YBLsXgM3z4X4ToHXFBGROqXQJiIiUsu8XsvLX2Xx4oIs+iU2ZeptKcTHNHKm+KIXYe08uOQp6Hy5MzVFRKROKbSJiIjUosKSMn757io+WbObG1La88cxPWgU6tD5Zpvmw4LfQ/frfbNFiohIg6DQJiIiUkt+OFTEvTOWk5l7hCeu6so9QzsGPuHIMfuzYN490LonjJ6ii2aLiDQgCm0iIiK1YGX2Qe6bsYLiUg9v3JHKiOSWzhV3H4Z3JoArDMbPgvAo52qLiEidU2gTERGpYe+tzOHX89bQpkkE79w7gE6tYp0r7vXAvHvh4Da4/QNokuBcbRERCQoKbSIiIjXE47U8+/kG/v71Vgaf35wpN/elaXS4szv56v8g63O4+gVIGuJsbRERCQoKbSIiIjUg313K/8z+ni837OW2gYn87tpuhLlCnN3J2nnw7QuQciek3u1sbRERCRoKbSIiIg7LzivknhkZbNl3lGdGd+e2QUnO7yR3FfznYUgYCFc+53x9EREJGgptIiIiDlqyJY+H3l6B18I/7+rP4Avind/J0f0w+xaIagbj/gmhDh9yKSIiQUWhTURExCGzlmbzu/fXkhQfzeu39yMpPtr5nXhKYe7tcHQf3PUZxDg4C6WIiAQlhTYREZEAeb2WP3y0numLtzO8SwtentCHxhFhNbOzz34NOxbB9a9D2z41sw8REQkqCm0iIiIB+mhNLtMXb2fikCSevLobrpAaurD18n9Axusw+FHodWPN7ENERIKOQpuIiEgASsq8/GX+RpJbx/Lbq7sRUlOBbccS+OQxuOBSuPTpmtmHiIgEJYfnHhYRETm3zFm+kx15hfxqZHLNBbbDOTD3NmjSAca+DiGumtmPiIgEJY20iYiIVFNhSRkvf5lF/6RmDO/SomZ2Ulrkmymy1A13fgyRTWtmPyIiErQ00iYiIlJN/1i0nX35xfzqyi4YUwOjbNbCB4/6rsk29jVo0cX5fYiISNBTaBMREamGQ4UlTP16C5d2bUVKYrOa2cniV2DNXLj4CehyZc3sQ0REgp5Cm4iISDW8+vUWCorLeOyKGhr9yloAC56CbtfB0F/WzD5ERKReUGgTERE5S7mHi5i+aDtj+rSjS+tY53ewfzP86y5o2Q2u+xvUxKGXIiJSbyi0iYiInKWXv8zCay0/u7Sz88XdR2D2zb4ZIsfPgvBo5/chIiL1imaPFBEROQtb9hUwd3kOtw1MJKFZlLPFvV547z7I2wy3vw9NE52tLyIi9ZJCm4iIyFl4Yf4mIkJDmHTxBc4XT/t/sOlTuOp56DjU+foiIlIv6fBIERGRKlqdc4iP1+Ryz9DziI9p5Gzxdf+Ghc9Bn9sg9R5na4uISL2m0CYiIlJFz32+kWbR4dwztKOzhXevgf88BO37w9V/0cQjIiJyAoU2ERGRKli0eT/fZO3n4REXEBsR5lzho3nwzs0Q0QTG/RNCHR7BExGRek/ntImIiFTCWsuzn22gXZNIbhnQwbnCnlJ49w4o2AN3fQqxrZ2rLSIiDYZG2kRERCrx2drdrMo5zP9c2omIMJdzhT9/ArZ/A6NehnYpztUVEZEGRaFNRETkDMo8Xp6bv5FOLWO4vm975wqvnAHL/g6DJkHv8c7VFRGRBkehTURE5Azmrcxh676jPHZFF1whDk0Qkr0UPvo5nDcCLv29MzVFRKTBUmgTERE5DXephxcXZNGnQxMu69bKmaKHd8GcWyGuPdzwJrh0ermIiJyZQpuIiMhp/HPJDnIPu/nVyGSME9PwlxbBnFugtBAmvANRzQKvKSIiDZ7+e09EROQUjrhLmZK2mZ90bsHA85oHXtBa+PB/4IfvYPwsaNk18JoiInJO0EibiIjIKUz7eiuHCkt57IouzhRM/xusng3D/xeSr3ampoiInBOqFNqMMSONMRuNMZuNMb8+xfoOxpj/GmO+M8asNsZc5XyrIiIitWNvvps3vt3Gtb3b0qNdXOAFt3wF85+ErtfCsMcCryciIueUSkObMcYFTAGuBLoBE4wx3U7a7ElgrrW2DzAe+JvTjYqIiNSWyV9tptTj5ReXdQ68WN4WeHcitEiG66ZCiA5yERGRs1OV3xz9gc3W2q3W2hJgNjD6pG0s0Nh/Pw74wbkWRUREak92XiGzlmYzvn8CSfHRgRUrzofZN4MxvvPYGsU406SIiJxTjLX2zBsYcwMw0lp7j//xbcAAa+2kCtu0AeYDTYFo4FJr7YpT1LoPuA+gVatWKbNnz3bqdTimoKCAmBj9UpW6oe+f1DV9B+Hvq9ys2OPh2WGRNIkIYFTMeumx9k80z8tgVe+nOdS0t3NNNlD6/kld0vdP6sKIESNWWGv7VbadU7NHTgCmW2v/YowZBPzTGNPDWuutuJG1dhowDaBfv352+PDhDu3eOWlpaQRjX3Ju0PdP6tq5/h1c/8MR0j//hgd/cj7XjUwOrNh//x/kLYWRf+bCgQ8402ADd65//6Ru6fsnwawq/4W4C0io8Li9f1lFdwNzAay1S4AIIN6JBkVERGrLc59voHFEGPf/5PzACq3/AL7+M1x4Cwy435nmRETknFWV0JYBdDLGdDTGhOObaOSDk7bJBi4BMMZ0xRfa9jnZqIiISE1aujWP/27cx4PDzycuMqz6hfasg38/AO36wdUv+M5nExERCUCloc1aWwZMAj4HMvHNErnOGPMHY8wo/2a/AO41xqwC3gHutJWdLCciIhIkrLU8+/lGWjVuxJ2Dk6pfqPAAvDMBGsXCuJkQFuFYjyIicu6q0jlt1tpPgE9OWva7CvfXA0OcbU1ERKR2fJm5lxU7DvL/Xd+TiDBX9Yp4yuDdOyA/FyZ+Co3bONukiIics5yaiERERKRe8ngtz32+kfPio7kxpX31C81/ErYthOtehfaVTgQmIiJSZbrCp4iInNP+890uNu7J5xeXdyHUVc1fi9lLYemrMOABuPBmZxsUEZFznkKbiIics4rLPLzwxSZ6tovjyh6tq19o0UsQ2RQu+V3l24qIiJwlhTYRETlnzVqaza5DRTw+sgshIdWc5XF/Fmz8BFLvhfBoZxsUERFBoU1ERM5RBcVlTP5qM4PPb85FFwRwadHFr4ArHPrf51xzIiIiFSi0iYjIOemNb7aRd7SEX41MxlT3WmoFe2HVbLhwAsS0cLZBERERP4U2ERE55+QVFPPaN1u5skdreic0qX6hZdPAUwKDHnGuORERkZMotImIyDlnyn+3UFhSxi8u71L9IiVHIeN1SL4a4i9wrjkREZGTKLSJiMg5JedgITPTd3BjSgIXtIypfqHvZkLRQRj8qHPNiYiInIJCm4iInFNeXJAFBn56aafqF/GUwZLJkDAAOgxwrjkREZFTUGgTEZFzxqY9+by3Moc7BiXStklk9Qtlvg+HsjXKJiIitUKhTUREzhnPf76R6PBQHhoewDlo1sKil6H5BdDlKueaExEROQ2FNhEROSeszD7I/PV7uP8n59E0Orz6hbZ/A7nfw6BJEKJfoyIiUvP020ZERBo8ay1//nQD8TGNmDikY2DFFr8CUfHQe7wzzYmIiFRCoU1ERBq8rzftY+m2Azx6yQVENwqtfqG9mZA1HwbcD2EBnBMnIiJyFhTaRESkQfN6Lc9+tpGEZpGMT+0QWLHFr0BYFKTe40xzIiIiVaDQJiIiDdpHa3JZn3uEX1zWhfDQAH7tHfkBVs+FPrdCVDPnGhQREamEQpuIiDRYpR4vf5m/keTWsYzq3TawYkungvXAoIedaU5ERKSKFNpERKTBmpOxkx15hfxqZDIhIab6hdxHYPk/oNtoaJrkWH8iIiJVodAmIiINUlGJh5e+zKJ/UjOGd2kRWLGVb0HxEV1MW0RE6oRCm4iINEhvLtrGvvxiHh/ZBWMCGGXzlEL6q5A0FNr1da5BERGRKlJoExGRBudQYQlTv97CpV1b0i8pwElD1r4HR3ZplE1EROqMQpuIiDQ4r369hYLiMn55RZfAClkLi1+GFslwwaXONCciInKWFNpERKRB2X3YzfRF2xnTpx3JrRsHVmzLV7BnLQx+BEL0K1NEROqGfgOJiEiD8tKXWXit5WeXdg682OKXIaY19Lwx8FoiIiLVpNAmIiINxtZ9BcxdvpNbBiSS0CwqsGK5q2BrGgx8AEIbOdKfiIhIdSi0iYhIg/GX+ZtoFBrCpIsvCLzY4lcgPAZSJgZeS0REJAAKbSIi0iCszjnEx2tyuWfoecTHBDgydijbN2tkyp0Q2cSR/kRERKpLoU1ERBqE5z7fSNOoMO4d2jHwYumvgjEw8MHAa4mIiARIoU1EROq9RZv3803Wfh4ecQGxEWGBFSs6CCvegh5jIa69Mw2KiIgEQKFNRETqNWstz362gXZNIrl1YGLgBZf/A0qP+qb5FxERCQIKbSIiUq99vm43q3IO8z+XdiIizBVYsbJiWDoVzhsBrXs606CIiEiAFNpERKTeKvN4efbzjXRqGcP1fR04lHH1XCjYA0MeDbyWiIiIQxTaRESk3pq3Moet+47yyyu64AoxgRXzen3T/Lfu6RtpExERCRIKbSIiUi+5Sz28uCCLCxOacHm3VoEXzJoP+zfC4Ed9M0eKiIgECYU2ERGpl/65ZAe5h938amQyxomQtfhlaNweuo8JvJaIiIiDFNpERKTeOeIuZUraZn7SuQWDzm8eeMGcFbBjEQx6CFwBXjJARETEYQptIiJS77y2cCuHCkt57IouzhRc/DI0ioO+tztTT0RExEEKbSIiUq/szXfz+jfbuLZ3W3q0iwu84IFtkPkB9JsIjWIDryciIuIwhTYREalXJn+1mVKPl19c1tmZgkumgHHBgAecqSciIuIwhTYREak3svMKmbU0m3GpCSTFRwde8GgefDcTeo2Dxm0CryciIlIDFNpERKTeeOGLjYS6DI9e0smZghmvQ1kRDH7EmXoiIiI1QKFNRETqhfU/HOH9VT9w15COtGocEXjB0iJYNg06XQEtkwOvJyIiUkMU2kREpF54fv5GGkeEcf9Pznem4PezoHA/DHnUmXoiIiI1RKFNRESC3vc7D/HVhr088JPziYt04DpqXg8smQxt+0LikMDriYiI1CCFNhERCXozlmwnOtzFbYMSnSm48RM4sNU3ymaMMzVFRERqiEKbiIgEtYNHS/hodS7X9WlHTKNQZ4ouehmaJELytc7UExERqUEKbSIiEtT+tSKHkjIvtw50aJQtOx1ylsGgSeByKASKiIjUIIU2EREJWl6vZdaybFISm9K1TWNnii56GSKbQp9bnKknIiJSwxTaREQkaC3ekse2/Ue5dWAHZwruz/Kdz5Z6L4Q7cHFuERGRWqDQJiIiQWtm+g6aRoVxZY82zhRc/AqENoL+9zlTT0REpBYotImISFDafdjNF5l7uKlfAhFhrsALFuyFVbOh9wSIaRF4PRERkVqi0CYiIkFpdkY2Hq/l5gEOHRq59O/gKYHBjzhTT0REpJYotImISNAp83iZvWwnwzq3ILG5A+eelRyFjNch+Wpofn7g9URERGqRQpuIiASdBZl72X3Eza1OjbJ9NxPch2Dwo87UExERqUUKbSIiEnTeXrqDNnERXJzcMvBinjJYMhkSBkCHAYHXExERqWUKbSIiElS27T/KN1n7mdC/A6EuB35NZb4Ph7I1yiYiIvWWQpuIiASVWUt3EBpiGJ+aEHgxa30X025+AXS5KvB6IiIidUChTUREgoa71MO7K3K4vHsrWjaOCLzg9m8g93sYNAlC9CtPRETqJ/0GExGRoPHx6lwOFZZy64BEZwouehmiW/iuzSYiIlJPVSm0GWNGGmM2GmM2G2N+fZptbjLGrDfGrDPGzHK2TRERORfMXLqD81pEM+j85oEX27MeNn8B/e+HMAdG7UREROpIaGUbGGNcwBTgMiAHyDDGfGCtXV9hm07Ab4Ah1tqDxhgHpvsSEZFzybofDvNd9iF+e003jDGBF1wyGcKiIPXuwGuJiIjUoaqMtPUHNltrt1prS4DZwOiTtrkXmGKtPQhgrd3rbJsiItLQzUzPJiIshBv6tg+82JEfYPVc6HMbRDULvJ6IiEgdqnSkDWgH7KzwOAc4+UI3nQGMMYsAF/C0tfazkwsZY+4D7gNo1aoVaWlp1Wi5ZhUUFARlX3Ju0PdP6lpdfQeLyizvrSikf+tQvlu2KOB65215iwSvh6UmBbf+TNUb+jtQ6pK+fxLMqhLaqlqnEzAcaA8sNMb0tNYeqriRtXYaMA2gX79+dvjw4Q7t3jlpaWkEY19ybtD3T+paXX0HZyzZTrFnHb+4bgAXJjQJrJj7CCy5DbqPZuCV4xzpT2qH/g6UuqTvnwSzqhweuQuoeLGc9v5lFeUAH1hrS62124BN+EKciIjIGVlrmZm+gx7tGtO7fVzgBVe+BcVHdDFtERFpMKoS2jKATsaYjsaYcGA88MFJ2/wH3ygbxph4fIdLbnWwTxERaaAyth9k054Cbh2QGPgEJJ5SSH8VkoZCu77ONCgiIlLHKg1t1toyYBLwOZAJzLXWrjPG/MEYM8q/2edAnjFmPfBf4DFrbV5NNS0iIg3HzPQdxEaEMurCtoEXWzsPjuzSKJuIiDQoVTqnzVr7CfDJSct+V+G+BX7uv4mIiFTJ/oJiPl2byy0DEokKD/A0a2t9F9Nu0RU6XeZMgyIiIkGgShfXFhERqQlzl++k1GO5ZUCHwItt+Qr2roPBj4AT13kTEREJEgptIiJSJzxey6yl2Qzo2IxOrWIDL7j4ZYhtAz1vDLyWiIhIEFFoExGROrFw0z5yDhZx68DEwIvlroKtaTDgfggND7yeiIhIEFFoExGROjEzfQfxMY24onvrwIstfgXCYyBlYuC1REREgoxCm4iI1Lqcg4V8tXEv41LbEx4a4K+iQ9mw9j1IuRMiA7wwt4iISBBSaBMRkVr3zrJsACb0d2ACkvRXfROPDHww8FoiIiJBSKFNRERqVUmZlzkZO7m4S0vaN40KrFjRQVjxFvQYC3HtnWlQREQkyCi0iYhIrfp83W72F5Q4MwHJ8jeh9Khvmn8REZEGSqFNRERq1cz0HbRvGsmwzi0CK1RWDEv/DudfDK17OtOciIhIEFJoExGRWpO1J5+l2w5w84AOuEICvAD26rlQsAcGP+pMcyIiIkFKoU1ERGrN20uzCXMZbuqXEFghr9c3zX/rnnDecCdaExERCVoKbSIiUisKS8qYtzKHK3u0IT6mUWDFsubD/o2+UTYT4IidiIhIkFNoExGRWvHhqh/Id5c5MwHJ4pehcXvoPibwWiIiIkFOoU1ERGrFzPRsOreKITWpaWCFclbAjkUw6CFwhTnTnIiISBBTaBMRkRq3auch1uw6zK0DEzGBHs64+CVoFAd9b3emORERkSCn0CYiIjVuZvoOosJdjOnTLrBCB7ZC5oeQehc0inWmORERkSCn0CYiIjXqcGEpH67+gdEXtiM2IsDDGZf8DUJCYcADzjQnIiJSDyi0iYhIjfrXyhzcpV5uHdghsEJH8+C7mdDrJoht7UxzIiIi9YBCm4iI1BhrLW8v3UGfDk3o3jYusGIZr0NZEQx6xJnmRERE6gmFNhERqTFLtuSxdd9Rbh0Q4DT/pUWwbBp0ugJaJjvTnIiISD2h0CYiIjVm5tIdNIkK4+pebQIrtHYeFO6HwZOcaUxERKQeUWgTEZEasfeIm/nr9nBjSnsiwlyBFct4A1okQ9JQZ5oTERGpRxTaRESkRszO2EmZ13JzoIdG/vAd/LAS+t0FgV7jTUREpB5SaBMREceVeby8syyboZ3i6RgfHVixjDcgLAp6j3emORERkXpGoU1ERBz31Ya95B52c0ugo2xFh3zns/W8ASICnH1SRESknlJoExERx81cmk3rxhFc2rVlYIVWz4HSQuh3tzONiYiI1EMKbSIi4qgdeUdZuGkf4/snEOoK4NeMtb5DI9ulQNsLnWtQRESknlFoExERR81amo0rxDA+tUNghXYsgv0bNcomIiLnPIU2ERFxjLvUw9zlO7msaytax0UEVizjDd95bN3HONOciIhIPaXQJiIijvl0bS4HC0u5ZWCAo2wFeyHzQ7jwFgiPcqY5ERGRekqhTUREHDMzPZuk5lEMOT8+sEIrZ4C31HdtNhERkXOcQpuIiDgiM/cIK3Yc5JYBiYSEBHARbK8HVkyHjsMgvpNj/YmIiNRXCm0iIuKImek7CA8N4YaU9oEV2rwADu/UBCQiIiJ+Cm0iIhKwguIy/vPdLq7p1Yam0eGBFct4A2JaQ/LVzjQnIiJSzym0iYhIwP793S6Olni4dWBiYIUO7oCs+dD3dnCFOdOciIhIPafQJiIiAbHW8nb6Drq1aUyfhCaBFVsxHYyBlDsc6U1ERKQhUGgTEZGArNhxkA2787l1YCLGBDABSVkJfPdP6DwS4gI8L05ERKQBUWgTEZGAzEzfQUyjUEZf2DawQpkfwNF9moBERETkJAptIiJSbXkFxXyyZjfX921HdKPQwIotfxOaJsH5FzvSm4iISEOh0CYiItX27oocSjzewCcg2ZsJOxZBykQI0a8mERGRivSbUUREqsXrtcxamk3/pGZ0bhUbWLHl/wBXOPS51ZnmREREGhCFNhERqZaFWfvIPlDILQM7BFao5Cisege6XQfR8c40JyIi0oAotImISLW8vTSb5tHhjOzROrBCa/4FxUcgVROQiIiInIpCm4iInLUfDhXxZeYebkpNoFGoq/qFrIXlb0DL7pAwwLkGRUREGhCFNhEROWuzl2VjgZv7B3ho5K6VkLsK+k30XVRbREREfkShTUREzkqpx8vsjJ0M79yChGZRgRVb/gaERUOvcc40JyIi0gAptImIyFn5Yv0e9uYXBz7Nf+EBWDsPet0EEY2daU5ERKQBUmgTEZGzMjN9B+2aRDK8S8vACq2aDWVuTUAiIiJSCYU2ERGpsi37Cli8JY+bB3TAFRLAOWjWwvI3oX1/aN3TuQZFREQaIIU2ERGpsrfTswlzGW7qlxBYoW0LIS9Lo2wiIiJVoNAmIiJVUlTi4V8rdnJF99a0iG0UWLHlb0BkM98FtUVEROSMFNpERKRKPlz9A0fcZYFPQJK/GzZ8DH1ugbAIZ5oTERFpwBTaRESkSt5O30GnljEM6NgssEIrZ4C3DFImOtOYiIhIA6fQJiIilVqTc5hVOYe5ZUAHTCAXwfaUwYrpcN4IaH6+Y/2JiIg0ZAptIiJSqZnpO4gMc3F9SvvACmV9Dkd2aQISERGRs6DQJvL/t3ffUVaV9/7H38806gBSHHqTYlAsgIIdMf40saWQxBaVaExT48015Sb3571p99400zD+LGCJJsYYE9EYjREmGr0gzQZSBpAqIx2GMvX5/XGOkRiQgbNnzpk579daLM7eZ8/3fGE9ax8+PHs/W9J72ra7lkdfXstFx/WmU9vizIrNmQqlvWHYB5JpTpKkPGBokyS9p0fmrWFPbUPmC5BsXgEVz8DoK6GwKJnmJEnKA4Y2SdJ+xRh5YNYqju3XhaP7dM6s2Ny7IRTAqCuSaU6SpDxhaJMk7dfM5ZupeKuKy8f2z6xQXTXMvx+O/CB06p1Mc5Ik5QlDmyRpv+6ftZLO7Yq54NgMg9bCR2HXJhjjAiSSJB2sRoW2EMK5IYTFIYSKEMLX3uO4j4YQYghhTHItSpKy4a0de3jqtfVMHN2XtsWFmRWbPQW6DoZBZyTTnCRJeeSAoS2EUAjcCnwAGAFcEkIYsY/jSoEvArOSblKS1Pwemr2auobIZZleGlm5AFbPhDGfggIv8JAk6WA15tvzRKAixrg8xlgDPAhctI/jvg18D9iTYH+SpCyob4j8+sXVnDKkG4N7dMys2OwpUNgGjrssmeYkScozjVlzuQ+weq/tNcDYvQ8IIYwC+sUY/xhC+PL+CoUQrgWuBSgrK6O8vPygG25qVVVVOdmX8oPjT9n29hic/1Yda7dW86GBDRmNycK6XZw0/wE2dj+ZRS++klyjapU8ByqbHH/KZRk/KCeEUADcAlx1oGNjjHcAdwCMGTMmjh8/PtOPT1x5eTm52Jfyg+NP2fb2GLzn7hc5vHQ7N35sAsWFGVzSOGcq1O+h5/nfoGe/E5JrVK2S50Blk+NPuawx38RrgX57bfdN73tbKXA0UB5CeAMYB0xzMRJJaplWb97FX5ds4OIT+mUW2GKE2VOh50jo61eCJEmHqjHfxrOBoSGEQSGEEuBiYNrbb8YYt8UYu8cYB8YYBwIzgQtjjHOapGNJUpN6YNYqAnDxiRkuQLJmNlS+mlrmP4REepMkKR8dMLTFGOuA64CngNeBh2KMC0II3wohXNjUDUqSmk9tQ+ShOas5631l9O7SLrNis6dASSmM/FgyzUmSlKcadU9bjPEJ4Il37bt5P8eOz7wtSVI2zFlfz+adNVw+bkBmhXZthgW/hwapFbUAAB7FSURBVFGfhDYZrj4pSVKe84E5kqS/m7G6lgHd2nPakO6ZFZp/P9RXpy6NlCRJGTG0SZIAWLR+O0u2NHDpif0pKMjgHrSGhtSqkf1PgrIRyTUoSVKeMrRJkgB4YOYqigrgY2P6Hfjg97KiHLascJZNkqSEGNokSeysruP389dyQs9CunYoyazY7CnQvjuMcK0qSZKSYGiTJPHg7NVUVddxVv/izAptXweL/wTHXw5FbZJpTpKkPGdok6Q8t6e2njueXca4wV0Z0qUws2Jz74XYAKOvSqQ3SZJkaJOkvPfw3DVUbq/m+glDMytUXwvz7oUhZ0HXQck0J0mSDG2SlM9q6xu4rXwZx/fvwslHdMus2OI/wY43XYBEkqSEGdokKY/9Yf5a1m7dzfUThhBCBsv8A8yZAp36wrBzkmlOkiQBhjZJylv1DZFflC9jRK9OnDn88MyKbVoGy8tT97IVZHhfnCRJ+geGNknKU4+/so4VG3cmNMs2FQqKYNQVyTQnSZL+ztAmSXmooSFy64wKhh7ekXOO6plZsdrd8NIDcOT5UFqWTIOSJOnvDG2SlIf+vLCSJZVVXDdhCAUFGc6yLfgD7N4CJ7gAiSRJTcHQJkl5JsbI5BlLGditPeeN7JV5wTlToNtQGHha5rUkSdI/MbRJUp4pX7KB19Zu5/Pjh1BUmOHXwJuvwJrZMOZTkOl9cZIkaZ8MbZKUR2KM/PyZpfTp0o4PHd8n84JzpkBROzjuksxrSZKkfTK0SVIe+d/lm5i3aiufPWMwJUUZfgXs2Q6v/BaO/ii0OyyZBiVJ0j8xtElSHpk8vYIepW342Jh+mRd75TdQuxNO+FTmtSRJ0n4Z2iQpT8xduZkXlm3iM6cPpm1xhg/AjjH1bLZex0Gf0ck0KEmS9snQJkl5YvL0Crp2KOHSsf0zL7ZqJry10GX+JUlqBoY2ScoDr67ZxozFG7j61EG0LynKvOCcKdCmc+p+NkmS1KQMbZKUBybPWEqntkVccdKAzIvt3AgLH02tGFnSIfN6kiTpPRnaJKmVW7x+B08tqOSqUwZR2rY484Lzfwn1NTB6Uua1JEnSARnaJKmVu3VGBR1KCpl08sDMizU0wJy7YcCpcPiRmdeTJEkHZGiTpFZs+YYqHn9lHZefNIDDOpRkXnDZM7B1pcv8S5LUjAxtktSK3Va+jOLCAq45dXAyBedMhQ6Hw5EXJFNPkiQdkKFNklqp1Zt38fv5a7nkxP70KG2TecGtq2HJkzDqk1CUwKydJElqFEObJLVStz+7jBDgM2ckNMs2797UQ7VHX5VMPUmS1CiGNklqhSq37+Gh2WuYOLofvTq3y7xgfS3Muw+GnQNdEng4tyRJajRDmyS1Qnc8u5z6GPn8+COSKbjocaiqhDFXJ1NPkiQ1mqFNklqZjVXVPDBrJR86rg/9urZPpujsKdC5Pww5K5l6kiSp0QxtktTKTPnbCqrrGvj8mQnNsm1YAm88B2OugoLCZGpKkqRGM7RJUiuydVcN973wBueN7MURPTomU3Tu3VBQDMdfkUw9SZJ0UAxtktSK3PPCG+ysqecLZw5JpmDNLnjpARhxIXTskUxNSZJ0UAxtktRK7NhTy93Pv8HZI8p4X69OyRRd8Ajs2eYCJJIkZZGhTZJaiftnrmLb7lquS2qWDVILkPR4Hww4ObmakiTpoBjaJKkV2F1Tz13PLef0YT04tl+XZIqumw/r5sGYT0EIydSUJEkHzdAmSa3Ar19cxaadNVw/IeFZtuL2cOwnkqspSZIOmqFNklq46rp6bn92GeMGd+WEgV2TKbp7K7z6MIycCG07J1NTkiQdEkObJLVwD89dQ+X2aq6fMDS5oi8/CHW7XYBEkqQcYGiTpBastr6B28qXcXz/Lpx8RLdkisYIc6ZCn9HQ+7hkakqSpENmaJOkFuwP89eyZsturp8whJDUYiErn4eNi51lkyQpRxjaJKmFqm+I/KJ8GSN6deLM4YcnV3j2FGjbBY7+SHI1JUnSITO0SVIL9cdX32TFxp2JzrIV12yF1x+D4y6D4naJ1JQkSZkxtElSC9TQELl1egVDDu/IOUf1TKxurzefhoba1LPZJElSTjC0SVIL9PTrlSyu3MF1Zw6hoCChe9ka6um97ikYdAZ0T/B5b5IkKSOGNklqYWKMTJ5ewYBu7Tn/mF7JFV76NG2rNzjLJklSjjG0SVIL89clG3h17Ta+MH4IRYUJnsbnTKG65DA48rzkakqSpIwZ2iSpBYkx8vPpFfTp0o4PHd8nucKblsHSp3mz19lQWJxcXUmSlDFDmyS1IDOXb2buyi189ozBlBQldAqPEf74JSjpwLre5yZTU5IkJaYo2w1Ikhpv8oyl9Chtw8fG9Euu6EsPwPJyOO9H1OzsllxdSZKUCGfaJKmFmLtyC89XbOIzpw+mbXFhMkV3VMJTX4f+J8NoFyCRJCkXGdokqYWYPH0ph7Uv5tKx/ZMr+sRNULsHLvw5FPiVIElSLvIbWpJagNfWbmPG4g1cc9pg2pckdGX7wkfh9Wkw/ms+l02SpBxmaJOkFmDy9ApK2xbxyZMGJFNw9xZ44svQ8xg4+fpkakqSpCZhaJOkHLekcgdPLljPpJMH0qltQsvx//nfYedGuGiyS/xLkpTjDG2SlONunVFBh5JCJp0yKJmCy2bA/PvhlBug17HJ1JQkSU3G0CZJOWzFxp089vI6Lj9pAId1KMm8YM1OeOwG6DYEzvhq5vUkSVKT8zltkpTDbiuvoLiwgGtOHZxMwenfga2rYNKfoLhdMjUlSVKTcqZNknLUmi27eGTeWi45sT89SttkXnD1bJh5G5xwDQw4OfN6kiSpWTQqtIUQzg0hLA4hVIQQvraP978UQlgYQnglhPBMCCGh5c0kKX/d/tflhACfOSOBWba6aph2HXTqDWf9R+b1JElSszlgaAshFAK3Ah8ARgCXhBBGvOuw+cCYGOMxwMPA95NuVJLySeX2Pfxmzmomju5Hr84JXMb43C2wYRGc/2No2ynzepIkqdk0ZqbtRKAixrg8xlgDPAhctPcBMcYZMcZd6c2ZQN9k25Sk/HLHs8upb4h87owjMi9WuRCe+xGM/DgMOyfzepIkqVk1ZiGSPsDqvbbXAGPf4/irgT/t640QwrXAtQBlZWWUl5c3rstmVFVVlZN9KT84/gSwvSbyyxd2Ma5nEctffZHlmRSL9Yya91XaFrZjdukF1B5gfDkGlU2OP2WT40+5LNHVI0MIlwNjgDP29X6M8Q7gDoAxY8bE8ePHJ/nxiSgvLycX+1J+cPwJ4PtPLqI2LuObF5/CkMM7Zlbsf2+FHUvho1M4ZeSFBzzcMahscvwpmxx/ymWNCW1rgX57bfdN7/sHIYT3A98AzogxVifTniTll227arnvf1fywZG9Mg9sm1fAM9+GYefC0R9NpkFJktTsGnNP22xgaAhhUAihBLgYmLb3ASGE44HbgQtjjG8l36Yk5Yd7XniDquo6rjtzSGaFYkw9RLugCM67BUJIpkFJktTsDhjaYox1wHXAU8DrwEMxxgUhhG+FEN6+1uYHQEfgtyGEl0II0/ZTTpK0H1XVdUx9fgVnjyjjfb0yXOFx/v2w4lk4+5vQuU8yDUqSpKxo1D1tMcYngCfete/mvV6/P+G+JCnv3D9zJdt212Y+y7b9TXjqGzDgFBg9KZnmJElS1jTq4dqSpKa1u6aeu55bzunDenBsvy6HXihGeOImqK+GC34GBZ7mJUlq6fw2l6Qc8ODsVWysquH6CRnOsi18FBY9DuP/DbpnWEuSJOUEQ5skZVl1XT23/3U5Ywd15YSBXQ+90K7N8MSXodexcNJ1yTUoSZKyytAmSVn28Nw1rN++h+snDM2s0J//HXZtggsnQ2Gij+GUJElZZGiTpCyqrW/gtvJlHNevC6cM6XbohSqegZcegFNvhF7HJNegJEnKOkObJGXRoy+tY82W3Vw/YQjhUJ+lVl0Fj90I3YbC6V9JtkFJkpR1Xj8jSVlS3xD5xYwKRvTqxIQjDz/0QtO/A9tWwaQnobhtcg1KkqSc4EybJGXJE6++yfKNOzObZVv9Isz6f3DCp2HASck2KEmScoKhTZKyoKEhMnl6BUMO78g5R/U8tCJ11fDoddCpD7z/P5JtUJIk5QxDmyRlwV9er2Rx5Q6uO3MIBQWHOMv27A9h42K44CfQpjTZBiVJUs4wtElSM4sxMnlGBQO6tef8Y3odWpH1r8HfboFjPgFDz062QUmSlFMMbZLUzJ5dupFX1mzj8+OPoKjwEE7D9XUw7Tpo2wXO+e/kG5QkSTnF1SMlqRnFGPn5M0vp3bktHz6+76EVmXUbrJsPE6dChwye7SZJkloEZ9okqRnNWrGZOSu38NnxR1BSdAin4M3LYfp3YfgH4aiPJN+gJEnKOYY2SWpGP5++lB6lbfj4mH4H/8MxwrQboLAYzvsRHOpjAiRJUotiaJOkZjJ35Raer9jEZ04fTNviwoMvMO8+eOM5OPtb0Kl38g1KkqScZGiTpGZy64wKDmtfzKVj+x/8D29fB3/+dxh4Goy6MvnmJElSzjK0SVIzeHbJBqYveotrThtM+5KDXAMqRvjjTVBfAxf8FAo8dUuSlE/85pekJjZj8Vtcc98chpeVcsVJAw6+wMI/wOI/wplfh25HJN+gJEnKaYY2SWpCTy1Yz7X3zWFYWUcevHYcpW2LD67Ars3wxJeh13Ew7gtN06QkScppPqdNkprIYy+v48bfvMQxfTtzz6QT6dzuIAMbwFNfh91b4JO/h0JP2ZIk5SNn2iSpCfxu7hq++OB8Rg84jF9ePfbQAtvSv8DLv4ZT/wV6jky+SUmS1CL437aSlLBfzVrFN/7wKqcc0Z07rxhDu5JDWN6/egc8fiN0Hwanfzn5JiVJUothaJOkBN3z/Ar+87GFnDm8B7ddPvrQnscG8My3Ydsa+NSTUNQm2SYlSVKLYmiTpIT8v78u43/+tIhzjirj55eMoqToEK9AXzUTXrwDTrwW+o9LtklJktTiGNokKUMxRn72TAU//ssSLji2N7d8/FiKCw8xsNXugWnXQ+e+cNbNyTYqSZJaJEObJGUgxsgPnlrML8qXMXF0X7730WMoLAiHXvC5H8LGJXD576BNx+QalSRJLZahTZIOUYyRbz/+OlOfX8GlY/vznYuOpiCTwLb+Vfjbj+HYS2DI+5NrVJIktWiGNkk6BA0Nkf/76Gs8MGsVk04ZyM3njyCEDAJbfR08eh20OwzO+a/kGpUkSS2eoU2SDlJ9Q+Srv3uFh+eu4bNnHMFXzx2eWWADmPkLePMlmHg3tO+aTKOSJKlVMLRJ0kGorW/gXx96mWkvr+PG9w/li2cNzTywbVoGM74Lw8+Doz6cTKOSJKnVMLRJUiPV1DVww6/n8+SC9Xz13CP53PgjMi/a0ADTboDCNnDejyDTAChJklodQ5skNcKe2no+/8A8pi96i5vPH8GnTh2UTOF598LKv8EFP4NOvZKpKUmSWhVDmyQdwO6aeq795RyeW7qR7374aC4bOyCZwtvXwdM3w8DTYNQVydSUJEmtjqFNkt5DVXUdn7pnNnPe2MwPJh7Dx8b0S6ZwjPD4l6C+Fi74qZdFSpKk/TK0SdJ+bNtdy1V3v8gra7bxk4uP58JjeydXfMEjsORP8H++A90SuDdOkiS1WoY2SdqHLTtruGLqiyxav51bLx3FuUf3TK74rs3wxFeg9ygY+7nk6kqSpFbJ0CZJ77KxqprL75rF8o07uf2To5lwZFmyH/Dkv8GerXDho1DoaViSJL03/7UgSXup3L6HS++cydqtu5l65QmcOrR7sh+w9Gl45UE4/SvQ8+hka0uSpFbJ0CZJaWu37ubSO2eycUc19046kbGDuyX7AdU74LEboftwOP2mZGtLkqRWy9AmScCqTbu45M6ZbN9Tyy+vGcuo/ocl/yF/+SZsXwtX/xmK2iRfX5IktUqGNkl5b9mGKi67cxZ76ur51TXjGNm3c7IfsHMjvPBzmH0XjP0M9Dsx2fqSJKlVM7RJymuL1+/gsrtmEWPk158ex/t6dUqueNUGeOFnqbBWuxtGToSzbk6uviRJyguGNkl567W12/jklFkUFxbwq2vHMeTw0mQK76hMh7UpUF8NR09M3cPWY3gy9SVJUl4xtEnKS/NXbeHKqS/SsU0Rv/r0OAZ275B50R3r4fmfwpypUF8DIz+eCmvdh2ZeW5Ik5S1Dm6S8M/uNzUy6ezZdO5TwwDVj6de1fWYFt78Jz/8E5t4D9bVwzCdSYa3bEYn0K0mS8puhTVJeeaFiI1ffO4dendvyq0+Po2fntodebPs6+NuPYe690FAHx14Cp33JsCZJkhJlaJOUN8oXv8VnfjmXAd3ac/81Yzm89BAD27Y1qbA27z6IDemw9q/QdVCyDUuSJGFok5Qn/rxgPV/41TyGHl7K/deMpWuHkoMvsnU1/O0WmH9/Kqwdd1lqZu2wgYn3K0mS9DZDm6RW7/FX1nHjgy9xVJ/O3DfpRDq3Lz64AltXwXM/gvkPpLaPvzwV1rr0T75ZSZKkdzG0SWrVHpm3hpt++zKjBxzG1KtOoLTtQQS2LW+kwtpLv4JQAKOugFP/Bbr0a7J+JUmS3s3QJqnV+vWLq/j671/lpMHduOvKMbQvaeQpb/MKeO6H8PKDqbA2ehKceiN07tu0DUuSJO2DoU1Sq3TvC2/wH9MWcMawHtz+ydG0LS488A9tWpaaWXv5QSgogjFXp8Jap95N37AkSdJ+GNoktTq3/3UZ//2nRZw9oozJlx5Pm6IDBLZNy+DZH8ArD0FhMZx4LZzyRejUq3kaliRJeg+GNkmtys+eWcotTy/hvGN68ZNPHEdxYcH+D964NBXWXv0tFLaBsZ+FU26A0p7N17AkSdIBGNoktQq19Q385C9LuHXGMj5yfB++P/EYivYX2DYsgWe/D6/9LhXWxn0eTr4BSsuat2lJkqRGMLRJalEaGiJrt+5m8fodLK7cwZLKHSxev4PlG3ZSU9/AxSf0478+PJKCgvDPP/zWonRYewSK28FJ16XCWscezf8HkSRJaiRDm6ScFGNkw45qFqdD2ZLKHSyurGJp5Q521dT//bg+XdoxrKwjZwzvwbF9u3DuUT3/ObBVLkyFtQV/gOL2qfvVTr4eOnRv5j+VJEnSwTO0Scq6rbtqWFJZlZo522sGbeuu2r8f071jCcPKSvn4mH4M71nKsLJShpZ1pNN7PXetcgH89Xuw8FEo6Zh6xtpJ10GHbs3wp5IkSUqGoU1Ss9lVU8fSfYSzyu3Vfz+mtE0Rw3qW8oGjezG8rCPD0gGte8c2jf+g9a+mwtrrj0FJKZx2E5z0BWjftQn+VJIkSU2rUaEthHAu8FOgELgrxvg/73q/DXAfMBrYBHwixvhGsq1Kailq6hpYvrHqncsa11expHIHq7fsIsbUMW2KChha1pFThnRneFkpw3qWMryslF6d2xLCPu5HO5CG+lRYe/YHsOhxaNMJTv8KjPucYU2SJLVoBwxtIYRC4FbgbGANMDuEMC3GuHCvw64GtsQYh4QQLga+B3yiKRqWlDvqGyKrNu/a656z1Azaio07qWtIpbPCgsDg7h0Y2bczE0f3ZXiPNhzZtZC+HRoorNsNNVVQ+ybUVMCanbB8J9TuSu2v2QU1O6F2Z+r3mvT+2l3/vF23J9VUm85wxtdg3Geh3WFZ/NuRJElKRmNm2k4EKmKMywFCCA8CFwF7h7aLgP9Mv34YmBxCCDG+/X/qLcPKxS+xbdUrLHhhV7ZbUZ7avmopC1/Yme029qumppbKTZvZsHkTW7ZuZcf2bZQ07KF9qKYTe/hgm3qualtH18Nr6VxYQ4eCato07KGgdhesqoKKXdBQe+APelsohJIOqV/F7d953e4w6NQndZ9aSXp/cYfUvWpHT4R2XZruL0GSJKmZNSa09QFW77W9Bhi7v2NijHUhhG1AN2BjEk02l7ee+C8u2vYULM92J8pXR0HLGn+FqV8NBSWEkvaEd4eokm7v2u6Q3u74jyFsf+8XtYFDuVRSkiSpFWnWhUhCCNcC1wKUlZVRXl7enB9/QFV9z+O3HUZT0uYgFjyQElRTXU1Jm7bZbmO/CgoCXTq0o6RtOxoK2lJfmPoVCw7iVNIA7En/+oedO9K/lE1VVVU5d25W/nD8KZscf8pljfmX1lqg317bfdP79nXMmhBCEdCZ1IIk/yDGeAdwB8CYMWPi+PHjD6HlplVeXk4u9qX84PhTtjkGlU2OP2WT40+5rKARx8wGhoYQBoUQSoCLgWnvOmYacGX69URgeku7n02SJEmSctEBZ9rS96hdBzxF6g6WqTHGBSGEbwFzYozTgCnAL0MIFcBmUsFOkiRJkpShRt2IEmN8AnjiXftu3uv1HuBjybYmSZIkSWrM5ZGSJEmSpCwxtEmSJElSDjO0SZIkSVIOM7RJkiRJUg4ztEmSJElSDjO0SZIkSVIOM7RJkiRJUg4ztEmSJElSDjO0SZIkSVIOM7RJkiRJUg4ztEmSJElSDjO0SZIkSVIOCzHG7HxwCBuAlVn58PfWHdiY7SaUtxx/yjbHoLLJ8adscvwpGwbEGHsc6KCshbZcFUKYE2Mck+0+lJ8cf8o2x6CyyfGnbHL8KZd5eaQkSZIk5TBDmyRJkiTlMEPbP7sj2w0orzn+lG2OQWWT40/Z5PhTzvKeNkmSJEnKYc60SZIkSVIOM7RJkiRJUg4ztO0lhHBuCGFxCKEihPC1bPej1i2E0C+EMCOEsDCEsCCE8MX0/q4hhKdDCEvTvx+W7V7VeoUQCkMI80MIj6e3B4UQZqXPg78JIZRku0e1TiGELiGEh0MIi0IIr4cQTvL8p+YSQviX9HfvayGEX4cQ2nr+Uy4ztKWFEAqBW4EPACOAS0III7LblVq5OuBfY4wjgHHAF9Jj7mvAMzHGocAz6W2pqXwReH2v7e8BP44xDgG2AFdnpSvlg58CT8YYjwSOJTUOPf+pyYUQ+gA3AGNijEcDhcDFeP5TDjO0veNEoCLGuDzGWAM8CFyU5Z7UisUY34wxzku/3kHqHyx9SI27e9OH3Qt8KDsdqrULIfQFzgPuSm8HYALwcPoQx5+aRAihM3A6MAUgxlgTY9yK5z81nyKgXQihCGgPvInnP+UwQ9s7+gCr99pek94nNbkQwkDgeGAWUBZjfDP91nqgLEttqfX7CfAVoCG93Q3YGmOsS297HlRTGQRsAO5OX557VwihA57/1AxijGuBHwKrSIW1bcBcPP8phxnapCwLIXQEfgfcGGPcvvd7MfVMDp/LocSFEM4H3ooxzs12L8pLRcAo4LYY4/HATt51KaTnPzWV9L2SF5H6z4PeQAfg3Kw2JR2Aoe0da4F+e233Te+TmkwIoZhUYHsgxvhIendlCKFX+v1ewFvZ6k+t2inAhSGEN0hdDj6B1D1GXdKXC4HnQTWdNcCaGOOs9PbDpEKc5z81h/cDK2KMG2KMtcAjpM6Jnv+Uswxt75gNDE2vHFRC6obUaVnuSa1Y+v6hKcDrMcZb9nprGnBl+vWVwKPN3Ztavxjjv8UY+8YYB5I6302PMV4GzAAmpg9z/KlJxBjXA6tDCMPTu84CFuL5T81jFTAuhNA+/V389vjz/KecFVJXHwgghPBBUvd4FAJTY4zfzXJLasVCCKcCzwGv8s49RV8ndV/bQ0B/YCXw8Rjj5qw0qbwQQhgP3BRjPD+EMJjUzFtXYD5weYyxOpv9qXUKIRxHahGcEmA5MInUfyZ7/lOTCyF8E/gEqZWc5wPXkLqHzfOfcpKhTZIkSZJymJdHSpIkSVIOM7RJkiRJUg4ztEmSJElSDjO0SZIkSVIOM7RJkiRJUg4ztEmS8k4IYXwI4fH06wtDCF/Ldk+SJO1P0YEPkSSpZUg/KDfEGBsOeHBajHEaqYc6S5KUk5xpkyS1aCGEgSGExSGE+4DXgCkhhDkhhAXpB+i+fdy5IYRFIYR5wEf22n9VCGFy+vU9IYSJe71Xlf69Vwjh2RDCSyGE10IIpzXbH1CSlPecaZMktQZDgStjjDNDCF1jjJtDCIXAMyGEY4AlwJ3ABKAC+M1B1r8UeCrG+N103fZJNi9J0ntxpk2S1BqsjDHOTL/+eHo2bT5wFDACOBJYEWNcGmOMwP0HWX82MCmE8J/AyBjjjoT6liTpgAxtkqTWYCdACGEQcBNwVozxGOCPQNuDqFNH+rsxhFAAlADEGJ8FTgfWAveEEK5IrnVJkt6boU2S1Jp0IhXgtoUQyoAPpPcvAgaGEI5Ib1+yn59/Axidfn0hUAwQQhgAVMYY7wTuAkYl37okSfvmPW2SpFYjxvhyCGE+qZC2Gng+vX9PCOFa4I8hhF3Ac0DpPkrcCTwaQngZeJL0DB4wHvhyCKEWqAKcaZMkNZuQurRfkiRJkpSLvDxSkiRJknKYoU2SJEmScpihTZIkSZJymKFNkiRJknKYoU2SJEmScpihTZIkSZJymKFNkiRJknLY/wftj0LO46u3UwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 1080x504 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "acc = 0\n",
    "radius = 0\n",
    "results = {} \n",
    "count_neighbours = {}\n",
    "    \n",
    "while acc != 1:\n",
    "    for embedding, labels in test[['vectors', 'labels']].to_records(index=False):\n",
    "        result = 0\n",
    "        count = 0\n",
    "        neighbors = find_within_radius(embedding, radius)\n",
    "        \n",
    "        if neighbors:\n",
    "            count = len(neighbors)\n",
    "            for neighbour in neighbors:\n",
    "                if code.loc[neighbour, 'label'] in labels:\n",
    "                    result = 1\n",
    "        \n",
    "        if radius not in count_neighbours:\n",
    "            count_neighbours[radius] = [count]\n",
    "        else:\n",
    "            count_neighbours[radius].append(count)\n",
    "            \n",
    "        if radius not in results:\n",
    "            results[radius] = [result]\n",
    "        else:\n",
    "            results[radius].append(result)\n",
    "    acc = pd.np.mean(results[radius])\n",
    "    n_size = pd.np.mean(count_neighbours[radius]) / code.shape[0]\n",
    "    print(f\"radius: {radius}\\tacc: {acc:.5f}\\tn_size: {n_size:.5f}\")\n",
    "    radius += 5\n",
    "    \n",
    "plt.figure(figsize=(15, 7))\n",
    "plt.plot(\n",
    "    [r for r in results], \n",
    "    [pd.np.mean(results[r]) for r in results], \n",
    "    label=\"avg accuracy\", #  (mean % matching label found in the neighbourhood)\n",
    ")\n",
    "plt.plot(\n",
    "    [r for r in count_neighbours],\n",
    "    [pd.np.mean(count_neighbours[r]) / code.shape[0] for r in count_neighbours],\n",
    "    label=\"avg neighbourhood size\", # (mean size of the neighbourhood in % from all samples),\n",
    ")\n",
    "plt.title(\"Embedding perfomance\")\n",
    "plt.legend(loc=\"best\")\n",
    "plt.xlabel(\"radius\")\n",
    "plt.grid()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.6 (testing)",
   "language": "python",
   "name": "testing"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}