LogisticRegression_model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import time\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from IPython.display import display\n",
    "from sklearn import metrics\n",
    "from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer\n",
    "from sklearn.model_selection import GridSearchCV, train_test_split\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "\n",
    "t = time.time()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 0 - Création des fonctions utiles"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def sentiment140(stop_words, lemmatization, negation, nb_data):\n",
    "    \"\"\"Fonction permettant d'ouvrir un échantillon de la base Sentiment140 préprocessée avec la combinaison \n",
    "    de booléens (stop_words, lemmatization, negation)\"\"\"\n",
    "    \n",
    "    file = \"train\"\n",
    "    if stop_words:\n",
    "        file += \"_stop\"\n",
    "    if lemmatization:\n",
    "        file += \"_lemm\"\n",
    "    if negation:\n",
    "        file += \"_neg\"\n",
    "    return pd.read_pickle(os.path.join(\"data\", \"Sentiment140\", file + \".bz2\")).sample(nb_data, random_state=1234)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "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>sentiment</th>\n",
       "      <th>id</th>\n",
       "      <th>date</th>\n",
       "      <th>query_string</th>\n",
       "      <th>user</th>\n",
       "      <th>text</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>599410</th>\n",
       "      <td>0</td>\n",
       "      <td>2220663113</td>\n",
       "      <td>Thu Jun 18 03:39:49 PDT 2009</td>\n",
       "      <td>NO_QUERY</td>\n",
       "      <td>gemgembam</td>\n",
       "      <td>open itunes will not open naooo the silence is...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>549856</th>\n",
       "      <td>0</td>\n",
       "      <td>2202759103</td>\n",
       "      <td>Tue Jun 16 22:21:43 PDT 2009</td>\n",
       "      <td>NO_QUERY</td>\n",
       "      <td>twistofcolour</td>\n",
       "      <td>the sun is already gone for the day</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1363048</th>\n",
       "      <td>4</td>\n",
       "      <td>2049441111</td>\n",
       "      <td>Fri Jun 05 16:53:10 PDT 2009</td>\n",
       "      <td>NO_QUERY</td>\n",
       "      <td>Niico1e</td>\n",
       "      <td>hiya this is my st tweet i am so tired though ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>558776</th>\n",
       "      <td>0</td>\n",
       "      <td>2204830292</td>\n",
       "      <td>Wed Jun 17 03:14:38 PDT 2009</td>\n",
       "      <td>NO_QUERY</td>\n",
       "      <td>MeriEsteban</td>\n",
       "      <td>i have the bac today at p m i do not know if i...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1582398</th>\n",
       "      <td>4</td>\n",
       "      <td>2190306594</td>\n",
       "      <td>Tue Jun 16 02:21:03 PDT 2009</td>\n",
       "      <td>NO_QUERY</td>\n",
       "      <td>MeghanMagnolia</td>\n",
       "      <td>as far as i remember you were not half bad ine...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1353343</th>\n",
       "      <td>4</td>\n",
       "      <td>2046706853</td>\n",
       "      <td>Fri Jun 05 12:34:26 PDT 2009</td>\n",
       "      <td>NO_QUERY</td>\n",
       "      <td>WahooPooh</td>\n",
       "      <td>as tempting as that is i am just waiting on my...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>586354</th>\n",
       "      <td>0</td>\n",
       "      <td>2215937155</td>\n",
       "      <td>Wed Jun 17 18:49:10 PDT 2009</td>\n",
       "      <td>NO_QUERY</td>\n",
       "      <td>amandarie</td>\n",
       "      <td>cool how i have to be stuck at home i want to ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>860355</th>\n",
       "      <td>4</td>\n",
       "      <td>1676343245</td>\n",
       "      <td>Fri May 01 21:46:07 PDT 2009</td>\n",
       "      <td>NO_QUERY</td>\n",
       "      <td>MatchesMalone</td>\n",
       "      <td>oh and followfriday my writing partner because...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>521668</th>\n",
       "      <td>0</td>\n",
       "      <td>2192486836</td>\n",
       "      <td>Tue Jun 16 07:06:35 PDT 2009</td>\n",
       "      <td>NO_QUERY</td>\n",
       "      <td>hockeyfan45214</td>\n",
       "      <td>you would think if i continue to block someone...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>243888</th>\n",
       "      <td>0</td>\n",
       "      <td>1981647618</td>\n",
       "      <td>Sun May 31 09:35:54 PDT 2009</td>\n",
       "      <td>NO_QUERY</td>\n",
       "      <td>ilovemytroops</td>\n",
       "      <td>nevaeh buchanan missing friends of her mother ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         sentiment          id                          date query_string  \\\n",
       "599410           0  2220663113  Thu Jun 18 03:39:49 PDT 2009     NO_QUERY   \n",
       "549856           0  2202759103  Tue Jun 16 22:21:43 PDT 2009     NO_QUERY   \n",
       "1363048          4  2049441111  Fri Jun 05 16:53:10 PDT 2009     NO_QUERY   \n",
       "558776           0  2204830292  Wed Jun 17 03:14:38 PDT 2009     NO_QUERY   \n",
       "1582398          4  2190306594  Tue Jun 16 02:21:03 PDT 2009     NO_QUERY   \n",
       "1353343          4  2046706853  Fri Jun 05 12:34:26 PDT 2009     NO_QUERY   \n",
       "586354           0  2215937155  Wed Jun 17 18:49:10 PDT 2009     NO_QUERY   \n",
       "860355           4  1676343245  Fri May 01 21:46:07 PDT 2009     NO_QUERY   \n",
       "521668           0  2192486836  Tue Jun 16 07:06:35 PDT 2009     NO_QUERY   \n",
       "243888           0  1981647618  Sun May 31 09:35:54 PDT 2009     NO_QUERY   \n",
       "\n",
       "                   user                                               text  \n",
       "599410        gemgembam  open itunes will not open naooo the silence is...  \n",
       "549856    twistofcolour                the sun is already gone for the day  \n",
       "1363048         Niico1e  hiya this is my st tweet i am so tired though ...  \n",
       "558776      MeriEsteban  i have the bac today at p m i do not know if i...  \n",
       "1582398  MeghanMagnolia  as far as i remember you were not half bad ine...  \n",
       "1353343       WahooPooh  as tempting as that is i am just waiting on my...  \n",
       "586354        amandarie  cool how i have to be stuck at home i want to ...  \n",
       "860355    MatchesMalone  oh and followfriday my writing partner because...  \n",
       "521668   hockeyfan45214  you would think if i continue to block someone...  \n",
       "243888    ilovemytroops  nevaeh buchanan missing friends of her mother ...  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sentiment140(False, False, False, 10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Modelisation():\n",
    "    \"\"\"Classe permettant d'entrainer un modèle et de l'évaluer\"\"\"\n",
    "    \n",
    "    def __init__(self, X, y, vectorizer, model, scaling=True):\n",
    "        X = vectorizer.fit_transform(X)\n",
    "\n",
    "        if scaling:\n",
    "            scaler = StandardScaler(with_mean=False)\n",
    "            X = scaler.fit_transform(X)\n",
    "\n",
    "        X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.80, random_state=1234)\n",
    "\n",
    "        model.fit(X_train, y_train)\n",
    "        y_pred = model.predict(X_test)\n",
    "\n",
    "        sc_accuracy = metrics.accuracy_score(y_pred, y_test)    \n",
    "        sc_balanced_accuracy = metrics.balanced_accuracy_score(y_pred, y_test)\n",
    "        sc_roc_auc = metrics.roc_auc_score(y_test, model.predict_proba(X_test)[:, 1])\n",
    "        probs = model.predict_proba(X_test)[:, 1]\n",
    "\n",
    "        metrics_score = {'accuracy': sc_accuracy, 'balanced_accuracy': sc_balanced_accuracy, 'roc_auc': sc_roc_auc}\n",
    "\n",
    "        self.X_train, self.X_test, self.y_train, self.y_test = X_train, X_test, y_train, y_test\n",
    "        self.model = model\n",
    "        self.probs = probs\n",
    "        self.metrics_score = metrics_score\n",
    "    \n",
    "    def get_data(self):\n",
    "        return self.X_train, self.X_test, self.y_train, self.y_test\n",
    "    \n",
    "    def show_conf_matrix(self):\n",
    "        metrics.plot_confusion_matrix(self.model, self.X_test, self.y_test, cmap='Blues')\n",
    "        plt.show()\n",
    "\n",
    "        print(f\"Accuracy : {self.metrics_score['accuracy']:.4f}\")\n",
    "        print(f\"Balanced accuracy : {self.metrics_score['balanced_accuracy']:.4f}\")\n",
    "        print(f\"ROC AUC : {self.metrics_score['roc_auc']:.4f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def comparaison(X, y, N, model, scaling=True, show_graph=True):\n",
    "    \"\"\"\n",
    "    Création d'un DataFrame qui contient les performances des différents couples (vectorizer, model) étudiés\n",
    "    Les performances regardées sont l'aire sous la courbe ROC, la précision et la \"balanced accuracy\"\n",
    "    N correspond au nombre de features que vont utiliser les vectorizers\n",
    "    \"\"\"\n",
    "    table = []\n",
    "    \n",
    "    m = Modelisation(X, y, CountVectorizer(max_features = N), model)\n",
    "    table.append(m.metrics_score)\n",
    "\n",
    "    m = Modelisation(X, y, TfidfVectorizer(max_features = N), model)\n",
    "    table.append(m.metrics_score)\n",
    "\n",
    "    m = Modelisation(X, y, CountVectorizer(max_features = N, ngram_range=(1,2)), model)\n",
    "    table.append(m.metrics_score)\n",
    "\n",
    "    m = Modelisation(X, y, TfidfVectorizer(max_features = N, ngram_range=(1,2)), model)\n",
    "    table.append(m.metrics_score)\n",
    "    \n",
    "    \n",
    "    df = pd.DataFrame(table, index=[f\"CountVectorizer {N}\", f\"TfidfVectorizer {N}\", f\"CountVectorizer {N} ngram 2\", f\"TfidfVectorizer {N} ngram 2\"])\n",
    "    \n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def influence_preprocessing(N, model, show_graph=False, show_table=False):\n",
    "    \"\"\"Création d'un DataFrame qui regroupe les performances des couples (vectorizer, model) sur les différents \n",
    "    \"états de preprocessing\" de la base Sentiment140, en fonction de la combinaison (stop_words, lemmatization, negation) \"\"\"\n",
    "    \n",
    "    tables = []\n",
    "    for stop_words in [False, True]:\n",
    "        for lemmatization in [False, True]:\n",
    "            for negation in [False, True]:\n",
    "                print(f\"stop_words: {stop_words}, lemmatization: {lemmatization}, negation: {negation}\")\n",
    "                df_temp = sentiment140(stop_words, lemmatization, negation, NB_DATA)\n",
    "                X = df_temp.text.to_list()\n",
    "                y = df_temp.sentiment.to_list()            \n",
    "                df_metrics = comparaison(X, y, N, model, show_graph=show_graph)\n",
    "                if show_table:\n",
    "                    display(df_metrics)\n",
    "                df_metrics['stop_words'] = stop_words\n",
    "                df_metrics['lemmatization'] = lemmatization\n",
    "                df_metrics['negation'] = negation\n",
    "                tables.append(df_metrics)\n",
    "    return pd.concat(tables)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def influence_features(X, y, max_features_list, model, scaling=True):\n",
    "    \"\"\"Comparaison des performances des couples (vectorizer, model) en faisant varier le nombre de features\"\"\"\n",
    "    \n",
    "    table = []\n",
    "    for max_features in max_features_list:\n",
    "        m = Modelisation(X, y, CountVectorizer(max_features=max_features), model)\n",
    "        metrics_score = m.metrics_score\n",
    "        metrics_score['features'] = max_features\n",
    "        metrics_score['method'] = \"CountVectorizer\"\n",
    "        table.append(metrics_score)\n",
    "\n",
    "        m = Modelisation(X, y, TfidfVectorizer(max_features=max_features), model)\n",
    "        metrics_score = m.metrics_score\n",
    "        metrics_score['features'] = max_features\n",
    "        metrics_score['method'] = \"TfidfVectorizer\"\n",
    "        table.append(metrics_score)\n",
    "\n",
    "        m = Modelisation(X, y, CountVectorizer(max_features=max_features, ngram_range=(1, 2)), model)\n",
    "        metrics_score = m.metrics_score\n",
    "        metrics_score['features'] = max_features\n",
    "        metrics_score['method'] = \"CountVectorizer ngram 2\"\n",
    "        table.append(metrics_score)\n",
    "\n",
    "        m = Modelisation(X, y, TfidfVectorizer(max_features=max_features, ngram_range=(1, 2)), model)\n",
    "        metrics_score = m.metrics_score\n",
    "        metrics_score['features'] = max_features\n",
    "        metrics_score['method'] = \"TfidfVectorizer ngram 2\"\n",
    "        table.append(metrics_score)\n",
    "\n",
    "    df = pd.DataFrame(table)\n",
    "\n",
    "    methods = list(set(df.method))\n",
    "    scores = list(df.columns[:-2])\n",
    "    fig, axes = plt.subplots(1, len(scores), figsize=(20, 6))\n",
    "    for i_score in range(len(scores)):\n",
    "        for j_method in range(len(methods)):\n",
    "            df[df.method == methods[j_method]].plot(x='features', y=scores[i_score], label=methods[j_method], ax=axes[i_score])\n",
    "        axes[i_score].set(xlabel='Nombre de features')\n",
    "        axes[i_score].legend()\n",
    "        axes[i_score].set_title(scores[i_score])\n",
    "    plt.show()\n",
    "    return df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 1 - Sélection des meilleurs paramètres du modèle Logistic Regression"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On commence par choisir le nombre de lignes que l'on extrait de la base Sentiment140, sur laquelle on va entraîner et évaluer notre modèle."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "NB_DATA = 50000"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1.1 - Comparaison des méthodes de preprocessing et de \"feature extraction\" (vectorizer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "stop_words: False, lemmatization: False, negation: False\n",
      "stop_words: False, lemmatization: False, negation: True\n",
      "stop_words: False, lemmatization: True, negation: False\n",
      "stop_words: False, lemmatization: True, negation: True\n",
      "stop_words: True, lemmatization: False, negation: False\n",
      "stop_words: True, lemmatization: False, negation: True\n",
      "stop_words: True, lemmatization: True, negation: False\n",
      "stop_words: True, lemmatization: True, negation: True\n",
      "Wall time: 3min 4s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "df_metrics = influence_preprocessing(2500, LogisticRegression(max_iter=100000))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "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>accuracy</th>\n",
       "      <th>balanced_accuracy</th>\n",
       "      <th>roc_auc</th>\n",
       "      <th>stop_words</th>\n",
       "      <th>lemmatization</th>\n",
       "      <th>negation</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7690</td>\n",
       "      <td>0.769546</td>\n",
       "      <td>0.849718</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7676</td>\n",
       "      <td>0.768166</td>\n",
       "      <td>0.847014</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7595</td>\n",
       "      <td>0.759925</td>\n",
       "      <td>0.845374</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7641</td>\n",
       "      <td>0.764576</td>\n",
       "      <td>0.845358</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7641</td>\n",
       "      <td>0.764490</td>\n",
       "      <td>0.844999</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7622</td>\n",
       "      <td>0.763052</td>\n",
       "      <td>0.843606</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7587</td>\n",
       "      <td>0.759043</td>\n",
       "      <td>0.843216</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7690</td>\n",
       "      <td>0.770321</td>\n",
       "      <td>0.842537</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7630</td>\n",
       "      <td>0.763778</td>\n",
       "      <td>0.842365</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7655</td>\n",
       "      <td>0.766850</td>\n",
       "      <td>0.840311</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7610</td>\n",
       "      <td>0.761350</td>\n",
       "      <td>0.838031</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7633</td>\n",
       "      <td>0.764461</td>\n",
       "      <td>0.837683</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7618</td>\n",
       "      <td>0.762782</td>\n",
       "      <td>0.837304</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7580</td>\n",
       "      <td>0.758747</td>\n",
       "      <td>0.836850</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7617</td>\n",
       "      <td>0.763102</td>\n",
       "      <td>0.836785</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7595</td>\n",
       "      <td>0.761025</td>\n",
       "      <td>0.835864</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7601</td>\n",
       "      <td>0.760315</td>\n",
       "      <td>0.835281</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7575</td>\n",
       "      <td>0.758171</td>\n",
       "      <td>0.834830</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7559</td>\n",
       "      <td>0.756176</td>\n",
       "      <td>0.832724</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7570</td>\n",
       "      <td>0.757661</td>\n",
       "      <td>0.832571</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7544</td>\n",
       "      <td>0.754591</td>\n",
       "      <td>0.832483</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7606</td>\n",
       "      <td>0.761022</td>\n",
       "      <td>0.830478</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7548</td>\n",
       "      <td>0.755065</td>\n",
       "      <td>0.830049</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7520</td>\n",
       "      <td>0.752225</td>\n",
       "      <td>0.829729</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7523</td>\n",
       "      <td>0.752487</td>\n",
       "      <td>0.828855</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7567</td>\n",
       "      <td>0.757397</td>\n",
       "      <td>0.827529</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7553</td>\n",
       "      <td>0.755952</td>\n",
       "      <td>0.827102</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7562</td>\n",
       "      <td>0.756849</td>\n",
       "      <td>0.825032</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7499</td>\n",
       "      <td>0.749967</td>\n",
       "      <td>0.824392</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7540</td>\n",
       "      <td>0.754728</td>\n",
       "      <td>0.824199</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7533</td>\n",
       "      <td>0.753775</td>\n",
       "      <td>0.823632</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7491</td>\n",
       "      <td>0.749411</td>\n",
       "      <td>0.819069</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                              accuracy  balanced_accuracy   roc_auc  \\\n",
       "TfidfVectorizer 2500 ngram 2    0.7690           0.769546  0.849718   \n",
       "TfidfVectorizer 2500 ngram 2    0.7676           0.768166  0.847014   \n",
       "TfidfVectorizer 2500            0.7595           0.759925  0.845374   \n",
       "TfidfVectorizer 2500            0.7641           0.764576  0.845358   \n",
       "TfidfVectorizer 2500            0.7641           0.764490  0.844999   \n",
       "TfidfVectorizer 2500 ngram 2    0.7622           0.763052  0.843606   \n",
       "TfidfVectorizer 2500            0.7587           0.759043  0.843216   \n",
       "CountVectorizer 2500 ngram 2    0.7690           0.770321  0.842537   \n",
       "TfidfVectorizer 2500 ngram 2    0.7630           0.763778  0.842365   \n",
       "CountVectorizer 2500 ngram 2    0.7655           0.766850  0.840311   \n",
       "TfidfVectorizer 2500 ngram 2    0.7610           0.761350  0.838031   \n",
       "CountVectorizer 2500            0.7633           0.764461  0.837683   \n",
       "CountVectorizer 2500            0.7618           0.762782  0.837304   \n",
       "CountVectorizer 2500            0.7580           0.758747  0.836850   \n",
       "CountVectorizer 2500 ngram 2    0.7617           0.763102  0.836785   \n",
       "CountVectorizer 2500 ngram 2    0.7595           0.761025  0.835864   \n",
       "TfidfVectorizer 2500 ngram 2    0.7601           0.760315  0.835281   \n",
       "CountVectorizer 2500            0.7575           0.758171  0.834830   \n",
       "TfidfVectorizer 2500 ngram 2    0.7559           0.756176  0.832724   \n",
       "CountVectorizer 2500 ngram 2    0.7570           0.757661  0.832571   \n",
       "TfidfVectorizer 2500            0.7544           0.754591  0.832483   \n",
       "CountVectorizer 2500 ngram 2    0.7606           0.761022  0.830478   \n",
       "TfidfVectorizer 2500 ngram 2    0.7548           0.755065  0.830049   \n",
       "TfidfVectorizer 2500            0.7520           0.752225  0.829729   \n",
       "TfidfVectorizer 2500            0.7523           0.752487  0.828855   \n",
       "CountVectorizer 2500 ngram 2    0.7567           0.757397  0.827529   \n",
       "CountVectorizer 2500            0.7553           0.755952  0.827102   \n",
       "CountVectorizer 2500 ngram 2    0.7562           0.756849  0.825032   \n",
       "TfidfVectorizer 2500            0.7499           0.749967  0.824392   \n",
       "CountVectorizer 2500            0.7540           0.754728  0.824199   \n",
       "CountVectorizer 2500            0.7533           0.753775  0.823632   \n",
       "CountVectorizer 2500            0.7491           0.749411  0.819069   \n",
       "\n",
       "                              stop_words  lemmatization  negation  \n",
       "TfidfVectorizer 2500 ngram 2       False           True     False  \n",
       "TfidfVectorizer 2500 ngram 2       False          False     False  \n",
       "TfidfVectorizer 2500               False           True      True  \n",
       "TfidfVectorizer 2500               False          False      True  \n",
       "TfidfVectorizer 2500               False          False     False  \n",
       "TfidfVectorizer 2500 ngram 2       False           True      True  \n",
       "TfidfVectorizer 2500               False           True     False  \n",
       "CountVectorizer 2500 ngram 2       False           True     False  \n",
       "TfidfVectorizer 2500 ngram 2       False          False      True  \n",
       "CountVectorizer 2500 ngram 2       False          False     False  \n",
       "TfidfVectorizer 2500 ngram 2        True          False     False  \n",
       "CountVectorizer 2500               False          False      True  \n",
       "CountVectorizer 2500               False          False     False  \n",
       "CountVectorizer 2500               False           True      True  \n",
       "CountVectorizer 2500 ngram 2       False           True      True  \n",
       "CountVectorizer 2500 ngram 2       False          False      True  \n",
       "TfidfVectorizer 2500 ngram 2        True           True     False  \n",
       "CountVectorizer 2500               False           True     False  \n",
       "TfidfVectorizer 2500 ngram 2        True          False      True  \n",
       "CountVectorizer 2500 ngram 2        True          False     False  \n",
       "TfidfVectorizer 2500                True          False      True  \n",
       "CountVectorizer 2500 ngram 2        True           True     False  \n",
       "TfidfVectorizer 2500 ngram 2        True           True      True  \n",
       "TfidfVectorizer 2500                True           True      True  \n",
       "TfidfVectorizer 2500                True          False     False  \n",
       "CountVectorizer 2500 ngram 2        True          False      True  \n",
       "CountVectorizer 2500                True          False      True  \n",
       "CountVectorizer 2500 ngram 2        True           True      True  \n",
       "TfidfVectorizer 2500                True           True     False  \n",
       "CountVectorizer 2500                True           True      True  \n",
       "CountVectorizer 2500                True          False     False  \n",
       "CountVectorizer 2500                True           True     False  "
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_metrics.sort_values(by=\"roc_auc\", ascending=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "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>accuracy</th>\n",
       "      <th>balanced_accuracy</th>\n",
       "      <th>roc_auc</th>\n",
       "      <th>stop_words</th>\n",
       "      <th>lemmatization</th>\n",
       "      <th>negation</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7690</td>\n",
       "      <td>0.770321</td>\n",
       "      <td>0.842537</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7690</td>\n",
       "      <td>0.769546</td>\n",
       "      <td>0.849718</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7676</td>\n",
       "      <td>0.768166</td>\n",
       "      <td>0.847014</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7655</td>\n",
       "      <td>0.766850</td>\n",
       "      <td>0.840311</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7641</td>\n",
       "      <td>0.764576</td>\n",
       "      <td>0.845358</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7641</td>\n",
       "      <td>0.764490</td>\n",
       "      <td>0.844999</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7633</td>\n",
       "      <td>0.764461</td>\n",
       "      <td>0.837683</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7630</td>\n",
       "      <td>0.763778</td>\n",
       "      <td>0.842365</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7617</td>\n",
       "      <td>0.763102</td>\n",
       "      <td>0.836785</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7622</td>\n",
       "      <td>0.763052</td>\n",
       "      <td>0.843606</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7618</td>\n",
       "      <td>0.762782</td>\n",
       "      <td>0.837304</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7610</td>\n",
       "      <td>0.761350</td>\n",
       "      <td>0.838031</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7595</td>\n",
       "      <td>0.761025</td>\n",
       "      <td>0.835864</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7606</td>\n",
       "      <td>0.761022</td>\n",
       "      <td>0.830478</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7601</td>\n",
       "      <td>0.760315</td>\n",
       "      <td>0.835281</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7595</td>\n",
       "      <td>0.759925</td>\n",
       "      <td>0.845374</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7587</td>\n",
       "      <td>0.759043</td>\n",
       "      <td>0.843216</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7580</td>\n",
       "      <td>0.758747</td>\n",
       "      <td>0.836850</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7575</td>\n",
       "      <td>0.758171</td>\n",
       "      <td>0.834830</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7570</td>\n",
       "      <td>0.757661</td>\n",
       "      <td>0.832571</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7567</td>\n",
       "      <td>0.757397</td>\n",
       "      <td>0.827529</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7562</td>\n",
       "      <td>0.756849</td>\n",
       "      <td>0.825032</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7559</td>\n",
       "      <td>0.756176</td>\n",
       "      <td>0.832724</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7553</td>\n",
       "      <td>0.755952</td>\n",
       "      <td>0.827102</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7548</td>\n",
       "      <td>0.755065</td>\n",
       "      <td>0.830049</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7540</td>\n",
       "      <td>0.754728</td>\n",
       "      <td>0.824199</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7544</td>\n",
       "      <td>0.754591</td>\n",
       "      <td>0.832483</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7533</td>\n",
       "      <td>0.753775</td>\n",
       "      <td>0.823632</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7523</td>\n",
       "      <td>0.752487</td>\n",
       "      <td>0.828855</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7520</td>\n",
       "      <td>0.752225</td>\n",
       "      <td>0.829729</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7499</td>\n",
       "      <td>0.749967</td>\n",
       "      <td>0.824392</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7491</td>\n",
       "      <td>0.749411</td>\n",
       "      <td>0.819069</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                              accuracy  balanced_accuracy   roc_auc  \\\n",
       "CountVectorizer 2500 ngram 2    0.7690           0.770321  0.842537   \n",
       "TfidfVectorizer 2500 ngram 2    0.7690           0.769546  0.849718   \n",
       "TfidfVectorizer 2500 ngram 2    0.7676           0.768166  0.847014   \n",
       "CountVectorizer 2500 ngram 2    0.7655           0.766850  0.840311   \n",
       "TfidfVectorizer 2500            0.7641           0.764576  0.845358   \n",
       "TfidfVectorizer 2500            0.7641           0.764490  0.844999   \n",
       "CountVectorizer 2500            0.7633           0.764461  0.837683   \n",
       "TfidfVectorizer 2500 ngram 2    0.7630           0.763778  0.842365   \n",
       "CountVectorizer 2500 ngram 2    0.7617           0.763102  0.836785   \n",
       "TfidfVectorizer 2500 ngram 2    0.7622           0.763052  0.843606   \n",
       "CountVectorizer 2500            0.7618           0.762782  0.837304   \n",
       "TfidfVectorizer 2500 ngram 2    0.7610           0.761350  0.838031   \n",
       "CountVectorizer 2500 ngram 2    0.7595           0.761025  0.835864   \n",
       "CountVectorizer 2500 ngram 2    0.7606           0.761022  0.830478   \n",
       "TfidfVectorizer 2500 ngram 2    0.7601           0.760315  0.835281   \n",
       "TfidfVectorizer 2500            0.7595           0.759925  0.845374   \n",
       "TfidfVectorizer 2500            0.7587           0.759043  0.843216   \n",
       "CountVectorizer 2500            0.7580           0.758747  0.836850   \n",
       "CountVectorizer 2500            0.7575           0.758171  0.834830   \n",
       "CountVectorizer 2500 ngram 2    0.7570           0.757661  0.832571   \n",
       "CountVectorizer 2500 ngram 2    0.7567           0.757397  0.827529   \n",
       "CountVectorizer 2500 ngram 2    0.7562           0.756849  0.825032   \n",
       "TfidfVectorizer 2500 ngram 2    0.7559           0.756176  0.832724   \n",
       "CountVectorizer 2500            0.7553           0.755952  0.827102   \n",
       "TfidfVectorizer 2500 ngram 2    0.7548           0.755065  0.830049   \n",
       "CountVectorizer 2500            0.7540           0.754728  0.824199   \n",
       "TfidfVectorizer 2500            0.7544           0.754591  0.832483   \n",
       "CountVectorizer 2500            0.7533           0.753775  0.823632   \n",
       "TfidfVectorizer 2500            0.7523           0.752487  0.828855   \n",
       "TfidfVectorizer 2500            0.7520           0.752225  0.829729   \n",
       "TfidfVectorizer 2500            0.7499           0.749967  0.824392   \n",
       "CountVectorizer 2500            0.7491           0.749411  0.819069   \n",
       "\n",
       "                              stop_words  lemmatization  negation  \n",
       "CountVectorizer 2500 ngram 2       False           True     False  \n",
       "TfidfVectorizer 2500 ngram 2       False           True     False  \n",
       "TfidfVectorizer 2500 ngram 2       False          False     False  \n",
       "CountVectorizer 2500 ngram 2       False          False     False  \n",
       "TfidfVectorizer 2500               False          False      True  \n",
       "TfidfVectorizer 2500               False          False     False  \n",
       "CountVectorizer 2500               False          False      True  \n",
       "TfidfVectorizer 2500 ngram 2       False          False      True  \n",
       "CountVectorizer 2500 ngram 2       False           True      True  \n",
       "TfidfVectorizer 2500 ngram 2       False           True      True  \n",
       "CountVectorizer 2500               False          False     False  \n",
       "TfidfVectorizer 2500 ngram 2        True          False     False  \n",
       "CountVectorizer 2500 ngram 2       False          False      True  \n",
       "CountVectorizer 2500 ngram 2        True           True     False  \n",
       "TfidfVectorizer 2500 ngram 2        True           True     False  \n",
       "TfidfVectorizer 2500               False           True      True  \n",
       "TfidfVectorizer 2500               False           True     False  \n",
       "CountVectorizer 2500               False           True      True  \n",
       "CountVectorizer 2500               False           True     False  \n",
       "CountVectorizer 2500 ngram 2        True          False     False  \n",
       "CountVectorizer 2500 ngram 2        True          False      True  \n",
       "CountVectorizer 2500 ngram 2        True           True      True  \n",
       "TfidfVectorizer 2500 ngram 2        True          False      True  \n",
       "CountVectorizer 2500                True          False      True  \n",
       "TfidfVectorizer 2500 ngram 2        True           True      True  \n",
       "CountVectorizer 2500                True           True      True  \n",
       "TfidfVectorizer 2500                True          False      True  \n",
       "CountVectorizer 2500                True          False     False  \n",
       "TfidfVectorizer 2500                True          False     False  \n",
       "TfidfVectorizer 2500                True           True      True  \n",
       "TfidfVectorizer 2500                True           True     False  \n",
       "CountVectorizer 2500                True           True     False  "
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_metrics.sort_values(by=\"balanced_accuracy\", ascending=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "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>accuracy</th>\n",
       "      <th>balanced_accuracy</th>\n",
       "      <th>roc_auc</th>\n",
       "      <th>stop_words</th>\n",
       "      <th>lemmatization</th>\n",
       "      <th>negation</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7690</td>\n",
       "      <td>0.770321</td>\n",
       "      <td>0.842537</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7690</td>\n",
       "      <td>0.769546</td>\n",
       "      <td>0.849718</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7676</td>\n",
       "      <td>0.768166</td>\n",
       "      <td>0.847014</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7655</td>\n",
       "      <td>0.766850</td>\n",
       "      <td>0.840311</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7641</td>\n",
       "      <td>0.764490</td>\n",
       "      <td>0.844999</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7641</td>\n",
       "      <td>0.764576</td>\n",
       "      <td>0.845358</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7633</td>\n",
       "      <td>0.764461</td>\n",
       "      <td>0.837683</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7630</td>\n",
       "      <td>0.763778</td>\n",
       "      <td>0.842365</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7622</td>\n",
       "      <td>0.763052</td>\n",
       "      <td>0.843606</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7618</td>\n",
       "      <td>0.762782</td>\n",
       "      <td>0.837304</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7617</td>\n",
       "      <td>0.763102</td>\n",
       "      <td>0.836785</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7610</td>\n",
       "      <td>0.761350</td>\n",
       "      <td>0.838031</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7606</td>\n",
       "      <td>0.761022</td>\n",
       "      <td>0.830478</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7601</td>\n",
       "      <td>0.760315</td>\n",
       "      <td>0.835281</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7595</td>\n",
       "      <td>0.759925</td>\n",
       "      <td>0.845374</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7595</td>\n",
       "      <td>0.761025</td>\n",
       "      <td>0.835864</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7587</td>\n",
       "      <td>0.759043</td>\n",
       "      <td>0.843216</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7580</td>\n",
       "      <td>0.758747</td>\n",
       "      <td>0.836850</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7575</td>\n",
       "      <td>0.758171</td>\n",
       "      <td>0.834830</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7570</td>\n",
       "      <td>0.757661</td>\n",
       "      <td>0.832571</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7567</td>\n",
       "      <td>0.757397</td>\n",
       "      <td>0.827529</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7562</td>\n",
       "      <td>0.756849</td>\n",
       "      <td>0.825032</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7559</td>\n",
       "      <td>0.756176</td>\n",
       "      <td>0.832724</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7553</td>\n",
       "      <td>0.755952</td>\n",
       "      <td>0.827102</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500 ngram 2</th>\n",
       "      <td>0.7548</td>\n",
       "      <td>0.755065</td>\n",
       "      <td>0.830049</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7544</td>\n",
       "      <td>0.754591</td>\n",
       "      <td>0.832483</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7540</td>\n",
       "      <td>0.754728</td>\n",
       "      <td>0.824199</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7533</td>\n",
       "      <td>0.753775</td>\n",
       "      <td>0.823632</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7523</td>\n",
       "      <td>0.752487</td>\n",
       "      <td>0.828855</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7520</td>\n",
       "      <td>0.752225</td>\n",
       "      <td>0.829729</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TfidfVectorizer 2500</th>\n",
       "      <td>0.7499</td>\n",
       "      <td>0.749967</td>\n",
       "      <td>0.824392</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CountVectorizer 2500</th>\n",
       "      <td>0.7491</td>\n",
       "      <td>0.749411</td>\n",
       "      <td>0.819069</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                              accuracy  balanced_accuracy   roc_auc  \\\n",
       "CountVectorizer 2500 ngram 2    0.7690           0.770321  0.842537   \n",
       "TfidfVectorizer 2500 ngram 2    0.7690           0.769546  0.849718   \n",
       "TfidfVectorizer 2500 ngram 2    0.7676           0.768166  0.847014   \n",
       "CountVectorizer 2500 ngram 2    0.7655           0.766850  0.840311   \n",
       "TfidfVectorizer 2500            0.7641           0.764490  0.844999   \n",
       "TfidfVectorizer 2500            0.7641           0.764576  0.845358   \n",
       "CountVectorizer 2500            0.7633           0.764461  0.837683   \n",
       "TfidfVectorizer 2500 ngram 2    0.7630           0.763778  0.842365   \n",
       "TfidfVectorizer 2500 ngram 2    0.7622           0.763052  0.843606   \n",
       "CountVectorizer 2500            0.7618           0.762782  0.837304   \n",
       "CountVectorizer 2500 ngram 2    0.7617           0.763102  0.836785   \n",
       "TfidfVectorizer 2500 ngram 2    0.7610           0.761350  0.838031   \n",
       "CountVectorizer 2500 ngram 2    0.7606           0.761022  0.830478   \n",
       "TfidfVectorizer 2500 ngram 2    0.7601           0.760315  0.835281   \n",
       "TfidfVectorizer 2500            0.7595           0.759925  0.845374   \n",
       "CountVectorizer 2500 ngram 2    0.7595           0.761025  0.835864   \n",
       "TfidfVectorizer 2500            0.7587           0.759043  0.843216   \n",
       "CountVectorizer 2500            0.7580           0.758747  0.836850   \n",
       "CountVectorizer 2500            0.7575           0.758171  0.834830   \n",
       "CountVectorizer 2500 ngram 2    0.7570           0.757661  0.832571   \n",
       "CountVectorizer 2500 ngram 2    0.7567           0.757397  0.827529   \n",
       "CountVectorizer 2500 ngram 2    0.7562           0.756849  0.825032   \n",
       "TfidfVectorizer 2500 ngram 2    0.7559           0.756176  0.832724   \n",
       "CountVectorizer 2500            0.7553           0.755952  0.827102   \n",
       "TfidfVectorizer 2500 ngram 2    0.7548           0.755065  0.830049   \n",
       "TfidfVectorizer 2500            0.7544           0.754591  0.832483   \n",
       "CountVectorizer 2500            0.7540           0.754728  0.824199   \n",
       "CountVectorizer 2500            0.7533           0.753775  0.823632   \n",
       "TfidfVectorizer 2500            0.7523           0.752487  0.828855   \n",
       "TfidfVectorizer 2500            0.7520           0.752225  0.829729   \n",
       "TfidfVectorizer 2500            0.7499           0.749967  0.824392   \n",
       "CountVectorizer 2500            0.7491           0.749411  0.819069   \n",
       "\n",
       "                              stop_words  lemmatization  negation  \n",
       "CountVectorizer 2500 ngram 2       False           True     False  \n",
       "TfidfVectorizer 2500 ngram 2       False           True     False  \n",
       "TfidfVectorizer 2500 ngram 2       False          False     False  \n",
       "CountVectorizer 2500 ngram 2       False          False     False  \n",
       "TfidfVectorizer 2500               False          False     False  \n",
       "TfidfVectorizer 2500               False          False      True  \n",
       "CountVectorizer 2500               False          False      True  \n",
       "TfidfVectorizer 2500 ngram 2       False          False      True  \n",
       "TfidfVectorizer 2500 ngram 2       False           True      True  \n",
       "CountVectorizer 2500               False          False     False  \n",
       "CountVectorizer 2500 ngram 2       False           True      True  \n",
       "TfidfVectorizer 2500 ngram 2        True          False     False  \n",
       "CountVectorizer 2500 ngram 2        True           True     False  \n",
       "TfidfVectorizer 2500 ngram 2        True           True     False  \n",
       "TfidfVectorizer 2500               False           True      True  \n",
       "CountVectorizer 2500 ngram 2       False          False      True  \n",
       "TfidfVectorizer 2500               False           True     False  \n",
       "CountVectorizer 2500               False           True      True  \n",
       "CountVectorizer 2500               False           True     False  \n",
       "CountVectorizer 2500 ngram 2        True          False     False  \n",
       "CountVectorizer 2500 ngram 2        True          False      True  \n",
       "CountVectorizer 2500 ngram 2        True           True      True  \n",
       "TfidfVectorizer 2500 ngram 2        True          False      True  \n",
       "CountVectorizer 2500                True          False      True  \n",
       "TfidfVectorizer 2500 ngram 2        True           True      True  \n",
       "TfidfVectorizer 2500                True          False      True  \n",
       "CountVectorizer 2500                True           True      True  \n",
       "CountVectorizer 2500                True          False     False  \n",
       "TfidfVectorizer 2500                True          False     False  \n",
       "TfidfVectorizer 2500                True           True      True  \n",
       "TfidfVectorizer 2500                True           True     False  \n",
       "CountVectorizer 2500                True           True     False  "
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_metrics.sort_values(by=\"accuracy\", ascending=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Pour N=500, la combinaison optimale semble être stop_words=False, lemmatization=True, negation=False\n",
    "\n",
    "Cependant, N=500 est un nombre de features assez faible. Il est donc pertinent de regarder si de nouvelles tendances se dégagent lorsque l'on fait augmenter le nombre de features N.\n",
    "\n",
    "Pour N=1000, la combinaison optimale est de nouveau stop_words=False, lemmatization=True, negation=False\n",
    "\n",
    "Pour N=2500, la combinaison optimale est encore stop_words=False, lemmatization=True, negation=False\n",
    "\n",
    "Finalement, pour la suite, nous retiendrons la combinaison stop_words=False, lemmatization=True, negation=False\n",
    "\n",
    "Enfin, il apparaît que le vectorizer le plus performant est le TfidfVectorizer avec ngram. Pour la combinaison qui nous intéresse (stop_words=False, lemmatization=True, negation=False), ce vectorizer est presque toujours le plus performant."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Paramètres retenus :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Wall time: 11.8 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "NB_DATA = 50000\n",
    "STOP_WORDS = False\n",
    "LEMMATIZATION = True\n",
    "NEGATION = False\n",
    "\n",
    "df = sentiment140(STOP_WORDS, LEMMATIZATION, NEGATION, NB_DATA)\n",
    "X = df.text.to_list()\n",
    "y = df.sentiment.to_list()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1.2 - Comparaison par rapport au nombre de features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1440x432 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Wall time: 49.4 s\n"
     ]
    },
    {
     "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>accuracy</th>\n",
       "      <th>balanced_accuracy</th>\n",
       "      <th>roc_auc</th>\n",
       "      <th>features</th>\n",
       "      <th>method</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.7218</td>\n",
       "      <td>0.725297</td>\n",
       "      <td>0.795604</td>\n",
       "      <td>199</td>\n",
       "      <td>CountVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.7234</td>\n",
       "      <td>0.724577</td>\n",
       "      <td>0.800729</td>\n",
       "      <td>199</td>\n",
       "      <td>TfidfVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.7106</td>\n",
       "      <td>0.713485</td>\n",
       "      <td>0.784691</td>\n",
       "      <td>199</td>\n",
       "      <td>CountVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.7131</td>\n",
       "      <td>0.713956</td>\n",
       "      <td>0.789587</td>\n",
       "      <td>199</td>\n",
       "      <td>TfidfVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.7398</td>\n",
       "      <td>0.742304</td>\n",
       "      <td>0.816651</td>\n",
       "      <td>380</td>\n",
       "      <td>CountVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>0.7387</td>\n",
       "      <td>0.739681</td>\n",
       "      <td>0.823818</td>\n",
       "      <td>380</td>\n",
       "      <td>TfidfVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>0.7348</td>\n",
       "      <td>0.738131</td>\n",
       "      <td>0.812344</td>\n",
       "      <td>380</td>\n",
       "      <td>CountVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>0.7368</td>\n",
       "      <td>0.737986</td>\n",
       "      <td>0.819207</td>\n",
       "      <td>380</td>\n",
       "      <td>TfidfVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>0.7551</td>\n",
       "      <td>0.756917</td>\n",
       "      <td>0.830562</td>\n",
       "      <td>724</td>\n",
       "      <td>CountVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>0.7550</td>\n",
       "      <td>0.755708</td>\n",
       "      <td>0.837628</td>\n",
       "      <td>724</td>\n",
       "      <td>TfidfVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.7468</td>\n",
       "      <td>0.749169</td>\n",
       "      <td>0.827869</td>\n",
       "      <td>724</td>\n",
       "      <td>CountVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>0.7516</td>\n",
       "      <td>0.752270</td>\n",
       "      <td>0.834258</td>\n",
       "      <td>724</td>\n",
       "      <td>TfidfVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>0.7579</td>\n",
       "      <td>0.759092</td>\n",
       "      <td>0.837043</td>\n",
       "      <td>1380</td>\n",
       "      <td>CountVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>0.7589</td>\n",
       "      <td>0.759412</td>\n",
       "      <td>0.844080</td>\n",
       "      <td>1380</td>\n",
       "      <td>TfidfVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>0.7625</td>\n",
       "      <td>0.764206</td>\n",
       "      <td>0.838551</td>\n",
       "      <td>1380</td>\n",
       "      <td>CountVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>0.7644</td>\n",
       "      <td>0.765040</td>\n",
       "      <td>0.845492</td>\n",
       "      <td>1380</td>\n",
       "      <td>TfidfVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>0.7604</td>\n",
       "      <td>0.760991</td>\n",
       "      <td>0.834509</td>\n",
       "      <td>2630</td>\n",
       "      <td>CountVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>0.7586</td>\n",
       "      <td>0.758922</td>\n",
       "      <td>0.842790</td>\n",
       "      <td>2630</td>\n",
       "      <td>TfidfVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>0.7692</td>\n",
       "      <td>0.770490</td>\n",
       "      <td>0.843332</td>\n",
       "      <td>2630</td>\n",
       "      <td>CountVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>0.7695</td>\n",
       "      <td>0.769990</td>\n",
       "      <td>0.850510</td>\n",
       "      <td>2630</td>\n",
       "      <td>TfidfVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>0.7505</td>\n",
       "      <td>0.750848</td>\n",
       "      <td>0.810947</td>\n",
       "      <td>5011</td>\n",
       "      <td>CountVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>0.7498</td>\n",
       "      <td>0.749976</td>\n",
       "      <td>0.820606</td>\n",
       "      <td>5011</td>\n",
       "      <td>TfidfVectorizer</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>0.7601</td>\n",
       "      <td>0.760633</td>\n",
       "      <td>0.831286</td>\n",
       "      <td>5011</td>\n",
       "      <td>CountVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>0.7616</td>\n",
       "      <td>0.761932</td>\n",
       "      <td>0.841681</td>\n",
       "      <td>5011</td>\n",
       "      <td>TfidfVectorizer ngram 2</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    accuracy  balanced_accuracy   roc_auc  features                   method\n",
       "0     0.7218           0.725297  0.795604       199          CountVectorizer\n",
       "1     0.7234           0.724577  0.800729       199          TfidfVectorizer\n",
       "2     0.7106           0.713485  0.784691       199  CountVectorizer ngram 2\n",
       "3     0.7131           0.713956  0.789587       199  TfidfVectorizer ngram 2\n",
       "4     0.7398           0.742304  0.816651       380          CountVectorizer\n",
       "5     0.7387           0.739681  0.823818       380          TfidfVectorizer\n",
       "6     0.7348           0.738131  0.812344       380  CountVectorizer ngram 2\n",
       "7     0.7368           0.737986  0.819207       380  TfidfVectorizer ngram 2\n",
       "8     0.7551           0.756917  0.830562       724          CountVectorizer\n",
       "9     0.7550           0.755708  0.837628       724          TfidfVectorizer\n",
       "10    0.7468           0.749169  0.827869       724  CountVectorizer ngram 2\n",
       "11    0.7516           0.752270  0.834258       724  TfidfVectorizer ngram 2\n",
       "12    0.7579           0.759092  0.837043      1380          CountVectorizer\n",
       "13    0.7589           0.759412  0.844080      1380          TfidfVectorizer\n",
       "14    0.7625           0.764206  0.838551      1380  CountVectorizer ngram 2\n",
       "15    0.7644           0.765040  0.845492      1380  TfidfVectorizer ngram 2\n",
       "16    0.7604           0.760991  0.834509      2630          CountVectorizer\n",
       "17    0.7586           0.758922  0.842790      2630          TfidfVectorizer\n",
       "18    0.7692           0.770490  0.843332      2630  CountVectorizer ngram 2\n",
       "19    0.7695           0.769990  0.850510      2630  TfidfVectorizer ngram 2\n",
       "20    0.7505           0.750848  0.810947      5011          CountVectorizer\n",
       "21    0.7498           0.749976  0.820606      5011          TfidfVectorizer\n",
       "22    0.7601           0.760633  0.831286      5011  CountVectorizer ngram 2\n",
       "23    0.7616           0.761932  0.841681      5011  TfidfVectorizer ngram 2"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "%%time\n",
    "max_features_list = np.logspace(2.3, 3.7, 6, endpoint=True, dtype=int)\n",
    "influence_features(X, y, max_features_list, LogisticRegression(max_iter=1000000))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Comme annoncé précédemment, le vectorizer le plus performant est le TfidfVectorizer avec ngram. C'est donc ce vectorizer que nous utiliserons pour nos estimations futures."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1.3 - GridSearch pour obtenir les meilleurs paramètres"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On choisit les paramètres au vu des études précédentes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.80, random_state=1234)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Balanced Accuracy : 0.851\n",
      "Best Balanced Accuracy Through Grid Search : 0.781\n",
      "Best Parameters :  {'clf__C': 1.0, 'idf__max_features': 26211, 'idf__ngram_range': (1, 2)}\n",
      "Wall time: 18min 28s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "#Tfidfvectorizer\n",
    "text_clf = Pipeline([('idf', TfidfVectorizer()), ('clf', LogisticRegression(max_iter=1000000, random_state=1234))])\n",
    "nb_max_words = len(TfidfVectorizer().fit(X_train).get_feature_names())\n",
    "\n",
    "parameters = {'idf__max_features' : np.logspace(2.3, np.log10(nb_max_words), 6, endpoint=True, dtype=int),\n",
    "           'idf__ngram_range' : [(1,1),(1,2)],\n",
    "           'clf__C' : np.logspace(-3,3,7)\n",
    "}\n",
    "\n",
    "grid_clf = GridSearchCV(text_clf, parameters, scoring='balanced_accuracy', cv=5)\n",
    "_ = grid_clf.fit(X_train, y_train)\n",
    "\n",
    "print('Balanced Accuracy : %.3f' % grid_clf.best_estimator_.score(X_train, y_train))\n",
    "print('Best Balanced Accuracy Through Grid Search : %.3f' % grid_clf.best_score_)\n",
    "print('Best Parameters : ', grid_clf.best_params_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 2 - Estimation finale du modèle"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On applique finalement le modèle avec les paramètres suggérés par le GridSearch."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy : 0.6854\n",
      "Balanced accuracy : 0.6854\n",
      "ROC AUC : 0.7417\n",
      "Wall time: 9.16 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "m = Modelisation(X, y, TfidfVectorizer(max_features=26211, ngram_range=(1,2)), LogisticRegression(C=1, max_iter=1000000, random_state=1234))\n",
    "m.show_conf_matrix()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Les résultats obtenus avec les paramètres suggérés par le GridSearch ne sont pas très bons. Ils sont nettement meilleurs lorsque le nombre maximal de features vaut 2725."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy : 0.7694\n",
      "Balanced accuracy : 0.7699\n",
      "ROC AUC : 0.8492\n",
      "Wall time: 4.08 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "m = Modelisation(X, y, TfidfVectorizer(max_features=2725, ngram_range=(1,2)), LogisticRegression(C=1, max_iter=1000000, random_state=1234))\n",
    "m.show_conf_matrix()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "En conclusion, le modèle \"LogisticRegression\" couplé à la méthode de \"feature extraction\" TfidfVectorizer avec ngram donne les meilleurs résultats, tous modèles confondus. \n",
    "\n",
    "La précision ainsi obtenue avoisine les 77%, contre 74,3% et 73,4% pour les modèles faisant intervenir la classification bayésienne.\n",
    "\n",
    "Nous utiliserons donc les outils suivants pour traiter notre base webscrapée :\n",
    "\n",
    "- modèle : LogisticRegression(max_iter=1000000, random_state=1234)\n",
    "- vectorizer : TfidfVectorizer(max_features=2725, ngram_range=(1, 2))\n",
    "- combinaison : stop_words=False, lemmatization=True, negation=False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Temps d'exécution total : 00:23:01\n"
     ]
    }
   ],
   "source": [
    "print(f\"Temps d'exécution total : {time.strftime('%H:%M:%S', time.gmtime(time.time()-t))}\")"
   ]
  }
 ],
 "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}