Wine-quality.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Nous allons travailler avec un jeu de données qui contient des informations physico-chimiques de vins portugais (vinho verde), ainsi que leur qualité telle que notée par des humains.\n",
    "Le problème, sur ces données, est de prédire automatiquement la qualité sur la base de ces informations, afin d’assister le travail d’évaluation des œnologues, d’améliorer la production de vin, et de cibler le goût des consommateurs sur des marchés de niche.\n",
    "Ce jeu de données est disponible dans les archives UCI : http://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# coding: utf-8\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn import dummy\n",
    "from sklearn import model_selection\n",
    "from sklearn.model_selection import KFold\n",
    "from sklearn import preprocessing\n",
    "from sklearn import neighbors, metrics\n",
    "\n",
    "import warnings\n",
    "warnings.filterwarnings(action='ignore')\n",
    "\n",
    "seed = 42"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Chargement et préparation des données"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "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>fixed acidity</th>\n",
       "      <th>volatile acidity</th>\n",
       "      <th>citric acid</th>\n",
       "      <th>residual sugar</th>\n",
       "      <th>chlorides</th>\n",
       "      <th>free sulfur dioxide</th>\n",
       "      <th>total sulfur dioxide</th>\n",
       "      <th>density</th>\n",
       "      <th>pH</th>\n",
       "      <th>sulphates</th>\n",
       "      <th>alcohol</th>\n",
       "      <th>quality</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>7.0</td>\n",
       "      <td>0.27</td>\n",
       "      <td>0.36</td>\n",
       "      <td>20.7</td>\n",
       "      <td>0.045</td>\n",
       "      <td>45.0</td>\n",
       "      <td>170.0</td>\n",
       "      <td>1.0010</td>\n",
       "      <td>3.00</td>\n",
       "      <td>0.45</td>\n",
       "      <td>8.8</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6.3</td>\n",
       "      <td>0.30</td>\n",
       "      <td>0.34</td>\n",
       "      <td>1.6</td>\n",
       "      <td>0.049</td>\n",
       "      <td>14.0</td>\n",
       "      <td>132.0</td>\n",
       "      <td>0.9940</td>\n",
       "      <td>3.30</td>\n",
       "      <td>0.49</td>\n",
       "      <td>9.5</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>8.1</td>\n",
       "      <td>0.28</td>\n",
       "      <td>0.40</td>\n",
       "      <td>6.9</td>\n",
       "      <td>0.050</td>\n",
       "      <td>30.0</td>\n",
       "      <td>97.0</td>\n",
       "      <td>0.9951</td>\n",
       "      <td>3.26</td>\n",
       "      <td>0.44</td>\n",
       "      <td>10.1</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>7.2</td>\n",
       "      <td>0.23</td>\n",
       "      <td>0.32</td>\n",
       "      <td>8.5</td>\n",
       "      <td>0.058</td>\n",
       "      <td>47.0</td>\n",
       "      <td>186.0</td>\n",
       "      <td>0.9956</td>\n",
       "      <td>3.19</td>\n",
       "      <td>0.40</td>\n",
       "      <td>9.9</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>7.2</td>\n",
       "      <td>0.23</td>\n",
       "      <td>0.32</td>\n",
       "      <td>8.5</td>\n",
       "      <td>0.058</td>\n",
       "      <td>47.0</td>\n",
       "      <td>186.0</td>\n",
       "      <td>0.9956</td>\n",
       "      <td>3.19</td>\n",
       "      <td>0.40</td>\n",
       "      <td>9.9</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   fixed acidity  volatile acidity  citric acid  residual sugar  chlorides  \\\n",
       "0            7.0              0.27         0.36            20.7      0.045   \n",
       "1            6.3              0.30         0.34             1.6      0.049   \n",
       "2            8.1              0.28         0.40             6.9      0.050   \n",
       "3            7.2              0.23         0.32             8.5      0.058   \n",
       "4            7.2              0.23         0.32             8.5      0.058   \n",
       "\n",
       "   free sulfur dioxide  total sulfur dioxide  density    pH  sulphates  \\\n",
       "0                 45.0                 170.0   1.0010  3.00       0.45   \n",
       "1                 14.0                 132.0   0.9940  3.30       0.49   \n",
       "2                 30.0                  97.0   0.9951  3.26       0.44   \n",
       "3                 47.0                 186.0   0.9956  3.19       0.40   \n",
       "4                 47.0                 186.0   0.9956  3.19       0.40   \n",
       "\n",
       "   alcohol  quality  \n",
       "0      8.8        6  \n",
       "1      9.5        6  \n",
       "2     10.1        6  \n",
       "3      9.9        6  \n",
       "4      9.9        6  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = pd.read_csv('winequality-white.csv', sep=\";\")\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Nos données contiennent 12 colonnes, 11 qui correspondent à divers indicateurs physico-chimiques et 1 qui est la qualité du vin.\n",
    "\n",
    "Nous allons extraire deux arrays numpy de ces données, un qui contient les points et l’autre qui contient les étiquettes :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = data.as_matrix(data.columns[:-1])\n",
    "y_m = data.as_matrix([data.columns[-1]])\n",
    "y = y_m.flatten()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On peut maintenant afficher un histogramme pour chacune de nos variables :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x864 with 12 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(16, 12))\n",
    "for feat_idx in range(X.shape[1]):\n",
    "    ax = fig.add_subplot(3,4, (feat_idx+1))\n",
    "    h = ax.hist(X[:, feat_idx], bins=50, color='steelblue',\n",
    "                normed=True, edgecolor='none')\n",
    "    ax.set_title(data.columns[feat_idx], fontsize=14)\n",
    "\n",
    "ax = fig.add_subplot(3,4, 12)\n",
    "h = ax.hist(y, bins=50, color='steelblue',\n",
    "            normed=True, edgecolor='none')\n",
    "ax.set_title('Quality', fontsize=14)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On remarque en particulier que ces variables prennent des valeurs dans des ensembles différents. Par exemple, “sulphates” varie de 0 à 1 tandis que “total sulfur dioxide” varie de 0 à 440. Il va donc nous falloir standardiser les données pour que la deuxième ne domine pas complètement la première."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Nous allons commencer par transformer ce problème en un problème de classification : il s’agira de séparer les bons vins des vins médiocres :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_class = np.where(y<6, 0, 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Séparons nos données en un jeu d’entraînement et un jeu de test. Le jeu de test contiendra 30% des données.\n",
    "On paramètre 'random_state' avec la valeur du seed pour que la séparation des données soit fixe et reproductible."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = \\\n",
    "    model_selection.train_test_split(X, y_class,\n",
    "                                    test_size=0.3, # 30% des données dans le jeu de test\n",
    "                                     random_state = seed\n",
    "                                    )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Nous pouvons maintenant standardiser les données d’entraînement et appliquer la même transformation aux données de test :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "std_scale = preprocessing.StandardScaler().fit(X_train)\n",
    "X_train_std = std_scale.transform(X_train)\n",
    "X_test_std = std_scale.transform(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On peut visualiser de nouveau les données pour vérifier que les différentes variables prennent des valeurs qui ont maintenant des ordres de grandeur similaires."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x864 with 11 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(16, 12))\n",
    "for feat_idx in range(X_train_std.shape[1]):\n",
    "    ax = fig.add_subplot(3,4, (feat_idx+1))\n",
    "    h = ax.hist(X_train_std[:, feat_idx], bins=50, color='steelblue',\n",
    "        normed=True, edgecolor='none')\n",
    "    ax.set_title(data.columns[feat_idx], fontsize=14)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Sélection de modèles avec GridSearchCV"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On crée les 5 k-fold en utilisant la seed afin que ces folds soient reproductibles :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "TRAIN: taille 2742 , 5ers indices [1 2 3 4 5] || TEST: taille 686 , 5ers indices [ 0 14 17 25 26]\n",
      "TRAIN: taille 2742 , 5ers indices [0 1 2 3 4] || TEST: taille 686 , 5ers indices [ 7 12 22 29 41]\n",
      "TRAIN: taille 2742 , 5ers indices [0 1 3 4 5] || TEST: taille 686 , 5ers indices [ 2  6  8 13 18]\n",
      "TRAIN: taille 2743 , 5ers indices [0 1 2 4 6] || TEST: taille 685 , 5ers indices [ 3  5  9 10 15]\n",
      "TRAIN: taille 2743 , 5ers indices [0 2 3 5 6] || TEST: taille 685 , 5ers indices [ 1  4 11 16 19]\n"
     ]
    }
   ],
   "source": [
    "kf = KFold(n_splits=5, shuffle=True, random_state=seed)\n",
    "\n",
    "for train_index, test_index in kf.split(X_train_std):\n",
    "    print(\"TRAIN:\", \"taille\", len(train_index), \", 5ers indices\", train_index[0:5], \n",
    "          \"|| TEST:\", \"taille\", len(test_index), \", 5ers indices\", test_index[0:5])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Nous allons maintenant utiliser la méthode \"GridSearchCV\" pour faire une validation croisée du paramètre k d’un kNN (le nombre de plus proches voisins) sur le jeu d’entraînement :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Meilleur(s) hyperparamètre(s) sur le jeu d'entraînement:\n",
      "{'n_neighbors': 5}\n",
      "Résultats de la validation croisée :\n",
      "\taccuracy = 0.762 (+/-0.013) for {'n_neighbors': 3}\n",
      "\taccuracy = 0.766 (+/-0.015) for {'n_neighbors': 5}\n",
      "\taccuracy = 0.764 (+/-0.008) for {'n_neighbors': 7}\n",
      "\taccuracy = 0.757 (+/-0.019) for {'n_neighbors': 9}\n",
      "\taccuracy = 0.759 (+/-0.010) for {'n_neighbors': 11}\n",
      "\taccuracy = 0.761 (+/-0.011) for {'n_neighbors': 13}\n",
      "\taccuracy = 0.762 (+/-0.014) for {'n_neighbors': 15}\n"
     ]
    }
   ],
   "source": [
    "# Fixer les valeurs des hyperparamètres à tester\n",
    "param_grid = {'n_neighbors':[3, 5, 7, 9, 11, 13, 15]}\n",
    "\n",
    "# Choisir un score à optimiser, ici l'accuracy (proportion de prédictions correctes)\n",
    "score = 'accuracy'\n",
    "\n",
    "# Créer un classifieur kNN avec recherche d'hyperparamètre par validation croisée\n",
    "clf = model_selection.GridSearchCV(neighbors.KNeighborsClassifier(), # un classifieur kNN\n",
    "                                    param_grid, # hyperparamètres à tester\n",
    "                                    cv=kf, # folds pour la validation croisée\n",
    "                                    scoring=score # score à optimiser\n",
    "                                    )\n",
    "\n",
    "# Optimiser ce classifieur sur le jeu d'entraînement\n",
    "clf.fit(X_train_std, y_train)\n",
    "\n",
    "# Afficher le(s) hyperparamètre(s) optimaux\n",
    "print(\"Meilleur(s) hyperparamètre(s) sur le jeu d'entraînement:\")\n",
    "print(clf.best_params_)\n",
    "\n",
    "# Afficher les performances correspondantes\n",
    "print(\"Résultats de la validation croisée :\")\n",
    "for mean, std, params in zip(clf.cv_results_['mean_test_score'], # score moyen\n",
    "                                clf.cv_results_['std_test_score'], # écart-type du score\n",
    "                                clf.cv_results_['params'] # valeur de l'hyperparamètre\n",
    "                                ):\n",
    "    print(\"\\t%s = %0.3f (+/-%0.03f) for %r\" % (score, # critère utilisé\n",
    "                                                mean, # score moyen\n",
    "                                                std * 2, # barre d'erreur\n",
    "                                                params # hyperparamètre\n",
    "                                                ))"
   ]
  },
  {
   "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>params</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>{'n_neighbors': 3}</td>\n",
       "      <td>0.761669</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>{'n_neighbors': 5}</td>\n",
       "      <td>0.765753</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>{'n_neighbors': 7}</td>\n",
       "      <td>0.763711</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>{'n_neighbors': 9}</td>\n",
       "      <td>0.756709</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>{'n_neighbors': 11}</td>\n",
       "      <td>0.758751</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>{'n_neighbors': 13}</td>\n",
       "      <td>0.761377</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>{'n_neighbors': 15}</td>\n",
       "      <td>0.762252</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                params  mean_test_score  rank_test_score\n",
       "0   {'n_neighbors': 3}         0.761669                4\n",
       "1   {'n_neighbors': 5}         0.765753                1\n",
       "2   {'n_neighbors': 7}         0.763711                2\n",
       "3   {'n_neighbors': 9}         0.756709                7\n",
       "4  {'n_neighbors': 11}         0.758751                6\n",
       "5  {'n_neighbors': 13}         0.761377                5\n",
       "6  {'n_neighbors': 15}         0.762252                3"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.DataFrame(clf.cv_results_)[['params', 'mean_test_score', 'rank_test_score']]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "La meilleure performance (~0.766) est ici atteinte avec 5 voisins.\n",
    "\n",
    "Nous pouvons maintenant regarder la performance sur le jeu de test. GridSearchCV a automatiquement ré-entraîné le meilleur modèle sur l’intégralité du jeu d’entraînement."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Accuracy score sur le jeu de test : 0.770\n"
     ]
    }
   ],
   "source": [
    "y_pred = clf.predict(X_test_std)\n",
    "print(\"\\nAccuracy score sur le jeu de test : %0.3f\" % metrics.accuracy_score(y_test, y_pred))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Sélection des modèles manuellement"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On commence par créer la fonction qui va effectuer la validation croisée :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def cvf(X, y, n_neighbors, n_fold, verbose=False):\n",
    "    \"\"\"\n",
    "    X : données \n",
    "    y : valeur à prédire \n",
    "    n_neighbors : liste des K à tester pour le K-NN\n",
    "    n_fold : nombre de folds à créer\n",
    "    verbose : False par défaut, permet d'afficher dees informations du code\n",
    "    \"\"\"\n",
    "    \n",
    "    # on recrée les folds. Etant donné que pour l'exercice on va utiliser n_fold = 5 comme dans la partie 1,\n",
    "    # et qu'on utilise la même seed, on aura les mêmes folds que dans la partie 1.\n",
    "    kf = KFold(n_splits=n_fold, shuffle=True, random_state=seed)\n",
    "    if verbose: \n",
    "        for train_index, test_index in kf.split(X):\n",
    "            print(\"TRAIN:\", \"taille\", len(train_index), \", 5ers indices\", train_index[0:5], \n",
    "                  \"|| TEST:\", \"taille\", len(test_index), \", 5ers indices\", test_index[0:5])\n",
    "    \n",
    "    # on initialise le tableau qui contiendra les différents résultats\n",
    "    accuracy_mean = [] \n",
    "    \n",
    "    # on boucle pour tester les différentes valeurs de K pour le KNN\n",
    "    for kn in n_neighbors:\n",
    "        if verbose: print(\"kn\", kn)    \n",
    "        knn = neighbors.KNeighborsClassifier(kn) # initialisation du classifieur\n",
    "        \n",
    "        accuracy = [] # initialisation d'une liste qui contiendra l'accuracy pour chaque k-fold et chaque knn\n",
    "        \n",
    "        # on boucle sur les différents folds\n",
    "        for train_index, test_index in kf.split(X):\n",
    "            if verbose: print(\"TRAIN:\", \"taille\", len(train_index), \", 5ers indices\", train_index[0:5], \n",
    "                      \"|| TEST:\", \"taille\", len(test_index), \", 5ers indices\", test_index[0:5])\n",
    "            X_train, X_test = X[train_index], X[test_index]\n",
    "            y_train, y_test = y[train_index], y[test_index]\n",
    "                        \n",
    "            # on fit le knn sur le fold d'entrainement\n",
    "            knn.fit(X_train, y_train)\n",
    "            \n",
    "            # on test l'accuracy sur le fold de test et on l'ajoute au tableau 'accuracy':\n",
    "            accuracy.append([knn.score(X_test, y_test)])\n",
    "        if verbose: print(np.mean(accuracy))\n",
    "        \n",
    "        # on aggrège les résultats en ne gardant que la moyenne des accuracy des différents folds pour chaque k\n",
    "        accuracy_mean.append([kn, round(np.mean(accuracy), 3)])\n",
    "    \n",
    "    return accuracy_mean            "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On fait tourner notre fonction avec les mêmes valeurs que celles utilisées pour le GridSearchCV :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_neighbors = [3, 5, 7, 9, 11, 13, 15]\n",
    "n_fold = 5\n",
    "\n",
    "cvf = cvf(X_train_std, y_train, n_neighbors, n_fold, verbose=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On obtient le tableau des accuracy moyennes pour chaque K :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "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>knn</th>\n",
       "      <th>mean_accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>3</td>\n",
       "      <td>0.762</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>5</td>\n",
       "      <td>0.766</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>7</td>\n",
       "      <td>0.764</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>9</td>\n",
       "      <td>0.757</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>11</td>\n",
       "      <td>0.759</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>13</td>\n",
       "      <td>0.761</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>15</td>\n",
       "      <td>0.762</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   knn  mean_accuracy\n",
       "0    3          0.762\n",
       "1    5          0.766\n",
       "2    7          0.764\n",
       "3    9          0.757\n",
       "4   11          0.759\n",
       "5   13          0.761\n",
       "6   15          0.762"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cvf_df = pd.DataFrame(cvf, columns=['knn', 'mean_accuracy'])\n",
    "cvf_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "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>knn</th>\n",
       "      <th>mean_accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>5</td>\n",
       "      <td>0.766</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   knn  mean_accuracy\n",
       "1    5          0.766"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cvf_df[cvf_df['mean_accuracy'] == cvf_df['mean_accuracy'].max()]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Le K à sélectionner est donc 5, avec une accuracy moyenne de 0.766, ce qui correspond à ce que renvoyait GridSearchCV avec les mêmes données."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Nous pouvons maintenant regarder la performance sur le jeu de test. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Accuracy score sur le jeu de test : 0.770\n"
     ]
    }
   ],
   "source": [
    "knn5 = neighbors.KNeighborsClassifier(5)\n",
    "knn5.fit(X_train_std, y_train)\n",
    "print(\"\\nAccuracy score sur le jeu de test : %0.3f\" % knn5.score(X_test_std, y_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On retrouve la même accuracy qu'avec le GridSearchCV."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Evaluation de la classification knn"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Courbe ROC"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y_pred_proba = clf.predict_proba(X_test_std)[:, 1]\n",
    "[fpr, tpr, thr] = metrics.roc_curve(y_test, y_pred_proba)\n",
    "plt.plot(fpr, tpr, color='coral', lw=2)\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('1 - specificite', fontsize=14)\n",
    "plt.ylabel('Sensibilite', fontsize=14)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Calculons l'aire sous la courbee ROC :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUROC = 0.812\n"
     ]
    }
   ],
   "source": [
    "print(\"AUROC =\", round(metrics.auc(fpr, tpr), 3))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Prenons l’exemple du vinho verde. Imaginons que l’algorithme doit être capable de détecter efficacement les vins de mauvaise qualité, qui ne seront pas ensuite examinés par un expert humain. On veut alors limiter le nombre de faux négatifs, pour limiter le nombre de rejets infondés. Fixons-nous un taux de faux négatifs tolérable (la proportion de positifs incorrectement prédits négatifs) de 5%. Cela équivaut à une sensibilité de 0.95 :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sensibilité : 0.98\n",
      "Spécificité : 0.18\n",
      "Seuil : 0.20\n"
     ]
    }
   ],
   "source": [
    "idx = np.min(np.where(tpr > 0.95)) # indice du premier seuil pour lequel\n",
    "                                   # la sensibilité est supérieure à 0.95\n",
    "print(\"Sensibilité : %.2f\" % tpr[idx])\n",
    "print(\"Spécificité : %.2f\" % (1-fpr[idx]))\n",
    "print(\"Seuil : %.2f\" % thr[idx])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Utiliser un seuil de 0.20 nous garantit une sensibilité de 0.98 et une spécificité de 0.18, soit un taux de faux positifs de… 82%."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evaluer en comparant avec des approches naïves"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = \\\n",
    "    model_selection.train_test_split(X, y,\n",
    "                                    test_size=0.3, # 30% des données dans le jeu de test\n",
    "                                     random_state = seed\n",
    "                                    )\n",
    "std_scale = preprocessing.StandardScaler().fit(X_train)\n",
    "X_train_std = std_scale.transform(X_train)\n",
    "X_test_std = std_scale.transform(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Entraînons un kNN avec k=11 sur ces données :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "KNeighborsRegressor(algorithm='auto', leaf_size=30, metric='minkowski',\n",
       "          metric_params=None, n_jobs=None, n_neighbors=11, p=2,\n",
       "          weights='uniform')"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "knn2 = neighbors.KNeighborsRegressor(n_neighbors=11)\n",
    "\n",
    "knn2.fit(X_train_std, y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Et appliquons le pour prédire les étiquettes de notre jeu de test :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = knn2.predict(X_test_std)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Calculons la RMSE correspondante :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSE : 0.71\n"
     ]
    }
   ],
   "source": [
    "print(\"RMSE : %.2f\" % np.sqrt(metrics.mean_squared_error(y_test, y_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Valeurs prédites VS vraies valeurs :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.scatter(y_test, y_pred, color='coral')\n",
    "plt.axis([2, 9, 2, 9])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Comme nos étiquettes prennent des valeurs entières entre 3 et 8, nous avons beaucoup de points superposés aux même coordonnées. Pour mieux visualiser les données, nous pouvons utiliser comme marqueurs des cercles dont la taille est proportionnelle au nombre de points qui sont présents à ces coordonnées."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sizes = {} # clé : coordonnées ; valeur : nombre de points à ces coordonnées\n",
    "for (yt, yp) in zip(list(y_test), list(y_pred)):\n",
    "    if (yt, yp) in sizes:\n",
    "        sizes[(yt, yp)] += 1\n",
    "    else:\n",
    "        sizes[(yt, yp)] = 1\n",
    "\n",
    "keys = sizes.keys()\n",
    "plt.scatter([k[0] for k in keys], # vraie valeur (abscisse)\n",
    "            [k[1] for k in keys], # valeur predite (ordonnee)\n",
    "            s=[sizes[k] for k in keys], # taille du marqueur\n",
    "            color='coral')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On note ainsi une accumulation de prédictions correctes sur la diagonale. Néanmoins le modèle n'est pas très précis dans ses prédictions.\n",
    "\n",
    "Pour mieux comprendre notre modèle, comparons-le à une première approche naïve, qui consiste à prédire des valeurs aléatoires, distribuées uniformément entre les valeurs basse et haute des étiquettes du jeu de données d'entraînement."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred_random = np.random.randint(np.min(y), np.max(y), y_test.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Calculons la RMSE correspondante :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSE : 1.89\n"
     ]
    }
   ],
   "source": [
    "print(\"RMSE : %.2f\" % np.sqrt(metrics.mean_squared_error(y_test, y_pred_random)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On obtient une RMSE bien supérieure à la RMSE obtenue par notre modèle kNN. Notre modèle a ainsi réussi à bien mieux apprendre qu'un modèle aléatoire.\n",
    "\n",
    "Cependant, beaucoup de nos vins ont une note de 6, et beaucoup de nos prédictions sont autour de cette valeur. Comparons maintenant notre modèle à un modèle aléatoire qui retourne systématiquement la valeur moyenne des étiquettes du jeu de données d'entraînement.\n",
    "\n",
    "Nous pouvons utiliser pour cela la fonction correspondante du module \"dummy\" de scikit-learn."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSE : 0.87\n"
     ]
    }
   ],
   "source": [
    "dum = dummy.DummyRegressor(strategy='mean')\n",
    "\n",
    "# Entraînement\n",
    "dum.fit(X_train_std, y_train)\n",
    "\n",
    "# Prédiction sur le jeu de test\n",
    "y_pred_dum = dum.predict(X_test_std)\n",
    "\n",
    "# Evaluate\n",
    "print(\"RMSE : %.2f\" % np.sqrt(metrics.mean_squared_error(y_test, y_pred_dum)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On obtient une RMSE légèrement supérieure à celle obtenue par le kNN. Le kNN a donc appris plus que la moyenne des étiquettes."
   ]
  }
 ],
 "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": "3.7.3"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}