meta_content_to_dmoz_classes.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from lib import jsoncorpus, datastuff, analysis\n",
    "import gensim\n",
    "import traceback\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import sklearn.pipeline\n",
    "import sklearn.preprocessing\n",
    "import sklearn.cross_validation\n",
    "import sklearn.ensemble\n",
    "import sklearn.svm\n",
    "import sklearn.decomposition\n",
    "from lib.scikitComponents import *\n",
    "import itertools\n",
    "import collections"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# LDA"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data preprocessing\n",
    "\n",
    "Load all data, build a list of class labels using DMOZ top-level classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Done.\n"
     ]
    }
   ],
   "source": [
    "# Load in the previously created corpus and dictionary of scraped sites\n",
    "dictionary, corpus, meta_corpus, dmoz_data = jsoncorpus.load_or_create('docs/sites.jl', stemmed=True, prefix=\"stemmed_\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# Link topics to URLs\n",
    "meta = list(zip(dmoz_data['urls'], dmoz_data['dmoz_categories']))\n",
    "# Build array of class labels\n",
    "classes, top_categories, dmoz_encoder = datastuff.encode_dmoz_categories(dmoz_data['dmoz_categories'])\n",
    "# Set up datasets\n",
    "X = dmoz_data['meta']\n",
    "y = classes"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Define a classification pipeline for the corpus data\n",
    "\n",
    "The Pipeline() object chains together objects from the lib.scikitComponents file, so that they can be used as part of a scikit-learn classification pipeline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Define a classifier (decision trees), and chain the preprocessing step to the classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# classifier = sklearn.svm.LinearSVC(C=1.0, class_weight=None, dual=True, fit_intercept=True,\n",
    "#      intercept_scaling=1, loss='squared_hinge', max_iter=1000,\n",
    "#      multi_class='ovr', penalty='l2', random_state=None, tol=0.0001,\n",
    "#      verbose=0)\n",
    "\n",
    "# classifier = sklearn.svm.SVC(C=1.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
    "# gamma=0.0, kernel='rbf', max_iter=-1, probability=False, random_state=None,\n",
    "# shrinking=True, tol=0.001, verbose=False)\n",
    "\n",
    "classifier = sklearn.ensemble.ExtraTreesClassifier(random_state=0, n_estimators=100, oob_score=True, bootstrap=True, n_jobs=4)\n",
    "# Use stem=True to remove word suffixes\n",
    "builder = MetaMatrixBuilder()\n",
    "clf = sklearn.pipeline.Pipeline([\n",
    "    ('sanitiser', MetaSanitiser(stem=False, meta_selection_flags=13)),\n",
    "    ('matrix_builder', builder),\n",
    "    ('classification', classifier)])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Run the classifier\n",
    "\n",
    "We can use the meta corpus or the full body corpus here, just by replacing a single argument\n",
    "\n",
    "The data is split in to training and test sets, and then fit to the training set. The LDA model is generated ***only*** from the training set, not the testing set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.39 (+/- 0.07)\n"
     ]
    }
   ],
   "source": [
    "scores = sklearn.cross_validation.cross_val_score(clf, X, classes, cv=5)\n",
    "print(\"Accuracy: %0.2f (+/- %0.2f)\" % (scores.mean(), scores.std() * 2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = sklearn.cross_validation.train_test_split(np.array(X), classes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Do some analysis\n",
    "### Find the score for each category. \n",
    "\n",
    "Note that if the classifier assigned *random* categories, the score would be $\\frac{1}{\\textrm{num categories}}$. Instead, it is actually quite high in some cases, indicating a moderate amount of success (given how naive this is!)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.513333333333\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "       arts       0.61      0.39      0.47        44\n",
      "   business       0.40      0.56      0.46        45\n",
      "  computers       0.66      0.59      0.62        59\n",
      "      games       0.57      0.63      0.60        19\n",
      "     health       0.46      0.46      0.46        24\n",
      "       home       0.52      0.42      0.46        31\n",
      "kids and teens       0.44      0.37      0.40        19\n",
      "       news       0.60      0.60      0.60        40\n",
      " recreation       0.53      0.17      0.26        46\n",
      "  reference       0.41      0.56      0.47        34\n",
      "   regional       0.00      0.00      0.00         0\n",
      "    science       0.56      0.50      0.53        56\n",
      "   shopping       0.66      0.59      0.62        32\n",
      "    society       0.41      0.16      0.23        44\n",
      "     sports       0.48      0.78      0.59       107\n",
      "\n",
      "avg / total       0.52      0.51      0.50       600\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Library/Python/2.7/site-packages/sklearn/metrics/classification.py:960: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in labels with no true samples.\n",
      "  'recall', 'true', average, warn_for)\n"
     ]
    }
   ],
   "source": [
    "model = clf.fit(X_train, y_train)\n",
    "print model.score(X_test, y_test)\n",
    "predicted = model.predict(X_test)\n",
    "print sklearn.metrics.classification_report(dmoz_encoder.inverse_transform(y_test), dmoz_encoder.inverse_transform(predicted), dmoz_encoder.classes_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Is there a correlation between test-set proportion and score?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": [
       "iVBORw0KGgoAAAANSUhEUgAAAaIAAAEZCAYAAADVBiHZAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\n",
       "AAALEgAACxIB0t1+/AAAIABJREFUeJztnXe81FT2wL/DozdpVkCfYkexITbUp64KiF3sXVzWVXFd\n",
       "C+Lqz2dZdXXXvmsvrL1hF3FVHvZCkyIqVUAQFZSmIsL9/XFuTGZeZibzXjLJvDnfzyefSW5ubk4y\n",
       "yT255557LiiKoiiKoiiKoiiKoiiKoiiKoiiKoiiKoiiKoiiKoiiKoihKSXACMDKG8+4BTAOWAYfE\n",
       "cP6omAzsVQJlKorSwJkN/IRUst8ADwKt4hTIUgmsARrFLAfAm8C5EZQ7G9g3gnJLlTXAJiGUUw08\n",
       "HEI5fjwEXB1R2aVEFTA3biGUwklCheqHAfoDbYAdgZ7AZT75GhdRJu+5UkU8bzY2BD6LoFxDPNdX\n",
       "zP+yUJLwfyeFirgFSChJe36TJk9JMov0r/IbgRft+hrgz4hZaoZNO9NuLwJeANb3HLsGaTnMAL4D\n",
       "bsCtWFKIgpsNLASGAW3tvkp77OnAV8Bo+7sGaaktBXYFTgXe8Zxvd+AT4EfgY2A3z74a4CrgXXv8\n",
       "SKBjjvuQ7bpmAKuRVuNSoInPsV2B4cC3wPfA7Ta9G/CWTfsOeARYy+572FPuMuBCm74r8D7wAzAB\n",
       "2Ntzno2Bt60c/wP+TfqX/yHAFHvsKGBLz77ZwMXAROBnpJKbDexn96eAS4DpVt4ngfZ2X3Mr+/e2\n",
       "7I+BdXzug3Me53mqBp5C/uuliNlupyzHvY3838uR+zHApvdH7sMPwHvAtp5jhgDzbNmf2/P2AVYC\n",
       "v9pyxmc536bIc/Yj8t884dm3JXJ/F9lyHVn+aMtdact+wafcO5F3yMsLwPl2fQPgWeRZmUl6S7sa\n",
       "eAb5T5cg70MvYIzd/gb4l81bRe0WyWzce5/tOD8ORe7xEuT/P9Cmn4Z8gC1F3oM/2vRWyDO0Gvf9\n",
       "XI/czxDAych7/T1uXeA8f82AW4Cv7XIz0NRzrfOQ53cB8F9gEvJsODSx5W7nc32dgJeRZ2gR8qw5\n",
       "9VK2d7cRweurGpt+ur1fi4HXkA9YJSCzcB+GrkhlcaXdXoNU4O2QB2Vf5KXdHnlIbkNeZjz537T5\n",
       "uwJfAGfYfacjFX0l8iA/izxQ4P6xDwEt7Lk2orZp7lRcRdQBebBOsHmORR4A58GvsefbFKlIRwHX\n",
       "ZbkH+a4rU1l7qQA+RV50R/Y97L5uyL1tgrwMo5EXLFu5nZGXoY/d/oPddhToB4hyb2zPsQT3Hm6O\n",
       "VOL7WZkustfvfK3NBsbZczTzOf95iALcwMp7F/CY3TcI+ThpjrzAOyAtaD+8ZVYjFVYfe9y19hqy\n",
       "kWma2wGpBHa2x59sy28CbAHMQSpAkJfeOfYK3PuSjceBoXa9KfJRA/JszgVOQZ6r7ZFnYyu7/0Hk\n",
       "Aycbe1q5HNojHxvr2fLGIhVcY+TDYgZwgM1bjSg6px+yOXK/TrDbLREFA/6KyHvvM4/bJYu8vRBl\n",
       "7NQBGyD3FqCflRGk328F8p+AfCBlnj/XM7Q1orR2t/tutNfqyHuVPbaTXd7Dvc9VwCrk/W2C3JeL\n",
       "SP94OBR5D/24DvlAqLCL8376vbvOc1BIfdXcnn8acu8aAX+z16AEZDbygPxg1+/ArajWIA+Bw/3A\n",
       "9Z7tVsjDtKEn/wGe/WcBb9j1N4E/efZtbo9thPvHVnr2O2nZFNFJwIcZ1/I+UoGAKJ5LM2QZgT/5\n",
       "riuXItoN+ZoKYno9DFEGDpnlDqF2BfoaUgFviLyMzT37Hvbkv5z0FzOFfEU6jgOzkPvnxXv+zzJk\n",
       "WR+5BxXIl3FmayQbmYrodc++rZFKORuZiuhOalf6nyPX1A1RUo6i91JN/j6iYcDdiGL2cgzyxezl\n",
       "buD/7PpD5O4jSiFfyXva7TNx34Fd7D4vQ4EHPHLXZOwfbdM7ZaRXkVsRZTsuk7vJ3Vry8hwwOMf5\n",
       "cz1D/wc86tnXAmlZOvmn436AgdQjszznWonbQgJRdsuA1nb7GVyrQiZXAs8jz4yXXO9uofXVCER5\n",
       "OTRCFHfXLDLFRpL7iA5FvtwqgXOQP93B+7CtT/qLtAJp6nbOkn8O8sD4HTsH+SpcN8ux+diA9C9P\n",
       "bPkbeLa/8az/jPvQZhLkurLRFdeMmMm6iHKYh7ReHia3eXAjxAz0g2fZA/ma3gBp8f3iyT/Ps555\n",
       "PwxyP7P9N5lUIhWNc97PgN8QE9zDSMv4CcRs8g+C28UXetZ/QhRp0HdhI+AC0u9HF+T/mgH8Bals\n",
       "FyItnPV9SxGlsMwuk2zaxYjS+BixApzmOecuGec8Hvc5NXlkNsh9Os5uH49bAW+E/E/esoeSbub0\n",
       "/qcgFoXNgalW1oPynL/Q47rgmt0z6Yt87C2ysvYj9/Nbif8ztC7y33iv7WdbrsMG1K4fvO/yd4gi\n",
       "cJiPfBwdhVhg+pCu6LzciCi615FrHWLTc727hdZXGwG34l67c21B6pCiklRFlA/vizef9K+AVsiD\n",
       "+bUnbcOMdWdf5rEbIg+pt6IyWdb9+Br5871slCFLUIJcVzbmItfi17F8LWJH3wbpGzqJ9Ocg8xrn\n",
       "IJV+e8/SBjHHLUDMkS08+b1fW5n3I2X3e68h1z2dg7zM3nO3tOf9DWmZdEdMF/2RVlrUzAH+niFT\n",
       "a6TvAUT57Ilct0EUJNS+zneQ+9gGt1W3EOnz6IyYHv+DfDHPQVoTmf/B2VnK9uNxpILcCDF9Peu5\n",
       "nlkZZbfF7eswPuVPR5TZ2vb6nkGegRXI/+NQYfPkOy6TuYj5OpNmVu4bEEXZHngVt2/F7z5ke4bm\n",
       "I89RF0/eFqQrNb/6Yb5n2+98w4ATkY+39+05/FiOtJa6IWbPvyItsTlkf3cLra/mIM+T99pbUdtq\n",
       "Ezulqoi8PI58OW6HPKjXIjfa+yV+IW4f0WDSK43zkT+3tT32Cfy/RkC+gNZQuzntMAL54jsO+VI5\n",
       "BulkftmTJ6gHVpDrysZHyAtwPfLSNce1M7dGKoylSIV3UcaxC0m/vkeAgxGzRIUtq8oe+xXS+VyN\n",
       "mKJ2I72z9mnkq3dfu/8CpPX0foBrALHnX4v7IbE2bl9FFVKBVyCtilWIgg2bzPtxL2Ie6YX8l62Q\n",
       "a2yN/Pf7Iv/XSuRaHZm+QZ6zXP//ANyK8UekUlmNPD+bIxVcE7vsjOv4sZD8LuYTkL69+xDT6lKb\n",
       "/jFy/y5GKuIK5COlp93vJ++JuApmiZVzDfAl8nz0szJehmtSz3VcJvcjz/6+SB3VGennaGqX7+1x\n",
       "fUk3uy9EFElbT1quZ+gZ5NnezZZbnXG9j9trcPqI/o/85tXnEE/fweTuEzwIUbYp5L9YbZePyf7u\n",
       "Flpf3YV0BWxtt9fCdXJRApCr/2M1tV+6QcjX1iKkA9vbfF6DmPZmIA/wjbgKOIX0Y8xB7LL/xfUg\n",
       "q7TnylTWV9q8ixFzySmk2+/3QCrnHxHvud09+0aRbrPNPDaTXNeV6x6BKN3ncL3jbrHpW1v5liF9\n",
       "Q38lXbkdgiiYH+w+kEq3xsrxLfASbstnE1yvuTcQ+/59nvIOQ7zmfkSufyvPPr9r8KalkBfvc1v+\n",
       "dOAau+9Ym74cqeRvIfuHlbfMTKeBSvz/Z4dByJfoD0iLAsSD62ObNh/5sGmNKMaPrKzOf+Y4LnRA\n",
       "WkGLkfvvxz8QU9Eye60DPfs2RxSS40n1BtDD7tsU8cT7AfG2ysZl9lqPzEhfH+nAX2Dle5/s9wuk\n",
       "Ml6Ia1b0Dqg+BbknC5EPj5mesnIdl8lhSKf9UqTDfX+b/mfk//7ByvUY6X129yP3ZzGu11y2Z8iR\n",
       "1+s1Nw/XcaAZYtqab5dbSPeay/ZReJ+9xpZZ9oOYcGchz+9cxJHAIdu7W5f66kTEK3WJPe4+ypA+\n",
       "yAMwDdcG6qUT8nU2AbGJnxqBDGENSFSC8SRSeSlKqdEaaVlnmtcL5XLye0gqRaIC+fqoRJrpE0j/\n",
       "GgZpCjvuy52Qr8iwB2KpIoqWnojpqhFiKvkZ/3ETipJEDkZaLq0QU9bYepbXAWnp9K5nOWVFlH1E\n",
       "vRBFNBv5yngC8YTzsgDXntsWUUS/hSxHkI5cpe6sh5jcliHjkf5E9rETipI0DsEdsNoNMfnWlTMR\n",
       "89cIZNC6kgCOQjp2HU7EHSHs0Ajpe5iPVGR9iyKZoiiKkhiibBEFaYlcipjsNkBGi/+b7KPjFUVR\n",
       "lAZIlIHxviZ9TElXag+M2x0ZkwHi1TYLcdPM9CqaTnaXaUVRFMWfGfiPyXLYCRl/d15xxCk+jZGb\n",
       "UIm4PPo5K9yE62G1LqKoOviUVSr9PNVxCxCQ6rgFCEh13AIEoDpuAQJSHbcAAamOW4CAVMctQEBK\n",
       "ou6M0jT3GzJ+ZyQSVuNJJLTHILuADMjqiXRuv4EMqlscoUyKoijlQCvgFbs+CTgaGQT9PtIo+Ahx\n",
       "V69CxgU6xzxg943DHed1KjI+bQQyaNmJFgIyRGesLfONPOVkJeo5K0ZQO6jn3Z717xH3SUVRFCU8\n",
       "+uCG0toW8UoejyiksYgS+jnjmL8hgVVPRyLRfISrXLZD+vF/RWYwuM2u34OEtPrKHpOrnKzBhXXy\n",
       "pHCpiVuAgNTELUBAauIWIAA1cQsQkJq4BQhITdwCBKQmbgHS6d4Pug6GNs1h2S8w6Rn42olG0RuJ\n",
       "rLAAd5zUcp9CDkAaBk7E8GZIaCSDKJZlNv0zpMulAxJVxQnE+mOOcpwpeEqakrBzKoqiFJ/u/WDg\n",
       "NDDGXQZOg00GIHVnDdIX7ze2qQrXNDcG2MwnzymkD715CZn7qT8SizKTbOVkpSEEPVUURSljug6G\n",
       "ezM8467aFLo5sQr/iQQYWA83mG0bakf4Hok7txO4Ew76Bb41SBDmvXAjgjuOZtnKyYqa5hRFUUoW\n",
       "k4LT1qmdPgn4wAneejkyCWcjpGXTAumv2Z/0aT6uRgKsTrR5ZyKOBn5TgYD08f8RcWRohAS0PTBH\n",
       "OSWPmuYURVEAUT5mOzDXg5kNFy5LN8s5S98RaN0ZKnozFUUpc8ymYC4H85koIHO9KCS/PqIzpkt6\n",
       "adSdappTFEVJLKYzMsHmcYgH21PIPFUfQMpRMjbIcL9zoXULWP4zzLkdprwag8ANmpLQ6oqiKPXH\n",
       "dAQzCEwNmMVgHgCzP5i6NBy07gwRvZmKojRgTBswJ4J5BcwSME+COQxMs/zH5i44FPEUQG+moigN\n",
       "DtMczOFgnrLK52UwJ4hSCu8kIZZV9ujNVBSlAWAagzkAzIPW7DYKzB/FHBfNCSMqtyzRm6koSoli\n",
       "GoHZA8wdYBaC+RjM+dYRIfKTF+EcZYPeTEVRSgiTArM9mH+A+QrMFDCXiQt2cQUp8vkaNHozFUUp\n",
       "AcxmYP4PzFQws8BcB6aHKKZ4BIrpvA0SvZmKoiQU0wXMBWDGgPkGzG1gdotR+XjRujNE9GYqipIg\n",
       "TCcwfwIzGswiMPeD+UMdx/pEidadIaI3U1GUmDFtwJwEZgSYH8E8DuaQEMb6RInWnSGiN1NRlBgw\n",
       "zcEcCeYZO9bnRTDHgWkdt2QB0bozRPRmKopSJExjMH3ADAPzA5g3wQwE0yH/sYlD684Q0ZupKEqE\n",
       "mEZg9gTzHzDfgvkQzHlg1o9bsnqidWeI6M1UFCVkTArMjmBuBDMHzCQwl4LZJG7JQkTrzhDRm6ko\n",
       "SkiYLcBUg/kczEwwfwezTdxSRYTWnSGiN1NRlHpguoK5CMw4MAvA3AJml4SM9YkSrTuBPsDnwDRg\n",
       "iM/+C4HxdpkE/Aa088mnN1NRlAIxa4P5M5h37Fife8HsC6YibsmKSNnXnRXAdKASaAJMALbKkb8/\n",
       "8EaWfWV/MxVFCYJpC+YUMK/ZsT6PgTkYTNO4JYuJsq87dwNe82xfYpdsPAackWVf2d9MRVGyYVqA\n",
       "OQrMs3aszwtgjgXTKm7JEkDZ151HAfd6tk8Ebs+StyWwCH+zHOjNVBQlDdMETF8w/7Vjfd4AcwaY\n",
       "9nFLljBKou6MMi5SITfgYOBd4Mcceao96zV2URSlbDCNgN7AccCRwAzgceBiSH0Tp2QJosouimVX\n",
       "0k1zQ/F3WAB4Djg2R1klodUVRQkbkwKzE5h/gpkLZiKYoWA2jluyEqHs687GyBdLJdCU7M4KayFm\n",
       "uRY5yir7m6ko5YXZEsyVYL4EMwPMNWC6xy1VCaJ1J9AX+ALxnhtq0wbZxeEUxFEhF3ozFaXBYzYE\n",
       "czGY8WDmg7kZTK8yGOsTJVp3hojeTEVpkJh1wJwN5l0w34O5B8w+ZTbWJ0q07gwRvZmK0mAwa4E5\n",
       "FcxIO9bnUTD9y3isT5Ro3RkiejMVpaQxLcAMADPcjvV5HswxYFrGLVkDR+vOENGbqSglh2kCph+Y\n",
       "h+1Yn9fBnA4m23hBJXy07gwRvZmKUhKYRmD2BnMXmO/AvA/mXDDrxi1ZmaJ1Z4jozVSUxGJSYHqC\n",
       "+ReYeWAmgBkCpjJuyRStO8NEb6aiJA6zFZirwEyzy1Vgto5bKiUNrTtDRG+moiQCs5Ft7UywrZ9/\n",
       "2daQjvVJJlp3hojeTEWJDbMumHPAvGf7fe6y/UCN4pZMyYvWnSGiN1NRioppZz3cXrcebw9bD7gm\n",
       "cUumFITWnSGiN1NRIse0tGN7nrdjfYbbsT861qd00bozROK8mbOBDhGf4xWgbcTnUBQfTFMb1eBR\n",
       "G+VgpI16sFbckimhoIooROK8mbOAjjGeX1FCxlTYeG732Phu79p4b+vELZkSOqqIQqRYN7MV0jqZ\n",
       "AEwCjkYUUTUwFpgIbGHzdgCeBz4FPgC2tenVwMPA+8CXwECbXgW8DbwMfA7cCTieRrNteZXAVOAe\n",
       "YDIwEmhu8+xszz8euNHKpygBMSkbyfpmMF/bCNcXixec0oBRRRQixbqZRyJKwKEtoojOtttn4U5/\n",
       "fjtwuV3fB1EQIIpoPNAMaUnNAdZHFNHPiLJpBLxuz4c9h6OIVgE9bPqTwAl2fTKwi12/DlFKipIH\n",
       "0x2Zy2c6MrfPlWC2jFsqpWiUhCKKcqrwEqB7P+g6GNo0h2W/wKRn4Ov9geuRlsu7NuNw+zsOOMKu\n",
       "7+FZH4UonTbIH/8CsNIuo4BeyDToHyOtH5ApjnsDz2YINQtXyYxFlNNaQGvgI5v+GNC/rletNHTM\n",
       "xsiMx8cB7ZEPmmOAcZAqiYpJKS/KWBF17we73Qr3buqmndkN3roYZjYFrgHesjtW2t/VpN+zoIP4\n",
       "1thfbyWQ8qR7WelZX43/zLU6eFDJwKyHmJKPAzYFngHOAd6FlN9zpiiJoYwHpHUdnK6EAK7aFLoN\n",
       "BB5F+mF2yFHAO7hmsyrgO2AZoiQOxTXNVQGf2PReuKa5Y3BbXPlYYsvuZbePDXic0qAx7cGcAeYN\n",
       "pG+xJ3AVsAGkzoLU26qElFKgjFtEbZrXTpsEfLAH0sfzK/Bn4GlPBoPbqqkGHkCcFVYgU547eSYi\n",
       "JrlOSMXwDbAlopDuQL5Y3wKe8xyDz7p3+wykf2oNMBpRTkrZYVoBByMtnyrgDeAu4BVI/RyjYIrS\n",
       "4InArt3nNTCm9tJ3RD0LvgK4wCe9CnipHuW28qxfAtxcj7KUksI0BXMwmMfsWJ/XwJyiY32UAJRE\n",
       "n2AZm+bm3gZnTk9PGzgD5tweQuF+f77Jkh6Ug5CW2iTEUeKaepSlJB5TAWZfMPcC84GLEFPuZpDq\n",
       "A6lhkNJWsaIUkYi0evd+cMRo+L81cMnPsO1B0ZxHCZG9gd0824OAk2KSJWRMCswuYG4BMx/MODAX\n",
       "gdkwbsmUkqUkWkSlQoQ303QDMwPMVBnwpxRIirp78VXU4Zhq/E2fJYzZBszf7XP4OZgrwGyR/zhF\n",
       "yYsqohAphiK6Fsx10Z0nUirxj7SwE3BrlmNmU/cYepXAF8AwZKDt/yFjpD5FFIXDyTZtMeK8AfAQ\n",
       "0rn+ITAC2NX+jkEiTzgV8ME2zzjgf8A69rwLgHmImbI36Yppe3vMp8jYr3Y2vQYZG/aRlbt3Ha87\n",
       "RMwmYC4FMwnMHDA3gNkBnddHCRdVREAfJJzNNGBIljxVSKUyGakw/CiGItpZvkZLkkoKD/njRHOo\n",
       "6/lWI+7k+wN32/RGiEPGnkB3pNLvADwInGjzPAi8iLSiRiEKzHGj3wV40647SgQkTNI/7foVwF89\n",
       "+7zbE+25Aa7EdegYhbjjA/RFFFsMmPXBnAfmQzDfgvk3mN7ovD5KdJS9IqoApiOVVhMkfttWGXna\n",
       "AVOALna7U5ayiqGIUsiMk5kylgKVuIpoE6QVsRPpnnodkbBCkxE38NmIkvCLr5dBm1uh1Y/Qejl0\n",
       "+AY2+CMwE2ndjAV+QQbizkTi652GuJh/j1T6r+CGM3oQ6dM5ChkbtQb4yZ7/c8QVfgzSMV+DKJcZ\n",
       "wLc2fTbSusGe/0NEqc5ExnKBhFT62JY7yd4Pp19pXeTDKAwq8Y8N2I3fW3lN3ofhf4Nf34SNVoN5\n",
       "CMYegShyp2X2tj1mb+SjbLyVuXVIcirlS0kooii/xHohimg2Ej/tCWSgp5fjkRA38+z29xHKk4eU\n",
       "QYKYHh6fDPVmC2RE/SmIgvByBVLhbYOMX3I6wPsAXyNmrW2B19IP694PDukPy9eCZa3gz+tCx2uh\n",
       "kfPstAIG2+NXA5sjY5zWQUxwJwO7k/5C/GTlHA8sAloiQV0XW7l6IqFpfkXi7v2IKJqepMfoA6ms\n",
       "b0cUqNOKOt5e61R7/HKyR8eoL5siY8O2sXIeCRX3wrMjxeGgpgdcNRia/BvmvQGpG2CnX5H/Zy9k\n",
       "4HMXRNlegIxd2wFRUjouSCkLolREnYG5nu15Ns3LZshX+Sjkazdu76fngMNiliEA3fvJOKgBNfLb\n",
       "pQqp+J9HKmE/M92ewCN2/VXgB7s+ETGvXY9UfkvTD+s6GM7cRA7vgQSd6NERmnVElMtw4HQkuOu6\n",
       "yH98IBKBfACiAN4ifRyUw2pECR6FKNHuSEtoPNKya22P287uGw/0Q2LvYc8/FTHzjUGe595Ia+g0\n",
       "e+4e9jxRYWMDmmaw789w1sXQdB8YciWs3QP2mgETvofUcFg9GlE+eyKBa3sjyvUTW9Z7iDnxXEQR\n",
       "Rym3oiSGKCMrBGkSNgF2BPZDvoo/QEwtfqaTas96Ddn7k+qDNZGYLpCalzd3LPjFyDt6C3h2Faz5\n",
       "CqnksvV1+XWET0O+wA9Cxia9CVzt7l6rJZyKdOtsi/gnvAQ0csoai0SO+ABRGk/Z8y8A/o6Y6Loi\n",
       "rZWH7THeZ+NvSEy0jZHn4VErxyFIpfwOYvr7BNgX+Xh5BjFdLUDMe055vyB9QS2R5+g1xHzXLOOa\n",
       "62iuyAyS+/lTsKgxLL0POBx2/RHemwkrF8L09XwKeBtp8ayPOHhchDs9CMA/kGC7ByFK6UCkn01R\n",
       "glJlF8WyK+lmnqHUdlgYQrqCuQ/5Os6kCH1Ev28PA3NOdOerL34RIWYZaLEMqYDfQcK/QHof0a1I\n",
       "pQ/SYb8GaY2ujzvnUX/csEOAORAuXgGdDHxr4FcDfzBwmhOB4kHSzWTL7O/hyH/fyJa/GDdSuZcX\n",
       "cV+apohS3NVuNwG2tuvv4T4XKdxpMrKdf0Nc1/CzgZt8zl0g3fvBwGnp9/1Pv0HnlWAukI8XLkBM\n",
       "oO9myLud5xpnI2F5QOakmoM7l1U3zwmfRpSxotSHsu8jGoN8vVYiL+AxSMXj5QXEPFGBVKK7AJ9F\n",
       "KFMQniPR/UR+MfLAtlB+QpTJ+fbXG83hSsQsNBm5vq9s+raIW/N4ZH6lq8GsK+FkuBNaXQ/dvpO/\n",
       "pjfibzJ6mScChV+cvOcQpfIZ0oR6P8vFPIT0I41DnsWjkFbBBCuP42BwAhJrb4KV31tB+51/H5t3\n",
       "HNJ3lM2FvQD8guQOqYDFKyH1L08L2iAegl55D7b7fkUUz4d2+23E/OiYUs+z65/avPUNN6UoCvLl\n",
       "/QXitDDUpg2yi8OFiOfcJKTT249itohaglkKJqHTg0cWIw9xIzZnWtfi6+VegLQGzp4DZ30p5+ne\n",
       "r/7nKjUG1Pjf9wE1cUumKDkoiRZRqVBERQRgngNzcnTnrA9+JqIzptdfOZitwbxjx7j08Nl/BZir\n",
       "a6eXC1F+AChKZKgiCpFiK6KTRRkllW0PgssMHPsODPkZLhua/5hsmOaiYMx3YP4MJkvYHXMmmPvq\n",
       "fp5Sp3s/uHBF+B8AihIpqohCpNiKqAOYJa5pKmmYdiIfgDkEzJeiUAouZz8w08A8A2aDPHn7g3m1\n",
       "8HMA0lnft47HJgSzDry1HPq/Lua4cjVRKiWGKqIQKbIiAjBvgEmo04KpBPOVZ/tZMFcWcPza1jvw\n",
       "KzAH588PYHYEM74gMV1ORQadeqkkd1iioEMLvGUfRnr0jhokwkQImLPBPJI/n6IkipJQRBrjKjtJ\n",
       "jrLQDndAKoiTx9lkD0/kBB+dAD3ehZmfwdb7QMVSSJ2HjPMB8WL7DzImaAbiWj0M+AzWuhhwWk3L\n",
       "EZfoyYgrshOaqQa34u+EDPZsgsxSewziCTcAGW90A+KuPA7XC+5UxLPyTSQ00HqIZ5kzD5NfsFLv\n",
       "i3YYrst35r76cjzwWIjlKYpSYsTRIuoCZhGYJtGdu66YfcDUZKSdC+ZtagfQtMFHn+oFZhR8Ox46\n",
       "vo0bxeI03LFDD+FWtocgURa683vkggmrwDRGxiA5Y5Uux22RjEIGKIOriEBCDt3mkelaxFX5M3vO\n",
       "lYjiOROZBO5/cj5m4I4BOhiJmOCNxu2UfTvi6r0ICQU0DonMMIpQom6bjW0fWgKfBUXJibaISpvU\n",
       "PKQi3CtuSXxoh8Q18/IfZGDqqenJLQ6AwxfAgFeBF2DtnWHRVrgK5xHcCtrgDoCdjERMmGLTp8C0\n",
       "pUgYnzXAkz7HZyNzzqIDgLMQM9ouSISEVUhryth9PZExRX9CBon+gMQv3NGe+2JP2SCtuBeR4QA7\n",
       "IgoJZIzaLsBfbDl14TjgaUitquPxiqLkIMoQPw0BZ3Drm/kyFhkfRZRaDeaP8FYNHHIiNGsEFS2g\n",
       "zWbyNc8OkPLG/ss2782v9ncNbqBQu73iR1zznLcc56vrN9yPm3zOE39ColY75sSLEUWzDhJVwGEu\n",
       "Eo/uXlu+QQZIz8SfzOsabn/HIf1SBWJSyIDaQflyKopSN7RFlBsbBDVx88X4tYiAbTaAlw28sA88\n",
       "tTc83gtrt3ksAAAgAElEQVSWrYYHWkFqhc3UAYl0cKzdPgE31lkeflmMhOxphPT1gPSdvGPXZyMt\n",
       "GUgP1bQUmm7pBmrt0h6a/A1X0e2ABPhsbdN2sMuhSJ/PfbjTivRAlEKLLEJmmiLqG3V7WytXtugQ\n",
       "iqLUk6RVsAkj9TnSMd8zX84ik0URdR0MN7VLT3u8E2y+CAk+OgGZYO5cpG/oU0QRnec5wC9kjuWn\n",
       "H5AW0Qqk9TIJcWi4ymb4J2JWG4fMf2SP36QxtN4TvjkQBuwN0zaBDXZBIm5PRsIPgYQoWoqrxKqQ\n",
       "AKrjEAV4r00/tfa1AxJrrm2WfXXleOBxSK0JuVxFUUqMGJwVft9/HYmbQtzcCuYvtdOjDkNjrrRu\n",
       "4svyZk0ja6BW75QTFyARqSuRGGsTkP6py+z+Q5A+uzGIx91bNt3rCLG7PWYsrrOC13kimzkvC6aR\n",
       "dXHfNn9eRUkkJeGsUCrEqYh6gZka3fnrghkG5tTa6VGHoTGDwNxDrTmL8lGqcdpMbzCFTsGuKEmi\n",
       "JBSRmubyMwZoA2bLuAXxkMU0N/c2OHN6etrAGZ5I2fVlAWKaK9D8tewX//TlSZ+BVMcOKYryOzG2\n",
       "iADMHWDqEc8tbMxoMFX++7r3kxZQFGFoTE8wmVOQByCqQK1RYprYsUMbxy2JotSDkmgRlQpxK6L9\n",
       "wHwcnQyFYj4Fs30M5+0MZkHdju3eD87/Di5YWBpx2kw/MO/FLYWi1BNVRCEStyJqYqMsdIlOjkIw\n",
       "X0m8uaKftzGYVWSN0J33+I/ADM+fLwmYRyW+nKKUNCWhiLSPKBCpVcAryLiWJJCljyhqUr8hYXTW\n",
       "yZczC+3JP9A1AZhWwEGkD6xVFCUiVBEFJyFTiJsKZIBlgZ5roeE4LNSFdkCzEGWJikOADyD1bdyC\n",
       "KEo5oIooOCOBncF0iFmOtsCyGAdYzkcGlxaISSEtolJQRMcDj8YthKKUC6qIApP6CRlE2T9mQWIy\n",
       "y/1OXVtErZAQOwk3zZmOwJ7AC3FLoijlgiqiwkjCHEVJUER1aBHhhB5KeovoKOA1SBUYPUJRlLqi\n",
       "iqgwXgL2I94pxDMnxSs2dTTN0R6Z6iHpikgHsSpKkVFFVBCpxcAnyHw6cZGEFlFdTHPtgIUk2jRn\n",
       "ugLbAK/FLYmilBOqiAonbu+5uBVRfVpE35DsFtFxwLOQGoJE/X4HaR1dAAxEZoidADyDOw3FQ2Sd\n",
       "Xp0HPWUfgEwlMRZ4CukzA5lBdgoSCf3GSK5KUcqcPsgLPQ0Y4rO/ClgCjLfLZT55IPYBrWn5u8Y7\n",
       "hbg5H8zN8Zwb7PV/XYfjTgbzEpg4lWgezAS4bBDyLDZF3OS/BP6KzOPkcDVwjl1/kJzTq7MdEvl7\n",
       "NK7yGoJMsd4BeT8cwp7CQlFKYkBrlDO0VgB3AH9AZtj8BJnKOTOS9WjkBS4RUnPBzESmEI9j5ta4\n",
       "W0QLgbVlPFNqdQHHOS2ihJnmuveTeZzW7gCXbAH/3glxSvnVLi8hSmVb4BpgLURBec13ftOrY38r\n",
       "ga7IBH/O5HpN7foS4BfgfuBluyhK2RGlIuqFzKg5224/gUQmyFRE2aasTjJxTiHejoLn1QmT1K+2\n",
       "VdMJUUpBcfqImsqYolQCvtS694PdboV7N3XT3jwcprwOtQKDP4g8v5OQOZCqPPtyTK9OY2R22P8h\n",
       "jhCZ9AL2Q7z1zrHrilJWRNlH1BmY69meZ9O8GGQys0+BV5GvxlLAmUI8DiXannhbRFA3h4X2SHig\n",
       "1UBMZs1Mug5OV0IAd3aCRv2RvqzWuOPG2iCtnSbAiQQ3eRjgQ2APoJtNawVsZn/bIRMB/hUx4ylK\n",
       "2RFliyjIizoOMVv8BPRFTCKbZ8lb7VmvsUtMpKaCcaYQ/6TIJ4/bNAeuw8L4Ao5ph3T0r0TMc7/m\n",
       "zl4M2viYCXsC6y+G6RORFtwkxIR2OfAR8J39be05KMf06gB8j0xv/jius8bfkJluX0DuRwo4v65X\n",
       "oiiWKtJb62XPrqTb0Yfi77DgZRbpncIOCXJW+P2468BcG748ec/7Npi9i3/eNBkeADOwwGNeAHMY\n",
       "mO/BrB2NXIWSbUbbP7xuM7REPjRimHJDUUIhASbw/AQxzW2K28G8DzAYd5R8LsYg5odKpHP2GMRZ\n",
       "wcu6uH1Evez64gBlJ4G4oiwkoUVUl+gK7ZGBuCtJjAt3thltRzdBWntjEVftCcWXTVHKhyCmuWeB\n",
       "nRCFdDdiSngMyDex2W9I5+tIxIPufsRRYZDdfzfSQXuWzfsTcGxh4sfKJ0BbmUI89Xne3OGRBEU0\n",
       "H3FRLgRHbsc0lwCmvAp9t4SLr4TZY2Xq8jm3w6pX45ZMUZR0nH6Ai4FzM9KKRQJNcwDm38WfQtws\n",
       "BbNWcc9ZS4YjwDxf4DFzwWwI5jMwhSqxCDE3gamOWwpFiYgGY5r7FXE7PRl3nENCvJ5ix3rPFQvT\n",
       "GPG0ijsgZ12iK3hbRAkxzZlGwAB0AjxFiZUgiuh0xPHg74gzwcbAw1EKVUKMBjYDk+mWHpTBSCiY\n",
       "oPezLbA0xrmIHAp03zZNkKgCy0iUaY5dgSWQmpI3p6IosdMS2DLG8yfUNAdgHgZzdo4MKbIP2p1K\n",
       "QRX63M3AzAqePypMMzC/2hZFkPydxFsOwNSA2Scy0QrC3AzmirilUJQIaTCmuUOQPiHHFXsHanu/\n",
       "lTN+QVArgS+Q4JeTkDEoHyMDd6ttnruATZD7+hdE2T+AjFEZhxv26FTkfr8J2zwGy5bkyDccGRz5\n",
       "JfAPjzx9EA+wCcAbNq1VlnICkFqJxFTrGPAA7yDchJjm1CynKKXEOMS+73VQmFxkGZLcImplHQi8\n",
       "458qkQgCvYD9EQ9BEMX/MjIDKKSPm7oWOMGut0MUWUtEwcyVNLMvDPoqR74ZSASAZkhopc7A2sAc\n",
       "YCPPMbnOF/S6J4EJGAnA7AzGDvw1L8p4orgxe8g1KEqDpsG0iFZR21047j6KBJFagUwhflDGjq+Q\n",
       "VtCByBQAzriUzRFX+EwOAC6x+UYhymRD5EH6H/IftIO32ubI9yZuP8xniELcFXjbygPuf+l3vq4F\n",
       "XHghY4m8LaJfSESLiKOR6RgURYmZIOOIpiBfzo2RAaqDcaMIKwDcMgNm3QgDzoBlv8DkJ2DeCk+G\n",
       "64B7AhR0BDJlhpddAKesdrB6dY583oCbq5H/LNcXkV85QZlP8P4t76yyCTDN/W6W2zdeORRFgWAt\n",
       "onOQwYsrkVhZS5E+DQWQCM5TD4db14Wn9oYRB8JuV0CjNjbDSMTz0JkIzTGXZTISUfIOO9hfr6ND\n",
       "O+g5I0A+Byfg5l5I6whcU2C28wWlri2iJHjN7QF8V+SByIqiZCFfi6gx8AoS2ufS6MUpRboOhrs3\n",
       "Tk+7oRJeXmanEvgfsBUygyfAcqSF+R3prZWrgVuAicgHwkzEgcB48rWDu0fCU+vkyefle+CPiCND\n",
       "IySQ54E5zheU+QT3pHTC+0AyTHNqllOUEuNNgsWWi5IEOysMqPEPnHnSR6GJ+DvmNjCD8+crBuYo\n",
       "MMMD5r3ejUBh/gXmwujkyitLBZj5YLJFeVeUhkRJOCsE6SNagbgg/w+3r8KQbtYpY5b94p/eZXsw\n",
       "w4CrITXdP0/BJCHOnEOhprnZdj1u09wewLeQ+jJGGRRF8RBEEQ23i6NZU5SIli0Oc2+DM7ulT7A2\n",
       "cAbMHYpM9Pehjct2DaRm1/NkSVJEhTorJMVrTs1yipIwgiiih5CKwzFlfI64dCuARHAG6HcutG7h\n",
       "RnCe8qrV2bcBFwBjwTwN/B1Sc7OXl5MkKaIFwHoBp/329hGtJH1SuSJiKpCI73vmy6koSrKoQsag\n",
       "vG2X2UCxJ2ZLcB9R4PN0AvMPMIvB3A6m0Km2ATMRTI/wZasrZrFcV958H4PZxa7/xSrnGDBVYMbF\n",
       "c25FiYWSsF4Fcd++CRn8uJddDgBujlKohknqe0gNQTzofgUm2ykI1i2gEKdFVIn029WHDah/eJug\n",
       "/URJ8ZpTs5yiJJAgiqgxEv7F4UuCmfQUX1ILIXUBMjarMTAVzA3BWhahmubmI4M660NQReSVO6YB\n",
       "raYxcCQaW05REkcQRTQWuA8x0e1j18dEKFOZkFoAqcFADyQ+3Bdg/p4Rs86DaYzEgltuExoDjyCh\n",
       "fJ62+2bjDljtiYTuATGljrfLOGRwbSVuq+pUsgdMPQCJpDEWaU04A3Ovh649odN9wA02bYAtcwIy\n",
       "RQbSh1QrskIcXnN7AXMhVQQzrKIoYdMc6Wx3vOfOp/hftA2gjyivHJVg7pXpEkw1mIyxW6aj9MkA\n",
       "okTWALvZ7fuBC0kPoupVRC968rZEpm6vJF0R+QVM7YQolBY23xAkkngH4HPb5zUUmScJZHCs00Ky\n",
       "aaY1GEd5Ymd3fS7f3QgfcyeYIcU/r6LESoPpI6pARuAfYZfbbJoSKqnZkDoTiRlXCUwDcxmYNhJG\n",
       "6Ijn4LIW0Oc16FKFROR2ojU8AvTOUfh7SL/euUh/zWqfPNkCpm6NtIjGI7P0bggsAX6BXfaCm3pj\n",
       "Q0jY8wwDBuKab73hfSAW05ya5RQlyQRRRG/hfhGDfFG/kSWvUm9SMyB1KjLwcit4ay784b8wfE+4\n",
       "prkbyy7lNW+lkBbSb7j/qXf/P4AzkP/xPWALnxP7BUwFGci8g126A2fy+xQXe42Ct7bGnavqLOAy\n",
       "JIr3WKTl5HVUcM5TbNPc3sBXkJpZ5PMqihKAIIqoGW6/BMhXcwHz1ih1I/UlpE6A/0yGWzImoLuh\n",
       "Esw6SIsF4HjgXcSk1tOmHek5oBsSRf0G4BP8FVEmTsDUPezxIP1Dm9nfdnDjCHhoIeDMS9QNmfri\n",
       "CiSWXhdqO1jE4TWn3nKKkmCCeL/9BOyEfOGCVHQ/Z8+uhEuj32qnpYCWP8FPZyOzrE4B/oMogfuR\n",
       "COk1uPbh8xBHkzXIpIYjkD4gZ3+ugKmnIlHXHeXxN+Rj5AVo3hY27oa0tkAU3WZWwDeQPqNDqd0i\n",
       "KqIiMo0Rk/LOxTunoihhszPSkf2uXabjfnUXizJwVshGn9f8g6r2HRG3ZGBagvnFesZly3MKmP96\n",
       "tnuAKeIMv2Z/MBEEoFWUkqDBOCtsjPQPnIX0F3xB8Bla+yAhgaYhHlfZ2Bnp3zgiYLllxNzb4MyM\n",
       "oKkDZ0gYobhJ/YSY2trnyJTZR1Rs05ya5RSlAeC4+PZGzD39gSBfmBVI66kSaIKMLdkqS763gJdJ\n",
       "79fwUsYtIhCvub4jZMqJviNkOymYqWC659hfDeZKz/ZGYObU8WQ7AbfmybMd0NeeqwmY7+ScilKW\n",
       "lESLKEgfkePq2x+4F1EYVwc4rheiiGbb7SeQ/oKpGfnOBZ5Bbfg5mPKqG1w1cTjRFaZk2d8eGd/k\n",
       "UB+vubG4fZXZ2AFRWCOQfrEZkPqqjudTFKUIBDHNfQ3cAxyDzNbaPOBxnZGxLg7zbFpmnkOBO+12\n",
       "SWhvJY1800F4veZaQbNh0KMj0tI+GtgPifYwEXG0aGrz7oy4mk9AWuCtkegeL7ll8YDdNw6ZXbYJ\n",
       "cBXyrI6Dfw6BjpsgA3NBnttpQIYXoqIocRKkRXQ00tdzI1KhrA9cFOC4IErlFuASmzdll2xUe9Zr\n",
       "7KLET754c94+oj5g5sHEVZDaFlgLUUj7Iq3nYUhf5J1IC/popAXUmtqemn9DBuGejii7jxBPvcuB\n",
       "ncBcILLdMgyZmv1W4A+IYltUnwtWlARTZRfFsivuQEeAodR2WJiJmG1mIS7BC5Ev20zKvI8oyZjz\n",
       "wWTpt+neD85fDKeNF++/zgOBWTDEQEVvpD9ntOeAfYFngW0QD81MqnBbRGMQJebE0JsNbIm4m98O\n",
       "pg+Y95GxTI457wkgQf1rihI5JWFlijKK9hhkTEklYr45BjguI88mnvUHkUrmxQhlUsJnAW4cOw/d\n",
       "+8Fut8JN7fndq+7MbvDWxdD9cQnwypsZB+VqEftxBGJq82LnPfrdW24e8oGzL2Luy3wGFUWJmSB9\n",
       "PXXlN+AcYCQSu+xJxFFhkF1Kmb+QHvaonMlimus6OH36dICrNoVuA+Gkn6HzHYgC2wg3csNJiMn1\n",
       "C1umM16tDbXjG44EBnu2d7C/y6BiLaTv8Rmbdh8Sj+8pSuQLUVGU5JEk01wFYkrUDm8AzGb+929A\n",
       "Te1BuCMNtF4O2/4GTcYCOyItFcdZ4T7E4QBECX2A9Om8jzgn7I3bYm4O3GWPm+xJbw8dvoAtliOt\n",
       "ImyZS3Cnu1eUckE/vEKkmIroOcSsOBkJ8AkSa++fSKV4OeKCPBExLTUCHkL6KyYiraUywrQG81Pt\n",
       "6Aq5IkKYeWC6RijTg2DO8yT0JL0vSlHKBVVEIVJMReRECWiBKJcOSCSJozx5vPP+7AS87tm3VkSC\n",
       "JhiztPb8Sd37wcBp6UrojOmSbmaA2dS/rHrL0lTmbTLOUIFLEEeG3aM5n6IkGlVEIRLRzezeDw5/\n",
       "Gy79Wb7gu/dD3MQn2OUHpPN7Fekd6V5F1A5xPb4NOJDCO9wbAOYLMD5RM7r3g0t/g2PeSY8IYT4D\n",
       "s01EshwE5p1oylaUkkMVUYhEcDP9vtgPnAcVk3FH/o9C+iWWZRzsVUQg02IcgZj17g9f1qRjasDs\n",
       "m2XfMpncLy1tHJidIpJlGJhzoylbUUoOVUQhEsHN9OvDeMFA+4U2w1bIIEo/RTQRcUsHcVpwpsre\n",
       "BhnTUmaYx8CcmGWfnyL6AEwEpjLTDMwPYHJFelCUcqIkFFGU44gSThufeGd9gEaNEXfzL3Cn4s78\n",
       "M+9BBut+DZyPjIFyXOEvCV/WxJMvukImUc3SegAwEVLzIyhbUZSIKGNFtOyX2mlNgV4fw4i+GTva\n",
       "ZmzfYReHiMxMJcN8JIJBUKKaCmIA8HQE5SqKEiFRDmhNOEme56fkqEuLKGRFZJoDByMhghRFKSHK\n",
       "uEXkTKvQ71xo3QKW/yxKKLHTLSSZJJjmDgA+hdSCkMtVFCViylgRQcLn+Skl8k0FkUkUpjmdiVVR\n",
       "SpQyNs0pIWJbRJnRFbISsmnOtEAmbhweXpmKohQLVURKGDju7W1y5nIJ2zR3IDAOUt+EWKaiKEVC\n",
       "FZESAilDYf1EYZvm1CynKCWMKiIlLApRRCGa5kwLZLI7NcspSomiikgJi0IcFsI0zfUBxkLq25DK\n",
       "UxSlyKgiUsIiLtOcmuUUpcRRRaSERaEtohAUkWkJ9EXNcopS0qgiUsKi0D6iMExzfYFPIPVdCGUp\n",
       "ihITqoiUsIjDNKdmOUVpAKgiUsKiyKY50xJxVHiufuUoihI3qoiUsCi2aa4f8BGkvq9nOYqixIwq\n",
       "IiUslgBNwLQOkDcM05ya5RSlgaCKSAmJlEHMc0FaRfU0zZlWSFgfNcspSgMgakXUB/gcmAYM8dl/\n",
       "KPApMr32WGDfiOVRoiWoea6+prmDgA8gtageZSiKUgZUANOBSqAJMAHYKiNPK8/6tja/HyUx77pi\n",
       "ngJzbEbaMjAZwVBNTzBj63GeZ8CcXvfjFaVsKIm6M8oWUS9EscwGVgFPIC0gLys8660B7XgubQpp\n",
       "EdXRNGdaA/sDz9fteEVRkkaUiqgzMNezPc+mZXIYMBUYAQyOUB4lehYQzIW7Pqa5/sB7kFpcx+MV\n",
       "RUkYUc7QGrRJ+Lxd9gQeBrbIkq/as15jFyVZzAe2CZCvPl5zA4Cn63isojR0quyiWHYFXvNsD8Xf\n",
       "YcHLDKCjT3pJ2DkVsz+YNzPS/PqI1gVTh2jZpg2YJWDa111GRSkryr7ubIwolkqgKf7OCt0AZ3rp\n",
       "HW1+P8r+ZpYGpjuYqRlpfoqoHZildSj/ODCv1F0+RSk7SqLujNI09xtwDjAS8aC7H+kLGmT33w0c\n",
       "CZyMODMsB46tXYxSQgR1VqiraU4HsSqKEhslodUVkwLzi40D56T5tYgagTGSP3DZba1Zrl04sipK\n",
       "WVASdadGVlBCJGUI1CpKrUFawYW0ig4G3obUj3WVTlGUZKKKSAmbqMxzapZTlAaKKiIlbIJOB1HA\n",
       "oFbTFnFJfbGuQimKklxUESlhE0W8uUOA0ZBaUmepFEVJLKqIlLAJGl2hENOcmuUUpQGjikgJm5Cn\n",
       "gjDtgL1Rs5yiNFhUESlhE7Zp7hBgFKTqMABWUZRSQBWREjZBnRWCmubULKcoDRxVRErYhDgVhGmP\n",
       "BMN9qd5SKYqSWFQRKWGzCGgFpkWefEFMc4cCb0FqWSiSKYqSSFQRKSGTMsA3wHp5MgYxzQ1AzXKK\n",
       "0uBRRaREQRAX7jymOdMe6A28HJpUiqIkElVEShQEceHOZ5o7DHhDzXKK0vBRRaREQRCHhXymOfWW\n",
       "U5QyQRWREgX1NM2ZjsDugE6CpyhlgCoiJQrqa5o7DHgdUstDlUpRlESiikiJgiAtolymuaOBp0OV\n",
       "SFGUxKKKSImCoC0iH0VkOgK7omY5RSkbVBEpURDEWSGbae5wYCSkVoQulaIoiUQVkRIF3wNtweTy\n",
       "istmmlNvOUUpM1QRKRGQWgMsJHd0BR/TnFkb2AV4NSrJFEVJHqqIlKjI57DgZ5o7HBgBqZ8ik0pR\n",
       "lMShikiJinwOC36mOTXLKUoZUgxF1Af4HJgGDPHZfwLwKTAReA/oUQSZlOjJ57CQYZoz6wA9gRGR\n",
       "SqUoSuJoHHH5FcAdwB+Ar4FPkCmfp3ryzAT2ApYgSusexH1XKW0KNc05ZrmfI5VKUZTEEXWLqBcw\n",
       "HZgNrAKeQOaY8fIBooQAPgK6RCyTUhwKNc2pWU5RypSoFVFnYK5ne55Ny8YZqMdUQyFIi8gqIrMu\n",
       "sBPwWuRSKYqSOKI2zZkC8u4DnA7skWV/tWe9xi5KcsnXIvKa5o4AXlGznKLUmyq7KB52Jf0rdyj+\n",
       "Dgs9EBPeplnKKUShKYnArAfmWzDLwLTx2b8NmCl2fRQMPxOYVFQRFaXho3Un0uKaAVQCTYEJwFYZ\n",
       "eTZElFAuBwW9mSWHqQCzCszKLIpoMzDTrcL6AbbdAlVEihI2Wnda+gJfIMpmqE0bZBeA+4BFwHi7\n",
       "fOxTht7MksR8DcZkUUQbgpkH5mwwDyMfK58hXpOTgZGI6W574EPExX840M4WUAPchHhiTgV2Bp4D\n",
       "vgSu9pzoRMQJZjxwFzp2TikvtO4MEb2ZJYkZk0MRrQPmOzA1YA5BFNEq3HFkT+KOMdvTpl0J3GzX\n",
       "RwHX2fXBSJ/UukjLey7QHml9v4gMIwD4D3BSONemKCVBSdSdUTsrKOXNfMQbzocD94aeHaFiDxj7\n",
       "K3R5DObNQgY2A4wFuiEtoHds2jDS5yl60f5OtstCuz0TMfnuac8/xqa3AL6p5zUpihIyqoiUKFng\n",
       "n9y9H2x0Hfw9hTyD+8PRm8GzKVjjZFqNa4ZzSGVsrwT2Bra06w5rcJ/tYcCldb0ARVGiR+3lSpRk\n",
       "UURdB8M93dLTbqiEZh0zMi4BFgO97fZJpLvtVyBu/9v4nMQAbwJHAWvbtA5IS0lRFKVgSsLOqWRi\n",
       "/pjRR9QKeAVaL4dtDDxpYCMDFxvYwkCj1Yg5DuAaxMT2BfAj4sgwHHgUcTpYgiiaBcC3drs3MABY\n",
       "bo8bjURsGI/0NY1Bon0oSrmgdWeI6M0sSczBGYroSOAe6POapC8xUGngWiPbW34KvGTzvoTrWHAa\n",
       "4hEH8BDSN+SY6a4A/uo56UTcgbRtw78mRSkpSqLuVNOcEiF/2RguA459VZRP5/bA/lCzFPrNc/XE\n",
       "ccDAGUjm3WzirsBjdv0RXPOcQRwWvC+Yt+/oPaRfaCDaB6ooSoiUhFZXvHTvB4Nm2RaRXQZOg00G\n",
       "ACdAxUTY8Eto/jP0rpH8NAG+swV8h6tIvOkPIi0rhyuACzJO3gtx9Z6F9AspSrmidWeI6M0sORzz\n",
       "m3eZb2D/kTZDf8TcNgs37NOJwAt2/QW7DXAq8Kxdz1REfyU9DqHXCeJjdH4rpbzRujNE9GaWHANq\n",
       "aiuikUYcFRiPRDvYCVFE1yPOBB8Bm9gCNkScET4F/oc7PciDSJBUh81snnGI+e5ZpJ9oEu7gV0Up\n",
       "V7TuDBG9mSWHX4vIGOibOQOrms8UJTpKou5UZwUlIubeBmdOT08bOAPm3J6RsSReFEVRoiNzpHpS\n",
       "MZSOrMrvdO8HG54LrVvA8p9FCU3RiQ8VpXho3Rki+tWsKIpSOCVRd6ppTlEURYkVVUSKoihKrKgi\n",
       "UhRFUWJFFZGiKIoSK6qIFEVRlFhRRaQoiqLEiioiRVEUJVZUESmKoiixoopIURRFiZViKKI+wOfA\n",
       "NNxw/162BD4AfqH2vDKKoiiKUi8qgOlAJTK52QRgq4w8awM9gWvIrohKIkwFUBW3AAGpiluAgFTF\n",
       "LUAAquIWICBVcQsQkKq4BQhIVdwCBKQk6s6oW0S9EEU0G1gFPAEcmpHnO2CM3V/qVMUtQECq4hYg\n",
       "IFVxCxCAqrgFCEhV3AIEpCpuAQJSFbcADYmoFVFnYK5ne55NUxRFURQgekVUEs1CRVEUJT6inqdi\n",
       "V6AacVgAGAqsAf7hk/cKYDnwL59904FuEcinKIrSkJkBbBq3EPloHHH5Y4DNEGeF+cAxwHFZ8uZS\n",
       "iom/kYqiKEpy6Qt8gbRqhtq0QXYBWA/pR1oC/ADMAVoXWUZFURRFURRFUZR4yDfgFeA2u/9TYAeb\n",
       "1hUYBUwBJgODEyijQwUwHngpKgEt9ZGzHfAMMBX4DOnfi4r6yDkU+c8nAY8BzaITs16DsYNcY1jU\n",
       "Vc4kvUP5BrYn5R3KJWeS3qFcchbzHSoJggx47Qe8atd3AT606+sB29v11oj5L/PYuGV0+CvwKPBi\n",
       "BPI51FfOYcDpdr0xsFYC5awEZuK+OE8Cp8QoZ7bB2EGOTYKcSXqH8g1sT8o7lEvOJL1D2eSspHjv\n",
       "UGDijjUXZMDrIcgfDPAR8tWxLvAN8geAeNtNBTZImIwAXZCK9T6i9VKsj5xrAXsCD9h9vyF9dkmT\n",
       "c6k9piXyorcEvo5RzmyDsYMcmwQ5k/QO5RrYnqR3KJucSXuHsslZzHcoMHEroiADXv3ydMnIU4mY\n",
       "bz4KWb5s5w8io5PnZuAixG09SupzLzdGHtwHgXHAvcgDmiQ5OwOLEff+OYgX5o/AGzHKGcWxhRLW\n",
       "uZvrtLYAAAQLSURBVCqJ9x3KRZLeoWwk7R3KRjHfocDErYiCDnjN/AryHtcascueh3zVhU1dZUwB\n",
       "/YFvEdt21GO26nMvGwM7Av+xvyuAS8ITrdb5guB3v7oBf0EqzQ2Q//6EcMSqRX0GYxdzIHcY50rK\n",
       "O+RHEt8hP5L4DvlRzHcoMHEroq+RDlOHroh2z5WnC25TsgnwLPAI8HwCZdwdMTPNAh4H9gX+m0A5\n",
       "59nlE5v+DPIyRUF95OwJvA8sQkwfw5F7HJecURxbKPU9V1LeoWwk7R3KRtLeoWwU8x0qGRojI38r\n",
       "gabk77jeFbfjOoU8kDcnWEYvexOtx0995Xwb2NyuV+Mf/SJuObdHvLtaIP//MODsGOV0qCa9Q7iQ\n",
       "Y+tLfeRM0jvkUE32KPxJeIccqqktZ5LeIYdq0uXcjuK9QyVFvgGvAHfY/Z/ifmX0RmzGE5Bm+3jc\n",
       "UEJJkdHL3kTr8QP1k3M75GvuU+QrKSqPn/rKeTGu6+kw5Is+LjlzDcb2OzZpcibpHQoysD0J71Au\n",
       "OZP0DuWSs5jvkKIoiqIoiqIoiqIoiqIoiqIoiqIoiqIoiqIoiqIoiqIoipIcKuIWQFEaIIciUUu+\n",
       "t9tXIu/arNgkUhRFURJHVOGtGgMPAUdGVL6iKIpSAlQik4Y9gkxQ9jQS0mQ2cD0wFjgGOA6YiIww\n",
       "v95z/HLgJiQUyhtAJ5u+PRJuyBk5386m1yBhcj4BLkXieM1EojBvQrpi2s+mTwTuR0K0YGWrtrJN\n",
       "BLaw6XvjRj0YR+1oA4qiKEoCqURC1+xmt+8HLkRMYxfatA2Ar4COiNnsTdw5XdYgSgrgcuB2uz4R\n",
       "mXMGxNzmxGgbhYQkcngQOMJnuzkSamVTmz4MiXiNlc2J+XUWMo0ASEgb5zpaouZ0pQESd/RtRYmK\n",
       "uchUySAto952/Un7uzOiQBYBq5HZP/ey+9Z48jnHtkVih71j04d58nvLdfCbFmQLROFMz1LGcPs7\n",
       "DlGmAO8hCu9coL2VVVEaFKqIlIaKd86WFO6kais8+1MZefzmecmV7mVFxrbfMZlpmWWvtL+rkb4m\n",
       "kAjOZyCmxfdwTXaK0mBQRaQ0VDZEppAAOB54N2P/J0j/i2OaOxYYbfc1AgZ4jn0HmWL5B9yW1UlI\n",
       "35CDVzEtQ1pQXgwSLbkSmZzMKWM0uemGREq+wcqsikhpcKgiUhoqXyB9Lp8hJrU7M/YvQGbQHIVM\n",
       "gzAGd66bFUAvxImhCrjKpp8C3Ig4K/TwpEN6y+YJZGrrsYizgsNK4DTEeWIiMjHZXT7He7fPs3J8\n",
       "CvwKjMhxzYqiKEpCqEQq77qyLCQ5FEUJgLaIlIaKXx9NMY5VFEVRFEVRFEVRFEVRFEVRFEVRFEVR\n",
       "FEVRFEVRFEVRFEVRFEVJIv8PxIPplWxYk5MAAAAASUVORK5CYII=\n"
      ],
      "text/plain": [
       "<matplotlib.figure.Figure at 0x108284690>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "analysis.plot_proportion_investigation(predicted, dmoz_encoder, y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Which things get mistaken for other things?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Library/Python/2.7/site-packages/IPython/kernel/__main__.py:2: RuntimeWarning: divide by zero encountered in log\n",
      "  from IPython.kernel.zmq import kernelapp as app\n"
     ]
    },
    {
     "data": {
      "image/png": [
       "iVBORw0KGgoAAAANSUhEUgAAAUcAAAEoCAYAAADL6zPkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\n",
       "AAALEgAACxIB0t1+/AAAIABJREFUeJztnXecFOX9x993iCKgKIq9XEREVFREQLBwaord2GOMSmKL\n",
       "iSWJpqrxbD81ligaewFjN3YjdimKIHBUKdaTxBYLKIIV7/fH5xlmdm5mdmbZ3du7/b7vNa+d8uyz\n",
       "z87tfvf7PN8GhmEYhmEYhmEYhmEYhmEYhmEYhmEYhmEYhmEYRpWzMvAosBC4Zzn6OQJ4sigjan12\n",
       "Bua29iAMw0jHT4HJwCLgXeBxYMci9HskMBGoLUJfbYHvgE1aexBGOqrlQ2kUzu+AvwPnA2sBGwL/\n",
       "APYrQt8bA68ioVEt1CRcW6FsozAMY7nohrTFgxLarARcAbzjtr8DK7pr9cB/kYD9AGmdw9y1c4Cv\n",
       "gK/da/wCaAD+Gei7DglO70d8GPAG8BnwJtJovfPjAs8bAkxC0/WXgcGBa6OBc4EXXD9PAmvEvDdv\n",
       "/L8H/ufG/2NgLyTUPwb+FGg/EHgJWODaXgV0dNfGuvfyuXu/hwT6/wPwHjDSnfuPe05P9xr93PF6\n",
       "wIfALjHjNQyjTOwBfEPyDONcYDywpttedOdAX/RvkNDrAOwJLEZCF+Bs4LZAX2cTLxy7AJ8Cvdy1\n",
       "tYEt3P4wfOHYHQmnI9zzfgJ8Aqzuro8GXgM2BToBzwMXxrw3b/xnuvEfC3wE3OHGswWwBGnAANsh\n",
       "AVnrzs0GTg30F55We/1fiIRoJ3KFI+41X0Hrs08Cf4sZq2EYZeQIpNEk8ToSoh4/BN5y+/VIeASF\n",
       "6wdIgEBLTTF8XEeucFwAHIgERZBh+MLxSGBC6Pp44Gi3/zzwl8C1E4FR4TcVGr83FV7FjWdAoM1k\n",
       "YP+Y5/8GeCBwHCUcv8LXtL1zQeEI8DAwE5iGr4kaJcbWHI0kPkbaYNLnZD3g7cDxfHcu2EdwTXEJ\n",
       "0LWAsSwGDgN+iaasjwG9Y8YzP3Tu7dCY3g/sf5FnPB8DzYG2IAEffH4Xt7+ZG9d7SMu9gPgpu8eH\n",
       "aGkhiZuALdE0/Zs8bY0iYcLRSOIlpNkckNDmXaTheWzkzhXC50DnwPE6oetPIc10HeTucmNEH+/g\n",
       "T3M9NnbnS821aCq9KVo6OIP837HmPNe7ojXdm9A67erJzY1iYcLRSOJT4K/IOr0/Elwd0drhxa7N\n",
       "XWhNzltz/Cu5U+MsTEPGhg2RcPlz4NpabgxdkPa0GFga0ccopMEdjqy/hwGbI43OI8livDx0RcaW\n",
       "Je41Twxd/wAZWbJwJTIqHQ/8G7huOcdopMSEo5GPy5G1+UxksZ0P/Ap40F0/H627zXDbZHfOI0kz\n",
       "ag5dfwY5g89A1uZHA9drgd8iDfBj5Cx9YkQ/HwP7AKch48np7viTmDGFxxA1xqTjIKcjC/pnwA3A\n",
       "3aH2DcgivQA4OOG1vXP7I03Ze5+/Q0afwxPGYBiGYRiGYRiGYRiGYRiGYRiGYRiGkZFVN9nWs17a\n",
       "Zlub2uq2HthMseiwUprX/CT2+UWiVP5ehvgNcD1+ZEU+mne+bFyLk28/eQsb/+gXLc7fcHi/Fueu\n",
       "vvQCTjr9jBbnFyyODqy4afhFHHvKn1qcH9Jr9bjPRoPb0hDb9otvmlt8mc4/t4Ez/9qy+codazKN\n",
       "ZfxrCyK/qHHvtV/dai3OxY0F4N5p4eg+ePCGyzng+N+1OH/0gI0yjf3hGe+1GPtd11zK4b86PbKT\n",
       "/bdeN6r/yL6j7jnEv9dLR7/e4txztw1nt6NOaXH+rB/0guLJk+ZO/U5ObPDl1KuK+XqRmJ9j6eiA\n",
       "kg50ztfQMIwQtR2St3IMoSyv0j55EDk8zwKOc+c+By5FkR5/QfG8zwPPons9AiUQmIG0SsMwoqip\n",
       "Sd7KgCXXLJxfoEiHlVF41/1IS5yAIiW8NvVofaQ/EpZ93bVupKRbz5bT5zgGDtk5dVuA7QbtlKk9\n",
       "SvlVirbsMrQ+S/PM/Wd5r1nHsnn/wfkb5TI6bcOtBgwpWd+Q7b1+b5tBGYdSIGXSDhOH0NoDaMOc\n",
       "ijTEl4ANUJ7BpUhIRvEGSlc1HPgRCjFLxWqbZhGO2fKgmnCMJutY+pRQOPatRuFYU5u8lQHTHAuj\n",
       "Htgd2AH4Ek2dO7n9OKvdQmBrlPvwl8ChwDHhRm8/ecuy/W49+2USjIZRLt6aPpG3pk8s3QtUgOZo\n",
       "wrEwVkVT6i+BPkhIRrHItf0E5fX7BiU/fZWYzDVRVmnDqDS+t82gHC3y+X9eVdwXKNO6YhImHAvj\n",
       "CaT9zQbmoak1tNQab3Bt30EZZW7FX8po6VNiGIYo09Q5CROOhfE1KrIUZtXQ8dVu8+hfshEZRnvC\n",
       "ptWGYRgRmOZohBl32s6pF1v2/2BU6pCtMw7ZsrABpWTOu59nCh+b2rSwVEOJjHgpJglRL8vNZmuu\n",
       "UqqukyKNIhk+7s3ihQRmpUPra46tL54NwzDCFO4E3gGYirLIRzEcleadjl8PPJL2KBzrUBTK8rAe\n",
       "cN/yD8UwjIIoPHzwVGQojdJ690LFz3qhmjzXJg6hwKG3d94FDmntQRhG1VKYE/gGSADeRHRSiv1Q\n",
       "DR+AicBqwNpxnbVX4bgCcDv6BbkPhfU1Ad3d9e2R4zbAUKSGTwUaUXW7OnztcxjyTRyF/BO9qnug\n",
       "4kfjgSnAvfj1iy8CXkGq+9/cuUPwC7OPWf63aBjtmMI0x78Dvye3TnqQ9YFgSqX/IoEaSXs1yPRG\n",
       "cc0vATejanlxi8unuesvISH6VUSbbYBtkQvPPLRu8RWqS7w7Skn2R1Qd7h/Aj1FpTvDde85CwvQ9\n",
       "Wrr8GIYRJLSuuPTj1/nuk5Yp1ALsg6pjTkURbLE9h45jjU7tVTj+B98x+3a0DhHHi+gX5w6kIUYV\n",
       "f38WRbuAtNE6VFx9C6Q5Aqzo9j9FkTM3o1rJXr3kF5FKf697nTgaAvujyRgnaxjl4LWpE3ht6oTS\n",
       "vUBIO+zQozcdevRedrz0jafCzxiCps17oVDeVYHbgKMCbd5BNdE9NiD6+w60X+EY/DWoQWr2t/jL\n",
       "CJ0C1y9GAmxvJMB+REvtMXi8FP++PY3qFIcZiDTKg4GT3P6J7vzeaBren+hsxg2x78owKoRe/Xag\n",
       "Vz8/avaJW4cX9wWy+zn+xW2gpbLTyRWMAI+g7+PdKOR3IfBBXIftVThuhN78BCS8XgBWQWuNTwAH\n",
       "Bdr2ROuDrwAD0JR8Rp7+m13f/3DPfwOtN66HjDld0BrleHfNe52X3bYn+tUqeap3w2iTLH+EjKcg\n",
       "neAerwceR5rl68Bi4OdJHbRH4diM1gV/DdyChN41SCjdjFKFjca/eacCuyLtchYSausHrjcTvS7x\n",
       "ETLW3AWs5M6dgabfDyPttAbFVIMMM73cuWfIL4ANo3pZvgiZMfhGz+tD105K20l7FI5vo0w5YV5A\n",
       "WmGYlgUxZNne2u2PxDf/A+wb2H8eTZXDRCW9OyjinGEYUVhWHiNM1jC8tAw96/FM7eOKMRWLuh6l\n",
       "7L20ZPkf9Vmva6Zv+cvvZFtpyfJ/yho++OmXSzONpahY4gnDMIyW1Jjm2CbZBhleRrX2QAyjvVJT\n",
       "2/rCsb1GyJSSfkTnckzCfoQMIwM1NTWJWzloi8LxKBSWNw0ZSjYGnnPnnsF38hyBrNQvIXeaetd+\n",
       "NsrI7fE5cDmyVD8DrOnOj8ZPTrsm8BbQETgXOAx54h+C3HZuQbGajcgRFWTJfgQ5kD8NrAOMdc+b\n",
       "CWSubGUY1UJtbW3iVpYxlOVViseWyF1mVxTO9xuUaftWNN29A4X2eawGDEbuNI8gd5otUXlUzxrd\n",
       "GZgEbIXM/2e781EuPN+gMMC7kQZ5nxvPs8hCvRtwiesT1+YgN94jkI9lP/fa0wq8B4bR7qkEzbGt\n",
       "Tfd2Q+F3nklvAXL2/rE7vh0/0UMzfk63WcD7yOcR91iHfA2/A+4JPD8ptA/kpxj87/wQufd4tapX\n",
       "Qk7ozUhj9LK6vow0zI7AQ0jTNQwjgkpYc2xrwrGZ6FREcXfya/f4HbkhgN8R/d5r8LXFuHDDKA5E\n",
       "CTSDDEJe+B7jgJ1RgPwINJVvUYHw6ksvWLY/cMjOmetQG0Y5KHVpVrNWZ+c54EEkWD5BKcjGAz9B\n",
       "Wt8RaF0vC7Vo7fAeFGo4zp1vQuGGk1GMtMdnKBTR40nkSH6yO+6H1hXD/92NUJD7TUi77EeEcDzp\n",
       "9DMyDt8wyk+pS7OWa10xcQytPYCMzAYuQGuD04BLkVD6OZqmHkFuBp7mmP0gi1GUy0xktDnXnb8U\n",
       "JYtoRDWnvec/j7LxeAaZ89BUeQaavp8TeL3ga9a7MTcChwJXpni/hlGd1OTZykBb0xxBaYhuC53b\n",
       "PaJdMKi8Cd8AE74GyukYZh4y8nic5R4X0DJk8JcRzw+HHUaN2zCMCArQHDshpWkllD7wYeDPoTb1\n",
       "7vyb7vh+4Py4DtuicCw2rVdhzTCMSApYc/wSeYUsQXLtBeQu90Ko3Rh8d7tETDhWWFbuuh5d8jdy\n",
       "ZI2VzcKwnRsz/WjcM2y71l9BL5Cs97GUcedZy74eXaqBAFuv3bWEveehsE/TEve4IqpCGBWonrrn\n",
       "trbmaBhGFVCgE3gtWtf/ANkGZoeuN6OM4dNRbsctksZgmqNhGBVHeFr91buz+Pq9V2JaL+M7FBzS\n",
       "DXmR1JNbZqQRRdAtQQmnHwI2i+vMhKNhGBVH2Am80wZ96bRB32XHixsTy8p/CvwbueKNDpxfFNgf\n",
       "hcKLuxOTkb/ap9VnAXORb+OdyGp9LIpmmQb8C1jZtR1BuljtNOVaLynN2zGM9kEB4YNronBh0Hf2\n",
       "B8jdLsja+GuOA91+bALNatYcB6DIlq3RAm4jcvh+ADlqg3wYj0Hx2834sdr7oVjtwUg4TkJuP++Q\n",
       "rVyrYRgRFODKsy5SVmrd9k+U8yBYQ+Zg5Lv8LZpa/ySpw2oWjjuiNYev3fYo+iXpi3yfugFdUbII\n",
       "j3yx2huSrVyrYRhRZLdWzwS2izgfrCHzD7elopqFY1yc9q3A/uhmH01ugfB8sdpLyVautQXnn9uw\n",
       "bH+XofXsMrQ+qplhtCozJ41n1qTx+RsWSCWED1azcHwR/apciML/9gFuQHHT77tzPwP+k7K/Qsq1\n",
       "tuDMvzZkfiOGUW76DhhC3wFDlh3ffd1lRe3fEk+0LpPRuuEM5Bc1E01/z0KJaz90j0FP2Hyx2lnL\n",
       "tRqGEYGlLGt9LkWJIjqjsKLJyEp9XUTbtLHaWcq1GoYRgWmOrc8NyIDSCbnqWHZuw6gATDi2Pke0\n",
       "9gDCZInzLWX95Kyx0kfePi1TvPENh22Tv5Gj6cPF+RsFKHWsdClj2rNSys/Awq++yT6gIlFr02rD\n",
       "MIyWVMLPT+vbyyuDOmSQWV6GAV5K5B8DfQLXRuNXMzQMI4Ha2prErSxjKMurVA/BKc6Pyc36YXkj\n",
       "DSMlJhwriw7IQDMLZfTohPwVRyEr9ligt2u7L/JpbERO32uF+hrs2lzi2mzizh+C3IPmYXWrDSMW\n",
       "E46VRS8UQ70VKqd6EHISPxll9/g9SjwBSlSxAwpXugf4gzvv/ddeQj6Up7s2Xlr2Dsil5zf49bEN\n",
       "wwhRU5O8lQMzyPi8hRzCQRl16lBizGBupBXd44Yo48467tybRBP+N3o1sRtd/1E0BPZHk5tyyTAq\n",
       "gjlTXmLulJdK1n+JasgADEe5HJcgG0E4c88yTDj6BGOll6L0RgtRCdUwVyEH8seAoeQKtCDhdUbv\n",
       "NZYSf+/j+jKMiqFP/8H06T942fFDN15R1P4L0A7T1JDZC9gUzRIHAdeiGWAkNq2O5zOkEXo1q2vw\n",
       "o2JWRfHSoF+fKBZhqckMoyAKXHPMV0NmP/yKoBNRCsK1Y8dQ+PDbHWEtrxklnjgGRc7Mwq9a1oCm\n",
       "25NRDHZz4Dne/t1onXIKvkEm6fUMw3AUkOwW8teQWZ/cRDL/BTaI68ym1aKJ3FjpYIqRPSPaP+K2\n",
       "MMFa1eOBLQPXdg3sf0S0wDQMg5YRMp+9OY1Fb+WN7s1XQwZa2gFilRQTjoZhVBxh5bBbz23p1nPb\n",
       "ZcfvPT+SBOJqyLyDjKkeG7hzkZhwbMMsWFy62NdSxxsv+mJS6v4v3LtP/kYBSj32UsYzl5Is4wZ4\n",
       "+Z3Y8iolp4DEE2ui8gcL8WvInBNq8whKNH03MsQsRFPwSEw4GoZRcRTg6J2mhszjyGL9OrCY3FSD\n",
       "LTDh2JI6VCumb552hmGUiAJcedLUkAFpjqkw4WgYRsVRCSnLzJUnmqg4621RPPV0FOni1cgdDVyO\n",
       "yrPOQSVfHwReRaVdPX6GfKumokzjdu8NI4YCXXmKin1Bo4mKsx6J/Ba3QSq8FxvdjCJfBiCP+4eB\n",
       "X7rnDgNWR6nLDkXhiP2Qy0HFJdo1jEqhEhJP2LQ6mnCcdU+kKY5z50aSG3Pt+TzOcptnAXsT2AjY\n",
       "GeVynOzOr4wqHEbRENgfjcVWGxVIqWOrK71MwlUJ15qBU4o8lkoiHGe9Wuh6+D/ntY+rZw0SqH9J\n",
       "8doN6YZoGK1HqWOrK2HNMUk4TsH3HvdG2uz2qy307VMUp+kFsh9Jeo2uGbkUPAz8HYUbdkclX+cX\n",
       "e6CG0R6oAMUxUTiOCB13Qb5B1UBUnPUwZEjpDLxBtI9UMLY6yBzgTOAptM77DfArTDgaRiQdKlxz\n",
       "9BgC3ASsgkJvtgWOR1/u9kgT8XHWg2lJMGZ6jNuirt3rNsMw8lDpa44eVwB7oGkhKOvF0JKNyEjN\n",
       "kF6rl+wTlLUcailD9vo//Vqmvs/6Qa+SfrMqKSQwy1iy/o9GNP43+4CKRFvRHKHl9O/bYg/EMAzD\n",
       "owIUx1R+jvOBHd3+iqguypwUz6sjutxpf+DKmOc0IWNFORiB/BfDHI3iNA3DaCVq8vxFsCHK4fgK\n",
       "cqeL8qapR8bVqW47M2kMaTTHE5EwWx+l93kK+HWK58UxxW1RlNMKHmc8GYZu7ntlHIthGAEKmFZ/\n",
       "A/wWLft1RTLmaVoqcmPwk1YnkkZz/BD4KSo/2gNFdnycbrzL2AQVleqPpPej7vwaSNjOAm7Edxnq\n",
       "gvKxTUPa56ERfR4HvOza/As5VoM0wiuBF5FV2dMOa1DUy1z8cqrh/8DBKAfcHW68ndyYRyMH7idQ\n",
       "US2IL9sa9/rrunZT3Xuy0qyGEUMB1QffR7IA4HMkFNeL6jrtGNIIx55ImH2EBOXDZMti3RsJr6Np\n",
       "qTGejQTGVigeeSN3fg+kpW6LsuM8EdHv/cBA12YOKmfgsQ5aCtgHuMidOwDYDIXyHYWs8GHN8V9I\n",
       "2P0UZfhYipzhD0JC81bgAtf2BqLLtsa9/k/d++iHrOF50xobRrXSobYmcctDHfqeTQydb0bf++ko\n",
       "fdkWSZ2kmVbfiTSuA93xYcBdqHpXPtYCHkKCaW7E9Z3dNdxgF7j9Gai630Wowt8LLZ9KX+B8lBK9\n",
       "K74AbXavCRKaXgGdXdx7aUZT5ucSxu3d/d6o1MEz7rgDKqzVhfiyrXGv/zJwC9DRXZ+e8PqGUdUs\n",
       "hytPV6TknIo0yCCNaG1yCSp/8hBSmCJJIxxXRokjPW5HmlIaFgJvIyEYJRwhWs19DUn+vZEAfJbc\n",
       "DDeg6et+aIp6NJque3wd0X8z6VXqYGTQK0gQBlkVCfKosq1xrz8O3Yd93NgvJ/e+ejQE9kdjsdVG\n",
       "BTJ/xkTmz3y5ZP2HtcP/zZnM/+ZOjmm9jI5oRnk7voISZFFgfxSa7XWnZZVCIFk4dkdf7FGoOPZd\n",
       "7vxh7lwavkYa55NIit8Vuj4WTTcvQJJ8dXd+XSR87kDWpWNoSVe0ztARpQP7T0Sb8GudgGKc10YO\n",
       "2ndEtAuWVJ2H1ll3QOnKOqKMPbNRcoqD0a9UDdJkZ4Q7C7ARWiq4CRUe70d+4WgYFclGWw9io639\n",
       "yeOLd/2jqP2HtZi1+2zP2n22X3Y8++Ebop5yM/puxgV6rw38Dyk/A91zYmtBJAnHRnLX5I4PDKIZ\n",
       "+FPCcz2akQq7DzKCLHKb1+85SGAejqr1ve3O9wUuQYkbvkYW8zBnoTWFD91j19DrhvcfBHZDN2++\n",
       "e70oRqAwwSVIYzwYGI6m7yug+OjZyDB1LXIH6Ojehycco15/V+QG9Y27B0fFvL5hVD0FWKt3RErS\n",
       "DGT0BCV68ewY16Pv8onIT3sJ8JOkDivA1dIIkGXqX1KyFmOq69ElU/9ZImTOq7AImbZK1giZhifn\n",
       "pW578T6bQ/E+u81H/DPZXnnHkdsW8/UiSRshsxWy7HQKnLut+MMxDMOo/JRlHg0olnpL5Hu4J7Ie\n",
       "m3BsZca/tiC1JpA1Djtr/HBWrSQLWTXBzf/0RKaxzL1oj9b/JjpKeR+zlqA95p6ZrZaasAJkYyo/\n",
       "x4OB7yP3l5+jMgHh5K+GYRhFo7amJnErB2k0xy+QM/S3yCjxP+QrZBiGURLKJQCTSCMcJyEXmxtR\n",
       "9Mhi4i29hmEYy00FyMZU0+pfIZ/D64AfIofrqCzY7Y06FOESLtEaFVPdARXTAi05LMWPnR7rnjMU\n",
       "PxtII7muR4ZhBKj06oP9ic+Ssx36grd3NkVO78cD96AY65+j0quvoxDKa4DdkcP4FijufAoKV5wE\n",
       "bIASUPwd/dC8hEotBAtxGYYRoNKn1ZeRnEJs14Rr7YVwidY64mOqxyGB+D3gQpQ1aAwSkKAsPX9H\n",
       "UTkPoGiZKBoC+6Ox8EGjAnnvlUm8N3tS/oYFUumuPPXlGkQFEy7RujaKF4+KqR6LNMN1gb+i+PN6\n",
       "dx7gYpREY28kKH+EtM0wDcs/bMMoLetuOYB1txyw7Hjqv64tav9p1vtKTSWMoS3xGVpbPNgd1yDX\n",
       "JlDWnSFIiH6Fsu6cgC8ce6IkFn9D2qSX/9EwjBDLmbKsKJhwTCaqROvPUCKMachQs6+79jWK2Z7g\n",
       "jscio4tXKuJUtz/dtU2bvMMwqo7amuStHKQNH6xGmogv0bpnzHN2CezfRW4WoqiaFoZhRFCAdrgh\n",
       "itpbCykxN6CEMWGGo+/vElQSZWpEGyCd5lgLHInW0UBZLgamHbFhGEZWCiiT4NWQ2RKlGPw1yvof\n",
       "ZC/kgdILeaAkLpSm0RyvQanDdgPORXkZr0HlAYxW5LWFi/I3cmSN2c1at3pq08JM7UdOmp96PEcP\n",
       "2CiTGjH1vB9lGkvDTnMz3Zth222Qum3WGPWs933B4m9St836GTh8RN7ksiVjheyuPO+7DXJryAQL\n",
       "bO2H8rmC0hyuhoysH0SOIcWLDkLWWU/9/ATlLzQMwygJy+nmWEd0DZn1yU2K/V/kh1ywcPwaRYB4\n",
       "9ECaZDUzFN2Xl9zxCWgNIyqzt2EYGVkOJ/CkGjLQMgdkrDadRjhehbJorwX8H3JjSSyG3QoE68Rk\n",
       "pQNyv8nCriibtyccry/gdQ3DiKFDyBrSNH0iTTPCimAL8tWQeYfcpDkbEB+MkUo43o6iQ3Z3x/vT\n",
       "slB2a1CH4p0noFDHe1E5hpWQMG9w7Y4CTkOCczqKDR8BfInKur6AFmavRlrxEhTdMg+56ZyBomA+\n",
       "RqUROiNNcSly6zkZpXRbhCza26I49JVR2OAvkOP4aDfWXdFaxzFEV1U0jKonrDlusu0ObLLtDsuO\n",
       "x9x+dfgpaWrIPAKcBNyNjDYLiZlSQzrhuBHKxPOoO2525+aneG6p2RRZ0rshjXYgsq4/jCr9fYKE\n",
       "22C37+WhbEaLtYPd/rNI4IXjpcehmwhwLPAHVAfmOiQML3fXdsfXWm9DlrJxqEbO2ciK1oy01EHI\n",
       "leBs4AfFuQ2G0b4Ia44pSFND5nFksX4dybTEBDpphOPj+F/8Tih2eB4ymbc2b6PIlEtRxiDvpnRB\n",
       "grML0ii9CmNBk+p96H11RUIyKl56Q/f8ddy5NwNtohZFVkWCepw7Hhnq9wH32Ig03ygaAvujsdhq\n",
       "owL5cN4UPpo3pWT912QvD/MC6VwTT0rbYRrhuFXoeDukGVUCQb+HC5HjZ5CTiC/Cs8Q91hIfL30V\n",
       "EryPISNMQ8bxhV/bi9VeSvy9z/oahlF2evTuT4/e/Zcdz3v0xqL2v0IFxO4VMoRGNDWsJJ5Ea3te\n",
       "Cbz10frhc8AhqAY3+HWxg3yGX4MaJNC8yJhVgXfd/rDAcxYBq4T6qXF9LcDP5XgkpvkZRmZqamoS\n",
       "t3KQRnM8LbBfizTHWAtPmfGm+08jb3jPerwIrT/MBi5AqcOWIsH+i9BzIb4GdQOaFi9AgnZj1/5R\n",
       "5C6wH35YoNff0WhNsjMyyMSta7Ra8SLDqHQKWHMsOmmEYzBj9bdoinl/aYaTiSZyY5+HEx1LeRst\n",
       "KyWGBVYT0fHSj7gtzGv42Xgg1+o8Ha1hhgnmv/wIJcU1DCOCSk92C7Kurkqu9mgYhlFSKl1zXAFp\n",
       "ijui9TSbBpaBLLWoV++SPooza83iSmLOu5+X9LN38T6bZ7o3oy8cXbLx1PXokr9RTvv0bbPGbf+h\n",
       "ftPUbaM8rpeH2uzW6qKTJBxfRuuL05Df4H34Ft5mfLeUtsIpqPbLFGQoMQyjQql0zdET3Z1QdMhu\n",
       "oeuVKByTwghPRM7a70Zci8LTnA3DKDOVsOaYJJ97AL9D2atnRWyVQh1ySh+JxnoW0nqn4/sMXocM\n",
       "IE8Av0GW5FtQ1o5GZHUGues8giJmns7T7gGUzftVVB/GYw+knU4DnnHnusT0YxhGBAXkcyw6SZpj\n",
       "B1r68lUqcWGEj6Awwl+iglb1KFrm/5AA/AUKKZyIL8j6AX2RY3hSu21QHPXXSDgPd/s3uNd8Gz9c\n",
       "8YyYfrxlCsMwApSrTkwSScLxfRQb3BbIF0Y4LtT+hyipxOnueCUUg9mMNMaFKdo9i/wpQf6UdcjZ\n",
       "fKwbD3n62ZDo6oOGUfVUwJJju6khky+MMIoDkb9ikEGhvpLahcu2rkCyRT+qnxbcNPyiZfvbDdqJ\n",
       "7QbtlNDaMFqHxokv0DixdEmlClhzvAWVPf4fmvmFqUeGZS8/wv3A+UkdJgnH72cdXQXwJHAecAcS\n",
       "cuujqe6HEe1OQenGwM90Hv6PpG0HEowTUEafOuRY3h1N4+P6acGxp/wp6f0ZRkUQ/uG+5aqLE1pn\n",
       "pwDheCvKhRAO+Agyhgzr/Una68dpO6kAgmGEd6IwwhnI/ahrqA1IgHZ0bWbhLx80F9jO4yNUuOcB\n",
       "ZJDxqg+pFUR3AAAgAElEQVTG9WMYRgQFlGYdh8J8k8gkcdvDtLqJdGGEwXC9L5GRJsxI/AI8Wdrt\n",
       "G9h/wm1B4voxDCOCEiSXaAaGIC+Wd9D6/+ykJ7QH4WgYRjujQ/GFYyMygi5BeRQeAjZLeoIJxwpj\n",
       "SK/VW9+HocLIWt601Ez8c33JxtOGwzyLGlIZvgmvTB7PK5NfimybkmAd41HINuDZBFKNwWhdmrH/\n",
       "idE2KeZnt/n+acmBbAdtux4Rr1eH0glGWavXRpbsZuQHfS/x2fgB0xzTcA7yXXy2iH3Wo0xH++Zp\n",
       "ZxhVSQFrjnehbP1rotrUZyMjKKh+zMEohPhbNLX+Sb4Oq004FlLC9exSDMQwjHgKCJA5PM/1f7gt\n",
       "/RgyD6HtUUf+2Gvc+bnIJeBO/ByWI4CD3P7uaGF3BioD6RXianJ9TXHXervzA4Hx7jkvkmcB2DAM\n",
       "UUtN4laeMVQHm6Jfjd8ix/CByBG7P4qDHoAiWLZGlqzt8bVLz6exE3I0PdS1WwGp6V6bD11/1+KH\n",
       "Cc5x/W+HNND/K9H7M4x2RaUnnmhPJMVe90IJNh5C0TRf49fo9qhB2uBbqOYtSBP9NXClOw6WXT3Q\n",
       "7a+GPPY3RQI0fXZaw6hiKiFlWbUIx3yx16eSa/mKCw8k1CZ4Lqrs6nnIkHMAKs41OsVYGwL7o1M+\n",
       "xzDKTb3bSkKlZwJvj8TFXr+ILFoXIu1ub3fs0YzWLeuAnqiq4JEoVjOJYGnXuCqEYRpStjOM1mQ0\n",
       "uT/cRTVc1lbAgl+1CMd8JVwno9yPM4APkOHm01AfXyEBdx+6by+jJLrB/r197/hvaPp9JvDviHaG\n",
       "YURQUwGaY+uPoHLogrTJzkgjPA4ljygn5gRutFWK6gT+7JyPEhvs3mdNivh6kVSL5piGG4AtkFV6\n",
       "BOUXjIZhOCrAHmPCMcARrT0AyFaGdMHib0o2jlLHeH/xTXPFLCtMbVqYv1GAZ5uStZogp2cobwqw\n",
       "+emPZWr/9pX7pv4/Zb3n9077T+q2Rw/YKEvXeSlB4onMmHA0DKPiqIQ1xwqwCVUc/fF9Fw3DaAUK\n",
       "SHZb/DGU52XaFFOQ36NhGK1EbU1N4hbBLfieJnEMR3WcpqMIueQxFDLwNkoX5E4zDd3AQ1HY4Hh3\n",
       "biIqqVCPHyETV296GFa32jBKRk2eLYJb0Xcvjr1QpFovVMrk2nxjqKY1xz1QevS93fGqKIzwUCTM\n",
       "ugJfhJ4TV28arG61YZSMAlKWjSM5P+N++KVNJqLv4dpI24ykmoTjDBRbfRHwGHLyfg8JRoDPI55T\n",
       "9rrVV196wbL9gUN2ZuCQXVK/QcMoF3OmvMTcKcuVmTuREhir10d5Hj3+C2yACUdAaw39kOZ4PvB8\n",
       "yueVtW71SaefkXJYhtF69Ok/mD79By87fujGK4raf1g4TpkwjikTlrtOdljkJro2VZNwXBeVbrwD\n",
       "aY0nAuug9GSTUWae8DS37HWrDcNo6cqz/Q67sP0O/izqpuGZ62S/g2ZrHhu4c7FUk3DsC1wCfIfW\n",
       "Bk9EBqmrgJWRYPwBubHR5wFXoCl5LfAmWrtIU7e6FqnsP0roxzCMCErgrvMIcBJwN7ADWvKKnVJD\n",
       "dQnHp9wWZnDoeAx+th2rW20YrUF24ZivhszjyGL9OsqhkDdLVjUJR8Mw2ggFJLvNV0MGpDmmxoRj\n",
       "G+a1hYvyN3Icuu2G+RstB+NfW5ApbjdLfeasMcGVVPv5rBL3P+jC0anvTdb7kuW+H52l4xRUwj+w\n",
       "GpzAm5BhpJT8G/lNGoZRBGpqahK3clANmmM5ciTunb+JYRhpqQTdv71pjlEhgiAXmnDZ1O6oqNZ0\n",
       "lBm8rzvfAPwThRW+ChzrztcjB+/HUAnXa/GFbpPrrw5VHLwBmIVceDq5NgPc609FVvOkGFDDqGoq\n",
       "ofpgexOOXojgtkjYeVbjqLKp5yCBuQ3wF1Ql0GMrYFdkyf4r8pEECbiTUFLcnvhVBoNrM5sCV7s+\n",
       "FuLXvL4VZRfvB3yLlUkwjFhq8vyVg/YmHGcgX8WLgJ2Az9z5YNnUOre/I9IQQdEyayBH8GbgYRQB\n",
       "87G7NtCdfxlpid8h14GdIsbwlhsHSPjWAd1Q7PZEd/5OKmPN2TAqkkrQHNvbmmM4RPA5dz6qbCqk\n",
       "F1DfucegtlcTOB8kHFa4ckSb2Ne12GqjLTB2zGjGjhldsv4rYc2xvQnHYIjgQvz1wijGodII56P1\n",
       "xA9RIokaYH9UptVLYfZHYHOkQdYB84HD8KsP5uNT1/dApH3+JK6hxVYbbYFdhtazy9D6ZccXnHdO\n",
       "UfuvhEzg7U04hkMEf4VKqXoEw/4aUI7F6chj/uhAmxloOr0mcC7wPhKOk9B64qZIK30w8Bwi9oPH\n",
       "xwA3urGNoWXpV8MwHOXK9p1EexOOUSGCmwT2pwC7uf0FwAEx/cwg2q/1M3JDBcOv8QmwdeD8ZYH9\n",
       "V5DxB+BPSNAahhGFCceKJcqSHJdsIi17A39G97wJZRM3DCOCAsIHi44Jx5bELZ4EE1IUwr1uMwwj\n",
       "DwWKxj1Q9qsOwE3kli8B2Q8eRlmxAO5HNodITDhWGH3W65r6c/HHx+ZWjK9kv7rV8jcqkKwxwQ/P\n",
       "eC/Tfdl/63Uz9Z8ljjxr/e+Rk+ZnGvvRAzZK3f95T7+W7b5cPyFL86JSQIhgB2QP+D7ydZ6E0pTN\n",
       "CbUbQ8p0ge3Nz7EcpCndug2wZxnGYhjtkgL8HAeidGRNwDcob+P+UV2nHYMJx+ykKd3aD+WOMwyj\n",
       "AAoQjlE1YtYPtWkGhiAPlcdRpFss1Tit7oLW/tZHqvh5KBLmEnQ/JqEs4V+jcMEr3HO+AnZHZRVO\n",
       "Q1brLiiT+JYosWYDKtd6Loqp3hFF65yP/ikfoR+keSgb8celfauG0TYJ+zm+9MIYJrw4NukpaZYM\n",
       "GlGphCVoZvcQsFlc42oUjuESrd1QEojdkFo+EgnHa5FqXkjp1rPQ9PsU125z5HB+JVoTmYYJRsOI\n",
       "JeznuOPOQ9lx56HLjq+85ILQM1rUiNkQaY9BgglQR6F6T16dp5ZjyDDe9kI4/roOxUO/7q6PBHZB\n",
       "vyjh0q1LQ339EPksTkVO417p1nDt8VuAo9z+L1ASCsMwYihgWj0Z6IW+zyuiCLZHQm3Wxv9eDnT7\n",
       "kYIRqlNzjIu/9shqJosr3Rrkv6iYz25oqp6U0r0hsD/abYZRUXzyaiOfvNZYwlfIbK3+FmXMehIt\n",
       "l92MLNUnuOvXAwejWeG3aGodG8YL1SkcwyVafw1sjFKQvQEciQTSPNe2kNKti1z7IDcBtyPNNGl9\n",
       "pCHzOzKMMtN9s+3ovtl2y47fHHVzUfsvMHxwlNuCXB/Y/4fb0o2hoCG0bfqitcGpaG3wDDTVvQ9N\n",
       "ub9FCSW+Qar5VWiN0EtcGy7d2tE9bxa+A/nzyBI2FT/h7qPIgGNTasPIg6Usax3iSrRuF3FuMoWV\n",
       "bl2A1jSCbIOE7KupR2oYVUq56sQkUY3CsTX4ExKiP23tgRhGW6D1RaMJx3JxkdvykqUc5tSmhakH\n",
       "8NSc91O3heylVktJ1tKsTR8uLtVQgGwlcbMycP3SFco8eMt18zcKcMHvryjRSPJjiScMwzCiaH3Z\n",
       "aMKxhAxFUTYvtfZADKOtYclu2y8roOqFizDhaBiZsTIJlUNUvPXfgHtQDOYXyJjyBvLAvwVVK/wQ\n",
       "+DkKeB+BrNfbolCmISii5mfID3JdVOZ1KfKv9GOhDMPIoQKWHE04OsLx1quiRJkLUdmDI1ECin2R\n",
       "3+OtqKzrz4Hh+OUW1kOuP83A2UhzvNxdm4HCDd9z/RuGEYMJx8phBnApsig/Brzgzt/lHu8G/u72\n",
       "dwB+7PZvRxomSCDeR8vyrR4vouiYe/HraLfg/HMblu2HK7wZRqWwdNE7fPf5OyXr36bVlUO+eGuI\n",
       "F3pBwuGFQU5EjuF7o2QW/YkIej/zrw35R2sYrUyHVdanwyp+usSlHxS3XlwlGGSqMXwwinXReuEd\n",
       "SIPs584fFngc7/bH4wesHwHEJZkLx1f3RDWrz0ZrlRukHVyW4umNE1/I3yjAzEnj8zcqsP8Cir7X\n",
       "Z2mctf+XxyfmAwxTn6XxnCmZ7W6p+8847kx9Z+1/6aLSaYs51OTZotkDmIuUnT/GtBnurk/H/55H\n",
       "YsJRhOOtvaI7q6ObeDLwW3fuZLTWOB0Jx2BW8KB2+Shai2xEqdH+hqbvM9EUe0bawZVSOM6qKuE4\n",
       "Lkvz+iyN55ZUOGYad6a+s/Zfyql0kNqamsQtAq+GzB4or8HhQJ9Qm71QzflewPEoZ2ssNq0WcfHW\n",
       "f0Ohf0Hmo4zgYX4eOn4Nv041+OuYhmHkoYBZdbCGDPg1ZIIFtvZD6/4gZWg1lOPxg6gOTXOMp2LC\n",
       "5wyj2qipqUncIkhTQyaqTerlLaN1GY2fEs0229rSNprikeb1Pgs95yDgxsDxz5DbXZBHUV0nj2eI\n",
       "zsYF2LS60qhv7QEYRgVQiK06TQ2ZcJsN3DnDMIx2ywr4EWwrotypUQaZx93+DsCEcg3OMAyjNdkT\n",
       "lTd5HfizO3cCfh0ZkEX7deRtEjulNgzDMGKoAD90o4h0R+soqX0oW5la5Cv6PeBcVNZ2HeQsH8VO\n",
       "aLr0OYp374dqgb8d096rNb4U6O22Uag+UDHojNaw5iW0eTThWjNyLym070JZg8Lqpre1z5fRDvkN\n",
       "0A39eN2MnNN/FNN2DEpk0R3V334ZPw48ip2Q0AAJmMtR9cU4LnH9dwSeBT5yz4vr+2nk4/mW295M\n",
       "6Ps6VFh9rjvujur2xDET3ZNt0D35NX49nygakZBZH/m/3YeioIox9v2Q4Gpyx/1oWScZZGRL2pan\n",
       "b4+Z+AEG3vYC+hysEdH+NXQv9iK/gpT182UYJcX7Zf4R8CCwFRIGUUxzj8fiVz+cmdB3VgEz3T0e\n",
       "gAR1N+I1h3lo3WdtYM3AFsfU0GPw9ZLan43eL0gA5mt/MvCHPP1nHXsjciIOjn1WQvssZO37EuBC\n",
       "FOm1NfB/KIvUn4jWXGtRhqi7kRHjQmCzmL6zfr4Mo6R4H77hwIFuP044zkSx4U/hVzxMmvZkFTCv\n",
       "uMebkfCAeAEzMaGfuPYdAmPqQfz7BMWx/wVpPuu45yZ9UaeiFHITgC3dubj2hYzdew2PpPu+GfAv\n",
       "FLGRTzPN2nfUPfPO5RNkuwHvohyjY1Ae0iBZP1/tBouQqUymoA/jXsATaFrzXUzbc1FN7TfQlKcn\n",
       "Eh5xLEIC5mcoPVsHNGWO41E07e2PptVroSQdQfq77XmkxQxGlkBvi+MqpBmvhbSdF5EWE8dhwFeo\n",
       "zvj7aLp8aUL73yCr5YNIyPd0Y4wi69hfQeulK6BY3avwk5NEcSt+PfR6FMYWN8XP2ncHYFDgeCD+\n",
       "d/vbiPZropwAU4DTgZPcudOAO0Nts36+DKOk1CJhs5o7XgNNl4rBOuhLsLM73gg4Ks9zuqMvIChr\n",
       "+jqh66ORcInbkuiDvpwn0dIvbXnpmaHtaLKNvQsS6JPddgHQKaG9p53PjDi3vH0PQNPuJrfNRAKy\n",
       "C3BoRPtXUVb6qNC5cC6BnSLaRJ0zjLLwbMpzoOQYaQ0mK5BfWIXpgjIVeaFZvYB9YtpukvJckNXR\n",
       "+md/8mtrByGt5TOkAS+iZRhZkLFo6noPWlvtm2cspWQ8+oF5EP0QHEjxLdHd3JaPKIEZdQ6ip+xJ\n",
       "yzCGURJWRlriDKSteVsdvkU3TBaDCUiArpZwPcy9KDeet/bYhfg1x6gvzZSEvs9DiQDGkE5be4Ps\n",
       "2uVKKJ72DJRRqUWCYcdqyAo7xW2XkSxoniH3PnZH0884BqL8nhuiekMPoCiNYvTdCU3Dz0BryWcj\n",
       "zTCOqP9TWAgORjOM/wK/c/unAQ0kG83aDRZbXVmcgNaC1iNXqCxCnv1ReP/DfdCC/6fIfy6OxWja\n",
       "9bTbx7U/JaZ9T6RVeAl+F0e06YNy6K2GNKIa1+eqJE8HD3P9f53QJsj75KagysdOwC7ucTXg38Qn\n",
       "J74F3ZdD0PiPROuEB8a0XxPVGPL4BFm64/B8NxcBw/KMO2vfD7v2U2i5HhxkT7SOvQEy9nluPKvQ\n",
       "0vdzRXe+A7lJmz8DDk4efvvAhGNlcQVafP8L0qrS4BlMvkSlGKIMJkEecJsnQGtIFqZfIY3Wo6c7\n",
       "F6Q3Kj7WzT16LAKOS+j7FTStjsynF8FkNEV+CF+gNhNfk2cMEhgXopja8LiD9CRXEDaQrCEtRf6h\n",
       "ngN6HfFGM9A9Ot218753zchavLx9r0+8H2yQd9H92N89ev/7RfjJnD3GIANZX3wXHsNodablb5JD\n",
       "PoNJmM7A5in7/iH6onyILJlvo5rcUQxO2afHAPSFfQoJ+UdJdnYe4bZbQ1scqyGN+mI0XX8GP8t7\n",
       "mAn4RiqQtpmU3nsPNE2/3W3z3bk4ZqAfr0HA9m7rX6S+byCbwa4j6T8DE6jSSLqqfNNtgEvRh/J+\n",
       "krU6kDD8HbI6H4cMJr2Rm04U+yGXlZWQRtIPaQZxYWygaZ63PjYBGX2ChPPmBUmass9Bqepn4WtG\n",
       "zSQ7pWdlCzS13gX58M13+2G2BW7DX2dcABxNsvbYA92XZqLvSxCvqFpasvQ9B6X/fwtfO24mXmBm\n",
       "+Qxch5Z57sMvIJekrbcbTDhWJp+jX/al+FNkbw0vzL3oi3cUcnTugiyj20S0BS3G74Y0Ka/A0CwU\n",
       "hRPHNvjTQU9YB78cw8idpgdpxk9NH2YS0h7T0huFG66D3uvW6Asdpw2+iSzC49Ba48vkX9/07nGS\n",
       "FdxjfVrel7g1zQakfT9A7vQ+zkCUpe+6mPNNMeezfAZGuMfwj3S4LEi7w9YcK5OuaKrci2SDBqQz\n",
       "mAT5htzFfkhez7oVrTu9EmoXFI4j8rxmHOPQeuAj5AqMOFeRG4HfI20GZEC5i3jh2Av9wCRxJPBP\n",
       "ZIkNCgBvPe7ymOddjAxKs0OvESfAhrn+Tg+cayba1Slt36siIZ5GkAfJ8hkYlrHvdoMJx8rkODQV\n",
       "3QCtP+6A1r+iFu/TGEyChKMvTiE5+mIQ0tLyTe9BxqA/oKmsN6Y4owPIp7GZli4tcWuanckN82sm\n",
       "OcPOpuRqmtsgg1FQmHZ2j6uQ7j16HIA02aR7HaSuBH3fheqgN9Jy7HGCF7J9BjZElm3P8Xss8qgI\n",
       "Z9k2jLIwCwkXzzDTBzkPR5HFYALZoy9G4scl5+NpFK89FxiKtM6/pXxuGkYhgef55B3szsUxFgl3\n",
       "r30Nvr9mmKyRIKPIdXHJx4pIqNyPXK5OJj5sM2vfWcnyGXgGTaE7um0Y+j8bRqvgpe2ahv+hnZ3Q\n",
       "fk1kld2H5EwyID++NOc86tG07VX8dFhxTubedDh4PSkFWVbH657Iif0LZOV+kWSNzHvtoINznCdA\n",
       "1kiQB5BT+g3IIHUV0rDiuBn90OyGSvuOAG4qUt81KHro7+geHpDQNitRBilzAjdajf8g/7+H0K/0\n",
       "AuIX10FWxwXo/7mFOxe39vUXZHnMd87jZpSkImhRjsMzdryPBPW76H3EkdXx+g0kWLog16V8a20f\n",
       "Ik3T42DgvVCbwciK3QNZ/YOO0R2I5xG3pfUXHUCu9fhZ4n9ksvZ9DfrhuMu1/SXwA+BXoXZXIu01\n",
       "Ko1ZXOLdj9H/5U7X909ItpwbRtmoRx/aFWOuX4wE5+P4voJRH/49kQbyP6SFeBrJCOIzb0Oyr1+Y\n",
       "fZE22Bclcmgk2UUoq1ayDhLWT7jjLYBjEtr3RNPCJUhQv0BLTXMosiS/hx96dzYSlL0S+oZs/qKN\n",
       "5ArqniRrpln6nktuhq1aosNNPVei+ohtaEzfdejz9KHbHkZuY4ZR8byKNMd8bIPWi+Yj/71hbjuQ\n",
       "ZO3uGqQ1HI6mbgcRr9llJavj9RPIiutpXB1JTgIbjDm+jOSY47r8w80ha7bu3dG9H+O2t4k3VGXt\n",
       "+zFyx19HvJ8ryBsiqBV3QNq4YbQrsi7eJ+VujGIELSNS4qJSeqPpomf02Bo4M6HvbZGm+LbbphLv\n",
       "nwnZ1hBByRruQRb00wJbFGsh5/vH8RNgPJfQdyGZwDuh97c1yT9oWfsei9ZhxyCNfYnbj4s4mohf\n",
       "KgP0+YmzVvd0/XyErznmy7TULrA1x7bPF0hAPEtudERcVMqrEeeS3D6GZRhLVj/Eucia3RMJg09R\n",
       "3G/c1Ppzcmui7OCeE0famGNQ4tl70FrpCeh9f5jQPq2v4O7of3MQus/emqY3xY6KNMnqi+ppw1GO\n",
       "+FFrlSuhe+mxCN+lKcydKOmJN1s4DP1PB8W0bzeYcGz7eIv3QfIZBjw6ISNFVBEmj5XRup7nu+j1\n",
       "/YuItln9EIPZZN5JaOdxGtJiNkGaTg+SM8SMR1pamrT+ayDr8Sn4U98kS3taX8FdkHDcl+j/S5Rw\n",
       "zOqLOhqVMhiIhOgkZBSLYzFaf/QyP22PfmSjWBk5yXvcjn4ADaMqSDIM/AtlCHoTrVU+TbxbSVY/\n",
       "xKwFqTqhZYGtkNFnRZJ9NOcg4ZzGDWmCe3wKaY/bIet4HJ3J5i+ahay+qMei9cyRbnubZEPVAPTe\n",
       "XnDbG0hARnExKjVR57Y/Ahfh5xptt1hsddvlPuQCE1VAKSnpQH98DaYWfSlOJH6tbxpaG5zh+uyI\n",
       "vlBR06qewPXINWYhEqg/I94N6QY0ZUtbsKmRlpnCo8551MWcjxrPviiccUNkxV8VWbGj1uxWQD8S\n",
       "Sc72YcLhiaAlgSlkz8IU5lXkkuTVol4DGbbiKgqCfli86/OI1/CbiJ+JJC3HtHlsWt12OdU97pvY\n",
       "qiWX4X/Yv0Uf/rgU+eD7Ln6KtLX30XQ2ineQseZ5pFV8hhJinBtq5wn0Dij6Il82mXVRZpjOSBAG\n",
       "k+nGrZVBsm9oGM/9aSHx9aQ9vkXT19VouTYYR3/0Q/QoGv/e6D78EmnnF1OYHyLIWBJcQ/ycZF/E\n",
       "FdEPopedaDR+8a8wf0BeAp+htc1+aA05KcN7u8A0x7ZPF5S5ZymyFvdGU9mktb4sHIvWxfoiwbcK\n",
       "qilzXUTbJ/HXEIMJEy4LtavL85pNoWPP9Wh7ctcBF+GXHFhesmb8eQQJirQZ1cchX1NPiHVFlvE9\n",
       "0P3qg78OWB/x/KRUbv9ESw0Pu+P9kTY+g+jkGTcjxWgkvvP9t/ileoPMRP/7ndC9uBT9/9u9QcaE\n",
       "Y9unEX1wV0fhdJOQtndETPs1kb/fTuiLMw5pdh/HtPemg8HPyqdISIWng/lSny0vByMtqxSMxbe0\n",
       "90PvdxbxceXD3GM4iiUuPdtcJHA9TXwlJLx6ozXafoG2XZGBxPuB6YDWHOMyLjXEjMUjnMnbWyLJ\n",
       "dw78ZZWLkKC8I2K8hlGReMaPk9EUCJKjTJ5Bv/zfQ+tFZ7pzcdyJ1rQuc9s8JKAmocX5IFkzUlcS\n",
       "WX0oIVsUy1mu77ORMJvi9rvQsn51Fj/EIKuQzuc1S7TOv9H/9S20jNCJKomtNto+U9Fi/AR8LSfK\n",
       "SOMRZSFOaj+O3C9qV6RldcYvduVZgmeT3jpcaWS1tGeNYgFZiU91W5x1GKKFcpKg7ovGPd9tU0jW\n",
       "4MPROk3ER+t0QT6aXijluigTlGFUPEPRl9LT4nqSnMHlchQKWOu2w2i5JhhkLrlx3Svh11v2BEld\n",
       "nq0t4GX88eKw82X8KSRCZmd8/9AeSHuPYjy5JRW2Jzms8iVyLef1JGuah6LsR9sgI8sokuuFG0ZV\n",
       "8DmytH6Lb3Vd5LaoLDdZpoOlwovn9h7D2/LSARkaQJpxVDmKMJ6ze1A4JmnJDcgK7UUorY8EcBRh\n",
       "P8TXSdY0sybw8GYKOyFL9T7kOu8bmCtPe+D5iHNJ2be7xpyP4zzkyrGj6/cE/PW5OKNPsfGiS9ZC\n",
       "PpRezPOuSENaXmv1UiQoash1iUkiaxTLAWjq7bnAvEP8+uD3XNuNkfD3Il/ieAv9iP0TvYcjkI9p\n",
       "HJ6hZx8U8vkY6UsBG0abYfvAthNKeHpJnud4birF1L7KwdNozctjXRTRUgyuQ8sTR5Iu+1DWCBkv\n",
       "LZynaXYhXtPMqtl1R47rjW67kuRMS2ZkMaqWSQnXbkVf5pGkq/1cScwl16UoLm9hIYwgXfYhL874\n",
       "Nxn6rkFre9cjgXQ8MqDF+UR6xpeL8LXzqEzlhWJGlhSYn2PbJxjf6oUDXon856KYTfqCWZXG1Sjk\n",
       "zctKfRjwGnJjKhezge+jpYb6iOtRpVZrkDb4W/wsQU8SX4vl32ja/QM0vf4SaY5xIZ69UVXDOvyl\n",
       "sqSlFcOoCpqQNvIWEhRPk1wYKkvBrEqjBk11r0DLB8WslTISTTE9VkdlHMKcglyYvsK/796WtM43\n",
       "Eq0dpiGrZjcDhQMOwl9i6Z/Q3jCqgs5Ia3gIGSZ+R/LaVz3pC2ZVE1l9C6PCJ5OYhwwhb1L8+97u\n",
       "45xbA5tWt33uQ8LudvT//CnyYYurKPgGmt6FC2Y1lW6IReMgtA63Nv5n10tAsbxMR9Zvb1rcHTlI\n",
       "9014zs7IcfxW5LfYFWmQUdTFnG/KOM4g3dF9OBkl5n2A3FrXUVN8w6gaokq2JpVxzVIwq9J4AyVo\n",
       "KAVHIe3uPJRgYZ47F0cDcoEJ+i2mCfErJk20nNqnmeIbRlVwOwof9NiB3MzNYUpZMKvUxDlNF4st\n",
       "gZPctkWettORASytE3gpORRfe/4r8CC25rjcmBN428XzhVsBCY3/oCnmRvjhfVF0Rplhwgv8xUj7\n",
       "VWomozovD+Fnt2mmeGPvjjLfeNPk7xE/Tf6K3GWJ1qzedxZwLzLE7Yb8XK+hCtKKlRITjm2XpCS3\n",
       "SW46w4o8jnLSDaXyKoVgb0DaVm8kHFdEWvmOEW1r0JT6emThPh7FTN9UhHEUQjji5d/E56E0DCOG\n",
       "DdG0yyvSfj+wQauOqDLIMk32cj3+AMVkX+r2WwuLeDGMIvAMKk3Q0W3DiHdGrjRWRuuB1yDt7hai\n",
       "fRELIUt4H2TzWyw1FvFiGEUgawaXSiJLJcQsZA3vg9L6LRqG0Qo8h5IrdEBrzj9DeQzbAp5TtieE\n",
       "OlKcVFuFTJPrYjbDMNooG6O8gt6a48PIwt0W8Ka+45Bzdg+K589XSdNkwzBagZHkprPqTvHW7UrN\n",
       "cWEHGdgAAARHSURBVGi8Q9H090NU2rQY2DTZyMHCB6sPr5pcvnPVRl3M+aYyjsGoIMzPsfqoQdpX\n",
       "MIa4Q+sNp2Joau0BGJWFCcfq4zIUX30vEpSHoCzWhmEEsGl1dbIlCjNrRtbrpEQVhmEYRoUTTLBw\n",
       "For0sZKihmFUPVZS1Cgbta09AMPIQFRJ0RVbbziGYRiVgSVYMAzDiMASLBhlw6zVRluge57rVivF\n",
       "KDomHI22QBNyO6pBceAL3PnVgbdRxm7DMIyq5UZgr8DxnmgN0jAMo6qZlfKcYSw3Fj5otCXeBc4k\n",
       "t0b3O606IsMwjApgDZT5e6rbriS/scYwDMMwjGJh1mqjLXAlcCrKYB6mGdivvMMxqgFbczTaAre5\n",
       "x8siriXV6DYMw6gK+kec26fsozAMw6gwGlFhLY/D8YtuGYZhVC2bIAG5OSq2NQ7o1qojMgzDqBB6\n",
       "A3OAJ4DOrTwWox1j1mqjLTAzdLwWsBD4Ghlkti77iIx2jwlHoy1Ql+d6UxnGYBiGYRiGYRiGYRiG\n",
       "YRiGYRiGUZEsRdlzZgL3AisvR18jUP0YUNLbPglthwKDC3iNJqKz/MSdD/J5xtdqAE7L+ByjjFhp\n",
       "VqOULAH6oaiWr4Ffhq5nie1vxo+jPg75OsaxKzAkQ9/B18hyPmub5WlvlBkTjka5GAdsirS6ccDD\n",
       "KIt3LXAJCgOcDhzv2tcAVwNzgaeRb6PHaPw46z2AKcA0125j4ATgt0hr3RHoAfzLvcbL+IJzDeAp\n",
       "N44bSefa9iAw2T3nuNC1y935Z4A13bmewCj3nLHIid0wjCpnkXtcAQnDE5Bw/BwJMZAwPMPtrwRM\n",
       "Qn6NByLBVYNKsC5w5wCeB7ZDQm9+oK/V3OPZwO8C47gTCUlQga7Zbn84yiwOqk3zHdHT57cC51d3\n",
       "jyuj5QLv+DsU6w1wFnCV238W/SgADHLH3hhtWl3BWMoyo5SsjLQ3kNZ0CxJSL6OqgaC6032Bg93x\n",
       "qqgu9c5IqDUD7wHPhfquAXZw/Xp9LQxd9/g+uWuUq6Aa2DsDB7hzj+NXNUziVODHbn9DN9aXkXC8\n",
       "x52/HXjAvcYQ4L7A81dM8RpGBWDC0SglX6A1xzCLQ8cnoSlxkL3IP81Nu25Xg7S2r2OupaUe2B0J\n",
       "5S+RBtspps9mtGSwgOh7YFQ4tuZotDZPAr/C/6HeDCWUGAschj6j6yIjS5BmYAKwC354oTf1XYS0\n",
       "Q4+ngFMCx9u4x7GoSBeozOvqJLMqEnZfosxAOwSu1QKHuP2fonXVRWhK7mnFNVgceJvBhKNRSqI0\n",
       "u+bQ+ZvQGmAjWsO7FuiADB+vuWsjgfERfX2E1iwfQAaZu9z5R9F02TPInAJsjww+r6C1T4BzkHCd\n",
       "5dp70/O49/EEEuKzgQuBlwJtFgMD3XuoB851548AjnHjm0VuSQezWBuGYRiGYRiGYRiGYRiGYRiG\n",
       "YRiGYRiGYRiGYRiGYRiGYRhGMfh/HSmKfXmcDQIAAAAASUVORK5CYII=\n"
      ],
      "text/plain": [
       "<matplotlib.figure.Figure at 0x10ae61590>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[17  2  3  0  0  1  0  3  1  5  0  4  0  1  7]\n",
      " [ 0 25  5  0  2  3  0  1  1  0  0  2  0  0  6]\n",
      " [ 2  2 35  1  1  1  0  1  1  2  0  5  1  0  7]\n",
      " [ 1  0  0 12  0  0  0  1  0  1  0  0  0  0  4]\n",
      " [ 0  2  0  0 11  1  2  0  1  0  0  0  2  0  5]\n",
      " [ 0  3  0  1  0 13  2  1  0  2  0  0  2  2  5]\n",
      " [ 0  0  0  2  1  0  7  0  0  0  0  0  0  1  8]\n",
      " [ 0  3  0  0  1  0  0 24  0  0  0  1  1  0 10]\n",
      " [ 0  5  1  2  3  5  0  1  8  2  0  2  3  1 13]\n",
      " [ 3  2  0  0  0  0  2  1  0 19  0  1  1  2  3]\n",
      " [ 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0]\n",
      " [ 2  3  3  0  0  1  1  1  3  7  1 28  0  1  5]\n",
      " [ 2  5  1  0  3  0  1  0  0  0  0  0 19  0  1]\n",
      " [ 0  8  1  0  1  0  0  4  0  3  0  5  0  7 15]\n",
      " [ 1  3  4  3  1  0  1  2  0  5  0  2  0  2 83]]\n"
     ]
    }
   ],
   "source": [
    "cm = sklearn.metrics.confusion_matrix(dmoz_encoder.inverse_transform(y_test), dmoz_encoder.inverse_transform(predicted), dmoz_encoder.classes_)\n",
    "analysis.plot_confusion_matrix(np.log(cm), labels= dmoz_encoder.classes_)\n",
    "plt.show()\n",
    "print cm"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## Searching over parameters using GridSearch...\n",
    "\n",
    "Which combination of keywords, description and key phrases should we use? Let's check every combination!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- It seems that we want to use **description**, **keywords**, and **title**, but not **keyphrases** (flags = 13).\n",
    "- The worst single feature is **keyphrases**.\n",
    "- The single best feature is **keywords**\n",
    "- Adding **keyphrases** seems to decrease the accuracy!\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "GridSearchCV(cv=None, error_score='raise',\n",
       "       estimator=Pipeline(steps=[('sanitiser', MetaSanitiser(meta_selection_flags=13, stem=False)), ('matrix_builder', MetaMatrixBuilder(min_word_count=4)), ('classification', ExtraTreesClassifier(bootstrap=True, class_weight=None, criterion='gini',\n",
       "           max_depth=None, max_features='auto', max_leaf_...estimators=100, n_jobs=4,\n",
       "           oob_score=True, random_state=0, verbose=0, warm_start=False))]),\n",
       "       fit_params={}, iid=True, loss_func=None, n_jobs=4,\n",
       "       param_grid={'sanitiser__meta_selection_flags': [1, 2, 4, 5, 7, 8, 9, 12, 13, 15]},\n",
       "       pre_dispatch='2*n_jobs', refit=True, score_func=None, scoring=None,\n",
       "       verbose=0)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Gridsearch\n",
    "from sklearn.grid_search import GridSearchCV\n",
    "param_grid = dict(sanitiser__meta_selection_flags=[1, 2, 4, 5, 7, 8, 9, 12, 13, 15])\n",
    "search = GridSearchCV(clf, param_grid=param_grid, n_jobs=4)\n",
    "search.fit(X, classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'sanitiser__meta_selection_flags': 13}\n"
     ]
    }
   ],
   "source": [
    "print search.best_params_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[mean: 0.28691, std: 0.00626, params: {'sanitiser__meta_selection_flags': 1},\n",
       " mean: 0.21726, std: 0.00568, params: {'sanitiser__meta_selection_flags': 2},\n",
       " mean: 0.28357, std: 0.00914, params: {'sanitiser__meta_selection_flags': 4},\n",
       " mean: 0.33862, std: 0.01369, params: {'sanitiser__meta_selection_flags': 5},\n",
       " mean: 0.33778, std: 0.00939, params: {'sanitiser__meta_selection_flags': 7},\n",
       " mean: 0.26647, std: 0.00674, params: {'sanitiser__meta_selection_flags': 8},\n",
       " mean: 0.33486, std: 0.00590, params: {'sanitiser__meta_selection_flags': 9},\n",
       " mean: 0.33361, std: 0.00670, params: {'sanitiser__meta_selection_flags': 12},\n",
       " mean: 0.38032, std: 0.00682, params: {'sanitiser__meta_selection_flags': 13},\n",
       " mean: 0.37823, std: 0.00223, params: {'sanitiser__meta_selection_flags': 15}]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "search.grid_scores_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of documents with no metadata: 173\n",
      "arts 14\n",
      "regional 0\n",
      "shopping 1\n",
      "business 10\n",
      "reference 12\n",
      "science 15\n",
      "computers 14\n",
      "recreation 10\n",
      "kids and teens 3\n",
      "sports 41\n",
      "society 14\n",
      "games 7\n",
      "health 13\n",
      "home 8\n",
      "news 11\n",
      "Meta-data-less documents get predicated as: \"sports\"\n"
     ]
    }
   ],
   "source": [
    "trf = MetaSanitiser(stem=False, meta_selection_flags=15).fit_transform(X)\n",
    "m = MetaMatrixBuilder(min_word_count=4)\n",
    "count = collections.defaultdict(int)\n",
    "for name, empty in zip(dmoz_encoder.inverse_transform(classes), np.sum(m.fit_transform(trf), axis=1)==0):\n",
    "    count[name] += 1 if empty else 0\n",
    "print \"Number of documents with no metadata: {}\".format(np.sum(np.sum(m.fit_transform(trf), axis=1)==0))\n",
    "for x in count:\n",
    "    print x, count[x]\n",
    "print \"Meta-data-less documents get predicated as: \\\"{}\\\"\".format(dmoz_encoder.inverse_transform(search.best_estimator_.predict([{}]))[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "#### Separate pipelines for keywords/title seem to make no difference"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.39 (+/- 0.07)\n"
     ]
    }
   ],
   "source": [
    "feature_meta = sklearn.pipeline.Pipeline([\n",
    "    ('sanitiser', MetaSanitiser(stem=False, meta_selection_flags=5)),\n",
    "    ('matrix_builder', builder)])\n",
    "\n",
    "feature_title = sklearn.pipeline.Pipeline([\n",
    "    ('sanitiser', MetaSanitiser(stem=False, meta_selection_flags=8)),\n",
    "    ('matrix_builder', builder)])\n",
    "feature_union = sklearn.pipeline.FeatureUnion([\n",
    "    ('meta', feature_meta),\n",
    "    ('title', feature_title)])\n",
    "clf_union = sklearn.pipeline.Pipeline([\n",
    "    ('union', feature_union),\n",
    "    ('classifier', clf)])\n",
    "\n",
    "scores = sklearn.cross_validation.cross_val_score(clf, X, classes, cv=5)\n",
    "print(\"Accuracy: %0.2f (+/- %0.2f)\" % (scores.mean(), scores.std() * 2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}