Notebooks/Classification_and_Categorical_Data.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Classifiers and Categorical Data\n",
    "\n",
    "Lecture 2/27/2020\n",
    "Materials and code courtesy of Na-Rae Han"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## A Brief Review of Count Vectors and TD-IDF Vectorizer\n",
    "\n",
    "### What is a Count Vector?\n",
    "\n",
    "Info about the CountVector() function: https://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.CountVectorizer.html\n",
    "\n",
    "\"Convert a collection of text documents to a matrix of token counts\"\n",
    "* Each row represents a document\n",
    "* Each column represents a token count\n",
    "\n",
    "### What is TF-IDF Vectorizer?\n",
    "\n",
    "Info about the TfidfTransformer() function: http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.TfidfTransformer.html\n",
    "\n",
    "\"Scale down the impact of tokens that occur very frequently in a given corpus and that are hence empirically less informative than features that occur in a small fraction of the training corpus.\"\n",
    "\n",
    "Info about the TfidfVectorizer() function: http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.TfidfVectorizer.html\n",
    "\n",
    "\"Equivalent to CountVectorizer followed by TfidfTransformer.\"\n",
    "\n",
    "### Why do we use them?\n",
    "Text can't be put directly into a classifier. They require numerical data. These functions allow us to (1) convert the text into token counts vectors and (2) convert these token count vectors into numbers that even out effects of very large counts\n",
    "\n",
    "### How can we use them effectively?\n",
    "\n",
    "Let's take a brief detour into Tuesday's code, with slight modifications..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Pretty printing has been turned OFF\n"
     ]
    }
   ],
   "source": [
    "# Turns on/off pretty printing \n",
    "%pprint\n",
    "\n",
    "# Every returned Out[] is displayed, not just the last one. \n",
    "from IPython.core.interactiveshell import InteractiveShell\n",
    "InteractiveShell.ast_node_interactivity = \"all\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import sklearn               \n",
    "import nltk \n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns        \n",
    "sns.set_style('darkgrid')\n",
    "\n",
    "import statsmodels.api as sm\n",
    "import statsmodels.formula.api as smf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Remove the -1 from your path\n",
    "english = pd.read_csv('../../Class-Exercise-Repo-1/activity3/english_updated.csv', index_col='Index')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['alt.atheism', 'comp.graphics', 'comp.os.ms-windows.misc', 'comp.sys.ibm.pc.hardware', 'comp.sys.mac.hardware', 'comp.windows.x', 'misc.forsale', 'rec.autos', 'rec.motorcycles', 'rec.sport.baseball', 'rec.sport.hockey', 'sci.crypt', 'sci.electronics', 'sci.med', 'sci.space', 'soc.religion.christian', 'talk.politics.guns', 'talk.politics.mideast', 'talk.politics.misc', 'talk.religion.misc']"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.datasets import fetch_20newsgroups\n",
    "data = fetch_20newsgroups()   # downloads training data by default: subset='train'. test', 'all'\n",
    "data.target_names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "categories = ['talk.religion.misc', 'soc.religion.christian',\n",
    "              'sci.space', 'comp.graphics']\n",
    "train = fetch_20newsgroups(subset='train', categories=categories)\n",
    "test = fetch_20newsgroups(subset='test', categories=categories)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sklearn.naive_bayes import MultinomialNB\n",
    "from sklearn.pipeline import make_pipeline\n",
    "\n",
    "# pipeline! See textbook. \n",
    "model = make_pipeline(TfidfVectorizer(), MultinomialNB())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Pipeline(memory=None,\n",
       "     steps=[('tfidfvectorizer', TfidfVectorizer(analyzer='word', binary=False, decode_error='strict',\n",
       "        dtype=<class 'numpy.float64'>, encoding='utf-8', input='content',\n",
       "        lowercase=True, max_df=1.0, max_features=None, min_df=1,\n",
       "        ngram_range=(1, 1), norm='l2', preprocessor=None, smooth...   vocabulary=None)), ('multinomialnb', MultinomialNB(alpha=1.0, class_prior=None, fit_prior=True))])"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# train model\n",
    "model.fit(train.data, train.target)\n",
    "\n",
    "# predict labels on test data\n",
    "labels = model.predict(test.data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[344,  13,  32,   0],\n",
       "       [  6, 364,  24,   0],\n",
       "       [  1,   5, 392,   0],\n",
       "       [  4,  12, 187,  48]], dtype=int64)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "mat1 = confusion_matrix(test.target, labels)\n",
    "mat1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8016759776536313"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.metrics import accuracy_score\n",
    "accuracy_score(test.target, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We got 80% accuracy by using the standard TfidfVectorizer function. What happens when we modify it?\n",
    "\n",
    "### The TfidfVectorizer() function:\n",
    "It has many parameters. Let's look at a few...\n",
    "* min_df: Only include words that occur above this minimum document frequency (between 0 and 1 OR an int)\n",
    "    * a cut-off\n",
    "* max_df: Only include words that occur below this maximum document frequency (between 0 and 1 OR an int)\n",
    "    * Corpus-specific stop words\n",
    "* max_features: Only include the top N features (ordered by frequency)\n",
    "* stop_words: a list of words to be ignored for tokenization\n",
    "    * How might you choose your stop words?\n",
    "\n",
    "Questions to ask when adjusting these arguments:\n",
    "1. How am I limiting my data?\n",
    "2. Why am I limiting my data in this way?\n",
    "\n",
    "### Fudging around with these arguments:\n",
    "Let's start simple--what if we get rid of english stop words?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Pipeline(memory=None,\n",
       "     steps=[('tfidfvectorizer', TfidfVectorizer(analyzer='word', binary=False, decode_error='strict',\n",
       "        dtype=<class 'numpy.float64'>, encoding='utf-8', input='content',\n",
       "        lowercase=True, max_df=1.0, max_features=None, min_df=1,\n",
       "        ngram_range=(1, 1), norm='l2', preprocessor=None, smooth...   vocabulary=None)), ('multinomialnb', MultinomialNB(alpha=1.0, class_prior=None, fit_prior=True))])"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model2 = make_pipeline(TfidfVectorizer(stop_words = 'english'), MultinomialNB())\n",
    "\n",
    "# train model\n",
    "model2.fit(train.data, train.target)\n",
    "\n",
    "# predict labels on test data\n",
    "labels2 = model2.predict(test.data)\n",
    "\n",
    "mat2 = confusion_matrix(test.target, labels2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Including Stop Words:\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.8016759776536313"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "array([[344,  13,  32,   0],\n",
       "       [  6, 364,  24,   0],\n",
       "       [  1,   5, 392,   0],\n",
       "       [  4,  12, 187,  48]], dtype=int64)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Excluding Stop Words:\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.8561452513966481"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "array([[367,  13,   9,   0],\n",
       "       [  9, 379,   6,   0],\n",
       "       [  2,   5, 389,   2],\n",
       "       [  5,  20, 135,  91]], dtype=int64)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(\"Including Stop Words:\")\n",
    "accuracy_score(test.target, labels)\n",
    "mat1\n",
    "\n",
    "print(\"Excluding Stop Words:\")\n",
    "accuracy_score(test.target, labels2)\n",
    "mat2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot object at 0x0E96FB10>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 16.0, 'true label')"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "Text(92.68, 0.5, 'predicted label')"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAVoAAAFaCAYAAABfUZV2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4wLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvqOYd8AAAIABJREFUeJzt3Xd8VFX6x/FPQijSFVGwi7qPvayCvQP2XsGGoBS74LoKItWu6LKuIoJlXRFFXcEuP+yignWl+NgVCyJFBOlJfn+cGwyYwMDMzZ2ZfN+vF69pd2a+NwnPnDn33HMKSktLERGR+BQmHUBEJN+p0IqIxEyFVkQkZiq0IiIxU6EVEYmZCq2ISMxUaEVEYqZCKyISMxVaEZGYFSUdQOKzcNzQvDvtr/FRA5KOEJvikpKkI8SisKAg6QixWbL4+5R2Ti1aEZGYqdCKiMRMhVZEJGYqtCIiMVOhFRGJmQqtiEjMVGhFRGKmQisiEjMVWhGRmKnQiojETIVWRCRmKrQiIjFToRURiZkKrYhIzFRoRURipkIrIhIzFVoRkZip0IqIxEyFVkQkZiq0IiIxU6EVEYmZCq2ISMxUaCUlxSUl9HnoBc659RE6DhrJtF9+Xf7YcxOncvYtI1bYvqSklAvvfIJRr39c1VEz5m9/u5DXXn2Kt8c/S4cOpyUdJ6NatdyNcWNHJR0j45o2bcKXX0zAbKuko6xAhTYhZja9gvs6mNmxSeRZndf+9yUAD17Rjm5H78ttT7wKwKfTZvDU+EmUlq64/Z1Pv8ncBYuqOGXmHHDAXuy91+4cdPAJtG5zCptsslHSkTLmih7duOeeW6hTp07SUTKqqKiIu/51E4sWZd/fnQptFnH3B9x9TNI5KnLIrtvQu31bAH6a/RvrNajLr/MXMnj0G/zt5INW2HbsB59RWFDAvttvmUDSzGjT5kAmTXZGPTaMJ5+8n+eeG5d0pIz58qtvOeXU85OOkXE33dSbofc+xI8//px0lD8pSjpAOsxsHeB+YHOgJnA50BnYCqgBDHL3R83sVeBjYEdgPvAGcBjQGGgLHBf9awisD/R39ydWeq/ewAnAL0BdoDdwELAPUB/oBJwN7AE0AKa6+7lm1hfYFtgAWBe42N3fBGqb2QhgM2AWcDLQC5gODAUGA62AWkAfYDzwKOHDsSbQ1d0/SfuHuAaKahRyzYPP88rHX3DLecfQ9z8vcsVJB1G71h9/Rl/8OJPnJ07l1vOP5Z7n3q7KeBnVpMl6bL7ZJhx/Qge23GJTnnjiPnba+aCkY2XEf//7HJtvvknSMTLqrLNOYeYvsxg79jWu/NtFScf5k1xv0XYFvnH3vYEOwIHATHffB2gNDDSz9aNtJ7j7oUBtYIG7twGmRM+BUCzbEArvIDNbXj3MbBfgCKAlcDzQvFyGqdH7/QDMiV53H2AvM9s42maBux8CnAn8q9z79XT3/YBGwG7lXvM4YH13bwUcHr1vK2BulOMSwodClRt4zhGM7tORy+8ZzWc//MJ1I/+Pq4Y/y1fTZ3HzqFd4+p3JzJg7n/P/8Rhj3pnMQy+/z1uTv04ialpmz5rD2LGvsXTpUj77/CsWLVpM06ZNko4llehwzmkceugBjH1pFLvssj33Df8HG27YNOlYy+V0ixYw4HkAd59kZt2A/4tuzzOzKYTWLcAH0eWvhAILMAco66h6zd1LgJ/NbA7QFPgpemw7QqEuBhaa2XvlMnh0uRDYwMweIbSa6xNangAvR5kmm1mz6L7Z7v5NdH06oZVcfr/ejp4zHbjGzAqBbYDRwFJgYEo/oQx55t0p/DxnHp0O35M6tYpo0rAu/732XGrXLOKHWXO5avizXHnKwSs85+5nxrN+w3rsu0PudSGMHz+Riy7qyB3/GErz5htSr15dZs2ak3QsqcShrU9efn3sS6O46OKr+PnnXxJMtKJcb9FOJbT2MLMWQDtg/+h2A2AnoKw5VVrRC5Sze/S8DQmtxRnlHpsMtDSzQjOrzYqtz5Lo8ghgU3dvB/QE1gEKVnrtHQkt39XlKb9fjczsRUI3xU/u3pZQZK9fzf5k1KG7bsOn38+g46CRXHDnE/zt5IOpXTPXP6cr99zz4/joo8m89eYzPPnEfVx66TWUlJSs/okiFcj1/yn3APeZ2WuEPtnDgQvN7E1Coevn7jPMLJXXamZm4whf4y9w92Iz6w584e5jzOw54B1gJqFFuXSl508AepvZO8Bi4Cug7FD1btFr1wNSOQoxBmgd7UcR0I/Qx/yomV0GFAP9U9mpTFmndk1uOe+YCh/buEkjHrqy/Z/u73b0PnHHilXPXlX6WValvv32e/bdv+LfZ65r0/aUpCP8SUHpyuNyqiEz6wBs6+5XVfL4BsDJ7n5X1KKdDBzi7t+l8Np9genuPiSDkVOycNzQvPvlNj5qQNIRYlOcpy3mwoKC1W+Uo5Ys/j6lncv1Fm1VmUnoOphI+Mo/LJUiKyICatHmNbVoc4tatLkn1RZtrh8MExHJeiq0IiIxU6EVEYmZCq2ISMxUaEVEYqZCKyISMxVaEZGYqdCKiMRMhVZEJGYqtCIiMVOhFRGJmQqtiEjMVGhFRGKmQisiEjMVWhGRmKnQiojETIVWRCRmKrQiIjFToRURiZkKrYhIzLQ4Yx5rWK9F3v1yZ4y/M+kIsWm0Z5ekI8SiZmH+LrY9f8HXWpxRRCQbqNCKiMRMhVZEJGYqtCIiMVOhFRGJmQqtiEjMVGhFRGKmQisiEjMVWhGRmKnQiojETIVWRCRmKrQiIjFToRURiZkKrYhIzFRoRURipkIrIhIzFVoRkZip0IqIxEyFVkQkZiq0IiIxU6EVEYmZCq2ISMxUaEVEYlbpgutm1rmyx9x9aDxxJFfssccu9Bvwd446oj227dYM/uf1FBTApE8+5YoefSkpKUk6YkqKS0roN+RRvv1pBoWFhfTv1o6669Sm/z2P8dvvCygpKWHghWewabP1ASgpKeGiG+/loD125NS2+yacfu28+87zzP1tHgDffDONzp17JJwoPXu03JUBA/7OEYe3Y6edt+O22/pRXFzM4sVL6Hx+D2bMmJl0xMoLLdC8ylLkKDPbFTjW3fsnnaUqXXp5Z05vdwILfl8AQJ++V9Cv7y2Mf2sid99zM0ce1Zpnnn4p4ZSpee29yQA8OOBSJk7+glv/PZqG9dbhyP3+ymH77MaESZ/z9Y8/Ly+0d458nrnzFyQZOS21a9cGoG3bUxNOkhmXXd6Fdu2O5/cFCwG45ZY+9OjRh0/+N5WOndpxefcuXH3VdQmnXEWhdfd+ZdfNrDWwJfAu8FkV5MoJ7v4R8FHSOara1199x5ntujF02G0AnNn+AkpKSqhZsyYbbtg0K1oQqTqk1U4csPv2APz0y2yaNKrPxMlfsM3mG9F5wF1s1HQ9ruxwAgBj3/mIwsIC9tt1uyQjp2Xnnbejbt11ePaZh6lRVINrr72JCRM+TDrWWvv6q29p364b9w4fBMA551zMz9N/AaCoqIhFixcnGW+5VbVoATCz64FNgO2AJcDVQLuYcyXOzP4CPAAsBZYBZxP2vRVQC+gDzAW6uvvp5Z7XFHiU0P9dE+gKzANGAT8RfpbPu3svM9sRGBRt2xi4xN3Hm1knoBtQAxjt7n3N7BSgO1AMvOnuV8X7E6jcmNEvsNlmGy+/XVJSwqabbsSYZ/7D3N/m8cXnXyUVba0U1ajBNXc+zMsTP+HW7h0Y/eoEGtZbh6G9L2DI4y9y/+iXabv3rjz35gfc1r0D9zyeG631iixYsIjb77iH++57hG223pIxYx5ix50OpLi4OOloa2X0Sn+LZUV2zz3/SpcuZ3NY29OSiraCVA6G7efuZwPz3f1BQsu2OmgDvA+0Bq4DOgLru3sr4HCgZSXPa0UowEcAlwANo/u3ADpEzzvEzP4K7AD0cPfWhIJ7rpltAFwF7A/sDjQys82AfsCh7r4fsLGZtcno3qZp2rQf2W2XQ7hv2MNcf2OvpOOssYEXncGYf/Sk/z2P0aDuOhy0x44AHLj7Dkz5ahrPvD6RGbPncn7/uxjz2gQeevY13vpoasKp19znn3/FiBH/Dde/+JpZs+bQvPkGCafKrJNOOop/DL6Ok07syMyZs5OOA6TQogWKzKwOUGpmNQgtqupgOPB34AVC4ZwAvA3g7tOBa8zsoAqe9zywDTCa0BoeGN3/sbvPBjCzdwEDpgG9zWwh0AD4DWgBTHL3hdHzLjezVkBT4DkzI9q2RSZ3Nh0jHxtKr6uv58svv2H+/N8pKSlNOlLKnn59IjNmzaXTCa2pU6sWBQUF7L79Vrzx4RSOOaAlH0z9kq02acblZx67/Dl3P/YCTRo3YN8c7ELocM5p7LjjtlxyaS+aN9+Qhg3r89NPM5KOlTGnnX48nTq154jDT2fOnLlJx1kulUJ7O6Fl15TQR3t7rImyx3HAG+7ez8zaAdcDbwGYWSPgMeCGCp53EPCTu7c1s72j550LbGdmdYHFwJ7A/dG/M9x9qpn1I7R6vwS2NbPa7r7YzB4HehCKcht3X2pmHciivuFBtw3h7ntuZsmSpSxcuIiLLkysV2ONHdpqZ/rc9Qjn9vkny5YVc2WH47EtNqbfkEcZ9dJ46tetw42XnJV0zIy5/4GRDBs2iJdffoLSUujc5Yqc7TZYWWFhIbfc2ofvv/+REY8MAeDNN9/luoF3JJwMCkpLV9/6MLN1ga2Ar919VuypsoCZbQX8h9A/W0LoH+0A7Eb4gOoHLCTqozWzm4HHCYXyUaAeofXfn3AA8RVgCrAh8Li732hm3Ql9uD8D3xO6JtpEhbQrUAo87e7Xm9mZwAWEfttvgHPdfZWHvxvWa5E7TcsUzRh/Z9IRYtNozy5JR4hFzcJU2nO5af6CrwtS2W61hdbM9gDuIRSI74Au7v5J2gmrETPbAhjp7ntV5fuq0OYWFdrck2qhTeVg2GDgLHffBOgC3JVOMBGR6iaVQrvQ3acARC3ZJfFGyj/u/k1Vt2ZFJHukcgruUjO7C3idMHTpt6oIJiKSL1I5Bfft6NIIw5yy5mi3iEguSPUU3OaEs5wKgI2qIJeISN5I5RTc4cDehOFK6wBfAepvFBFJUSoHw7YjnCr6IrA9sCjWRCIieSaVQjvP3UuBeu4+kzChioiIpCiVQvu+mV0B/GhmI0nttF0REYmstmi6e08zq0/oMjiCMN+BiIikaFXjaG8gnGu/sr2BnrElEhHJM6tq0X5aZSlERPLYqsbRPliVQURE8pWWGxcRiZkKrYhIzFZ1MOzayh6rbstri4ikY1Ut2p+jf3sDzQgrB6wH7FoFuURE8saqDobdA2BmJ7r7BdHdD5vZ2CpJJiKSJ1Lpo20SrZ+FhSVYG65mexERKSeV02kvAx4xs42A6UD+LAkqIlIFVtuidfc3gTbAkcBB7j419lQiInkklVVwTwKuIbR+HwNK3X1gFWSTNG3eZOe8WwV3cfHSpCPEZtoXzyYdIRbv7Hhl0hFis//0xzO2Cm53wkTfM4GBwAlp5BIRqXZSKbQl7r6Y0JItBX6POZOISF5JpdC+YWYjgE3MbAgwMeZMIiJ5JdX5aA8HPgQ+dfen448lIpI/VnUKbg2gBjASOA14GahhZi+7+yFVlE9EJOetqkXbkTDBdzPAo/tKgDfjDiUikk9WdQruvcC9ZtbR3e8DMLNa7r6kytKJiOSBVA6GFZnZ7dH1Z8xMZ4aJiKyBVE7B7QrsE10/CngdeCi2RCIieSaVFm2xuy8CcPelVLxgo4iIVCKVFu1oM3sDmAD8FRgTbyQRkfySyjjagWb2DGDAv9394/hjiYjkj0q7DszsvOjyBuBUYBfgNDO7voqyiYjkhVW1aKdFl59WRRARkXy1qkK70MwOAL6uqjAiIvloVYW2W3S5FVCLMJnMbsB84KB4Y4mI5I9K+2jdvZ27twN+AfZw9/OBPYFFVRVORCQfpDKOtnm560XABjFlERHJS6mMox0OTDazScD2QJ94I4mI5JdUxtH+y8weArYFvnL3mfHHEhHJH6vtOjCzHYBnCS3b88zs6NhTiYjkkVT6aAcD5xIWZxwO9I0zkIhIvkml0OLuXxAWZ/wFmBdvJBGR/JLKwbDZZtYFqGdmpwO/xpxJslxhYSE33dGHFltvQXFxCVdc3Jv69evR78arKS4uZsmSJXS/oBczf5mddNQ1UlRUxO3/uo5NN92I2rVrcfutQ3jp+VcAOOHko+jU+UyObtsu4ZSpKy4ups9N/+Cb736gRmEhA3pezvzfF9D/ln9Sq2ZNtt2mBVdd1pXCwkJuvXMYH/5vMsuKiznluCM4+dgjko6/Wg1224Ytep/JJyf2od4OW7D1zZ0pXVbMwq9+4vPud1Nv+81pMeDc5ds3/Os2TDn3Zua88lGVZ02l0HYiLGkzE9gjup01zGy6uzczszuAQe7+XSXbjQTOzuQKEWb2ADDS3V9YxTaV5jKzOsCZ7j7MzDoAs90962dHa334gQCcdOQ57LXvHvQe+DcaNmxAn6tuYMokp/05J9Ptko4M6H1rwknXzMmnHcOc2b9ycZe/s+66jRn7+hO89Pwr7LDTtrQ/6yQKCgqSjrhGXn3rXQD+M+Q2JnzwP2755738/Mssrr68K7vttD2Dhz7Is2NfZcP1m/DdDz/y8NDbWbJkCced2ZU2B+1Ho4YNEt6Dym1y4XFscPIBFC9YDMBmPU7hu0GjmDPuQ+xfl7Je678ye+z7fHJiGCS1/jF7s2T67ESKLKRWaO929zNiT5Imd79sNY+fXlVZVnrfVeVqBpwHDHP3B6omUfpeeu4Vxr34OgAbb7oRM2fMolePAcz4OQxIKSqqwaLFubfi0ZinXuTp0S8uv11cXMy66zamV5/u9L76Bm77x4AE0625Qw/YhwP32ROAn6b/TJP1GvO/Kc5uO20PwG47bc/Lb7zDVZd2wbZpAUBBQQElJSXULEqlNCRn4TfTmdLxFuzOSwD4fdLXFDWuD0CN+nUoXVa8fNvCurXZ/G+n8vHx1yaSFVIrtHXMbGfgM8LijKyuVWhmfwEeAJYCy4Czge7AftEmI9z9H2a2DTCMcIrvAuD0qB+47HUmRe+7mLDSw3CgSfTwJe7+SbltX422mQmMAGoTFpU8xN23NrNvCEPUmkWvU5Mwifkl7v6xmX0OvEWYDvJn4CR3Ly73+n/KGj3UxcyuBBoRTlueATwNzAKeA46McjUBbot+JnOAM4BewPZmdi2hv3w6cC9wD7Bp9Jzn3b131HpeDGxBOImkg7t/sKrfQ5yKi4u57V8DOeyoQ+jWocfyIrt7y10457x2nHL0uat5heyz4PcFANSrX5dh/76Dm64bzKA7B9Cn540sWpSbJ0QWFdWg54BbGff6eAYN7MXnX33LxA//R8vddubVt95l4aJF1K5di9q1a7F02TJ6DryNk489grp110k6+irNevZdam/adPnthV/9xFY3nMdml53MsnkL+HX85OWPNWt3KL88/TbLZid3eCmVg2F/AUYDUwmFK5XZvNoA7wOtgeuAE4Atgb0Ixba9me0E3Arc4O57E4rLbiu9Tn1gQHQqcE9gnLsfDHQG7q7kvXsBT7n7gcAo/vxhcisw2N0PAC4lFF2AFkDvKEtToGUFz6so6/vR8uv/BDpE9zUD2rr7zeWefzzwJHAgcB+wbvSzmeLu/ctttynwjrsfFv2supV77Nvo/n9GP4NE9bjwGg5udQw33tGHdequw9HHH8Z1t/Wmw+kXMnvWnKTjrZWNNm7Gk08/yOOPjuHrL7+lRYstuGlQH4YMH8RfbCv633B10hHX2PW9r+DZkcPoe9Ngel9xEcMeeoxuV1zLeus2Zt1GDQGY+9s8una/hq222Izzzz4t4cRrrsXAjvzv+N68v/+lzBj1Gi36nrP8saYn7c/0h8clmC61ExZ2MrMahOIzw91LUnjd4cDfgReAucCHwBvuXgosNbN3CGeZGfB29D6PVRYhutwJOMTMyv4K1q1k++2AB6Prb1Ty+OvRe35kZptG989097KpIacBdVZ63p+ymll7wgcKhNZo3ej61xW0+q8nfAiMA34A3iW0ulc2G2hpZgcDv620zYfl8u1bwXOrxAmnHk3zjTbkrjuGs3DhIkpLSjn8qENof84pnHZsR+b++ltS0dKyftMmjHxyGD3/NpA3X38HgAP3PgaATTfbiCHDB3Ht1TckGXGNjHlhHD/PmMn5Z59GnTq1KSws4PXxExhw9eVs0LQJ1w+6i/322oNFixdz3qVXc87pJ3L0YYckHXutLJszn2XzFgKwZPpsGrY0AGo0qEthrSKW/DgryXgpnbBwAvAF4WvwZ2bWJoXXPY5QWA8ltCo7EnUbmFlNwmKPnxNayS2j+88ws4sreK2ywv4pcLu7H0SYiPzhSt57ErB3dH2vCh6fCuwfveeuhAIJq18LrbKsFT2vog+jM4AHohb5ZEKLtIQ//w46AL9G/eK3AXXNrOwoTFas1/bCM+PYYadteezp+3lo1BD69bqJvjdcRb36dbnnwdsZOXo4l//9gqRjrrFLe3SmceOGdL+yG08+8yBPPvMgdepU9FmYG1ofuC+ffv4l51zwN7pcfg1/v7QLf9lqC7pdcS1ndOlOvXp1OWCfVjz23+f4/sfpPP70C3S46Eo6XHQl3/84ffVvkEU+73E32w65nJ3/24/mHQ7jmxtGALDOVs1ZNO2X1Tw7fgWlpav+v2tmHwKHufsMM9sQeNrdW63mOVsB/yH0z5YAlwPtCa2wWsBj7n6TmW1N+Bpeg9DveSawK7Cfu/cv61d190Vm1oTQUm4MNAT6uvuYcqMOXuWPPtqHCC3SH4FW7r7NSn209xJaijWBi939vbLXifKPBIYA/yMcqDqxkqyDiEYdmNnhhH7bvtF9e0WvVZarEXAXYZrJJYRC+xPwDvAisJBQ9N8ARhLGK/9O6Eo4lNDNsMJ7uXuHVf0eNm+yc1YU5kxaXLw06QixmfbFs0lHiMU7O16ZdITY7D/98ZSGoqRSaP/P3VtXdjvbmNmRwC/uPtHMWgM9oz7UakeFNreo0OaeVAttKqMOfjOzF4HXgN0JX2WvB3D3nmsfMTZfA/eZ2TJC6/OShPOISDWX0nLj5a7/EFeQTHH3qfzRRysikrhURh08uLptRESkcilNKiMiImtPhVZEJGYqtCIiMVOhFRGJmQqtiEjMVGhFRGKmQisiEjMVWhGRmKnQiojETIVWRCRmKrQiIjFToRURiZkKrYhIzFRoRURipkIrIhIzFVoRkZip0IqIxEyFVkQkZiq0IiIxW+1y45K7imptrF+uJG6H9TZPOkJsPp4+PqXlxtWiFRGJmQqtiEjMVGhFRGKmQisiEjMVWhGRmKnQiojETIVWRCRmKrQiIjFToRURiZkKrYhIzFRoRURipkIrIhIzFVoRkZip0IqIxEyFVkQkZiq0IiIxU6EVEYmZCq2ISMxUaEVEYqZCKyISMxVaEZGYqdCKiMRMhVbWWkFBAf+680befH0M48aOYquttkg6Usbk677l436tt/66vPj+f9li682xHbbhoWeH8sDou+l3e08KClJaDTx2WVlozayOmZ23ise/ibZ5wMwOT/O9pkeXd5jZZqvYbqSZ1UrnvVaTY1czuzau14/DcccdTp06tdnvgGPp2esGbrk5p+KvUr7uW77tV1FRDXrffCWLFy0GoGuPjtwz6H46HNeNmrVqckDrfRJOGBQlHaASzYDzgGFV9YbuftlqHj895vf/CPgozvfItP32acWLL70CwLsTPmD3v+6ccKLMydd9y7f96t7nYkb9+yk6XnIWAJ9O+oxGjRsCUK9+XZYuW5ZkvOWytdD2AraPWngtgTpAE6C/uz+18sZmticwGDjZ3aeVu38S8BmwGOgKDI9eB+ASd/+k3LavRtvMBEYAtQEHDnH3rc3sG2BbwofAcKAmUBq9zsdm9jnwFmDAz8BJ7l5c7vUfAJYCm0evPRI4BtgMOA7YFOjq7qdH224V7fet7v6omR0N9Ile7sNo25KUf6IxaNCwPr/Nnbf8dnFxCTVq1KC4uHgVz8oN+bpv+bRfx552JHNm/cr4V99dXmi//ep7et7Qg/Mv68D8efN5b/yHCacMsrLrALgOmAKMB25z9zbARcCFFWy7DzAIOKZ8kY3UBwa4ezugJzDO3Q8GOgN3V/LevYCn3P1AYBR//jC6FRjs7gcAlxKKLkALoLe77w00JXxArOwbd28LTAW2dPcjgScIBRcAM2sAHAycCBwB1DCzIuBO4Ch3bwl8D2xSSf4qM++3+dRvUH/57cLCwpz8D1uRfN23fNqv408/mr0ObMmwJ+/EdtiG6/7Zm4GDr+Hc47tx/P7teHrU8/Toe3HSMYHsLbRlfgK6mNlDhNZmzQq2aQs0JrQWK+LR5U5Ax6jlei+wbiXbb0co8ABvVPL467D86/6m0f0zyxX6aYTW6Mo+iC5/JXyQAMwpv627zyN8qAwFHiW0ftcH5rj7jGib/u7+XSX5q8xbb0/kiMMPAWDPVn9l0qSpCSfKnHzdt3zar44nXECnEy7kvBMvwid/Tq+LB/D9tz8yf97vAPwyfSYNGzVIOGWQrV0HJYQPgQHAve7+vJmdC3SoYNu+hNbd3UBF/ahlX68/Bf7j7iPMbANCH3BFJgF7E/pL96rg8anA/sAYM9sVmB7dX7qafUppGzNrDuzu7ieYWR1C0X4YaGxm67n7bDMbHO3LhBTeMzZPPfU8rQ89gDdeG01BQQGdzr88yTgZla/7lq/7VaZfjxu4aUh/iouLWbpkKf2vuDHpSED2FtoZQC1gB2BwNDJgGqFl9yfuPszMTjaz9oTCt5+7919ps+uA4WbWGWhIKNAVuRF4yMxOBX7kzy3lK4B7zewKQgu7U2U7YWbrAcPc/cRK9/TPpgPNzOxDYD6hj3aJmV0APGtmxYQ+2olr8JqxKC0t5cKLrko6Rizydd/ydb/OO/EiAL4BOhzbNdEsFSkoLU2lIVZ9mNmRwC/uPtHMWgM93f2QpHOtjaJaG+uXK4nbYb3Nk44Qm4+nj09poG62tmiT9DVwn5ktA2oAlyScR0RynArtStx9KqGPVkQkI7J91IGISM5ToRXapIAAAAAQkklEQVQRiZkKrYhIzFRoRURipkIrIhIzFVoRkZip0IqIxEyFVkQkZiq0IiIxU6EVEYmZCq2ISMxUaEVEYqZCKyISMxVaEZGYqdCKiMRMhVZEJGYqtCIiMVOhFRGJmQqtiEjMVGhFRGKm5cZFRGKmFq2ISMxUaEVEYqZCKyISMxVaEZGYqdCKiMRMhVZEJGYqtCIiMVOhFRGJmQqtiEjMipIOILnLzHYAGgIlwPXA9e4+LtlUmWFmhwItgHeBz9x9UcKRZDXMbBN3/97M9nD395LOU54KraRjCHAp0A/oBdwM5HyhNbPrgU2A7YAlwNVAu0RDZYCZ7Qp0BuqU3efuHZNLlDlmNgT4HhgInGVmZ7n7pQnHWk5dB5KOpcBkoJa7v0P+fHDv5+5nA/Pd/UFgy6QDZcgDwAfAo+X+5Yvd3H0gQFRgd0s4zwry5T+GJKMUGAE8Z2anAr8nnCdTisysDlBqZjWA4qQDZch0dx+WdIiYFJhZE3efZWaNybLallVhJOecBrQCngcOjG7ng9uB94GmhD7a25ONkzHfmNlVwIeED0nc/aVkI2VMf+A9M5sNNAYuSDjPCtR1IOnYG9jL3UuBq4A9Es6TEe4+CjgcOAo40d0fTjhSptQGDDid0Od8erJxMsfdnwG2Bo4EzN1fTDjSCtSilXT0IxQkCK3Z54GcbyGZ2bVAI3fvYWajzOw9d78p6Vzpcvdzy982s+ZJZck0MzsFqEH4MLnZzG5x91sTjrWcWrSSjqXuPgPA3eeSP32Zx7l7DwB3PwU4NuE8GWFm/czsFzOba2ZLgf9LOlMGXQGMBc4ENgOOSTbOilRoJR0TzGyEmV1sZg8R+v7yQYmZ1QIws5rkz/+TIwjD1h4mDF37Idk4GVU2znmeuy8GGiQZZmX58gckybgEeAyoCzzm7pcknCdThgCTzOwJ4KPodj6YVVaE3P0Lwu8tX3wNvAfcZ2Z9CAcxs4bWDJM1ZmZHu/szZtZ55cfcfWgSmTLNzJoSzgz70t1nJp0nE8zsXuBtwkiROcBR7r5zsqkyx8zqu/t8M9vQ3X9OOk95Ohgma6NJdLnywZS8+NQ2s72Ac4GahPGZG7n7YQnHyoQuwKbAKKADeTAcz8yucfeBZvYIYdxz2f24e/tk0/1BhVbWWHS2FMB1wK7k11dQgMGEsbMnA58AtZKNk56ybyDAeeXuXgzsD0xNJlXGPB1dZnX3jgqtpGMU0Ago+5pWCryeXJyM+dXdHzGztu7e18xeSzpQmvL2G4i7fxxd/Y4w0qBOuYez5vemQivpWN/d9086RAxKo5nJ6lr4Ltos6UDpKPcNpLhsPgAAM7shoUhxGA08Seh7zjoqtJKOb81sU3eflnSQDOsO7EDoQhhBln8tXR0z60ToNtjOzI6M7q5B6IO+OrFgmTXN3fsmHaIyGnUga8zMfiJ87awD1AdmR7dL3X2jJLNlSjSloAFT3P2TpPOkw8xqE7oNehL61SHMITwjGu6V88ysK7AFMKXsPnf/d2KBVqJCK7ISMxsAHAJMAPYE/uvutySbKn1mVg9YlzC9ZWfg3+7+bbKpMsPMXiUc2Ps1uqvU3Xsml2hF6jqQtWZm+wB3ARsSzjI6z90/SjZVRhwBtHL3kmiaxLeBnC+0hDPC7gdOIrT8hgL5MGwNYLG7d0s6RGV0Zpik459Ae3dvThiXeVeycTLme/44hbMmf4yqyHXrAmOATdz9RsIELPniWzO72swOM7O2ZtY26UDlqUUr6fjV3acAuPskM1uQdKAM2Qj4zMw+BrYHlpjZeAB33yfRZOmpBfQA3jez7Qn96/miJvCX6B+EYwZZM5OcCq2kY4aZDQNeBnYHCstOy83xU3FPSTpATHoAxxMOiJ1Blk2OnY6Vp4DMNuo6kHR8CkwDtgF+IwwQb06OjzsltPQ2IuzHfcDW7v5trh44MrNNoqszgWHABoQpBX+t9Ek5zsweTzpDeWrRylpz937R5NE1gQJgI3d/O+FYmZBvq/t2j/7dQ/hKXRDdX0oYXZGPzk86QHkqtLLWzGw4YTmbesA6wFfAXomGyowVVvc1s5z+f+Lu3aOrz+XDMLWKmFkB0JIVT8HNmtPB1XUg6diOcAbVi4SDRotWvXnOyNfVfY+IhqvloyeAQUC36F/XZOOsKKc/qSVx89y91MzqufvMslUJ8kD51X0PIg+mE4w0BX40s6/540y+XB5FUV6zbN4XFVpJx/tmdgXhP+9IwvnzOS+a6Pu56OYrZrYH4TTjXHd00gFi9Gk0b/CPSQepiAqtpONB4EdgIeFsqgnJxonNKYRlUnJdU8KJJeXnD+6YTJSM2w/4zsxmkoXzbmiuA1lrZvamu++XdA5JjZlNBO4Eppfd5+4vJpeo+lCLVtLxu5ndDjhhNqicPlFh5WVRyj+WTcuipOG3cnPT5hUz24kw5nkTwgdJR3fPmlWZVWglHeOjyw2jy1z/epQTy6KsqXLn/c81s57A+0S/K3fPmtNU0zSYMKnRx9EUl/8C9k0403IqtJKOV1a6vdTMNnH37xNJk6Zyy6I0BFq6+7Vm9gJh/bBc1i66nEs4i2+b6HZWzQeQpsKy35+7f2Rmy5IOVJ4KraRjIOE01feB3YAlQB0zuzfHB8b3Aw6Prp9GGOaVs32ZZfMAmNn6wG7uPtbMLgL+k2yyjFpqZkcDbwAHEBafzBo6YUHSsQDY2d3bAbsQFsjbkTDfaS5b6u4zANx9LlCccJ5MeYSwmCaE4Wr5VGg7AecAbwFnkWWn4KrQSjqauvsigGhJlPXdfQm5/3c1wcxGmNnFZvZv4IOkA2VIPXd/HMDdRxBOnc5p5U6P/okwI9nuwJnR7ayhrgNJx1Nm9iZh/GxLYIyZdQMmJRsrbTcQxpfWJYzPzPUWepklZtYGeIdw5ls+tNT/DbQnjHwpOxhbEF1vkVSolWkcraTFzHYmzHkwOZr8uykw091z9g/LzMYC1wMXAo8DXdz94GRTpc/MtgZuJVp0ErjS3b9MNlX1oEIrGWNmR7v7M0nnSJeZvQK0Bl5w9zZm9rq7H5B0LqmcmX3Git/QlxLmSr7S3RPv+sn1vjTJLtusfpOcUIswE9QbZnYwedrFlm2TY6fpFcLKvtsRun0mErqABicZqowKraTFzArNbEMzK3D3XB9vWqYDoc/vJsL8AGcmmiY+WXVkPk1/cff/c/fF7v4q0NzdxxGdsZi0vPyklqphZicSWn5zgAZm1s3dxyYcK23u/jnweXTzsSSzZFK2T46dpiVm1pVwtuI+wGIz250sqXFZEUJyVm+glbvPMLMNCaew5nyhzWNPENYLmxbdLiV/Cm17wrJDxwGfEMbStiJLZidToZV0zCo3sP9nM/st6UCySlk9OfbaKHfKdxNWnKOiibs/n1CsP1GhlXT8ZmYvEla/3QOoa2bXA7h7z0STSUWyenLstdQDuJyw8GR5WbXwpIZ3yVozs3Mqeyxfp+PLZdEQqBaEZcezbnLsfKZCK2vNzBoBB1Lu4Iq7583BI8l+ZvYTf56es4As+xBR14Gk4yXCGUa/RrdLyaOj9Pkm2yfHXhvu3jzpDKlQoZV0zC2bgk9yQlZPjp0OM9uBcDCsMfAwMCmbzlJUoZV0vBiNXZxSdoe758twoXyU1ZNjp2kwcC5wLzCcMIewCq3khf2B2oR+WsivcZn5KKsnx06Xu39hZqXu/ouZzUs6T3kqtJKO+u7eOukQkrJOhNm7biR8C8mnU3Bnm1kXoJ6Znc4fxw2yggqtpGNS9Ef9IX8s9vdZspFkZWZW5O7L+GNy7LL5WvNJJ6AnYejaHtHtrKFCK+nYJfpXJqsGictyOTE5dprudvczkg5RGY2jlbSYWRNgK+Ard5+ZdB6pnszsCcKimp8RzdgVLauUFdSilbVmZqcQVsKdCuxoZn3dPZ8W/Msr2T45dpr+AowmtNKzrrWuQivp6A7s7u7zzawB8DL5tbJqvnkFGEUYdbA3cB5wP2Fo1H4J5kqbu++UdIZV0cTfko4Sd58P4O7zgEUJ55FVy+rJsTPJzK5JOkN5atFKOr40s9sIY2f3B7TQX3bL6smxM+ytpAOUpxatpGMoMBtoQzgr585k48hqtCf0Zd5E6L88izAReFZMjp0OM9t4pbs2TCRIJfLxk0yqziCgg7tPMbNBwAOEM44ki+TK5Nhpejw6620ZYR8bASOTjfQHFVpJxzJ3nwLg7l+ZWd719eWJnJgcO02XEkYdNAJud/f7Es6zAo2jlbVmZo8AXwNvE9ZnapHNg8Yl/5hZ23I39yV0Y/UFcPeXkshUEbVoJR3nAl2BIwljaQcmG0cqkiuTY6+ldivd9ui+UsJ8yVlBLVoRyVlmVquyx3RmmIhUuWyfHHstlZ+/oYzODBORxGT15Nhrw923TDpDKlRoRaqRbJ4cOx1mdixwIVCT0KJt4u47J5vqDzphQaT6yOrJsdN0LWG0wTTgQeCTRNOsRIVWpProBGxJlk6OnaZZ7v42gLs/QFjpN2uo60Ck+sjqybHTtNjMDgBqmtlhQLOkA5WnQitSfdQxs53J0smx0zSBUM8GAgPIshnJVGhFqo+snhx7bZhZJ8K8utvxx7L3NYCFiYWqgE5YEJGcZWa1geaEhRmvi+4uAWa4e9Ysp65CK1JNmdk17q7TpquARh2IVF9ZNTl2PlOhFakmzKyJmbWOrl8IfJhwpGpDhVak+hhJmOcAYA5aSLPKqNCKVB913f1xAHcfAdRNOE+1oeFdItXHUjNrA7xDmKg9q8aa5jO1aEWqj/MIE6+8C1wAdEk2TvWh4V0i1YiZ7QhsD3zm7h8lnae6UItWpJows0sIc9HuAww1sysSjlRtqNCKVB/tgP3d/TLCQoanJZyn2lChFak+Ctx9GYC7LwWWJpyn2tCoA5Hq400zexx4A9gPeDPhPNWGWrQi1ccg4BWgAbA7MCLZONWHCq1I9fEgYSrBXQmzXd2ebJzqQ4VWpPooAl4HGrn7SMK8rVIFVGhFqo9ahO6DN8zsYHSMpsqo0IpUHx0AB24CmgJnJpqmGtGZYSIiMVOLVkQkZiq0IiIxU6EVWUtmVsfMzsvg6x1kZiNX8XgHM7sxxddKeVuJnwqtyNprRph6UGSVNLxDZO31ArY3s2sJjZZ9gPpAJ+B+d98LwMzeAU4nLB8zHGgSPf8Sd/+kohc2s4uAE4GawNzoOsDeZjYOaAj0dfdnzexAwlLbxcCXaJ7ZrKMWrcjauw6Y4u79o9tT3X0fYGEl2/cExrn7wUBn4O6KNjKzQkIxbu3u+xOKbcvo4d+B1sBRwJ1mVoMw9eGJ7n4g8ANhGJdkEbVoRTLHK7m/ILrcCTjEzMqmJ1y3whdxLzGzJcAjZjYf2IRQbAHedPdSYIaZzQXWB5oDj5kZwDrAS4SWrWQJFVqRtVfCit8Ky9bgWgRsELU2GwBbRvd/CvzH3UeY2QZU0r9rZjsDx7v7nmZWF3ifP4p1y2ibZoRuipnA98Bx7j7XzI4F5gObZWgfJQPUdSCy9mYAtczspvJ3uvt0YCwwERgKfBE9dB1wqpm9CrwATKrkdb8Afjez96LX+QnYKHpsHTN7GRgDdHH3YuBS4FkzG09YC6yy15WE6MwwEZGYqUUrIhIzFVoRkZip0IqIxEyFVkQkZiq0IiIxU6EVEYmZCq2ISMz+H5th04xn3kx6AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot object at 0x0E978F30>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 16.0, 'true label')"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "Text(92.68, 0.5, 'predicted label')"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.heatmap(mat1.T, square=True, annot=True, fmt='d', cbar=False,\n",
    "            xticklabels=train.target_names, yticklabels=train.target_names)\n",
    "plt.xlabel('true label')\n",
    "plt.ylabel('predicted label')\n",
    "plt.show()\n",
    "\n",
    "sns.heatmap(mat2.T, square=True, annot=True, fmt='d', cbar=False,\n",
    "            xticklabels=train.target_names, yticklabels=train.target_names)\n",
    "plt.xlabel('true label')\n",
    "plt.ylabel('predicted label')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "That got our accuracy score up about 5%! What if we change other things?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Pipeline(memory=None,\n",
       "     steps=[('tfidfvectorizer', TfidfVectorizer(analyzer='word', binary=False, decode_error='strict',\n",
       "        dtype=<class 'numpy.float64'>, encoding='utf-8', input='content',\n",
       "        lowercase=True, max_df=1.0, max_features=3000, min_df=1,\n",
       "        ngram_range=(1, 1), norm='l2', preprocessor=None, smooth...   vocabulary=None)), ('multinomialnb', MultinomialNB(alpha=1.0, class_prior=None, fit_prior=True))])"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Only include the top 3000 features\n",
    "model3 = make_pipeline(TfidfVectorizer(max_features = 3000), MultinomialNB())\n",
    "\n",
    "# train model\n",
    "model3.fit(train.data, train.target)\n",
    "\n",
    "# predict labels on test data\n",
    "labels3 = model3.predict(test.data)\n",
    "\n",
    "mat3 = confusion_matrix(test.target, labels3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[373,   7,   7,   2],\n",
       "       [ 21, 366,   7,   0],\n",
       "       [  6,   3, 386,   3],\n",
       "       [  8,  16, 116, 111]], dtype=int64)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mat3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8631284916201117"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(test.target, labels3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "An even higher accuracy score! But what has changed in how it categorizes?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot object at 0x0EB2EE50>"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 16.0, 'true label')"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "Text(92.68, 0.5, 'predicted label')"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot object at 0x0EA0CCD0>"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 16.0, 'true label')"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "Text(92.68, 0.5, 'predicted label')"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.heatmap(mat2.T, square=True, annot=True, fmt='d', cbar=False,\n",
    "            xticklabels=train.target_names, yticklabels=train.target_names)\n",
    "plt.xlabel('true label')\n",
    "plt.ylabel('predicted label')\n",
    "plt.show()\n",
    "\n",
    "sns.heatmap(mat3.T, square=True, annot=True, fmt='d', cbar=False,\n",
    "            xticklabels=train.target_names, yticklabels=train.target_names)\n",
    "plt.xlabel('true label')\n",
    "plt.ylabel('predicted label')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Suggests that including too many features causes the classifier to take features into account that aren't truly relevant\n",
    "\n",
    "Porque no las dos?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Pipeline(memory=None,\n",
       "     steps=[('tfidfvectorizer', TfidfVectorizer(analyzer='word', binary=False, decode_error='strict',\n",
       "        dtype=<class 'numpy.float64'>, encoding='utf-8', input='content',\n",
       "        lowercase=True, max_df=1.0, max_features=3000, min_df=1,\n",
       "        ngram_range=(1, 1), norm='l2', preprocessor=None, smooth...   vocabulary=None)), ('multinomialnb', MultinomialNB(alpha=1.0, class_prior=None, fit_prior=True))])"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model4 = make_pipeline(TfidfVectorizer(max_features = 3000, stop_words = 'english'), MultinomialNB())\n",
    "\n",
    "# train model\n",
    "model4.fit(train.data, train.target)\n",
    "\n",
    "# predict labels on test data\n",
    "labels4 = model4.predict(test.data)\n",
    "\n",
    "mat4 = confusion_matrix(test.target, labels4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[375,   9,   3,   2],\n",
       "       [ 25, 368,   1,   0],\n",
       "       [  8,   4, 378,   8],\n",
       "       [ 10,  17,  85, 139]], dtype=int64)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "0.8798882681564246"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mat4\n",
    "accuracy_score(test.target, labels4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([0, 1, 2, 3], dtype=int32), array([584, 593, 599, 377]))"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Wow! Now we're up to almost 88% accuracy! Recall, what was our baseline?\n",
    "np.unique(train.target, return_counts=True) #returns two arrays, one with value counts of each category"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.20499657768651608"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "599 / np.sum(train.target) # our baseline if we were looking purely based on frequency in our training data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Maybe we think that 3000 features isn't nearly enough. What if I choose 50,000?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Pipeline(memory=None,\n",
       "     steps=[('tfidfvectorizer', TfidfVectorizer(analyzer='word', binary=False, decode_error='strict',\n",
       "        dtype=<class 'numpy.float64'>, encoding='utf-8', input='content',\n",
       "        lowercase=True, max_df=1.0, max_features=50000, min_df=1,\n",
       "        ngram_range=(1, 1), norm='l2', preprocessor=None, smoot...   vocabulary=None)), ('multinomialnb', MultinomialNB(alpha=1.0, class_prior=None, fit_prior=True))])"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model5 = make_pipeline(TfidfVectorizer(max_features = 50000, stop_words = 'english'), MultinomialNB())\n",
    "\n",
    "# train model\n",
    "model5.fit(train.data, train.target)\n",
    "\n",
    "# predict labels on test data\n",
    "labels5 = model5.predict(test.data)\n",
    "\n",
    "mat5 = confusion_matrix(test.target, labels5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8561452513966481"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(test.target, labels5) # a lower accuracy score here..."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "What about min_df and max_df?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Pipeline(memory=None,\n",
       "     steps=[('tfidfvectorizer', TfidfVectorizer(analyzer='word', binary=False, decode_error='strict',\n",
       "        dtype=<class 'numpy.float64'>, encoding='utf-8', input='content',\n",
       "        lowercase=True, max_df=1.0, max_features=None, min_df=0.5,\n",
       "        ngram_range=(1, 1), norm='l2', preprocessor=None, smoo...   vocabulary=None)), ('multinomialnb', MultinomialNB(alpha=1.0, class_prior=None, fit_prior=True))])"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model6 = make_pipeline(TfidfVectorizer(min_df = 0.5, stop_words = 'english'), MultinomialNB())\n",
    "\n",
    "# train model\n",
    "model6.fit(train.data, train.target)\n",
    "\n",
    "# predict labels on test data\n",
    "labels6 = model6.predict(test.data)\n",
    "\n",
    "mat6 = confusion_matrix(test.target, labels6)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.31075418994413406"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(test.target, labels6) #uh oh..."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "What happened? Recall what min_df DOES!\n",
    "Only looking at words that occur at least 50% of documents!?!? Motivation "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Support Vector Classifier: an alternative\n",
    "\n",
    "Let's take a look at how well this does"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\users\\cassi\\appdata\\local\\programs\\python\\python36-32\\lib\\site-packages\\sklearn\\svm\\base.py:196: FutureWarning: The default value of gamma will change from 'auto' to 'scale' in version 0.22 to account better for unscaled features. Set gamma explicitly to 'auto' or 'scale' to avoid this warning.\n",
      "  \"avoid this warning.\", FutureWarning)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Pipeline(memory=None,\n",
       "     steps=[('tfidfvectorizer', TfidfVectorizer(analyzer='word', binary=False, decode_error='strict',\n",
       "        dtype=<class 'numpy.float64'>, encoding='utf-8', input='content',\n",
       "        lowercase=True, max_df=1.0, max_features=3000, min_df=1,\n",
       "        ngram_range=(1, 1), norm='l2', preprocessor=None, smooth...f', max_iter=-1, probability=False, random_state=None,\n",
       "  shrinking=True, tol=0.001, verbose=False))])"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.svm import SVC # \"Support vector classifier\"\n",
    "\n",
    "model = make_pipeline(TfidfVectorizer(max_features = 3000, stop_words = 'english'), SVC())\n",
    "# model = SVC(kernel='linear', C=1E5)  # 10^5: a large & hard margin\n",
    "                                     # keeps training accuracy high\n",
    "model.fit(train.data, train.target)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([2, 2, 2, ..., 2, 2, 2], dtype=int32)"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels = model.predict(test.data)\n",
    "labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.2779329608938548"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(test.target, labels) #oh my--that's awful! What happened?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's take a look at SVC and what it does: https://scikit-learn.org/stable/modules/generated/sklearn.svm.SVC.html\n",
    "\n",
    "Was that really helpful for what it is? Good ol' Wikipedia: https://en.wikipedia.org/wiki/Support-vector_machine\n",
    "* Essentially, maps each category to a point in space based on features, then finds the optimal line to separate each category"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Pipeline(memory=None,\n",
       "     steps=[('tfidfvectorizer', TfidfVectorizer(analyzer='word', binary=False, decode_error='strict',\n",
       "        dtype=<class 'numpy.float64'>, encoding='utf-8', input='content',\n",
       "        lowercase=True, max_df=1.0, max_features=3000, min_df=1,\n",
       "        ngram_range=(1, 1), norm='l2', preprocessor=None, smooth...f', max_iter=-1, probability=False, random_state=None,\n",
       "  shrinking=True, tol=0.001, verbose=False))])"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = make_pipeline(TfidfVectorizer(max_features = 3000, stop_words = 'english'), SVC(kernel='rbf', C=1E5))\n",
    "# model = SVC(kernel='linear', C=1E5)  # 10^5: a large & hard margin\n",
    "                                     # keeps training accuracy high\n",
    "model.fit(train.data, train.target)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "labels = model.predict(test.data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[373,   7,   3,   6],\n",
       "       [ 15, 372,   5,   2],\n",
       "       [  4,   4, 366,  24],\n",
       "       [ 15,  16,  36, 184]], dtype=int64)"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cm = confusion_matrix(test.target, labels)\n",
    "cm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9043296089385475"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(test.target, labels) # A LOT better! What happened?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot object at 0x16559CD0>"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 16.0, 'true label')"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "Text(84.40000000000006, 0.5, 'predicted label')"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYoAAAFaCAYAAAAAQBo/AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4wLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvqOYd8AAAIABJREFUeJzt3Xd0VOXWx/FvEpooRQEFxV62WHj1KipcUETsBbtYUBQLWBCxoCAICtgbegVBFCxYsFwQRWxcBa9d9IrgtmKlSJPekrx/PCcakUyGzISTmfw+a83KzDlnzuzJJLPP03MKCwsREREpSW7cAYiISMWmRCEiIgkpUYiISEJKFCIikpAShYiIJKREISIiCSlRiIhIQkoUIiKSkBKFiIgkVCXuAKT8dM6pnXXD7ocs/DruEMpPlepxR1A+1qyMO4LyU2eLnFRPsT7/p0MKF6X8emWhRCEiEqNMqNZRohARiVFuTiyFhPWiRCEiEiOVKEREJKHcil+gUKIQEYmTShQiIpJQFbVRiIhIIqp6EhGRhFT1JCIiCeWo6klERBJRiUJERBJSG4WIiCSkXk8iIpKQqp5ERCQhVT2JiEhCKlGIiEhCuVT8IoUShYhIjKpU/DyhRCEiEie1UYiISEKqehIRkYRUohARkYTU60lERBLKhBJFJiQzqQBycnPpMPxfXD35Va58azz1d9ieTk8+QveJL9F94ksM+P5zOj35CADt+vemx3tvcs27b7Bds31ijnz9fTZ1Gh06dwXgiy+dVkefSIfOXenQuSsvv/ZGzNGlx7z58znoyGP59vsZcYeSskz/vKrk5CR9iy3G2F65kjOzWe7ecK1tHYH57j42nqhK1vTYIwG4veVh7HJQS065ayCDjz8dgJp163LFxHGMvuJatt6rKdsf0IxbD2hDvW23ocuYJ+m/1z/jDH29DHt0FGPHT2CjjTYCYNqXX3PuGady3pntY44sfVavXkOfAbdQo3r1uENJWTZ8XhlQoFCJoiJx9xEVMUkAfDbmJZ64MFy1bbbtNiyaPeePfcf068nE+x5k0azZ/PTp/xh0+AnRcVuzaPZvscRbVts03pL7bu3/x+OpXzr/mfwuZ154KT1vuoUlS5fFGF163HrPvbQ/6UQ2b9Ag7lBSlg2fV25O8re4ZHSJwsw2Ah4BtgWqAlcAFwI7AnnAXe7+tJn9B/gM2ANYAkwCDgfqAocB7aJbbaA+cKO7P7fWa/UGTgB+A2oCvYHWQAtgE6ATcDawL1ALmO7u55pZX2BXYHNgU+Ayd58MVDezUcA2wDzgZKAXMAsYCgwC9gOqATcA/wWeJiT3qkBnd/885V/ieijIz+ecEUPY64RjGHry2QDUalCfXQ85iNFXXPuX49r1783BXTvz9GVXb8gQU3Z4m9b8/OvMPx433b0Jp7Q7hj2aGIMffpR/PfQIPS6/JL4AU/T82HFstummtGpxAEMfGRl3OCnLhs8rE7rHZnqJojMww92bAx2Bg4C57t4CaAv0N7P60bEfuPshQHVgmbsfCkyLngPhy/5QQuK4y8z+SKJm9n/AkUAz4HigUbEYpkev9wuwIDpvC+AAM9sqOmaZu7cBzgL+Vez1erp7S6AOsHexc7YD6rv7fsAR0evuB/wexdGVkNQ2uJEdO3PDLv/grGGDqFazJv84+Xg+HDWawoKCvxw35vqb6LGlcejVl1N/h+3jCDUtDm3dij2aWHT/QKb51zFHlJrnxrzIf9/7gA4XdGG6f0WPPv34be68uMNKm0z8vFSiKH8GjAdw96lm1gV4PXq82MymEUoXAJ9EPxcSEgTAAqBGdP8tdy8AZpvZAqABUHSp0oSQaPKB5Wb2UbEYPPq5HNjczJ4klFo2IVz5A7wZxfSFmRW1S8x39xnR/VmEUkrx9/Vu9JxZwPVmlgvsDIwBVgP92YD2P6s9dRtvyYRb7mLVsuUUFhRQkJ/Prm1b83L/2/8M/OAD2fukdjx16ZWsXrGC/NWr/5ZEMkmnrlfR+6rLabr7brz74cfsvqvFHVJKnhj+4B/3O1zQhb49e9Cgfr0YI0qvTPy80nW1bmZ5wDDC90c+cC6hduO+6PFK4Gx3n21mFwAXAWuA/u4+LtG5Mz1RTCdcbY8xsx2A0wm/jBfMrBawJ/B9dGxhKefaB8DMtiBcrc8ptu8L4LLoy7oqf736L/oWPBLY2t1PM7MGhGqqomuAfYDHzWwPQsmjtHimA6dE8dQBngFuBWa6+2Fm1hwYCBxcyntKmynPj+WcRx7gyrfGk1e1KqO7XcualSvZwnZm7ncz/jjuq7cm849Tjufqya+Sk5fHW/8axrwZP2yoMNOub4/u3HT7PVStWpX69TbjpusyqyqtssnEzysvfb2ZjgVw93+aWWvgLkL1+mXu/qmZXQT0MLPbCLUS+xIulCeb2WvuvrKkE2d6ongQeNjM3iK0SRwBXGJmk4GNgH7uPscsqauKhmb2BqEa6GJ3zzez7sA37j7WzF4G3gPmEq7oV6/1/A+A3mb2HiFZfQdsGe3bOzr3xsAFScQyFmgbvY8qQD9CG8vTZtaNcHVwYzJvKl1WLVvGsNM6/m37jXvs/5fHhQUFPHlx9w0UVflovGUjnnl4CAC772o8NXxwzBGVj8eGZcf7yvTPK11pwt3/bWZFJYNtgdmEtsyimpEqwApCNfY7UWJYaWbfAE2BD0uMsbCwtAvt7Bd1S93V3a8tYf/mwMnu/oCZVSeUMNq4+49JnLsvMMvdh6Qx5KR0zqmddR/ukIUVv865zKpkfnfVdVpT4oVq5quzRcrf86M33SLp/9PrN697EaHDTpGh7j60+DFmNpJQo3Gyu78abWsBDAcOJHTk2dPde0T7HgUedffXS3rdTC9RbChzgWZm9iGhyuihZJKEiEhp1ifTRElhaCnHnGNmPYD3zWw34BhCj8qj3f03M1tEaLsoUovQdlsiJQrC+IVS9hcQGobKcu6+ZXmeiFQOOWlqozCzDkBjd78ZWEZoPz2B0Gjd2t3nR4d+AAwwsxqEXqBNgKmJzq1EISISo7z0nep54BEze5vQ6aYbYZzZj8DzUVvtW+5+g5kNIownywV6ufuKRCdWohARiVG6Oj25+1Lg1LU2b1bCscMIXWmTokQhIhKjnAwYma1EISISo4qfJpQoRERipUQhIiIJZcLCRUoUIiIxyoTZY5UoRERiFOPCdUlTohARiVEG5AklChGROKnqSUREEqr4aUKJQkQkVur1JCIiCWlktoiIJKQShYiIJJQBeUKJQkQkTkoUIiKSkNooREQkobyKnyeUKERE4pQBeUKJQkQkTulaM7s8KVFksSGLZ8QdQtp1rrVd3CGUmyFLfow7hPJRGHcAFVvFTxNKFCIisVKiEBGRhFT1JCIiCeVlQLcnJQoRkRhlQIFCiUJEJE5KFCIikpDaKEREJKEMyBNKFCIiccrNgHnGlShERGKUmwFFCiUKEZEYZUCeUKIQEYmTGrNFRCShnNy4IyidEoWISIxUohARkYTU60lERBLKgAKFEoWISJzUPVZERBLKgDyhRCEiEic1ZouISEIZkCeUKERE4pSrhYtERCSRdJUozKwq8DCwHVAd6O/uY6N9ZwCXuXvz6PEFwEXAmui4cYnOnQFjAkVEslduTk7St1KcBcxz91bAkcD9AGa2F9AJyIkeNwS6Av8EDgduNrPqCWNM6R2KiEhKcnKSv5ViNNC72OM1ZlYPuAXoVmz7fsA77r7S3X8HvgGaJjpxiVVPZnZhSfvcfWipIUul8PzYl3hh3EsArFy5iulffc07r46jdq1aMUeWnJzcXM4adh8NbWcK8vMZee7FtBvQhzoNNweg3nbb8N17HzH89HM58bab2Kllc3KrVGHy0EeY/NDImKMvu3nz53Pimefw8AP3seP228UdTko+mzqNO/41hMcGD2Le/AVcP/A2Fi1eTH5BAbfd0IttGm8Vd4gJrU+vp+h7ufh389Ci72N3XxIdUwt4lpA0hgNXAMuLPac28Huxx4uBOoleN1EbRaNkg6+soiLdce5+Y9yxxOXE447mxOOOBqDfLbdzUrtjMiZJADQ99kgAbm95GLsc1JJT7hrI4ONPB6Bm3bpcMXEco6+4ll1at2LznXbgthZtqVKtGn2++IBPnh3DsoUL4wy/TFavXkOfAbdQo3rC2oaMMOyxUYwdP4GNamwEwO33D+bYIw7lqLZteO+jT/huxo8VPlGszxQeUVIo8ULdzLYGXgAeAL4GdgYGAzWA3czsHuBNoPg/aS0g4R9yiYnC3fsVe/G2wPbA+8BXpbyXSsPdPwU+jTuOiuDzadP55rvvueHaq+MOZb18NuYlPh/3CgCbbbsNi2bP+WPfMf16MvG+B1k0azbLFizk508/B6CwsJDcvFzyV6+OJeZU3XrPvbQ/6USGPpK5JaIi22y1Jffd0p9r+g4A4JPPPsd22pGOl17BVo0a0qt715gjLF0aG7O3AF4FLnX3N6LNu0f7tgOecvduURvFADOrQWj0bgJMTXTuUns9mdlAoHF0slXAdcDpZXsrmcPMdgFGAKsJPQPOJrz3/YBqwA2E4ltnd29f7HkNgKcJ7T9Vgc6Eot1oYCbhdzne3XuZ2R7AXdGxdYGu7v5fM+sEdAHygDHu3tfMTgG6A/nAZHe/tnx/A+vnwYdHcskFneIOo0wK8vM5Z8QQ9jrhGIaefDYAtRrUZ9dDDmL0FeHXvGblStasXElulSp0HPkgk4aOYOXSpXGGXSbPjx3HZptuSqsWB2RFoji8TWt+/nXmH49/mTmL2rVqMeL+u7n/oREMe3QUl19Usf8u0zjgriewKdDbzIraKo509+LVTrj7LDMbBEwifPf0cvcViU6cTPfYlu5+oJlNdPeRZtalDG8gEx0KfEz4cm4FnAfUd/f9oox8KfD6Op63HyGBnAHsRqgPXEzosnZ4tG+ymf2DUCy80t0/j7qvnWtm3wDXEhqXVgJ3mtk2QD9gX3dfZmaPmdmh7v5aOb339bJo8WK+m/EDBzTbJ+5Qymxkx8680GNzerz/Jv12249/nHw8H44aTWFBwR/H1KxblwuffZSv/jOZCbfcFWO0ZffcmBfJycnh3fc/ZLp/RY8+/Rh89x00qF8v7tDSom6dOrQ58J8AtGnVgrsHD4s5otKlaz0Kd78cuLyEfTOAA4o9HgYk/ctJJlFUiYoohWaWR7iirQyGAz2AVwhf7h8A70LIyMD1ZtZ6Hc8bT0gAYwilkf7R9s/cfT6Amb0PGPATIfsvJ9QTLgJ2AKYWuwq4wsz2AxoAL5sZ0bE7pPPNpuLDTz6lxX7N4g6jTPY/qz11G2/JhFvuYtWy5RQWFFCQn8+ubVvzcv/b/ziuao0adHtjLK/feT8fjHomxohT88TwB/+43+GCLvTt2SNrkgTAPv+3J2+98x7HH3U4H075jJ122D7ukEqVCVN4JJPL7iZcWe9BaKN4oFwjqjjaAZPc/RBCtdFFQDMAM6tjZhNKeF5rYKa7H0ZIEgOj7U3MrGaUbPcHpgGDgBvc/Rzgc0I/52+BXYv6NZvZs8BsQlI51N1bA/cRPosK4fsffqBx4y3jDqNMpjw/lm32bsqVb42n64QXGN3tWtasXMkWtjNzv5vxx3EHdu5E/R22o+UF59B94kt0n/gS9bbbNr7AZZ16XH4JY8ZPoP35XZj03gd07tgh7pBKl5uT/C0mOYWFhaUeZGabAjsC37v7vHKPqgIwsx2BxwntEwWEKqiOwN6Eklg/Qpezzu7e3sxuI3RJ+5bQRrExofR1I6EDwERCctgCeNbdbzGz7oQ2jNnAz4SqrUPNrGO0vRB40d0HmtlZwMWEdosZwLnuvizhm1gyv/QPN8N0rrVd3CGUmyFLfow7hPKxemXcEZSfuluk/O29qM3eSf+f1n5zSizZotREYWb7Ag8SvuB+BC5y9883QGxZo1iPgwNKOzatlCgyihJFBkpHomj7j+QTxeufxJIokql6GgR0cPfGhOqXylL1JCJS/tI4NLu8JNOYvdzdpwFEvXNWlXNMWWftHgciIkVyMnnN7GJTeKw2sweAtwldPxdtiMBERCqFDOj1lMwUHu9GP43QTVQjkUVE0iSjSxRrTeHRiDDKOAfIzH6QIiIVUV7Fn8Q7mSk8hgPNCd09NwK+Q/XtIiJpkS0D7poQJpaaQJiSIuGcICIish4yYMBdMolisbsXAhu7+1zChHgiIpIOWdI99mMzuwr41cyeSvI5IiKShHRNClieSv3Sd/eeZrYJocrpSCrQHEMiIpkuJ5Mbs83sZsJcQ2trTpj3XEREUpUBjdmJShRfbrAoREQqqwwfR5H5y1+JiFRwmdA9Vg3TIiJxyuQShYiIbACZXKIwsz4l7XP3G8snHBGRyiUTej0linB2dGsONCSs3LYZsNcGiEtEpFLIyclJ+haXRI3ZDwKY2YnufnG0+Qkze22DRCYiUhlkQBtFMmWeetH60ZiZAbXLNyQRkUokS6bw6AY8aWZbArOADuUbkohI5ZHR61EUcffJZnYosC3wnbsvKf+wREQqiUzu9VTEzE4Cro+OfcbMCt29f7lHJqnLzYs7grQbsuSHuEMoN91qbxt3COXi7p8/ijuEcpNTd4vUz5HhvZ6KdCcsVDQX6A+cUK4RiYhUJlmyHkWBu68ECqN1KZaWc0wiIpVHljRmTzKzUUBjMxsCfFjOMYmIVB7Z0EYRrUdxBDAF+NLdXyz/sEREKolMThRmlgfkAU8BpwFvAnlm9qa7t9lA8YmIZLe8it/pJFGJ4jzCAkUNAY+2FQCTyzsoEZFKI5NLFO4+DBhmZue5+8MAZlbN3VdtsOhERLJdBiSKZHo9VTGzu6P748xMI7NFRNIlNzf5W0yS6fXUGWgR3T8aeBt4rNwiEhGpTLKkRJHv7isA3H01UFi+IYmIVCJZMo5ijJlNAj4A/gGMLd+QREQqkQzv9QSAu/c3s3GAAY+6+2flH5aISCWRyVVPZnZ+9PNm4FTg/4DTzGzgBopNRCT7ZXjV00/Rzy83RCAiIpVSmhOAme0P3Orurc1sc2AYsClhAPXZ7v6tmV0AXASsAfq7+7hE50yUKJab2YHA9+kJX0RE1paTxm6vZnYNYXG5oslbbwOecPdnzOxgYFczWwp0BfYFagCTzey1aPLXdUqUKLpEP3cEqhEmA9wbWAK0TuG9iIhIkfSOj/gWOJE/hzD8E/ifmb0OzAAuBw4B3okSw0oz+wZoSoIJX0uM0N1Pd/fTgd+Afd39AmB/YEXq70VERID1aqMwswvN7KNitwuLn8rdnwNWF9u0HbDA3dsCPwI9gNrA78WOWQzUSRRiMt1jG611/OZJPEdERJKxHiUKdx8KDF2Ps8/jzyENLwIDgI+AWsWOqQUsTHSSZBLFcOALM5sK7AbcsB5BiohIIuXbm2kycBShKupA4AvCmLgBZlYDqA40AaYmOkky4yj+ZWaPAbsC37n73BQDFxGRIuWbKK4EHjKzLoTqpjPcfYGZDQImEZofehXNvlGSUhOFme0ODAHqAk+Y2dTSulKJiEiS0pwo3H0GcEB0/wfg0HUcM4zQbTYpyVSODQLOBeYSqqH6JntyEREpRV5e8reYJNWK4u7fAIXu/huhhVxERNIhw0dmF5lvZhcBG5tZe0ppHZfKIz8/n+tvGsj3M34gLzeXm/v1YZutG8cdVtocf/rZ1NpkYwAab7klN/frHXNEycvJzaX90EE02GVnCvPzGdXpElYuXsxpQwdRs25dcvLyeKJjZ+Z99z1NjmjL4b2vBeDnKZ/x7KVXxhx98vLz8+l9x318/9Mv5OXmMrBHNzauuRG977iPRYuXkF9QwK3XdWebrRqVfrK4xLjORLKSSRSdCEuiziWM5OtUrhGtJzOb5e4Nzewe4C53/7GE454iDF9P2wp9ZjYCeMrdX0lwTIlxRb0OznL3h8ysIzDf3TNmdt6Jb08C4KkRD/H+Rx9z8533MPieO2KOKj1WrgyDVB8bNjjmSMpmj2OPBGDQgYez00EtOf7OASxfsJCPR43m09EvsFPrVmyx684smfMbx916E/e3OZql8+bT5qrL2bh+PZbOnRfzO0jOxP9+AMCT99/O+1P+xy0PPETtTTbh2LatOfLgVrw35X989+PPFTtRZMCkgMkkisHufma5R5Iid+9Wyv72GyqWtV43UVwNgfOBh9x9xIaJKH3aHtya1q1aAvDrrzOpX2+zmCNKny+/+prlK1Zw3sVdWbMmn+6XdmGvpnvEHVbSPh/zEl+MC9cvm267NYtn/8bOrVvx6/++oMuEMcz/4Ude6NaDHVo2Z+bUabS7YwD1tt+O9x5+NGOSBEDbVs1p3Xw/AH6d/Rv1Nq3LB1P+h+24Hed278VWDbeg52UXlnKWmGVJoqhhZk2Br4ACgNKuys1sF2AEYYTgGuBsoDvQMjpklLvfa2Y7Aw8RpghZBrSP2kGKzjM1et2VhJX2hgP1ot1d3f3zYsf+JzpmLjCK0D/YgTbuvpOZzSB08W0YnacqYRGmru7+mZl9DbxDmE59NnCSu+cXO//fYo12XRTNr1KHMO3JHMLAlnnAy4Q+zJ2juO+MficLgDOBXsBuZtaH0F40i9AT4UFg6+g54929d1R6WUkYadkI6OjunyT6HDaEKlWq0KN3X16b+BaDbr857nDSpkaNGnTqcAannNCOGT/+xAWXXcErzz9NlSrJ/MtUDAX5+ZzxyGCaHn8Mj5x6Dvt3PJNlCxcy+PB2HH79NRxyTTfmfPU1O7Vuxe3/aMnKJUvp+tYrzHj3A377+tu4w09alSp59Lj5Ll6f9C739ruOF8a/Tu1am/DIXQP418gneejJZ+l63llxh1myDEgUyVSO7QKMAaYTvniTmU32UOBjoC1hJOAJwPaELlstgTPMbE/gDuBmd29O+HLce63zbALcFE0l0hN4w90PBi4ESqoT6AX8290PAkbz92R4BzDI3Q8kzHsyPNq+A9A7iqUB0Gwdz1tXrB+7exvgPqBjtK0hcJi731bs+ccDzwMHAQ8TZnMcAExz9xuLHbc18J67Hx79rroU2/dDtP2+6HdQIdx6U18m/Hs0vW8cyLLly+MOJy2233YbjjvqCHJycth+222oW6cOv2XQlXaRUed2YcCu+3Dag/eyfOHvTB37MgBTx73C1vvuzdJ58/nxo09YPHsOq5Yu5dtJ77DVXk1jjnr93Xpdd155bCh97riPWptsTJsW+wNwcIv9mOpfxxxdKbKh15O77wnsBDQHdnT3HZI473DClf0rwKWEL8VJ7l4YLaf6HmGUtwHvRq/zjLu/uq4Qop97AudFJYeiaXPXpQnw3+j+pBL2vx295qeEL2aAue5eNLX6T4RZFYsrKdaPo5+zgJrR/e/XUeoaSJj+5A3gZP46H0tx84FmZvYEcDehZFRkSoL4Nrh/j3uZB4ePAGCjGjXIyc0hLwMa5pLx7JgXueXuQQDM/u03lixdSoP69Up5VsWx71mn0bZHdwBWLVtOYUEh37z9DrsddRgAO7ZqwawvvuSnjz+l0e67sXG9zcjNy2O7/Zsxa1rmrCww5tU3efCJZwDYqEZ1cnJyafZ/e/D2+x8B8NFnU9lpu23jDLF0GdDrqdT/ajM7AfiGUI3ylZn9bfDGOrQjJIZDCFf15xFVO5lZVaAF8DWhlNIs2n6mmV22jnMVRD+/BO5299aEhZSeKOG1pxKSGkSDTtYyHWgVveZehC94KH0t8JJiXdfzCtax7UxgRFQi+oJQIijg759BR2Bh1C50J1DTzIr+QirUeuWHHXIw09w587wL6XRxV3pe1Z3q1auX/sQMcPLxx7F48RJOP+9CruhxPQNv6JVR1U7/e/5Fttq7KZdNfJku45/nhe7XMuaqXjTrcDqXT3qVJke05bWb72Tp3HmM69WPzuNf4Ip33+R/L7zIrC+mxx1+0g5t1YLpX3/HWV170Onq3vS89AJ6XHw+Yya8SftLrmLSB59w0Vmnxh1mYjm5yd9iksxffh9gf3efY2ZbEOrfXyvlOR8Bj5vZGsKX4UmE6qZ3CXX8z7j7J2Z2NfCgmV1PqPc/y8zaAC3Xqo6BUE0zPJotsTYlD/y7BXjMzE4FfuXvV+5XAcPM7CpCO0WJvbjMbDNCQ/OJwN9iBfYp5fdQ3IfASDNbAqwiJIo5QDUzuxUoqrN5A3jKzFoR5pT/GthyPV5ng6m50Ubce1v2tEsUV61qVe4cuPafYOZYtWwZI9t3/Nv2wYcf/7dtU55+jilPP7cBokq/mhvV4J6+1/5t+8N39o8hmjLKrfhtFDmFhYkvUs3s9WiK2nU+rmjM7CjgN3f/0MzaAj2jNoTKZ9nvFaoEkhaF6yqsZYdutbeLO4RycffPH8UdQrnJabRzyt/y+Q/2TPr/NO+igbFklWRKFIvMbALwFuEKumbRutnu3rM8gyuj74GHo9JMHmElJxGRiinGRupkJZMoxhS7/0t5BZIu7j6dP9soREQqtgzoHpvMNOMjN0QgIiKVUoyN1MnKnG4cIiLZKBtKFCIiUo4yYOyREoWISJxUohARkYSypNeTiIiUFzVmi4hIQqp6EhGRhFSiEBGRhDJgriclChGROKnqSUREElKvJxERSUhtFCIikpCqnkREJCGVKEREJCH1ehIRkYRy1ZgtIiKJaPZYERFJSI3ZIiKSkBqzRUQkIZUoJFaFBXFHkH4ZcPVVVvfMnhp3COVizY2XxB1CuakyaGzqJ1EbhYiIJKReTyIikpCqnkREJKEMqE5VohARiZNGZouISEIqUYiISEJpaqMws6rASGA7IB+4AFgDjAAKganAJe6+3t0hK34qExHJYjm5eUnfSnEUUMXdWwA3AgOAu4Dr3b0VkAO0K0uMShQiInHKyU3+lthXQBUzywVqA6uBfYC3ov3jgbZlCVFVTyIicVqPNgozuxC4sNimoe4+NLq/hFDt9CVQHzgGONDdC6P9i4E6ZQlRiUJEJE7r0espSgpDS9h9BTDB3a8zs62BN4FqxfbXAhaWKcSyPElERNIkfVVPC4Dfo/vzgarAFDNrHW07EphUlhBVohARiVP6pvC4G3jYzCYRShI9gY+AYWZWDZgOPFuWEytRiIjEKU3dY919CXDqOnYHhZnxAAAXrklEQVQdlOq5lShEROKkAXciIpKQpvAQEZGEVKIQEZGENM24iIgkpIWLREQkIVU9iYhIQlozW0REEslRG4WIiCSkqifJdg8+PJI335rE6tWrOf3Ukzjl+OPiDiktVq9eQ89+N/HLr7+yatVqupx/Hoe0PjDusMpk9Zo19Bx4B7/MnMWq1avpcs6ZHNKqBQAvvvoGjz/7b54eel/MUZbBtruQd9w55N/XC7banrzTLob8fAp/+5WCJ++DwmjS1Jwcci/qQ+Hn71P4zivxxrwuGdCYXSFTmZnVMLPzE+yfER0zwsyOSPG1ZkU/7zGzbRIc91Q0X0q5MLO9zKxPeZ2/PLz/0cdM+ex/PPnIUB57aDCzZs2OO6S0GfvyeOrWqcOoh4cx7P57uOnW2+MOqczGTnidurVrM2rwPQy7cyA33R2SwvSvvuHZca9QWPSFmkFyDjmRvNMvharhXzL3iPYUvPIU+fdeC1WqkLP7vn8cm3v0WeTU3CSuUEuXk5P8LSYVtUTREDgfeGhDvaC7dytlf/tyfv1PgU/L8zXSbfK777PLTjtxyZU9WLJ0Kdd0uyzukNLmiEMP4fC2bf54nJdX8a/6SnLEwQdxeLHSUF5eHgt+/507hjxEz8u70PuWu2KMrozmziR/+M3kdegOQOEv30HNWmFf9Y0gPx+AnL1aQGEBhdM/iSvS0qkxu8x6AbtFV9jNgBpAPeBGd//32geb2f7AIOBkd/+p2PaphFWfVgKdgeHReQC6uvvnxY79T3TMXGAUUB1woI2772RmM4BdCUlsOGEK38LoPJ+Z2dfAO4ABs4GT3D2/2PlHEFac2jY691PAscA2hOUJtwY6u3v76Ngdo/d9h7s/bWbHADdEp5sSHbvea9+m04KFC/l15iyG3HsnP//yK12uuJpXnn86IxrnSrNxzZoALFm6lK5XX0e3SzrHHFHZbVxzIwCWLF1G1143cvn559Lr5jvp2bUL1atXjzm6sin87F3YbPM/N8yZSe4pF8Hhp8LypRR8/Tk02oacfQ6i4OFbyD2iXK/zUpMB/y8VNZUNAKYB/wXudPdDgUuBS9ZxbAvCurDHFk8SkU2Am9z9dMKUu2+4+8GEFaIGl/DavYB/u/tBwGj+nkzvAAa5+4HA5YSkAbAD0NvdmwMNCAlubTPc/TDCdL/bu/tRwHOEhAGAmdUCDgZOJMwfn2dmVYD7gaPdvRnwM9C4hPg3mLp16tCy+f5Uq1qVHbbblurVqjF/wYK4w0qbmbNmc/YFXWh39JEce2RKNZyxmzl7DmdfdhXtjmjLdltvxQ8//ULfO+6le5/+fDPjRwbc80DcIaYk96Tzyb/3WvIHXEzBhxPJPeE8cpu1IafOZuRd2p+c/dqQe3A7cpr8I+5Q/y5961GUm4paoigyE7jezDoRrt6rruOYwwgrN60u4Rwe/dwTaGNmp0WPNy3h+CbAyOj+uhb5aAK8DaG6KFpJCmBusUT1E6E0sLai8u9CwnKFEBYb+eNYd19sZpcSVrGqDTxOWNZwgbvPiY65sYTYN6h99vo/Hn3yac496wzmzJ3L8uUrqFunTCstVjhz583jvIsvo0+Pq2i+/35xh5OSufMXcN4V19Kn+6U03zd8Ub70RLi++XnmLLr36U+vbhfHGWLqli2BFcvD/d/nw/ZNKBj94B+7c488ncJFCypmFZRKFGVWQIjtJuBRd+8ATATW9RvtS1iwo6QSQlH1zJfA3e7emjBn+xMlHD8VaB7dP2Ad+6cDrSA0QAOzou3JtAiWeoyZNQL2cfcTgKOB2wirVdU1s82iYwaZWezfXgcf2JImtgsndziPLpdfRZ9rr8rouvzihgwfwaJFi3hg2MN0OL8zHc7vzIoVK+IOq0yGPDqKRYsX88CIx+lwaXc6XNqdFStXxh1WWuU/eR95Ha8mr+tAcloeRcG4x+IOKXkqUZTZHMIKTbsDg6KeST8Rrqz/xt0fMrOTzewMwhd3y3VcdQ8AhkeLk9cmJJh1uQV4zMxOBX7l7yWVqwgrRl1FKOF0KulNRF/sD7n7iSW+07+bBTQ0symExdLvcPdVZnYx8JKZ5RPaKD5cj3OWm2xqwC7u+muu5Pprrow7jLS4vtslXN9tXbW20LhRQ54Zdv8GjihN5s8h/66rw/3vppN/T48SDy0Y/+QGCqoMMqBEkZOJXePKk5kdBfzm7h+aWVugp7u3Ke15FdLSBdn34WbA4KQyW7Yo7gjKxZob152kskGVQWNT/pYv/PnLpP9PcxrvGktWqaglijh9T1h3dg2QB3SNOR4RyWYZUKJQoliLu0/nzzYKEZHyVfHzhBKFiEi8Kn6mUKIQEYmTRmaLiEhiKlGIiEgiaswWEZHElChERCQRlShERCShDBhEqkQhIhKjTJiWX4lCRCROShQiIpKYEoWIiCSiEoWIiCSkRCEiIgmp15OIiCRU8QsUShQiIvGq+JlCiUJEJE5qoxARkYSUKEREJDElChERSUQLF4mISGIVv0SRU1hYGHcMIiJSgVX8Mo+IiMRKiUJERBJSohARkYSUKEREJCElChERSUiJQkREElKiEBGRhJQoREQkISUKERFJSFN4SJmZ2e5AbaAAGAgMdPc34o0qPczsEGAH4H3gK3dfEXNIUgoza+zuP5vZvu7+UdzxZBMlCknFEOByoB/QC7gNyPhEYWYDgcZAE2AVcB1weqxBpYGZ7QVcCNQo2ubu58UXUfqY2RDgZ6A/0MHMOrj75TGHlTVU9SSpWA18AVRz9/fInguPlu5+NrDE3UcC28cdUJqMAD4Bni52yxZ7u3t/gChB7B1zPFklW/6xJR6FwCjgZTM7FVgaczzpUsXMagCFZpYH5McdUJrMcveH4g6inOSYWT13n2dmddF3W1rplympOA3YDxgPHBQ9zgZ3Ax8DDQhtFHfHG07azDCza4EphCSPu78ab0hpcyPwkZnNB+oCF8ccT1ZR1ZOkojlwgLsXAtcC+8YcT1q4+2jgCOBo4ER3fyLmkNKlOmBAe0KbS/t4w0kfdx8H7AQcBZi7T4g5pKyiEoWkoh/hCxVCaWI8kPFXqGbWB6jj7lea2Wgz+8jdb407rlS5+7nFH5tZo7hiSTczOwXIIyTD28zsdne/I+awsoZKFJKK1e4+B8Ddfyd76vLbufuVAO5+CnBczPGkhZn1M7PfzOx3M1sNvB53TGl0FfAacBawDXBsvOFkFyUKScUHZjbKzC4zs8cIdd/ZoMDMqgGYWVWy5//kSEK33ycIXX9/iTectCoa57LY3VcCteIMJttkyz+AxKMr8AxQE3jG3bvGHE+6DAGmmtlzwKfR42wwr+hL1N2/IXxu2eJ74CPgYTO7gdAJQdJEa2bLejOzY9x9nJlduPY+dx8aR0zpZmYNCCOzv3X3uXHHkw5mNgx4l9BTbQFwtLs3jTeq9DGzTdx9iZlt4e6z444nm6gxW8qiXvRz7cbQrLjqMLMDgHOBqoT++Vu6++Exh5UOFwFbA6OBjmRBd2Yzu97d+5vZk4RxL0Xbcfcz4o0ueyhRyHqLRisDDAD2IruqMAAGEcZOnAx8DlSLN5zUFJUAgfOLbV4JtAKmxxNV2rwY/cyW6sEKSYlCUjEaqAMUFfMLgbfjCydtFrr7k2Z2mLv3NbO34g4oRVlbAnT3z6K7PxJ6OtUotjvTP7cKQ4lCUlHf3VvFHUQ5KIxmxq1poS6jYdwBpaJYCTC/aD4kADO7OaaQysMY4HlC24ukmRKFpOIHM9va3X+KO5A06w7sTqiCGkWGV2uYWSdCtVMTMzsq2pxHaIO5LrbA0usnd+8bdxDZSr2eZL2Z2UxCtUUNYBNgfvS40N23jDO2dImm5DZgmrt/Hnc8qTCz6oRqp56EdiUIa4jMibrLZjwz6wxsB0wr2ubuj8YWUJZRohBZi5ndBLQBPgD2B15w99vjjSp1ZrYxsClhevgLgUfd/Yd4o0oPM/sPoWF+YbSp0N17xhdRdlHVk5SZmbUAHgC2IIzyPd/dP403qrQ4EtjP3QuiacbfBTI+URBGZD8CnES48h4KZEO3X4CV7t4l7iCylUZmSyruA85w90aEfvkPxBtO2vzMn1NAVOXPXl2ZblNgLNDY3W8hTKCXLX4ws+vM7HAzO8zMDos7oGyiEoWkYqG7TwNw96lmtizugNJkS+ArM/sM2A1YZWb/BXD3FrFGlppqwJXAx2a2G6F9KVtUBXaJbhDazDJ+JuOKQolCUjHHzB4C3gT2AXKLpvXI8Kk8Tok7gHJyJXA8oUH7TLJocZ+1p1CX9FLVk6TiS+AnYGdgEWGAUyMyfNwB4Up7S8L7eBjYyd1/yNSGXzNrHN2dCzwEbE6YknthiU/KcGb2bNwxZBOVKKTM3L1ftPhNVSAH2NLd3405rHQYAlxOWJipF3Ab8EasEaWme3R7kFAlkxNtLyT07spGF8QdQDZRopAyM7PhhOVQNwY2Ar4DDog1qPRYDXwBVHP398wso/9P3L17dPflbOjmuy5mlgM0469TeGTDdDIVgqqeJBVNCCOYJxAafVckPjxjFBJGZL9sZqcCS2OOJ12OjLr7ZqPngLuALtGtc7zhZJeMvlKS2C1290Iz29jd5xatCpcFTiOs2TAeaE0WTMcdaQD8ambf8+dI+kzuxVVcwyx6LxWOEoWk4mMzu4rw5fMUYf6gjBctVPRy9HCime1LmKYk0x0TdwDl6Mto3ZBf4w4kGylRSCpGAr8CywmjmT+IN5xycwphmc1M14AwMLL4+iHnxRNK2rUEfjSzuWTZvGMVgeZ6kjIzs8nu3jLuOCQ5ZvYhcD8wq2ibu0+ILyLJFCpRSCqWmtndgBNmI83ogXZrL6tZfF+WLKu5qNjaFFnFzPYkjHlpTEiE57n7lHijyh5KFJKK/0Y/t4h+ZnrxNCuX1Sw279HvZtYT+Jjos3L3bJnmYhBhUsrPoini/wX8M+aYsoYShaRi4lqPV5tZY3f/OZZoUlRsWc3aQDN372NmrxDWz85kp0c/fyeMot85epxN8yHlFn1+7v6pma2JO6BsokQhqehPmObiY2BvYBVQw8yGZfjArn7AEdH90wjdZDO2Lr9oHiQzqw/s7e6vmdmlwOPxRpZWq83sGGAScCCQFQsyVRQacCepWAY0dffTgf8jLHC/B2G9g0y22t3nALj770B+zPGky5NAnej+fLIrUXQCzgHeATqgKTzSSolCUtHA3VcAREtq1nf3VWT+39UHZjbKzC4zs0eBT+IOKE02dvdnAdx9FGHqlYxWbHqVmYQZcfcBzooeS5qo6klS8W8zm0wYP9EMGGtmXYCp8YaVspsJ4wtqEvrnZ3oJqcgqMzsUeI8w8jwbSkqPAmcQet4VdabIie7vEFdQ2UbjKCQlZtaUMOfTF9HiRQ2Aue6esX9YZvYaMBC4BHgWuMjdD443qtSZ2U7AHYARlkK9xt2/jTcqyQRKFJI2ZnaMu4+LO45UmdlEoC3wirsfamZvu/uBccclJTOzr/hrDclqwlop17h7tlQdxibT65KlYtm59EMyQjXCTKSTzOxgsrSKNssW95kIXEgo3Z4HfEioQhwUZ1DZQolCUmJmuWa2hZnluHumjzco0pFQ530rYX6ks2KNpvxkU8+gXdz9dXdf6e7/ARq5+xtEMwZIarLySkk2DDM7kXDlvQCoZWZd3P21mMNKmbt/DXwdPXwmzljSKcsX91llZp0JswW0AFaa2T7oOy4t9EuUVPQG9nP3OWa2BWEKjIxPFFnsOcJ62T9FjwvJnkRxBmHZ2nbA54SxFPuRPbPjxkqJQlIxr9jAtNlmtijugCShrFvcp9iUMfX46xxd9dx9fExhZR0lCknFIjObALwF7AvUNLOBAO7eM9bIZF2ycXGfK4ErgAfX2l4ItNnw4WQndY+VMjOzc0ral63TWWeyqAvpDoAW95H1okQhZWZmdYCDKNY46u5Z0/grFZ+ZzeTv09vnoCSYVqp6klS8ShjhuzB6XEgW9RLKNtm4uI+7N4o7hspAiUJS8XvRFNaSEbJ2cR8z253QmF0XeAKYmg2zBFQUShSSiglR3/VpRRvcPVu6W2ajbF7cZxBwLjAMGE5YQ0SJIk2UKCQVrYDqhHYKyK5++dkoqxf3cfdvzKzQ3X8zs8Vxx5NNlCgkFZu4e9u4g5CkdSLMHnsLoRSYTVN4zDezi4CNzaw9f7abSRooUUgqpkb/lFOIep64+1fxhiRrM7Mq7r6GPxf3KVqvIZt0AnoSuv7uGz2WNFGikFT8X3QrokFOFVNlWNxnsLufGXcQ2UrjKCQlZlYP2BH4zt3nxh2PVE5m9hzQD/iKaMbYaFleSQOVKKTMzOwUoD8wHdjDzPq6++MxhyUlyPLFfXYBxhBKSdlWWoqdEoWkojuwj7svMbNawJuAEkXFNREYTej11Bw4H3iE0LW0ZYxxpczd94w7hmymhYskFQXuvgTA3RcDK2KORxKrNIv7mNn1cceQTVSikFR8a2Z3EsZOtAK+jTkeSawyLe7zTtwBZBOVKCQVQ4H5wKGEUbH3xxuOlOIMQl3+rYT6+w6EhYwyfnEfM9tqrU1bxBJIlsrGKwnZcO4COrr7NDO7CxhBGPErFUglWdzn2WjU+RrCe6wDPBVvSNlDiUJSscbdpwG4+3dmlnV13VmiMizuczmh11Md4G53fzjmeLKKxlFImZnZk8D3wLuE9Yl30KAn2ZDM7LBiD/9JqAbtC+Dur8YRUzZSiUJScS7QGTiKMJaif7zhyLpk+eI+p6/12KNthYT1UiQNVKIQkYxlZtVK2qeR2emjEoVIJZGli/sUn7+qiEZmp5kShUjlkXWL+7j79nHHUBkoUYhUItm6uI+ZHQdcAlQllCjquXvTeKPKHhpwJ1J5ZPPiPn0IvZ1+AkYCn8caTZZRohCpPDoB25Odi/vMc/d3Adx9BNA43nCyi6qeRCqPbF7cZ6WZHQhUNbPDgYZxB5RNlChEKo8aZtaU7Fzc5wPC91l/4CaycEbcOClRiFQeWbe4j5l1Iqyr0QSYFm3OA5bHFlQW0oA7EclYZlYdaAT0BAZEmwuAOe6+MrbAsowShUglZWbXu7umXZFSqdeTSOWlxX0kKUoUIpWEmdUzs7bR/UuAKTGHJBlCiUKk8niKMM8TwALg8RhjkQyiRCFSedR092cB3H0UUDPmeCRDqHusSOWx2swOBd4jLDSlsQaSFJUoRCqP8wkT570PXAxcFG84kinUPVakEjGzPYDdgK/c/dO445HMoBKFSCVhZl0Ja1G0AIaa2VUxhyQZQolCpPI4HWjl7t2AfwKnxRyPZAglCpHKI8fd1wC4+2pgdczxSIZQryeRymOymT0LTAJaApNjjkcyhEoUIpXHXcBEoBawDzAq3nAkUyhRiFQeIwlTce9FmG317njDkUyhRCFSeVQB3gbquPtThHUbREqlRCFSeVQjVD9NMrODURulJEmJQqTy6Ag4cCvQADgr1mgkY2hktoiIJKQShYiIJKREISIiCSlRiJSRmdUws/PTeL7WZvZUgv0dzeyWJM+V9LEipVGiECm7hoSpu0WymrrHiZRdL2A3M+tDuOhqAWwCdAIecfcDAMzsPaA9YfnR4UC96Pld3f3zdZ3YzC4FTgSqAr9H9wGam9kbQG2gr7u/ZGYHAQOAfOBbtM6EpJlKFCJlNwCY5u43Ro+nu3sLYHkJx/cE3nD3g4ELgcHrOsjMcgnJpK27tyIki2bR7qVAW+Bo4H4zyyNMHX6iux8E/ELoBiuSNipRiKSPl7A9J/q5J9DGzIqm9950nSdxLzCzVcCTZrYEaExIFgCT3b0QmGNmvwP1gUbAM2YGsBHwKqFkIZIWShQiZVfAX0vlRWtQrwA2j672awHbR9u/BB5391FmtjkltG+YWVPgeHff38xqAh/zZ7JpFh3TkFDNNRf4GWjn7r+b2XHAEmCbNL1HEVU9iaRgDlDNzG4tvtHdZwGvAR8CQ4Fvol0DgFPN7D/AK8DUEs77DbDUzD6KzjMT2DLat5GZvQmMBS5y93zgcuAlM/svYS3sks4rUiYamS0iIgmpRCEiIgkpUYiISEJKFCIikpAShYiIJKREISIiCSlRiIhIQkoUIiKS0P8DssNW3K777vgAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.heatmap(cm.T, square=True, annot=True, fmt='d', cbar=True, cmap=\"Reds\", \n",
    "            xticklabels=train.target_names, yticklabels=train.target_names)\n",
    "plt.xlabel('true label')\n",
    "plt.ylabel('predicted label')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "How do these results differ from Multinomial Bayes? Look at the differences in confusion between christian texts and misc. religious texts..."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Categorical Data\n",
    "\n",
    "Central Questions:\n",
    "* What data is truly categorical?\n",
    "* Can we put the data directly into a classifier for machine learning?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "# transfer file!\n",
    "essay_df = pd.read_pickle(r\"../../HW2-Repo-1/jevon/ets_corpus_essay_df.pkl\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Filename</th>\n",
       "      <th>Prompt</th>\n",
       "      <th>L1</th>\n",
       "      <th>Score</th>\n",
       "      <th>Partition</th>\n",
       "      <th>Text</th>\n",
       "      <th>Tokens</th>\n",
       "      <th>Token_count</th>\n",
       "      <th>Score123</th>\n",
       "      <th>Avg_slen</th>\n",
       "      <th>TTR</th>\n",
       "      <th>TTR2</th>\n",
       "      <th>Avg_wlen</th>\n",
       "      <th>Avg_google_kband</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>88.txt</td>\n",
       "      <td>P6</td>\n",
       "      <td>KOR</td>\n",
       "      <td>high</td>\n",
       "      <td>DV</td>\n",
       "      <td>Some people might think that traveling in a gr...</td>\n",
       "      <td>[Some, people, might, think, that, traveling, ...</td>\n",
       "      <td>416</td>\n",
       "      <td>3</td>\n",
       "      <td>24.470588</td>\n",
       "      <td>0.367788</td>\n",
       "      <td>0.420</td>\n",
       "      <td>3.942708</td>\n",
       "      <td>1.539267</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>278.txt</td>\n",
       "      <td>P6</td>\n",
       "      <td>DEU</td>\n",
       "      <td>medium</td>\n",
       "      <td>TR</td>\n",
       "      <td>IThe importance and popularity of travelling i...</td>\n",
       "      <td>[IThe, importance, and, popularity, of, travel...</td>\n",
       "      <td>339</td>\n",
       "      <td>2</td>\n",
       "      <td>26.076923</td>\n",
       "      <td>0.365782</td>\n",
       "      <td>0.436</td>\n",
       "      <td>4.028037</td>\n",
       "      <td>1.438095</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>348.txt</td>\n",
       "      <td>P1</td>\n",
       "      <td>TUR</td>\n",
       "      <td>high</td>\n",
       "      <td>TR</td>\n",
       "      <td>It is an important decision, how to plan your ...</td>\n",
       "      <td>[It, is, an, important, decision, ,, how, to, ...</td>\n",
       "      <td>394</td>\n",
       "      <td>3</td>\n",
       "      <td>28.142857</td>\n",
       "      <td>0.479695</td>\n",
       "      <td>0.556</td>\n",
       "      <td>4.424581</td>\n",
       "      <td>1.844193</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>666.txt</td>\n",
       "      <td>P2</td>\n",
       "      <td>ZHO</td>\n",
       "      <td>medium</td>\n",
       "      <td>TR</td>\n",
       "      <td>Some people believe that young people can enjo...</td>\n",
       "      <td>[Some, people, believe, that, young, people, c...</td>\n",
       "      <td>379</td>\n",
       "      <td>2</td>\n",
       "      <td>27.071429</td>\n",
       "      <td>0.437995</td>\n",
       "      <td>0.532</td>\n",
       "      <td>4.133136</td>\n",
       "      <td>1.546828</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>733.txt</td>\n",
       "      <td>P6</td>\n",
       "      <td>TEL</td>\n",
       "      <td>medium</td>\n",
       "      <td>TR</td>\n",
       "      <td>Travelling is  usually considered as good recr...</td>\n",
       "      <td>[Travelling, is, usually, considered, as, good...</td>\n",
       "      <td>354</td>\n",
       "      <td>2</td>\n",
       "      <td>59.000000</td>\n",
       "      <td>0.418079</td>\n",
       "      <td>0.480</td>\n",
       "      <td>4.252976</td>\n",
       "      <td>1.733746</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  Filename Prompt   L1   Score Partition  \\\n",
       "0   88.txt     P6  KOR    high        DV   \n",
       "1  278.txt     P6  DEU  medium        TR   \n",
       "2  348.txt     P1  TUR    high        TR   \n",
       "3  666.txt     P2  ZHO  medium        TR   \n",
       "4  733.txt     P6  TEL  medium        TR   \n",
       "\n",
       "                                                Text  \\\n",
       "0  Some people might think that traveling in a gr...   \n",
       "1  IThe importance and popularity of travelling i...   \n",
       "2  It is an important decision, how to plan your ...   \n",
       "3  Some people believe that young people can enjo...   \n",
       "4  Travelling is  usually considered as good recr...   \n",
       "\n",
       "                                              Tokens  Token_count Score123  \\\n",
       "0  [Some, people, might, think, that, traveling, ...          416        3   \n",
       "1  [IThe, importance, and, popularity, of, travel...          339        2   \n",
       "2  [It, is, an, important, decision, ,, how, to, ...          394        3   \n",
       "3  [Some, people, believe, that, young, people, c...          379        2   \n",
       "4  [Travelling, is, usually, considered, as, good...          354        2   \n",
       "\n",
       "    Avg_slen       TTR   TTR2  Avg_wlen  Avg_google_kband  \n",
       "0  24.470588  0.367788  0.420  3.942708          1.539267  \n",
       "1  26.076923  0.365782  0.436  4.028037          1.438095  \n",
       "2  28.142857  0.479695  0.556  4.424581          1.844193  \n",
       "3  27.071429  0.437995  0.532  4.133136          1.546828  \n",
       "4  59.000000  0.418079  0.480  4.252976          1.733746  "
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "essay_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Which columns contain categorical data?\n",
    "If a column contains categorical data, is it nominal or ordinal?\n",
    "\n",
    "#### Ordinal Categorial Data\n",
    "* Scores: low, medium, high\n",
    "    * 3 target labels (can be used as-is)\n",
    "    * to use them as a feature, must turn these into numbers (strings cannot be directly used as features)\n",
    "* Score123: 1, 2, or 3\n",
    "    * if treated as real numbers, a regression model is appropriate (continuous variables then must be binned)\n",
    "    * if treated as 3 discrete targets, can use a classification model with 3 target labels\n",
    "    \n",
    "#### Nominal Categorial Data\n",
    "* L1\n",
    "* Prompt\n",
    "    * Can both be used as targets as-is\n",
    "    * To be used as features, must be decomposed\n",
    "    \n",
    "#### Decomposing strings into a set of binary, independent features\n",
    "3 options\n",
    "1. Encode value labels as integers via LabelEncoder. THEN, map integer lables into set of binary-valued columns via OneHotEncoder\n",
    "2. Transform column of categorical data into set of dummy variables via pandas.get_dummies\n",
    "3. Use sparse vector format via DictVectorizer (in textbook)\n",
    "\n",
    "\n",
    "##### Option 1: LabelEncoder + OneHotEncoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import LabelEncoder, OneHotEncoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Prompt</th>\n",
       "      <th>L1</th>\n",
       "      <th>Score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>12090</td>\n",
       "      <td>P6</td>\n",
       "      <td>HIN</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12091</td>\n",
       "      <td>P6</td>\n",
       "      <td>ZHO</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12092</td>\n",
       "      <td>P8</td>\n",
       "      <td>DEU</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12093</td>\n",
       "      <td>P2</td>\n",
       "      <td>KOR</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12094</td>\n",
       "      <td>P7</td>\n",
       "      <td>JPN</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12095</td>\n",
       "      <td>P1</td>\n",
       "      <td>ZHO</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12096</td>\n",
       "      <td>P8</td>\n",
       "      <td>KOR</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12097</td>\n",
       "      <td>P5</td>\n",
       "      <td>SPA</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12098</td>\n",
       "      <td>P5</td>\n",
       "      <td>HIN</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12099</td>\n",
       "      <td>P1</td>\n",
       "      <td>ZHO</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      Prompt   L1   Score\n",
       "12090     P6  HIN  medium\n",
       "12091     P6  ZHO  medium\n",
       "12092     P8  DEU    high\n",
       "12093     P2  KOR  medium\n",
       "12094     P7  JPN  medium\n",
       "12095     P1  ZHO  medium\n",
       "12096     P8  KOR    high\n",
       "12097     P5  SPA    high\n",
       "12098     P5  HIN    high\n",
       "12099     P1  ZHO  medium"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mini = essay_df[-10:][['Prompt', 'L1', 'Score']]\n",
    "mini\n",
    "# 6 different L1 represented."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LabelEncoder()"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "array([1, 5, 0, 3, 2, 5, 3, 4, 1, 5])"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Maps string values to integers. \n",
    "lenc = LabelEncoder()\n",
    "lenc.fit(mini['L1'])\n",
    "lenc.transform(mini['L1'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Prompt</th>\n",
       "      <th>L1</th>\n",
       "      <th>Score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>12090</td>\n",
       "      <td>P6</td>\n",
       "      <td>1</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12091</td>\n",
       "      <td>P6</td>\n",
       "      <td>5</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12092</td>\n",
       "      <td>P8</td>\n",
       "      <td>0</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12093</td>\n",
       "      <td>P2</td>\n",
       "      <td>3</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12094</td>\n",
       "      <td>P7</td>\n",
       "      <td>2</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12095</td>\n",
       "      <td>P1</td>\n",
       "      <td>5</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12096</td>\n",
       "      <td>P8</td>\n",
       "      <td>3</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12097</td>\n",
       "      <td>P5</td>\n",
       "      <td>4</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12098</td>\n",
       "      <td>P5</td>\n",
       "      <td>1</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12099</td>\n",
       "      <td>P1</td>\n",
       "      <td>5</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      Prompt  L1   Score\n",
       "12090     P6   1  medium\n",
       "12091     P6   5  medium\n",
       "12092     P8   0    high\n",
       "12093     P2   3  medium\n",
       "12094     P7   2  medium\n",
       "12095     P1   5  medium\n",
       "12096     P8   3    high\n",
       "12097     P5   4    high\n",
       "12098     P5   1    high\n",
       "12099     P1   5  medium"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mini['L1'] = lenc.transform(mini['L1'])\n",
    "mini"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\users\\cassi\\appdata\\local\\programs\\python\\python36-32\\lib\\site-packages\\sklearn\\preprocessing\\_encoders.py:368: FutureWarning: The handling of integer data will change in version 0.22. Currently, the categories are determined based on the range [0, max(values)], while in the future they will be determined based on the unique values.\n",
      "If you want the future behaviour and silence this warning, you can specify \"categories='auto'\".\n",
      "In case you used a LabelEncoder before this OneHotEncoder to convert the categories to integers, then you can now use the OneHotEncoder directly.\n",
      "  warnings.warn(msg, FutureWarning)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[0., 1., 0., 0., 0., 0.],\n",
       "       [0., 0., 0., 0., 0., 1.],\n",
       "       [1., 0., 0., 0., 0., 0.],\n",
       "       [0., 0., 0., 1., 0., 0.],\n",
       "       [0., 0., 1., 0., 0., 0.],\n",
       "       [0., 0., 0., 0., 0., 1.],\n",
       "       [0., 0., 0., 1., 0., 0.],\n",
       "       [0., 0., 0., 0., 1., 0.],\n",
       "       [0., 1., 0., 0., 0., 0.],\n",
       "       [0., 0., 0., 0., 0., 1.]])"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Turns multi-valued L1 column into 6 columns of binary values, \n",
    "#     in sparse vector format. \n",
    "ohenc = OneHotEncoder()\n",
    "\n",
    "# In sparse vector, but small enough to cast to array\n",
    "ohenc.fit_transform(mini[['L1']]).toarray()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Option 2: Use pandas.get_dummies"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Prompt</th>\n",
       "      <th>L1</th>\n",
       "      <th>Score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>12090</td>\n",
       "      <td>P6</td>\n",
       "      <td>HIN</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12091</td>\n",
       "      <td>P6</td>\n",
       "      <td>ZHO</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12092</td>\n",
       "      <td>P8</td>\n",
       "      <td>DEU</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12093</td>\n",
       "      <td>P2</td>\n",
       "      <td>KOR</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12094</td>\n",
       "      <td>P7</td>\n",
       "      <td>JPN</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12095</td>\n",
       "      <td>P1</td>\n",
       "      <td>ZHO</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12096</td>\n",
       "      <td>P8</td>\n",
       "      <td>KOR</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12097</td>\n",
       "      <td>P5</td>\n",
       "      <td>SPA</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12098</td>\n",
       "      <td>P5</td>\n",
       "      <td>HIN</td>\n",
       "      <td>high</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12099</td>\n",
       "      <td>P1</td>\n",
       "      <td>ZHO</td>\n",
       "      <td>medium</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      Prompt   L1   Score\n",
       "12090     P6  HIN  medium\n",
       "12091     P6  ZHO  medium\n",
       "12092     P8  DEU    high\n",
       "12093     P2  KOR  medium\n",
       "12094     P7  JPN  medium\n",
       "12095     P1  ZHO  medium\n",
       "12096     P8  KOR    high\n",
       "12097     P5  SPA    high\n",
       "12098     P5  HIN    high\n",
       "12099     P1  ZHO  medium"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Re-build mini DF\n",
    "mini = essay_df[-10:][['Prompt', 'L1', 'Score']]\n",
    "mini"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Prompt</th>\n",
       "      <th>Score</th>\n",
       "      <th>L1_DEU</th>\n",
       "      <th>L1_HIN</th>\n",
       "      <th>L1_JPN</th>\n",
       "      <th>L1_KOR</th>\n",
       "      <th>L1_SPA</th>\n",
       "      <th>L1_ZHO</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>12090</td>\n",
       "      <td>P6</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12091</td>\n",
       "      <td>P6</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12092</td>\n",
       "      <td>P8</td>\n",
       "      <td>high</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12093</td>\n",
       "      <td>P2</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12094</td>\n",
       "      <td>P7</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12095</td>\n",
       "      <td>P1</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12096</td>\n",
       "      <td>P8</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12097</td>\n",
       "      <td>P5</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12098</td>\n",
       "      <td>P5</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12099</td>\n",
       "      <td>P1</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      Prompt   Score  L1_DEU  L1_HIN  L1_JPN  L1_KOR  L1_SPA  L1_ZHO\n",
       "12090     P6  medium       0       1       0       0       0       0\n",
       "12091     P6  medium       0       0       0       0       0       1\n",
       "12092     P8    high       1       0       0       0       0       0\n",
       "12093     P2  medium       0       0       0       1       0       0\n",
       "12094     P7  medium       0       0       1       0       0       0\n",
       "12095     P1  medium       0       0       0       0       0       1\n",
       "12096     P8    high       0       0       0       1       0       0\n",
       "12097     P5    high       0       0       0       0       1       0\n",
       "12098     P5    high       0       1       0       0       0       0\n",
       "12099     P1  medium       0       0       0       0       0       1"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Decomposes L1 column into 6 binary columns. \n",
    "pd.get_dummies(mini, columns=['L1'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Score</th>\n",
       "      <th>Prompt_P1</th>\n",
       "      <th>Prompt_P2</th>\n",
       "      <th>Prompt_P5</th>\n",
       "      <th>Prompt_P6</th>\n",
       "      <th>Prompt_P7</th>\n",
       "      <th>Prompt_P8</th>\n",
       "      <th>L1_DEU</th>\n",
       "      <th>L1_HIN</th>\n",
       "      <th>L1_JPN</th>\n",
       "      <th>L1_KOR</th>\n",
       "      <th>L1_SPA</th>\n",
       "      <th>L1_ZHO</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>12090</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12091</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12092</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12093</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12094</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12095</td>\n",
       "      <td>medium</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12096</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12097</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12098</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12099</td>\n",
       "      <td>medium</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        Score  Prompt_P1  Prompt_P2  Prompt_P5  Prompt_P6  Prompt_P7  \\\n",
       "12090  medium          0          0          0          1          0   \n",
       "12091  medium          0          0          0          1          0   \n",
       "12092    high          0          0          0          0          0   \n",
       "12093  medium          0          1          0          0          0   \n",
       "12094  medium          0          0          0          0          1   \n",
       "12095  medium          1          0          0          0          0   \n",
       "12096    high          0          0          0          0          0   \n",
       "12097    high          0          0          1          0          0   \n",
       "12098    high          0          0          1          0          0   \n",
       "12099  medium          1          0          0          0          0   \n",
       "\n",
       "       Prompt_P8  L1_DEU  L1_HIN  L1_JPN  L1_KOR  L1_SPA  L1_ZHO  \n",
       "12090          0       0       1       0       0       0       0  \n",
       "12091          0       0       0       0       0       0       1  \n",
       "12092          1       1       0       0       0       0       0  \n",
       "12093          0       0       0       0       1       0       0  \n",
       "12094          0       0       0       1       0       0       0  \n",
       "12095          0       0       0       0       0       0       1  \n",
       "12096          1       0       0       0       1       0       0  \n",
       "12097          0       0       0       0       0       1       0  \n",
       "12098          0       0       1       0       0       0       0  \n",
       "12099          0       0       0       0       0       0       1  "
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Decomposition on both Prompt and L1.  \n",
    "pd.get_dummies(mini, columns=['Prompt', 'L1'])\n",
    "\n",
    "# Works, looking tidy, but lots of wasted space with 0s\n",
    "# It would be more efficient to use a sparse vector..."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Looks good...but there's something wrong here..."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### The Dummy Variable Trap\n",
    "* Recall, we want these features to be INDEPENDENT\n",
    "* For Prompt and L1, there's one redundant column\n",
    "    * If all others are labeled, can predict the last L1 or prompt based on the others\n",
    "    \n",
    "This is *multi-collinearity* and causes problems in certain machine learning models (not to mention you don't want to enter redundant info in general)\n",
    "\n",
    "We must REMOVE a column for each feature category. For more info, see https://shanlodh.wordpress.com/2018/03/05/beward-silent-multi-collinearity-from-dummy-variables-in-sklearn/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Score</th>\n",
       "      <th>Prompt_P2</th>\n",
       "      <th>Prompt_P5</th>\n",
       "      <th>Prompt_P6</th>\n",
       "      <th>Prompt_P7</th>\n",
       "      <th>Prompt_P8</th>\n",
       "      <th>L1_HIN</th>\n",
       "      <th>L1_JPN</th>\n",
       "      <th>L1_KOR</th>\n",
       "      <th>L1_SPA</th>\n",
       "      <th>L1_ZHO</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>12090</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12091</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12092</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12093</td>\n",
       "      <td>medium</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12094</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12095</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12096</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12097</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12098</td>\n",
       "      <td>high</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12099</td>\n",
       "      <td>medium</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        Score  Prompt_P2  Prompt_P5  Prompt_P6  Prompt_P7  Prompt_P8  L1_HIN  \\\n",
       "12090  medium          0          0          1          0          0       1   \n",
       "12091  medium          0          0          1          0          0       0   \n",
       "12092    high          0          0          0          0          1       0   \n",
       "12093  medium          1          0          0          0          0       0   \n",
       "12094  medium          0          0          0          1          0       0   \n",
       "12095  medium          0          0          0          0          0       0   \n",
       "12096    high          0          0          0          0          1       0   \n",
       "12097    high          0          1          0          0          0       0   \n",
       "12098    high          0          1          0          0          0       1   \n",
       "12099  medium          0          0          0          0          0       0   \n",
       "\n",
       "       L1_JPN  L1_KOR  L1_SPA  L1_ZHO  \n",
       "12090       0       0       0       0  \n",
       "12091       0       0       0       1  \n",
       "12092       0       0       0       0  \n",
       "12093       0       1       0       0  \n",
       "12094       1       0       0       0  \n",
       "12095       0       0       0       1  \n",
       "12096       0       1       0       0  \n",
       "12097       0       0       1       0  \n",
       "12098       0       0       0       0  \n",
       "12099       0       0       0       1  "
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Use drop_first=True to avoid creating perfectly correlated variables\n",
    "pd.get_dummies(mini, columns=['Prompt', 'L1'], drop_first=True)\n",
    "\n",
    "# All right -- it takes only 5 columns to encode 6 distinct values"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Wait, doesn't OneHotEncoder also run into the dummy variable trap?\n",
    "Yes! But people don't typically use OneHotCoder for our purposes. Instead, they use it for other methods, like unsupervised classification. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Option 3: sklearn's DictVectorizer\n",
    "See the [textbook example](https://jakevdp.github.io/PythonDataScienceHandbook/05.04-feature-engineering.html#Categorical-Features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}