bagging_boosting_stacking.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Apprentissage d'ensemble\n",
    "Un ensemble est un modèle d'apprentissage machine qui combine les prédictions de deux ou plusieurs modèles.\n",
    "\n",
    "Les modèles qui contribuent à l'ensemble, appelés membres de l'ensemble, peuvent être du même type ou de types différents et peuvent ou non être formés sur les mêmes données de formation.\n",
    "\n",
    "Les prédictions faites par les membres de l'ensemble peuvent être combinées à l'aide de statistiques, telles que le mode ou la moyenne, ou par des méthodes plus sophistiquées qui permettent de savoir dans quelle mesure il faut faire confiance à chaque membre et dans quelles conditions.\n",
    "\n",
    "L'étude des méthodes d'ensemble a réellement pris son essor dans les années 1990, et cette décennie a vu la publication d'articles sur les méthodes les plus populaires et les plus utilisées, telles que les méthodes de mise en sac des noyaux, de renforcement et d'empilage.\n",
    "\n",
    "À la fin des années 2000, l'adoption des ensembles s'est accélérée en partie grâce à leur énorme succès dans les concours d'apprentissage automatique, comme le prix Netflix et, plus tard, les concours Kaggle.\n",
    "\n",
    "Au cours des deux dernières décennies, les systèmes de classification multiples, également appelés systèmes d'ensemble, ont bénéficié d'une attention croissante au sein de la communauté de l'intelligence informatique et de l'apprentissage machine.\n",
    "\n",
    "- Page 1, Apprentissage machine d'ensemble, 2012.\n",
    "\n",
    "Les méthodes d'ensemble augmentent considérablement le coût et la complexité des calculs. Cette augmentation provient de l'expertise et du temps nécessaires à la formation et à la maintenance de plusieurs modèles plutôt que d'un seul modèle. Cela oblige à se poser la question :\n",
    "\n",
    "Pourquoi devrions-nous envisager d'utiliser un ensemble ?\n",
    "Il y a deux raisons principales pour utiliser un ensemble plutôt qu'un modèle unique, et elles sont liées ; elles le sont :\n",
    "\n",
    "La performance : Un ensemble peut faire de meilleures prévisions et atteindre de meilleures performances qu'un modèle unique.\n",
    "Robustesse : Un ensemble réduit la dispersion des prévisions et des performances du modèle.\n",
    "Les ensembles sont utilisés pour obtenir de meilleures performances prédictives sur un problème de modélisation prédictive qu'un modèle prédictif unique. La manière dont cela est réalisé peut être comprise comme le modèle réduisant la composante de variance de l'erreur de prédiction en ajoutant un biais (c'est-à-dire dans le contexte du compromis biais-variance).\n",
    "\n",
    "Développé à l'origine pour réduire la variance - et donc améliorer la précision - d'un système de prise de décision automatisé ...\n",
    "\n",
    "- Page 1, Ensemble Machine Learning, 2012.\n",
    "\n",
    "Un autre avantage important et moins discuté des méthodes d'ensemble est l'amélioration de la robustesse ou de la fiabilité des performances moyennes d'un modèle.\n",
    "\n",
    "Ce sont deux préoccupations importantes dans un projet d'apprentissage machine et parfois nous pouvons préférer une ou les deux propriétés d'un modèle.\n",
    "\n",
    "Examinons de plus près ces deux propriétés afin de mieux comprendre les avantages de l'utilisation de l'apprentissage d'ensemble dans un projet. \n",
    "\n",
    "# Utiliser des ensembles pour améliorer la robustesse\n",
    "\n",
    "Dans le cadre d'un projet de modélisation prédictive, nous évaluons souvent plusieurs modèles ou filières de modélisation et choisissons celui qui fonctionne bien ou le mieux comme modèle final.\n",
    "\n",
    "L'algorithme ou le pipeline est alors adapté à toutes les données disponibles et utilisé pour faire des prévisions sur de nouvelles données.\n",
    "\n",
    "Nous avons une idée de la performance moyenne du modèle à partir de notre harnais de test, généralement estimée en utilisant une validation croisée répétée de k fois comme étalon-or. Le problème est que la performance moyenne pourrait ne pas être suffisante.\n",
    "\n",
    "La précision ou l'erreur moyenne d'un modèle est un résumé des performances attendues, alors qu'en fait, certains modèles ont mieux fonctionné et d'autres moins bien sur différents sous-ensembles de données.\n",
    "\n",
    "L'écart-type est la différence moyenne entre une observation et la moyenne et résume la dispersion ou la dispersion des données. Pour une mesure de la précision ou de l'erreur d'un modèle, il peut vous donner une idée de la dispersion du comportement du modèle.\n",
    "\n",
    "L'examen des notes minimales et maximales de performance du modèle vous donnera une idée de la pire et de la meilleure performance que vous pouvez attendre du modèle, ce qui pourrait ne pas être acceptable pour votre application.\n",
    "\n",
    "L'ensemble le plus simple consiste à ajuster le modèle plusieurs fois sur les ensembles de données de formation et à combiner les prédictions en utilisant une statistique sommaire, comme la moyenne de régression ou le mode de classification. Il est important de noter que chaque modèle doit être légèrement différent en raison de l'algorithme d'apprentissage stochastique, de la différence de composition de l'ensemble de données de formation ou des différences dans le modèle lui-même.\n",
    "\n",
    "Cela permettra de réduire la dispersion des prédictions faites par le modèle. La performance moyenne sera probablement à peu près la même, bien que la performance la plus mauvaise et la plus bonne soit plus proche de la performance moyenne.\n",
    "\n",
    "En effet, cela permet de lisser la performance attendue du modèle.\n",
    "\n",
    "On peut parler de \"robustesse\" de la performance attendue du modèle et c'est un avantage minimal de l'utilisation d'une méthode d'ensemble.\n",
    "\n",
    "Un ensemble peut ou non améliorer la performance de la modélisation par rapport à un seul membre contributeur, ce qui est discuté plus en détail, mais au minimum, il devrait réduire l'écart dans la performance moyenne du modèle. \n",
    "\n",
    "# Biais, écarts et ensembles\n",
    "Les modèles d'apprentissage machine pour la classification et la régression apprennent une fonction de cartographie des entrées aux sorties.\n",
    "\n",
    "Cette cartographie est apprise à partir d'exemples du domaine du problème, l'ensemble de données de formation, et est évaluée sur des données non utilisées pendant la formation, l'ensemble de données de test.\n",
    "\n",
    "Les erreurs commises par un modèle d'apprentissage machine sont souvent décrites en termes de deux propriétés : le biais et la variance.\n",
    "\n",
    "Le biais est une mesure de la proximité du modèle par rapport à la fonction de correspondance entre les entrées et les sorties. Il rend compte de la rigidité du modèle : la force de l'hypothèse que le modèle a sur la forme fonctionnelle de la correspondance entre les entrées et les sorties.\n",
    "\n",
    "La variance du modèle est la quantité de changement de la performance du modèle lorsqu'il est ajusté sur différentes données de formation. Elle saisit l'impact des spécificités des données sur le modèle.\n",
    "\n",
    "La variance fait référence à la quantité de changement [du modèle] si nous l'estimons en utilisant un ensemble de données de formation différent.\n",
    "\n",
    "- Page 34, Une introduction à l'apprentissage statistique avec applications en R, 2014.\n",
    "\n",
    "Le biais et la variance de la performance d'un modèle sont liés.\n",
    "\n",
    "Idéalement, nous préférerions un modèle avec un faible biais et une faible variance, bien que dans la pratique, cela soit très difficile. En fait, cela pourrait être décrit comme l'objectif de l'apprentissage machine appliqué pour un problème de modélisation prédictive donné.\n",
    "\n",
    "La réduction du biais peut souvent être facilement obtenue en augmentant la variance. Inversement, la réduction de la variance peut facilement être obtenue en augmentant le biais.\n",
    "\n",
    "On parle alors de compromis car il est facile d'obtenir une méthode avec un biais extrêmement faible mais une variance élevée [...] ou une méthode avec une variance très faible mais un biais élevé ...\n",
    "\n",
    "- Page 36, Une introduction à l'apprentissage statistique avec des applications en R, 2014.\n",
    "\n",
    "Certains modèles présentent naturellement un biais ou une variance élevés, qui peuvent souvent être relâchés ou augmentés à l'aide d'hyperparamètres qui modifient le comportement d'apprentissage de l'algorithme.\n",
    "\n",
    "Les ensembles fournissent un moyen de réduire la variance des prédictions, c'est-à-dire la quantité d'erreur dans les prédictions faites qui peut être attribuée à la \"variance\".\n",
    "\n",
    "Ce n'est pas toujours le cas, mais lorsque c'est le cas, cette réduction de la variance entraîne à son tour une amélioration des performances prédictives.\n",
    "\n",
    "Les preuves empiriques et théoriques montrent que certaines techniques d'ensemble (comme l'ensachage) agissent comme un mécanisme de réduction de la variance, c'est-à-dire qu'elles réduisent la composante de variance de l'erreur. De plus, les résultats empiriques suggèrent que d'autres techniques d'ensemble (telles que AdaBoost) réduisent à la fois le biais et la partie variance de l'erreur.\n",
    "\n",
    "- Page 39, Classification des modèles à l'aide de méthodes d'ensemble, 2010.\n",
    "\n",
    "L'utilisation d'ensembles pour réduire les propriétés de variance des erreurs de prédiction conduit à l'avantage clé de l'utilisation d'ensembles en premier lieu : améliorer les performances de prédiction. \n",
    "\n",
    "# Utiliser des ensembles pour améliorer les performances\n",
    "La réduction de l'élément de variance de l'erreur de prédiction améliore la performance prédictive.\n",
    "\n",
    "Nous utilisons explicitement l'apprentissage d'ensemble pour rechercher une meilleure performance prédictive, comme une erreur plus faible sur la régression ou une précision élevée pour la classification.\n",
    "\n",
    "... il existe un moyen d'améliorer la précision des modèles qui est plus facile et plus puissant que la sélection judicieuse d'algorithmes : on peut rassembler les modèles en ensembles.\n",
    "\n",
    "- Page 2, Ensemble Methods in Data Mining, 2010.\n",
    "\n",
    "Il s'agit de l'utilisation principale des méthodes d'apprentissage d'ensemble et de l'avantage démontré par l'utilisation d'ensembles par la majorité des gagnants des concours d'apprentissage machine, tels que le prix Netflix et les concours sur Kaggle.\n",
    "\n",
    "Dans le cadre du prix Netflix, un concours s'est déroulé pendant deux ans dans lequel la première équipe à soumettre un modèle améliorant de 10 % le système de recommandation interne de Netflix remportait 1 000 000 de dollars. [...] l'avantage final a été obtenu en pesant les contributions des modèles de 30 concurrents au maximum.\n",
    "\n",
    "- Page 8, Ensemble Methods in Data Mining, 2010.\n",
    "\n",
    "Cet avantage a également été démontré dans le cadre de concours universitaires, comme les solutions de pointe pour le célèbre ensemble de données ImageNet en vision par ordinateur.\n",
    "\n",
    "Un ensemble de ces réseaux résiduels atteint un taux d'erreur de 3,57 % sur l'ensemble de tests d'ImageNet. Ce résultat a permis de remporter la première place dans la tâche de classification ILSVRC 2015.\n",
    "\n",
    "- Deep Residual Learning for Image Recognition, 2015.\n",
    "\n",
    "Lorsqu'il est utilisé de cette manière, un ensemble ne devrait être adopté que s'il est plus performant en moyenne que tout autre membre de l'ensemble. Si ce n'est pas le cas, le membre contributeur le plus performant doit être utilisé à la place.\n",
    "\n",
    "Considérez la distribution des scores attendus calculés par un modèle sur un harnais de test, comme la validation croisée répétée par k fois, comme nous l'avons fait ci-dessus en considérant la \"robustesse\" offerte par un ensemble. Un ensemble qui réduit la variance de l'erreur, en fait, modifiera la distribution plutôt que de simplement réduire la dispersion de la distribution.\n",
    "\n",
    "Il peut en résulter une meilleure performance moyenne par rapport à un modèle unique.\n",
    "\n",
    "Ce n'est pas toujours le cas, et avoir cette attente est une erreur courante des débutants.\n",
    "\n",
    "Il est possible, et même courant, que la performance d'un ensemble ne soit pas meilleure que celle du membre le plus performant de l'ensemble. Cela peut se produire si l'ensemble a un modèle performant et que les autres membres n'offrent aucun avantage ou si l'ensemble n'est pas en mesure d'exploiter efficacement leur contribution.\n",
    "\n",
    "Il est également possible qu'un ensemble se produise moins bien que le membre le plus performant de l'ensemble. Cela aussi est courant, car il s'agit généralement d'un modèle très performant dont les prévisions sont rendues plus mauvaises par un ou plusieurs autres modèles peu performants et l'ensemble n'est pas en mesure d'exploiter efficacement ses contributions.\n",
    "\n",
    "Il est donc important de tester une série de méthodes d'ensemble et de régler leur comportement, comme nous le faisons pour tout modèle individuel d'apprentissage machine. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:32.197826Z",
     "start_time": "2020-04-25T20:09:29.465479Z"
    }
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.datasets import make_moons\n",
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:33.508799Z",
     "start_time": "2020-04-25T20:09:33.303756Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x1a15ed72d0>"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "X,y=make_moons(n_samples=500, noise=0.3, random_state=0)\n",
    "plt.scatter(X[:,0],X[:,1],c=y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:34.523916Z",
     "start_time": "2020-04-25T20:09:34.518973Z"
    }
   },
   "outputs": [],
   "source": [
    "Xtrain,Xtest,ytrain,ytest =train_test_split(X,y,test_size=0.3,random_state=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# VOTING"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:36.272646Z",
     "start_time": "2020-04-25T20:09:36.102804Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.linear_model import SGDClassifier\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "\n",
    "from sklearn.ensemble import VotingClassifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:36.794986Z",
     "start_time": "2020-04-25T20:09:36.779823Z"
    }
   },
   "outputs": [],
   "source": [
    "model1=SGDClassifier(random_state=0)\n",
    "model2=DecisionTreeClassifier(random_state=0)\n",
    "model3=KNeighborsClassifier(n_neighbors=2)\n",
    "\n",
    "model4=VotingClassifier(estimators=[('SGD',model1),('Tree',model2),('KNN',model3)],voting='hard')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:37.069481Z",
     "start_time": "2020-04-25T20:09:37.029785Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "SGDClassifier 0.84\n",
      "DecisionTreeClassifier 0.86\n",
      "KNeighborsClassifier 0.86\n",
      "VotingClassifier 0.8733333333333333\n"
     ]
    }
   ],
   "source": [
    "for model in (model1,model2,model3,model4):\n",
    "    model.fit(Xtrain,ytrain)\n",
    "    print(model.__class__.__name__,model.score(Xtest,ytest))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# BAGGING\n",
    "avec modèle qui font de l'overfitting"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:38.981116Z",
     "start_time": "2020-04-25T20:09:38.973745Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.ensemble import BaggingClassifier, RandomForestClassifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:39.609300Z",
     "start_time": "2020-04-25T20:09:39.603648Z"
    }
   },
   "outputs": [],
   "source": [
    "model=BaggingClassifier(base_estimator=KNeighborsClassifier(),n_estimators=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:40.015790Z",
     "start_time": "2020-04-25T20:09:39.833551Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8866666666666667"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(Xtrain,ytrain)\n",
    "model.score(Xtest,ytest)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:40.557771Z",
     "start_time": "2020-04-25T20:09:40.400629Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.88"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "RandomForestClassifier(n_estimators=100,max_depth=10).fit(Xtrain,ytrain).score(Xtest,ytest)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# BOOSTING\n",
    "avec modèles qui font de l'underfitting"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:41.613910Z",
     "start_time": "2020-04-25T20:09:41.610270Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.ensemble import AdaBoostClassifier,GradientBoostingClassifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:41.829677Z",
     "start_time": "2020-04-25T20:09:41.824396Z"
    }
   },
   "outputs": [],
   "source": [
    "model=AdaBoostClassifier(n_estimators=100,learning_rate=1.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:09:42.491143Z",
     "start_time": "2020-04-25T20:09:42.346730Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8866666666666667"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(Xtrain,ytrain).score(Xtest,ytest)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# STACKING\n",
    "permet d'associer déjà des modèles puissants attention, très lente"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:11:01.094431Z",
     "start_time": "2020-04-25T20:11:01.091159Z"
    }
   },
   "outputs": [],
   "source": [
    "from sklearn.ensemble import StackingClassifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:11:05.430620Z",
     "start_time": "2020-04-25T20:11:05.426792Z"
    }
   },
   "outputs": [],
   "source": [
    "model4=StackingClassifier(estimators=[('SGD',model1),('Tree',model2),('KNN',model3)],\n",
    "                          final_estimator=KNeighborsClassifier())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:11:06.372415Z",
     "start_time": "2020-04-25T20:11:06.322433Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8666666666666667"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model4.fit(Xtrain,ytrain).score(Xtest,ytest)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:11:09.351007Z",
     "start_time": "2020-04-25T20:11:09.344703Z"
    }
   },
   "outputs": [],
   "source": [
    "SGD = SGDClassifier(random_state = 0)\n",
    "Tree = DecisionTreeClassifier(random_state = 0)\n",
    "KNN = KNeighborsClassifier()\n",
    "\n",
    "modele_1 = BaggingClassifier(base_estimator = KNN, n_estimators = 100)\n",
    "modele_2 = BaggingClassifier(base_estimator = Tree, n_estimators = 100)\n",
    "modele_3 = BaggingClassifier(base_estimator = SGD, n_estimators = 100)\n",
    "\n",
    "final_model = StackingClassifier([('Bag_KNN', modele_1),\n",
    "                                  ('Bag_Tree', modele_2),\n",
    "                                  ('Bag_SGD', modele_3)],\n",
    "                                  final_estimator = KNN\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:11:49.146147Z",
     "start_time": "2020-04-25T20:11:46.680177Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8866666666666667"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "final_model.fit(Xtrain,ytrain).score(Xtest,ytest)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:15:43.508191Z",
     "start_time": "2020-04-25T20:15:43.503449Z"
    }
   },
   "outputs": [],
   "source": [
    "ADA=AdaBoostClassifier(n_estimators=100,learning_rate=1.5)\n",
    "RDF=RandomForestClassifier(n_estimators=100,max_depth=10)\n",
    "\n",
    "stacking=StackingClassifier(estimators=[('Ada',ADA),('Rdf',RDF)],final_estimator=KNeighborsClassifier())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-04-25T20:16:17.622206Z",
     "start_time": "2020-04-25T20:16:16.196845Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8133333333333334"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "stacking.fit(Xtrain,ytrain).score(Xtest,ytest)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.6"
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}