Lecture-4/Lecture-4.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Text classifcation applications\n",
    "\n",
    "## Sentiment analysis\n",
    "\n",
    "In this lecture we use IMDB dataset for sentiment analysis classification:\n",
    "\n",
    "http://ai.stanford.edu/~amaas/data/sentiment/\n",
    "\n",
    "Download the data in tsv format:\n",
    "https://github.com/Poyuli/sentiment.analysis\n",
    "\n",
    "Load data with pandas:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "df = pd.read_csv(\"labeledTrainData.tsv\", header=0, delimiter=\"\\t\", quoting=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Printing top 10 training data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>sentiment</th>\n",
       "      <th>review</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>5814_8</td>\n",
       "      <td>1</td>\n",
       "      <td>With all this stuff going down at the moment w...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2381_9</td>\n",
       "      <td>1</td>\n",
       "      <td>\\The Classic War of the Worlds\\\" by Timothy Hi...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>7759_3</td>\n",
       "      <td>0</td>\n",
       "      <td>The film starts with a manager (Nicholas Bell)...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3630_4</td>\n",
       "      <td>0</td>\n",
       "      <td>It must be assumed that those who praised this...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>9495_8</td>\n",
       "      <td>1</td>\n",
       "      <td>Superbly trashy and wondrously unpretentious 8...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>8196_8</td>\n",
       "      <td>1</td>\n",
       "      <td>I dont know why people think this is such a ba...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>7166_2</td>\n",
       "      <td>0</td>\n",
       "      <td>This movie could have been very good, but come...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>10633_1</td>\n",
       "      <td>0</td>\n",
       "      <td>I watched this video at a friend's house. I'm ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>319_1</td>\n",
       "      <td>0</td>\n",
       "      <td>A friend of mine bought this film for £1, and ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>8713_10</td>\n",
       "      <td>1</td>\n",
       "      <td>&lt;br /&gt;&lt;br /&gt;This movie is full of references. ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        id  sentiment                                             review\n",
       "0   5814_8          1  With all this stuff going down at the moment w...\n",
       "1   2381_9          1  \\The Classic War of the Worlds\\\" by Timothy Hi...\n",
       "2   7759_3          0  The film starts with a manager (Nicholas Bell)...\n",
       "3   3630_4          0  It must be assumed that those who praised this...\n",
       "4   9495_8          1  Superbly trashy and wondrously unpretentious 8...\n",
       "5   8196_8          1  I dont know why people think this is such a ba...\n",
       "6   7166_2          0  This movie could have been very good, but come...\n",
       "7  10633_1          0  I watched this video at a friend's house. I'm ...\n",
       "8    319_1          0  A friend of mine bought this film for £1, and ...\n",
       "9  8713_10          1  <br /><br />This movie is full of references. ..."
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(25000, 3)"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Histogram plot:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZUAAAD8CAYAAAC/1zkdAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ngg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1dc4b67e2b0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "df['sentiment'].plot.hist()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Creating 80% of data for the training, and 20% for validation:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(20000,) (5000,)\n"
     ]
    }
   ],
   "source": [
    "#First shuffle data\n",
    "df = df.sample(frac=1, random_state= 100)\n",
    "\n",
    "X_train_text = df['review'][:20000]\n",
    "y_train = df['sentiment'][:20000]\n",
    "\n",
    "X_valid_text = df['review'][20000:]\n",
    "y_valid = df['sentiment'][20000:]\n",
    "\n",
    "print(y_train.shape, y_valid.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Vectorizing the training data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(20000, 16414)\n"
     ]
    }
   ],
   "source": [
    "from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer\n",
    "\n",
    "#vect = CountVectorizer(binary = False, min_df = 10)\n",
    "vect = TfidfVectorizer(min_df = 10)\n",
    "\n",
    "X_train = vect.fit_transform(X_train_text)\n",
    "print(X_train.shape)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Vectorizing the validation data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(5000, 16414)\n"
     ]
    }
   ],
   "source": [
    "X_valid = vect.transform(X_valid_text)\n",
    "print(X_valid.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Training logistic regression:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LogisticRegression(C=0.1, class_weight=None, dual=False, fit_intercept=True,\n",
       "          intercept_scaling=1, max_iter=100, multi_class='ovr', n_jobs=1,\n",
       "          penalty='l1', random_state=None, solver='liblinear', tol=0.0001,\n",
       "          verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "\n",
    "LR = LogisticRegression(C = .1, penalty = 'l1')\n",
    "LR.fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Predicting the validation set and report the accuracy:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.7972\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics  import accuracy_score, classification_report, confusion_matrix, precision_score, recall_score, f1_score\n",
    "\n",
    "p_valid = LR.predict(X_valid)\n",
    "print(accuracy_score(y_valid, p_valid))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Printing clasification report (precision, recall, F1):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "             precision    recall  f1-score   support\n",
      "\n",
      "   Negative      0.889     0.874     0.881      2502\n",
      "   Positive      0.876     0.891     0.883      2498\n",
      "\n",
      "avg / total      0.882     0.882     0.882      5000\n",
      "\n"
     ]
    }
   ],
   "source": [
    "target_names = ['Negative', 'Positive']\n",
    "print(classification_report(y_valid, p_valid, digits = 3, target_names  = target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Confusion matrix:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[2186  316]\n",
      " [ 273 2225]]\n"
     ]
    }
   ],
   "source": [
    "print(confusion_matrix(y_valid, p_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "80"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(LR.coef_ != 0).sum()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Feature importance:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Positive: great, excellent, best, wonderful, love, and, perfect, well, amazing, very, will, favorite, also, life, it, is, loved, fun, his, as\n",
      "\n",
      "Negative: worst, bad, waste, awful, boring, no, nothing, poor, terrible, minutes, even, just, script, stupid, plot, worse, horrible, why, any, they\n",
      "\n"
     ]
    }
   ],
   "source": [
    "vocab = np.array(vect.get_feature_names())\n",
    "order = LR.coef_[0].argsort()\n",
    "print(\"Positive: %s\\n\" % (\", \".join(vocab[order[-20:][::-1]])))\n",
    "print(\"Negative: %s\\n\" % (\", \".join(vocab[order[:20]])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Learning curve:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0.8166     0.79030481 0.86148919 0.82436315]\n",
      " [0.845      0.82157521 0.88110488 0.8502994 ]\n",
      " [0.8524     0.83587786 0.87670136 0.85580305]\n",
      " [0.8616     0.84730769 0.88190552 0.86426049]\n",
      " [0.8652     0.85266821 0.88270616 0.86742722]\n",
      " [0.868      0.85427911 0.88710969 0.87038492]\n",
      " [0.872      0.86091686 0.88710969 0.87381703]\n",
      " [0.8726     0.86192143 0.88710969 0.87433419]\n",
      " [0.8778     0.86927593 0.88911129 0.87908173]\n",
      " [0.8766     0.86495925 0.89231385 0.87842365]\n",
      " [0.8788     0.87039937 0.88991193 0.88004751]\n",
      " [0.883      0.87613055 0.89191353 0.8839516 ]\n",
      " [0.8844     0.87735849 0.89351481 0.88536295]\n",
      " [0.884      0.88055556 0.88831065 0.8844161 ]\n",
      " [0.8862     0.88077379 0.89311449 0.88690121]\n",
      " [0.8888     0.88440222 0.89431545 0.88933121]\n",
      " [0.893      0.88779139 0.89951962 0.89361702]\n",
      " [0.8922     0.88547816 0.90072058 0.89303433]\n",
      " [0.8922     0.88639053 0.89951962 0.89290682]\n",
      " [0.8924     0.88704581 0.8991193  0.89304175]]\n"
     ]
    }
   ],
   "source": [
    "scores = []\n",
    "metrics = [accuracy_score, precision_score, recall_score, f1_score]\n",
    "for i in range(1000, 20001, 1000):\n",
    "    X_train_text = df['review'][:i]\n",
    "    y_train = df['sentiment'][:i]\n",
    "\n",
    "    vect = TfidfVectorizer(min_df = .0005)\n",
    "\n",
    "    X_train = vect.fit_transform(X_train_text)\n",
    "    X_valid = vect.transform(X_valid_text)\n",
    "\n",
    "    LR = LogisticRegression(C = 5)\n",
    "    LR.fit(X_train, y_train)\n",
    "\n",
    "    p_valid = LR.predict(X_valid)\n",
    "    scores.append([metric(y_valid, p_valid) for metric in metrics])\n",
    "\n",
    "scores = np.array(scores)\n",
    "print(scores)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZgAAAEWCAYAAABbgYH9AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvNQv5yAAAIABJREFUeJzs3Xd8lFWi//HPmZnUSa+kBxJagFCkSFFpQWyrWMGGrr2vbv3dq6u792653r1217WhdEUBRUUEXBAVpdeEFgIhjfQ2M5l+fn88QwgQIEAmZTjv12te0555njO8Qr45XUgpURRFUZT2puvsAiiKoii+SQWMoiiK4hUqYBRFURSvUAGjKIqieIUKGEVRFMUrVMAoiqIoXqECRlEugBDiayHEzM4uh6J0RULNg1G6IyHEYeB+KeXqzi6LoiitUzUYRTkNIYShs8twoXzhOyjdlwoYxecIIa4VQmwXQtQJIdYLIbJbvPcHIcRBIUSjECJPCDGtxXv3CCF+FEK8LISoAV7wvPaDEOIfQohaIcQhIcRVLT6zVghxf4vPn+nYnkKIdZ5rrxZCvCmEmHeG73G953s0eMo81fP6YSHE5BbHvXDsPEKIdCGEFELcJ4Q4AvxbCLFCCPH4SefeIYS40fO4nxBilRCiRgixTwhx6/n/6yvKcSpgFJ8ihBgGzAIeAqKBt4FlQogAzyEHgcuAcOBPwDwhREKLU4wCCoA44C8tXtsHxAAvAu8LIcRpinCmYxcAGz3legG46wzfYyQwB/gtEAFcDhw+2/dv4QqgP3Cl57ozWpw7C0gDvhJCGIFVnmPiPMf9Uwgx4ByupSitUgGj+JoHgLellBuklC4p5WzABlwKIKX8REpZKqV0Syk/Bg4AI1t8vlRK+bqU0imlbPK8ViilfFdK6QJmAwlA/Gmu3+qxQohUYATwRymlXUr5A7DsDN/jPmCWlHKVp6wlUsq95/Dv8IKU0uz5DkuBIUKINM97dwBLpJQ24FrgsJTyA8933gosBm4+h2spSqtUwCi+Jg34tad5rE4IUQekAIkAQoi7WzSf1QED0WobxxS1cs6jxx5IKS2ehyGnuf7pjk0Ealq8drprHZOCVts6X83nllI2Al8B0z0vTQfmex6nAaNO+ve6A+hxAddWFABUB6Dia4qAv0gp/3LyG56/4N8FJgE/SSldQojtQMvmLm8NqywDooQQwS1CJuUMxxcBGad5zwwEt3jeWhic/D0WAs8LIdYBQcCaFtf5TkqZc6bCK8r5UDUYpTvzE0IEtrgZ0ALkYSHEKKExCiGuEUKEAka0X7yVAEKIe9FqMF4npSwENqMNHPAXQowGrjvDR94H7hVCTBJC6IQQSUKIfp73tgPThRB+QojhtK05azlabeXPwMdSSrfn9S+BPkKIuzzn8xNCjBBC9D+f76koLamAUbqz5UBTi9sLUsrNaP0wbwC1QD5wD4CUMg/4P+AnoBwYBPzYgeW9AxgNVAP/DXyM1j90CinlRuBe4GWgHvgOLSAAnkOr3dSiDVRYcLYLe/pblgCTWx7vaT6bgtZsVorWxPc/QEArp1GUc6ImWipKJxFCfAzslVI+39llURRvUDUYRekgnqanDE+T11TgeuCzzi6XoniL6uRXlI7TA62ZKhooBh6RUm7r3CIpiveoJjJFURTFK1QTmaIoiuIVPtNEFhMTI9PT0zu7GIqiKN3Kli1bqqSUsd44t88ETHp6Ops3b+7sYiiKonQrQohCb51bNZEpiqIoXqECRlEURfEKFTCKoiiKV6iAURRFUbxCBYyiKIriFSpgFEVRFK/wasAIIaZ69vjOF0L8oZX304QQ3wohdnr2Nk9u8d5MIcQBz22mN8upKIqitD+vzYMRQuiBN4EctHWXNgkhlnmWTD/mH8AcKeVsIcRE4G/AXUKIKOB5YDja/h1bPJ+t9VZ5FUVRvMEt3VQ1VVFqKqXUVEqZuQx/vT+JxkQSQhJINCYSHhCOEOLsJ+tmvDnRciSQL6UsABBCfIS2emzLgMkCnvY8XsPxlWWvBFZJKWs8n10FTEXblU9RFKXLcLgdVFgqmgOk1FxKmams+fFR81EcbscZzxFkCCIpJIkEYwKJIYnN94khiSQaE4kOikYnul+PhjcDJokT9xwvBkaddMwO4CbgVWAaECqEiD7NZ5O8V1RFUZQzq7RUsq54HSWmkuMhYi6lwlKBu3mDUE1sUCwJIQkMiB7A5LTJJBmTmmsrCSEJOFwOSs2lJ9Rqjt3vqNxBg73hhPP56fxIMCaQEJLQHEQ9jD0I8Qsh0BBIkCGo+T5IH9T8PEAf0Kk1I28GTGvf6uSlm38DvCGEuAdYB5QAzjZ+FiHEg8CDAKmpqRdSVkVRlFbV2+r5YPcHzN8zH6vLil7oiQ+OJyEkgRHxI5prGsdqHT2MPQjQn2VDUD+ICIwgKzqr1bfNDnNz4JSYSprDrMxUxrridVQ1VbWp7AJxPHgMQQTqTwyjQEPguf5znBNvBkwxkNLieTLalqzNpJSlwI0AQogQ4CYpZb0QohgYf9Jn1558ASnlO8A7AMOHD1f7DiiK0m6anE3M3zOfWbtnYbKbuKrnVdw/6H56hvfEoPPuMo5GPyO9I3vTO7J3q+/bXDYqzBVYnBaanE00OZuwOq1YXdbm58deO917NdYampxNXv0e3vxX2gT0FkL0RKuZTAdub3mAECIGqJFSuoH/B8zyvPUN8FchRKTn+RTP+4qiKF7lcDlYfGAxb+98m6qmKq5IvoInhj5B36i+nV20ZgH6AFLCUs5+YBuIVhuM2ofXAkZK6RRCPI4WFnpglpQyVwjxZ2CzlHIZWi3lb0IIidZE9pjnszVCiP9CCymAPx/r8FcURfEGt3Sz/NBy3tz2JsWmYobFDeOl8S8xNG5oZxet2/KZHS2HDx8u1XL9iqKcKykl64rX8dq219hfu5++kX15athTjEsa55NDh08mhNgipRzujXP7zH4wiqIo52pL+RZe3foq2yq2kRKawouXv8iV6Vd2yyHBXZEKGEVRLjp7a/by6tZX+aHkB+KC4nju0ueY1nsafjq/zi6aT1EBoyjKReNIwxHe2PYGXx/+mjD/MJ6+5Glm9JtBkCGos4vmk1TAKIri88rN5by9822WHliKn96PBwY9wD0D7yHMP6yzi+bTVMAoitLtmeymE2bXnzxDvtpajUFn4Ja+t/Bg9oPEBMV0dpEvCipgFOUiJ6WkxlpDQX0BBXUFFDYWMixuGJNSJ3WJUVRSSupsdc0BUmIqOWFplRJTCY32xhM+E6AP0JZWMSYwPmU8SSFJXNXzKpJDk09zlTNzlJVRu2ABls1b0AUHowsNRR8a6rkPQRcSii40pMVrnvuQEHQhIQjDxfmr9uL81opyEZJSUm4pp6CugIL6Ag7WH2x+XGeraz7OIAzMzZvLgOgBPDXsKUYnju6U8m4t38q7u95lS/mWU2acG/2MzUuzDIkdoi3V4lnrKzEkkejA6AsORyklTdu3UzNnDo0rVyHdkrLkTPxlI4H2w/jbLBiaLOjstrOeSwQHe0InBH1IKCIgAKHXgd6A0OlAr0fo9dr9Cc91CJ0eDHrtXq9D6A3aZ3V6cLuQThfS7YJj9y4X0uUGlxPpciNdTnC5ka5j7514700qYBTlPDXYG/ih+Ad0Oh3D44d3mWYXt3RTYio5HiR1B7XaSX0BZoe5+bjwgHAywjOYnDaZXuG9yAjPoFdEL2KCYvji4Be8teMtHlz1IKN6jOKpYU8xKHaQ18supWTj0Y28vfNtNh3dRFRgFNMyp5ESmnJCgIT5h3mtdiXtdhq+WUnN3LlYd+7EHhjM15mXsyR1NH5JSTQ5XNRa7BybQmhwOzE6rAQ7rBidVmKwE693EiMcREkbEW47YW4bIU4rQXYrgTYzOlMTwuUGTyDgdiPcbs9j7Tkul/baKc89x0k36HRInRY2UqdDnnTv1umRQodbp8Pd2r2Xh2OriZaKcg5qrbWsKVrDqsJV/Fz2M063s/m99LB0Lom/hEviL2F4/HASQhK8Xh6r08r+2v3kVueSW5XLvtp9HK4/jNVlbT4mJiimOTx6hfciIyKDXuG9iAqMOuMvabvLzqJ9i3h317vUWGuYlDqJJ4Y+QUZERrt/DyklP5T8wDs732F75XbiguK4d+C93NTnpg4b4eWsqaFu0SJq5i/AVVlJdVQPFiaP5vueI5k8LJ2ZY9IZnBIBgMstqbPYqbXYqTbZqTHbqTZr9y1v1WY7tZ7Hdpf7LCXwPoNOYNAL/HQ6DHqBQa9j87M5XptoqQJGUc6iqqmKbwu/ZVXhKjaXb8YlXSSFJJGTlsPktMnohZ7NRzezpXwLWyq2NPcHJBoTGd5jeHPgpISmXNBf3Q6Xg/11+8mtyiWvOo/c6lzya/NxSi3kogKj6B/VvzlAMiIy6Bnek/CA8Av6/maHmTl5c5idO5smZxPX9bqOR4c8SmJI4gWdF7RgWVO0hnd2vkNudS4JxgTuG3gfN/S+4ewrErcT67791MydQ/2yL8BuJzcpi49SR1Ocmc0dl/ZkxqhUYkIurCxSSkw2Z3Po2J3tEzZ+eoFBp8NPr9Me63UYdOKE58ePEa3+/HlzJr8KGEVpxVHzUVYXrmZV4Sq2VWxDIkkPS28Olf5R/Vv9z+pyu8ivy2dzuSdwyrdQY9WW0YsNimV4vCdwegynV3iv0waOw+2goK6guWaSW53L/tr9zRtXhQeEMyB6wPFbzADig+O92ilfa63l/V3vs3DvQiSSW/veygODHiA6KPqcz+WWblYVruKdne+wv3Y/ySHJPJD9ANf1ug4/vfcnO0qXC9N331EzZy6Wn3/G6efPt6nDWZw+lviB/Zg5Jp2pA3vgp/f9Gf0qYNpABYxyoYoailh1ZBWrC1ezq2oXAL0je5OTmkNOWg4ZERnn/AtcSsmh+kNsLt+shc7RLVQ0VQAQGRDJsPhhDI8fTv/o/hQ3FmuBUp3Lvpp92Fxa53GoXyhZ0VlkxWQ1B0pSSFKnjfA6aj7Kv3b8i6X5SwnQB3B31t3MHDCTUP/Qs37W6Xay4vAK3t35LgX1BaSHpfNg9oNc1fMqry+BD+AymahfsoSaufNwFBXREBrJpymj+TZzNJNG9GbmmHQGJl1Yja+7UQHTBipglPNRUFfAqsJVrD6ymr01ewHIis7Saiqpk0kPT2/X60kpKW4sPh445VsoMZU0vx9sCKZ/dP8TaiYpoSldcm2sQ/WHeGPbG6wsXEl4QDj3D7yf6f2mt7qJlcPt4MuDX/Lervc40niEzIhMHsp+iJy0HPQ6vdfLai8spGb+fOoWL0GazeTH9WJR2hgO9RvBjDG9mDEylSijv9fL0RWpgGkDFTDK2TjdTg7VH2pudtp0dBMH6w8CMDh2MDlpOUxKnXTecyXOV5mpjH21+0gNTSUtLK1DfuG2p9zqXF7b+hrrS9cTFxzHI4Mf4YbMGzDoDNhddj7L/4xZu2dRYiqhf1R/Hhr8EBNSJng1NB0lJZg3bcLiuTmOFOHW6fk+eTBLeo4jctgQ7hmTTk5WPIaLoBnsTFTAtIEKGKUlt3RzuOHwCR3ie2v2Ns+nMPoZGRgzkAkpE5icOpl4Y3wnl7j723R0E69sfYWdlTtJD0tnSvoUPs//nHJLOdmx2TyU/RCXJV3W7k17UkocxcVYNrYIlBKtVmgNMpIXk8Gm8HQ2pA/litFZzByTTv8EtUTMMSpg2kAFzMVLSklRY9EJHeJ7avY0z/kIMgTRL6pfc5PTgOgBpIWldclmp+7u2Kiw17e9Tn5dPpfEX8JD2Q9xacKl7RYsUkochYXHaygbN+E8ehQAqzGMvbEZ/BSayq6YDBoSUrk0M5ZxmTFcNbAHEcEX0AxWlQ/VByBxGIT6zh8kKmDaQAWM90gpKTOXnTDnozM53U4O1h9kd9Vucqu1GsqxocH+On/6RfUjKzqrOUw6Yg915UQut4vKpkp6GHtc8LmklNgPHTpeQ9m4EWdlJQC20HD2x/fme2MqO6J7URWVwIheMYzNjGZsZgz9e4Sh011AsNUVQe4S2L0YynYcfz2qF6SOhtRLtfvoTOgCy+qcMykROp3acEzpeA63g5WHVzInbw551XmdXZxTGHQG+kT2YWr61ObaSUZEhtrTowvQ6/TnHS7HA2Uj5g0bsGzajKuqCgBbRBT5CX1Yl3IFWyN7URYWx+CUCMZlxnBHZgxDUyMIMFxgH1ZjOeR9roVK0c/aa0mXwJV/hYQhULoVjvwM+1fA9vna+8Exx8MmdTQkZEMHDLcGQEqwNYK1DprqPPe1LR6f6b7eq0VTNRjlFA32BhbvX8z8PfMpt5STHpbOrX1vJSIgorOLBoAQgp5hPekd2Rt//cU58seXNDd5bdiIZePGE2oo9shoCpL7sS4ohQ1haZQaY+gdH8rYzBjGZcYwqlcUoYHt8IvcUgN7vtBC5fD32jIscQNg4I0w8CaI6tlawaHqABz5SQucIz9B7SHtPb9gLZTSxmjBkzwCAs4+jLuZ0wbmSjCVg6nCc3/sebnnvQotSKz1IM+wppjQQ1AEBEa0ei8mP6+ayM5GBcyFK24sZv6e+Sw5sASL08LIHiOZOWAm45LGqf6Ki4iU0qtzbKSUOIqKPDUUT6CUlwNgDYtkf2JffghJY3NET8qM0fQID9ICpXc0YzJiiA87dRj0ebE1wr6vtVDJ/xbcDq3pa+DNWrDE9T/3czYePR42R36Co7u0sBI66DHoeLOaf6gnKCpaBEiLx9a61s8fFAnGOAjx3IIiTxsczff+IWdsvlN9MG2gAub87ajcwZzcOaw+shodOqb2nMrdWXfTP/o8/oMp3YJ0uXAePYq9sNBzO9L82FFUhPDzwxAbiyE2Fn1sTPNjQ4zn3vOaPiJCW/33LOzFJVg2bGhu9jrWKW8JCScvLpP1oensjM2kPDSWrKRwhqVGMjQ1gmGpkSRHBrVf4Dma4MBKLVT2fwNOK4Qlw8BpWk0lYUj79qXYGqF4kxY6heuheDOctDI0/qEQEgsh8Z7giG8RIvHHw8QYC4b2Xz5HBUwbqIA5Ny63i38X/ZvZubPZUbmDUP9QbulzC7f3u10N2fUR0u0+Y4hIu735WBEYiF9qKu6EJEzR8UinC7/6GvR1tehqqpA11WA2n3oRgwFDdHSLAIppDiDh54dly1ZMGzbgKi0FwBwUyo6YDLZG9mRnbCaW+GSGpUcxLDWSYakRZCdHEOTfzvOAXA44uAZ2fwp7vwK7SftlPcATKskjoQ0h2W5lOboTXM7jweFv7Jhrn4YKmDZQAdM2FoeFpflLmZs3lxJTCckhydyZdSfTMqcR7Bfc2cVTzpN0uTD/+CPmnzd4QuQwjiMnhUhAQHOINMYkUBEWR2FQFHt0EeTZ/Thcaz3jIowBThtRtkZi7Y3EOc3E2U3E2huJtDYSaW0gzFJPqKWBYHMDAu33iinAyPaoXuyMzWBXbCbGPr0ZlhbFsDStdpIaFey95riynbDjI9i1SOuzCAyH/r/QQiX9MtCrMU7g3YBR/8IXiaPmoyzYu4BP931Ko6ORIbFD+M3w3zAhZUK3mzmuHOc4epS6xYupW7wYZ2kZwt8fv5QUXInJWAaNoDw8jiMBUew1hLPbHkDhsRBxA3UQYNKRFh1AeqyR8f17kBYdTHq0EYNO0ORw0WR3affHHnueW+wurA4Xhx0u8jyPLZ737TY7+oZ6/J024vr2Ylh6NDNSI/h7cgTGAC//ymk8Crs+0YKlfDfo/KDvVMieDr2ngEENCulIKmB8kFu6abA1UGOrocJSwWf5n/HNoW9w42Zy6mTuHnA3g2MHd3YxlfMknU5M69ZRt+gTTOvWaZtVjRjFjxNnsMi/J4frHdreIxKog0A/HenRwfSKC2ZiVg/So42kRQfTM8ZIfGjghc0T6QocTVrT146FcPDfWqd60nC4+h9abSU4qrNLeNFSAdMNONwO6qx11FhrqLXVUmv13DyPa6w11FprqbNpx9Tb6nG1GLYYbAhmer/p3Jl1J0khSZ34TZQL4Sgp0Worny7GWVGBPiYG0423MzdqMMsqBH5NgitSwpk4KJT06GDSY4ykRxuJCw3o/iFyMim1UVo7FkLuZ2Br0Drrxz0Ng2dATO/OLqGCCpguqcZaw3u73mNd8TpqrDXNs9RPJhCEB4QTGRhJZEAkaWFpDIkbQmRAJFGBUc2vD4od1Kal1JWuRzocNK5ZQ90nn2L+4QcAAsaMZfctD/K6JZ6CWhtxTQE8k5PGjJGpxIZ2zCZdnaamAHZ8rAVLXSH4GSHrehgyA9LGdVxnvdImXg0YIcRU4FVAD7wnpfz7Se+nArOBCM8xf5BSLhdC+AHvAcM8ZZwjpfybN8sKcKD2AHaXnQExA7x9qVaZ7CZm581mTu4crC4rlydfzpjEMUQGRhIVEEVEYIQWHAGRRAZGEh4QrpZA8VH2I0eo++RT6pYuxVVVhSE+HnH3fSztMZS5h+xYSlxckhbM61OzfH9jrKY6yPtM61c58hMgoNcVMOE/of+1nT4KSzk9r/12EkLogTeBHKAY2CSEWCalbLnmyLPAIinlW0KILGA5kA7cAgRIKQcJIYKBPCHEQinlYW+V1+KwcN8391Frq2VI7BBmDpjZYR3gNpeNj/Z+xHu73qPOVkdOWg6PD32cXuG9vH5tpeuQdjuN335L7aJFWH76GXQ6jFdcQeHoHN6xxvN9QS3+Zhu/GJzIPV19Y6zK/VD4ozbPxGXXhue6HNpjd4vHbXm9PBdcNojpC5NfgEG3Qrhq6u0OvPnn70ggX0pZACCE+Ai4HmgZMBI4tm52OFDa4nWjEMIABAF2oMGLZWXh3oXU2mq5d8C9rCxcydNrn/b6EF6n28nn+Z/z1o63KLeUMzphNE8Ne6rTalBKx3Lb7dgPH8Z+8CBN23dQ/8UXuGpq8EtMJPTRx1mdNpz395kpymsiIbyJ317Zl+kjUoi+wP3hvUJKLQj2LNPW8arc2/pxOgPo/bV1unR+xx/r/U593RAAASEw/F7Ivg0Sh3bPBSUvYl6bByOEuBmYKqW83/P8LmCUlPLxFsckACuBSMAITJZSbvE0kc0FJgHBwNNSyndaucaDwIMAqamplxQWFp5XWc0OM1MXT2VgzEDemvwWTreTfx/5N3Py5nhlEuKx/cjf2PYGhxsOkx2TzVPDnmJkwsgLPrfS9bhtNuyHDmHLP4jtYD72/Hxs+QexHzkCLs9gDIOB0AkTMOVcyxx7PEt3lNHkcDGyZxT3jElnSlfcGEtKbeHHvGVasNQUaEuipI3V5pv0ztGWKmkZJCogupzuOg+mtZ+kk9NsBvChlPL/hBCjgblCiIFotR8XkIgWPt8LIVYfqw01n0wLnXdAm2h5vgVduHchdbY6Hh38KKCt0jslfQpT0qewvWI7c/Lm8GHuh8zJncPUnlOZOWAm/aL6nfN1pJT8VPoTr257lbzqPDLCM3hlwitMTJnYafurX4xcJhO1CxZi/vFHdEYj+rAw9OFh6MLC0IeFow8PQx8erj0PD9feDwtD+J95DoW7qQlbQQH2gwc9YXIQW/4BHEXF4PZMYNTr8U9NJSAzk9CpVxKQkYkjJZ1NzhBmbynl5/U1BBhKuWFIEjPHpJOV2MU2xnK7oWiDFih7voD6Iq1W0vNyGPMk9LtWW/ZEUfBuwBQDKS2eJ3O8CeyY+4CpAFLKn4QQgUAMcDuwQkrpACqEED8Cw4EC2pnZYebD3A+5LOkyBsUOOuX9IXFDGBI3hKLGIhbsWcDiA4v5suDLc14IckflDl7d+iqbjm4i0ZjIX8b9hWt6XqMmOXYgZ20ttXPnUTNvHu6GBgKy+uOqq8O6Zw/u+nrcFssZPy+CgprDRhfuCaOwMFx1ddgOHsRRXKz9VQ9gMOCfnkZgv/6EX3MtAb0z8c/IwJCaxuEGO+sL69h6pJYt+2vJ//EwUkJSRBB/uKoftw1PIbIr7Q/vcmr9KcdCxVQO+gDImAgT/gP6TFVzTZRWebOJzADsR2vmKgE2AbdLKXNbHPM18LGU8kMhRH/gWyAJ+B3QD/glWhPZJmC6lHLn6a53vkvFvLfrPV7d+ioLr1nIwJiBZz3+5KXse4b35K6su7iu13UEGk5d5fVA7QFe3/Y6a4rWEBUYxYPZD3JLn1vUMvMdyFlZSfWHH1K78COkxUJozmTMN9/Jj/o44sICSIoIIikyiJgAHe7GRlz1Dbgb6nE1NOCqb8DVUI+7+bHneZ3n/YYG9KEh+GdmEpCZSUBGJgGZGfinpSH8/Gi0OthRVK+FSWEt247U0mDVNm4LD/JjaGoEl6RGcklaJCN7RnWdZjCnHQ59p/Wn7FsOlmptCfrMydqw4N5TILCL1a6U89Jt1yITQlwNvII2BHmWlPIvQog/A5ullMs8I8feBULQms9+J6VcKYQIAT4AstCa2j6QUv7vma51PgFjspuYumQqg2MH8+akN8/ps8c245qdO5s9NXuIDIjktn63cVvf24gJiqG4sZh/bv8nXxZ8idHPyD0D7uGurLvUel8dyFFaSvX7s6j79FOkw0HY1VfjuO0uXjvkZum2Ek7+0Q8w6JrDJilCuyVHBZEUEUxSZBDxoQFnDAApJYeqzGw9UtccJvvKG5FS63roHRfCJWmRDE2NZFhqJL1ijJ0/AdLRpO3aWHdEm1dSd0TrSyn4Dmz12kq/fadqfSqZk8Ff/fz6mm4bMB3pfALm3Z3v8tq219pce2mNlJLN5ZuZkzuHtcVr8df5MyJhBBvKNqAXem7vdzu/HPhLIgK7xmZdFwP74cNUvfsu9Z8vAyEIv/4X6G6fyVsHbCzYeAQhBDNHpzFzTDqNVicltU0U11ooqWvSbrVNFNc2UW22n3BevU7QIyyQ5EgthJI9YVRlsjcHSq3FAUBooMETJNqijkNSIwhrj42xzlVrAdLyZq448XidH4Qnaxtl9f8FZEzwyhLxStehAqYNzjVgTHYTVy6+kqFxQ3lj0hvtUoZD9YeYlzePb498y4TUCTyc/bBa+r4DWffvp/rtd2j4+muEnx8RN9+M/513M2t/E7N+PITN6ebW4ck8Oak3CeFBZz1fk911UuhYmh+X1DVxtMHaXAvKiDVySZpWMxk+gp/VAAAgAElEQVSWFklmbEjH106KNmprcp0tQCJSICK1xS3t+OOQeFD9ghcVFTBtcK4B887Od3h92+t8dM1Hat5JN9e0azdVb/8L0+pv0QUHEzFjOsF33M38AybeWnuQ+iYH12Yn8ExOH3rFhrTbde1ON0frrYQFGYgI7sQ+tYZSWPVHbRVhvT+EHwuQlFYCpIdaTkU5QXcdptxlNdobmZ07m/HJ41W4dGOWzZup+tfbmH/4AV1YGDGPPkroHXewOL+R1z7cRXmDjSv6xPLbK/t6Zda7v0FHanQn9kk4rPDT6/D9S+B2wWW/0RZ7DGi/EFWUC3FRBsyCPQtosDfw8JCHO7soyllItxu3pQm32YTbbMZtMuE4epTaOXOxbN6MPiqK2F8/Q/ht01le0MDLs3ZwuNrCJWmRvDZ9KKN6RXf2V2h/UmpNYd/8h9av0u9amPLfENWzs0umKCe46AKm0d7I7LzZjE8Zz4BoVXvpKFJKrHl52A4cwG0yN4eF22zGbTbhMpuPv97yPYuFU4Z7AYb4eOL/4z8Iv/km1h0x8eIH29lT1kC/HqG8P3M4E/vF+ebk1Yq9sOL3ULAWYvvDXZ9pHfGK0gVddAEzf898Gu2NPDL4kc4uykXBVnCIhq++ouGrr7AfPnzimwYDeqMRXUgIOs+9PiICv+QkbYa90fP6CccY0YeGEjhoEFtKTbw4ZzubDteSGhXMK7cN4ReDEzt/6K83NNXC2r/Dxne1JrCrXoTh96ltf5Uu7aL66Wy0NzInbw4TUiaQFZ3V2cXxWY6yMhqWf039V19iy9sDQhA8ciRRv7wX48iR6MLC0BmNCH//NtcypJTYnG4amhwU1Tbx5oId/HtvBbGhAfzX9QO4bUQq/gYf7Lx2u2DrbPj2v7SQGX4vTHgWjD7Y9Kf4nIsqYObtmadqL17irK2l8ZtvqP/yS5o2bwEgcNAg4v7we8Kuugq/+HicLjfljTYamhw01JlotDppsDpoaHK0eOyk0ea5tzposDqb37e73M3XCws08LupfblnTDrB/j76Y1y4Hr7+HRzdBalj4Kr/gYTszi6VorSZj/7PPFWDvYG5eXOZkDKB/tH9O7s4PsFlMmP6djX1X32Fef1P4HTin5FB7FNPEnb11finpQFQVGNh4Yq9LNpcRJXJftrzBfnpCQ00EBbkR1igNvQ3NdqovRboR1iQgdBAPyKC/Li8dyzhwZ0wcbEj1BfDyucgd4m2DfDNs2DAjWolYqXbuWgCZn6e6ntpD26bDdO6dTR8tRzTmjVImw1DYgLR995D2DXXENC3L0IIXG7J6rxy5m8oZO3+SgQwsV88k/rHERHkR1iQX4vg0B779K6MbeFogvWeYcdIuOL3MPZXankWpdu6KALmWO1lYspEVXs5D9LhwLxxIw1fLadx1SrcjY3oo6KIuOkmwq69hqAhQxCeyXvlDVY+3lTERxuPUFpvJS40gCcm9mb6iBQSI84+e/6iJKW2UvE3z0L9EW0xyZz/gsi0zi6ZolyQiyJg5uXNo9HRyCNDVO2lrVx1dZi+/wHTmjWYvv8ed2MjupAQQidPJuzaazFeOgph0H583G7JDwcqmf/zEVbtKcflllzWO4Y/XjeASf3jVM3kTIo3w8pntb3m4wbAzC+0vVUUxQf4fMA02BuYlzePSamTzmuTsIuJreCQFihr1mDZtg1cLvTR0YROySF0wgSMl12GLuD4wofVJhufbilmwcYjFFZbiDL6c/9lPZkxIpX0GGMnfpM2cDm1JehDO2mtuJpD8O2ftX4WYxxc+zIMvVsNO1Z8is//NM/Nm6vVXlTfyymkw4Fl67bmULF7tpwO6NuX6AfuJ3TCBAIHDWpu/gJtuPCmw7XM31DI17uOYne5Gdkzimdy+jB1YA8CDN1gocTaw/DJvdp2v2lj4dJHoe9VHbPIo6UGvv8/2PC2thPk5b+DsU9CQKj3r60oHcynA6beVs+8vHlMTp1M36i+nV2cLsFVX39i01dDA8LPj+BRo4i8+y5Cx4/HLynplM81WB0s3VrC/A2F7C83ERpo4PZRqdwxKpXe8d3ol+OeL+Czx7THY56E3M/g4zu0BSFHPQxD7/TORlpOmzZJct3/grUeht4BE/4TwhLb/1qK0kX4dMDMzZuLyWHi4cEX95pj9sOHaVyzVmv62rLleNPX5MmETBhPyJgx6IynNmkV1VhYu7+S7/ZV8EN+FVaHm8EpEbx4czbXZScS5N8NaivHOG3a0N+Nb0PiMLjlA4hMh0nPw76v4Oe34Jv/B2v+qoXMqIfaZ20vKbVmsNV/0tYNy5gEOX+GHue3/5CidCc+GzD1tnrm75lPTlrORVd7kVJizc2jcdUqGletwl5QAEBAnz5E338/oRPGE5idfULTF4DV4WLjoRrW7qtk7f4KCirNAKREBXHb8BRuGZ7ilVWJva7mEHxyD5Rt15rDJv8JDJ7l9fUGbdRW1vVQslULmk3vwoZ/Qb9r4NJHtGa085mDUrhe68Av2QLxA+HOJZA5qV2/mqJ0ZT4bMHPy5mBymHgo+6HOLkqHkC4XTdu307hSCxVHaSno9RhHjSTy9tsJGT8e/+RTm74Kq82s3VfJd/sr+elgNU0OF/4GHZf2iubOUWmM7xtLzxhj9104MvczWPaEFhDTF2ihcTpJw+Cmd7Uaxqb3YPMs2Psl9MjWgmngjW3b3bEqH1Y/r302NAGu/ycMnq428lIuOj654Vi9rZ4rF1/JmMQxvDT+pU4umfccm5/SuGoVjau/xVVVhfD3xzh2LKE5OYRMGI8hMvKEz1gdLn4uqG4OlUNVWi0lPTqY8X3juKJvLJf2jO5ezV+tcVi12sOmdyFpuNYkFpF6buewW2DXIq1WU7lXG+014n4Y/ksIiT31eHOVtiDllg/AEAjjfgWXPqYmSipdmtpw7BzNzp2N2WH2yb4Xt9WKef16Gr9ZSePatbjr6xHBwYRccTlhOTkYL78CfciJ/SmHqsys3VfB2n2V/FxQjc3pJtBPx+he0cwcncb4vnFdf1jxuag+qDWJHd0Jox/X+lkM57HjpH8wXHIPDJsJBWvgp3/C2r9qo8Cyb4FRj2h9KY4m+Pmf8P3L4LBonxn/BwiJa+cvpijdi88FTJ21jgV7FzAlbQp9Ivt0dnHahctkxrzuOxpWrsK0bh3SYkEXFkbohAmEXjkF45gx6AIDT/hMUY2FpdtK+Gx7SXNfSq8YI7ePSmV83zhG9Ywi0K+b11Jas3sxLHtK61uZ8ZE2/PhCCQEZE7Vb5X6tf2b7Atg2D9Iv0/p4Goqh79Va/06sb/zcKcqF8rmAmZM3B4vD0u1rL26LhYYV39C4ciXm9euRdjv66GjCr7uO0Ck5GEeORPiduNhjg9XB8p1lLNlawsbDNQCM7hXNPWPSGd8nrnO39/U2h1UbBbZ5FiSP1BaIjEhp/+vE9oFrX4KJz2rL6G/+AEJ7wI1vQ/q49r+eonRjPhUwddY65u+Zz5T0KfSO7N3ZxTlvjpISih5+GNuBfAyJCUTOmE5oTg5BQ4ci9CfWOhwuN98fqGTx1hJW5ZVjd7rJiDXy2yv7csPQJJIuhvW/qvK1JrHyXTD2KZj4HOi9vNJycBSMe1q7KYrSKp8KmDl5c2hyNvFwdvetvTTt2kXRI48ibTZS3v4XxssvP2UEl5SS3NIGFm8tZtn2UqrNdqKM/tw+MpVpQ5PITg7vvqO+ztWuT+GLp0DvD7d/An2mdHaJFEXx8JmAcUkX8/fM58r0K8mMzOzs4pyXhlWrKP3t7zBER5My+0MCMjJOeL+svonPtpWyZGsxBypM+Ot1TM6KY9rQZK7oE+ubOzqejqMJvv691kyVcqnWJBZ+6jBsRVE6j88ETHVTNRHOiG4570VKSc2Hs6l48UWCsrNJ/uebGKK1LXHNNicrdh9lybZi1h+sRkoYnhbJX6cN4ppBCb676daZVO7XmsQqcmHcM9qSK2qRSEXpcnzmf2W1tZrb0m/rdrUX6XRy9C9/oW7hR4ROnUri3/+GLjCQ/IpG3lxzkBW7j9LkcJEaFcxTk3ozbWgSadE+NKS4JZcDTOXQeBQaSrX7xrJT7611EBwNdyyG3pM7u9SKopyGVwNGCDEVeBXQA+9JKf9+0vupwGwgwnPMH6SUyz3vZQNvA2GAGxghpbSe7lpSym43csxlMlHy9DOYv/+e6AceIPbpXyF0OrYU1nLvBxuRwLRhSdw4NIlL0iK7f7+KuUrb/6SxrMWtRXiYK0/9jM4AIT20kVoxmdDzMm2ByOzb1EKRitLFeS1ghBB64E0gBygGNgkhlkkp81oc9iywSEr5lhAiC1gOpAshDMA84C4p5Q4hRDTgONP1UsNSyYjIONMhXYqjrIyihx/Blp9Pjz//ichbbwXg+wOVPDhnC/FhAcy9bxQpUT4wtFhK2LkIvv6ttpIwAEKbiBjaA8KSIOkSbVmV5lsP7T44GnQXUd+SovgQb9ZgRgL5UsoCACHER8D1QMuAkWg1FIBwoNTzeAqwU0q5A0BKWX22i4X4hbRTsb2vKTeX4ocfwd3URMo7bxMydiwAK3aX8eTC7fSKNTLnvpHEhQae5UzdgKkCvviVtmJxyihtVn1kOoTEq34TRfFx3vwfngQUtXheDIw66ZgXgJVCiCcAI3CsQb0PIIUQ3wCxwEdSyhdPvoAQ4kHgQYDU1HNcZ6qTNP57DSW//jX6yAjS3p9PYB9t1vcnm4v4/eKdDEmJ4IN7RvpG5/3uxfDVb8Bu1vaYH/2YWvBRUS4i3mx7aK3D4OSVNWcAH0opk4GrgblCCB1a8I0D7vDcTxNCnLLOuZTyHSnlcCnl8NjYVhYf7GJq5syl+LHHCMjIoOfHHzeHy6wfDvHbT3cyNjOGefeP6v7hYq6GRTPh019qtZWH1mm7NqpwUZSLijdrMMVAy7U6kjneBHbMfcBUACnlT0KIQCDG89nvpJRVAEKI5cAw4FsvltdrpNNJ+d/+Tu38+YTmTCbxxRfRBQUhpeSV1Qd49dsDTB3Qg1dnDOkeWw6fyZ4v4MunoalOm1E/9leqKUxRLlLerMFsAnoLIXoKIfyB6cCyk445AkwCEEL0BwKBSuAbIFsIEezp8L+CE/tuug232UzxY49TO38+UffeS9Irr6ALCsLtlvzpizxe/fYAt1ySzBu3D+3e4WKpgcUPwMd3ap3zD30Hl/9GhYuiXMS89r9fSukUQjyOFhZ6YJaUMlcI8Wdgs5RyGfBr4F0hxNNozWf3SG2DmlohxEtoISWB5VLKr7xVVm9xlJdrI8X276fHC88TOX06AE6Xm98v3sXircXcN64n/3l1f3S6bjwEef83sOxJsFTBFX/wBEs3b+ZTFOWC+eSGY12Bdc8eih5+BLfJRNIrLxNy2WXa6w4XTy7cxsq8cp7J6cMTEzO77/wWaz2s+A/YPg/isuCGtyBxSGeXSlGUc6A2HOtmGteupeSZX6MPCyNtwXwC+/YFtGVfHpy7mR/zq3nhuizuGduzk0t6AfK/1bYibiyDy34NV/y+bdsJK4py0VAB044cFRVUvvYa9YuXENi/P8lvvYVfvLarYZ3Fzr0fbmJncT3/d8tgbrokuZNLe55sjbDyOW1b4Jg+cN9qSL6ks0ulKEoXpAKmHbgtFqo/+IDq92chHQ6i7r6b2CefQGfU1gyraLBy1/sbOVRl5q07hjFlQI9OLvF5OvQ9fP4o1BXBmCe0RSb9LoL9ZhRFOS8qYC6AdLup/3wZlS+/jLOigtApU4j79TP4p6U1H1NUY+HO9zdQ2Wjjw3tHMCYzphNLfB6a6qBqP+z6BDa+A1G94JcrIPXSzi6ZoihdnAqY82T+eQPlL/4Ptrw9BGZnk/TySwRfcmJT0YHyRu58fwNWh5v5949iaGpkJ5W2DczVULnXc9t3/N509Pgxox7Wlnrx94H10RRF8ToVMOfIVlBAxf/+A9OaNRgSE0j8xz8Iu/oqxEkLMu4oquOeDzbip9ex6KHR9O0R2kklbkFKbW2wk0Okcq82xPgY/xCI7QuZk7T7mL4QP8A7e9wriuKzVMC0kbOmhqo33qT244/RBQYS+8wzRN19F7rAUxek/OlgNffP3kRUiD/z7hvVefu3OO2waxEUbTgeJM2rGQOB4RDbD/pdrd3H9tXuw5Kguw6dVhSly2hTwAghpgH/llLWe55HAOOllJ95s3Bdgdtmo3buXKr+9TbupiYibr2F2Mcfb95xsiUpJYs2F/Hc57mkRQUz7/5RxId1worIUkLeZ7D6T1B7SFvyPrY/DLz5xCAJiVNBoiiK17S1BvO8lHLpsSdSyjohxPOAzwaMlJKG5cupfOllHCUlhFxxBXG/+y0BGa3vOWO2OXn2s90s3VbCuMwYXp8xlEijfweXGjj8A6z6I5Rs0SY/3v4J9M5RQaIoSodra8C0tmaZzzavWbZuo+J//oemHTsI6NeP1A9mYRw9+rTH7zvayKPzt3CoyswzOX14bEIm+o5e+qViD6x+AfavgNBEuP5NGDxDrWCsKEqnaWtIbPasDfYm2tpgTwBbvFaqTuIoKaH8f/9B44oVGGJjSfjLfxN+ww0Ifeu/pKWUfLKlmD9+vpvQQD/m3T+KMRkdPAy5oRTW/BW2z9c65yc9D5c+ouanKIrS6doaME8AzwEfo+3zshJ4zFuF6gxSSo48+BCO0lJiHnuM6F/e2zxRsjUWu9YktmRrCWMzo3nltqHEhnbgUinWevjxVfjpn+B2wqhHtEUmg6M6rgyKoihn0KaAkVKagT94uSydqmnrVuwHD5Lw178SceO0Mx67v7yRR+dv5WCliacn9+HxiR3YJOa0w+ZZ8N3/QFON1nE/6TltYy9FUZQu5IwBI4R4RUr5KyHEF5y6GyVSyl94rWQdrG7JEnTBwYRNvfKMx32yuYjnPt9NSIAf8+8b1XEz891uyFsK3/4Zag9Dz8sh58+QOLRjrq8oinKOzlaDmeu5/4e3C9KZ3BYLjV+vIPSqqeiCW5+lbrE7ee6zXBZvLWZ0r2henTGEuNAOGoJ8aJ02Mqx0G8QNgDsWa5Mg1cgwRVG6sDMGjJRyixBCDzwgpbyzg8rU4Rq+WYnbYiHixhtbff+Ap0ksv9LEk5N689Sk3h3TJFaeq40MO7BSm/x4w1uQfZsaGaYoSrdw1j4YKaVLCBErhPCXUto7olAdrX7JEvzT0ggaNuyU9xZvKebZz3ZjDNAz95ejGNe7g5rEfnhFC5eAMJj8Jxj1kBoZpihKt9LWUWSHgR+FEMsA87EXpZQveaNQHcl+5AiWTZuI/dWvTthZssnu4vllu1m0uZhLe0Xx2vShxHXUrPydn8Dq5yHrerj2FTUyTFGUbqmtAVPquemAY6s2+sRey3VLl4JOR/gN1ze/ll+hNYkdqDDx5MRMnpzUG4O+tbmmXlC4XttzJW0c3Piu2iVSUZRuq60Bkyel/KTlC0KIW7xQng4lXS7qP/sc49ix+PXQNgFbuq2Y/1y6myA/PbPvHcnlfWI7rkBV+fDR7RCRBrfNVeGiKEq31tY/y/9fG1/rVsw//4yzrKx53sv/frOXpz/ewcCkcJY/dVnHhou5GhbcAkIHdyxSzWKKonR7Z5sHcxVwNZAkhHitxVthgNObBesI9UuWogsPJ2TiRBqsDt79/hDXZifwym1DOq5JDMBh1Wou9SVwz5farpGKoijd3NmayEqBzcAvOHHtsUbgaW8VqiO4GhpoXL2aiJtuQhcQwMotxdidbn45rmfHhovbrfW5FP0Mt3wIKSM77tqKoihedLZ5MDuAHUKIBZ5jU6WU+zqkZF7WsHw50mYj3DP3ZdmOUlKighiaEtGxBVnzF9i9GCa/AAPOvESNoihKd9LWP9WnAtuBFQBCiCGeIcvdVt3iJQT06UPggCyqTTZ+zK/iuuzEE4Yqe922efD9P2DY3TD2Vx13XUVRlA7Q1oB5ARgJ1AFIKbcD6d4pkvfZDhzAumsX4TdOQwjB8t1Hcbkl1w1O7LhCFKyFL56CXhPgmpfUsi+KovictgaM89h2yb6gbslSMBgI/4W2VucX20vpHRdCvx6hZ/lkO6nYCx/fDTF94NbZoPfrmOsqiqJ0oLYGzG4hxO2AXgjRWwjxOrD+bB8SQkwVQuwTQuQLIU5Z7l8IkSqEWCOE2CaE2CmEuLqV901CiN+0sZxnJR0O6pctI3TCeAxRUZTVN7HxcA2/GNxBzWOmCm04sl8g3L4IAsO9f01FUZRO0NaAeQIYANiAhUADcMZOA88imW8CVwFZwAwhRNZJhz0LLJJSDgWmA/886f2Xga/bWMY2MX3/Pa7qasKnaZ37X+4oA+iY5jG7BRZOB3MVzPgIIlK8f01FUZRO0tYNxyzAf3pubTUSyJdSFgAIIT4CrgfyWp4abU4NQDjasGg8x98AFNBi7bP2ULdkCfqYGEIuvwzQRo9lJ4eTHnP63SvbhdsNSx+Ekq0wfT4knbqwpqIoii8520TLM44UO8uGY0lAUYvnxcCok455AVgphHgCMAKTPdc1Ar8HcoDTNo8JIR4EHgRITU09U1EBcFZXY1r7HVF3340wGDhUZWZXST3PXtP/rJ+9YKv/CHu+gCv/Bv2u8f71FEVROtnZajCj0UJiIbABOJdOitaOPXmBzBnAh1LK/xNCjAbmCiEGAn8CXpZSms7ULyKlfAd4B2D48OFnXXyzftkX4HQ2Lw3zxY5ShIBrs73cPLbpfVj/Oox8EC59xLvXUhRF6SLOFjA90GoRM4Dbga+AhVLK3Dacuxho2cmQTIsmMI/70ObYIKX8SQgRCMSg1XRuFkK8CEQAbiGEVUr5Rhuu2yopJfVLlhCYnU1AZiZSSpbtKGVEehQ9wr24DP+B1bD8t9D7Sq32ooYjK4pykThjJ7+U0iWlXCGlnAlcCuQDaz1NWmezCegthOgphPBH68Q/ucntCDAJQAjRHwgEKqWUl0kp06WU6cArwF8vJFwArLtzsR040Lxr5Z6yRvIrTPzCm537R3fBJzMhfgDcPAv0bV28WlEUpfs76288IUQAcA1aLSYdeA1YcrbPSSmdQojHgW8APTBLSpkrhPgzsFlKuQz4NfCuEOJptOaze6SUXtlnpn7pEkRAAGHXaCOhl+0oxaATXD0owRuXg4YyWHCbtiPl7R9DQIh3rqMoitJFna2TfzYwEG2o8J+klLvP5eRSyuXA8pNe+2OLx3nA2LOc44VzuWZr3DYb9V9+RWhODvrQUKSUfLGjlHG9Y4gy+l/o6U9lM8GCW8FaD79cAWEduEKAoihKF3G2GsxdaMOE+wBPtuhwF4CUUoad7oNdSePq1bgbGpo797ceqaOkrolncvq0/8XcLlh8H5TnajWXHoPa/xqKoijdwNlWU+7Adeu9p37JUgyJCQRfeimgjR4LMOiYMiC+/S+2/jXYvwKu/gf0zmn/8yuKonQTPhEgZ+IoK8O8fj0RN9yA0Olwutx8ubOMif3iCA1s5zXAju6Gf/8Fsq6HEfe377kVRVG6GZ8PmPrPPwcpCZ+mNY9tOFRDlcnW/kvDOG2w9CEIioRrXlbDkRVFuej59LhZKSV1S5cSPHIk/inalJxl20sJCTAwsV9c+15s7d+gfDfM+BiM0e17bkVRlG7Ip2swTVu24Cg8Qrinc9/mdPH17jKmZMUT6Kdvvwsd+Rl+fFXbOKzv1PY7r6IoSjfm0wFTt2QpOqORsClTAPh+fxUNVifXDWnH5jGbSWsaC0+GK//afudVFEXp5ny2icxtNtOwYgVhV1+FLjgY0CZXRgb7MS4zpv0utPJZqC2Ee5dDQAdtWKYoitIN+GwNpmHFN0iLpXlpGIvdyaq8cq4alICfvp2+9oFVsOUDGPMEpI1pn3MqiqL4CJ8NmLqlS/BPTydo6FAAVu+poMnhar+1xyw18PnjEJcFE85lmxxFUZSLg08GjL2wkKbNWwi/8cbmbZC/2FFKfFgAI9Kj2uciX/0aLNUw7W1t+2NFURTlBD4ZMHVLl4JOR/j11wNQ3+Tgu32VXJudiF7XDvNTdn0KuUtg/B8gIfvCz6coiuKDfC5gpMtF/WefYxw3Fr94ba7LN7uPYne526d5rKEUvnoGkkfA2F9d+PkURVF8lM8FjHn9TziPHm3u3Adt9FhadDDZyeEXdnIp4fPHwOXQmsbU/i6Koiin5XMBU790CfrwcEImTgSgstHG+oNVXJedyJm2X26Tze/DwX/DlP+C6Ix2KK2iKIrv8qmAcdXX07j6W8KuvRadv7bPy/JdZbgl/OJCJ1dWH4SVz0HGJBh+XzuUVlEUxbf5VMDUf/UV0m5vXhoGtOaxfj1C6RN/AZMgXU5Y+jDo/eD6N9RCloqiKG3gWwGzZCkB/foRmJUFQHGthS2FtRe+cvKPr0DxRrjmJbU7paIoShv5TMBIqxXr7t1E3Ditua/ly51lAFyXfQGhULZDWyl5wDQYeFN7FFVRFOWi4DMB46yrAz8/wq67rvm1ZdtLGZISQWp08Pmd1GHVmsaCY7Tai2oaUxRF+f/t3Xt0VdW1+PHvJAHCWxIoAkFAQSC8gqY2CD4oRsEqNr7QcosKba+2INb+flcs3hYr3quYOir3x8CiYoR6w8Nb+PFUBKlWCkKAACEECAgSwstQ3iSBZN4/9k48CUkg4exzwjnzM0ZGdtZ+zb3OyZlnrb332pctZBJM8fHjNLvzTiJbtgQg58hpsg6evLLusVWvwpEs57xLYz+NAGCMMWEiZBIMF4rLndxftDkPEbivT9vabW/fP+Af/wU3PwVdk/wUpDHGhI+QSTD1mjah6W23Ac6TLBdtziOxcwxtmtdinLDCU07XWMuOcPckP0dqjDHhIWQSTINOnZBI5876bXkn2fPtmdrf+/LJb+H4N87d+g2b+jFKY4wJHyGTYHwt3JxHZD1hSM9ra77yjo9h40wYMA6uS/R/cMYYEyZCbjCtkhJl8eY8br+xNbRypo4AABi2SURBVC2bNKjZymfyYeFYaNMLBv3WmwCNMTV2/vx5cnNzKSgoCHYoV62oqChiY2OpX79+wPYZcglmwzf/JO9EAf82pHvNV/7kRTj3T/jpfIhs6P/gjDG1kpubS7NmzejUqdOVjykYhlSV/Px8cnNz6dy5c8D262kXmYgMEZEdIpIjIuMrmX+diKwSkU0iskVE7nXLk0Rkg4hsdX//8HL3uTAjj4aR9bgrrk3Ngi08BdvmQ8IouLZXzdY1xniqoKCAmJgYSy61JCLExMQEvAXoWQtGRCKAqUASkAusF5GFqprls9hLwFxVnSYiccBSoBPwLXC/quaJSC/gE6D9pfZ5obiEpVsPclePNjRtWMND2/UpFBdBzx/XbD1jTEBYcrkywag/L1swtwA5qrpHVYuA2cADFZZRoLk73QLIA1DVTaqa55ZvA6JE5JJ9Vv/YnU/+maLa3VyZvdi5Y7/DD2q+rjHGmIt4mWDaA/t9/s7l4lbIROBfRCQXp/UytpLtPARsUtXCijNE5Bciki4i6UePHmXh5jyaNYzkzm6taxbphULYuRy6DYV6ETVb1xgTNubPn4+IkJ2dHexQrgpeJpjK2mNa4e/HgVRVjQXuBWaJSFlMItITeB3418p2oKrTVTVBVRNatWrNJ5mHuLvntUTVr2GS+PrvUHQKut9Xs/WMMWElLS2NgQMHMnv2bM/2UVxc7Nm2A83Lq8hygQ4+f8fidoH5GA0MAVDVNSISBbQCjohILDAfGKmquy+1s1MF5ykqvFC7myuzF0ODpnD9nTVf1xgTUC8v2kZW3km/bjOuXXN+f3/Papc5ffo0q1evZtWqVQwbNoyJEycCMHnyZGbNmkW9evUYOnQor732Gjk5OTz99NMcPXqUiIgI5s2bx/79+0lJSWHx4sUAjBkzhoSEBJ588kk6derEqFGjWL58OWPGjOHUqVNMnz6doqIiunTpwqxZs2jcuDGHDx/m6aefZs+ePQBMmzaNZcuW0apVK8aNGwfAhAkTaNOmDc8++6xf66g2vEww64GuItIZOAA8BvykwjLfAIOBVBHpAUQBR0XkGmAJ8KKqrr6cnR0/d57YJg0YcENMzaIsKYEdS6HLXVC/FsPKGGPCwoIFCxgyZAg33ngj0dHRbNy4kcOHD7NgwQK++uorGjduzLFjxwAYMWIE48ePJzk5mYKCAkpKSti/f3+124+KiuLLL78EID8/n5///OcAvPTSS7z33nuMHTuWZ599ljvuuIP58+dTXFzM6dOnadeuHQ8++CDjxo2jpKSE2bNns27dOm8r4zJ5lmBU9YKIjMG5AiwCmKGq20TkD0C6qi4EfgO8IyK/xuk+e1JV1V2vC/DvIvLv7ibvVtUjVe3vZMF57u19LZERNez1O5AOpw9b95gxV4lLtTS8kpaWxnPPPQfAY489RlpaGiUlJTz11FM0buw8EiQ6OppTp05x4MABkpOdwXejoi7vi+vw4cPLpjMzM3nppZc4fvw4p0+f5p577gHgs88+Y+bMmQBERETQokULWrRoQUxMDJs2beLw4cP069ePmJgaftH2iKc3WqrqUpyT975lv/OZzgIGVLLeJKBGo0yqwrC+l7yS+WLbF0G9+nDj3TVf1xgTFvLz8/nss8/IzMxERCguLkZEeOihhy66/Fe14qlmR2RkJCUlJWV/V7wnpUmTJmXTTz75JAsWLKBv376kpqbyt7/9rdr4fvazn5GamsqhQ4cYNWpUDY/OOyEzFlnLxg1I6NiyZiupOudfOt8GUS28CcwYc9X76KOPGDlyJPv27WPv3r3s37+fzp07Ex0dzYwZMzh79iwAx44do3nz5sTGxrJgwQIACgsLOXv2LB07diQrK4vCwkJOnDjBypUrq9zfqVOnaNu2LefPn+fDDz8sKx88eDDTpk0DnIsBTp50zkUlJyfz8ccfs379+rLWTl0QMgkmtmUj6tWr4Y1ER7Ph2B7rHjPGVCstLa2sy6vUQw89RF5eHsOGDSMhIYH4+HhSUlIAmDVrFlOmTKFPnz7ceuutHDp0iA4dOvDoo4/Sp08fRowYQb9+/arc3yuvvMIPfvADkpKS6N79u2Gv3nrrLVatWkXv3r25+eab2bZtGwANGjRg0KBBPProo0RE1J1bLaSq5tzVJiEhQdPT02u20udvwKpJ8Hw2NK/lg8mMMZ7bvn07PXr0CHYYdVZJSQk33XQT8+bNo2vXrlUuV1k9isgGVU3wIq6QacHUSvZiiP2+JRdjzFUrKyuLLl26MHjw4GqTSzCE3GjKl+34fjiYAXe9HOxIjDGm1uLi4srui6lrwrcFk73E+W3nX4wxxhNhnGAWQ+vu0KpLsCMxxpiQFJ4J5uwx2PcP6P6jYEdijDEhKzwTzM6PQYute8wYYzwUnglm+2Jo3h7aVX0dujHG+IqIiCA+Pp5evXrxyCOPlN1ceSXS09OrHZQyLy+Phx9++Ir3Eyzhl2CKzsLuz5zuMXtCnjHmMjVq1IiMjAwyMzNp0KABb7/9drn5qlpuKJjLkZCQwJQpU6qc365dOz766KNaxVsXhN9lyrtXwoVz1j1mzNVq2Xg4tNW/27y2Nwx97bIXv+2229iyZQt79+5l6NChDBo0iDVr1rBgwQJ27NjB73//ewoLC7nhhht4//33adq0KevXr2fcuHGcOXOGhg0bsnLlSjZs2FA2hP/nn39eNuS+iPDFF1+Qn5/PfffdR2ZmJgUFBTzzzDOkp6cTGRnJm2++yaBBg0hNTWXhwoWcPXuW3bt3k5yczOTJk/1bP7UUfi2Y7Ysh6hroeGuwIzHGXIUuXLjAsmXL6N27NwA7duxg5MiRbNq0iSZNmjBp0iRWrFjBxo0bSUhI4M0336SoqIjhw4fz1ltvsXnzZlasWEGjRo3KbTclJYWpU6eSkZHB3//+94vmT506FYCtW7eSlpbGE088UTZgZkZGBnPmzGHr1q3MmTPnko8GCJTwasEUn3dO8HcbChH1gx2NMaY2atDS8Kdz584RHx8POC2Y0aNHk5eXR8eOHUlMTARg7dq1ZGVlMWCAM0h8UVER/fv3Z8eOHbRt25bvf//7ADRv3vyi7Q8YMIDnn3+eESNG8OCDDxIbG1tu/pdffsnYsc5T5bt3707Hjh3ZuXMn4AyC2aKFM2BvXFwc+/bto0OHDgRbeCWYfauh4LhdnmyMqbHSczAV+Q6zr6okJSWRlpZWbpktW7ZcNKx/RePHj+dHP/oRS5cuJTExkRUrVpR7lkx140Y2bNiwbDoiIoILFy5c8ngCIby6yLKXQGQjuGFwsCMxxoSgxMREVq9eTU5ODgBnz55l586ddO/enby8PNavXw84w/FXTAK7d++md+/evPDCCyQkJJCdnV1u/u233142dP/OnTv55ptv6NatWwCOqvbCJ8GoOgmmy2Bo0DjY0RhjQlDr1q1JTU3l8ccfp0+fPiQmJpKdnU2DBg2YM2cOY8eOpW/fviQlJV30wLE//elP9OrVi759+9KoUSOGDh1abv4vf/lLiouL6d27N8OHDyc1NbVcy6UuCp/h+g9sgHd+CD+eBvE/CVxgxpgrZsP1+4cN1++V7CUgEXDjkGBHYowxYSF8Esz2xc6lyY2jgx2JMcaEhfBIMN/ugm93QI/7gx2JMcaEjfBIMNmLnd/d7g1uHMYYE0bCJMEsgbbxcE3wbzwyxphwEfoJ5uRByF0PPWzsMWOMCaTQTzA77NHIxpgr5ztc//3338/x48f9uv3U1FTGjBkDwMSJE0lJSfHr9oMh9BNM9hKIvsF5PLIxxtSS73D90dHRZYNPmqp5OhaZiAwB3gIigHdV9bUK868DPgCucZcZr6pL3XkvAqOBYuBZVf2kxgGcOw5ffwGJv7RnvxgTIl5f9zrZx7IvvWANdI/uzgu3vHDZy/fv358tW7aU/f3GG28wd+5cCgsLSU5O5uWXXwZg5syZpKSkICL06dOHWbNmsWjRIiZNmkRRURExMTF8+OGHtGnTxq/HU1d4lmBEJAKYCiQBucB6EVmoqlk+i70EzFXVaSISBywFOrnTjwE9gXbAChG5UVWLaxTErk+h5IJdnmyM8Zvi4mJWrlzJ6NGjAVi+fDm7du1i3bp1qCrDhg3jiy++ICYmhldffZXVq1fTqlUrjh07BsDAgQNZu3YtIsK7777L5MmT+eMf/xjMQ/KMly2YW4AcVd0DICKzgQcA3wSjQOm41S2APHf6AWC2qhYCX4tIjru9NTWKIHsRNG0D7T0ZBcEYEwQ1aWn4U+lw/Xv37uXmm28mKSkJcBLM8uXL6dfPeQT76dOn2bVrF5s3b+bhhx+mVatWAERHOzd55+bmMnz4cA4ePEhRURGdO3cOyvEEgpfnYNoDvk+9yXXLfE0E/kVEcnFaL2NrsG71zp+DXSuce1/qhf6pJmOMt0rPwezbt4+ioqKyczCqyosvvkhGRgYZGRnk5OQwevRoVLXSIfrHjh3LmDFj2Lp1K3/+858vGvQylHj5yVvZSY+KI2s+DqSqaixwLzBLROpd5rqIyC9EJF1E0o8ePVp+5p7P4fwZuzzZGONXLVq0YMqUKaSkpHD+/HnuueceZsyYwenTpwE4cOAAR44cYfDgwcydO5f8/HyAsi6yEydO0L698335gw8+CM5BBIiXXWS5gO+djbF81wVWajQwBEBV14hIFNDqMtdFVacD08EZTbnczOxF0LA5dLr9yo7CGGMq6NevH3379mX27Nn89Kc/Zfv27fTv3x+Apk2b8pe//IWePXsyYcIE7rjjDiIiIujXrx+pqalMnDiRRx55hPbt25OYmMjXX38d5KPxjmfD9YtIJLATGAwcANYDP1HVbT7LLAPmqGqqiPQAVuJ0hcUB/41z3qWdW961upP85YbrLymGlK5w/SB4+D0vDs8YE0A2XL9/BHq4fs9aMKp6QUTGAJ/gXII8Q1W3icgfgHRVXQj8BnhHRH6N0wX2pDoZb5uIzMW5IOAC8KsaXUH2zVo4m2+PRjbGmCDy9D4Y956WpRXKfucznQUMqGLdV4FXa7Xj7CUQ0RC6JtVqdWOMMVcu9C6vUnXOv1x/JzRsFuxojDEmbIVegjm0FY5/Y91jxhgTZKGXYLKXAGLPfjHGmCALwQSzGK5LhKatgx2JMcaEtdBKMMe+hsOZNjS/McbvSofrL/3Zu3cv+fn5DBo0iKZNm5YNtW++4+lVZAGXXfrsFzv/Yozxr9KhYnydOXOGV155hczMTDIzM4MUWd0VYglmMbTpBdGhO3icMeHu0H/8B4Xb/Ttcf8Me3bn2t7+t8XpNmjRh4MCB5OTk+DWeUBE6XWQlF5wbLK17zBjjgdLRlOPj40lOTg52OFeF0GnBFJwA1LrHjAlxtWlp+ENlXWSmeqHTgik4AddcB9f2DnYkxhhjCKUEU3jK6R6zRyMbY0ydEDpdZFpi51+MMQHXqVMnTp48SVFREQsWLGD58uXExcUFO6w6IXQSTFQL5wZLY4zxQOkDxSrau3dvYAO5ioROF1n09VAvIthRGGOMcYVOgjHGGFOnWIIxxlwVvHr6brgIRv1ZgjHG1HlRUVHk5+dbkqklVSU/P5+oqKiA7jd0TvIbY0JWbGwsubm5HD16NNihXLWioqKIjY0N6D4twRhj6rz69evTubONMXi1sS4yY4wxnrAEY4wxxhOWYIwxxnhCQuWqDBE5CuwL0O5aAd8GaF/+YPF6y+L1lsXrrW6q2syLDYfMSX5VbR2ofYlIuqomBGp/V8ri9ZbF6y2L11siku7Vtq2LzBhjjCcswRhjjPGEJZjamR7sAGrI4vWWxesti9dbnsUbMif5jTHG1C3WgjHGGOMJSzDGGGM8YQkGEJEOIrJKRLaLyDYRGeeWTxSRAyKS4f7c67POiyKSIyI7ROQen/IhblmOiIz3MOa9IrLVjSvdLYsWkU9FZJf7u6VbLiIyxY1pi4jc5LOdJ9zld4nIEx7F2s2nDjNE5KSIPFeX6ldEZojIERHJ9CnzW32KyM3u65XjrisexPuGiGS7Mc0XkWvc8k4ics6nnt++VFxVHbuf4/Xb6y8inUXkKzfeOSLSwIN45/jEuldEMtzyulC/VX2GBfc9rKph/wO0BW5yp5sBO4E4YCLwfypZPg7YDDQEOgO7gQj3ZzdwPdDAXSbOo5j3Aq0qlE0GxrvT44HX3el7gWWAAInAV255NLDH/d3SnW7pcV1HAIeAjnWpfoHbgZuATC/qE1gH9HfXWQYM9SDeu4FId/p1n3g7+S5XYTuVxlXVsfs5Xr+9/sBc4DF3+m3gGX/HW2H+H4Hf1aH6reozLKjvYWvBAKp6UFU3utOngO1A+2pWeQCYraqFqvo1kAPc4v7kqOoeVS0CZrvLBsoDwAfu9AfAj33KZ6pjLXCNiLQF7gE+VdVjqvpP4FNgiMcxDgZ2q2p1oy4EvH5V9QvgWCVxXHF9uvOaq+oadf5TZ/psy2/xqupyVb3g/rkWqHZs9kvEVdWx+y3eatTo9Xe/Sf8Q+CgQ8br7exRIq24bAa7fqj7DgvoetgRTgYh0AvoBX7lFY9wm5AyfZmx7YL/ParluWVXlXlBguYhsEJFfuGVtVPUgOG844Ht1KN5Sj1H+H7Ou1i/4rz7bu9MVy700CudbZqnOIrJJRD4XkdvcsuriqurY/c0fr38McNwnuXpdv7cBh1V1l09ZnanfCp9hQX0PW4LxISJNgf8BnlPVk8A04AYgHjiI0ywGp4lYkVZT7oUBqnoTMBT4lYjcXs2ydSFe3H7xYcA8t6gu1291ahpfoOt5AnAB+NAtOghcp6r9gOeB/xaR5oGOqxL+ev0DfRyPU/5LUp2p30o+w6pctJIyv9exJRiXiNTHeWE+VNW/AqjqYVUtVtUS4B2cJjo42buDz+qxQF415X6nqnnu7yPAfDe2w25TtrR5fqSuxOsaCmxU1cNu7HW2fl3+qs9cyndXeRa3e1L2PmCE25WB29WU705vwDmPceMl4qrq2P3Gj6//tzhdPJEVyv3O3ceDwByf46gT9VvZZ1g1+wnIe9gSDGV9qu8B21X1TZ/ytj6LJQOlV5QsBB4TkYYi0hnoinMCbD3Q1b2ipQFOd9BCD+JtIiLNSqdxTu5muvsqverjCeD/+8Q70r1yJBE44TaXPwHuFpGWbvfE3W6ZV8p986ur9evDL/XpzjslIonue22kz7b8RkSGAC8Aw1T1rE95axGJcKevx6nPPZeIq6pj92e8fnn93US6CnjYy3hddwHZqlrWXVQX6reqz7Bq9hOY9/ClrgIIhx9gIE5zbwuQ4f7cC8wCtrrlC4G2PutMwPmmsgOfqync9Xa68yZ4FO/1OFfQbAa2le4Hpy96JbDL/R3tlgsw1Y1pK5Dgs61ROCdRc4CnPKzjxkA+0MKnrM7UL07iOwicx/m2Ntqf9Qkk4HyA7gb+H+4oGn6ONwen/7z0Pfy2u+xD7vtkM7ARuP9ScVV17H6O12+vv/s/sc6tg3lAQ3/H65anAk9XWLYu1G9Vn2FBfQ/bUDHGGGM8YV1kxhhjPGEJxhhjjCcswRhjjPGEJRhjjDGesARjjDHGE5ZgTFgQkRj5brTbQ1J+FN/LGnlXRN4XkW6XWOZXIjLCP1EHfvvG+JNdpmzCjohMBE6rakqFcsH5nygJSmDGhBhrwZiwJiJdRCRTnGd4bATaish0EUkX57kav/NZ9ksRiReRSBE5LiKvichmEVkjIt9zl5kkIs/5LP+aiKwT5xkmt7rlTUTkf9x109x9xVcS2xsikiXOYJCv+25fnOd/+D5jp0RE2otIGxH5q7vNde5d2sYEhSUYY5znZrynqv1U9QDO8zMSgL5AkojEVbJOC+BzVe0LrMG5+7kyoqq3AP8XKE1WY4FD7rqv4Yx8W34lkTY4d2L3VNU+wH/6zlfV/aoar6rxwPs4w9sfAKYAk934HwXevfxqMMa/Ii+9iDEhb7eqrvf5+3ERGY3z/9EOJwFlVVjnnKqWDoe/AWcI98r81WeZTu70QJwHgqGqm0VkWyXrHQNKgHdEZAmwuLKNizOK9hPuNsEZK6ubfPewwZYi0khVz1URnzGesQRjDJwpnRCRrsA44BZVPS4ifwGiKlmnyGe6mKr/lworWeaSj5pV1fMikgAk4Qzq+AzOwINlRKQ9MB24T78b3FLc2H3jMyYorIvMmPKaA6eAk/LdE/787Uuc7itEpDdOC6kccUbLbq6qi4FfU6Ebzb3ybR7OI4dzfGatAH7ls9xF53aMCRRLMMaUtxGnOywT5xklqz3Yx38B7UVkC/Abd18nKizTAlgiIpuBz3AeZOXrNpykM8nnRP/3cJLLAPfCgCzg5x7Eb8xlscuUjQkwcR5aFamqBW6X3HKgq373yF9jQoKdgzEm8JoCK91EI8C/WnIxochaMMYYYzxh52CMMcZ4whKMMcYYT1iCMcYY4wlLMMYYYzxhCcYYY4wn/hdiZ6qWxpEU5wAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x17b8ff50080>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "size = np.arange(1000, 20001, 1000)\n",
    "plt.title(\"Learning curve\")\n",
    "\n",
    "plt.plot(size, scores[:, 0], label = 'Accuracy')\n",
    "plt.plot(size, scores[:, 1], label = 'Precision')\n",
    "plt.plot(size, scores[:, 2], label = 'Recall')\n",
    "plt.plot(size, scores[:, 3], label = 'F1')\n",
    "\n",
    "plt.xlabel('Training size')\n",
    "plt.ylabel('Metric')\n",
    "plt.xlim(size.min(), size.max())\n",
    "#plt.ylim(0, 200)\n",
    "plt.legend(loc = 'lower right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Cross-validation:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import KFold\n",
    "\n",
    "scores = []\n",
    "kf = KFold(n_splits=10)\n",
    "for train, valid in kf.split(df['sentiment']):\n",
    "    X_train_text = df['review'][train]\n",
    "    y_train = df['sentiment'][train]\n",
    "\n",
    "    X_valid_text = df['review'][valid]\n",
    "    y_valid = df['sentiment'][valid]\n",
    "\n",
    "    vect = TfidfVectorizer(min_df = .0005)\n",
    "\n",
    "    X_train = vect.fit_transform(X_train_text)\n",
    "    X_valid = vect.transform(X_valid_text)\n",
    "\n",
    "    LR = LogisticRegression(C = 5)\n",
    "    LR.fit(X_train, y_train)\n",
    "\n",
    "    p_valid = LR.predict(X_valid)\n",
    "    scores.append([metric(y_valid, p_valid) for metric in metrics])\n",
    "\n",
    "scores = np.array(scores).mean(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy:  0.893\n",
      "Precision: 0.887\n",
      "Recall:    0.901\n",
      "F1:        0.894\n"
     ]
    }
   ],
   "source": [
    "print(\"Accuracy:  %.03f\" % scores[0])\n",
    "print(\"Precision: %.03f\" % scores[1])\n",
    "print(\"Recall:    %.03f\" % scores[2])\n",
    "print(\"F1:        %.03f\" % scores[3])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Demographic classification\n",
    "\n",
    "Blog gender classification, download data from:\n",
    "\n",
    "https://github.com/mdimarco/blog_gender_identification\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>target</th>\n",
       "      <th>text</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>chris having friday afternoon relaxing plannin...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>originally uploaded fun contrary think given w...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1</td>\n",
       "      <td>literature got taste 1ST_PERSON read 1ST_PERSO...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1</td>\n",
       "      <td>saturday february la food 1ST_PERSON setting u...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>i've get eating solid foods frustrating he's s...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>1</td>\n",
       "      <td>just make around home keep expectations 1ST_PE...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>1</td>\n",
       "      <td>really enjoyed mark confidence basically setti...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>0</td>\n",
       "      <td>took trip home weekend friends family going so...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>1</td>\n",
       "      <td>quite abilities unknown character guy since 1s...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>1</td>\n",
       "      <td>waiting hear town reported wednesday had guys ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   target                                               text\n",
       "0       1  chris having friday afternoon relaxing plannin...\n",
       "1       1  originally uploaded fun contrary think given w...\n",
       "2       1  literature got taste 1ST_PERSON read 1ST_PERSO...\n",
       "3       1  saturday february la food 1ST_PERSON setting u...\n",
       "4       0  i've get eating solid foods frustrating he's s...\n",
       "5       1  just make around home keep expectations 1ST_PE...\n",
       "6       1  really enjoyed mark confidence basically setti...\n",
       "7       0  took trip home weekend friends family going so...\n",
       "8       1  quite abilities unknown character guy since 1s...\n",
       "9       1  waiting hear town reported wednesday had guys ..."
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df = pd.read_csv(\"train.csv\", names = ['target','text'], delimiter=\",\", quoting=1, encoding = 'latin-1')\n",
    "valid_df = pd.read_csv(\"test.csv\", names = ['target','text'], delimiter=\",\", quoting=1, encoding = 'latin-1')\n",
    "valid_df.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2903, 2)\n",
      "(322, 2)\n"
     ]
    }
   ],
   "source": [
    "print(train_df.shape)\n",
    "print(valid_df.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Vectorization and training logistic regression:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "             precision    recall  f1-score   support\n",
      "\n",
      "     Female      0.698     0.669     0.683       145\n",
      "       Male      0.738     0.763     0.750       177\n",
      "\n",
      "avg / total      0.720     0.720     0.720       322\n",
      "\n"
     ]
    }
   ],
   "source": [
    "vect = CountVectorizer(binary = True, min_df = 5)\n",
    "\n",
    "X_train = vect.fit_transform(train_df['text'])\n",
    "y_train = train_df['target']\n",
    "\n",
    "X_valid = vect.transform(valid_df['text'])\n",
    "y_valid = valid_df['target']\n",
    "\n",
    "LR = LogisticRegression(C = .01)\n",
    "LR.fit(X_train, y_train)\n",
    "\n",
    "p_valid = LR.predict(X_valid)\n",
    "target_names = ['Female', 'Male']\n",
    "print(classification_report(y_valid, p_valid, digits = 3, target_names = target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Top features:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Male: its, game, wife, john, guy, set, similar, future, guys, team, character, you, important, test, case, obviously, film, bill, video, possible\n",
      "\n",
      "Female: love, female, food, mom, husband, little, can, beautiful, lunch, eating, felt, lovely, really, week, eat, children, myself, baby, crazy, oh\n",
      "\n"
     ]
    }
   ],
   "source": [
    "vocab = np.array(vect.get_feature_names())\n",
    "order = LR.coef_[0].argsort()\n",
    "print(\"Male: %s\\n\" % (\", \".join(vocab[order[-20:][::-1]])))\n",
    "print(\"Female: %s\\n\" % (\", \".join(vocab[order[:20]])))"
   ]
  }
 ],
 "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}