LogReg.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h3>Assuming that data is lineary seperable We will try to train model using Logistic regression</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Import data using get_data.py module \n",
    "from get_data import load_data\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from matplotlib import pyplot "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Load the data\n",
    "X_train, X_val, X_test, y_train, y_val, y_test = load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Conerting text to TFIDF\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "vect = TfidfVectorizer(decode_error='ignore',stop_words='english')\n",
    "train_tfidf = vect.fit_transform(X_train)\n",
    "val_tfidf = vect.transform(X_val)\n",
    "test_tfidf = vect.transform(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Creating empty prediction array\n",
    "# Training a different model for each toxicity level\n",
    "col = ['toxic', 'severe_toxic', 'obscene', 'threat','insult', 'identity_hate']\n",
    "pred_train = np.zeros((X_train.shape[0],len(col)))\n",
    "pred_test = np.zeros((X_test.shape[0],len(col)))\n",
    "pred_val = np.zeros((X_val.shape[0],len(col)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Load Logistic Regression model\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "LogR = LogisticRegression()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "toxic predicted!\n",
      "severe_toxic predicted!\n",
      "obscene predicted!\n",
      "threat predicted!\n",
      "insult predicted!\n",
      "identity_hate predicted!\n"
     ]
    }
   ],
   "source": [
    "# Predict on train, val \n",
    "# Assuming that Data is linearly seperable\n",
    "for i,x in enumerate(col):\n",
    "    LogR.fit(train_tfidf, y_train[x])\n",
    "    pred_train[:,i] = LogR.predict_proba(train_tfidf)[:,1]\n",
    "    pred_val[:,i] = LogR.predict_proba(val_tfidf)[:,1]\n",
    "    print(x,\"predicted!\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We are comparing the AUC score for predictions on train data and validation data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "toxic <Train,Val> AUC: 0.985669142993 0.969281957227\n",
      "severe_toxic <Train,Val> AUC: 0.992689695181 0.985720379516\n",
      "obscene <Train,Val> AUC: 0.993966841859 0.982284100316\n",
      "threat <Train,Val> AUC: 0.9956640451 0.971606244941\n",
      "insult <Train,Val> AUC: 0.988622467896 0.973958296445\n",
      "identity_hate <Train,Val> AUC: 0.991503895837 0.975949369693\n"
     ]
    }
   ],
   "source": [
    "from sklearn import metrics\n",
    "for i,x in enumerate(col):\n",
    "    print(x,\"<Train,Val> AUC:\",metrics.roc_auc_score(y_train[x], pred_train[:,i]),metrics.roc_auc_score(y_val[x], pred_val[:,i]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Avobe results show that AUC scores of train and validation data are almost comparable and there is no case of overfitting.\n",
    "However in order to get best results we can change the regularasation and get best model that does not overfit"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tuning the Model By Changing the regularisation(C) value"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "c=[0.01,0.1,1,10,100]\n",
    "resultDict ={}\n",
    "resultValDict ={}\n",
    "for i in c:\n",
    "    LogR = LogisticRegression(C=i)\n",
    "    temp=[]\n",
    "    temp2=[]\n",
    "    for j,x in enumerate(col):\n",
    "        LogR.fit(train_tfidf, y_train[x])\n",
    "        pred_train[:,j] = LogR.predict_proba(train_tfidf)[:,1]\n",
    "        pred_val[:,j] = LogR.predict_proba(val_tfidf)[:,1]\n",
    "        temp.append(metrics.roc_auc_score(y_train[x], pred_train[:,1]))\n",
    "        temp2.append(metrics.roc_auc_score(y_val[x], pred_val[:,1]))\n",
    "    resultDict[i]=temp\n",
    "    resultValDict[i]=temp2\n",
    "\n",
    "result = pd.DataFrame(resultDict)\n",
    "valResult =pd.DataFrame(resultValDict)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(     0.01      0.10      1.00      10.00     100.00\n",
       " 0  0.801645  0.867584  0.898566  0.903493  0.880183\n",
       " 1  0.982275  0.986849  0.992690  0.998630  0.999576\n",
       " 2  0.943818  0.960118  0.955245  0.933972  0.899600\n",
       " 3  0.925848  0.941271  0.951133  0.943425  0.923306\n",
       " 4  0.919243  0.940872  0.944313  0.927895  0.894769\n",
       " 5  0.925177  0.941028  0.949261  0.938981  0.912521,\n",
       "      0.01      0.10      1.00      10.00     100.00\n",
       " 0  0.800974  0.874828  0.906041  0.909949  0.885901\n",
       " 1  0.983627  0.984865  0.985720  0.980200  0.964196\n",
       " 2  0.943817  0.960962  0.954879  0.932323  0.894569\n",
       " 3  0.917733  0.937900  0.941677  0.924766  0.886858\n",
       " 4  0.920785  0.943354  0.945361  0.926712  0.891177\n",
       " 5  0.928420  0.943376  0.949501  0.936101  0.907708,\n",
       "          0.01      0.10      1.00      10.00     100.00\n",
       " 0  6.709623e-04 -0.007244 -0.007475 -0.006456 -0.005718\n",
       " 1 -1.351729e-03  0.001984  0.006969  0.018429  0.035381\n",
       " 2  4.747003e-07 -0.000845  0.000366  0.001649  0.005030\n",
       " 3  8.114294e-03  0.003371  0.009455  0.018659  0.036448\n",
       " 4 -1.541671e-03 -0.002482 -0.001048  0.001183  0.003592\n",
       " 5 -3.242322e-03 -0.002348 -0.000240  0.002880  0.004813)"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "result , valResult , result-valResult"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see that for lower values of C i.e for high regularisation, we are getting almost same AUC scores for both the data.But there is also a trade off as we are getting better solution to overfitting with the expense of overall accuracy of classification."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEOCAYAAACaQSCZAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0xf37c438>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "pyplot.subplot()\n",
    "for i in result.index:\n",
    "    pyplot.plot(c,(result - valResult).loc[i])\n",
    "    \n",
    "pyplot.xscale(\"log\")\n",
    "pyplot.xlabel(\"Value of C\")\n",
    "pyplot.ylabel(\"ROC_AUC Score Difference\")\n",
    "pyplot.legend(col)\n",
    "pyplot.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<p8>We can see that the AUC score of predictions on training data is greater than that of Validation data. This is expected\n",
    "as validations data is new to Trained model. Although it might indicate towards overfitting the difference is not that\n",
    "much and thus we can say that model is not overfitted </p8>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h4> Finally running the model on testing data using tunned parameters</h4>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "toxic <Test> AUC: 0.961251532981\n",
      "severe_toxic <Test> AUC: 0.986932468014\n",
      "obscene <Test> AUC: 0.985689700797\n",
      "threat <Test> AUC: 0.992805943004\n",
      "insult <Test> AUC: 0.97516894943\n",
      "identity_hate <Test> AUC: 0.976127176656\n"
     ]
    }
   ],
   "source": [
    "LogR = LogisticRegression(C=0.1)\n",
    "for i,x in enumerate(col):\n",
    "    LogR.fit(test_tfidf, y_test[x])\n",
    "    pred_test[:,i] = LogR.predict_proba(test_tfidf)[:,1]\n",
    "\n",
    "for i,x in enumerate(col):\n",
    "    print(x,\"<Test> AUC:\",metrics.roc_auc_score(y_test[x], pred_test[:,i]))\n"
   ]
  }
 ],
 "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": "3.6.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}