GloVe_Eval.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# [Double-Hard Debias: Tailoring Word Embeddings for Gender Bias Mitigation](https://arxiv.org/abs/2005.00965)\n",
    "\n",
    "For more detailed explanations, please refer to the paper."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import codecs, os, json, operator, pickle\n",
    "from random import shuffle\n",
    "import numpy as np\n",
    "from numpy import linalg as LA\n",
    "import scipy\n",
    "\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def load_glove(path):\n",
    "    with open(path) as f:\n",
    "        lines = f.readlines()\n",
    "    \n",
    "    wv = []\n",
    "    vocab = []\n",
    "    for line in lines:\n",
    "        tokens = line.strip().split(\" \")\n",
    "        try:\n",
    "            assert len(tokens) == 301\n",
    "            vocab.append(tokens[0])\n",
    "            wv.append([float(elem) for elem in tokens[1:]])\n",
    "        except:\n",
    "            print(line)\n",
    "        \n",
    "    w2i = {w: i for i, w in enumerate(vocab)}\n",
    "    wv = np.array(wv).astype(float)\n",
    "    print(len(vocab), wv.shape, len(w2i))\n",
    "    \n",
    "    return wv, w2i, vocab\n",
    "\n",
    "def normalize(wv):\n",
    "    \n",
    "    # normalize vectors\n",
    "    norms = np.apply_along_axis(LA.norm, 1, wv)\n",
    "    wv = wv / norms[:, np.newaxis]\n",
    "    return wv\n",
    "\n",
    "def debias_gn(wv):\n",
    "    for v in wv:\n",
    "        assert(len(v) == 300)\n",
    "    \n",
    "    wv = wv[:,:-1]\n",
    "\n",
    "    for v in wv:\n",
    "        assert(len(v) == 299)\n",
    "    return wv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "322636 (322636, 300) 322636\n",
      "322636 (322636, 300) 322636\n"
     ]
    }
   ],
   "source": [
    "glove, glove_w2i, vocab = load_glove(\"./data/vectors.txt\")\n",
    "gn_glove, gn_glove_w2i, gn_vocab = load_glove(\"./data/vectors300.txt\")\n",
    "gn_glove_a = debias_gn(gn_glove)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "322636 (322636, 300) 322636\n"
     ]
    }
   ],
   "source": [
    "def load_glove_p(path):\n",
    "    \n",
    "    debiased_embeds = pickle.load(open(path, 'rb'))\n",
    "    wv = []\n",
    "    vocab = []\n",
    "    for w in debiased_embeds:\n",
    "        wv.append(np.array(debiased_embeds[w]))\n",
    "        vocab.append(str(w))\n",
    "        \n",
    "    w2i = {w: i for i, w in enumerate(vocab)}\n",
    "    wv = np.array(wv).astype(float)\n",
    "    print(len(vocab), wv.shape, len(w2i))\n",
    "        \n",
    "    return wv, w2i, vocab\n",
    "    \n",
    "dhd_glove, dhd_glove_w2i, dhd_vocab = load_glove_p(\"./data/glove_dhd.p\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "322636 (322636, 300) 322636\n",
      "322636 (322636, 300) 322636\n"
     ]
    }
   ],
   "source": [
    "hd_glove, hd_glove_w2i, hd_vocab = load_glove(\n",
    "    \"./data/vectors_hd.txt\")\n",
    "hd_glove_a, hd_glove_a_w2i, hd_vocab_a = load_glove(\n",
    "    \"./data/vectors_hd_a.txt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "322636 300\n",
      "\n",
      "322636 (322636, 300) 322636\n",
      "322636 300\n",
      "\n",
      "322636 (322636, 300) 322636\n"
     ]
    }
   ],
   "source": [
    "gp_glove, gp_glove_w2i, gp_vocab = load_glove(\n",
    "    \"./data/gp_glove.txt\")\n",
    "gp_gn_glove, gp_gn_glove_w2i, gp_gn_vocab = load_glove(\n",
    "    \"./data/gp_gn_glove.txt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def simi(a, b):\n",
    "    return 1-scipy.spatial.distance.cosine(a, b)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Word Embedding Utility: Analogy & Concept Categorization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from eval import evaluate_cate, evaluate_ana, evaluate_analogy_google, evaluate_analogy_msr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Calculating categorization benchmarks\n",
      "Sample data from AP, num of samples: 402 : \"['ball']\" is assigned class social_occasion\n",
      "exist 401 in 402\n",
      "Cluster purity on AP 0.5561097256857855\n",
      "Sample data from ESSLI_1a, num of samples: 44 : \"['mushroom']\" is assigned class green-vegetable\n",
      "exist 44 in 44\n",
      "Cluster purity on ESSLI_1a 0.7272727272727273\n",
      "Sample data from Battig, num of samples: 5231 : \"new york\" is assigned class state\n",
      "exist 4125 in 5231\n",
      "Cluster purity on Battig 0.4976969696969697\n",
      "Sample data from BLESS, num of samples: 200 : \"['acacia']\" is assigned class tree\n",
      "exist 200 in 200\n",
      "Cluster purity on BLESS 0.81\n"
     ]
    }
   ],
   "source": [
    "evaluate_cate(glove, glove_w2i, vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 158,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Calculating categorization benchmarks\n",
      "Sample data from AP, num of samples: 402 : \"['anger']\" is assigned class feeling\n",
      "exist 401 in 402\n",
      "Cluster purity on AP 0.6234413965087282\n",
      "Sample data from ESSLI_1a, num of samples: 44 : \"['mushroom']\" is assigned class green-vegetable\n",
      "exist 44 in 44\n",
      "Cluster purity on ESSLI_1a 0.7954545454545455\n",
      "Sample data from Battig, num of samples: 5231 : \"jazz\" is assigned class type_of_music\n",
      "exist 4125 in 5231\n",
      "Cluster purity on Battig 0.5003636363636363\n",
      "Sample data from BLESS, num of samples: 200 : \"['blouse']\" is assigned class clothing\n",
      "exist 200 in 200\n",
      "Cluster purity on BLESS 0.845\n",
      "Calculating categorization benchmarks\n",
      "Sample data from AP, num of samples: 402 : \"['anger']\" is assigned class feeling\n",
      "exist 401 in 402\n",
      "Cluster purity on AP 0.6408977556109725\n",
      "Sample data from ESSLI_1a, num of samples: 44 : \"['mushroom']\" is assigned class green-vegetable\n",
      "exist 44 in 44\n",
      "Cluster purity on ESSLI_1a 0.7954545454545455\n",
      "Sample data from Battig, num of samples: 5231 : \"jazz\" is assigned class type_of_music\n",
      "exist 4125 in 5231\n",
      "Cluster purity on Battig 0.49163636363636365\n",
      "Sample data from BLESS, num of samples: 200 : \"['blouse']\" is assigned class clothing\n",
      "exist 200 in 200\n",
      "Cluster purity on BLESS 0.845\n"
     ]
    }
   ],
   "source": [
    "evaluate_cate(hd_glove, hd_glove_w2i, hd_vocab)\n",
    "evaluate_cate(hd_glove_a, hd_glove_a_w2i, hd_vocab_a)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 159,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Calculating categorization benchmarks\n",
      "Sample data from AP, num of samples: 402 : \"['anger']\" is assigned class feeling\n",
      "exist 401 in 402\n",
      "Cluster purity on AP 0.5685785536159601\n",
      "Sample data from ESSLI_1a, num of samples: 44 : \"['mushroom']\" is assigned class green-vegetable\n",
      "exist 44 in 44\n",
      "Cluster purity on ESSLI_1a 0.7045454545454546\n",
      "Sample data from Battig, num of samples: 5231 : \"jazz\" is assigned class type_of_music\n",
      "exist 4125 in 5231\n",
      "Cluster purity on Battig 0.4945454545454545\n",
      "Sample data from BLESS, num of samples: 200 : \"['blouse']\" is assigned class clothing\n",
      "exist 200 in 200\n",
      "Cluster purity on BLESS 0.85\n",
      "Calculating categorization benchmarks\n",
      "Sample data from AP, num of samples: 402 : \"['anger']\" is assigned class feeling\n",
      "exist 401 in 402\n",
      "Cluster purity on AP 0.5685785536159601\n",
      "Sample data from ESSLI_1a, num of samples: 44 : \"['mushroom']\" is assigned class green-vegetable\n",
      "exist 44 in 44\n",
      "Cluster purity on ESSLI_1a 0.75\n",
      "Sample data from Battig, num of samples: 5231 : \"jazz\" is assigned class type_of_music\n",
      "exist 4125 in 5231\n",
      "Cluster purity on Battig 0.5127272727272727\n",
      "Sample data from BLESS, num of samples: 200 : \"['blouse']\" is assigned class clothing\n",
      "exist 200 in 200\n",
      "Cluster purity on BLESS 0.8250000000000001\n"
     ]
    }
   ],
   "source": [
    "evaluate_cate(gn_glove, gn_glove_w2i, gn_vocab)\n",
    "evaluate_cate(gn_glove_a, gn_glove_w2i, gn_vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Calculating categorization benchmarks\n",
      "Sample data from AP, num of samples: 402 : \"['ball']\" is assigned class social_occasion\n",
      "exist 401 in 402\n",
      "Cluster purity on AP 0.5610972568578554\n",
      "Sample data from ESSLI_1a, num of samples: 44 : \"['mushroom']\" is assigned class green-vegetable\n",
      "exist 44 in 44\n",
      "Cluster purity on ESSLI_1a 0.7272727272727273\n",
      "Sample data from Battig, num of samples: 5231 : \"new york\" is assigned class state\n",
      "exist 4125 in 5231\n",
      "Cluster purity on Battig 0.5081212121212121\n",
      "Sample data from BLESS, num of samples: 200 : \"['acacia']\" is assigned class tree\n",
      "exist 200 in 200\n",
      "Cluster purity on BLESS 0.785\n",
      "Calculating categorization benchmarks\n",
      "Sample data from AP, num of samples: 402 : \"['ball']\" is assigned class social_occasion\n",
      "exist 401 in 402\n",
      "Cluster purity on AP 0.6109725685785536\n",
      "Sample data from ESSLI_1a, num of samples: 44 : \"['mushroom']\" is assigned class green-vegetable\n",
      "exist 44 in 44\n",
      "Cluster purity on ESSLI_1a 0.7272727272727273\n",
      "Sample data from Battig, num of samples: 5231 : \"new york\" is assigned class state\n",
      "exist 4125 in 5231\n",
      "Cluster purity on Battig 0.5023030303030303\n",
      "Sample data from BLESS, num of samples: 200 : \"['acacia']\" is assigned class tree\n",
      "exist 200 in 200\n",
      "Cluster purity on BLESS 0.775\n",
      "Calculating categorization benchmarks\n",
      "Sample data from AP, num of samples: 402 : \"['ball']\" is assigned class social_occasion\n",
      "exist 401 in 402\n",
      "Cluster purity on AP 0.5960099750623441\n",
      "Sample data from ESSLI_1a, num of samples: 44 : \"['mushroom']\" is assigned class green-vegetable\n",
      "exist 44 in 44\n",
      "Cluster purity on ESSLI_1a 0.7727272727272727\n",
      "Sample data from Battig, num of samples: 5231 : \"new york\" is assigned class state\n",
      "exist 4125 in 5231\n",
      "Cluster purity on Battig 0.4761212121212121\n",
      "Sample data from BLESS, num of samples: 200 : \"['acacia']\" is assigned class tree\n",
      "exist 200 in 200\n",
      "Cluster purity on BLESS 0.795\n"
     ]
    }
   ],
   "source": [
    "evaluate_cate(gp_glove, gp_glove_w2i, gp_vocab)\n",
    "evaluate_cate(gp_gn_glove, gp_gn_glove_w2i, gp_gn_vocab)\n",
    "evaluate_cate(dhd_glove, dhd_glove_w2i, dhd_vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 151,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "6750\n",
      "ACCURACY TOP1-MSR: 53.94% (3641/6750)\n",
      "capital-common-countries.txt:\n",
      "ACCURACY TOP1: 96.44% (488/506)\n",
      "capital-world.txt:\n",
      "ACCURACY TOP1: 96.35% (4359/4524)\n",
      "currency.txt:\n",
      "ACCURACY TOP1: 6.47% (56/866)\n",
      "city-in-state.txt:\n",
      "ACCURACY TOP1: 72.68% (1793/2467)\n",
      "family.txt:\n",
      "ACCURACY TOP1: 53.95% (273/506)\n",
      "gram1-adjective-to-adverb.txt:\n",
      "ACCURACY TOP1: 20.97% (208/992)\n",
      "gram2-opposite.txt:\n",
      "ACCURACY TOP1: 28.82% (234/812)\n",
      "gram3-comparative.txt:\n",
      "ACCURACY TOP1: 85.36% (1137/1332)\n",
      "gram4-superlative.txt:\n",
      "ACCURACY TOP1: 48.93% (549/1122)\n",
      "gram5-present-participle.txt:\n",
      "ACCURACY TOP1: 63.26% (668/1056)\n",
      "gram6-nationality-adjective.txt:\n",
      "ACCURACY TOP1: 91.99% (1471/1599)\n",
      "gram7-past-tense.txt:\n",
      "ACCURACY TOP1: 56.54% (882/1560)\n",
      "gram8-plural.txt:\n",
      "ACCURACY TOP1: 75.68% (1008/1332)\n",
      "gram9-plural-verbs.txt:\n",
      "ACCURACY TOP1: 58.28% (507/870)\n",
      "Questions seen/total: 100.00% (19544/19544)\n",
      "Semantic accuracy: 78.58%  (6969/8869)\n",
      "Syntactic accuracy: 62.43%  (6664/10675)\n",
      "Total accuracy: 69.76%  (13633/19544)\n",
      "Analogy prediction accuracy on SemEval2012 0.17625085732716383\n",
      "6750\n",
      "ACCURACY TOP1-MSR: 51.90% (3503/6750)\n",
      "capital-common-countries.txt:\n",
      "ACCURACY TOP1: 96.44% (488/506)\n",
      "capital-world.txt:\n",
      "ACCURACY TOP1: 96.22% (4353/4524)\n",
      "currency.txt:\n",
      "ACCURACY TOP1: 7.04% (61/866)\n",
      "city-in-state.txt:\n",
      "ACCURACY TOP1: 65.59% (1618/2467)\n",
      "family.txt:\n",
      "ACCURACY TOP1: 72.92% (369/506)\n",
      "gram1-adjective-to-adverb.txt:\n",
      "ACCURACY TOP1: 20.67% (205/992)\n",
      "gram2-opposite.txt:\n",
      "ACCURACY TOP1: 27.71% (225/812)\n",
      "gram3-comparative.txt:\n",
      "ACCURACY TOP1: 82.43% (1098/1332)\n",
      "gram4-superlative.txt:\n",
      "ACCURACY TOP1: 48.93% (549/1122)\n",
      "gram5-present-participle.txt:\n",
      "ACCURACY TOP1: 57.77% (610/1056)\n",
      "gram6-nationality-adjective.txt:\n",
      "ACCURACY TOP1: 91.31% (1460/1599)\n",
      "gram7-past-tense.txt:\n",
      "ACCURACY TOP1: 56.28% (878/1560)\n",
      "gram8-plural.txt:\n",
      "ACCURACY TOP1: 77.33% (1030/1332)\n",
      "gram9-plural-verbs.txt:\n",
      "ACCURACY TOP1: 59.43% (517/870)\n",
      "Questions seen/total: 100.00% (19544/19544)\n",
      "Semantic accuracy: 77.68%  (6889/8869)\n",
      "Syntactic accuracy: 61.56%  (6572/10675)\n",
      "Total accuracy: 68.88%  (13461/19544)\n",
      "Analogy prediction accuracy on SemEval2012 0.17454325679918348\n",
      "6750\n",
      "ACCURACY TOP1-MSR: 51.29% (3462/6750)\n",
      "capital-common-countries.txt:\n",
      "ACCURACY TOP1: 96.64% (489/506)\n",
      "capital-world.txt:\n",
      "ACCURACY TOP1: 95.56% (4323/4524)\n",
      "currency.txt:\n",
      "ACCURACY TOP1: 6.12% (53/866)\n",
      "city-in-state.txt:\n",
      "ACCURACY TOP1: 75.03% (1851/2467)\n",
      "family.txt:\n",
      "ACCURACY TOP1: 85.38% (432/506)\n",
      "gram1-adjective-to-adverb.txt:\n",
      "ACCURACY TOP1: 22.18% (220/992)\n",
      "gram2-opposite.txt:\n",
      "ACCURACY TOP1: 26.48% (215/812)\n",
      "gram3-comparative.txt:\n",
      "ACCURACY TOP1: 84.38% (1124/1332)\n",
      "gram4-superlative.txt:\n",
      "ACCURACY TOP1: 46.88% (526/1122)\n",
      "gram5-present-participle.txt:\n",
      "ACCURACY TOP1: 62.97% (665/1056)\n",
      "gram6-nationality-adjective.txt:\n",
      "ACCURACY TOP1: 92.18% (1474/1599)\n",
      "gram7-past-tense.txt:\n",
      "ACCURACY TOP1: 53.46% (834/1560)\n",
      "gram8-plural.txt:\n",
      "ACCURACY TOP1: 75.90% (1011/1332)\n",
      "gram9-plural-verbs.txt:\n",
      "ACCURACY TOP1: 59.43% (517/870)\n",
      "Questions seen/total: 100.00% (19544/19544)\n",
      "Semantic accuracy: 80.60%  (7148/8869)\n",
      "Syntactic accuracy: 61.70%  (6586/10675)\n",
      "Total accuracy: 70.27%  (13734/19544)\n",
      "Analogy prediction accuracy on SemEval2012 0.1742715205641269\n",
      "6750\n",
      "ACCURACY TOP1-MSR: 51.81% (3497/6750)\n",
      "capital-common-countries.txt:\n",
      "ACCURACY TOP1: 96.84% (490/506)\n",
      "capital-world.txt:\n",
      "ACCURACY TOP1: 96.57% (4369/4524)\n",
      "currency.txt:\n",
      "ACCURACY TOP1: 5.89% (51/866)\n",
      "city-in-state.txt:\n",
      "ACCURACY TOP1: 66.03% (1629/2467)\n",
      "family.txt:\n",
      "ACCURACY TOP1: 68.58% (347/506)\n",
      "gram1-adjective-to-adverb.txt:\n",
      "ACCURACY TOP1: 20.46% (203/992)\n",
      "gram2-opposite.txt:\n",
      "ACCURACY TOP1: 27.83% (226/812)\n",
      "gram3-comparative.txt:\n",
      "ACCURACY TOP1: 84.01% (1119/1332)\n",
      "gram4-superlative.txt:\n",
      "ACCURACY TOP1: 48.40% (543/1122)\n",
      "gram5-present-participle.txt:\n",
      "ACCURACY TOP1: 56.44% (596/1056)\n",
      "gram6-nationality-adjective.txt:\n",
      "ACCURACY TOP1: 91.62% (1465/1599)\n",
      "gram7-past-tense.txt:\n",
      "ACCURACY TOP1: 56.47% (881/1560)\n",
      "gram8-plural.txt:\n",
      "ACCURACY TOP1: 76.05% (1013/1332)\n",
      "gram9-plural-verbs.txt:\n",
      "ACCURACY TOP1: 62.18% (541/870)\n",
      "Questions seen/total: 100.00% (19544/19544)\n",
      "Semantic accuracy: 77.64%  (6886/8869)\n",
      "Syntactic accuracy: 61.70%  (6587/10675)\n",
      "Total accuracy: 68.94%  (13473/19544)\n",
      "Analogy prediction accuracy on SemEval2012 0.17776766097974608\n"
     ]
    }
   ],
   "source": [
    "evaluate_ana(hd_glove_a, hd_glove_a_w2i, hd_vocab_a)\n",
    "evaluate_ana(gn_glove_a, gn_glove_w2i, gn_vocab)\n",
    "evaluate_ana(gp_glove, gp_glove_w2i, gp_vocab)\n",
    "evaluate_ana(gp_gn_glove, gp_gn_glove_w2i, gp_gn_vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "6750\n",
      "ACCURACY TOP1-MSR: 53.81% (3632/6750)\n",
      "capital-common-countries.txt:\n",
      "ACCURACY TOP1: 96.44% (488/506)\n",
      "capital-world.txt:\n",
      "ACCURACY TOP1: 96.68% (4374/4524)\n",
      "currency.txt:\n",
      "ACCURACY TOP1: 6.00% (52/866)\n",
      "city-in-state.txt:\n",
      "ACCURACY TOP1: 74.67% (1842/2467)\n",
      "family.txt:\n",
      "ACCURACY TOP1: 83.60% (423/506)\n",
      "gram1-adjective-to-adverb.txt:\n",
      "ACCURACY TOP1: 21.07% (209/992)\n",
      "gram2-opposite.txt:\n",
      "ACCURACY TOP1: 28.33% (230/812)\n",
      "gram3-comparative.txt:\n",
      "ACCURACY TOP1: 84.46% (1125/1332)\n",
      "gram4-superlative.txt:\n",
      "ACCURACY TOP1: 47.42% (532/1122)\n",
      "gram5-present-participle.txt:\n",
      "ACCURACY TOP1: 62.88% (664/1056)\n",
      "gram6-nationality-adjective.txt:\n",
      "ACCURACY TOP1: 91.99% (1471/1599)\n",
      "gram7-past-tense.txt:\n",
      "ACCURACY TOP1: 54.62% (852/1560)\n",
      "gram8-plural.txt:\n",
      "ACCURACY TOP1: 76.43% (1018/1332)\n",
      "gram9-plural-verbs.txt:\n",
      "ACCURACY TOP1: 55.06% (479/870)\n",
      "Questions seen/total: 100.00% (19544/19544)\n",
      "Semantic accuracy: 80.94%  (7179/8869)\n",
      "Syntactic accuracy: 61.64%  (6580/10675)\n",
      "Total accuracy: 70.40%  (13759/19544)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/zf15/tw8cb/anaconda3/lib/python3.7/site-packages/web-0.0.1-py3.7.egg/web/evaluate.py:158: FutureWarning: arrays to stack must be passed as a \"sequence\" type such as list or tuple. Support for non-sequence iterables such as generators is deprecated as of NumPy 1.16 and will raise an error in the future.\n",
      "  prot_left = np.mean(np.vstack(w.get(word, mean_vector) for word in prototypes[:, 0]), axis=0)\n",
      "/zf15/tw8cb/anaconda3/lib/python3.7/site-packages/web-0.0.1-py3.7.egg/web/evaluate.py:159: FutureWarning: arrays to stack must be passed as a \"sequence\" type such as list or tuple. Support for non-sequence iterables such as generators is deprecated as of NumPy 1.16 and will raise an error in the future.\n",
      "  prot_right = np.mean(np.vstack(w.get(word, mean_vector) for word in prototypes[:, 1]), axis=0)\n",
      "/zf15/tw8cb/anaconda3/lib/python3.7/site-packages/web-0.0.1-py3.7.egg/web/evaluate.py:162: FutureWarning: arrays to stack must be passed as a \"sequence\" type such as list or tuple. Support for non-sequence iterables such as generators is deprecated as of NumPy 1.16 and will raise an error in the future.\n",
      "  question_left, question_right = np.vstack(w.get(word, mean_vector) for word in questions[:, 0]), \\\n",
      "/zf15/tw8cb/anaconda3/lib/python3.7/site-packages/web-0.0.1-py3.7.egg/web/evaluate.py:163: FutureWarning: arrays to stack must be passed as a \"sequence\" type such as list or tuple. Support for non-sequence iterables such as generators is deprecated as of NumPy 1.16 and will raise an error in the future.\n",
      "  np.vstack(w.get(word, mean_vector) for word in questions[:, 1])\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Analogy prediction accuracy on SemEval2012 0.17303299182822923\n"
     ]
    }
   ],
   "source": [
    "evaluate_ana(dhd_glove, dhd_glove_w2i, dhd_vocab)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Gender Bias Evaluation:\n",
    "### Restrict vocabulary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 50000/50000 [00:00<00:00, 274368.85it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "442\n",
      "1883\n",
      "size of vocabulary: 47628\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "from tqdm import tqdm\n",
    "from utils import limit_vocab\n",
    "\n",
    "gender_specific = []\n",
    "\n",
    "with open('./data/male_word_file.txt') as f:\n",
    "    for l in f:\n",
    "        gender_specific.append(l.strip())\n",
    "with open('./data/female_word_file.txt') as f:\n",
    "    for l in f:\n",
    "        gender_specific.append(l.strip())\n",
    "print(len(gender_specific))\n",
    "\n",
    "with codecs.open('./data/gender_specific_full.json') as f:\n",
    "    gender_specific.extend(json.load(f))\n",
    "\n",
    "\n",
    "exclude_words = gender_specific\n",
    "print(len(exclude_words))\n",
    "\n",
    "glove_vocab_limit, glove_limit, glove_w2i_limit = limit_vocab(glove, glove_w2i, vocab, exclude = exclude_words)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 50000/50000 [00:00<00:00, 389222.82it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "size of vocabulary: 47628\n"
     ]
    }
   ],
   "source": [
    "gn_vocab_limit, gn_limit, gn_w2i_limit = limit_vocab(gn_glove, gn_glove_w2i, gn_vocab, exclude = exclude_words)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 50000/50000 [00:00<00:00, 286433.57it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "size of vocabulary: 47628\n"
     ]
    }
   ],
   "source": [
    "gn_a_vocab_limit, gn_a_limit, gn_a_w2i_limit = limit_vocab(\n",
    "    gn_glove_a, gn_glove_w2i, gn_vocab, exclude = exclude_words)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 50000/50000 [00:00<00:00, 316926.29it/s]\n",
      "  0%|          | 0/50000 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "size of vocabulary: 47628\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 50000/50000 [00:00<00:00, 507142.77it/s]\n",
      "  0%|          | 0/50000 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "size of vocabulary: 47628\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 50000/50000 [00:00<00:00, 362967.85it/s]\n",
      "  0%|          | 0/50000 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "size of vocabulary: 47628\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 50000/50000 [00:00<00:00, 376660.59it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "size of vocabulary: 47628\n"
     ]
    }
   ],
   "source": [
    "gp_vocab_limit, gp_limit, gp_w2i_limit = limit_vocab(\n",
    "    gp_glove, gp_glove_w2i, gp_vocab, exclude = exclude_words)\n",
    "gp_gn_vocab_limit, gp_gn_limit, gp_gn_w2i_limit = limit_vocab(\n",
    "    gp_gn_glove, gp_gn_glove_w2i, gp_gn_vocab, exclude = exclude_words)\n",
    "\n",
    "hd_vocab_limit, hd_limit, hd_w2i_limit = limit_vocab(\n",
    "    hd_glove, hd_glove_w2i, hd_vocab, exclude = exclude_words)\n",
    "hd_a_vocab_limit, hd_a_limit, hd_a_w2i_limit = limit_vocab(\n",
    "    hd_glove_a, hd_glove_a_w2i, hd_vocab_a, exclude = exclude_words)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 50000/50000 [00:00<00:00, 499453.90it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "size of vocabulary: 47628\n"
     ]
    }
   ],
   "source": [
    "dhd_vocab_limit, dhd_limit, dhd_w2i_limit = limit_vocab(\n",
    "    dhd_glove, dhd_glove_w2i, dhd_vocab, exclude = exclude_words)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Select originally top biased words"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 196,
   "metadata": {},
   "outputs": [],
   "source": [
    "he_embed = glove[glove_w2i['he'], :]\n",
    "she_embed = glove[glove_w2i['she'], :]\n",
    "\n",
    "def simi(a, b):\n",
    "    return 1-scipy.spatial.distance.cosine(a, b)\n",
    "\n",
    "def compute_bias_by_projection(wv, w2i, vocab):\n",
    "    d = {}\n",
    "    for w in vocab:\n",
    "        u = wv[w2i[w], :]\n",
    "        d[w] = simi(u, he_embed) - simi(u, she_embed)\n",
    "    return d\n",
    "\n",
    "gender_bias_bef = compute_bias_by_projection(glove_limit, glove_w2i_limit, glove_vocab_limit)\n",
    "sorted_g = sorted(gender_bias_bef.items(), key=operator.itemgetter(1))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### TSNE visualization of originally top biased words"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 243,
   "metadata": {},
   "outputs": [],
   "source": [
    "def visualize(vectors, y_true, y_pred, ax, title, random_state):\n",
    "    \n",
    "    # perform TSNE\n",
    "    vectors = normalize(vectors)\n",
    "    X_embedded = TSNE(n_components=2, random_state=random_state).fit_transform(vectors)\n",
    "    for x,p,y in zip(X_embedded, y_pred, y_true):\n",
    "        if y:\n",
    "            ax.scatter(x[0], x[1], marker = '.', c = 'c')\n",
    "        else:\n",
    "            ax.scatter(x[0], x[1], marker = 'x', c = 'darkviolet')\n",
    "    \n",
    "    return ax\n",
    "\n",
    "def cluster_and_visualize(words, X1, title, random_state, tsne_random_state, y_true, num=2):\n",
    "    \n",
    "    kmeans_1 = KMeans(n_clusters=num, random_state=random_state).fit(X1)\n",
    "    y_pred_1 = kmeans_1.predict(X1)\n",
    "    correct = [1 if item1 == item2 else 0 for (item1,item2) in zip(y_true, y_pred_1) ]\n",
    "    print('precision', max(sum(correct)/float(len(correct)), 1 - sum(correct)/float(len(correct))))\n",
    "    \n",
    "    fig, axs = plt.subplots(1, 1, figsize=(6, 3))\n",
    "    ax1 = visualize(X1, y_true, y_pred_1, axs, title, tsne_random_state)\n",
    "    \n",
    "#     fig.savefig(\"a_{}_{}_{}.pdf\".format(title, size, random_state))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.cluster import KMeans\n",
    "from sklearn.manifold import TSNE\n",
    "from utils import extract_vectors\n",
    "import operator\n",
    "import matplotlib as mpl\n",
    "import matplotlib.pyplot as plt\n",
    "from cycler import cycler\n",
    "%matplotlib inline\n",
    "\n",
    "size=200\n",
    "female = [item[0] for item in sorted_g[:size]]\n",
    "male = [item[0] for item in sorted_g[-size:]]\n",
    "y_true = [1]*size + [0]*size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 245,
   "metadata": {},
   "outputs": [],
   "source": [
    "# You may change the random states to get more visualization result\n",
    "random_state = 0\n",
    "tsne_random_state = 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 246,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 1.0\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cluster_and_visualize(male + female, extract_vectors(male + female, glove_limit, glove_w2i_limit), \n",
    "                          'GloVe', random_state, tsne_random_state, y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 247,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 0.7775\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX0AAADCCAYAAAClkhJFAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjMsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+AADFEAAAgAElEQVR4nO29e3Ac13no+TszJB58CE8CIGVJJCVZ17IIwuEDokg6TOysFd9YjiPJ692qK8d2Xd3YEsGHthKrvBFBZ13xraz5gGTL0d7YsaqS65iSvVZyvXFsl2mRFAU+ZBCkrOuQIilZBgYgnpJIDEDOnP2j+zRP93TP9GBmMAPM+VWxiJnp6T6nZ+Y73/meQkqJwWAwGMqDSLEHYDAYDIaZwwh9g8FgKCOM0DcYDIYywgh9g8FgKCOM0DcYDIYywgh9g8FgKCPmFXsAOo2NjXL58uXFHobBYDDMKk6ePDkkpVwS5tiSEvrLly/nxIkTxR6GwWAwzCqEEG+EPdaYdwwGg6GMMELfYDAYyggj9A0Gg6GMMELfYDAYyggj9MsYb7E9U3zPYJj7GKFfphzrHODwjpgj6KWUHN4Ro3tXzHWcWQgMhrmFEfolxExp3lJKJseS9O4fdgT/4R0xevcPc/GFd0kmk85xh3fEONY5kPX50z02GAzFwwj9PDNdgRekeWcrcMOMQQjBpr0ttG5roHf/MN+IvErv/mEa26oY6olzZOeAayGYHEsWZR4GgyH/GKGfR6Yr8NJp3tkI3GzGoAS/zoMnV6YsBK3bGti4pxkhhGu83vHnex4Gg6EwGKGfJ3IReEGad+u2BjbtbXEJ3HyNQb2mc2TnABv3NLueq6gRjuavv08tIvoiI4Rg455mGtuqcpqHwWAoHCVVhmE2o2vOvfuH6d0/DBBa4Kn3q/cBKe9TgtX7WP2vziGldI1hVUe961zJZJIjOwdcAlktDn2/uOwa18UX3mWoJ+6MRx3Xuq2BZDLpLDLq9SM7B5zjg+ZhMBiKh9H084ifycRPcOt4NWgd3UwTZLb5wZbzrucBvOJVfyyltAXzhCPwAUdDH+qJs6qjni8k30/rtgaGeuKBmnskEgnwDVS657G935h2DIYSwQj9PDJdwd29K+bSoJXAVWYaXaP2mm2mxj3Pb++nt2vENYberhFH8Kr3NbZVs3FPM8d3D3J4RwwhBMvvW8yqjnoEcHz3oCPQl9+32HU+70LmXeiGeiZpVQtHRz29XSM8v+G8EfwGQwlghH5IMkXE6AI1e8EtqaiJuDRvJXArayNpbf4pzldb4K/qqOfziTtZ1VEPWILfq6kDrjGt72xC2MdOjllhmxv3NDM1nnTN9ZC9gKhF5rm7X0+5X0m1sKn7M417ajAY8o8opR/a2rVrZSmWVj7WOcDkWNLRcJXArqyNsL6zOfRxuqBX6DZ/KSXHdw8651Coc6zb1cQ3Iq86z38h+X7nb/35VR31VNZGmBqXbNzTzOEdMU5r2r963+EdMSpqBFPj0ndMyhTUu3/Y2QHoiwrgnHfV1jo271/Goe39rmupYzfvW+raHYS9pwaDITNCiJNSyrVhjs3ZkSuEqAJeBCrt8z0npdwlhFgBfBeoB14B/pOUcirX6800ekQMpDozdWeqEuy6c1U3hYRx1gZda1VHPYe297vG9tyG8zStryLicZIOvHyF5rsXcLprxNae3Qu7Os/prhEnJNM7JsDX9g+W4FeCvbolykQsAWkctV6BH/aeGgyG/JOzpi+sX+dCKeW7Qoj5wGFgG7AT+L6U8rtCiG8Cp6SUT6c7V6lq+pk09LBaaxhNH0g5ZlVHPQPdVxjsjtPaUc+mfUtTNOoUzbujDinhzJOjzjENqysZPjXpevzgyZW8tHPA5Qdo7ahH4l4UIpGIMz59V/H5xJ0pOwnvdfRzKPQIIr97YTAYwpONpp+zTV9avGs/nG//k8DvA8/Zz38H+ONcr1UsVPy5jkpYChsb77X5r32i0YmKUTb/wztijgNVx3psh1t6Ful5C63nT9va96qOenuRiFNVF3Ud+8lXbnU9Hj41yXNrL9DbNUKDFnGjNPmG1ZUugX9oez/PbzjvOseRnQNU1ERSztvYVsXaJxodf8OBNeed8g5K4FfUuIV7piQwg8GQO3lx5AohokKIHmAQ+AnwOjAmpbxmH/IWcGPAex8WQpwQQpy4dOlSPoaTd7p3xTiwxi3sDqw5T/euWFaJVZW1EUfrnRxLOuGQFTURR+uNj1xLMeMc2TnA/S+tYMn6Ks48Oco3Iq9yumuEu7bW8R8+W+s6Vi0Yg90TXPjhO67Xvvc7bodrw+oKJ6Z+uGeSVVvrXK8Pn5p0ErPUzmKge+J6ZI4951efHnW9r6oxylBPnKlxybzF1x8f2TlAMpnkwJrzTp0f7z3Nte6PwWBIT14duUKIWuAHwBPAt6WUt9nP3wT8SEq5Kt37i23e8Ut+klJyYM15R0A/eHJlymOlCXudrOpcuvlHcXh7P7HuCQa7J5znmtZXkZiSDPdMOouDupZlvpGc7rouYO/aWucy3+g0tlUy1DPpOqeioa2SicFrXOlLuN7T0FbpOk7F7Tvja6+mpb2ajXaMvhLg6l7oxza2VfHAiRW89NggvfuHU0w+VY1R4kOJlHmqe+pNHjMmH4MhmBk17+hIKceAg8DdQK0QQjmK3wP05fNa+SYohv7Ely+x/L7FjlB7OvorRzgtv2+xI/CD4vO95h/1Wm/XCC3t1a73tNy9gOEeyzSycU+zk93a2FbFpV9O0PeLK67jlcBvaq/iLo+WLuZbi4g6p85wzyTVS1J9+MM9kzS0VfL5xJ1OYpbOA0dXstHOuj3WOUAkEmH5fYuc0FGd5fctIhqNsmlvC03t1S6BD3DDyvk0t1c7SV4PnlzpusdG4BsMhSFnoS+EWGJr+AghqoEPA68BPwcesA/7NPDDXK9VKDLZ5dd3NqUItQdPrqR9d3OKrd4bnw/4m39sZ6lrHFihj7rgW9VRz/3Hl3PtHcnwqUkaVruzXasaIzSurUzR+C8dn6SpvZp77Exb93uitGxyP6dYtnmBVkMnNbNWzTU+mkBKSfvuFmeB0pkal84C2uxZ3AAGj8VpWn99DErw6xiBbzDkn3xo+kuBnwsheoHjwE+klP8C/AWwUwhxDmgA/i4P1yoImezyQIpQU7ZuIYRjq9dLGqjEKoXXOSuldKJjVObq6a4R+g65tXkBvPJ/DbP8vkUpJhKAxSvmBwpGAbzkUwsnPpTgwvPvctejbn/AXY/WUlU/TyvVYGXWLrGFs3Lw3rW1zsna9cby64veoe39JJPJlHo+2o133Q/vPfaWlzAYDLmTc5y+lLIX+IDP8+eB9bmef6YIiqEHXJq8HlOujlHx+cd3D/LGv77r2L2FEBza3s9A9xW8FXHOHXibVVvrnGskpaS6OZpiiuntGqG1o56Ne1uYHEu6hP4qe6EYeXWKJeuruHTMbX9XAtxrb69qjHIlluDMU2PuexCJsPaJJa74/I17mjmyI+Y6d0QIa1zbGgDLQd3UXu3MUCV2DXRP8P2NFxk+NUl9awUjvdfTNOpbK5xsY+9uye8eG43fYMgPpgyDTZBdHnBp8vquQAktRXw0wWD3BL1dIxzZEXMiXga74wx2TzilERrbqpiIJeg7dMW57pknR6ludodY6mMDiL3s3gXIZJKqxiiJK5J3zl91vZaIJ1nzl41OlJDaUazqqCc+5HbgguV4Pd014kTWqNo8QogUM5RaiNT9WLeriZb2aqvGj33PBFYEkQCWrKtERNxCe2IwgVoIhRAM9Uw4vgy9RPNQz4QR+AZDHjFCn8x1c9btanLFkCuhpCdeCSHYvG+pq9aNSlhqaq92ShHoTsvhnknHdt/UXs28qkhK2OSCZVEGuic4bGvb73+01rG1n3lqjPhQwomEUTS2VTF8apLvb7zIgL3YOOYlKalucS8u+k5ALRBKoKvEKzUv555pfwsh2LRvaUoNoNZtDXziyHLkVRwHtbXoVTIRS3DxhXdIJpP2bqTa1bXruhO72ph4DIY8YurpQ4pdXi+RXFkbcdXDUeaIIzsHUjJuleD31p554OhK53W47rR8Ovor5xilKSem3AIuOWU5PYUQTv2bIY8JqPZ984kdui70HzixggNrznP5ratc/u01J0ro0PZ+TtsOX2Uy8mbFgjtJqrLWWoi8urYeaqrmFlRiwqrSKRwHNWBHPy1ysnRz6UVgMBjCYzR9m/Wdzb51ctbtasqqG5U3sQpS68n7mZKSUtLoEycfH0rQ2FbFQPcEp7tG6O0aSYnGGXvNbdp5bu0Fhk9NsuJPFjuljVVCV1N7tVPKIRKJOGYUHT1Jau0TS+g7dMXR3D+fuJPWjnpn9+FNpnLN275f7btbAqKfrju3w/QiMBgMuWOEvoZXwOjdqDJl3OpZq2Bp0rqp58Ca1x1Thjquqb3aMSWdeXKUSIX7+rd8bKGjEesoE4zyDygTj/56dXPUMkN5BOnNH1nIJrsAmm5G0c831BN3BP+RnQOOaUa1Ttxo3w/1uHvXgMs8puL81UKZSCQCo58UmXoRGAyG/GCEfgjCaKFCCKrqoi5NevO+pdy1tY4FLVGGeiYdwadMI8rssmlvi2W68Sw6b/zzZYZ64inlFKoao9zztSYnOaqxrYq7vuD2BSxeXuE4ZnUuvvCuI0i9Zq2gJKlVHfU8cGKFU4L58I6YXWdfOo1cKmqEvRBEnF67rdsamH+DsOr7BPhLnLr8Xp9KR33KMQaDIXeMTT8EQVqot3Lkul1NrNvVBFzfNXxw/zI271vq2M4de7W9MDhlmIGB7gmn1IGqclnZEHGFaSqB/NJjVmG29t0trPnLaxz9P9x1i5rWV5GcTKaURlDOUrVo6eWgwd/fIOznVSjmac1Jre94vJUzvVnFyleg+0t034E61/Hdg0j7HqlcB1Nr32DID0bTz0C6yB6/AmHHdw+m7ACUwNTZpNWYF0JQWRdlVUe949AVWHH40Sq39q86ZSmB2b1rgOfXXXSZVhrbqnzNRd736mPU5+s1xah2i5Daf1ff8Xh75urlFFSNInU9ld+gUD4VsHoKnO4aQWItpI4Pxc4C1sdqMBiyw2j6GfCL7Lnna030/eKyozUrjTaoCUjQTkEJOV3jtp9IiahRqOupmj9T49erdd7ztSZHs25oq0R6IuyP7Bzgnq81EY0G5wN4k6SU/6HX/udFzSObRjHqOL97rY4HK5JH7Sia7d2Pd6xG+zcYssMI/RDoJhBVMVOvIJkuxFAXpLrppnf/sGPWqayLsr6z2SX0vE1UNu1tcRKn1DG6qaR3/zDfnPeaNY6OepJScubJURa0RPn0b+9wFqW+X1xm+X2Lad+dKij9FrjN+5YCpDRs2bxvqW/WbLoFLmwkjt/C0dRezWl7B6SHmrZ21JNMJl1mNoPBEIz5pQTgZzrQC7O99NhgYGMVHSVI/Uw3KnvX22wlJexTSo7siGn190VKophrHHtbGLCdxVdiCY7Y/gdl058aTzrOUa+5ZN2uplQntec+6OUWdHNRpiS3sOYYv4VDYPkpertGrpuN7MXt+/dcMHX3DYaQGE3fh3TtD/2SiBS6g1QnnenG2y7RyYDdWoewa9x4E6p0rdbPBq+csHdtrXPq5PR6HK96naBNtjb/4rY+Lh2Lc/O9i1jf2XzdMWvH6G/a28Lh7f3WuWxt3GvaSZfkFkbTT1eHx1thVO1mwDb/mN66BkNG8tEY/SbgWaAFSALPSCn3CyHqgX8ClgMXgU9KKf07fpQQmZp2AymNxD+fuNMVtZLOhp3O3q2EZnN7Ndgx9rodvaI2kiLw9bFt3NPsirpRDcn1cyghrOoEqfDR2MtXGLSLqjW3VzsCX2+MrsotYI/TT8B6o4GU4M/GtOO7cEiZUnvojLYYbvI0XzcYDP7kQ9O/BjwmpXxFCLEYOCmE+Anwp8DPpJRfFUJ8EfgiVrnlguHnQM1WEHjt5Lq9XplR/BKN1GvpNNow9u51u5qIjyY43TVC/4tuIXfxhXdZtyvpioLRWzB6x3Voez8Rz1jU9XRbvb4oNLZVum3m2xq452tNKeakdDb0ME7bdHgXDrBq/QweizsZxjobs1hUCkU+vnsGw0yQ13aJAEKIHwJP2f+2SCn7hRBLgYNSyjvSvTeXdom+Jpnt/Y6TFLL7IXrbH659YglTYwknfl5l2w7Y2rISvCqLVz9PpvLBXgew3oZQoezxfs5iPT5ejeu0R5AP2Vm1+jkA1xy9qGSrqXF/U1c+omb0z0RPGvO+dqxzgMnRhMuk44yzyJq+97unPg91j7xzNIuDId+ILNol5tWmL4RYjlVbvxtollL2A9iCvymf19LxM8k8v+G808RbCZOwwspPI7/4wjsM9cRptitmCq6XGG5ur6ayNsKJL18K9AWs72wObe+2Mm0Xu4S+6hvrt5OIRCIpjVzU+MAq0KbvBlTCk1+dIB09FFXdV32hylVg6cLy+O5B4qMJJ5pJxeere7duVxMvbutzmXTUAqDmWQzB7/3uzb/heiZ167YGEokELz026NzzdN8Pg2EmyJvQF0IsAp4Htksp387Chvsw8DDAzTffPN1rBzpY1T4mrLBKp5GrwmcqMkYXsvo1wF9AKudoJnt3Mplkajzpek6Pz/fDq1Fu2rc0xZav3ydvnSA/DVo3W+W7+qVLWErp7KDUeNT49M+run6eE/aqnM9CCAa7J6isixZFY/Z+9xQqb+K5tRecxvb6QpzvBdRgCEtezDtCiPnAvwA/llLusZ/7NTNo3oFUk4zqLKUIK6yConcqagQnvjzkHPeF5PsDY/IVDasr+eQrtzrJVIe291OlmZy8dO8acHYVSjtXpp6w4/cbh4qtV+/t3hXjzR9fdrp8HbGbtS9oiXLnw3VOXR0/B7F33tPFb5w6QXkP4DYB6Y+LRTKZdN0jnca2Kh48uRIhxPXoJxtTPtqQD7Ix7+SjMbrA6n/7mhL4Ni9gNUSHGWiMHhTbrRP2x+VXZlkVGNPxxp7rWp9i+NSkc5zSXt/413fxW2y9Gbbe2jV6fH4QfrHyKrHpkF3i2bqO5OaPLHJKLFfWRWntqOfTv72D9t0tTgy+qqSZbt7Txe9+6QRFQblyCETmewI45TKCHueCX9iszgMnVjg7NO9dMwLfMNPkIzlrI/CfgN8XQvTY/z4KfBX4AyHEWeAP7McFIbBKoyfKIxthFaTB620HvUlHfgsPWGYLVc8erlfX9ENVp9SrXKraNXr9+XTj9tr31fVUeKaai25CWt/Z7MoBEMIqN6Fr/NNNtgoi6H4p8rW4/GDLeVedJOUo/8GW8xnemRnvd+PPrr0v5Zjn1l4gkUi4TGoKUz7aMNPkozH6YVKVasWHcj1/GPwEnfoZNbVX88DRlS4zgtc2nsmeenz3oNPjVneUNtkO3KDoHL8feVCkybHOAd7813dpaq9OieWvqBFZlRnws+8DTjMVSDUr+Jm0XnpsMDVOP8tkqyBc96ujPsWmr0pVQG7asPKPqB4BD55c6ZjLGtuqci7hoH/3lA1fp6oxylBP3CmRATgRVoPdE6b5u2HGmTNlGHSTjBBWbftVHfU8cHSlI6wsDXrC1cBDCZ+gNH7lcBywm3w7wqprhOb2atY+sQSwfvwVNe6FZ9PelpQsUj+dTkrJ5GjC6Y71vTWvu16/YPeSzQavCUQJfoW3AUxQdzC9QbpzLk+FzCC8GqzXFOYs1PuWOp9Xa0c9lXVRNts9d3NdXPx6BCiBr1f+zIX1nc1s3NNsL5LWuf/s2vto3daQ0oReD6lVvRdynaPBkA15j9PPhVwduV68Grwez54pVt57Hq/Dsbm9miXrKhGRCFVaiKGyuwfFlEOqU1Vd49C2PqfkgqLBbqGYjcMv3bwVfp2/vHPMxcmYrpSFvmCEjdPPFa+j9fOJO0MJ/Gzi6tWc1U7SL99CoXZ8UHwntGH2M6OO3FLGL549TOtDv/PMv8H92pJ1lZx5aozTXSPERxNWVIbdN3dyNEFv1wjnD7wNWMlR+v/Ktp7iBPZo4wCfPHlrKI1XnetY54Crd60SPJns8n5O1ekK/HQ7B29f4XRO2XwKfG8HMd3GH4S6l2F3hWq3qSK1lBO+ub2au7a6O5slpUzpvWAwzARlV3BNCTe/+jdBWl0ymeSNf37XdZ4zT405f3u7SNkXcq7hSo7aEaOyLsrx3YMpWZwHPGYduF42IZ1WqmuYSthaJZQXMTUuQ3WuClMiIixBeROZdlPqvX6Pp4uubSuTjnqsbPx+9zaZTLqSrjL1TNDnrv6vrI0E5kCceXKUVT5loU28vqHQlJ3QDxJuQeUG1PNDPXGqW6JMxBIBZ3ZrxkELi6oT43UsH1jzOkM9VltEvTqmWlC85iA1F8ARTlJKNu1tcRq8KLOCWnC8nav8TDtesxdSuqJ6wgrjdIurl2OdAykVPw9t72ewe8Kp+DldIpEIFTURlw1fCf6KmoivwNcXUZh+YppTqTQgQmmg+wqHd8Scz9Zk6BpmgrIS+nqsvB5ho7JtlZD02vorakSgxqajd8NKWVi29zuFzCprI74CZUFLlJUP3sAH9y9z3hfrnqDKJ9tUt5lv2pvau1bHbyfjNaP4lYgYePkKrz//NhKcAm2Ht/cT657glgzCOOzOQUqZUvFTj+TJR8nkTxxc6dKoleD307C9ZRW8FVX9eiakQ8+B0EOIV3XUO457db9Nhq5hJigroZ8u9PLmjyxk2e8uDNTqEolEil1Y4WjmtrYNltBSC0l1c5TerhH6XrxsmXo66m2Nzt228KG33kskct3konYFfhq+t9ZQOvGgmrjrRcC8+FW2bFpf5Qgmda96QwjjtDsHUstJB1X8zGchNa9Grz/WF1CwBHvfLy5n1TMhHet2NfnWObr/pRWOySifJS4MhnSUjdDXQy+b7WQlFXrZ2lHvCMIgW/9Ljw0yfGqShtWVDJ+adI6pX11xvayCEAz1THAtLp2GJ4e39zsROUrgK022sa3KNcbn1l5w0vWVI9BPSAfZzHWqGqPEhxI0tlU5Nn5VyiGTPVr9vXn/MoTHzAT+0Ufe82TTTEUJfu8uZSYKqOkL6MDLV2hqr0ZASsRNmJ4JQedXjXG8+RtCiJSdhLfaapC931TrNEyXshH6KYLSx/mazhyhWh4CLqE/r+p6f1vd2dfcXu2rpavr6mWO9fo6yrmYyWnoZzNf1VFPZW3Uqd2jm6yyqd3juoancBv4+xe8ZNNMRZndvBze3l9wwa/GpcxjqpieCpVVHNkRY+M0EtOs/A3h6liGlKzqqKeiJpLynTu0vZ/N+5ZyrHOAiy+86/Qz1u39YKp1GqbPnA7Z9BIUlgjuKpx+YY3rdjUhwNHYVKmHwe4J5xg9JFSVXvAKTMXy+xY5gsAvgShs7oBrfsD6ziYePLmS1m0NLL9vUcpcsxGgUlo9CbyoGj6Z8Fuo/K7hrfipFtferhEOh7xWLhzfPZjy3LDmVFf2eFV1NBvBeqxzgKlx6fhw4qMJTj85ysDLV5gcSzjzVuG8p7tGOLStjwv2wn3RTsxT38/4aMIKCQ4RDmsw+FE2mj6kdy6GMUdU1kUztg701cC31oEQLvOFrqnBdeeinkAUJlksqHetXyetbEIwlcBXi5ZeGjhdRFG2CGFlT3tLJgMzUjJZOZL9HOCgzdH+nLPJ4E3nexk8FnfaUyofk6p0qsyBuhIA/iHBxhdgyJayEfphnItAWnNEGJOF38LSd+gKwz2TjuCMeaI2dBu+TpCQ9rWZawsQEJh5rOYaJuSysi5Kc3s1Te3VjrMVgiOKpotqkqKuC9ejhWZCiGVygm/etzS/+Qo+xQD9TGnplICw4bAGg5e8CH0hxLeAPwIGpZR32c+VVGP0bJ2L6j2ZnvMT+H4aeGNbleuHqWywuk02GyGdaQHKdq5++AnjQpUOCGMKKgRqcfP2XvALqZzu+b0C2muCOd01gvQp/OatwZQ2JHiaiXSG8iNfmv7fY/XEfVZ77ovMcGP0TGTjXJwO6TRwbyKQN2xxOkI6naDM11zDCuOj4+McHBtjS20tG2pqsrpGsckUUplLQTS/nd/pJ0dpWF3BJ1+5zYns0TO8V22tc3aHKqlM7dyklNdbdU5zF2cob/Ii9KWULwohlnue/jiwxf77O8BBiiz0ofAaZVhhm63mPh1mSns+Oj7Oh06dYiqZpCIS4WerV88awe8NqdQjsFRE1nQrcXp3cBv3NPP3y37NxECC4VNTlpPa532b9y/jWOcAAsHy+xY7AQKAY74rRLlrQ3lQSJv+jDVGLzWmK2yLZeLIlYNjY0wlkySAqWSSg2Njs0boZ9pl5avWvjr3n/bdwfd+53WGT02mVFRVKFPNul1NaUtnFGrHapjbFN2RK/LQGN1QXLbU1lIRiTia/pba2mIPKSsKafbznjsSifDJV25N6aerso/9ggsU+nhmq4JgKD6FFPoDQoil8npj9NRgaEBK+QzwDFj19As4HkOB2FBTw89Wr561Nn0orBDNFN2lngeMqcZQcAop9FVj9K8yA43RDcVlQ03NrBT2M4lfIpoqyXH6yVHH8W9MNYZCkq+Qzf+O5bRtFEK8BezCEvbfE0J8DngTeDAf1zIYZivFTkQzGGCOt0s0lDelWpSsUA1jDOWLMO0SDeVOtq0OZxIhUttCGoFvmCmM0DfMObLp0WswlBtFD9k0GKZLUBZwYM0bU5TMYDCavmF28kxfH7/b08P/eeECHzp1iqPj467X/erlGIFvMBihb5iFHB0f55GzZ7kqJUlg0s4C1gkqo21MO4Zyxwh9g4uj4+P89RtvpGjOpcTBsTGSmvCOCuHKAvbWvPFrimMwlCvGpm9wyKVw2kxW2dxSW0tlJMJkMklECJ66/fYUm34+SksbDHMRI/TLjHTCebqF02a6ymaYsg+FLqNtMMxWjNAvI4KEs1oIGubPn1bhtGJU2QxT9sEUJTMYUjFCv4zwE86AayHYd9ttDF+9mpWZZrZX2TQYygkj9PNIqXeP8grnhvnz6bx4kclkkiTWQjB89SqP33JLVuedC1U2DYZywQj9PDFbukd9urkZgA8sXsz2c+ccgR+BnLR0U2UzlUxKQKkrCdky1+YzVzFCP0+UahwgBmkAACAASURBVPco3V6//dw5Z1ECa5xK4H+4ro7O5ctLYsxzgWf6+nj07FkSUlLpowSUspIwHeFdyvMxuCm40BdC3AvsB6LAf5NSfrXQ1ywGpWjX1n+IESFI2MlMU8kkgGu8RuDnD5U8ds3OB5j0UQJKWUmYjvAu1fkYUimo0BdCRIGvA38AvAUcF0K8IKX8Vfp3zj5K0a6t/xCllESEQEhJRSTCQy0tPNTSUlLjnStkSh6D0lQSYPrCu1TnY0il0Jr+euCclPI8gBDiu8DHgTkn9KH07NreH6JfZE6hxpuNiWCu2YIzJY9BaSoJMH3hnct85trnX+oUWujfCPxGe/wW0K4fYBqjF45iCZZM+QD6WNSxk8kkUVtAPrxs2YyMs1CEve8zqSSEFay5fmfejMd5NhZzzhVmXMYXMLMUWuj7ZcO4Cp+UQ2P0o+Pjzg/hoZaWGf1ShxEsz/T18fylS9y/ZEleBK5uIphMJum8eJH7lyxxOZLVruPNeNyJIEpKySNnz7Jq4cJZ/8MvpV1fNoJ1ulr30fFxtvT0MGWbtb4di/HztraM5wjKHTGaf+EotNB/C7hJe/weoK/A1ywppvtjmCme6evjv/z7vwPwb6OjADkLfmUiUML8p6Oj/HxszHEkTyaTPHr2LEkpmSeESzNISllUJ2C+F8B8kK0g9h4f1k4/Ha1bXevNeJyrmh9jKuTn6Jc7kmuyoCE9hRb6x4HbhRArgN8CnwL+9wJfs6Q4ODY2rR/DTPH8pUspj3MVdspE0HnxIj8dHSUJoDmShRZJhJR8rLGRfxkeJmmHNxbLCViIBTBXsjGVBR0f1k6faXHwXlO/VlQI5gFX7WMrfJzXQefRzUneXaJSDryLkPEDTJ+CCn0p5TUhxKPAj7FCNr8lpXy1kNcsNbbU1jJfCEfTT/djKAb3L1niCDj1OB9sqKmhc/lyDo2Pp2ht3pyBP7/pJv78ppuK/iMuxAKYK2FKZ+jC0O/4x2+5JZSdPt3i4LeY6NdCSv7z0qXO8cqMmW6h0Meuj0mNQXjCjNUiZPwAuVHwOH0p5Y+AHxX6OtNhJrSFDTU1HGxrK5pNPxNKqBXCpJHOKbhq4cKU54utxRVqAQwizDz9BHGQRn50fJxjb78NpGZYh/ExpPu8/K7pHdsHFi92mWIyLRTesatz/mz1ap6NxYhdvcqPhodJ2Jq+movJCciNss3InUltoZScen48vGxZwTTaoLkHPV9MLa6QC6CXsPMMEsTeheDo+Dgf/OUvuWa/Lwrsu+22rO9d0Ofit/joY/Pu3oIEvN95vPdi32238Z2BAaaSSeYJwX9eutSlLJmcgNwoW6E/W7SFcrNdFvtzKeQCqPNsLEY8mUSS/Tz9FoLP//rXjsAHSALDV686j3P9Hm2oqWHfbbc5C6J3d+Yt3Bck4P3G/tdvvOH6zJ+/dMllNrq5qso15lLNcZgtlK3Qnw3aQjnaLmfD55IrR8fH+XYs5sQu+2Xs6scGRbOkq4YaAeec+fgeHR0fdzT5Q+PjTlitnmfhLdyXbqHQr+/9zO9fssTlC/K7N6W+ey5lylbol6q2oGtk09F6i7UzyNd1S/Vz8ZLLfA+OjTl1eQTw2TR+Hm80yyNnzyJ9olk+sHgx0f5+ElimnW+8971pnbvTGbPfOdTzfoX71EIxmUzyc9sB7beL8vvM/Xw+hvxQtkIfCq8tTCe+2qvVZaP1FnpnkE2YYK6Cv5R/6LnO16vZPtTSEupYb9E83Qm6/dw5JDDfJ6s5H7unoHN4n9cL9x0cGwudeOf9zEv9OzCbKWuh7yWfWvJ0BIOuTcWTSX75zjtZab2FtIenm0+x7fAzTa7zzWY3k85Z6o1mSQJCSpctP9vrZTvmdOfeUltLVAin+FyxE+8MFkbo2+RbW52OYNhSW8s8W5uTwLdiMR5qaWFLbS3PxmI8az/2nifXHre5zqfYdviZNGk909fH/zs0RMR+PN35ZqPJ6sf6mT30+z9PCN6Mxzk6Ph5Kc87m3mUbibWhpoanbr+dR+wEq2Im3hmuY4S+Tb611UyC0O/HtqGmhs+0tPC3/f1IICElz8ZifCsWCyzjkM7Rl08BmG4+xbTDz6SzW8/YBfjjhgb+/OabZ3S+fgJW3X/1Xfl/+vv5zsBAxnuhSoRclZL5QnAwz+VBjo6PM3z1Kl+//fasv5PlFrU2kxihbxNWW01n19afTycI0wmqh1panBhl1eEqXRkH72I1nR633vn5JZJlEuzFssEWqmCX333wZuxeSSZLRiApp2pCytCKy7OaMjFlKxhBx+fqn8pmMS7HqLWZxAh9G6/tVAmPMMI6XWq538LwZjweuKvwClfApel7t+/5NK0cHR/n93p6mLSv9a1YzKX9zYRg14UL+Atv/ZhMBbumG57oVyRvpjN2s+Ho+DhvxuNEhQAtgzUXjTmo1Wa2/qlsd87l5iOaaYzQ11BfrGwdlmG+pPrCME+IlB+ndxz6+1UZB5WW7t2+ZzKthP3hHxwbcwQdWDsMv7kUqhKlt4CXAK55whP9Ftiggl25hCf67a7UDqoUq3Dq3y2VwQrB32WwdjDfthWKCiFcUURBrTbjyWTaHYEiF2Wk2D6iuY4R+h6m47AM8yX1K051c1VVaEG9ocbKXPznoaGUsaXTwLPZKm+praVCCEfTn++TNFTISpT6PVIRH96MVb/P5/FbbnHNKR/hiUFF8mYqYzcbvN8tlcHqzXTVv8vq+9UVYG/Xwy2llE75az3AIGzUUba7jGL6iMqBnIS+EOJBoBN4H7BeSnlCe+1x4HNAAuiQUv44l2vNFNNxWIYxDanzqhZ6H1i8OFB4BAnq6WhA2Wi+G2pq+HmG4nDZVqLMxrygh/hFsTJVr3l2Q9772DB/fsoc8hGeOBNF8nKtk6/IVhkJowg0zJ9vlb7GKunwwRtu4NDbbzsBBmF2ULmYA4vlIyoHctX0zwB/Avyt/qQQ4k6s2vnvB5YBPxVCvFdKmcjxegUnSGjoPzg/R6nXNDRPCD5jNx9XX+B9t93GI2fPkpCS7efOBSaqBAnq6Qi0bBeKTD+2bOza03HIKY0yKoSvFhrmPuZDYBRa6GR7b9Idr0fveOfg930JowgMX71KBJxM2zsXLuT4u++6fCd//cYbRhOfheQk9KWUrwEIkdIV8ePAd6WUk8AFIcQ5rCbpR3O53kyh/zgUYX6g+pY4ISV/67G9D1+9ivTJqPSSabdRzK2y0ur/rr+fZRUVrFq4MPDYbO3rqjyBxLLlB0Uihb2PxSAr/0mW9yadqUaFayakdH3n/L4vYRSBLbVWc3c9a/ihlpZpO3YNpUOhbPo3Ai9rj9+yn5sVeLWqTzc3hwoL1LfE4LZHA77RFX7kW1DnW2tdtXAhpy9f5uQ77/Dj0dHAH30680I2Zoqw550u+YoJf6avz5WIlNF/ksUcgsxa6ruqKnZCcJ36oAgxvzGmM2Wm8xUYSp+MQl8I8VPArzjIl6SUPwx6m89zvk3PhRAPAw8D3HzzzZmGMyN4tSpwOweDwgL1LTFYN8F7vF99cD90QZ3vRJWw5ws6LqyW6ic4wpgpMo0tn4viM319PGqbijIJ6nQcHR/n0bNnnUJqkyH8J9nMIcispT4L9eNS3zkVrpnuXk/XJm+ia2Y3GYW+lPLD0zhv6IboUspngGcA1q5d67swzDTeL7W+tU0XFqhviXWbflB0RRjynajiDe/T/Q5hrqviweeF2LGAT/hpBjNFriUBsuHo+DiPZCGog86hci8SWphnJERbzGzn4GfW8pZg0D/PMBq5HosfNmvWRNfMbgpl3nkB+EchxB4sR+7twLECXSvvpPtSqx+In6aT7n3T1Yzynaiin8/P76Afp/wTShgCrjh6fccSVnjkElESFr/Fw++5g2NjTmgoZK5r73dO/X7MF4KrUhK1K13mujh7r+d377xO3Gw6TPnVwg+72zHRNbOXXEM2PwE8CSwB/ocQokdK+REp5atCiO8BvwKuAY/MhsgdHe+XOmyNG78fQy6aUb630up8ygYc1LnJG7KnQlH9dizZCI9sIkrU89ncM7/FA/wd8WpnpuzkO97zHt9w26AFSR+zlJK1ixfzO4sW5RzimS7DO+h7pEp3eJ246b53avz652xs9HOfXKN3fgD8IOC1rwBfyeX8pYRXKGVb42a6mlEhnLp6cS5v02mFN2RPLXB+C1C2wiNMRMl0yykELR6ZQmDTRaSkM+cp52oSOPHOO5y+fDltffwwpNvd+d27bI9XeMfvbaZumJuYjNyQFNN5le+ttDqf7qfwE8K6f+LNeBzAdwHKh/DwLm7TNWsFfU6ZQmDT2b+DzqnGvP3cOY6/807eNOXpRvZk+930LnqFqM5qKD2ElCXhOwUsR+6JEycyH1gkyq3cqzf+O53GPR2HYKZr51KlMYxN3zvPb8diKbV+0p1TPa8XqavIU4nifGXrZnsvDLMTIcRJKeXaUMcaoW9Ix1+/8QZ/eeGC03v1r1asyKl0czbMhHDyOmM/GxDNFIR+fwTwX5Yu5ek77ij42MOcPxv/RthzGkqTbIS+Me8Y0jKXzFp+5BJOC/7hvTPRqzhshnhY/4apYV8+GKFfAuTbNJLpOtmcP1dHcqlrj7kuan73J5uM1TD9A7yE9Xdk49/Id2iwoXQxQr/IeMMdBVa8+Ndvv71gteqz1eSmq3GHTQQrJvmIjvLen7ALSZj+AX6EPX/Q3NI548spyzaZTBKJRAIfz1WM0C8y3nBHVWzs0bNnA6tw5nKdmdTkwiaCzRRBuw49NFN/PF3CLiT6/QnqH5DL+dWxfrkS+cwlmY38YMt5psaTPHhyJZFIhGQyyYE156moifCJgyuLPbyCYoR+kfGGOyrC1izP9jozqcmFTQTLN5myZ/0cmPm2Z4fZHemfiVfTz3cJhzDMhA+lFEgmk0yNJxnqiXNgzXkePLmSA2vOM9QTp7Gtas5r/EboFxldwxq7do29b73lFP/Kp2AuhiYXlAiWrhZ7rj6AMNmz3oWnWPZs72eixlIOmnYxiUQiLkH/dPRXADS2VfHAiRVzWuCDEfolga5h/XFjY8F++LloctMVxt5EsHSZr/nQuDNlz+oZv2rhSfdaocMZvZ+JEfYzQyQS4ZaPLWKoJ+4898CJFbz02CCVtRHWdzYXcXSFxQj9EqPYW2y/aJJ8NM0Ik/maD43bz4yl5qTqJfnNJ105BgjXRMcwe0gkErz69Kjrub9v+XfiQwlatzVYfYFTm0PNCYzQNzgERZNEhCCRp05V6XwL2US9BGndukkJ4PTlyylCXF9cJpNJOi9epHP5ch6/5RbfRQn8Y9sNs5NkMslzay8QH0pQ1RglPmTVglSP7/la05wV+GCEvkEjKJpE2oJfhHQypiOdbyGM3yGsCUhVnYzYvgS/GvTKef6T0VEOjY/zs9Wr2VKrNWfXSi2XWzjjXCYSiVBRE3Fs+N+c95rzWt2dFUSj0SKOrvDkWlr5b4CPAVPA68BnpJRj9muPA58DEkCHlPLHOY7VUGDSRZMElZKeDulMWOleOzo+TufFi46wDtK6vSWPvQvWhpoatt54I3/zm98A1sI2qS0ISsdT/5dbOONsx2ua0R+rvz9xcCXXrl3jyM4B13uXfGBBRtNOuvPPBnLV9H8CPC6lvCaE+K/A48BfCCHuBD4FvB+ricpPhRDvnW019cuNUo4m8avZH6R165p81K6TXztvnjOPo+Pj7H3rLVf/TtXp6uCYuzm7Xoa52PfAkJljnQNMjiXZuKeZSCSClJLDO2JU1EQQAibHkmza28KxzkEu/PBthk9N0tRezf0vreDAmvP07h8GYNPeFl/BHnT+2eT8zbWe/r9pD18GHrD//jjwXSnlJHBBCHEOWA8czeV6hsJTqtEkehJbBPhwXR2dy5f7jm9Djbuf7JO//W1KfXxXa0Pg61qnqyBHcCksfuVGNlq1lJLJsSS9+4fp+8VlHjy5kiM7B+jdP0xDWyXRCsHgsTjYi/rwqUkAmturObwj5sTpV9QI1zV0Qa+ff/l9i5katx7PJudvPm36nwX+yf77RqxFQPGW/VwKpdgY3VB6eJ28QQJf4ddPVi85UKntBJ7SSl747XZM5E5xUMJWad2ZtGohBBv3NNP3i8sp8fdDPXFWddTTcvcCertGXO87bT9u3dbgaPAKKSWTownnPdb532WoJ+6Ee7Zua0jZGZQyGYW+EOKngF8roC9JKX9oH/MlrLaI/6De5nO8bw3nUmyMbig9srWrp4sEynQufbeTTfE0Q/7QtXawzC2Hd8RStGqvdu0Xf680+MraKOs7m1KEvsJPcB/fPYgEWjvq6d0/7Iwn3ftUuXrdj6A/LjYZhb6U8sPpXhdCfBr4I+BD8npx/reAm7TD3gP0TXeQhtJnuiaQbN6XjV09G8GejnIsRFYKCCHYtNfSNXVh29hWxcY9zYGafzKZ5I1/fjflfEM9cZZ+cAGHt/cHXvPwjphLgKuF53TXCKs66kO971jnAG/867u0tFezad9SAA5t72ewe4Kb711UEnb/XKN37gX+AvhdKeUV7aUXgH8UQuzBcuTeDhzL5VqG0mW6mbSFquGuLyS5NnwxkTvFQwl+Xbse6olzZOeAr+YPcGTnAEM9cVf8PUBVYxQpJaefHKW1ox6JZtaxH3uduOr6UkrnWEVDWyWfPHmr4zNASjbubSE+mmCwe4LB7gkA13Wa26tLwu6fq03/KaAS+Ik9kZellH8mpXxVCPE94FdYZp9HTOTO3CWbTFrVmlCR7n3TafWXqbBaLqUkDDOL0uR1GtuqXJq/156u4u918w5YiVfnv/8OjW2V3LOnmZN/NcSqrXX0HbrCpZ44nzi4AiEElbWRVLOMZ1zVLVGGeyY5siPGxr0tICWx7glOfPkSm23t/nTXiMuM1NpRz6Z9S4su8CH36J3b0rz2FeAruZzfMDvIJpN2S08PU/aPab4QRIUAn6SvbFv9KfQFKJ5M8mws5oRplnpt/2KiL8alcF+UwFea/Ka9LRza3p+icXvt6e27m5HSqlerC349eue5tRd48ORKDu+IMdwzSWNbFcc6B1zn0iN2YrbWrqhuijIRSxDrnrDMTMBg9wQtdy8AYPO+panjLBGBD1a0msGQE8oE8lcrVqQ10RwcG+OqFip5TUo+29Li+z6/3UNQ+z+dLbW1zFOaGvCtWMzR8J3SC3Zt/w+dOsXR8fE83onZiVqMv9nfzzf7+/m9np6i3xeldSuBf3z3IAANqytdxz234TxS+05JKbn6tiXwW7c18IXk+2ntqGe4Z5LmuxfQsLrSiew53TXi7Aqmxt3nUE7kA2vOM9g9QUObdd3GtipGeqdoWF3JYPcE34i8yumuEad8A8AvOn6bMp/D2/tTxqkzk73KTRkGQ14IWz9+vhCOpl8hRKBWmU2rP+84PtPSwt/29yO53pegWLX9ZwPexXgqz70cpsv6zmZHGE6OJhzteVVHPQLo7RphsHuCQ9v72Wxr0n6LhYq+qayL8uDJla6yC2px0LV8rxMZYLhn0gnpPLJzgIoa4cT5g2U+em7teZo3VfPqU5YyUtUY5dZPLebVp8YcU8+mfUs5vnswq1DUfGOEvmHG2FBTw8G2tlBmhCAHahin6kMtLU7tHb30gl9tfxON478Yl8p9UYJ4076lDHRPMNA9keKAraqLukwnrsVCi75Z+8QSDqw5n3INFQ2kUO/1OpGVkFaCX6eqMcpQzyRDPZP24wh/GnsvkUiESCTCYPcElXXRlFBUda6ZTPASM7mtyMTatWvliRMnij0MwxwgndPWZNimUmo2fT+klHwj8qrz+AvJ9wPp499134DCG9nT2FbltE1UIZfN66sQQricsc3t1fzJSytcQnrT3hae33CeAY/d/66tdVTXz3MtQErD37inmcM7Yi67/6qOeme3Mh2EECellGvDHGts+oY5yYaaGh6/5ZbAMg1Br5UrG2pqePqOO3j6jjtK8r74RfJ4H6vjvChTjULVzP984k7Hpn9k5wDJZNIJuTz95Ci9XSOOLR9goHuCIztiVNQIR+ADLFnn9jMAnHlylPjINZfmrjT8799zAbzjlNLxWxQaY94xGAwlh7cy5uHt/fR2jTjCVtfe9cQoZStXqMQoHZXgpdomKht9JBJxhVyCZcsHWLW1zvIX1EUd7V0IQTKZpP+w+/yKvkNXnOMcP4GU9HaNpOwMnPyBGTDvGKFvMBhKBmlrvF7hHeueoLm92hHw6jUVV6/bygdevkJTezXYyVhghWwu/9hirr4t6d0/zJGdA47gr6iJMDWedATupr0tKSGXm/dbtZm8pR8ikQgVN0RoaKt0Fgh1vcqaqKuOjxCCjXtbAstAzJSh3Qh9g8FQEhzrHCA+mnAic1Q1zIHuKwx2x51SCLpwTom48WjSTeurSExJhnsmEQgeOLECsJK4lIY/NS5djtXvrXk9ZWyHt/ezad9S3yJwS9qqib18xXX8ss0LnIVCIaVMcQArVnXUpzikC4UR+gaDoeh469y0dtS7NGLl6AQCwxuFEGzat9T1vsFjVoKWst2/9NhgSsTMxj3WebwF1VZtrQMhnOxatdg44Zd7Wxyzkz5GZXoSkUhKWGbv/mHHjKN2IfYNYN2upjze0WCM0DcYDEXHLzbeD796O17bvx8PnFjBS48NBpZw8IZnrtpa59LUB7snqKqfZwlmIVznaW6vpqm92lmUvKYnNcbK2sj1uj9Pjjr5BjHbcawvEoXEhGwaDIaSwRuW6YeeeKXb/pXWveDGKNVL5qXY2Ffct5gTXx5ynlMhn0BKWKceQplMWmUdlH3eO8bPJ+50BPWh7f1U1UVZt8u/uXqQzyLX5KxsQjaNpm8wGEoCKSWHPJq6ZceXnO66bgpRwlJvbrJpb4tjV09OWlE3DW2VLNu8gHMH3ma4Z5LLb11znfu5Dedd8fitHfX0v3yF0VcnOd01ggA27m3hyI4Yse4Jbrl3Eet2NaWEin5vzet88uStTux9U3t1oKlGCOGK/tHnNFO1eXLS9IUQf4XVGjEJDAJ/KqXsE9bo9wMfBa7Yz7+S6XxG0zcYyhPd5l3dHOW2/9XKFTjdNUJlQ4TJ4aRzbOu2BipqBPHRBBFPAlXT+ire878s5M1/uewquKYSsvyKtzWtr6Ll7gUkpeSMZmdXz6vz37W1zrmer11eja8IFTVnUtP/GynlX9oX7QCeAP4M+EOsGvq3A+3A0/b/BoPBkIIQIqUs8sY9zZz9x3HiQwmqmyPc+skaS+juH6ZhdaXV1Hx9Vcq5rr0teeDECleNndr3zafpd2qdkgvK/j5ol3ZQDl8VIXS6a4TBY3HX85vtujl6YhbgEvylVEI5iFxLK7+tPVzI9VDTjwPP2p20XhZC1Aohlkopg9vWGAyGsqZ9dzPrdi1xImuUJt7YVknL5gWcsZ2fTe3VvHNxCrgenaNITlkhm30vukMor70DG/5v69zKdq5MKrp93pucpT/vNc1IKcEj3EvHQxpMzjZ9IcRXgIeAceD37KdvBH6jHaYaoxuhbzAYAonYESy6U/XBk7cihJUxm76GTiVDPZN28TO3aWeoJ87z6y46VTW7d8WYGkumCOnn7n6dZrsuvo63kqfyP3gXB/U4lzo6hSZj7R0hxE+FEGd8/n0cQEr5JSnlTVhN0R9Vb/M5le8iKIR4WAhxQghx4tKlS9Odh8FgmAP41dhRCU1+NXR0RIWgtaOeujsrfI9TAn/jnmamxpL0do248gLA2jnoJZxbNXPPIU9NfFXeobWj3qnbrz9fquTcGF3jH4H/Aewii8boUspngGfAcuSGvJbBYJhj+HXLckIp7ezcdFw6FqdpfRVL2qrpf9Ff8Dpx+QGlmvtfvsKV315j5f03OKYesGLp9YxZIQQ337vIKg1ha/WqEXrlDGXWTpdco3dul1Ketf/eitUg/QEhxH/E0vo/iuXA7ZJSrs90PhO9YzCUN35lDg5v7ydmNxtv3dbAPV9r4r/d8D+5duW67Gpsq6R5UzVDxycZsI/buKeZA2vOu0w9ekJWUKlmRUqv3IC4e28t/mII/JmM3vmqEOIOrJDNN7AidwB+hCXwz2GFbH4mx+sYDIYywBvDrjTo47sHabl7gVNCQRf4AEM9kyz74EJu+sgimrXjhnriNLZVsfy+RSk1drx1cA7viPnGy6cT4tkcWyrkGr1zf8DzEngkl3MbDIbyxE+Q6otBRY1whXaCVVunojZiN0aXrraJqpqm0thVsTVfMxIzmyhVDExGrsFgmBUok8zUuHT1tlUCe9nvLnTtEnx3DVrtfW/tHXDXy5mrGKFvMBhmDd7G55kEdtDjdAvCXMcIfYPBMKvIl8Cejfb4fGB65BoMhllHuQrsfGCEvsFgMJQRRugbDAZDGWGEvsFgMJQRRugbDAZDGWGEvsFgMJQRJdUjVwhxCaucQz5pBIYyHjX3KNd5Q/nO3cy7/FBzv0VKuSTMG0pK6BcCIcSJsIWI5hLlOm8o37mbeZcf05m7Me8YDAZDGWGEvsFgMJQR5SD0nyn2AIpEuc4bynfuZt7lR9Zzn/M2fYPBYDBcpxw0fYPBYDDYzEmhL4T4GyHE/xRC9AohfiCEqNVee1wIcU4I8WshxEeKOc5CIIR4UAjxqhAiKYRY63ltrs/9Xntu54QQXyz2eAqJEOJbQohBIcQZ7bl6IcRPhBBn7f/rijnGQiCEuEkI8XMhxGv293yb/fycnrsQokoIcUwIccqe9277+RVCiG573v8khKjIdK45KfSBnwB3SSlbgX8HHgcQQtwJfAp4P3Av8A0hRLRooywMZ4A/AV7Un5zrc7fn8nXgD4E7gf/NnvNc5e+xPkedLwI/k1LeDvzMfjzXuAY8JqV8H3A38Ij9Oc/1uU8Cvy+lXA20AfcKIe4G/iuw1573KPC5TCeak0JfSvlvUspr9sOXgffYf38c+K6UclJKeQGrh2/GgJfETAAAAmdJREFUhu2zCSnla1LKX/u8NNfnvh44J6U8L6WcAr6LNec5iZTyRWDE8/THge/Yf38H+OMZHdQMIKXsl1K+Yv/9DvAacCNzfO7S4l374Xz7nwR+H3jOfj7UvOek0PfwWeD/s/++EfiN9tpb9nPlwFyf+1yfXxiapZT9YAlHoKnI4ykoQojlwAeAbspg7kKIqBCiBxjEsma8DoxpCm6o7/ys7ZwlhPgp0OLz0peklD+0j/kS1nbwH9TbfI6fdeFLYebu9zaf52bd3NMw1+dn0BBCLAKeB7ZLKd8uhyYqUsoE0Gb7KH8AvM/vsEznmbVCX0r54XSvCyE+DfwR8CF5PS71LeAm7bD3AH2FGWHhyDT3AObE3NMw1+cXhgEhxFIpZb8QYimWRjjnEELMxxL4/yCl/L79dFnMHUBKOSaEOIjl06gVQsyztf1Q3/k5ad4RQtwL/AVwn5TyivbSC8CnhBCVQogVwO3AsWKMsQjM9bkfB263oxkqsJzWLxR5TDPNC8Cn7b8/DQTt+mYtwlLp/w54TUq5R3tpTs9dCLFERSEKIaqBD2P5M34OPGAfFm7eUso59w/LSfkboMf+903ttS9h2cJ+DfxhscdagLl/AkvrnQQGgB+X0dw/ihWt9TqWqavoYyrgXP870A9ctT/vzwENWJErZ+3/64s9zgLMexOWCaNX+31/dK7PHWgFfmnP+wzwhP38Sizl7RxwAKjMdC6TkWswGAxlxJw07xgMBoPBHyP0DQaDoYwwQt9gMBjKCCP0DQaDoYwwQt9gMBjKCCP0DQaDoYwwQt9gMBjKCCP0DQaDoYz4/wEIc00vc5zilwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cluster_and_visualize(male + female, extract_vectors(male + female, hd_limit, hd_w2i_limit), \n",
    "                          'Hard-GloVe', random_state, tsne_random_state, y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 248,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 0.7775\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cluster_and_visualize(male + female, extract_vectors(male + female, hd_a_limit, hd_a_w2i_limit), \n",
    "                          ' Strong-Hard-GloVe', random_state, tsne_random_state, y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 244,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 0.6225\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXkAAADCCAYAAACsebI/AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjMsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+AADFEAAAgAElEQVR4nO2de5BcV33nv79uzUuyPE9N94xBGgnZqdiWNGBJgzQS1i4kkFRib7BN2E2tHaBwYoLkkdkq8LJBo7Ap8igkWSYBxAYvrpAAtsgiKi+CC9mSGI0kk9HImBhkPYw90zOaV/uh6ZlR99k/7j1nzj19X/243bdb51Ol0vTr3nNn+v7O7/zO9/f7EWMMGo1Go6lOIuUegEaj0WiCQxt5jUajqWK0kddoNJoqRht5jUajqWK0kddoNJoqRht5jUajqWKWlHsAMm1tbayrq6vcw9BoNJqK4vnnn59gjK2wey1URr6rqwtnzpwp9zA0Go2moiCiy06v6XCNRqPRVDHayGs0Gk0Vo428RqPRVDHayGs0Gk0Vo428RqPJi4FkEl+4fBkDyWS5h6JxIVTqGo1GUxkMJJPYMTSEBcZQQ4Sj3d3Y0thY7mFpbNCevEajyZknEwnMMwYGYJ4xPJlIlHtIGge0kddoKggdItHkig7XaDQVwkAyifeePYv5TAa1kQie2bChbCGS++NxPGF687VEuD8eL8s4NN5oT16jqRCOzsxgPpNBGsB8JoOjMzNlG8uWxkYcvPlm/FpzMw7efLOOx4cY7clrNBXCjqYm1EYiwpPf0dRUtrEMJJPoO38e85kMjiWTWLdsmTb0IUUbeY0mRAwkkzg6M4MdTU1ZRnNLYyOe2bDB8fVSYreq0EY+nGgjr9GEBK+YuzwBAMBDL70EwIiPqwbWbbIoBmFaVWjc0UZeowkJbt6xPAEsIUKaMVwzP/dEIoEfSTr1UmzQhmlVoXFHb7xqNCGBe8dRIMs7tkwAkoGH+VjehM1lg7YQSeaWxkY8umqVNvAhR3vyGk0J8QqjPBCLAcgOwcjhEdWTryVCa00NvnD5MnY0NfkOpYRJkqkJDm3kNZoS4WZU1de47lyeFOTwCACRZfrO5cuF0oUf108oRW+eXh9oI6/RlAg3o+oUYlEnhUdXrRLH45/9wuXLls8+mUhgZX29Z6xcb55eHxRs5Ino7QCeBBAHkAFwiDH2GBG1APg2gC4AlwB8iDE2Xej5NJpKxc2o2r3m19OWPxslwhOJBK4x5hmC0Zun1wfF8OSvAfgUY+wnRLQcwPNE9G8Afh/AM4yxPyOizwD4DIBPF+F8Gk3gBCFBdDOqTq/58bTlz76SSuFro6O+QzBbGhu1ca9yiDFW3AMSfQ/Al8x/Oxhjo0TUAeAoY+xX3D67ceNGpht5a8pNmDYk3SYbu9fKOfagtfkaZ4joecbYRrvXihqTJ6IuAO8EMAggxhgbBQDT0LcX81waTVCEaUNS9bS5IW2tqcnabOXvLUYIxmkCcZtwwjIxaqwUzcgT0Q0ADgPoY4y9TkR+P/cggAcBYOXKlcUajkaTN2HdkJQNacSUUWaQPREVGoKxM9hA9ibwlsZGYfhfSaVCMzFqrBTFyBNRDQwD/03G2HfNp8eIqEMK14zbfZYxdgjAIcAI1xRjPBpNIYR1Q1JeYTDGECECmRusxZyInJQ+buqfKBGWEAEBjEdTGMVQ1xCAvwHwM8bYPumlIwAeAPBn5v/fK/RcGk2p4N4wzwj1Y+wPjYzg8JUruGfFCjzY2Vn0MakrjANr12JyYaHoE5HTSsZN/QPG8PGODl/STU1pKYYn3wvgvwM4R0RD5nP/E4Zx/w4RfQzAKwDuK8K5NJqSkUuc+dDICP7g5z8HAPxg2lAKuxn6fDYpC1lhOJ3P7nmn87ipf6JmeFYb+PBRdHVNIWh1jSYsDCST6L90CT+cnkYGQBTA51evtiQjybz/7Flh3AHg15ub8a9mLNvu2KXcpHRquu13HF4brk8mEr61+ZpgcFPX6AJlGo0CN37cwEeQXTBM5Z4VK1wfy5S6w5NT020/4+C/iz++eBHvPXvWtpDZhVQKC4yFomOVJhtd1kCjUeDGjxv49zU3o7+ry9U75aEZPzF5NeYtFxcrpab9lVTKCLO4bJb6KX88J/2u9KZr+NBGXqNRUI2wl4HnPNjZ6WvDVY55O+ndi4nadPudy5dbatN/vKPDtvEIALTW1ICIELGZCPKZDDWlRxt5jUYh3w3OXBKIuHpHLi6WMouLBdHg40fd3WIcqioGgAixqJuyfefPI8MYokQ4sHatY/njXCZDTWnRRl6jscEpochNpeI3gUhmR1OTqA/PAHw9kXD0qt3On+v1yLXpv55IIG2zaSp76sQYJhcWso4ZxnwCjRVt5DUan7ipUfwmENk15/5IPI6vjo6CAUibXZ6cJphiqHL8FjTzk/mrC5yFH62u0Wh84qZGsWvdx710ArCEyHFD8v54HPUObf84XNI5l4cqx67FH2/dd3887thyEDA6VX28oyNrQjk0MoL3nz2LQyMjvsagKR/ak9eAMQa51pD6WGPg5NnyEIqagTqQTIJnobhlo3iFPexULEuI8EoqhYFk0tWT9vL+1XMDRhMSdUOYd6oCck/80pQXbeSvc071j2FuJoNt++MgIjDGcHx3AnVNEWzuj5V7eIGSa3zbzhh7hXB4rN0tDMOPzT8jP+bPySqWztpaJObn8bXRUXxjbMw1bOOnoqZcwsFPAbTDV65YPn/4yhVt5EOMNvLXMYwxzM1kMPzYJABg2/44ju9OYPixSax/uLWqPfp849tqDNrJiHIdut+iXW7j4SsI7sm/Oj8vPudV8TGXipp+C6Dds2KFJbvXLfFLU360kb/O2bbfWIYPPzYpjP36h1uFZ1+tFKtmvJ0RlQ121EOH7qdUL19ByGUWAIDgnXyUiwLGbwG0XBK/NOVHG/nrBNUrH9yTwHySYdv+OLbtjwsDD6DqDTyQe814N727akRl7TsYw8r6ek+1jFqqV82C3dLYiP6uLhxLJoX08SPxuKvcUh5jIeEouxCS38QvTfnRRv46QI27ZzIZXDryJiaGUoAZM5Y51jeK7Qc6qtrQ5+Lh+tm8dEsScppAnEr1Bt31yQ35WnS3p+pAG/kyE7SyxS7ufuKRMUwMpdDWXYfhg1Pivet3tYABOHdwCkRU9R69Xw8319COX2OsTgbcK5dXAkF0ffI7SeR63brHazjRRj4g/BjvUihbuLEGsuPuvfti+HL0RfHebQc6xGfqmiJVbeBzIZ92gH6MsdNk4Od8+RjUXD3zXK5be/3hRRv5APBjvEupbOGGXo679+6L4fjuhOV9x3cnRIxeG/hFihEm8aphk8v58jWoQa1I8jm2pnRoI19k/BpvNw+72EaWMYbjfaOW575zx8uYHJozzmmGaeQxa6wUEibJxyi7nS9fgxrUiiTfY2tKgzbyRcbLeHO4sQ9a2cJXEcMHpywxd27g1+1sRq95zkoJ01Ra7LfYXm6+BjXIjVtdrCy8aCMfAE7GGzCUK5NnU2jrbnANmRQzVFPXFBGTzKn+ccvr2w504MQjY6hrilREmMaplV2YKbaXW4hBDbKgmC5WFk60kQ8Au/DI8b5RJE5exfipFFo31GH4sUm8dvRNTJ6dQ0MsitmxNNq66y0hk2IZ3M39MTDGwBjDpSNvWF576o4LmBhKYf3DrUU5V9DwVnaA0cruL155BZtvvLHosfJi4ifGnk9Tb21QNX7QRj4P5E1R3gidP06n0/jxp8YXwyOM4dzj0xapYuedy9B55zKcM5+bHUtj3a4WIW8MKmSyKJ2sNzTygHjcuy8Wei/eju9PTuL7k5N5KTqCUoTYGW0no6xVKZqgKYqRJ6KvA/gtAOOMsdvN51oAfBtAF4BLAD7EGJt2OkalICtnTu8dx9x0GgxAfXMUjAGXjryBmhtJSBTVcMz6XS0Wg8/hyUdBhUzksI0qney66wZEIuGtOi0bTbmVXQRGdUe7Ilp+CEIRkms4SatSNEFTrDv7/wL4gPLcZwA8wxi7GcAz5uOKRlbOHOsbxdx0GsMHp3Du4BRmp67h0pHXMTGUwsLrDFu/2I7juxPCWxfHcDj28d0JZDKZLAPPVwpOj3Nhc38MvftiOPHImOX5+SQr6LhBwj3dP754Ee89exYA8KPubvzp6tX461tuQZ1HHXY37GrAu41DrcluBw8nMRjhpCcTCdf35zKGfPE7dk11UhRPnjH2HBF1KU/fDWCH+fM3ABwF8OlinK9cqMoZmRceNxYpPBTylSU/E6+t39WCbQc6cKxvNMvor9vVIo438uxb6LprOXr2Glr6YidLMcZw4pExIeeU5Z1AOGvW2Hm6j65aJbzddcuW5R1T9xsrD7LZdqGqFHmMdsXEdDhIE2RMPsYYGwUAxtgoEbUHeK6SYaeckbnv+TWWUAg38IBRNRAAlt4URUPbEkyencO5g1NYt6sFrRvqMDGUQuedy4RX7aW35+PhqHsF6msALEobflz+vEoYSg17KVMK3YCUa6nLBcFk40hEyDjUVleRw0m1RJZmG15jyBW7ZiJ1PtoSaiN/fVH2jVciehDAgwCwcuXKMo/GG+5NO/HUHRes7zf/JyLUNUeF0efHOWeGe4DsRCg3vf3pveOOXj6ArNee3nIBBOCegTXC4PP39+6L4cyfXLHIN8PSPKQU+ms7b1c2jhHGEHWorW433h91d+c03nwVPnIzEcB+b0InKWmCNPJjRNRhevEdAMbt3sQYOwTgEABs3LgxnIFhE5FY9Ngk1u1qAQFiE/X2nc1IHLsq1Cr3Pb9GhEa498+ljDzxaPuBDkv4Rg2XuOntnbx8dVzb9sdxrG8U44OzAICnt7yMWM9S8Z71u1pwrG8UV06lMGa+J2zNQ4KWC9p5u35rqxcqwSwknDJz7ZpljyeC7Li+TlLSBGnkjwB4AMCfmf9/L8BzlQQ1sej03nGRRVrfHEXXXTcCeEOoVeRQCDeScjjFKxHK7T2eWbVEltd47P/cwSmMDxryST72Fx6fxu07mxHrabjumocA9t6uahyBxdZ8HDsDDSAno310ZkaEW+ZyCKccGhnBX/zyl+Lx77W347Zly2wNudbUX98US0L59zA2WduI6FUAe2AY9+8Q0ccAvALgvmKcq9zI3jj/GVg03pv2rBByRCdJpLwicNoABeD5HreSCOpr2819AXnlwL197v3XKnH5SjPw+XrVTt6uHK+3M9x2KwCYP/uNgbfW1FjCLa01Nb7GrPZZvbKwgEdXrfJ9zZrrh2Kpa/6rw0vvLcbxw4YaUpFR9eZ2RlJdEciqHdnrd3sPAEcv3+61Y32jcDLXPHTT1l1ve7xKMPSFqkjyKQjmFO/mzy0hwiupFAaSScdjTy4sIAKIjdPJhQVf49V9VjV+KfvG6/WKvCIA7L1+p/cAzl4+Y2wx3m6+Jks315lZuFzyCXADXyfKG5RbWpmPRy4b4lQmgycTiaIkNnF5op0xd1oBPLNhA55MJPD1RAJfGx3FN8bGHCedHU1NqMtjY1T3WTUIuulONaCNfEBkMhmLV68+BrK9fCev3+6xl5cvv7b9QIfYVAUWY/CyoY/UkqH8cVlZlAIvj9xpAtjR1IQoEdJmItITiURW/9NcJg91HE4br0714I/OzCDNmGfYppCN0eu9zyrPI+ndF0MkEhFh0NpGQs9eXS6bo418APzDjguYT2Zw3/NrEIlEkMlk8NQdF1DbGMHvHF1TlHN4rQTU1+4dMM57eu+4iMHLEJEoOWx3vFLhput2mwC2NDbio/E4vjo6CgbgGmO+P+tnHJM5xrxba2pARIg4yC7VCUdvjOaGnH0+8uxbFjWbEXYcE0mF1zvhLVhSoWQyGcwnM5gYSuGpOy4IAz8xlMJ8MoNMJuN9EJ+4rQTsXiMibNrTbgnnfCJzG9bvasHY4CxOPDJmKW9QjmWvW5q/3QQgp+zfH4+jPofP5jsOLwaSSfSdP4+Mqa8/sHatbRYqL9Wgyw3kDhGhd19MZJh/OfqiMPD8XgtrqY5Soz35IhOJRHDf82uEYefZr1w7X+5CYCIpSw71HOgAQtIwxC180VpTg4iZqLWECKdefx39ly4hbXrLz2zYIOLhKrkmBRUSRpGTlIixrM1UnYVaHPi9JmeYy/tKKn7j9dUW59dGPgDsvnxhMPAcP5u+5cQufMG947S5sZxmDN+bnBTJQHIS0zfGxjCfyVg2PLnR9ioY5jUOP3hNKGHJQh1IJsXvQ92/cHp/mJKqeC0mld59MdeMcLcM7mLXiwoD2sgHAA/RyDx1x4VQGXo/m75hwuIdm8+JkhFYDKk8mUgglckYVSBtvGS7CaCYcEPotFELhCMLlZdE5g1Ynkgk8CObssilKNKWD3KuidwfATD6F3duX4pzprDAbwY3Y0xUlhWf6xsVoc1K9ei1kS8ycgyeh2j447AZ+kpC9n6XEIHB8OaXEOEj8bgoBPb1REIY/yVEnnH5YhqqXDZ3/awSgvScj87MYEGKWfOyyPL55OuJmMqlfGv3FxsiQm0jCQPP+yTwe+2m9yzDul0tvjO4GWM4vXccDGbPB+lzsZ6GUK10c0Ub+SITiURQ2xixxOC5oa9tjGgDnwd23jEAS4ld/jjNM5ABfEQKQQwkk3gllUKUCPBRaCwfijmJBF0ieEdTE2qIhCe/hAhfTyQs+xvy9TDGEPFZpK1UGDLJMXTeuUwYYa6yqWuKYNOedtfaUJxT/WOi+c+5g1O4fWez5fUVmxeTBNUM90rw7rWRD4DfObrGoovnhl4b+NzxMnaqll2OdXPvXj7GEiJ8vKPDVww6V4oZa3dSAxXLs9/S2Iij3d2WPYqvjY7mVaStnPTste4vyTWjvGpDAZIU0yzWp+aPAEDi2FXh6aem0yAAdc1RbNrTXhHxem3kA0I16NrA54ebd2ynZbeLdcvvA2NYWV8fiJEqZqxdNbCtNTVF9+zlkNFAMin2K5yKtIXFsKvYedJedZ/sssjtekQYmeBzWXF+Xr31nI94fbnVOtrIa0KNm3fsVD1SNUalVLPko8hxavwtG9hcQ0G5xvO9irRVEkSEiaFZS4P63n0xjDz7FiaGZm1FB06NgOLblwIAJobmMDk0J54fdugBoRIGtY428ppQ4+ZNyq+11tSIkEYY1SxOeGXx8g3QXPYT8o3nV6JBt4MxhrZuo2z2iUfGsG1/HCd2J8QGrV1c/XjfqOUYvAw3j9FPSAZexs3Ay1m5/L3l6NOgjXwJKfeyrVJxMz78eS+jVgoDls/fV/XQ3RQufvcTrvdkKzUEI9Q1Zm0mYLErGo+rDx+cQqynAe09DZaGOrfvbMaVUymnU1kqv9r97d36PpTq3tdG3idq/1Qgtx12voO/7UDH4rKtbxR1zdFQb9qEFTkcEQajlu+yXJWGuilc3PYT5N9HWJKtyoldCIYLRmVvGgBqG8mSJcsYA4hQ2xjB/EwaY4OzFs8eWPT0hx+bROLkVcR6GrDd4d526/tQCrSR94F8A5/eOy7kVvU+d9gZY3jlX95cbK93oEMkWcR6Gox6Mnn80a/XlYFdhch8jVoxtOiFLMvlUNIrqZRQuPByyffH457XZheeCWt4qlTYdVVT+ynzzNj5JEPvvsV798QjY6htjGBzf7vR/c2cAE71Lxb34+oaALhw+HWcG5wFIfve3vi5FVlZucf7RtG7Py7EGKrTWGy0kfdAvoHlWu2AUZvdb0Zce08DxgZnMXxwSnyeP+92bicjHoYNnXLhV1XjRbG06IUuy+XY+xOmJ89gJHbdH497XpvdSubRVauuS+MOOHReM+9TTm1jBMd3JxbvZ/N3Pj44i7HBWcR6GjA3kxYd1QBgPpkR95fcgQ0wJhD13l6xuR4nzFAQH8fhLRcwfHAKrz33Fj70/DtARDjeN4rE4CxWfeCGQO5dbeQ9cJNYnfO5w85rusufAYxJgi/xVNyM+KY97a6NvNXQUjE9hDDUL/GrqvGimGEeu/BArsvyLY2N+IhULjltlkv2Mtg6PGNF7bwGAGo9yktH3sDEUArrdrUYGa7KfQkY9yr/uzqtzGRdvt29La8EAKB9cz3GBmcxOTRnnWRgZNYGsRrXRt4HbhIrwP/NnFXD3eF9Xst//hwYs3iO7ZvrLccstncfdBamX4qllimmcfTTmN0P98fjWXp1N/zUyrke4aEUwPg7nDs4JQyvXPNGNswc7pARUdbKrLYxIhQ73Pk6sTuBsZNXLcfgf3G1GOD2x4wmL+cen/bt8BXKdWPkC4lf293AMl43M9+IGVa+UKIQkvLH9bP8P9U/luWdpOeZcUwP7yNfwrDBySmWWuaBmDH5FZIB66cxu9/f/ZbGRhxYu1a09fPSwodh0g0b6iqY17ipa4pa7q3aRrIUNuPw+1l17Hr3xURjEgBZrTVl3O7t7Y91iqQqTlAGHrhOjPzgnjHMJxf/6JlMRtS38PJw5RuYb7rIMXkCfN3MCXPTlc/Y/MuRkNryybgt/zOZDFLT6awv1+TQHFq76wKTa1VTWEA1kLwEgtN73VYNanigkPaJvKTyfCaDY8kk1i1bJuL1bpm8bpNuGEJsdgQxLrtV8HySYWIohc47lwlnp3dfLMtxk+9nu4YjJx4ZExu08j3WEItidiyN23c2g2Dc61dOpWzvbTtNPgAc6xvVnny+DO5J4NKRN8WMLVeq8+Phqjfw6b3jQj7F1TVeDTeICKs+cIOQWQGLS8L65qjtZ+y+DMf7RlHbFMHcjHN3qc7tSy2ZecWUa4U5qShXcjGQfrzlYtXod6pZYzcGP5NuWL39oMblZxXM69CfOzhlCh8Y0ikmQjrrdjXj/LeTmB1L267MevfFLM7XDSuX4B0fuhEEIwzTvrket+9sRkPLEnE+WVqZ5SRKqp8gDH3gRp6IPgDgMQBRAP+HMfZnQZ+TwxgTs3hbd73ljy6nPHsh38D8Z2BR8uTnZuafk5sZcEOvxs0ZYzi85YLQ58qyrIZ4FLOJNJbetAStG+oweXbRoLesr8WF775hOW8+cWE3qiUr0u+qRDa6c5kM+i9dQn9Xl6NH7/Y433E5TUh+Jt1cQmyl9PiDDP15bYLLjlvvvhiO943i3OPTIqQzN30Ns2Npi42QQzyqJPLK6TlcOb14H8bfvVRIJLNCR01RLI1HsWxljUW5kxicRX1ztPI8eSKKAvgrAL8G4FUAp4noCGPsRfdPFu38jsqYXKtCqnE1p9e88Kun5otF9f/lK2vwjvtuxAuPT+Pqa9csx54angeAxYkhz7jw9YDfVQk3unNmw5IfTk/jWDLp6nkWYiydxuU0IXlNujuamhAlEv1mnbz9JxMJPJFI4JqUiBWkoQ8y9OdnE1x2ukAkFDZ8xR/racAHf7xa2Age4uEx+cX69S9bSh7we4+PQ73f52bSuJpI4x333Sg+s+1Ah1HKOaAihkF78psBnGeMXQAAIvoWgLsBlMTIm+fMWl4BsOyQl3IsdkvJdWa6tTyWewfWiLj9OWl5t938QqjlUDlt3XXoNY/F44c8lFSNyVK5GlT1/V6f4Ua3/9Il/HB62rNphhyGiBLho2ZDk1wMpjourwnJc89A+d9uvLybFjyur1gEFfrz2gTv3RezGFO+t8Vlk5z2noaseyUSiWRJM+Pbl1qMPMNiW8K6pojt/b5ic70lbFTbGLFo8ItN0Eb+JgC/lB6/CqBHfgMRPQjgQQBYuXJl0Qdg14qPh26A0ni4aqxWnXTks8ubwtsPdFg2V/nyzq6vJWB4EbVNEcsykX+pqzFZKte4biGFu/q7unAsmcwpvJNmDF8dHfVsNSi32HNrGZjPnsHRmRlcMxN9rpm6e/X1ecnAy60UgyaI0J/bJvjE0GyWc0cwjK4qYhAtJhXHSJZmHusbzXK2zh2cwsizb2Hy7JxF7mwpr2Cq4EaeuypCyX73CPMhaCNvN1rLtjVj7BCAQwCwcePG7C3tAuAzpdyKjy+32rrrUduYm/IhH9SYnN2kwzdievfHxaYwz6aVOd43KupnrNvVAjBmkWIxxsQyNAzV74Im13hz/6VLIuySq7eaa3iHe8ZOvWblcb337FkxrgiAuhwmIK/fQWtNDYgIEal6pVOdm3xXHmHDbhPcTv7IC5O1dddZPs9DN4nBWcR7Gqz1pqSExHFJMQcsJkNNnp0Tq3Mgu3kJ9/x5aIgb+KAczqCN/KsA3i49fhuAkYDPKVA3WOTstNpGQs/e3DzaXLX2qrGVlT2t3XWWJgRySnRdawQAw/DBaaHkGTdLIsR6GoRUa/hx6+vnHp8GmdcYhup3QeM3rmtnSPPxVnMJ76gxbi/Pn+ul1AnIKxTj9jvgckwejz+wdi2AbKVOtSimZOxCLU6VKRNKIhMDsG5nM8ZPpRzzTgBgpamY22aTzS6LKuTQkZOuPsj7MmgjfxrAzUS0GsBrAD4M4L8FfE4LxZK2edWKcZoA7L5YfFVBREZWnfJHn5vMYGwwZal8t25XC2I9DaIwkpouDSyqdGTVTzmr3wWNX+/66MyMxcC/r7nZUSFTrHFtaWzE/fG4Zxy9tabGsrErT0ADySR2DA1hgTHUEOFod3dOtfLlCYQYw+TCwnVd58ZOdcMAjJ9KZeWvrNvVgg/+eLUw0naOkqy0c9roVWP4TndfsVVwMoEaecbYNSL6JIB/hSGh/Dpj7KdBntOOQqVtXuGPwT0JzCeZ4wSgfrFkZU/v/niWkW/rrsf44KxYDvIvibwDv7k/hkwmY7kW+UtSrDR7v78fwFp6WX4cJH6869aaGoun7JVJGvTY7KpoTi4sZMXkH3rpJdFoe54xPJlI2B7P6TxOXn61JLTlCt/vkhkfnLXUmSEY4RcuZ1S/wXb3j7AFqqqNMeHli/eZ4aGJoTkRiw96jzBwnTxj7J8A/FPQ5wkStwQLx1ifOQHYfbH45g9jzHZTWE21FiWOlZWEmrWrGvhipNl7cap/DJf/5U0RuwSMDanxwVmsDKiqXq5MLiwgAghPeXJhoazjsaui+eiqVUU/j+zl885ZO5qaAg/PhCnDlq+o5aTIRfnjBVFxUm3ovWlPu20pkuO7E0I7z//VNkbQ3lMvNht5XanE4CxO7x0X9wD36msbCZ133iBsR8HXh4oAACAASURBVG1jBJ13Lss5O9ovVZ/xWizcEizsUp3VCSDL2JqKB7tN4fq2KFITaXGeY32jlnIKXhupxUyzd4MxhtR02rLqkBsrBFVVL1d2NDWhLkTeq9+9hPvjcTyRSGCeMdQSuZZeUFE3V9U4fBCTCj8v3/+IEuFLN9+MBzs7AzmXF7LCjCdF1rdFRUITv/dqm6LZMXub4mH8njv/7Rmsue9GvMcsNjY3k8brF+cxPpgS9xrfJ4u/e2lWIqUsZ5adtqDuE7Kr0VAuNm7cyM6cOVPuYdgie8ccPivPJxm2frEdX1nyM/Haxs+1oWdv3DWWD8Aic+TKG3m33aluDj+/fFzAGiLJZDIWTXAQXyTGmO1mEl+6ltvAc8LkXeYynnzGrYaDHojFRDOSKIDPr14dmJH/wuXL+F8XL4rw2BIiPGezlxA06mpWFj1wZEGG/Lm/jixGlOXqkMb9uZj8pN6Taga6k9AhiGY/RPQ8Y2yj7WvayHujfmFqGyOiHjUPr6jeN/fOuUbd6Y+qvqYWU1NlW/IX8KH0reL4T2+5AAJwz8Aa232BoH8/8rgA4BOZ20Jj4POhGEY434ml0AnpC5cv448vXhRG/eMdHZbyxUFmsw4kk7jT3CwGjPDY/w5wUnHDzjGTUb+jto6c4qxkMhnRCESGSya/HF3M87S7B4Jq9uNm5IPJo60yVCnmfDIjDPyq315ma+AnhlI48ciY7SztViKhZ28sq87Gtv1x0WZQ5qk7LiCTyYgY+NjgrKGll76sczMZ24p6xYJ78ip8HJUI94T/+OJFvPfsWQwkkzm/z+8x8j23GzwcFAVEhc1nNmzA51evDrxcwZbGRnzp5puxhEho/ssVHpPDlHYc6xtFJmOsOUTxMNOR+0TmNqx/uBXDB6dwfHdCfJcjkYhlM5WzbX88u82f9DnGmEXAcUy9T6fTgd0vOibvEzmeJsfu7OpR8/h6LvFvVeapTg5OS0/uOchLR+5lBK2LV0M1suTTqZ52Kcl3Wew3ycrtffkW4CpG4S4nWWWpQiYPdnZi3bJlZQ+P2SnM+Ar78NaLRnaq1Ibv/NNJRJcSam4ksdc28uxbmBiatYoabJyab7/zPKaG522FDrxsAc9fSZx8y9pv1rxv5E3aYlJVRl69iYsdk1a9a6dlIK877bfgkNcSzm4j9b7n11iWhjz5Ql5GBq2LJyLUN0fR3tNgUdcAxqZTXUBV9fxQyLLY78ao2/vyLcBVrMJd5a4WWu7zZ4dYSahrTuxOoH1THcbNNnwnTI/76oixGr/0/Tewub9dbM5ycQMAi+Lm9p3NiJh5LlPD82iIRW2rVs4nrQmR6ZTVY+eOUaWWNSgZ6k3NJVNddy1Hz95YUWPUbh6CKqf0+oP5LUGg7syrS0OuwJHJVRefj+e7uT8m6nnw98rt08pBoWUd/CZZ2b1PjqfnI1Wsppr95cTOMdq0Z7Eu1KY9xgrz3OPTFseotbsOk0NzwoFSV8O1TREsjUdxNZFGhAi9++MYee4tTAzNYXlXrW3eCp8gZPWdDE++Csopq4qNV7eddNX4FhrCcPMQZNlkLpOJk3LHaWfeLVVaTcjwe71BbQiVi1x+p35RN0TtHv+noSEhefyRoioJQuETNtVQ2PASPaiCgYfSt1pWyFzcwN/P/1c3X1u76/Ch59/hWAzQTrUjK9IKVaO5bbxWhSfvVj5AjlsXI0bt5SFwvW0u53DT4Ps5//YDHRgbnAVhMQaeiy6+Egqa8ZrngL9+rLn8Tv2eX81S5W36uGLlyUQCc6YhmFMyVIPohBTWrk9hwkn04CQY+M4dL1seP3XHBdz3/Bqc+ZMrSE2nQQDqmqNZmeqTQ3MiTCs7lE5lD0aefcvy2KkNaDGoCiMP2N/Uaty6WMshtVaNathzPYdTCQKnuL5dPZ57B9ZYzi0nWnjhltEbhno3vIYLT/F/IpHI8pJVil3WQd0QPXzlim2bPr+fL0a99qAbq1frKsFOMJDJZPDCl2ZEn+R7T6/G4U2XMDGUwlN3vIz49qWirPC6nc1Zk4HaeU6uV6NGGb7zrpcxeXZORBmO707gnKniCeJ+qxoJpd1NrZYM4JIm/q8Q3GSR8pjcHvPn+JegrbseD6VvNaRbj03iqTsuYHBPIuszTuf3MyYn7ORmYTDwgGHMFqTf3TxjrkZVDWkJOdxjkxZZWy6ossR7VqywPN7R1IT743HUklHvRM1QVT9fDFlhEMfkFEPKGVZkwYDRg8HIeF2xqQ4N8Si6fns5Bv7HFaz67RvMaMCcpW78ucenMTk0h/o2oz+zXSkSufSBvPKORCJYffdytHXXo+uu5YhEjL4R6x9u1WUN3HCLyS+NR7Hs7UuQWcBiOQHGMHYqhVUB1lbxG+MmMmpfyNr6RemWtcO8n99Dvpl0pSxolis7mppQQyQ8+Vqyb2PHsQtpFVrWwW5D1E4meLS729b7zWVD1a8HHcQmLT/3K6lUoKuEcsMFA1zEwGvWbB2I4cefGhe25N4zqy2Z7DK/n7hFNARXOb47ITJl1WKCPXvj2LQnY2ktWMmlhkuC3U3dddcNABgitYTxU8Ys29pdh8TJq+Ixr63Cj1Esco1x9+yNYdOeFSKWp9a/8WO4C9k4LWVBs3zY0tiIo93dOcXki1ViWh2HargBiFUFf91NjcM3aL9w+bJjlmwucfZiShXV1oVLiADmXg+/kuGetl3NGllE4cR3ey+ifXOD5bl1u1rw8uHXheGXu7nJ96Iahg3y/qoKIw9k39TGbNkOIhLa1kmpF+PtO5sthf2LqSLJJ8bN4/rynkJtY8TSrszJcOc6qdhNFF6ebxD1NnIhH2NWSPjKD24G2ckb99Our1wetHxuMIaPd3RgZX191cXkVez28ywbqFKSH7AosxwfTGF80BqmOXdwCq3ddbj62jVRtK/cIoaqMfKAfTcYwFCcqLUmIuZ7g/oD5KrusAuX8Po4/LNOY81lUnHz+O3KKRBR1ckri4WTQXYz5F5GvFjJULkykEzilVTK4r1XehtAv9jde7wEML+HTu8dF1nltU1RdG6/Zmm9KdOxfSk637MM5w5OCdlkOUUMVWXk7XBKQ3ZK/8/X0Ntl29rVsrBTzLiFS5x27e28VK9JxcvjV+G/j7DLK/0QhFLEySCrhvzJRMK2p6qdES9HMtShkRF88he/QJoxLCHCxzs6rjsDb3fvyeFSuQsUvy/sjDxvOLJpT7slVl/OkGdVG3lRdMj8ZduV6wUWd8Lz9VCdmnXbJWONPPsWuu66AT17F5UXbhuFtY1k2bl3+rLYTWbH+0YtCRZeHr8dflcJ5Q7nuFFsPblXVmtrTQ0i5vcpSmTp9eqnp2opSwIMJJP4o1/8AtdMA3aNMaysr/fVX7Ya8Nqklx0y+buuev7iPQA2fm5FqEQMVSOhtIOIUNccRaynQdSF7t0fz+rOzqs55lO5UfZ0uTyP17yob4taFDNcQTOfzJZxbu6PWTZZiUg0O5CxkwAyxnB4ywWhEPhE5jbRcf7wlguW9ztJJfmxT/VnbzR5yStP9Y9lVdxzOlY5sAuR5IsqLQRg6ZHKm2enGUOECL/Z2oprjGWFZ8LSV/XozAwy0vcjaiqXqllCqcJbdKrOkJ2jp0qeAVj+Hz44hafuuFBU+W6hVLUnD1hrqwAwDfAc1u9qQe/+eFY1x1xjZ26e7tYvtgs5luoBe7XzE9Iun4oXpv7PDa70mJ9HzfSTO085bdQ6eSYAQh/OKWac2yumrjbPjpuNur3OXS6vmXfNmstkECGjk9OWxkZ84fLlqpZQqvjdpJc9f6ON37KsNn4TQ7OBd2XLhaqoXZMLdqEVr0L/flBrU/Dj2D0PwDUO6LThyeP88kQgLyHVDk3rdjZju9mi7PjuhOhi5dhpSqmfIadk889st6mLI79HHCsk2bKcYjX38Ar92L0OwPX45S5PYHf95R5T2JGdJrf/1fcHBenOUFbkP0QxjJPTcdQm3/LxAX+GkY9V7leZSxEk1SjLta0B2E5MgHUyPL133FK3gzcwcTt3pXSGyseYeU0KTkbTqbjZ0ZkZSyenINvz5cL1EJOvFtyMfEHhGiK6D0A/gF8FsJkxdkZ67VEAHwOQBrCLMfavhZyrmKgGvpAEILfj8KxVp+P7kVj6VbgA2UWQLI0JbPrBOoVg1PP17ouJqntchaNuuoZpoykX8tGle22Mqq97FTfbedNNICJElMSjchvZcteE1xSHQmPyLwD4IICvyk8S0a0APgzgNgCdAH5IRLcwxtLZhygPXrvqTgoWOy/b6ThusTnA2cjy8cl4qWKyJhpJVcQ/Lx/TaWJKnLyKeE8Den3q7v1OlmGIzdtRCl26W3GzuUwG+199FRlTiXNg7VpPrb1GkwsFGXnG2M8A202KuwF8izE2B+AiEZ0HsBnAQCHnKza5pL57JQSpRkz1mvnxuYJGGEal/vvYyato72kQdS/k8/Tui2Vl5fH3yBMNYwxqEE72rIkIE0OzaOuuF8fo3RfDa0ffxFuvLmB4cBYwn7M7n4zdJFfbSGjrrkdtY8T2dxUmgqz/wo+nTiT3rFiBY8kk5s16JmnGxEbt5MICgPJmvmqqi6DUNTcBOCk9ftV8LnT42VXPJSFIngw43MABEK/VNUUsvR237Y8DjCExOItzB6eE4eUx/fbN9Xj5/1llbE/dcQE1NxJWvHOpOB9jTCiGeLcZeV+AG+q27gYMPzYpyiaceGQMk2fnFjdkbbrYyCUWZNSuVfNJZimuFja1jUpQ9V9kD9xuIjl85Qq6b7gBj7/2WtZKolyZr5rqw9PIE9EPAdhlynyWMfY9p4/ZPGe7w0tEDwJ4EABWrlzpNZyykEtCkONkoNS/2LZ/saOTiHNLtXTk86zb2YzXnnsLU2eNPpK/P/IrwpDXt0Ux+tysOCbX6Ld116OuKSI0+oCRWMXVOW7Xk8lkLKGeh9K3ZrU15L+XfH9X1YqTBy5PJFxLP5/J4FgyiQNr12JyYUEYcl68TLcB1BQDTyPPGHtfHsd9FcDbpcdvAzDicPxDAA4Bhromj3OVBD9lA/xklBKRZ6XJrV9st57nQAcOb70IAJgdSwvJp1GT+gYhixTH3NWCrfsWS6aCMfRKHr28eapeD2MMT2+8aLn2E4+MYesX20XnejuNv9p4PJe6PdWEHw9cnQgmFxbw6KpVtquAMKhsNJVNUBmvRwB8mIjqiGg1gJsBnAroXEXFLpuU/2+3Uaq+3y2j1O417l3z42QymSwj++XoixgfnMW6XS2W5+89sxo9e+PCUxdjxmJVS575+uXoi5bNUbukqOceHsFTd7wsVgjAYsebpzdeNFYIG+qRmk5bMvhEpvB0WmTyuv2uSi3b5aV9S5G1yUMzn1+92nGzVG72ESXCK6mUiOMXKzNXo+EUKqH8HQCPA1gB4B+JaIgx9n7G2E+J6DsAXgRwDcAfhUlZ44TT5qqcROQlt7QzcIe3XEB7T4MIp8jwJuDc216seVOHe8+ssTQsUI3j0xsv4t4zq7MmBTkkpJpTbuDluD1PiuLdb/i5xUoAsEhBgewVSaynAcwcI18xtHXXo+ZGwsLrzCrHzLHReSGUQ6XiR2LJe8I+kUjga6Oj+MbYGA6sXVvUODyfOFpravDvb7wBwF8tfk11cV0mQ9nhVY1OTiJyyj4FIKSL/BhyFmrrhjrR25HHzXn3qquJxTmwrbsOsW0NmDg9hzGlwS+vg8+NNKchFsXa320EHKrjceS9AV48jYgsiUy8Q71bgpNd4hXfX6htjIgyyfK1dt21HPPJTFZmb5B84fLlUCYaAfZj48lRhcbh+eQ2Z5ZY4NQRefbH1VQegSVDVRN+Ngy9sk8nhmZxLcUMWaQUlmnvMbrH8CYCE0MpUTvnxO4EEievWox8bFsDfvolY6nOFTLf3XoRY4OzGD12FUB2k/K1v9toGNldLVi3s9li6NftagZAOCeVV+bH5XJOmRO7E2JsMrKOX/0MAbjtk02WOLyq7uGTUik3YcOsUrEbW7GUPnINHRneH1cb+esHbeQlvDYMvbJPudcaM40678LOjZ1slJl5vF7z87wlIQD89EszWLG5HtwGEhE++OPV+M4dL2NyaA6Ht14U5xDHY0zE322uDNv2xy11bbYrSh7Zwx8+OIWR597ChNlJS5ZVwtTgyysBbsR5HH/xrMCZP7lS1k1YPzr4cpVJDrJ2PJ9AVE/eqz+upvrQRl7CT3q+m8cv69rlcgK9+2K2JQeYWY7WrhFw/N1LjcYEpmZ+2/44OrYvNduOzYpVQXtPA9KpDF54fBq372y2HOOh9K1iohl97i3La8f6RrH9QIdoIl4rxcjPHZyyGHg+IcBMfGKMWRqPb/1iO37+zRmkJqzbLsMHp7BuZ3N2nXuXzN4gcPOOy931KqjSAfIEomPy1zfayJvkUsvGzeNXn+/dF8PhrReFOmb7gQ4ce3gE5x6fFhudwGK8njM2OIsPnujK2uC8fWez5XOxngajr+SGOoyfssbveSeqUdMrb99cj/i7l2JYqmlDgEhcAgwPX/X4+XXLv4NNezJZjcftGDl2FZNDc2jvaUDc3JwdfmzS8JZhFDwrVxZsLklulYiuPaMBqrxpSC441aBZ/3BrVi0bJ4+fNx5Rn+d5YPwIdoaDb8g+lL4V63a1YHxwFt94288xdvKq5X2JY9bHBMPwT56dE02FWzbUAjA88h8/MgaqNc4Xe/dS9JqySv66vEm8OF7r+PmmsjxuLtGUUVcS63a1oK4xinW7WhDvacDwwSmQ+fz44CyGD07l1KCl2Mh/4+HHJvHXkZ+WdFNYoykFWl2joNZoB5D12K0fq13VyXWSUeXwju+c+rYoUhNpEd7h8Xc7WjfU4kM/WSs8aXWj9Q+v/SpOPDJmrS1v7gs4qWbcrstu89muvHJbd50I8wCL9ek5Ya05X6oyyeWuKhm2cWiKh5u6RnvyCnINGtmr5Ubt9N5xR49f7u4uP1/fHF2Ma5tMDs1h/cOteCh9K9q665GaSIvEoy9HX8Tk0BxalTaF0igBGOGFdTubcf6p1y2vPr3xYlaCFI+fO682YFNozIzXN5LFsA/uGbNMAPwaJoYWVyPrH27F8MEpcWynJLEwGHg/SW6FEpZ2emEZh6Z06Ji8DV6xWnXTUK5eaVfVEsgOg9S3RbH1i+2WWjOrfnuZRfu++q7l6HzPMusKwIzd85o0I8euYjaRNnXyN2Lk2auYGErhidhLlvNNvziH0eeuihr3sn7driaNUWgs41horLaRLBNC1103AAC67lpuCeXwUFcYa84Xq6eAH4pdVTJfb1xXt7z+0EbeBl4BEvBfZEs27IA1O1WWKcpZrTybta27Hvec7sLA/7hiOeb8TAZQPMrO9yxF553LLONq3VCHzjuXmXLNZrz56gLmJheFc2oSFrCYwSpnoOZSh0ed0Hr2xrFpT0Z0t1cnvlIZ01xw2ocBFiengWQSTyaMyakQZUqhen3ZqAPIakLCC5x5jS/MeQOaYNAxeRvkZCdZ277xcyvQs9dbCXKqfwyv/Muboi78qf4xXDryBiK1hFW/sRwbP7fCclwAWVmwcny7dUMdPvSTd4gYvKrEeSh9q6VGvQqXUtY1RXDmTybE83IfWicjm0u82k1vXm6pohtO4x5IJrFjaAjz5j1SaLZovt63WprhgVgMXxsdRRpGvDVi/j6L1b5QU3nomHwOyKGap+64YHnt0pE3kMmoOYQ2n59OY8ysC3+sbxTzMxlMDM1h/FQKs1PXsjJJ7Tzs+55/B9p76tEQi1rCM6qBByDq4ahxeM53t14Ey2SMlYHE4S0XPA2833g138NQC7qd6jfGtrk/ZitDLbeB52Oxe3x0ZgYL0rXybNF82dLYiEdXrcrZsKohFgCojUTEzZtmLKuomVtRtnzHoalMdLhGgYdqeOyaIyf/uIUXiEgoSmQ9OmBIDCNEFtmi2qYPWAxf3DvwDkvBL9lL55p3rjvPZDIYfc4qr1y/q8VIqHp8WtTAkTNbxwZncbxvFNskLTyH1+bhIRa1KbnvOvqS3tzJmIaVHU1NqCESnny5skXVEMv98TjeuXw5PvmLX+CamYEcgWH4W2tq8NBLL+GJRALXcvDuNdWLDtc4MLgnYQlt8MYZfsMLapgDMEIdci3203vHkZq6JhKGOO09Dbh3YI1zMbCdzUaSlFmrJsMYLjz1Oq4m0miIR7H2vhstrzPG8PJ338DV164tHsMsVWCXjMRDK7woGzfwvBqn3e/ATlIZFolkIRQrJl+MccghFrm4WQTA+5qbcc+KFeg7fx6pTEZUHw1bUTZNMOgCZTnCW9jJ8HAJ31iU36vGcwFkpfIDi6UEOHPTaaFvl0v+jg/OWt6rhkxg1ryBmQ3LkatK8vfVNUWwaU87tj8Gy0TBj60aYNkrVz14/lj9HfDjVGOjkLBkjarjUL37/q4uEdbh31wC9OaqRht5lVyUILabiX2jSJy8KgqOycabh254qYBtBzowNjgr4vfAYjiF94SVE6q2Sw2/eVhJNqr3Pb/GYoDVTU4ZuX6MjB9Vjdvvze4clW7ow4hTcTNu+KNE+Gg8rmvVaLSRV/EjqwNc4tAHpxDraUD75nrE3r3U4rknBmctx+DVJWWlzbYDHTjVP475ZEack08UouE3FjtKycj7BeoGKJdwbpMmCj5uu1h5Ll55WCWS1Y7q3QdZ1VJTuWgjbwNvAmKX7MTx8njl9wFW482PnclkshQ8x/pGhecPwLI5y3uzqiGU2kbCpSNvWrovHd56EQTgnoE1xqRlrhBO9Y9ZJi07cvXK/U6MXpSr5G81EZbwkiY8aAmlA36UIG6p+nZqEt4ViRczW2z1t1gK4NzBKTAYYRtRNEtS4xARIpGIMKq9+2KYTzJxnNrGiFGf3gwDHe8bxaY97UJRc+nIm2CMYdv+ODbtabfIHIFsr/wTmdtEAS+3dP9CJZJeEkyNRpMf2pMvgFw8XifPn2+WqqUANu1pt0gr1WPKqw35uFz2Ke8F8OOoMlAnmWO+Xnm+EslqL/mr0ZQTLaHME6+esE6hDVUOyfupyq8DzhUbAXvj6VRZUj2XrHWXj2unsill6KRaJZgaTSnQEsoA8OPxqoaRJxjJ2CVXcWMX62lAe09DVuu9ekXbbrei4LF99VyqIserFo/d4yAmgGqVYGo05aagmDwR/SUR/QcRDRPRPxBRk/Tao0R0noheIqL3Fz7U8OEWh1ZjzDwG7xXrFpPHrha0m12feIx+9ORVnJMabfB/agx93a6WxYYgu1os51I3enMtqxtU7LxUJX9VDo2M4P1nz+LQyEig59FoykWhnvy/AXiUMXaNiP4cwKMAPk1EtwL4MIDbAHQC+CER3cIYS7scqyKx83jtYsxySeHefTHXWDePt/Pjyd4tb/4BQBT4UlcU2039PQGiZIFcqiFfmWNQsfNySTAPjYzgD37+cwDAD6aNpLQHOzuLfh6NppwUZOQZYz+QHp4EcK/5890AvsUYmwNwkYjOA9gMYKCQ81UKbvJKOWPUTpopHwNAVgiDv9Orvv29A2ssz0UiEXTdtRyddy7LW+aYb6KUn+MWQ4KZK4evXMl6rI28ptoo2sYrEX0fwLcZY39LRF8CcJIx9rfma38D4J8ZY0+7HaOSNl79UGhbObvNSJl8jGsx4ulBtcsr9Wav7MkDwFdvuUUbeU1F4rbx6hmTJ6IfEtELNv/ult7zWQDXAHyTP2VzKNvZhIgeJKIzRHTmiuJZVTL5xJjl1xhjONY3mhVrl8nHe85X5iiPK6jYeaFjy5UHOzvx1Vtuwa83N2sDr6laPMM1jLH3ub1ORA8A+C0A72WLd/qrAN4uve1tAGx3thhjhwAcAgxP3seYQ08+MWa1Dg4AjA/Oor2nYTEco5zn+O6ECP/IDbaDMo7VWL7gwc5Obdw1VU1BMXki+gCATwO4kzEmFzM/AuDviGgfjI3XmwGcKuRclUSuMWanDc2xwVlRKvjEI2NCLVPbFBFlDEaefQv3nlmNH39qHDU3AguvQ5QCLrbBL1fsXKPR5E9BMXlzQ7UOAA8an2SM/aH52mcBfBRGGKePMfbPXserxpi83xizVzKQ3JKQJzTVt0WRmkiLTFb+WO3dyjX1xYp56xozGk24cIvJ64zXEOG1oSmHZJw2ZHmZBLmAGW9QEtYeqxqNpjAK2njVlAY/G5pq0pUdE0MpfDn6Ypa8koeD+DH5JMETqzQaTXWiyxqEgFw3NO0mBDvkzwWhb9doNOFHe/IhwGlDc/3DrVkbmvKE0NZdDwDi//q2qOW4arkEp7LIGo2metGefEjw06iEP88nhNrGCDrvXIatX2zH0xsvirIJXXfdgPkks6wEgOxesbo9n0ZT/WgjHyL8JgPJEwL/n5cskHXzQHav2GrRt2s0Gn9oI1+hyB4/APTsdV8JaH27RnN9oo18FeG2EvAbDtJoNNWF3ni9jih1bRiNRlN+tJHXaDSaKkYbeY1Go6litJHXaDSaKkYbeY1Go6litJHXaDSaKiZUVSiJ6AqAyzl8pA3AREDDKSX6OsKFvo5woa/Dm1WMsRV2L4TKyOcKEZ1xKq9ZSejrCBf6OsKFvo7C0OEajUajqWK0kddoNJoqptKN/KFyD6BI6OsIF/o6woW+jgKo6Ji8RqPRaNypdE9eo9FoNC5UpJEnos8T0TARDRHRD4io03yeiOggEZ03X39XucfqBhH9JRH9hznWfyCiJum1R83reImI3l/OcXpBRPcR0U+JKENEG5XXKuk6PmCO8zwRfabc48kFIvo6EY0T0QvScy1E9G9E9Avz/+ZyjtELIno7Ef2IiH5mfp8eNp+vtOuoJ6JTRHTWvI695vOriWjQvI5vE1FtSQbEGKu4fwBulH7eBeAr5s+/CeCfARCAdwMYLPdYPa7j1wEsMX/+cwB/bv58K4CzAOoArAbwMoBoucfrch2/CuBXABwFsFF6vmKugBrAqgAAA09JREFUA0DUHN8aALXmuG8t97hyGP97ALwLwAvSc38B4DPmz5/h36+w/gPQAeBd5s/LAfzc/A5V2nUQgBvMn2sADJr26DsAPmw+/xUAD5ViPBXpyTPGXpceLgPANxbuBvAkMzgJoImIOko+QJ8wxn7AGLtmPjwJ4G3mz3cD+BZjbI4xdhHAeQCbyzFGPzDGfsYYe8nmpUq6js0AzjPGLjDG5gF8C8b4KwLG2HMAppSn7wbwDfPnbwD4LyUdVI4wxkYZYz8xf34DwM8A3ITKuw7GGHvTfFhj/mMA/jOAp83nS3YdFWnkAYCI/pSIfgng9wB8znz6JgC/lN72qvlcJfBRGKsQoLKvQ6aSrqOSxuqXGGNsFDAMKID2Mo/HN0TUBeCdMLzgirsOIooS0RCAcQD/BmOVOCM5dSX7foXWyBPRD4noBZt/dwMAY+yzjLG3A/gmgE/yj9kcqqzyIa/rMN/zWQDXYFwLUKHXYfcxm+fCKueqpLFWNUR0A4DDAPqUVXvFwBhLM8a6YazON8MIaWa9rRRjCW37P8bY+3y+9e8A/COAPTBmx7dLr70NwEiRh5YTXtdBRA8A+C0A72VmsA4VeB0OhO46XKiksfpljIg6GGOjZthyvNwD8oKIamAY+G8yxr5rPl1x18FhjM0Q0VEYMfkmIlpievMl+36F1pN3g4hulh7eBeA/zJ+PALjfVNm8G0CSL/PCCBF9AMCnAdzFGLsqvXQEwIeJqI6IVgO4GcCpcoyxQCrpOk4DuNlUQNQC+DCM8VcyRwA8YP78AIDvlXEsnpDRj/JvAPyMMbZPeqnSrmMFV8oRUQOA98HYX/gRgHvNt5XuOsq9E53n7vVhAC8AGAbwfQA3SbvafwUj/nUOktIjjP9gbET+EsCQ+e8r0mufNa/jJQC/Ue6xelzH78DwhOcAjAH41wq9jt+Eoeh4GcBnyz2eHMf+9wBGASyYf4uPAWgF8AyAX5j/t5R7nB7XsA1GCGNYuid+swKvYz2Afzev4wUAnzOfXwPDyTkP4CkAdaUYj8541Wg0miqmIsM1Go1Go/GHNvIajUZTxWgjr9FoNFWMNvIajUZTxWgjr9FoNFWMNvIajUZTxWgjr9FoNFWMNvIajUZTxfx/0DOVisXrg+MAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cluster_and_visualize(male + female, extract_vectors(male + female, dhd_limit, dhd_w2i_limit), \n",
    "                          'Double-Hard-GloVe', random_state, tsne_random_state, y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 249,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 1.0\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXkAAADCCAYAAACsebI/AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjMsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+AADFEAAAgAElEQVR4nO29e3RU15Xu+80qVJIQWG+pJGIMGOJrYgnZxggM2PRJcpKT27HTscnNuHdcO905oePEyOCccRNfjxOEMzLcj2OM5Tzp23HsMdKdtk1yQk4/nHZGsAGDeCQC4bht3saWqoRehUDvqnX/2Hstrb1r73qoqlSl0vyNwUD12nvVrqq55prrm3OSEAIMwzBMfuLJ9gAYhmGYzMFGnmEYJo9hI88wDJPHsJFnGIbJY9jIMwzD5DFs5BmGYfKYedkegE5VVZVYsmRJtofBMAwzqzhx4kSvEKLa6bGcMvJLlizB8ePHsz0MhmGYWQURXXJ7jMM1DMMweQwbeYZhmDyGjTzDMEwew0aeYRgmj2Ejn4PYi8ZxETmGvxPMdGEjn2McbQ3i4PaA+hELIXBwewBHW4NZHhkz08jvgPxORCIRdX+s7wRPCIwOG/kcQgiBscEITj3Xpwz9we0BnHquD2ODEf6xznKSMb66YZffiVfuPI/2HcGY3wl2Ehg7OaWTZ4ANz/oBAKee68Op5/oAAI0tFdjwrB9EBMD44cq/mdnB0dYgRgfC2Li7DkQEIQQObOtGUbkXa1prLc/VJ3sAuPuZGnS9cR29HaPo7RgFADQ+Vmn5Tji9bsOzfjUhND5Wyd+bOQob+RxACIFjO3swNhjBhmf9WL+rVv1QAUBgyrBLz6ywzBNlHJjcQTeoQghc+tch9Bw1DPTG3XU4sK0bnW39qGkuxl07aizGl4gcJ3sdX6nH0WCv31Ub9TqnCYGZO7CRzzJHW4MYGwhDAOhs60ckEsG5l69antPZ1o+uN65j6f03YDwUYc8sxznaGlQTtpyYw+NG+KSzrR+dbf3quf7mYsdjEFHUZK9zcd8Q7tpRDY/Ho8556d+uwd9cjPXP+i2v85USf0/mMByTzyJqed3WDwJw29ZynH5+ACPBMACgYpUPDS3lAIC+k2M4/tQVZeDZM8tNnPZVDj0eRF/HGCqbCi3PbWypwHqHkIv8/9DjznH0qqYi9HaM4tDjQQghIITASP8ketpHcKqtH6/cec7y/Au/GlKbtvo5mLlByp48Ed0I4CUAfgARAHuEEM8RUQWAfwKwBMBFAF8QQgyker58wr4st9N/chz9J8ej7k+ngbevBnh1kBpuoZaGlgpACPR1jKnnBo4M49D2ADZocfqD2wPwlXosK7aCG4AT35n6fiz+0/kQQqiQTfuOALrfHEbVXYXoPTaGXu0cxbVe9J0cU+cBwOG+OUY6PPlJAN8QQtwKYC2ArxPRSgDfAvBbIcQKAL81bzM2dKOQKLp6IhVYiZEZHD9TIdD5vNXH6Tk6ilNt/Ti4rduipBoPReArJTQ+Von1u2px+TfDltdd+vV19J0cw/uvXUM4HMbFfdfQd3IMItofwM1fuAEAEGgfAQBWa81BUvbkhRDdALrNv4eI6B0AiwDcD2CT+bQXAewH8M1Uz5dvCCFwcFu35b6GreXoOjBs8fqqmoqw+cQyHHo8aFFPTNfrFkJgbCCMU2Z8eMOzfhzc1o1Tbf0c708RabB1zr5i7LMUVXlRvtKH6qZinGrrx3y/F6fa+tXnoIfiIpEIDm4PoMc00A0tFQCgYvo97SP40bx3AACVqwrRd3IMdgjG96nz+QH8wPN21DmY/CetG69EtATA7QDaAdSaEwCEEN1EVJPOc+UDyntr60dtczFqmostHl9lUyF8N3gwcVWoGKxUTxSWOasrEuXYzh4IGHFhPaxQ21zMBiAFdI+8qMqL0d4wiqq8GAmE4S0ijPaGMXFV4G7zc/SVeXD8qV71ev3aezweFJV70dBSAQLURAAANc1F6GkfVbf9G4stRv62R8tw+nuD6Hx+AA1byy1j1DeE+XPOf9Jm5IloAYC9ALYJIa4m+uUhoi0AtgDA4sWL0zWcWQERobDMozwrSfDoKAjA599aCo/Hg0gkgkOPB1FY5oHH40nZCMvNwc62fuUdSmpc1B5MYsjPtKGlAut31eLV1ReUtj08KtSKzOPxYP2z/qjN1YPbA5bPd01rrQqr6Ea+tnm+xci//b1B6zg8HjRsLUewfURJN/VzrN9Vq75THJvPb9Ji5ImoAIaB/5kQ4hfm3UEiqjO9+DoAPU6vFULsAbAHAFavXj3ngoTyR6wb7QcPLwMAi0en//BT9b5kzFgIYZHzAcbynkkdgvG5bT6xDD/0/lHdLw28VM/IzdX1u2pxyFwBAIbenYiUx22fDDrb+l1DNFVNRehs68dtW8tR01yM088PKEVOVVMRTj3Xp5KrODSX/6RDXUMA/h7AO0KIXdpD+wA8DOCvzP9/leq58hX7D8zpB5eJH6H9iI0tFYa3aE4C/MO3Ek+JJD1uKYuNCIHuA9ZN05fvPIfNx5fB6/WqVZyvlPCLuy+gek0RGlsq4Cv14ND2AALtI/AWQoXr5GTw4qJ3MRwIo/7e+QBgMfRFVV48cGwJDv+3K/CVEi6/dh01a4rwwGHrfo48Hn/O+U861DXrAfzfAP4TEXWY/z4Dw7h/kojOAPikeZtJM9MpRiWfIxUX6n4Yhj7VeH8+YlciyU1RqURq3xFQm60bnvWrnAe5eX7b1nJUNhWir2MMP/W/hyPf7saa1lqs31WL8cEIgu0jOP38ACJCYHzQ2BDvaR/B+NWI8sBliGU4EEZlUyECR0aiPPnR3jD23nURdz9Tg/GQQLB9BOFxQ0tvV/ywgZ8bpENdcxDuq/yPp3p8xh2nzMp4Gmg9w7anfQS3bS2HhwiB9hEVo79rh/Me+VzV1NtrwvhKPbi4b0h5w1LGKGPvTsaTiLD5+DL81P+euflqHNfj8Sj9+qm2fpzWZJYNZs0i6YHLsI/Uzl9+7bqaOCTe+YTejlGlupFhmldXn4d/43zLmF5dew4PHrl5TnyGcxnOeJ2lTKdipZ5h29M+goaWCniIcKqtH/7mYjS0VKCo3Ov4o5/Lmnq5h9H4WCVOPdeH409dsXjXb32jxxLv/oHn7ai9js62fvxo3jsY7Q2jek1R1EQgDb2OLGZm98DX76rFeEig5+go+jrG0NhSoTbQw8PWz/3B40tR1VSI3o4xNYFUmZm3PUdHccDU6DP5C9eumaW4ZVbGirPaXxM0wzUy1uvkpas4s11TP8eqG8prp2cm93aMWrzr9btqLZus0rjK/yXCDJ+o1de2bgSOWGP3gOFp166dH3X/3rsv4IG3lhrKGbOUAWCEhAIHhi3neusbPXjw+DLl2RvjNiYGAbhO6kz+wEZ+FuNkeNwMvDQqRARfqfVxGevt7RhBVVMxfKWE8ZBQ9/tKPRAwNPRztbqhU4KTjrxWOv6N81F3T0mUV9/bMVVmQCagSXRNfM/RUSV/tNzfPoKD2wN44K2llknFQ4TejrGphKvb5yslTdR4zc9tLnx2cx028i7Mhvizk+Gx66yFEDja2oPxkBG7j0QiuLBvyPKaV+48bwk3yDjuh/uNdHl5+zZTdy1xS6rJxWuVCnooTHrs8ppJXr7znBE6MSe+Nx/rwunnB1C5ylqUTG7I6lmutc3FEEKgdu18bNTCNt1HhjH84SSufzipJgrpgReWeXHAlindfWRYJWBNXBVY9z+qlVRSfoaSV+48j80nluXV58Q4w0begXgbmolI6TJt9OyGZ8OzfhzY1m0peXC0NYj3/+0awuNGYSwRiaDrzeEoRYY0Ag8eX4q3vtGjjiGfJw18wCYHNIppGV6/fq3szTBmu9HXk9amVj3GNVty3wJcfu06gu0jqGwqxN3PGJvWHiJUrynC9Q8mLMfyEClDL3nAzIuQ5wKsMXpZjkDeL3Xzp58fsIRorphef1VTodLj3/TZEgCwSDDlBHXo8eCcWYnNZXjj1Ua8DU3Zfs1tAzLTG5TyuMrwmAqMYzt7jDolpgRSCIGL+4bU5lxlUyE6nx9Qhru8scBy3CX3LYTX63UtliYNSVVTER4Jr1SbkBf3XbNcq1fXnUdnWz9GB8KqDG4+bNCuaa3Fhmf98Hg8yuBvPrEMa1prVZbwSM8k3no8aEy2bf2IjEUw3BVGzZoifC3yMeOatfU7TpZ2dOfC/lx90rnnuXpsPnGz5TmbT0wpZtY+VYcl9y1UjoBM0Gp8rJKlsnME9uRtxNrQlF6cW3s1vR+n0+NOHm0yXr99hXHXjhoc2NaNo61BjIeEUVzMlEAe3B6wbPrpMjsA+MimhRg4NRULHh8Mq/IJTshjSQ9RXiNfKaH+3pKoDkbyHeTTBq291IC8vdH0rmUYBjCKggVNz9q/1igNvH5XrSXTVP9+BI8Mo6a5WIVrhBDK425sqTDi9zIjVgjl0Qsh8OKidy3jfOXOc/BvnI/iinlY01qL5p3W8aajNAYze5iTRt4tU1GPL9u78sgfRTxFSzKKl2R07m79OzvNqpGWtm9aRUO74kPSaU4IG3bXYe+68zjV1o+uN69bJoaiGsJoz5S8rnZDseU96Bu0+rWqaS5yrayYL+jvhYhwz3P1lhCMLDLXsLUcd2t1YpbctxD198xX18NXSqhqKrSUmCAYiWq9HaMo9nvV93PDs35ACATaR/A//+QCKlcVoeuNaxgOGE1mKlcVYjgwid4Oo6b8bVvLLRvubuNn8ps5Z+SdDOvedechACz+1AKMhyK4+5kavLr6guV1sqiT9ILcFC2JKF4s6e+aZxbL6010gtHPu35XbVQZY8BoJDESDKvesTXNxQi2j2D4ShiNLRXolnK+iAdAWL3uwqtDKK7oAREsLQs/3H/NcvzwqFV3nYyBn42buLGUN8H2Eby46D2MBMNq0j24PYBjO3tw1w4jK1VKGmvXzrcocW7bWg6CMWGQ+b2TSWyyPk3JonmoXOUDQJa9lvmL5uGe5+pz/toxmWdOGXlHb3hbt1KMhEcj6Ds5hvd+NojRXqM0rKwcKKVoN322BBPWFqwWRYtTyEN/XJ9kpGeWqNcbawJxMjQv33FO/fArmwpRt6EYp783iJFgGBWrfCgs81omj05tHFKlYYRobsYrd55Db8cYLuy7ivqN840Sti0VUUWyKhp9URu7dsWPG9PJ4J1J3FaAcnK2Z58CUBLIylWFKtwnV18AYnYGu+e5egBGRclYocPrH0aP9UuXb2EDzwCYYxuv9szFH3jeVnHshpYKZZxGew3vVRp4I2vQkKBd+vV15W2v/naVmgAObg8gHA7jlTvPq8fVZpv5uB6zlwbZnmsYyxi6bcTJOir6eWubjfrixbVe3La1HJuPL0PPsVHMKyEU13qx9L6FWNNqxPR/cfcFFJZ5Lcctu9WHoiovejvG8EPvH5X+2neDBxufq0fjY5XobOuPMuiLNi0AYMgC7e8/VmbldDJ4ZxK3DfVjO3tUaeH6jdGJS5K+k8Z1tPfodcpolei1cHTkBmqsjmLp6h7GzH7mlCcPOHvDMjHEnrQCwFIHxKgY6EH9vSVRUjpfqceS3i4zSOUPUdaCt8TOHbw3J6/X7jE2tFRgo74RB6Oeil6XXoZgFt5UgI2763Boe0DVH5+8HsZEyJA6yvccHovow8DkkFCTnWS0N4zq2w1DZr+GkqBZLsGekh9PyTGdDN6Zwm0/RDfYgNGIpaGlwvF7pKO/n0gkgkO2iVtq4U8912esHmyv1+vB66S7exiTH8w5I+/kDUs1QjzsyT+6UZKJJnIp7fEYiyT5PD1UE7VRuaYIDx652WK0ncI70mMkGAZFN6BOig/ACMGo1PuWCkRsKhBgKv3erviQIZup5xWpScoe65eGqdPsciXR3388ksngzST6RrzEaXK2T0BS7aTT0FKOrjeuo09ryH5wWzc27K7D0dYeVeistrkY1WuKEDgwrFaXt20tR0/7CILtI1Gfja7S8ZWSKpCWzu5hTH4wp4y8c+biOaVGqGj0of/U1I/RbuTsXnY8o6QrG/SeqvY0c3spWPnjdNtDOKXFdO0Tin7OjbvrLF6lTLA5bWsoveS+hai/d4Ea+/pdtfjwjWtR8eXejlEc2h6wGPOa5mKVbi/DXvZ6KMlsusbL4M00R1uDuPRv1+BvLlbX68C2bvS0j+DGTy2wPNf+WUu1k2rlCIHONrOV4yoflty3EBOm1DVwZBj+tfPVyu/zby3FISV7LYLPnLiP7exB7dopNY78jvR2jFgmmbt2pLd7GJM/5JWRT0SZYW+3t/nEzaoJgzTwt20tR7fZSFtmNY6HRJSXLYtL6Ugv7djOHssmohxPsd9rSS+Xm3Uy+9BpEgGcPUj7OfSNSqmVt48t4hCnHR+MYL0W3z3+1BV4fRS1kVjs98JX5gXR1BjUNTYTdFZ/u1pNOsnglMHrtLLJJEIIjA6E0WMW/gKgJjQgOqSlT0B6gpKRbdyDsYFJpWTyb5yPiZCYUi4RGdecKKqMsD5x21do+soolvadDTwjyRsjH0+ZoT8ukY8//OEtFi35Pc/Vo31HEIvuXaB+cJFIBBDC4mW/uu48etpHppJVTC870D6M2uYpOZzStD8/gJo1RRgJTK0OvnDiZuWBAXCcpJykkYBWGVJKMKWX31Kh4u3KYNoKYdWsKVLlDk619atz+cq8GB8MOxbGGgmEMTYYVqEgfZKR11WWOmjeOXWdE5FB2o1kMvH8dCFXP4BVaQS4h7SAqQlI78c6HoqoJtoC1tXTbVvLlbwxkfCU223WvjOJkBdGPt7GWCQSwehA2GJ0lRFsqYjyeF9ddx7+5mLcbRp4IQR+cfcFCCHw4JGb1TmHLhqev/SNp3zkKW9L98ClwdQbK8sY6vGnrlg8QzlJyYQjnVfuPI+bPlsCYR7TXuxKriR0g7n+WT+C7SMY+mACN39+IWBuNFc2FcLrI/QcHVWxX/ncoNlIBEBUadpY17yqqQh37Yioa5doLZtEvdZYx0gVpzAXEB3ScpuA5N/xpJFusteZDk8x+U9eGPlEwhqyrov+eM2aoqj4MoRA5/MD6GkfwYdvXsfS+xaq9myAEZ/duLsOhx4PYiQYRuWqQnS29SujYFeW2FP9paetJz9J769T98o1g6kXl3r5jnPo7RjFtQ8mMNobRkWjz3ItZB0V3WDKVczn31pqMS61pgJHok8KDxxeFlUYS17rWNdcjleGn+Rqp6GlwqISctO+O3mnTqu0TBVBk8e2I0NaThOQE06fv0RXx2QzPMXMDfJGJ++kN5a3ZTck+0/Gv3a+oVgxu9p3mk2sG7aWo9jvRV/HGI4/1WvZVOxs6zf09eaPc/OJZZZjyuW+1K7rBNpHUNNcpDx+mdIePDKsio2dsh1/yX0LVRGy409dQf29JahY5VMbwvpGMWDtw2j3uKXkTibk1GgqGP16uXmYiVxzWfxK5iHI2La9lk2i2nd9/LKLkSyCNtI/mdYiaNLA66sX2XHpVFu/semsjTmWEXbar5HHk5+FLnvVczhYFcOkk7zw5IHYygw9s9TyGhiyt0PbA/CVeZVxcsKQvFmNyN3P1OClj7xnPee2bnQfvo7rlycxHAhrKh7Do5WeP2AYPtnFp3btfEBElwM4trNHlR8YG4wYIRZbjXKJnCSgeZluHndDSwWCtm5E0oOVHYecirLpm4JO11yGn/Tr2GgLKSWjfZfjDxwZtqyYAKMypirOloYiaESEonIvapqLLeoawCglUJhgFyW1iRxDgeQke01GbsowiZIWI09EPwHwpwB6hBC3mfdVAPgnAEsAXATwBSHEgNsxUkFXZtgThaQ00e4zSmPbbRblkgbNzcgf2NaF934WstwnmzIDUA2x7Zt1TvXHF21aEHUepwSag9u6VTgplrRRXQcYBtUpTuwrtS7ahBBqb0DXuANGgbGGlgr4Sj0W3bWvlNQmsVTvOIUb7BLRZLJ63fA3F6tVATC1ESonyXQlTa1prVWNzO05B4ke276JrD2grh1vnjIzBaUj9ZmI7gFwDcBLmpH/GwD9Qoi/IqJvASgXQnwz1nFWr14tjh8/Pq0xxNY3l6jGDk40tlRg/bN+7F133rIpWlzrRWRSYKxvSjrnKQQimo0tqvJi+RcX4vT3BtV91WuKULd2vsXg69I4IYQl3m0fi6VYmc0A2/FVEMb7hYqFy0kOmDIYkUjEUsdGvreRYDhKJlnbXIwHDi+DbEyh5xTYY8h2CWckEpkqj6s1MrGPPVmDbA+jOPFIeOW0pJuZZDYWW2NmJ0R0Qgix2umxtPwqhBBvArD/Au8H8KL594sAPpeOc7lx144a+JuLcaqtX4UQCEbo4eK+awi2j6CqqdBoeGHGWSURM34qDXxxrReVq3wYCYYtBh6wGngAWLh0HjaahaQkN316gUV3DkzVG3GK1evoZWUbH6uEr8xraQlnZ7xfoHJVIZbctwCNj1WiqNyLYzt7VO0SucqRdWwkctN483HrnsIDh42WcFJ3LUNYet0VWbJBNtLQtdp6gwpgKhafbC0b/XrIJKMG2+em88qd59G+w/26ZgP20plcIJMx+VohRDcACCG6iagmg+cy4pm761Ryia6wkY2rZcamPSFI1zBLz3YkaIRhvPMJ4eGp59uzYK8cG8MLdf9hOd75/3k1qo/qwe0B3P1MDV658zz6To5Z9OeAESMPHBk2ysqa70WWpd277nzM915/b4lFqWJXacgwx/L/o9TiDS+5bwHe+kZP1Dj1eL5b028ZU7YbLnuDisJyr1qdTEf7LkMf8nrpeIuBrwzdildXX0BvxyiEELhrR03OefQMk03SEq4BACJaAuB/aeGaQSFEmfb4gBCi3OF1WwBsAYDFixffeenSpZTGYQ+FfC3yMRVKOGRuhkkaTVlfp2bkHwmvdGyyoVNU5cWXAh9VKhUdvTTCVLleI5SiQiSrCvGF39+MYzuN2iUeH2HxpxeojdWa5mI8eHhZlNLDV+bFxX1XVRwagCojoEsJAauhl88DrGEfXZ5pL5mw4Vmj6bc0oG6vSXQjMpWwhVSqyLH5Sgl/+Ns+hEemvruVqwrhLfLgpk8vyImyxAwzk2Q8XONCkIjqzAHUAehxepIQYo8QYrUQYnV1dXVKJ3RT2AghjPCDLexx965awGZsXr7jnOW2xyZkIa9RjfGtbxgFwor91hK90sDP9xsTQUNLhTKScnVQf28JiAhjgxH0dozBv3Y+1rTWYuPuOjS0VKCnfQQ/8LytDL70hNe01kQVUut64zpWf7va8v714mU6MgNWhk30iplSxSM3boUQysBXNRWpY9irbCZCrLCF3clwcjqIyFgRmBPLmtZarPxKmeU59feWoKd9JCfKEjNMLpFJI78PwMPm3w8D+FUGzxVV+8Splrt9Avip/z0V6/3q5K3wzje661SuKsRfTvxvKKz0IDIGeIsJX528FVVNRRBhw0OXcWq9RIHOwx/eAo/HExViaNharrT20ojLUIaeVi958PAyNTkd3B7AabNZxyPhlahqKkLfyTG8cud5S035sYFwVEKPTEjSSyA3PlaJmz67QE04nW39U7JSTRH04PGllmMtuW9hWkIiyTQ9t8f/Ybuyqp0hSxAZxkJajDwR/SOAwwBuIaIPiOjLAP4KwCeJ6AyAT5q3M4Zb7RO5vNeVIdJgj/aGUVTlxd3P1OCtx4Mq9u4pBLxeL0qXG9mk4RGBH817B70do5jv92LhsgLLuQurPKi6y5p5+uq68zhgqxcDAF0HrNp0v5aQFGslYn9/Ho8Hm08sU+ETtTGqqXH0yc4oUzBsOd76XbWYuCqU5y8bgejJWA8eXxoVtx8PuXvLiXjm8v5km4To2bpOG7HsvzNMNGmLyaeDVCSUErf4r0yNl7VgZB9XPd6s11uXNLSUq3KxAPCxR8vwtiaX1PnYo2U49/OhqGYbskjVuZevqpDN1PGnvE83rb9sBiHVOfrGZyQSsewhfC3yMefqlNu6EdCSnNwaX+j7GY+EV7qm3jvF5N2KwPlKybJJKz8T+V4siVMJxPqPtgYxOhC2bFwDU/JP9uSZuUasmHzeZLxK3OK/0iDqRuXB40tV1ydgqkOUtd669XjBg85aewCOxr9hazlAhNNt/agwZZnqMbNMQvebw9h8YhkKyzyqefOxnT1Yv6tWNYeQtWB0oxurn6z+3pXyCEaClUV9ZE4y8rU6Ruo9JVQZUvfMA0eG4W8uViuKyqZCjA2EUVQxTyVR9XaMoHJVUVR9F18pxTXSd+2oidoklre5wBfDWMk7Tz4eeljATqMpY9QTogBDVrnkswtx6dfXzDh1oVXhsrXcotBxoqGlQpULkFSvKcTQ+UmM9urlD85ZStrqSUlqnC7JSbG8bGDKA9YVNjLFXi8NYD+WXspAXkPAeQPVnrSkJ1vpslFdaWRP0vpS1y1xY/653vSbYWaSWJ583ht5p/CNEMIS4ngkvDJKXmk35PJ2VVMhqMDQx0vsWaOVqwqx+cQyyypBeu322vNOr9fDM3LMTrLQWDXy9dR5eQy3rFHD0HsxHkrMaMYysHftqHHN5tXPp3dNcnpcVvLUPz/75BKJRKImH/bgmbnInArX6LiVqO2xlTc49HhQ1VCPCBFVkgAwConpMXu54VlUZVSrlJp4bzEpxYvOOy8MGLVstPruXW9ex/CVcFQtGruBt4dRZLljWeBKFumyN++wG2g380cA1rRa67W4FcuKVUe+oaUiqvKiE/K4wfZRy2fR0FIOgFTN+vYdQTXxHNvZg7GBsKppLxVA+ntkA88w0eRtaqCTekN6skGzm5MscSBLv37+raUWyaKdDbvrUFwxT6lOGh+rVJuso71hNLRU4Nb/aui3ZVarVIBMDgn0doypcrWGRHHMaOBh49DjQUtJAl0WKksiy7K7Qghc3HdNvU/AuZyv0prbFCmNLRWqumIiafi6akmWE5YGXoZiGloqos6j89P6dxGJRCwNvwGg641hrN9Vq/ZPLu4bUiWGZReszrZ+jA6E1d4C6+IZJjZ5Ha5xir/rZWSP7exRKo1C0zuUnr5TMTMZapFG2lfqwfGnrqjHvxb5GADg1bXnLHF92TDjitl9Sd2/tRxdZq0fQqoAACAASURBVC9ZiT2jVFfKAFBhHqkksTcXUWN1iMs7hWz00Eiy19YeQtKv5ymzJLIeb9dvy8JqdmQ2sNxvsL8vnXRVnmSY2U62Ml6zjq4Gkeieuj0BSPf0q5qMVFf5P2AYroPbupVhvehQn+Zoa9CoDW+jqHweHjhsLQYGItUs/JHwSksWqq/UE1UETKpkGs1N3FjNS5wMvNSX6/p5fVWgPzcWbnr+u3bUYOPuOpWd+oXf32xpTCJXNxWNPouBN1Y85eo5ejE0+/uK9R4ZholmznnyuvcXy9P3mRuR63fV4tD2ALqPDOPKUfcaLvI4du9V4lRvXjaU0I3VgW3dluJfTpuJdi9a17M7vU+J0x7Fq+vOgzBVfTKeSsUeQnJT9chxO3n89o1vuQKyrzKc3pcOe/IMYzAnN15jGSNgygu067QfNL1t3VBJz183VkvuW4D6e0ssxxFCuNasDxwYtkgj5Vhqm4txbGcPxkPC0IgDSkvuK/VgPBSxGFwnL9pew92tV6i9ExEAS3lmu8F2mmDcMosBq3ZenzB0ZCMUy31mz1M78n05VewkwPE9MgxjJW+NfCLGyC3sYE8mkvfrjIeEpUiXrDtzbGePMpySxpYK9J4cdRyLNOR6/LnrzWEVtpHG215d0m7Q9aJhscr56reJCL4yL6qaiqKacctwkROJtK1znGQ12ag9q1cmfTm9L9nfVpZskOoamJ8xG3iGcSdvjTwQ2xjF8vSFEBadtgwjxPOUiUjF9nUEgM/9bmnMsQBTnqncaHQq6es2cflKPUp26SZ/tCOEwHgoErWx2dsxivp7S2LqzuMpcRwn2d11CLaPWK6vXAH1uU6CRgMT+Vnq52IPnmHik9cx+XgkGqPeu+48BIxQTqy4tT5xyNi+TO2XIYZCrfa7jj12LZGJT/bnOtXnmQ56yz6JPRkrFexj+8W95zAeiuALv79Z1eJ5+Y5zKCzz4nO/W5q298Uwc4k5q66JR3T5WiNGHWwfsVRGDLaPWKpFSk/TbqyJyKzJUohaLWRTuaoQ5/dexam2fkddt1PYSOLUJi8RPXsiTOn1oz15qdVPFX1skUgEE1eFpTyy7JTlVNmSDTzDpM6cNvJAdIx6w+66qESfxscqVc13p9cBU+USqpqK0XdyDEIINJqlDPpOjuH6h5OO9c51718259D/d+uHmmhJ33jv3VdKlqYg8ryJFApLFqfyyHLvIV0rB4ZhrPCvyoaTtj6R0rfSE9/wrB8NLRU4/fxAVGkE+0Qhzydj17IZ9+YTy8zbRlNs++aiY7ONbd2WZhvhsLWkcSRibUguXzceEir2r3eMGg+JqInDfgynY8ZDGnodNvAMkznyeuN1OsRS3DgZeqdaLm7Tgdtx9A1iJdu06c1jnW/vuvOqVIMQAr/cdB4DfxzHbV8rR/NOvwqL+Eo9+LP9UwY2UTkkAPxy03mMhyKWuvZOx4yHfJ3OK3eeZ0PPMBmCjbxGotp6Hd0w6jJEnUR03brqxun/RM4nYBjRgT+OY7Q3jNM/GMDqb1dberXaKzcmIoeMRCJKhSMNstysdTqmG/omrwzRyNts6BkmM8xpdY0T061TblfHVK8pQt3a+Qmra5LFfj5Zylgiq2JKUo17p0uFk64VAcMwU8zJjNfpkohna8cpxFO3dr4qKyzDInqN91RkkE7ns7/yS4GPWurZp+oly1i6Xo5gOsf8s/3LLJ6/PC578Ey2OBwKYf/gIDaVlWFdaWm2h5N2+JflQDISRadywI0tFTjV1q9a80m5pTyO48apWdwsHrHOp/NT/3uW21KyOF3cYunT3XyNdZthZorDoRA+fvIk/vuFC/j4yZM4HAple0hpJ+O/LiL6NBG9S0RniehbmT7fTOOW2amrYtw2TnUtfqK10e3nA6BqwdQ0F+Ork7eqUE1RlRdfnbxVSRana5TtsfRHwitTPibD5AL7BwcxHokgDGA8EsH+weg+zbOdjIZriMgL4PsAPgngAwDHiGifEOKPsV85u0gmxOO2cZpMRUX7+YrKvZa68OUrfUpd4/V61QanXvogGTweD3ylHksMPtVjMkwusKmsDD6PB+ORCHweDzaVlWV7SGknoxuvRLQOQKsQ4lPm7ScAQAjxtNPzc2HjdaZw69uayvH014fDYXi9XnU7UQVMLOzHSMcxGSbbyJj84OQkOq5dwwPV1dhSX5/tYSVFNssaLAJwWbv9gXnfnMZNi5/KhGufIHQDD6Qn7s2xdGYmORwK4elLlzIeJ19XWorKggL8zeXL+M3AAP7yvfewp6sr4dfP1DinS6bVNU6uqcWSEdEWAFsAYPHixRkeTvaZjhafYfKZw6EQXgoYTs9Dfj/WlZaqDVEZRvntqlUZVb7svXIl6rbuzbspcGZ6nNMh00b+AwA3arc/AsAyRQoh9gDYAxjhmgyPJ+skk2XKMPnO4VAImzo6MG6uYl8IBPC7pibHDdF4xjOWIY4nkXyguhq/GRiw3NZf72bIpzPOmSbTRv4YgBVEtBTAhwC+COD/zPA5c57paPEZJh/ZPziICS1MOS6EMsjJbIi6GWL7/buXL0ffxESUwW8oKcHnKivRNT6OL9fVWbz4WIY8mXFmS4+fUSMvhJgkokcBvAbAC+AnQojooulzkHSVC2aY2cymsjIUEClP3kekjOBvV61K2CjqhngsEkHrxYtoXbIk6v5Hz5xBRIiYE0FDSUnUGN0MeaLjzGZYJ+MZr0KIfwHwL5k+D8Mws491paXY39QUFZOXjyVqCKUhHotEEAHw+sAADoRC2L18uTLQRISwEIjA6pG7aeV1wx3LkOvjdPLWD4dCaL14UY3NvhrItIfPZQ0YhskqyRjzWMf47apVaL14Ea8PDChj2jcxoQx0ZUEBtp09G+WRVxYUgIjgMT38yoKCmDF4eT47Tt46AHz85Ell4D2A5dwz4eGzkWcYJi9YV1qK1iVL8MbgIMaFwDwt9KMbzr1XruCB6moVqtl29iwiQsBLpGL2Tp59PGPstiIY1wz8J8rL0bpkiWXSkBPAWIY2btnIM8wsIt+LaaUDYftfIg36eCSCA6EQGkpKlGGOACAh1KasDP14iFBZUJCQikaP3c8jwvujo7h94UJLPF838ICxipBFQSLm7XTD2SwMM0uYC8W0nEgm2Wj/4CDCQkAAmBBCxfrlY7qhfikQwPujo5hHBC+mwijrSkuxe/lyEIBJIbD1zBlUFhTA5/FYnmdHhoy+UlcHAWBPdzcePXMGWxctwneWLnX0/vsmJpQR9pi30w178gwzS3ALBzhtWuYLycasN5WVwWtusAoYunt5Xeye9k8CAYTNMM1X6uos1+8PQ0OQ3RjGhcAfhoYSUtHIjdxJc4M3IgSeuXwZB26/3fE1m8rKUJjh2jls5BlmlmCX8lUWFDgmEuWSod/T1aVi4NOpB5NsstG60lL8hd+PH3d3Q8DwxOVrdJXM+6Oj+LvubsOQC4HFRUYz+6cvXXI1tPE2iGUorbKgwJLqH4YxEbupaZKRik4HNvIMM0uwGwS3RKJcMfJ7urrwl+8ZfQ1kNmmyhn46VSIf8vvxYjDoqmuXG676c+yKmt3Ll6MwEMC4EPAR4SG/P+Y57SuO9TfcgDevXo37PLkyyeRnxkaeYWYRdoPglEiUK8SrB5MI0/F0E3mN04Sprxj6JiZUeYVkk7HGIxGsLClB+9CQmiRuX7gQT1+6hPdHR2e8DAIbeYaZpcRKJMoFYtWDSYbpeLr21zipkuzPsa8YppOMJV//kN+Ph/z+KH2+lwjziABTkz8TkzIbeYaZxWR6qZ8K0mtPJSafDhLZvE01Nm5/PQDHVQKEwFfq6rC4qGjGZLBs5BmGyRhb6uunZdzTmQ+Q6OZtqhOmHu+3x/ftXn7eFChjGCb3SFXxkmnSneo/0y3+nOL7mVbQxIKNPMPMIdKheMk06a7RnkwoJt4KIpEVhtOkks2wGht5hplDpEPxkmky4XknYmT1FYSXCH9hbp7q+vZEVhgzoX1PBjbyDJNFZroWTboUL5nEaRPzkXffBZBZBZG+gggLgR91d+P/CwTw/RUrsKW+PqkVRqKe+0x8/mzkGSZLZKORRK4oXuKhb2L+SUcHxsxcgJ8EAtifoaxeuYIYjURUcbNJIfDomTNoKCmxFC4js3CZnWSM9kx9/lygjGGyhFstmljoxbpiFe6K9diW+nq8tmpVVKPqRIuAzST7zbLBkgkzqzdVnN6vXEHcX1lpeW5YyyTevXw5vESICIFtZ89aXp9sAbnpfP7TgT15hskSqfQx9RKpKol2LzBZDzGdHmW6ww+VBQWWksHzgJRj9LHe77rSUqy54Qb8qq9PnderZRL3TUwg4tBdCnCuchnrWsyU6oeNPMNkiWQ36F4KBFQoIWJ6twLRBiVZdUq61CzTmSziTQp/GBqy3P7fKyszrpvfVFaGIq2e/PdWrFCPxzLMblUuM5WAlShs5BkmiySzQfdCIDDlXcLwMCfNUrkvBALKq7cn38TzENPlUSY7WUxnUvD7fNMam0689xvL+MrH9Dr1Tq/Tq1xmMgErEVIy8kS0GUArgFsBrBFCHNceewLAl2FU2mwRQryWyrkYZq6he7myRjkAEID/atY/dzIoySbfpMujjGU8nTz2RCaFh/x+vJBkNchEar4n836djikrWL4YDEaFe5yqXGazcFyqnvxpAJ8H8GP9TiJaCeCLAD4GoB7A60T0USFEOPoQDMPYSTQ13s2gJOshpsOjdDOebh57IiuIdaWlCVeDTGZlEOv9xtv7SGRyyiWtfEpGXgjxDgAQkf2h+wH8XAgxBuACEZ0FsAbA4VTOxzBzhWRS4zNpUJLdSHUynvp7GYtE0Hrxoup1mmi3pUTOna69Bf049r0POVY5OXnNXq6HQyHLpCaf98RNNyV9/nSTqZj8IgBHtNsfmPcxDJMAyabGZyK2my7Vja4vjwB4fWAAB0KhtDfMSNfegt2I65683s3ppUAALwQC+LvubhW2ATDjuQ/xiGvkieh1AE6BsCeFEL9ye5nDffbm6fL4WwBsAYDFixfHGw7DzAlyYbmfzhoyD9fW4vfXruH40JCj/DAdpOuaxSobrMfe5T6JXefuds1mOrtZEtfICyE+MY3jfgDgRu32RwB0uRx/D4A9ALB69WrHiYBh5iLZLGoFGBp1IoInhQYX9vh2gakIytRmpL5PIfu1Tuca2q99Ijr3yoIC/GFoKKopyOFQSHn9TnkNmSZT4Zp9AP6BiHbB2HhdAeBohs7FMEyaORwKYdvZs4iYEs3dy5dPq3pjNhpm7OnqwtfPnEFECBRm0KDqHr+9+9NXTPUTYIRv9FIJM9X2T5KqhPLPADwPoBrAPxNRhxDiU0KIt4noZQB/BDAJ4OusrGGY2YM0zhEAJAT6JiZcnxsrdu/UFi+TG8OHQyE8euaMkpuOZdigSo//6UuXLJPZ4qIiy/3SwBMw45LKVNU1vwTwS5fHvgvgu6kcn2GY7JDMJmas2H2mlT/2yWX/4CDCWq0bT4rNzRONo7tdL/smrr188UzAGa8MMweYjhTSvvn49KVLqCwoQN/EhOU4TrFpPR7utLcw3U1Ie4KYnFxGzdIOD/n9KDSVPF5bSYJkORwKYVNHByaEQAFRzOqXbpNZLmygkxC5s9e5evVqcfz48fhPZBgmYZIxVm6v//jJk0oC6QGiYt3S+OqxabcNxulKM+2x9t3Ll6PlzBlVhthnvjcgWg0zHR559138qLtb3f5qXR1+eMstlvfhdJ5sqGiI6IQQYrXTY+zJM0ye85JZFgAAxoXAS4FAzGxPt/IDEfM5ThJIp9i0m4Tw/dHRpKWZTrH2vokJ/Lnfjx93d0NgqiTwEzfdlHHjKuvcy3ILvzMnTrcJLFvySYCNPMPMKNn8scfDTZViT2byAI5hGcA9Nq0bv3lE8NpkhvGIFWtPd40Y+RndvnAhCl3q5rwUCKgVxJg2cbrViM9mghQbeYaZIdy8v0yTSJGvWKoUu1Swb2LCNSzjFoNOVUq5qazMNdaezpi33RNvW7Eiag8i3jjtk1y6G5MnCxt5hpkh3Ly/TJNIka94qhT75mmssIzTRut0pZT6yidW7Z50XUenmkFO9Wce8vvxk0BA7XPIidNtkpuJ5iBusJFnmFmGzKAEEm9sHc8QxvKU3Z6vN8iwF+lyOn+yHrdTfDvZgl/JhscSlY6uKy3FfpeJ0ylbNpsKG1bXMMwMkarKRT+G3EgtTGPYJ1GDqCtp/rWvD782W+WlO7v06UuX8N8vXEAYRpOU7yxdmpSRn66KJ5f3TdxgdQ3D5ACxvL9E2T84iAnNMRvXmkynY3yJqFz0WjRhs0AXYI3jT2e1YSeVqpKHQyG0XryoNouTiYXr+wj67VjnyuVJgY08w8wgqcaPN5WVoYBIefK+BDM6UzVETvLHiC0KIOP49tXGC4HAtFYb0w1zOOn6k5kkklkBpLMJeqZgI88wWSYZAyxXA8l4yakaIrv3LqssylrrE+bfMo7/9KVLaVttxJsU4+n6PQA+UV6umpQkQjJqmGwrZxKBjTzDZJHpGOBkVwOpGqJY8sfO69ex98oVPFBdjS319QCmv9pIFqcWiVLeqYd5kjHwcvyJhonS1agkk7CRZ5gsMhOeYKqGyE3+KMsRj0ciOBAKoaGkRE1Aya42poO9reCjZiKXNPh/GBqa1nETDRPJVYScXDgmzzBMFDPhCaYq4UskwSkRrXy60a8dmZvAcpP1D0NDKhNWtuZLZjyJhIlSVUrNFGzkGSaLZEpDbY9VuxmtRPcDEklwmulQhT0TV8/ABaLb8AHpKVwGJFcPKNuwkWeYLJOK1+tkpBON86e6IZvtJB85BnnehpISS2lkvaZNZUFBzqtgMgUbeYaZpbgZabciWcmEWxJlJsIyiWIfiz4BpXvvI5F6QLkCG3mGmaW4GS57GGVwchL3dnQg7FJdMpeVIalgN/rpfK+J1APKFdjIM8wsJZaRfri2FgBw+8KF+HoC1SVz3VDZibWX4PRYJt5rLq1iYsFGnmFmKU6Gyx7CAayZqV4tK1VuWM4WEuk+FWufIdGyDbNx0otFSkaeiP4WwGcBjAM4B+DPhRCD5mNPAPgygDCAFiHEaymOlWHmPPFUM/YQDgBVXdJjZqUCiNvOL9fQjbfHJpfU4+upxN5nkywyGVL15P8dwBNCiEki+msATwD4JhGtBPBFAB8DUA/gdSL6qBAiHONYDMPEIBE1jFPi0kN+v0pMkgqUeO38Zur96GqYePXupfEWQsBDBHLoLJXKPoOTLDLeuGYDKRl5IcRvtJtHADxo/n0/gJ8LIcYAXCCiswDWADicyvkYZi4Ty0uN1VzjcChkSQzavXy5Yzu/mdx4tbcDlD1aE528ti5ahI5r1/BAdTUAWNoQ7l6+XJVasD+WDIGJibyQXaYzJv8XAP7J/HsRDKMv+cC8j2GYaZJI/1Sn5hpO3Y7s7fwy4anGim/rY5J7BgLuKwq3xKf9g4MgAJNaOQP52BuDg3EnD52H/H78vdbtyV9QkPPFxxIhrpEnotcBOIlAnxRC/Mp8zpMAJgH8TL7M4fmO3UmIaAuALQCwePHiBIbMMHOTZMsL6BuV9skhk8oQWUv+J4GAq4G1d5bSjXGsbkwALHXi7RPE3itXkpo87JD27/aFC+ELBjFmlk2YTZvUOnGNvBDiE7EeJ6KHAfwpgI+LqTZTHwC4UXvaRwB0uRx/D4A9gNEZKoExM8ycJdHyAm4VGjMdW97T1aUagssfs5OBtU9YQPzYt1OdeDlBTAqBeUR4oLoaB0IhNXlEhMCk+bx44SjZ51ZOOH0TE9i9fDkePXMGYSGw7exZVYRtNpGquubTAL4J4F4hxLD20D4A/0BEu2BsvK4AcDSVczEM44yTh29vtO3WkDqdHA6FLJp8wPCI3bxz+4QVT+/uVCe+acECPHP5MgDDY28oKbGEdbaeOQNoE04snCbL/YODiLgoeWYLqcbkvwegEMC/ExEAHBFCfFUI8TYRvQzgjzDCOF9nZQ3DpB/dEOpGPFb8PlNqEWkQJV4AX6mrm1apYad9Bvt7alqwAP/j8mWlEpo0FTGy1r3dM49noN3CYXKTOlbIZk9XV1Rd/VwhVXXN8hiPfRfAd1M5PsMw7sRL/NFVJk6JUulWi2wqK7No8r+/YsW0DZ7TPsMTN91k8dIfPXNGGXjAWDW8EAhYNmGTlVM6rS7ihWz2dHXhL997DwDwm4EBAMgpQ88ZrwwzC4nXqFpv6PHG4KBqoJGIWmS63n46SwdIr11OGNKDlkb46UuXENZWDR4An62qwq97ex1VRKmMp29iIipkA0ztIey9csXy/L1XrrCRZxhmCrtRjWdkE2lUrXvCYSHw4+5uFGj9WfUN2lhlEdLdbCOZ4+xevhxfd/Gg9VWD7C/bUFKC1/r71abr+6OjAJDyXoQ9TGQvW7x10SLlwQNQ+vxcgY08w2QRJxWMW10WSSKNqqVhGo1EIAClQFm9cCHuWLBAlca1G3R7mOSlQCBrGZ99ExMQLpuebquG365apeSbf9fdPa2uUHbs57Jfo7J58/Djj340P2PyDMOkht1g6Dpvt5CK3bN0alQtDZM0eJOmsTw+NITO69fxkN/vGPfWj+0lssS4ZzrjM16JAqdVg1ThhIVISxKT28a2U95Brhl3CRt5hskidkOm67xjbRbKUsJOyhXdMP3wllvwkN+PbWfP4tjQkMUrdjKiutf6/ugo/q67O2sZn9ON8aerTr5b6Gq2lWhmI88wWcTJYOht7JwMuG547B2JnAwTAHRcu6a04jIxyM1YSUOm17zJVlOR6cT402WEs92oPF2wkWeYLOMk23MzIPFKGLw/OurY+k8qUQjAn2vef6xzzTaPVScdRjhfOmexkWeYWUS8EgZeBwUNYI0hJ9OPdDZ5rOlmNk9yOmzkGWYWEa+EAYTAV+rqVNanNEz2xCgmMfJhkmMjzzCzDLvhcWoUoj+uJ0YdCIUyXmQr2f6rTGZhI88ws5x4YYVUWuIlS6xkqmxVxpzrsJFnmDwgVlhhJjcQY00o+mNjkQi+9t57iAB51U81F2EjzzB5Tjo3EPVGJE5eeKwJZVNZmarxDgCyLK3sp8pGPjOwkWeYOUA6NhBluEWWSiAARQ7VL90mlM7r1y215pmZgY08wzAJsX9wEGOmgQeMejhjLl2fnDZcHz1zBnpTCS+ACAAfUVKyTiY52MgzDJMQm8rK4NHCLQDgTaCtHjDVWk8iSwP7Cwqm1VSESRxPtgfAMMzsYF1pKb6/YgUKiEAwyiN8b8WKhAy0LA3sgeHBe4jw695evBgMZnrYcx725BmGSZgt9fUxa+u4kUuFz+YabOQZhkmK6W7i5lLhs7kEG3mGYWaUfKkJM1tIycgT0XcA3A9jk7wHwJeEEF1ERACeA/AZAMPm/b9PdbAMw+QH+VATZraQ6sbr3wohGoUQTQD+F4Bvm/f/FwArzH9bAPwwxfMwDMMw0yAlIy+EuKrdLAGUhPZ+AC8JgyMAyoioLpVzMQzDMMmTckyeiL4L4CEAIQB/Yt69CMBl7WkfmPd1p3o+hmEYJnHievJE9DoRnXb4dz8ACCGeFELcCOBnAB6VL3M4lGM+MxFtIaLjRHT8ypUr030fDMMwjAMk0lRLgohuAvDPQojbiOjHAPYLIf7RfOxdAJuEEDE9eSK6AuCSdlcVgN60DHDm4DHPDDzmmYHHPDOkOuabhBDVTg+kqq5ZIYQ4Y968D8B/mH/vA/AoEf0cQDOAUDwDDwD2QRLRcSHE6lTGONPwmGcGHvPMwGOeGTI55lRj8n9FRLfAkFBeAvBV8/5/gSGfPAtDQvnnKZ6HYRiGmQYpGXkhxAMu9wsAX0/l2AzDMEzq5HqBsj3ZHsA04DHPDDzmmYHHPDNkbMxp23hlGIZhco9c9+QZhmGYFMhpI09E/42IBBFVmbeJiNqI6CwRnSKiO7I9RgkRfcccUwcR/YaI6s37c3LMRPS3RPQf5ph+SURl2mNPmON9l4g+lc1x6hDRZiJ6m4giRLTa9lhOjhkAiOjT5rjOEtG3sj0eN4joJ0TUQ0SntfsqiOjfieiM+X95NseoQ0Q3EtHviOgd83vxmHl/Lo+5iIiOEtFJc8w7zfuXElG7OeZ/IiJf2k4qhMjJfwBuBPAaDNVOlXnfZwD8K4xkq7UA2rM9Tm28N2h/twD4US6PGcB/BjDP/PuvAfy1+fdKACcBFAJYCuAcAG+2x2uO7VYAtwDYD2C1dn8uj9lrjmcZAJ85zpXZHpfLWO8BcAeA09p9fwPgW+bf35Lfk1z4B6AOwB3m3wsBvGd+F3J5zARggfl3AYB20y68DOCL5v0/AvBIus6Zy578swD+H1gzZXO2Jo6YZXV8hBC/EUJMmjePAPiI+ff9AH4uhBgTQlyAIYNdk40x2hFCvCOEeNfhoZwdM4xxnBVCnBdCjAP4OYzx5hxCiDcB9Nvuvh/Ai+bfLwL43IwOKgZCiG5hVrcVQgwBeAdG+ZRcHrMQQlwzbxaY/wSA/wTgVfP+tI45J408Ed0H4EMhxEnbQ241cXICIvouEV0G8H9hqiJnTo/Z5C9grDaA2TFeO7k85lweWyLUCjOR0fy/JsvjcYSIlgC4HYZnnNNjJiIvEXXAKM/+7zBWeoOa05XW70jWmoYQ0esAnFq0Pwng/4URToh6mcN9MyYPijVmIcSvhBBPAniSiJ6AUcdnB7I45njjNZ/zJIBJGLWHgBy/xm4vc7gvV2RjuTy2vICIFgDYC2CbEOKq0c4idxFChAE0mftgv4QRhox6WrrOlzUjL4T4hNP9RNQAI6560vywPgLg90S0BsYMd6P29I8A6MrwUBVuY3bgHwD8Mwwjn7UxxxsvET0M4E8BfFyYwUDMnmusk9UxxyGXx5YIQSKqE0J0m2HGnmwPSIeICmAY+J8JIX5h3p3TY5YIIQaJaD+MmHwZEc0zVMy1egAAAW1JREFUvfm0fkdyLlwjhOgUQtQIIZYIIZbA+JHcIYQIwKiJ85CpWFmLBGvizAREtEK7aa/jk3NjJqJPA/gmgPuEEMPaQ/sAfJGIColoKYzGL0ezMcYkyOUxHwOwwlRP+AB8EcZ4Zwv7ADxs/v0wALfV1IxDhhf49wDeEULs0h7K5TFXSyUbERUD+ASMvYTfAXjQfFp6x5zt3eYEdqMvYkpdQwC+DyOG1QlNYZHtfzC8idMATgH4NYBFuTxmGJuTlwF0mP9+pD32pDnedwH8l2yPVRvXn8GY9McABAG8lutjNsf2GRjKj3Mwwk5ZH5PLOP8RRs+HCfM6fxlAJYDfAjhj/l+R7XFq490AI6xxSvsefybHx9wI4A/mmE8D+LZ5/zIYjslZAK8AKEzXOTnjlWEYJo/JuXANwzAMkz7YyDMMw+QxbOQZhmHyGDbyDMMweQwbeYZhmDyGjTzDMEwew0aeYRgmj2EjzzAMk8f8/7DHGWVdoldiAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cluster_and_visualize(male + female, extract_vectors(male + female, gn_limit, gn_w2i_limit), \n",
    "                          'GN-GloVe', random_state, tsne_random_state, y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 250,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 0.9975\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cluster_and_visualize(male + female, extract_vectors(male + female, gn_a_limit, gn_a_w2i_limit), \n",
    "                          'GN-GloVe_a', random_state, tsne_random_state,  y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 251,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 1.0\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cluster_and_visualize(male + female, extract_vectors(male + female, gp_limit, gp_w2i_limit), \n",
    "                          'GP-GloVe', random_state, tsne_random_state,  y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 252,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 1.0\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAADCCAYAAABHTgk8AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjMsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+AADFEAAAgAElEQVR4nO29eZRV5Z3v/fmdoiZAax4ARUZzoxZUIoIIKt3JvbH77Wg6Yt706nVjbrLajokUYNbqJG9WC3a/vZJOXwEhnYHcDLpu305UYos9ZPINERALISkGYysIiFp1qqhRhpqo87x/7P3sevY++9Q5VXVO1anD81mLVXXOnp69ge/z27/pEaUUFovFYsldIpM9AIvFYrFkFiv0FovFkuNYobdYLJYcxwq9xWKx5DhW6C0WiyXHsUJvsVgsOc60yR6ASWVlpZo3b95kD8NisVimFIcPH25XSlUl2j5uoReRIuBFoNA93zNKqU0iMh/4MVAO/Bb470qpgZHONW/ePA4dOjTeIVksFssVhYi8NdL2dLhu+oE/VEotBeqBu0TkVuDvga1KqcVAF/DZNFzLYrFYLKNk3EKvHC64H/PdPwr4Q+AZ9/sngI+N91oWi8ViGT1pCcaKSJ6INAFtwC+BN4FupdRld5d3gDkJjn1ARA6JyKFz586lYzgWi8ViMUiL0CulhpRS9cA1wHLg/WG7JTh2p1JqmVJqWVVVwliCxWKxWMZIWtMrlVLdwB7gVqBURHSw9xqgOZ3XslgsFoCdzc185MgRdjZbiUnEuIVeRKpEpNT9vRj4MPAa8Gtgrbvb/cBz472WxWKxmOxsbuYv33iDX3R18ZdvvOGJ/YGeHr721lsc6OmZ5BFmB+nIo58FPCEieTgTx1NKqX8Vkd8DPxaR/xf4HfD9NFzLYrFYPHYF4nq7zp2jbsYMPnTkCAOxGAWRCC8sXcrKkpJJGmF2MG6hV0odBT4Q8v0pHH+9xWLJIQ709LCnu5s1paWTLqD3VlXxi64u3+c93d0MxGIMAQOxGHu6uyd9nJNNVlXGWiyW7OZAT09WWcsPzJ4NOJb8vVVVPDB7Ngd6eiiIRLwxriktHfV5xzKZZdMEGMQKvcViSZlstJYfmD3bE3yAlSUlvLB06ZhFdyyTWbZNgEFsUzOLxZIya0pLKYhEyIMxW8vJSEcgdWVJCV+57roxie2e7m763cms353MUjkmOAFmE9ait1gsKTNeazkZk2kZa9dL9+XLxNzvYkBFfn7cPsF71xPgeNxFmcQKvcViGRUrS0oyJr6puIYy4Qs3JxgRQXAqPCNAx+Cgb5/+WIyICP+4eLHnMsr0BDherNBbLJYJZ2dzsy+AqglaxhX5+Xztrbc88cyUxf9kNEpfLOaIu1JMEyGmlM861y6dGBBTiodOnKBuxgzv+mETYKL7NJmIIK4VeovF4jERoqOLnAAvNfKB2bM9UVw3Zw6l06ZRkZ/PhpMnfaKeiWDwgZ4efhiNej1apomwY/FiOgYHfRPM2b4+xDhuSKkRr5/oPoPXnghXlRV6iyXHSSbeenuYsGZCdMKKnACfKH73+uvpGByME/VM+ML3dHdzWTkyL8Bnamt9gmyKMQxnsBQmeOPQ4wy7z6DQT1QWkxV6iyXHMMUGGNFiNEUsIsKQUsTIrOiEFTmFieLmefPiRD0TvvDg5PGp2lrfdlOMNdNEWDdnjm9i3LZoke/zujlz4u4z2bUzFcS1Qm+x5BBBV8D9NTU+i/HJaNQnkqaIKaWIiCAB33S6CStyAuJEMZGopzsYnGzy0GKsffjgPKumCxd8z3bXuXO+z6XTpvHd66/37rNuxgyf9Z/KtdOFFXqLJYcIugIAz2LME+GH0SiXXSF/YelSn0U5TYQ/Ki+ntqCAT9XWZjRzJFjklEj8xyvqqcYcRrqOFuMno1H+V0sLQ0CeCPdWVbG3p8ebVM3PeSKc7etjTWmpV62b6M0qk1lMGiv0FksOEXQFfOCqq3zbd7a0EGO4EOgr113HC0uX8o233+b59nZ2d3RQGOK+mAiC4j9ekgU6RxN41tt/GI0ypBQC1M2YwbZFi9h17hz1M2fSMTjItkWL+N358/wwGuV7LS080drqWew6Y6c/g26xRFiht1hyCNMVEAyurpszx1cI1H3ZWQDu2MWL7G5v97alQ4hSSStMlbFmAo0U6NzZ3Mzn33jDsc6Bb11/fdJx6qCtAi4rxZPRKE+0ttIfi/GLri4iOAHa+2tquKyU77oV+fkJi7AmAiv0FkuWkaqwHejp4cloFMDnatGugK+99ZZP6JouXPAKgQC2vvMOC4uL+cKJE54IgZN5cravjwM9PXGZJKkIbSpphakynvTDRIHOAz09nsgDDAGff+MNX058KucD57maAh50l+nr7unuJuLuYxZhTRRW6C2WLCJM2IA4oT3Q08OapiYG3LTAH0aj/Lq+PrQsv9+t9qyfOZP/z0glvKwU329pIaaGV/kUICLiuR2CmSSpCG0qaYWpMt70w/tragD/RLinu9s3sYEjwMnOHQycAp5FrwVcZ+18qrY27u+scBJbJIxb6EXkWuBJoBbnee1USj0uIuXAT4B5wBngE0qprkTnsVgs8cKm3QNBod3T3c2gIdADIcU7K0tK2LZoEQ+dOMGQUux4910evuYaHnv7bSfLBvjthQtME+Gym3HzJxUVPN/enjCTJBWhDUufHCvJKmUTYU6YeTJc5rSypIQ1paXki3iTJDhCaL7FJCIYODXdZGaBld7XPM7cVzc9myg/fTos+svAF5VSvxWRq4DDIvJL4NPAC0qpr4vIl4EvA19Kw/UslpwlkXsgKLRBsSoQCbUSOwYHiRm58aXTpvHRykr+pb0dcKz6j1VWsvyqq7zjf97ZGZpJkqolmiiDZiyYGS/RgQEaTpzwZQ0lEkpzwhxSiu8agdGVJSXsqa93zjk4SOfAAPvfe4+dgX2SMVqXlt5nMpq2pWOFqRagxf39vIi8BswB7gHWuLs9gbNouBV6i2UEErkHgkLrE6uBAWoLCkLPF+anPtvX59unNj+fr1x3nfc5mNddN2PGqIOh6c6geaK11ZfHnuztIpj7rgLH6D8Henq4s6nJ89enGohONXYQnAwmq59/Wn30IjIPZ1nBRqDGnQRQSrWISHU6r2Wx5CqJ3AOJhPY/Oju5rFSoNZqoIOcH0SiDSpEvEpdKGby+6ds2P6dCOnrnaHHUIi8k74VvvgmYtQPBY/Z0dzNkuHAiCd6MEo0pWZfN4GQwWe2M0yb0IjIT2AVsUEq9JyLJDtHHPQA8ADB37tx0DcdiyRnCCmq0iKRi5YYJ9576+pQFeKyZL0E/+WfcIGUiyzcsgwj8byWpnMf0mYcFRs3JZ01pKYVuwDpPhG8uXpzw3oLHJYsdhE0Gum5hotsZp0XoRSQfR+T/SSn1U/frVhGZ5Vrzs4C2sGOVUjuBnQDLli1TYftYLBY/Y7FyTUZTjWm28B2NuyGZn1yTLIMo0VtJMFff7Bevs2AK3YlJu6bCJi1t+YNTBBVGouOC9QoR4ANXXcVnZ80acTLQKZf6/jJNOrJuBPg+8JpSaouxaTdwP/B19+dz472WxWJxGI2VC2N3oQRb+Oal6NoApygo4vZ1D/OTa1LNIDI/h+Xq626Xwbx281xhVvaa0lIvDpIoGJvIOg/WKwwBB8+f5+D583z3+utDJ4M8d2GTVILK6SIdFv0q4L8Dx0Skyf3u/8ER+KdE5LPAWeC+NFzLYrEwumZY2mLWPvk9gXz7kQhr4Ztq0dQX3LTOCHiCH/bWkSiDaKTJaaRul9qiF5wuk+b1grUFOtUxmb/dPC7iHhfc1hvzZ+fr+oHgZKDrFkb7hjQe0pF1sw9I5JD/0HjPb7FcqSSzwlN1vzwZjXoiOqAU3zh7lmfr6lIagylweSJxvXMSjfshNw0SHEH7bG0tc4uKQu/FzCACvODwSHGBsFz9sOZjQV9wsLZgw8mTbFu0KGmAVB/3BeM4wMudf2HpUr5x9iz/0tHhHVM/c6bPVWO+gZkW/UQEZG1lrMWShWRy5aHnOzqSFgZpwgQuWauAsEyWZN0wg5NWsH2DafUe6OmhY3CQv7r2WpouXIjrdrmnu9tzFYWtAhWsLegYHEzp7ahjcBDlHtcfi/HQiRPeW8oLS5fybF2dFzeonzmTHe++G+rTN5cmnKiArBV6iyULSVe+9YGeHgCvzwo4Ajia85kCl8pY1pSWMk2EQddt848jZLKMdI5EfWrCWkSYGS/JUhjDtqfydmQeJwkWadH1A2ETlfbpayYq4was0FssWUk68q3NLBTwL4E3mvONZSzalzvNTbNOpW2BSaIYRKIWEWZ65AOzZ49ooY8mvpHouO7Ll9nyzjvgWvTB9MrJypdPhBV6iyULGasYmZg90MFpx/sXs2aNelGR0Y7FbOc7qBRfOHEC5QritkWL4nrCBDFjE2bFLoS3iND3GHOvpV1Lo3EVpYrZxiCmFHkhSwpqN9tk5Msnwgq9xTIJBNd1TeS7HY9ArCkt9bJdYDgwOZb87dFUx5pibK5DG+bXTqWadKRKX4DvR6PePcZCfPLpRr9VxAAJWVIw2GYhG7BCb7FMMKaYTRPxgobpzq9eWVLCPy5e7GWYTAtZSjDV848mOGyKsZk/nsivbZJKbCIooN9cvNjpqa/UqN1SY6kvCL5VjKXx20Rjhd5imWBMMTOtbb1EXTrzqx+YPdtrSna2r4/vuWmH2r+tRe7YxYsjdpscbXDYFGN9/eCKV2GCOBbftnmPoxHssWY2hbllxnL9icQKvcUywQQX5I4phV5vSBFe6DMetOge6OnxKkCnifADd/3TCHjXT7Qi1HiCi2GiP1JtwFgDpaMV2NFOXkHrf6TK3WzDCr3FMsEExezJaJTvtrR4PnRt3WfyuqZ1PxTYL2xFqHQFF1MRxIkSzdFMXpmsa5gIrNBbLJNAUMyeaG31ldBfxqloTbeYhFn3pkUPiVeEynarFUbncx/N5DVZfeTThRV6i2WS0YKz4eRJDp4/P6HXTNVHn4wvvfkmPz13jo9XVfH3CxemfbypCPhYrO5UJ69sy4sfLVboLZYsQLca+AO3XW9ByIIgmbimFjld1TkWvvTmm3zj7bcBvJ/pFPtUBTyTVne25cWPFiv0FkuWsLKkhF+PYkGQbOGngU6SPz13Lq1Cn6qAZ9rqngquq0RYobdYsoipKCYfr6ryLHn9OZ2kKuBT3erOJFboLRZLHKMJamrrPR0++rDrjkbAp+JEORGIykAa11hZtmyZOnTo0GQPw2KZNNKxmHY6xjAZqYRTPYVxMhGRw0qpZYm2RxJtGOVFfiAibSJy3PiuXER+KSIn3J9l6biWxZKraKH769On+dCRI16L4YnG9In3x2JsPnNmQsYS5ou3pIe0CD3wI+CuwHdfBl5QSi0GXnA/WyyWBIxX6A709PC1t94atyhrn7juYf+rrq6MTzwHeno429dHngh5jG6h89FeJx3PaKqRFh+9UupFEZkX+PoeYI37+xPAHuBL6biexZKLrCktJc9tiSDA2b4+T5BMd06Yeyedbo9gXn+qC46EMdr892kiY2qlnOpYdH/+iAgPX3MNpdOmXRGB20wGY2uUUi0ASqkWEakO20lEHgAeAJg7d24Gh2OxZD96wY7LwM6WFn4Qjfo6Wm5btCi093kmcsiPXLjg/Z43ht47Y8l/RynmFhVlRHjN/vwxpfjG228TwVmIJdfjAely3YwZpdROpdQypdSyqjSnZVksUwlzwQ5w3CaDSjGglCfgu86dC/Wfa3dLutweeizgTD6fSWJhh7lEUnVFpXvsidD9+U3Mt5VcJpNC3yoiswDcn20ZvJbFMuUxfePg/OfMF6HA8FvfW1UV6j8HeGHpUv52/vy0WKem+BZFIiNW6SYKIicS8J3NzXzkyBF2NjcDw66isYx9ND533Z8/X8T3jKdiS4PRkknXzW7gfuDr7s/nMngti2XKE1ywQy+5B34ffd2MGWw+c4ZfdXX5LNLg4tPpGksy//rmM2c8l0hwhaXgOXY2N/OXb7wB+FsiJ8p/H8nHP5a4hNm73nzGuey2gTQJvYj8M07gtVJE3gE24Qj8UyLyWeAscF86rmWx5DKJBC/Y+3zzvHkZX9UoWfGRGdyMMWwdV+Tn8+DrrwPwqdpa37qvuwLtEsJaIgfPn0jIxxqXuBKLqtKVdfNnCTZ9KB3nt1gsftJV7r+zuXnMXSvNtVMjwIfLyri3qoqGEyfod/37P4hG2VNf743v3qoqz5IHqJ85M+n5Ewn5VO8oOZHYFggWS5ZzoKeHJ6NRAF/aYdAyHW1VbSI3Stj1w84bFNrN8+Y54mxU2w8GFut+YPZs3uzt5X++/TYK2PHuu3yssjLULaPz6nEzjiry8/naW2/5VnhKNNllQ4VxNmGF3mLJYg709PAHTU0JLWRzv9H6q1Nxo4x03mBMQf8sEPHGmx+Sllk6bRoCCfPzTZeQAB+tqOCPKipC00rD3DC2lUI8VugtliwmmYXs22+U659Oj/iT7syVpbRFfLavb8Tz6t9NYd2+eDG/cxdQCSt8SuRyMa+p/f4A/9rZSW1BQcr3N9VXg8oEVugtlixmTWlpUgvZ22+U65/2u8sI/pfp01l/zTWeNR+sVNWpnIkWLA8Ka8fgIN9+3/sSXj9RNs4XTpwgphTTRDCz3WPuvY80OZguGuu7j8cKvcWSxejFSMJ89MH9RrP+qWkxn+jtpW7GjFArXrntGAAS9bkNCmvQl57ovsz2DQ+dOOEVaF1WirsrKvjXzk5iSpHvFjltW7SIjsFBz0107OLFhO6c8Qaqc83Hb4XeYslyUk0HHM36pxG3pw7AkFI8GY16i4XniTDNDYJGRBhyq3WHEriNgr76MPEdiT3d3QwZ7qmICH81dy5/NXcuT0aj/CAa5XstLRS0tvpaQOixaV//k9GoT5zH2pdnLPeQ7Viht1iuMHSF6EMnTjCkFIWur97sN/MXs2Yxt6goTvRGWt1pZUkJX3vrraT+8aC1vKa0lMJIhH53kvnm4sXeMXoSCGsBodyJSFx3zw+iUYbcDJ2xVNhqd5W4k+B4mrllG1boLZYpzljcDGaFqBZvbdEXuC0P9LnM/ZKdP5l/PFFGTCJXS/B891ZV+QrFtDvnbF8f32tpieurk+q4zThDRCny3AkkV3z8VugtlimKzq//YTTqdbccjSUbdG8kEtvRuEGS+ccTZcSMVBEcPF/djBlezKJuxgyvdbM5UVXk548qxTI4oegJxProLRbLpKEt475YzAuSjtfNkK7WACOdJx0ZMccuXuT7rpvmidbW0LeCRJ0zE01A6ao0zlas0FssUwTTRaOFTIu8MDW6MI5WUIOunm2LFvEFI0Onf4S3gmAmUDILP5d74Fiht1imAGGCp4UsT4TP1NamfVWm8fTBSQdhnTG/39LiiTzEL4hiTobbFi3yxt8xOHhFF1FZobdYpgBhRUmZdDWMtw9OIlJtTxDWGTNPhN8Zq15FgP+rvDz03NNEvJTQvT09volxKrz5pBsr9BbLFCDMt51JV8N4++AkItX2BGGdMRcUFfG9lhbAcVUJ8HxHBz/v6vImPX1uXSOgmJiJMduxQm+xTAEmOlgYbCd8b8gyn2PpKZNqMDasMyYMp4AGi6X0c9HHmBb9REyM2U7GhV5E7gIeB/KA/6WU+nqmr2mx5CITKVTaeh/JRz+WDJqwjpf6+0T7mRObeWzDiRMMuMVSeh/zGEg9jz7XEaUSdbBIw8lF8oA3gP8KvAO8AvyZUur3YfsvW7ZMHTp0KGPjsVgs6WWsPWHG20r4QE8Pa5qaGHR74YS1br6SEJHDSqllibZncnFwgOXASaXUKaXUAPBj4J4MX9NisUwQK0tKxrRWbaI899EcH+zBY0lMpoV+DvC28fkd9zuLxXIFo90+eYwt/3+8x19pZNpHLyHf+XxFIvIA8ADA3LlzMzwci8WSDYw3uJzrlazpJtNC/w5wrfH5GqDZ3EEptRPYCY6PPsPjsVgsWcJ4g8tXchbNaMm06+YVYLGIzBeRAuCTwO4MX9NisVgsBhm16JVSl0XkIeDnOOmVP1BKvZrJa1osFovFT8bz6JVS/w78e6avY7FYLJZwMu26sVgsFsskY4XeYrFYchwr9JYrjmA1eLLq8Ezvb7FkGiv0lpwgVXE9uLmVfRuj3nalFPs2Rjm4uXVS9rdYJgIr9JYpT6i4bmjxiatSCqUU/d0xjj7e4e2/b2OUo4930N8dC50sMrm/xTJR2DbFVyhKKUQk4eepgimuAKu31rJr5SlaG3tZ0lDuieu+jVEKSyOs3loLwNHHO7xjlqyvYPXW2tD7X7WlZsT99flFBBFh1ZYalFKh+wefcSwWIxKxtpYl89h/ZVcAQUuycVPuuBdEhNVba1myvoKjj3fwrcirtDb2AsO9NkyrGvDEXhP8DM5bwq6Vp9i3MeqJvaagxBH1g5tbeebWN9m3ocV7Y9i3MUpr46W48z+75jRP33yKWMwZQywW4+mbT/HsmlPpeAwWy4hYiz4HMS3Hg5tb6e8aYvW2Wd62M7vP097UB0qxetssTwiXrK+Ykpa9FnttQQPUNZRzbHsnx7Z3An6r+qe3nfYdv29DCwooKstj+WbHIu/vGqK1sZfWxl6aX7zo2//0c+dZ9kgVvZ2XaTvYR9vBPsCZWPT1TPZuaKbz1T76Oxxxv+/wAp6++RTtTX1U1hdZy96ScazQ5xgHN7fS3x3zrNT+riGObu/kzV3vseDeqwFcgSnk6PZOjgaEcKqIvDkhKaXYu6EluIPv46otNRzc3Mbp3e/R0dQPwE3ryojuveQ9g7qGck90V2+bhVKKYzu6vP01HUf6eenhVl/HvqMhAl/XUI4Y2worIrQ39fHtPGc5hsr6Iu47vMCKvCXj2H9hUxzTLaMtUR0MhGH3xcV3L3sWbl1DObW3T/edJ9tEXrs4wj6bwVct8se2d1K9opjPx26kenkRx3Z0+Y5/6uY3Ofuz83Q09VNcm8dN68o4vqOLdlfEq5YXUVgaYf/DrcRiMectwX0LMqlbVwY44n1sRxdLGspD99FvFFrkq1cU8z9a3+fbz4q8ZaKw/8omANMXrn/qP8F9RkMw2wQcYa9ZUez5q7Wwmxzb3snxgBBqP3M28OyaUwn92cHMFoA21ydfu6KYWCzG0IBzH3nFzvmKKvPoaOqn59QAAL3Robj7v/el+Qz0OEHUp28+ReOmKE/f/Gbc2IJPKBb2zFxXksnH98/jmWV+l5F5jxZLJrGumwyjXSkFJREGemKs2lLDrttOA4qaFdMpKsvjlk3VXlbI8s01Sc8JjviZ2SarttSw/+FWT9h1QBLCFwXQ+NwLrkBNpmUfi8UY6InR3tQX6s9WSoVnzjSUs3rbLESE+XdfxaWWy/S2DgHQ1+787G+PUbW8iHOuT93kmWWnWXtoPs2/uUh7U58Tw3C5aV0ZERGOhkyQ+nP18iJqbp3uvTUF/fo/qn2DvvYhz12j70nfYyQSyZlMKEv2YYU+g5jWZ2V9Ee1Nfby75wIdRxx3QVtjH3UN5ezb0MLR7Z2hwdCwz6882kZ/dyw09U8Lt8lRw6oPBgs9YRehoCQy6cISiUR8QpjInx0MvmqRB1i+uYb+rqE49w1ArN9vgdetK6N57yXam/r4zrTX4va/8aFS7nh8Nkop3n3xouevr1haSF6R0NboTAiX+2IUlAh168o4+bQTB9BxD/33W1SZx9pD8xER7x4LSiJEIhHPIFi1pcYT/X0boxSURFjxaGqTv8WSCOu6ySBm6p+2ELXIa7Qft66h3GdNK6VCC4H2bmjh7M8ucPTxDvY/3BqX+gd4k8bnYzdSvcLxXzT/5iLHtndSsbTQt+9TH3wTpRS3PVbNQE/MS7GcTDeOFnsTU+S1CJoEXViJJqyOI/1UryimrqGcJQ3lHNvRxexAvMKkdV8vBze3EYlEmH/31dQ1lFGxtJCOI/2eyBdVR+g8OsDge7Bqay03/EWZJ/IHN7cSU4olDeXc+GApkUiEfRujHPqbc9x3eAF/umeB7+3s6ZtPMTQ05GVCndl9nsZN0YTjs1hSwVr0GSYs9S8R2nrXlZ3Rly95qXurt9ayb2PUc81oP3zwvLpQSE8aaw8sYO+GFtqber23ipvWlSEiHNveSceRfp5ZeQo1oGh3rdBYLMb+h1vjrMmJciVon7yJdnGIiC8dVD+XoAvr6PZOKuoLfRkzxTV59LYOUbOi2Ffw1BZw5RRV5vHp6PU8s+y0O0Gf55ZNVax4tMbzqes3DYC+thh1DeUUlER4dtUZqt3zK6U4vdsJAFctL+K+lxf6xq5z8fu7Y9z2WLXnNtJvFvrva/adM6wbxzIurNBnGKUUz6xMXhRzbHsnJ3/Sw6eb3+cTqurlRQmrLM2UvgeHbnCOe7yDGteKB2eiud11azRuamX2nTN8gcJj2zs9n3VlfdGwULrupls2VY3oSki3AGmR1z75MH92YWnElw6q76ew1HGDFJZGWNJQTvRlf+GSTBPKlxRQWBrxRL557yXHzdJQTn6J8Oq3u+lrH/LEtqgyj+s+OsN7Bvs3Rjm5672QkSvO/uw8bQf7vPiIgDfRnDvYx7cizpo7S9ZXeG9iZpxl7aH5PvdRe1PflEt7tWQnMp5XdBG5D9gMvB9YrpQ6ZGz7CvBZYAhoUEr9PNn5li1bpg4dOpRstymDmfqnKavLp+vYoPf5xodKadl7ic4jA75jtTWn0/Q0n4/dCOCzYmFYPPY/3DpiUDeYf67FJ4h5/duNoqqKpYV84rcLfeI/miByKjy75hQDPTHPXaPFv6Akwp/uWRB3H2H3pZ+7DtLuXd/s+ex14FQHoSvrC6m9fTrF5dO4+a8r43z1QV97IuoaykARFxuoXl7ke2v43OX389IX2ygsjTiB+BHO++DQDTYF05IUETmslFqWaPt4LfrjwMeB7wYuegPO+rA3ArOBX4nI9UqpoXFeb0ohIhSV5XlZMG2NvZ7I500Xhi4pXv1mNwAVSwvoMMTec7EEzrl3Q4snUKa4m66LkYTBFMOgn9vkvsML2HXbaV91qfZN77rtNGsPLMhYRa32W+v70D57876C1zI/6+duWsO3P55vTJUAACAASURBVD4bEfHcYVp4lzSUE1OK4zu6qGso56WH/W0gKuv9b1TVK4q53DtE51H/xOxemVVba+KEXqd7anQGjn5uoSmaLmZWjsUyVsb1r0cp9ZpS6vWQTfcAP1ZK9SulTgMngeXjudZUZfnmGm7fNou1B/zBxb9477/4PucVxf9VtLx40RdYXbK+gmPbO4m6fvjbHqt2qjjdgG/+1aQkCGZXxSXrK3hw6AYq64t8++zbGPW5gABm3zkDcPLWvxV51ecnT7drIXgfoxW65ZtrfOPSBVBrX17o20+nTJpppubzNtMsAdYeWMCCj10dV5vg+OiFZ5bFu+k6mvqpqB8Ogve1D1FUmcdtjzlptcGUzcr6Qj53+f3eW9X+h1uzpsbBMjXJlJkwB3jb+PyO+10cIvKAiBwSkUPnzp3L0HAmn6D1HCye0RkcJh1HBqhYWsCqLTW+DJ7r7prJuaY+nll22qvivO2xat56/mJKTbJExPNz6zcC7RNf9kiVN6E0/8afCx5WfJXN/uOwcSV6i7l92ywKSvOorC8i/2q8TpRhE+CyR6oIlk4N9xAabq+gq2gBalf5z6PjAPptqXp5ETf/dQUV9YW0N/Xz7KozrD00nyXrK7yYgsUyVpK6bkTkV0B8ez/4qlLquUSHhXwXapIopXYCO8Hx0Scbz1QjaD2v2lITF2zctfJUXOaHZtYdM/xWqc7meO68L0CpM0RSbZKlm3cFRV/7xHUGSDBGEBT/fRujWS32Gt/fg+Gu0by4vpmIiGfB37JpiJeMCXDe3TMZcAOnr36vk6FLijq3DfLxHV3euYprIiz8RAlFZdPo7x7yXHKnnrkw4viG+hW3bKrm9HPnnfGC97ZmsYyXpEKvlPrwGM77DnCt8fkaoHkM55nymEKqBXHe3TMBmHf3VUQiEda+vJBnbn0TBb6qzbqGcorK8uL8z2bBzXiaZJkFRqaPPRKJMO/umcy6Y3rcjN1xpD8uQAvZbdmD8ffQUI7CqWjVTc3am/o9oa6sL6S9qY9nV51BKeV9nn3HdIZiMS9FEwCl4p7Pp965nkgk4lUp37SuFBBf/USkEGJGOYUUOM/VzPT5+P55w6m2GQh4W64sMuW62Q18UkQKRWQ+sBg4mKFrZT1Bf/GKR2u57/ACL01RRLj3wAJm3eov3BHglk3VoedMVlQ0WoIivXxzTZzPunpFkTcuwHMlTRXXwvLNNazeNssL1N7x+GzuO+z32d93eCFLdPD8YJ9TW+BODq9+s5uFn7jac8kc29EVF3jddcsZwH02DeUc39EdVyQX839EBeK6fe1DvPTFNrtClSVtjCvrRkT+FNgBVAH/JiJNSqmPKKVeFZGngN8Dl4EvXGkZN0GCQmgKslLKy52PKwJK0H9mpKKidGRoiAiFgcyVtQcWsm9DC4XGW0a2W/JBRMR7gwHYH8iy2f9wK6u21vrSHYOtnCE+hfJzl9/vuc/038NYZTmY6WNz6S3jZVx59OlmqufRp9KnBvwpjvrzwc2t9HUNecVNOhdcL4ZhMlJRUbp7nOdqo61g7MScXHW2SxBdwxCsjQAnTfO2LTVGNW08dQ3lvLvnQoLUTIewa38+dmNOPHNL5sh0Hr3FxVzwI8y3enBzK2/97AK1K4q9Pud7N7TQ1tjL3Lscn33wv3Ki/9qRSISCkohP1INNstLFSPnqQUaayLKNsNjJqi01XhBau2tMQX9xfTPCsDWv16Q9tqPLs/qD1a3VK4qoWTFcnFW9oojyJQU+sS+siHDT58sYfM9pUFdUmed13ISpE/C2ZC9W6NNA2ALVvgyPWIy+riHa3KIp8C87V728CHHb4GpXzb6NUV9HS/AL5sd+PR+l1IhFRRNJsoksGwOJ4UHoq5h9x3Tv70fn10cbezm+o4vqFcVULy+i9tbpwwuTiNDW2EtBaYSXvtjmu0bNiulxHUL7u4foPDo8gVz/5yWseNTJpvImmpA+PlbsLWPFCv0IpOq2MPutmL7VmhXFKIb7zehUPNP/q0v03ROF+mZ1W+JEbwuayRJ5pdSIE1nNiuKsdfkEx7Ti0RqvFbTpkwe8Z64D5PpY/XerK5T1cdrFo/99mGsGBNsqiDiplPPuvsrrRxTs45ONz88yNbBCn4BkrpggYV0qq1cUe2J3+7ZZREL+o5p91ON6rOt1XxO9LWTJYt56IgN8y+eBf0GQqYIZsDXHncii1imvQVeQjrdokQ7bZ/W2WWDsoyeaYO3EVHp+luzDBmNDGClQlygDwjxGE1aYE8S06MMalWmxT7RtsgUgWZO0Ky2QmMpbYK4GuC2Thw3GjoFErphURF5nTVTWF3mdEYOYgb6j2zu9/+ih6ZU4lmSYtT9WcUiX0JhvPQB718fXxO3b0DLlLPrxkErwejQBboslHVyxQp9M7MJcMSO9ugd7x+jj2pv81TGV9YWsMs7T1thLUbnz15CoxzrE92gZaybGaF1SiTAD0EopcLNPACrqC5l9+3RfNsqVJPYWS7ZxRQp9KmIX1sZ374YWz++q9wlrIxCcILyWAW7gbf/DrazeWuv5tcPOB8S5bXwBPMPaN88xEiNmB43S3x9869FU1BfyicMLvSyTtsZeX4GVxWKZeK44oU9F7CBeXJ9ZecoXWNX7mJODOWmYeC0DAoG3IPq74ERUUCJU1hdRUBLhlUfbUDjuH9PaT8UiH61LKtHzMwOFq7bU+IT+E4cXetk/wYnMYrFMDlfcagZa7Jasr+Do4x2hfdWD2REAtW5vdp0+GNaDJKzP+5KGco5u7/QWr169tXZEQTYnIn3MQI+ivamPge4h+rqGOLa9E4XTB8cch17P1DxX8LMp9ppURT64WHlYGwazd7p+lhaLZXK54ix6SM3/HkyvW71tlhdA9db+dBfhNs+rOyQWlES8vim4Vnlwmb9EcYKRrG59naOPd3hvGEvWV3jnH6kyt7875sUQTEx/f6IxBd+Ewtotm7GJbMgIslgsDjlt0YdZtPpnWHAzuL8pVK886q94BKco6F/+4LTv2Fs2VRNTird/foGjj3c4Yr+lhoEe5XsDMK1j/WffxigHNzsinMjqDrPIzfPrcwYtfS3ST998yssOAsenro97+ZEo+za0xD2ng5tb496Evp33e0/k1x6a71vpqqDEFvdYLNlEzlr0iQKuBSURBnpiXnuC1WZfdaVCs0N05WewkdWx7Z1ULC2M8/frpelqVhQntMq18La+fInqFcXDLYEbynlxfbOvLz0MW936dxM9mUBi33swcNre1EfedGHW7cXMvmMG+VfDq9/uoq99iGjjJadTpdHGIVGg+bqPzuClL7b5+sXsf7iVg5tbs7LtgcVyJZKTQp8s4FpQIl57Ar0dpYg29vLKo23hla8JrjX7zhnMvnNGQkE3q0RNd4YTYC2ktbGXVtfvX1lfyLsvXaDjkNPwKjgRqVgMRLwS+rCFwRO5o8JEeuiS4viObuoayjiz+6LXSKutsc/nnlI4bzQ6JmDy1vNObxalFLdvm+WNp86YHCwWy+SSk66bZAHX5ZtrvPYEWrgUTqA1bIEHEbc3e2C91CXuClA6u0STyPI2XTVOgNWfY9/e1E/HoQGmz87ztQ5YvbWW6hXFtB3so6BEfCKvP+uYQNj1wAmcBsdTVJkHwLHtXd7iGOVLC3z76LhEf9cQe920zuDi2cU1eV7sQou8EO7uslgsE8+4hF5E/kFE/lNEjorIsyJSamz7ioicFJHXReQj4x/qqMc2op/79m2zfBPBMddtYlrBZhbLLZuq433+wLJHqkLz7cNEUfvCAW57rNoT2iCL1pY4QVyD2hXFtDb28tbzF7ntsWrPcj6z+wL5VzPsjlpfwbJHKr3FK/ZtjDI0NOTzzT84dANL1lf4WuFq5tw50/dZd3A0V2bSz3X11lrqGsq5ap5/ctBuKLsqksWSHYzXov8lcJNSagnwBvAVABG5AfgkcCNwF/AtEQlXtQwxUsA1UZqhKUmNm6I8ffMpGje5wc0NLRzb0UXNimJPuI9t7wwV0GPbOzn5kx6qVxT5RFHnviuleOmLbaFCCxAzcvl1IHTV1lqvvcJ3pr3ma7cw+B4+S1+nY+rc+5e+2EZ7Ux9FlXm0N/Wx/+FWVv7PKvKK4t0qWtjDuGVTNau31nJwc5svXbR6eZFvv6Mhk6bFYpk8xuWjV0r9wvj4MrDW/f0e4MdKqX7gtIicBJYDB8ZzvVGMK2FTMh38XL21Ns7VoYOtq7bUcPxbXZ4Q37KpmqjrR6928+lXb3X6h7c39XqCq4Oiuqf49NppHNzcykCPYtWWGhTDywae+12vu/C0331TVB3h+I4uWvZeoqOp3yvi0v3m9ULggK9vuZkaaQZe9WpFS9ZXcNtj1bz0xTZfPAHgpnVltLx4yXPfvPubC74xnfxJDwUlwuB7zpvImd3naW/q493fXCCvQGgLBI6BMS+jZ7FY0k/auleKyPPAT5RS/1tEvgm8rJT63+627wP/oZR6ZqRzpLN75cHNrfR3DXl+br00X6vbM324+Vi82GqCK/3o8v5IJOJZ+YVleSx7pMoXFAV8zc1G+jm9No9L0XDLvmZFMfceWODLGjKvAfDg0A2hfeiDnSR1F8ng9zetK+OOx2ejlOLH9Sd4781Bhi7hvR089cGTdBwZ8J6Ffl7BZxN8jubYLRZLZknWvTKp60ZEfiUix0P+3GPs81WcRcD/SX8VcqrQGUVEHhCRQyJy6Ny5c8mGMyrCLjj3IzO8ICI4AdAlDeVU3hLfZXLRn13t+9zR1O9VfuoVoPq7HT9+0A103+EFgesM/1yyvoLrPjqDyvoiLkWHvA6XQZ+9fnvQk4qZ/655+uZTXiwhWZ1AWEDW1GGJCMWV07y3hEN/c45Zd8ygsr7QE3Ut5kG3U3tTP3UN5SxpKKd6eRGtjb2htQkWi2XiSeq6UUp9eKTtInI/8CfAh9Tw/+p3gGuN3a4B4nvYOuffCewEx6JPYcxJ0emV5uo++zZGfWmJpmW8yt3e/orfsg/2ka90i4u8NMqGcqcidWM0blLZddtprv1v00PHV1DiLGwh0sbsO2ewaksNeze0xF1Pa/Arj7YRffmSazX3eX3uTz39nucyKigRz0Wk3y4q64uYd/dVXqBWu5TqGspp/s1FOo70c2x7FyA0/+YinUcGqKwv4rbHnBWU+rtjHN/RxU3ryhK+9WgqljqTlfbP19w63a6KZLFkCePy0YvIXcCXgDuVUpeMTbuB/yMiW4DZwGLg4HiuNcpxJWwjoDNWTPZtjPLungtx54Fhd81Pbzvt5btrYkpx9ucXaGt0rHWdVnh0eydtjb30nAwXx9/v7KKva4g7Hp8NOBNTdO8l3z66R47CEfy2g31ULS/y8tq1AEfEWYd0oHvIy9nXTdDam/q8iQSgvanXs9aVUjz1wTddsXeOMxcbh+E4RLBQzHs2SwuZfecMjm3vpONIPx1H+n0LqViRt1iyg/EWTH0TKAR+6f6nflkp9Tml1Ksi8hTwexyXzheUUuGO6AwRViCUf7XwzLLTnvukoEQ4/q0uT8jK6vLpOjbo7V9Umcd1fzITEaHKdUeYaLEFpx2vPo9jMV/w+bbNc16KDnF8R5eX5rl/YzTOYtYdKgt0bxy3v42umK2sL/K1TN67ocWrxNUEK2PNgK2I8InfLvQFd02R9xqTBZ5rcU0eva1DFFXm0XGkn1l3+N9abN95iyX7GFd6pVJqkVLqWqVUvfvnc8a2v1NKLVRKvU8p9R/jH+qoxxZSxXnBC4Q62SMX6GsfYtp0ofKWQiJ5/schBYqBHicXPGy9V3AmlI/vn+f7rrA0QqRAWNJQTtkN/hzzGx8spaK+kBlzpg0XGbkTxI0PlfpSNKMvX2IgQQxAu2wAzy2lffqaYGWs+TOs86T29zduinr+9Whgcpte69gGVy3I58aHSjn19Hu+7dYvb7FkHznbAmHvhhbPJ796ay0vrm/m+I4uXy46DGfHaP+8dl9o8dR+86KyaVQsLfRSEDXRly/xxJw3fN+dfu48HUf6nZz9y/4J4szzF+ho6qd6eREX373sfX/TujJEhF23nWbuR2ZS11BOW2OvNwkEpVP32NEWvHYbmZj9cUwrW4u8rmpd9H+XeP77n3zgJBIROpr6ve8q6gvpaOr3/Tx3sM/3drH20HyeWXbadq+0WLKQnBT6Vx5to62x16no3FrLwc2tRPdeonp5EXPvmullwABxuen3HV7guTpaXrxIe1N/wsW9S2+a5ms+VrYkn2vWXOW5cNrdnjUmHU39lC8t4HKfv3e8mcde4+b5i4hTYWv4yHUxU9BvrmMDNSuKqVpeRMR19aAUCigqy/N6+EQiEQpKIj4fuz5vX/sQF9+97E2Aesxm24X8q53vNGsPzfeKsnSRlhV5iyV7yDmh1xk3rY29TuMypTiz+wLtriXa1+UPFfxo9uu+zzqDpqgsj/sOL2S/m0apqVhayKzbizn+zW66j1/2Hdt1dJCuo86+1cuLQguJAKYVRmg72EdRdYS+NkfwtcjXrSvzfO+6ZXEQM9iqaXn5ktdI7PiOLqrdwG3UrRsINhn72K/n+/z1Qd/+qi01vgkw2J3SRL8d6ePC8votFsvkkXP/I30NzbZ3Gn3THbeDts7r1pVRsbSQ3lZH+G9aV+Zluhzb3ulNCLGAv3nW7cX+5PMEDA0k9lOfP+sEfGfMyo/bptdXHak1sjNx9Xl56wDnDvb5hLztYJ+X/aPbIGt0/OLQ35xL2N8+LDMpFot5qZt6Ba3gcVbkLZbsIyf/V4aJ19pDC3yfC8vymH/P1VTWF1K1vIjjO7o8y123SNC95U2Of7M7oSvHpKOpn8LKCDc9VBq3rTc6RF4Rcf5+gOPf6mJoaOQEpbxCoXxJAYWleazeNssT++M7urzmbCa1K4p9yxkGFyUJBq11/55gQzYnXz/ic+OYmMsIWiyW7CEnhV4pxa6V/oySH9X6A6Zndl/glk1V3Hd4Ife9vNC3ba1buv/mMz2AY/0HrdfKZQVuamVi+ttjtLx8ySsmMhkK9+rQ1+50mlRKxaVzguOjv/DuZTqPDtDf7UwIwU6XQanVqZrBls1mcVWw9XBlfZGXf6/fkJxlCavjjtOZQuYKV77ru43kEn22WCyZJSd99PvcnjbgpCy++ePzIX1ZhpuQhbULuO2xatDxUpG4fWLxcVYfN60ro62xN6GfHqCwSug/Nyx45UsK6G0douNIv+cfD2b6mK4cce83mCapg6u3G4uWBDtSap+7uQi6+SaUf7Vw6G/Oeat0rdpSg7jPobA0kvC4YDXswc2tnP3ZBapXFHt9+/dtaCHa2Mt1d820q1BZLBNATgm99lHrRUKUUhwLcbPcd3ih54Yw2wVoS1a3C7gUHaKivtAnrnUNZby75yKdRwfoPDrgpkXithIYRsRZxWokoTdFHmDOmhms2lLrBTfBce+EBUfr3HjC0UBVq67g1VKrq1vbAm8HOvUyuAi6GJW2etERpRQCXmA3uFC5/hlMqVRK0d815FtFS2cHAV6w3GboWCyZJWeE3lwjdvnmmlDfs0Zb8pFIhIObW0OX5dPBzlVbanzCu3qrY5X+aM7r9LoVrhptkfe2xgdRk+FMKF20vOgXZF3ctT94LwHXh65qvffAAvaub/YFdXHdQMGWzc79xIvzQE/M61lT11AemOgS95kPficiXjsEHeQ2z3O7raK1WCaEtLUpTgdjbVMc1n9eF0yZmL1owvq4B1v4Vi8vYmhA+XLGi2vzuPGBcpY9UumbAG5aV4ZA3BtE+dICOo8M+3nCiq40w62Aw1sZL2kodxqwuYugmOj7eeXRtrj2zLtWnkIxHHvQz6tQt1cY4XkG0e2OR0PwuY71PBaLJZxkbYpzwqJP1MQMHLGuvXW654bQKYmmL9kUPxPtdqmoL2T27dM5tqOL3ugQp597j74ufw69tuyD+e1z7pjh5c2D44oprIjQ3zFcMKXFf7gVcN/wW8bGKCd3OW0G9JTc7DZAK67N49Pvvs97E9EulqPbO8Ho2tna2OvLxAlzs4Q9zzCh1y6fVEVax0yC7N3QYi16i2WCyAmLXhO0HM1OioBnxd6yqTrOXZHKGwE4ohwpFM4d7Av1ncOwW2KfW9UadH/o83zitwu9VgtBS99cKASIG0/cIijGvQWtcfPtJdXnuC9Qkau7Zh4LvA2N5jzmG5X5nKzYWyzj44qw6CG8iZmewrSQjORbDmaR6AyRoEB/4rcLnQU5Vjrtf3fddtq3XadSvvJoG6u3zUJBXCBU94Y59DfnENwMnYPhwVI93tu3zfKNRYu8Hr+5b9AaH7XIu9XANSuKvWIr02efap95HRjX57ndmHSjjb0UuXEEi8WSWXJC6EdaI9YUwZFEJZh9ooOYQfZvjDoC7l5Ti7jZg0b3ZdeulLBA6P6HWz0BrVhaQIe76Ed7Ux/lSwp8wVK9zqxvHA+3hnanTLS6VKpiH5z0jA2hb0PJWL65hls2VfvGaPvVWywTS04I/Uj54KNZ5Ujv17ipldO73/OCsHXrymh2F+vWbofV22ZRVJYX544AtymZ65IoLMtLOC696LieLNqb+rwAcNXyIgpLI16evLkQeKKsmZEmvOC+IxGc9EZz7EjPNdFni8WSWXJC6CFenJIFHBOh0ws7mvoprs1j0X1XIyJee968AvFSF/U1wS/05qLYYePSqZ1KKWpXFPtcOzE3y6eyvtBbdNysVB1pEkvXhKfPNdJni8UydRhXMFZE/ha4B6eGtA34tFKqWRxVeBz4Y+CS+/1vk51vvMHYdBGWXqgDr0EXUKJ9E00yZr4/4PXJN3Eqd4cDs2FdIUcqNApus0VJFktukywYO95eN/+glFqilKoH/hV4xP3+j3DWiV0MPAB8e5zXmVDCmqKt3lob15nRyyoJaQCWqOdLf7ezUPe+DS2hC4ID1N4eWJ4v5NojCbe1xi0Wi8m4XDdKKXMduRkMJ7rcAzypHKV7WURKRWSWUio+oToLSRTQLCgRBnqUz1qPun3vU3GVBPP9NUELPmx5Prtik8ViGSvj7l4pIn8nIm8Df86wRT8HeNvY7R33u6wnGNA0rfQzuy/4rHWddWOu1arFPFGzrrC3hXZ3BacHh26gsr6QS1GnOjZZV0iLxWJJhaQWvYj8CqgN2fRVpdRzSqmvAl8Vka8ADwGbIG75UojvnqvP/wCOe4e5c+emOu6MMVJAs6Akwuw7Z/iqb8P88SNZ3mFvCzrQGolEmHf31cB55t09k0gkMuZgqsVisWjSVhkrItcB/6aUuklEvgvsUUr9s7vtdWBNMtdNtgRjIXFAM1h9O5qeLSOlP5oTRiwWSznwarFYLBkNxorIYuPj3cB/ur/vBj4lDrcCPVPFP68J868n8t2nOlkmelvQi3roa44m8GqxWCzJGG8e/ddF5H046ZVvAZ9zv/93nNTKkzjplf9jnNeZdDJVjDTWfH+LxWJJlfFm3dyb4HsFfGE85842bDGSxWKZquRMZexEYK1xi8UyFcnJxcEzibXGLRbLVMMKvcViseQ4VugtFoslx7FCb7FYLDmOFXqLxWLJcbJqzVgROYeTj58pKoH2DJ5/IrH3kp3Ye8lOcv1erlNKVSU6IKuEPtOIyKGRyoSnEvZeshN7L9nJlX4v1nVjsVgsOY4VeovFYslxrjSh3znZA0gj9l6yE3sv2ckVfS9XlI/eYrFYrkSuNIveYrFYrjhyXuhF5G9F5KiINInIL0Rktvu9iMh2ETnpbv/gZI81GSLyDyLyn+54nxWRUmPbV9x7eV1EPjKZ40wFEblPRF4VkZiILAtsm1L3AiAid7njPSkiX57s8YwGEfmBiLSJyHHju3IR+aWInHB/lk3mGFNFRK4VkV+LyGvuv6/17vdT7n5EpEhEDorIEfdeHnW/ny8ije69/ERECpKeTCmV03+Aq43fG4DvuL//MfAfOMse3go0TvZYU7iX/wZMc3//e+Dv3d9vAI4AhcB84E0gb7LHm+Re3g+8D9gDLDO+n4r3kueOcwFQ4I7/hske1yjGfwfwQeC48d03gC+7v39Z/1vL9j/ALOCD7u9XAW+4/6am3P242jTT/T0faHS16ingk+733wEeTHaunLfolVLvGR9nMLx27T3Ak8rhZaBURGZN+ABHgVLqF0qpy+7Hl4Fr3N/vAX6slOpXSp3GWfBl+WSMMVWUUq8ppV4P2TTl7gVnfCeVUqeUUgPAj3HuY0qglHoR6Ax8fQ/whPv7E8DHJnRQY0Qp1aKU+q37+3ngNWAOU/B+XG264H7Md/8o4A+BZ9zvU7qXnBd6ABH5OxF5G/hz4BH36znA28Zu77jfTRU+g/NGAlP/Xkym4r1MxTEno0a5y3+6P6sneTyjRkTmAR/AsYSn5P2ISJ6INAFtwC9x3hy7DYMvpX9rOSH0IvIrETke8uceAKXUV5VS1wL/BDykDws51aSnICW7F3efrwKXce4HpvC9hB0W8t2k30sSpuKYcxoRmQnsAjYE3uqnFEqpIaVUPc7b+3Icl2fcbsnOkxMrTCmlPpzirv8H+DdgE85MeK2x7RqgOc1DGzXJ7kVE7gf+BPiQcp10TNF7SUBW3ksSpuKYk9EqIrOUUi2uS7NtsgeUKiKSjyPy/6SU+qn79ZS9HwClVLeI7MHx0ZeKyDTXqk/p31pOWPQjISKLjY93A//p/r4b+JSbfXMr0KNf7bIVEbkL+BJwt1LqkrFpN/BJESkUkfnAYuDgZIwxDUzFe3kFWOxmQxQAn8S5j6nMbuB+9/f7gecmcSwpI86Sb98HXlNKbTE2Tbn7EZEqnVknIsXAh3FiDr8G1rq7pXYvkx1ZnoDI9S7gOHAUeB6YY0S0/xHH53UMI/MjW//gBCbfBprcP98xtn3VvZfXgT+a7LGmcC9/imMJ9wOtwM+n6r24Y/5jnAyPN4GvTvZ4Rjn2fwZagEH37+SzQAXwAnDC/Vk+2eNMm2UO0AAAAHBJREFU8V5W47gyjhr/T/54Kt4PsAT4nXsvx4FH3O8X4Bg/J4GngcJk57KVsRaLxZLj5LzrxmKxWK50rNBbLBZLjmOF3mKxWHIcK/QWi8WS41iht1gslhzHCr3FYrHkOFboLRaLJcexQm+xWCw5zv8PoApXC9mLKhMAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cluster_and_visualize(male + female, extract_vectors(male + female, gp_gn_limit, gp_gn_w2i_limit), \n",
    "                          'GP-GN-GloVe', random_state, tsne_random_state, y_true)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Cluster originally top biased words\n",
    "Following results may be slightly different from the numbers reported in the paper. This is due to:\n",
    "\n",
    "1. how we select top biased words. Here, we select ones with largest similarities with **he** and **she** without normalizing embeddings, however, this may not be the perfect way. Other options can be:\n",
    "    - normalize word embeddings first and then pick top biased words.\n",
    "    - select according to projections on the difference vector of **he** and **she**.\n",
    "2. different random state used in the clustering algorithm.\n",
    "\n",
    "We care more about the relative clustering acc change and we see significant decrease with Double-Hard Debiased embeddings."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 220,
   "metadata": {},
   "outputs": [],
   "source": [
    "def cluster(words, X1, random_state, y_true, num=2):\n",
    "    \n",
    "    kmeans_1 = KMeans(n_clusters=num, random_state=random_state).fit(X1)\n",
    "    y_pred_1 = kmeans_1.predict(X1)\n",
    "    correct = [1 if item1 == item2 else 0 for (item1,item2) in zip(y_true, y_pred_1) ]\n",
    "    print('precision', max(sum(correct)/float(len(correct)), 1 - sum(correct)/float(len(correct))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 221,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cluster most biased words before and after debiasing\n",
    "\n",
    "def my_cluster(wv, w2i, start=0, size=500, random_state = 1):\n",
    "    \n",
    "    if start == 0:\n",
    "        female = [item[0] for item in sorted_g[:size]]\n",
    "        male = [item[0] for item in sorted_g[-size:]]\n",
    "    else:\n",
    "        female = [item[0] for item in sorted_g[start:size+start]]\n",
    "        male = [item[0] for item in sorted_g[-(start+size):-start]]\n",
    "\n",
    "    y_true = [1]*size + [0]*size\n",
    "    \n",
    "    cluster(male + female, extract_vectors(male + female, wv, w2i), random_state, y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 239,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 1.0\n",
      "precision 0.765\n",
      "precision 0.765\n",
      "precision 1.0\n",
      "precision 1.0\n",
      "precision 1.0\n",
      "precision 1.0\n",
      "precision 0.665\n"
     ]
    }
   ],
   "source": [
    "random_state_=0\n",
    "start=0\n",
    "end=100\n",
    "my_cluster(glove_limit, glove_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(hd_limit, hd_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(hd_a_limit, hd_a_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gn_limit, gn_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gn_a_limit, gn_a_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gp_limit, gp_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gp_gn_limit, gp_gn_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(dhd_limit, dhd_w2i_limit, start, end, random_state=random_state_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 224,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 1.0\n",
      "precision 0.8029999999999999\n",
      "precision 0.8029999999999999\n",
      "precision 1.0\n",
      "precision 0.995\n",
      "precision 1.0\n",
      "precision 1.0\n",
      "precision 0.741\n"
     ]
    }
   ],
   "source": [
    "random_state_=0\n",
    "start=0\n",
    "end=500\n",
    "my_cluster(glove_limit, glove_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(hd_limit, hd_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(hd_a_limit, hd_a_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gn_limit, gn_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gn_a_limit, gn_a_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gp_limit, gp_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gp_gn_limit, gp_gn_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(dhd_limit, dhd_w2i_limit, start, end, random_state=random_state_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision 1.0\n",
      "precision 0.8025\n",
      "precision 0.8025\n",
      "precision 0.998\n",
      "precision 0.8945\n",
      "precision 1.0\n",
      "precision 0.9995\n",
      "precision 0.704\n"
     ]
    }
   ],
   "source": [
    "random_state_= 0\n",
    "start=0\n",
    "end=1000\n",
    "my_cluster(glove_limit, glove_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(hd_limit, hd_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(hd_a_limit, hd_a_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gn_limit, gn_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gn_a_limit, gn_a_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gp_limit, gp_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(gp_gn_limit, gp_gn_w2i_limit, start, end, random_state=random_state_)\n",
    "my_cluster(dhd_limit, dhd_w2i_limit, start, end, random_state=random_state_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Classification experiment\n",
    "\n",
    "We also test classification experiment as described in [Lipstick on a Pig](https://arxiv.org/abs/1903.03862). However, we don't see any significant improvement on any debiasing approaches. We suspect that classification may not reflect gender bia correctly as it can always find a decision boundary with supervisions which may not be gender. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn import svm\n",
    "from random import shuffle\n",
    "import random\n",
    "from utils import train_and_predict\n",
    "\n",
    "def cls(wv, w2i, vocab, size_train, size_test):\n",
    "    size = size_train + size_test\n",
    "    females = [item[0] for item in sorted_g[:size]]\n",
    "    males = [item[0] for item in sorted_g[-size:]]\n",
    "    males.reverse()\n",
    "\n",
    "    shuffle(females)\n",
    "    shuffle(males)\n",
    "\n",
    "    train_and_predict(wv, w2i, vocab, size_train, size_test, males, females)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy: 1.0\n",
      "accuracy: 0.9105\n",
      "accuracy: 0.9265\n",
      "accuracy: 1.0\n",
      "accuracy: 0.9935\n",
      "accuracy: 0.995\n",
      "accuracy: 1.0\n",
      "accuracy: 0.8905\n"
     ]
    }
   ],
   "source": [
    "size_train=100\n",
    "size_test=1000\n",
    "cls(glove_limit, glove_w2i_limit, glove_vocab_limit, size_train, size_test)\n",
    "cls(hd_limit, hd_w2i_limit, hd_vocab_limit, size_train, size_test)\n",
    "cls(hd_a_limit, hd_a_w2i_limit, hd_a_vocab_limit, size_train, size_test)\n",
    "cls(gn_limit, gn_w2i_limit, gn_vocab_limit, size_train, size_test)\n",
    "cls(gn_a_limit, gn_a_w2i_limit, gn_a_vocab_limit, size_train, size_test)\n",
    "cls(gp_limit, gp_w2i_limit, gp_vocab_limit, size_train, size_test)\n",
    "cls(gp_gn_limit, gp_gn_w2i_limit, gp_gn_vocab_limit, size_train, size_test)\n",
    "cls(dhd_limit, dhd_w2i_limit, dhd_vocab_limit, size_train, size_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Association Experiments (Calisken et al.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['john', 'paul', 'mike', 'kevin', 'steve', 'greg', 'jeff', 'tom']\n",
      "['amy', 'joan', 'lisa', 'sarah', 'diana', 'kate', 'ann', 'donna']\n",
      "['executive', 'management', 'professional', 'corporation', 'salary', 'office', 'business', 'career']\n",
      "['home', 'parents', 'children', 'family', 'cousins', 'marriage', 'wedding', 'relatives']\n"
     ]
    }
   ],
   "source": [
    "from utils import p_value_exhust, effect_size\n",
    "from utils import WEAT_words\n",
    "\n",
    "A = WEAT_words['A']\n",
    "B = WEAT_words['B']\n",
    "C = WEAT_words['C']\n",
    "D = WEAT_words['D']\n",
    "# C = WEAT_words['E']\n",
    "# D = WEAT_words['F']\n",
    "# C = WEAT_words['G']\n",
    "# D = WEAT_words['H']\n",
    "\n",
    "A = [elem.lower() for elem in A]\n",
    "A = ['john', 'paul', 'mike', 'kevin', 'steve', 'greg', 'jeff', 'tom']\n",
    "B = [elem.lower() for elem in B]\n",
    "B = ['amy', 'joan', 'lisa', 'sarah', 'diana', 'kate', 'ann', 'donna']\n",
    "C = [elem.lower() for elem in C]\n",
    "D = [elem.lower() for elem in D]\n",
    "print(A)\n",
    "print(B)\n",
    "print(C)\n",
    "print(D)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.805996189486473\n",
      "1.546645694968245\n",
      "1.5466455821704757\n",
      "1.8211045958742451\n",
      "1.7554763167372835\n",
      "1.8058846500019743\n",
      "1.7974306940681066\n",
      "1.5313012260878656\n"
     ]
    }
   ],
   "source": [
    "print(effect_size(A, B, C, D, glove, glove_w2i, vocab))\n",
    "print(effect_size(A, B, C, D, hd_glove, hd_glove_w2i, hd_vocab))\n",
    "print(effect_size(A, B, C, D, hd_glove_a, hd_glove_a_w2i, hd_vocab_a))\n",
    "print(effect_size(A, B, C, D, gn_glove, gn_glove_w2i, gn_vocab))\n",
    "print(effect_size(A, B, C, D, gn_glove_a, gn_glove_w2i, gn_vocab))\n",
    "print(effect_size(A, B, C, D, gp_glove, gp_glove_w2i, gp_vocab))\n",
    "print(effect_size(A, B, C, D, gp_gn_glove, gp_gn_glove_w2i, gp_gn_vocab))\n",
    "print(effect_size(A, B, C, D, dhd_glove, dhd_glove_w2i, dhd_vocab))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "num of samples 12870\n",
      "0.0\n",
      "num of samples 12870\n",
      "0.0001554001554001554\n",
      "num of samples 12870\n",
      "0.0001554001554001554\n",
      "num of samples 12870\n",
      "0.0\n",
      "num of samples 12870\n",
      "0.0\n",
      "num of samples 12870\n",
      "0.0\n",
      "num of samples 12870\n",
      "0.0\n",
      "num of samples 12870\n",
      "0.0001554001554001554\n"
     ]
    }
   ],
   "source": [
    "print(p_value_exhust(A, B, C, D, glove, glove_w2i, vocab))\n",
    "print(p_value_exhust(A, B, C, D, hd_glove, hd_glove_w2i, hd_vocab))\n",
    "print(p_value_exhust(A, B, C, D, hd_glove_a, hd_glove_a_w2i, hd_vocab_a))\n",
    "print(p_value_exhust(A, B, C, D, gn_glove, gn_glove_w2i, gn_vocab))\n",
    "print(p_value_exhust(A, B, C, D, gn_glove_a, gn_glove_w2i, gn_vocab))\n",
    "print(p_value_exhust(A, B, C, D, gp_glove, gp_glove_w2i, gp_vocab))\n",
    "print(p_value_exhust(A, B, C, D, gp_gn_glove, gp_gn_glove_w2i, gp_gn_vocab))\n",
    "print(p_value_exhust(A, B, C, D, dhd_glove, dhd_glove_w2i, dhd_vocab))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}