model_comparison.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from sklearn.externals import joblib\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.metrics import accuracy_score\n",
    "import time\n",
    "from sklearn.metrics import accuracy_score\n",
    "from sklearn.metrics import precision_score,recall_score\n",
    "from sklearn.metrics import roc_curve\n",
    "from sklearn import metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "## Preprocess\n",
    "df_raw_data = pd.read_csv('reviews.csv',sep='|',header=0)\n",
    "df_train = df_raw_data.iloc[df_raw_data.index %5 != 0]\n",
    "df_test = df_raw_data.iloc[df_raw_data.index %5 == 0]\n",
    "\n",
    "# preprocess X\n",
    "# word of bag\n",
    "word_bag = joblib.load('word_bag.pkl')\n",
    "\n",
    "X_test = df_test.loc[:,'text'].tolist()\n",
    "X_test = word_bag.transform(X_test)\n",
    "\n",
    "X_train = df_train.loc[:,'text'].tolist()\n",
    "X_train = word_bag.transform(X_train)\n",
    "\n",
    "# preprocess y\n",
    "y = df_test['label']\n",
    "lec = LabelEncoder()\n",
    "y_test = lec.fit_transform(y)\n",
    "\n",
    "y = df_train['label']\n",
    "lec = LabelEncoder()\n",
    "y_train = lec.fit_transform(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# get the feature (v2)\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "\n",
    "## Preprocess\n",
    "df_raw_data = pd.read_csv('reviews.csv',sep='|',header=0)\n",
    "df_train = df_raw_data.iloc[df_raw_data.index %5 != 0]\n",
    "df_test = df_raw_data.iloc[df_raw_data.index %5 == 0]\n",
    "\n",
    "## Preprocess\n",
    "X_train = df_train.loc[:,'text'].tolist()\n",
    "X_test = df_test.loc[:,'text'].tolist()\n",
    "transformer = TfidfVectorizer()\n",
    "#ngram_range=(1, 3)\n",
    "X_train = transformer.fit_transform(X_train)\n",
    "feature_tfidf_name = transformer.get_feature_names()\n",
    "X_test = transformer.transform(X_test)\n",
    "\n",
    "# preprocess y\n",
    "y = df_test['label']\n",
    "lec = LabelEncoder()\n",
    "y_test = lec.fit_transform(y)\n",
    "\n",
    "y = df_train['label']\n",
    "lec = LabelEncoder()\n",
    "y_train = lec.fit_transform(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['tfidf_features.pkl']"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joblib.dump(feature_tfidf_name,'tfidf_features.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "joblib.dump(transformer,'final_model/Tfidf_n_3.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Decision Tree"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.tree import DecisionTreeClassifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training time: 841.031721830368 s\n"
     ]
    }
   ],
   "source": [
    "tic = time.time()\n",
    "clf = DecisionTreeClassifier(criterion='entropy',max_depth=50)\n",
    "clf.fit(X_train,y_train)\n",
    "tac = time.time()\n",
    "print('training time:',tac-tic,'s')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['final_model/decision_tree.pkl']"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joblib.dump(clf,'final_model/decision_tree.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# accurarcy\n",
    "y_pred = clf.predict(X_train)\n",
    "DT_train_ac = accuracy_score(y_train,y_pred)\n",
    "y_pred = clf.predict(X_test)\n",
    "DT_test_ac = accuracy_score(y_test,y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training accuracy: 93.4177420918 %\n",
      "testing accuracy: 76.157673651 %\n"
     ]
    }
   ],
   "source": [
    "print('training accuracy:',DT_train_ac*100,'%')\n",
    "print('testing accuracy:',DT_test_ac*100,'%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision:  0.756514678019  recall:  0.770140280561  f1:  0.763266674115\n"
     ]
    }
   ],
   "source": [
    "# precision and recall\n",
    "DT_p_score = precision_score(y_test,y_pred)\n",
    "DT_r_score = recall_score(y_test,y_pred)\n",
    "DT_f1_score = 2*(DT_p_score*DT_r_score)/(DT_p_score+DT_r_score)\n",
    "print(\"precision: \",DT_p_score,\" recall: \",DT_r_score,\" f1: \",DT_f1_score)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 0.          0.24695518  1.        ] [ 0.          0.77014028  1.        ] [2 1 0]\n"
     ]
    }
   ],
   "source": [
    "# ROC\n",
    "DT_fpr,DT_tpr, DT_thresholds = roc_curve(y_test,y_pred)\n",
    "print(DT_fpr,DT_tpr, DT_thresholds)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## NN model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.neural_network import MLPClassifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training time: 1177.0293114185333 s\n"
     ]
    }
   ],
   "source": [
    "tic = time.time()\n",
    "clf = MLPClassifier(hidden_layer_sizes = (10,10),activation='logistic',max_iter=200,\n",
    "                    learning_rate_init = 0.01,learning_rate='invscaling',\n",
    "                    solver='adam',early_stopping=True,random_state=42)\n",
    "clf.fit(X_train,y_train)\n",
    "tac = time.time()\n",
    "print('training time:',tac-tic,'s')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['final_model/nn_model.pkl']"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joblib.dump(clf,'final_model/nn_model.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training accuracy: 93.1561395339 %\n",
      "testing accuracy: 90.1235185278 %\n"
     ]
    }
   ],
   "source": [
    "# accurarcy\n",
    "y_pred = clf.predict(X_train)\n",
    "NN_train_ac = accuracy_score(y_train,y_pred)\n",
    "y_pred = clf.predict(X_test)\n",
    "NN_test_ac = accuracy_score(y_test,y_pred)\n",
    "print('training accuracy:',NN_train_ac*100,'%')\n",
    "print('testing accuracy:',NN_test_ac*100,'%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision:  0.917797494781  recall:  0.881012024048  f1:  0.899028629857\n"
     ]
    }
   ],
   "source": [
    "# precision and recall\n",
    "NN_p_score = precision_score(y_test,y_pred)\n",
    "NN_r_score = recall_score(y_test,y_pred)\n",
    "NN_f1_score = 2*(NN_p_score*NN_r_score)/(NN_p_score+NN_r_score)\n",
    "print(\"precision: \",NN_p_score,\" recall: \",NN_r_score,\" f1: \",NN_f1_score)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 0.          0.07861635  1.        ] [ 0.          0.88101202  1.        ] [2 1 0]\n"
     ]
    }
   ],
   "source": [
    "# ROC\n",
    "NN_fpr,NN_tpr, NN_thresholds = roc_curve(y_test,y_pred)\n",
    "print(NN_fpr,NN_tpr, NN_thresholds)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Logistic Regression"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.linear_model import LogisticRegression"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training time: 25.659657955169678 s\n"
     ]
    }
   ],
   "source": [
    "tic = time.time()\n",
    "clf = LogisticRegression()\n",
    "clf.fit(X_train,y_train)\n",
    "tac = time.time()\n",
    "print('training time:',tac-tic,'s')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['final_model/logistic.pkl']"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joblib.dump(clf,'final_model/logistic.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training accuracy: 91.9943991424 %\n",
      "testing accuracy: 90.443566535 %\n"
     ]
    }
   ],
   "source": [
    "# accurarcy\n",
    "y_pred = clf.predict(X_train)\n",
    "LG_train_ac = accuracy_score(y_train,y_pred)\n",
    "y_pred = clf.predict(X_test)\n",
    "LG_test_ac = accuracy_score(y_test,y_pred)\n",
    "print('training accuracy:',LG_train_ac*100,'%')\n",
    "print('testing accuracy:',LG_test_ac*100,'%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision:  0.903470173509  recall:  0.905235470942  f1:  0.90435196076\n"
     ]
    }
   ],
   "source": [
    "# precision and recall\n",
    "LG_p_score = precision_score(y_test,y_pred)\n",
    "LG_r_score = recall_score(y_test,y_pred)\n",
    "LG_f1_score = 2*(LG_p_score*LG_r_score)/(LG_p_score+LG_r_score)\n",
    "print(\"precision: \",LG_p_score,\" recall: \",LG_r_score,\" f1: \",LG_f1_score)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 0.          0.09636119  1.        ] [ 0.          0.90523547  1.        ] [2 1 0]\n"
     ]
    }
   ],
   "source": [
    "# ROC\n",
    "LG_fpr,LG_tpr, LG_thresholds = roc_curve(y_test,y_pred)\n",
    "print(LG_fpr,LG_tpr, LG_thresholds)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Ensemble model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## RandomForest"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.ensemble import RandomForestClassifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training time: 271.4661648273468 s\n"
     ]
    }
   ],
   "source": [
    "tic = time.time()\n",
    "clf = RandomForestClassifier(n_jobs=-1)\n",
    "clf.fit(X_train,y_train)\n",
    "tac = time.time()\n",
    "print('training time:',tac-tic,'s')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['final_model/random_forest.pkl']"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joblib.dump(clf,'final_model/random_forest.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training accuracy: 99.4077218074 %\n",
      "testing accuracy: 78.468020203 %\n"
     ]
    }
   ],
   "source": [
    "# accurarcy\n",
    "y_pred = clf.predict(X_train)\n",
    "RF_train_ac = accuracy_score(y_train,y_pred)\n",
    "y_pred = clf.predict(X_test)\n",
    "RF_test_ac = accuracy_score(y_test,y_pred)\n",
    "print('training accuracy:',RF_train_ac*100,'%')\n",
    "print('testing accuracy:',RF_test_ac*100,'%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "precision:  0.831082066692  recall:  0.713602204409  f1:  0.767874711916\n"
     ]
    }
   ],
   "source": [
    "# precision and recall\n",
    "RF_p_score = precision_score(y_test,y_pred)\n",
    "RF_r_score = recall_score(y_test,y_pred)\n",
    "RF_f1_score = 2*(RF_p_score*RF_r_score)/(RF_p_score+RF_r_score)\n",
    "print(\"precision: \",RF_p_score,\" recall: \",RF_r_score,\" f1: \",RF_f1_score)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 0.          0.14450434  1.        ] [ 0.         0.7136022  1.       ] [2 1 0]\n"
     ]
    }
   ],
   "source": [
    "# ROC\n",
    "RF_fpr,RF_tpr, RF_thresholds = roc_curve(y_test,y_pred)\n",
    "print(RF_fpr,RF_tpr, RF_thresholds)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## logistic+randomforest+NN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.ensemble import VotingClassifier\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.neural_network import MLPClassifier\n",
    "import time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training time: 1486.9226295948029 s\n"
     ]
    }
   ],
   "source": [
    "tic = time.time()\n",
    "clf1 = RandomForestClassifier(n_jobs=-1)\n",
    "clf2 = LogisticRegression(n_jobs=-1,random_state=42)\n",
    "clf3 = MLPClassifier(hidden_layer_sizes = (10,10),activation='logistic',max_iter=100,\n",
    "                    learning_rate_init = 0.01,learning_rate='invscaling',\n",
    "                    solver='adam',early_stopping=True,random_state=42)\n",
    "\n",
    "eclf1 = VotingClassifier(estimators=[\n",
    "        ('rf', clf1), ('lg', clf2),('nn',clf3)],n_jobs=-1,voting='soft')\n",
    "eclf1.fit(X_train,y_train)\n",
    "tac = time.time()\n",
    "print('training time:',tac-tic,'s')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['final_model/ensemble.pkl']"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joblib.dump(eclf1,'final_model/ensemble.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training accuracy: 95.8246731531 %\n",
      "testing accuracy: 90.4348152223 %\n"
     ]
    }
   ],
   "source": [
    "# accurarcy\n",
    "y_pred = eclf1.predict(X_train)\n",
    "ES_train_ac = accuracy_score(y_train,y_pred)\n",
    "y_pred = eclf1.predict(X_test)\n",
    "ES_test_ac = accuracy_score(y_test,y_pred)\n",
    "print('training accuracy:',RF_train_ac*100,'%')\n",
    "print('testing accuracy:',RF_test_ac*100,'%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# precision and recall\n",
    "RF_p_score = precision_score(y_test,y_pred)\n",
    "RF_r_score = recall_score(y_test,y_pred)\n",
    "RF_f1_score = 2*(RF_p_score*RF_r_score)/(RF_p_score+RF_r_score)\n",
    "print(\"precision: \",RF_p_score,\" recall: \",RF_r_score,\" f1: \",RF_f1_score)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ROC curve"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "DT_clf = joblib.load('final_model/decision_tree.pkl')\n",
    "NN_clf = joblib.load('final_model/nn_model.pkl')\n",
    "LG_clf = joblib.load('final_model/logistic.pkl')\n",
    "RF_clf = joblib.load('final_model/random_forest.pkl')\n",
    "ES_clf = joblib.load('final_model/ensemble.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[78.013854122090649, 96.510333860293201, 96.469514475691938, 86.421485309893029, 96.265507329923921]\n"
     ]
    }
   ],
   "source": [
    "print([DT_roc_auc*100,LG_roc_auc*100,NN_roc_auc*100,RF_roc_auc*100,ES_roc_auc*100])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYoAAAEWCAYAAAB42tAoAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#calculate the fpr and tpr for all thresholds of the classification\n",
    "#decision tree\n",
    "DT_probs = DT_clf.predict_proba(X_test)\n",
    "DT_preds = DT_probs[:,1]\n",
    "DT_fpr, DT_tpr, DT_threshold = metrics.roc_curve(y_test,DT_preds)\n",
    "DT_roc_auc = metrics.auc(DT_fpr, DT_tpr)\n",
    "\n",
    "LG_probs = LG_clf.predict_proba(X_test)\n",
    "LG_preds = LG_probs[:,1]\n",
    "LG_fpr, LG_tpr, LG_threshold = metrics.roc_curve(y_test,LG_preds)\n",
    "LG_roc_auc = metrics.auc(LG_fpr, LG_tpr)\n",
    "\n",
    "NN_probs = NN_clf.predict_proba(X_test)\n",
    "NN_preds = NN_probs[:,1]\n",
    "NN_fpr, NN_tpr, NN_threshold = metrics.roc_curve(y_test,NN_preds)\n",
    "NN_roc_auc = metrics.auc(NN_fpr, NN_tpr)\n",
    "\n",
    "RF_probs = RF_clf.predict_proba(X_test)\n",
    "RF_preds = RF_probs[:,1]\n",
    "RF_fpr, RF_tpr, RF_threshold = metrics.roc_curve(y_test,RF_preds)\n",
    "RF_roc_auc = metrics.auc(RF_fpr, RF_tpr)\n",
    "\n",
    "ES_probs = ES_clf.predict_proba(X_test)\n",
    "ES_preds = ES_probs[:,1]\n",
    "ES_fpr, ES_tpr, ES_threshold = metrics.roc_curve(y_test,ES_preds)\n",
    "ES_roc_auc = metrics.auc(ES_fpr, ES_tpr)\n",
    "\n",
    "# plot\n",
    "plt.title('Receiver Operating Characteristic')\n",
    "plt.plot(DT_fpr, DT_tpr, 'b', label = 'Decision_Tree_AUC = %0.2f' % DT_roc_auc)\n",
    "plt.plot(LG_fpr, LG_tpr, 'm', label = 'Logistic_Regression_AUC = %0.2f' % LG_roc_auc)\n",
    "plt.plot(NN_fpr, NN_tpr, 'g', label = 'Neural_net(10*10)_AUC = %0.2f' % NN_roc_auc)\n",
    "plt.plot(RF_fpr, RF_tpr, 'c', label = 'Random_Forest_AUC = %0.2f' % RF_roc_auc)\n",
    "plt.plot(ES_fpr, ES_tpr, 'y', label = 'Ensemble(logitistic+NN+Random_forest)_AUC = %0.2f' % ES_roc_auc)\n",
    "\n",
    "plt.legend(loc = 'lower right')\n",
    "plt.plot([0, 1], [0, 1],'r--')\n",
    "plt.xlim([0, 1])\n",
    "plt.ylim([0, 1])\n",
    "plt.ylabel('True Positive Rate (tpr)')\n",
    "plt.xlabel('False Positive Rate (fpr)')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## Evaluate model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[93.417742091757802, 91.994399142368678, 93.156139533866124, 99.407721807401757, 95.82467315307656]\n"
     ]
    }
   ],
   "source": [
    "y_pred = DT_clf.predict(X_train)\n",
    "DT_train_ac = accuracy_score(y_train,y_pred)*100\n",
    "y_pred = NN_clf.predict(X_train)\n",
    "NN_train_ac = accuracy_score(y_train,y_pred)*100\n",
    "y_pred = LG_clf.predict(X_train)\n",
    "LG_train_ac = accuracy_score(y_train,y_pred)*100\n",
    "y_pred = RF_clf.predict(X_train)\n",
    "RF_train_ac = accuracy_score(y_train,y_pred)*100\n",
    "y_pred = ES_clf.predict(X_train)\n",
    "ES_train_ac = accuracy_score(y_train,y_pred)*100\n",
    "Training_ac = [DT_train_ac,LG_train_ac,NN_train_ac,RF_train_ac,ES_train_ac]\n",
    "print(Training_ac)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[76.157673651047659, 90.443566534980249, 90.123518527779169, 78.468020203030449, 90.434815222283333]\n",
      "[75.651467801865209, 90.347017350867546, 91.779749478079324, 83.108206669195084, 91.320716060132654]\n",
      "[77.014028056112224, 90.523547094188373, 88.101202404809627, 71.360220440881761, 89.323647294589179]\n",
      "[76.32666741146241, 90.435196075977871, 89.902862985685061, 76.787471191557614, 90.311142629199537]\n"
     ]
    }
   ],
   "source": [
    "y_pred = DT_clf.predict(X_test)\n",
    "DT_test_ac = accuracy_score(y_test,y_pred)*100\n",
    "DT_p_score = precision_score(y_test,y_pred)*100\n",
    "DT_r_score = recall_score(y_test,y_pred)*100\n",
    "DT_f1_score = 2*(DT_p_score*DT_r_score)/(DT_p_score+DT_r_score)\n",
    "\n",
    "y_pred = LG_clf.predict(X_test)\n",
    "LG_test_ac = accuracy_score(y_test,y_pred)*100\n",
    "LG_p_score = precision_score(y_test,y_pred)*100\n",
    "LG_r_score = recall_score(y_test,y_pred)*100\n",
    "LG_f1_score = 2*(LG_p_score*LG_r_score)/(LG_p_score+LG_r_score)\n",
    "\n",
    "y_pred = NN_clf.predict(X_test)\n",
    "NN_test_ac = accuracy_score(y_test,y_pred)*100\n",
    "NN_p_score = precision_score(y_test,y_pred)*100\n",
    "NN_r_score = recall_score(y_test,y_pred)*100\n",
    "NN_f1_score = 2*(NN_p_score*NN_r_score)/(NN_p_score+NN_r_score)\n",
    "\n",
    "y_pred = RF_clf.predict(X_test)\n",
    "RF_test_ac = accuracy_score(y_test,y_pred)*100\n",
    "RF_test_ac = accuracy_score(y_test,y_pred)*100\n",
    "RF_p_score = precision_score(y_test,y_pred)*100\n",
    "RF_r_score = recall_score(y_test,y_pred)*100\n",
    "RF_f1_score = 2*(RF_p_score*RF_r_score)/(RF_p_score+RF_r_score)\n",
    "\n",
    "y_pred = ES_clf.predict(X_test)\n",
    "ES_test_ac = accuracy_score(y_test,y_pred)*100\n",
    "ES_test_ac = accuracy_score(y_test,y_pred)*100\n",
    "ES_p_score = precision_score(y_test,y_pred)*100\n",
    "ES_r_score = recall_score(y_test,y_pred)*100\n",
    "ES_f1_score = 2*(ES_p_score*ES_r_score)/(ES_p_score+ES_r_score)\n",
    "\n",
    "Testing_ac = [DT_test_ac,LG_test_ac,NN_test_ac,RF_test_ac,ES_test_ac]\n",
    "print(Testing_ac)\n",
    "p_score = [DT_p_score,LG_p_score,NN_p_score,RF_p_score,ES_p_score]\n",
    "print(p_score)\n",
    "r_score = [DT_r_score,LG_r_score,NN_r_score,RF_r_score,ES_r_score]\n",
    "print(r_score)\n",
    "f1_score = [DT_f1_score,LG_f1_score,NN_f1_score,RF_f1_score,ES_f1_score]\n",
    "print(f1_score)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feature importance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "values = sorted(zip(feature_tfidf_name, RF_clf.feature_importances_), key=lambda x: x[1] * -1)\n",
    "feature_imp = {}\n",
    "for i in zip(feature_tfidf_name, RF_clf.feature_importances_*100):\n",
    "    feature_imp[i[0]] = i[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[('not', 0.020812454729523897), ('great', 0.011469824135427315), ('waste', 0.0074298190663252368), ('and', 0.0054899028737268717), ('best', 0.0054132389636220033), ('money', 0.0048467353136542393), ('bad', 0.0043647785224612346), ('was', 0.0043067911473494373), ('good', 0.0042278323380913405), ('the', 0.0042183249253557424), ('love', 0.0042044315897036928), ('boring', 0.0041962503533852858), ('but', 0.0039754334912661263), ('easy', 0.0038757622973307356), ('it', 0.0037445684814688693), ('excellent', 0.0037414389727968927), ('for', 0.0036714666000187044), ('this', 0.0036158761734949572), ('is', 0.0036148620332547864), ('don', 0.003538723803220927), ('worst', 0.0035265968563098112), ('of', 0.003506178050248894), ('to', 0.0034763272718490042), ('perfect', 0.003454931381316334), ('poor', 0.0033924146584344158), ('my', 0.0033677523732797672), ('no', 0.003322701223017221), ('disappointed', 0.0031840367702931595), ('you', 0.0031353170782277159), ('disappointing', 0.0030334692688426608), ('highly', 0.0030215287832989048), ('do', 0.0030095305925312574), ('horrible', 0.0029908909655939147), ('in', 0.002990604973833441), ('well', 0.0029452258903113564), ('amazing', 0.0028173453248039326), ('awesome', 0.0027982857417187465), ('that', 0.0027726317085035846), ('on', 0.0026778703790062849), ('if', 0.0026457304878938818), ('didn', 0.0025656945427268182), ('with', 0.002505161413830931), ('be', 0.0024655732861665149), ('have', 0.0024510863521910809), ('awful', 0.002451049031773261), ('only', 0.0022683634569497417), ('or', 0.002212458940087831), ('beautiful', 0.0021824545398920984), ('would', 0.0021417861647770762), ('as', 0.0021207048048827548)]\n"
     ]
    }
   ],
   "source": [
    "print(values[:50])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "top_50_imp_words = []\n",
    "top_50_imp_rate = []\n",
    "for i in values[:50]:\n",
    "    top_50_imp_words.append(i[0])\n",
    "    top_50_imp_rate.append(i[1]*100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['not', 'great', 'waste', 'and', 'best', 'money', 'bad', 'was', 'good', 'the', 'love', 'boring', 'but', 'easy', 'it', 'excellent', 'for', 'this', 'is', 'don', 'worst', 'of', 'to', 'perfect', 'poor', 'my', 'no', 'disappointed', 'you', 'disappointing', 'highly', 'do', 'horrible', 'in', 'well', 'amazing', 'awesome', 'that', 'on', 'if', 'didn', 'with', 'be', 'have', 'awful', 'only', 'or', 'beautiful', 'would', 'as']\n",
      "[2.0812454729523897, 1.1469824135427316, 0.74298190663252373, 0.54899028737268718, 0.54132389636220035, 0.4846735313654239, 0.43647785224612345, 0.43067911473494375, 0.42278323380913407, 0.42183249253557425, 0.42044315897036927, 0.41962503533852857, 0.39754334912661266, 0.38757622973307354, 0.37445684814688696, 0.37414389727968927, 0.36714666000187046, 0.36158761734949574, 0.36148620332547865, 0.35387238032209267, 0.3526596856309811, 0.35061780502488937, 0.3476327271849004, 0.3454931381316334, 0.33924146584344156, 0.33677523732797671, 0.33227012230172209, 0.31840367702931593, 0.31353170782277157, 0.30334692688426607, 0.30215287832989046, 0.30095305925312577, 0.29908909655939148, 0.29906049738334411, 0.29452258903113565, 0.28173453248039326, 0.27982857417187468, 0.27726317085035845, 0.26778703790062847, 0.2645730487893882, 0.2565694542726818, 0.25051614138309308, 0.2465573286166515, 0.24510863521910808, 0.24510490317732608, 0.22683634569497418, 0.22124589400878308, 0.21824545398920983, 0.21417861647770761, 0.21207048048827548]\n"
     ]
    }
   ],
   "source": [
    "print(top_50_imp_words)\n",
    "print(top_50_imp_rate)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['top_50_imp_words.pkl']"
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joblib.dump(top_50_imp_words,'top_50_imp_words.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "top_50_words = joblib.load('top_50_words.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "feature_imp[''] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "importances = []\n",
    "for i in top_50_words:\n",
    "    importances.append(feature_imp[i])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.42183249253557425, 0.54899028737268718, 0.3476327271849004, 0.35061780502488937, 0.37445684814688696, 0.36158761734949574, 0.36148620332547865, 0.29906049738334411, 0.36714666000187046, 0.27726317085035845, 0.43067911473494375, 2.0812454729523897, 0.31353170782277157, 0.25051614138309308, 0.39754334912661266, 0.26778703790062847, 0.1976717872098287, 0.24510863521910808, 0.33677523732797671, 0.21207048048827548, 0.20095098497128552, 0.19798552855707593, 0.2465573286166515, 0.18356616475032722, 0.19444382252304371, 1.1469824135427316, 0.20259314360393488, 0.2645730487893882, 0.19962085628122597, 0.42278323380913407, 0.18087395050641653, 0.089112252863010055, 0.17404022444932099, 0.15965502092489214, 0.19234199004250196, 0.22124589400878308, 0.21417861647770761, 0.15807764204268424, 0.15458651448981373, 0.16900819877380632, 0.16677455431975585, 0.14429317084921181, 0.20813822697484033, 0.15471591427347614, 0.17173657558539837, 0.14459037202458996, 0.14473195127703872, 0.1858511370401377, 0.17042478769578531, 0.12313512079524856]\n"
     ]
    }
   ],
   "source": [
    "print(importances)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python3.5",
   "language": "python",
   "name": "py35"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}