xgb_classifier.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[MLENS] backend: threading\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "# -*-coding:utf-8-*-\n",
    "import os\n",
    "import glob\n",
    "import json\n",
    "import time\n",
    "import matplotlib\n",
    "# matplotlib.use('agg')\n",
    "%matplotlib inline\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import xgboost as xgb\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from mlens.ensemble import SuperLearner\n",
    "from sklearn.ensemble import AdaBoostClassifier, GradientBoostingClassifier, GradientBoostingClassifier\n",
    "from sklearn.ensemble import ExtraTreesClassifier, VotingClassifier, RandomForestClassifier\n",
    "from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor\n",
    "from sklearn.neural_network import MLPClassifier\n",
    "from sklearn.svm import SVC\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.model_selection import GridSearchCV, cross_val_score, StratifiedKFold, learning_curve, KFold\n",
    "from sklearn.externals import joblib\n",
    "\n",
    "from scipy.stats import pearsonr, spearmanr, kendalltau\n",
    "# from kappa_calculator import quadratic_weighted_kappa as qwk\n",
    "# from kappa_calculator import liaesr_weighted_kappa as lwk\n",
    "\n",
    "from PIL import Image\n",
    "from keras.preprocessing import image\n",
    "from keras.models import load_model\n",
    "from keras.applications.inception_v3 import preprocess_input\n",
    "\n",
    "sns.set(style='white', context='notebook', palette='deep')\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>correct</th>\n",
       "      <th>prob_avg</th>\n",
       "      <th>text_size</th>\n",
       "      <th>ref_size</th>\n",
       "      <th>text_word_size</th>\n",
       "      <th>ref_word_size</th>\n",
       "      <th>flag_short</th>\n",
       "      <th>flag_digit</th>\n",
       "      <th>prob_0</th>\n",
       "      <th>prob_1</th>\n",
       "      <th>...</th>\n",
       "      <th>nb_strip</th>\n",
       "      <th>flag_equal</th>\n",
       "      <th>feature_0</th>\n",
       "      <th>feature_1</th>\n",
       "      <th>feature_2</th>\n",
       "      <th>feature_3</th>\n",
       "      <th>feature_4</th>\n",
       "      <th>feature_5</th>\n",
       "      <th>feature_6</th>\n",
       "      <th>feature_7</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>0.993067</td>\n",
       "      <td>6</td>\n",
       "      <td>7</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>0.991831</td>\n",
       "      <td>15</td>\n",
       "      <td>15</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1</td>\n",
       "      <td>0.994514</td>\n",
       "      <td>5</td>\n",
       "      <td>5</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1</td>\n",
       "      <td>0.393408</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1</td>\n",
       "      <td>0.995023</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 28 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   correct  prob_avg  text_size  ref_size  text_word_size  ref_word_size  \\\n",
       "0        0  0.993067          6         7               1              1   \n",
       "1        1  0.991831         15        15               1              1   \n",
       "2        1  0.994514          5         5               1              1   \n",
       "3        1  0.393408          1         1               1              1   \n",
       "4        1  0.995023          4         4               1              1   \n",
       "\n",
       "   flag_short  flag_digit  prob_0  prob_1    ...      nb_strip  flag_equal  \\\n",
       "0           0           0       0       0    ...             0           0   \n",
       "1           0           0       0       0    ...             0           1   \n",
       "2           0           0       0       0    ...             0           1   \n",
       "3           1           0       1       0    ...             0           0   \n",
       "4           0           0       0       0    ...             0           1   \n",
       "\n",
       "   feature_0  feature_1  feature_2  feature_3  feature_4  feature_5  \\\n",
       "0          0          0          1          0          0          0   \n",
       "1          0          1          1          1          1          1   \n",
       "2          0          1          1          1          1          1   \n",
       "3          0          0          1          0          0          0   \n",
       "4          0          1          1          1          1          1   \n",
       "\n",
       "   feature_6  feature_7  \n",
       "0          0          1  \n",
       "1          1          1  \n",
       "2          1          1  \n",
       "3          0          0  \n",
       "4          1          1  \n",
       "\n",
       "[5 rows x 28 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "FLAG_gs = True\n",
    "\n",
    "dataset = pd.read_csv('./dataset/csv_data/blank_train.csv')\n",
    "valid_data = pd.read_csv('./dataset/csv_data/blank_valid.csv')\n",
    "\n",
    "# dataset = pd.read_csv('./dataset/archived_49_exams/csv_data/blank_train.csv')\n",
    "# valid_data = pd.read_csv('./dataset/archived_49_exams/csv_data/blank_valid.csv')\n",
    "\n",
    "experiment_name = 'first experiment'\n",
    "dataset.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "DataFrame shape: (786387, 28)\n"
     ]
    }
   ],
   "source": [
    "dataset.columns.tolist()\n",
    "print 'DataFrame shape:', dataset.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Split train and test dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Features count: 27\n"
     ]
    }
   ],
   "source": [
    "# dataset = dataset.drop(['other_avg'],axis=1)\n",
    "# valid_data = valid_data.drop(['other_avg'],axis=1)\n",
    "\n",
    "x_train = dataset.iloc[:,1:]\n",
    "y_train = dataset['correct']\n",
    "\n",
    "x_valid = valid_data.iloc[:,1:]\n",
    "y_valid = valid_data['correct']\n",
    "\n",
    "print 'Features count:', len(x_train.columns.tolist())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. xgboost classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "xgb_model = xgb.XGBClassifier().fit(x_train, y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 直接预测结果"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9134371328148446"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_pred = xgb_model.predict(x_valid)\n",
    "correct_count = 0\n",
    "for t1, t2 in zip(y_valid, y_pred):\n",
    "    if t1 == t2:\n",
    "        correct_count += 1\n",
    "1.0 * correct_count / len(y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "####  grid search 调参"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 3 folds for each of 64 candidates, totalling 192 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=11)]: Done  28 tasks      | elapsed:  7.1min\n",
      "[Parallel(n_jobs=11)]: Done 192 out of 192 | elapsed: 56.7min finished\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.922906914789\n",
      "{'colsample_bytree': 0.3, 'silent': 1, 'learning_rate': 0.1, 'min_child_weight': 6, 'n_estimators': 300, 'subsample': 0.7, 'objective': 'binary:logistic', 'max_depth': 8}\n"
     ]
    }
   ],
   "source": [
    "if FLAG_gs == True:\n",
    "    xgb1 = xgb.XGBClassifier()\n",
    "    parameters = {\n",
    "        'objective':['binary:logistic'],\n",
    "        'learning_rate': [0.08, 0.1], #so called `eta` value\n",
    "        'max_depth': [6, 8],\n",
    "        'min_child_weight': [6, 8],\n",
    "        'silent': [1],\n",
    "        'subsample': [0.7, 0.5],\n",
    "        'colsample_bytree': [0.2, 0.3],\n",
    "        'n_estimators': [200, 300]\n",
    "    }\n",
    "    xgb_grid = GridSearchCV(xgb1,\n",
    "                            parameters,\n",
    "                            n_jobs = 11,\n",
    "                            verbose=True)\n",
    "    xgb_grid.fit(x_train, y_train)\n",
    "    xgb_best = xgb_grid.best_estimator_\n",
    "\n",
    "    print xgb_grid.best_score_\n",
    "    print xgb_grid.best_params_\n",
    "    \n",
    "else:\n",
    "#     xgb_best = xgb_model\n",
    "    xgb_best = joblib.load('./models/xgb_clf.model')\n",
    "    \n",
    "weights = xgb_best.feature_importances_\n",
    "\n",
    "# 0.9038040937\n",
    "# {'colsample_bytree': 0.3, \n",
    "#  'silent': 1, \n",
    "#  'learning_rate': 0.1, \n",
    "#  'min_child_weight': 4, \n",
    "#  'n_estimators': 300, \n",
    "#  'subsample': 0.7, \n",
    "#  'objective': 'binary:logistic', \n",
    "#  'max_depth': 8}\n",
    "\n",
    "# Fitting 3 folds for each of 972 candidates, totalling 2916 fits\n",
    "# [Parallel(n_jobs=-1)]: Done  26 tasks      | elapsed:   30.7s\n",
    "# [Parallel(n_jobs=-1)]: Done 176 tasks      | elapsed:  3.7min\n",
    "# [Parallel(n_jobs=-1)]: Done 426 tasks      | elapsed:  9.6min\n",
    "# [Parallel(n_jobs=-1)]: Done 776 tasks      | elapsed: 18.1min\n",
    "# [Parallel(n_jobs=-1)]: Done 1226 tasks      | elapsed: 27.5min\n",
    "# [Parallel(n_jobs=-1)]: Done 1776 tasks      | elapsed: 37.3min\n",
    "# [Parallel(n_jobs=-1)]: Done 2426 tasks      | elapsed: 45.5min\n",
    "# [Parallel(n_jobs=-1)]: Done 2916 out of 2916 | elapsed: 50.4min finished"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9227760342222923"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_pred = xgb_best.predict(x_valid)\n",
    "correct_count = 0\n",
    "for t1, t2 in zip(1.0 *y_valid, y_pred):\n",
    "    if t1 == t2:\n",
    "        correct_count += 1\n",
    "1.0 * correct_count / len(y_pred)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feature importances of the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.text.Text at 0x7f056dfc73d0>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAx8AAALJCAYAAADVtJ4jAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n2ClWZywAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f05e4c46050>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "indices = np.argsort(xgb_best.feature_importances_)[::-1]\n",
    "fig, ax = plt.subplots(figsize=(12,12))\n",
    "g = sns.barplot(y=x_train.columns[indices], x=xgb_best.feature_importances_[indices], orient='h', ax=ax)\n",
    "\n",
    "g.set_xlabel('Importance')\n",
    "g.set_ylabel('features')\n",
    "g.set_title(\"xgb_best feature importance\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Save xgb_best"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['models/xgb_clf.model']"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joblib.dump(xgb_best, 'models/xgb_clf.model')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  评分函数"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def calc_correl(test_pred, test_origin):\n",
    "    test_prs, _ = pearsonr(test_pred, test_origin)\n",
    "    test_spr, _ = spearmanr(test_pred, test_origin)\n",
    "    test_tau, _ = kendalltau(test_pred, test_origin)\n",
    "    return test_spr, test_tau\n",
    "\n",
    "\n",
    "def calc_qwk(test_pred, test_origin, self_low, self_high):\n",
    "    test_pred_int = np.rint(test_pred).astype('int32')\n",
    "    test_origin_int = np.rint(test_origin).astype('int32')\n",
    "    test_qwk = qwk(test_origin_int, test_pred_int, self_low, self_high)\n",
    "    test_lwk = lwk(test_origin_int, test_pred_int, self_low, self_high)\n",
    "    return test_qwk,test_lwk"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## 2. 模型融合"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2.1 Initialize model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "ensemble = SuperLearner()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2.2 add 7 different estimators to the first layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "SuperLearner(array_check=2, backend=None, folds=2,\n",
       "       layers=[Layer(backend='threading', dtype=<type 'numpy.float32'>, n_jobs=-1,\n",
       "   name='layer-1', propagate_features=None, raise_on_exception=True,\n",
       "   random_state=None, shuffle=False,\n",
       "   stack=[Group(backend='threading', dtype=<type 'numpy.float32'>,\n",
       "   indexer=FoldIndex(X=None, folds=2, raise_on_exce...rer=None)],\n",
       "   n_jobs=-1, name='group-0', raise_on_exception=True, transformers=[])],\n",
       "   verbose=0)],\n",
       "       model_selection=False, n_jobs=None, raise_on_exception=True,\n",
       "       random_state=None, sample_size=20, scorer=None, shuffle=False,\n",
       "       verbose=False)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ensemble.add([\n",
    "    SVC(),\n",
    "    AdaBoostClassifier(),\n",
    "    GradientBoostingClassifier(),\n",
    "    RandomForestClassifier(),\n",
    "    ExtraTreesClassifier(),\n",
    "    LogisticRegression(),\n",
    "    xgb.XGBClassifier()\n",
    "])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2.3 add xgboost classifier as meta learner"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "SuperLearner(array_check=2, backend=None, folds=2,\n",
       "       layers=[Layer(backend='threading', dtype=<type 'numpy.float32'>, n_jobs=-1,\n",
       "   name='layer-1', propagate_features=None, raise_on_exception=True,\n",
       "   random_state=None, shuffle=False,\n",
       "   stack=[Group(backend='threading', dtype=<type 'numpy.float32'>,\n",
       "   indexer=FoldIndex(X=None, folds=2, raise_on_exce...rer=None)],\n",
       "   n_jobs=-1, name='group-1', raise_on_exception=True, transformers=[])],\n",
       "   verbose=0)],\n",
       "       model_selection=False, n_jobs=None, raise_on_exception=True,\n",
       "       random_state=None, sample_size=20, scorer=None, shuffle=False,\n",
       "       verbose=False)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ensemble.add_meta(xgb.XGBClassifier())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2.4 train super learner with data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "SuperLearner(array_check=2, backend=None, folds=2,\n",
       "       layers=[Layer(backend='threading', dtype=<type 'numpy.float32'>, n_jobs=-1,\n",
       "   name='layer-1', propagate_features=None, raise_on_exception=True,\n",
       "   random_state=None, shuffle=False,\n",
       "   stack=[Group(backend='threading', dtype=<type 'numpy.float32'>,\n",
       "   indexer=FoldIndex(X=None, folds=2, raise_on_exce...rer=None)],\n",
       "   n_jobs=-1, name='group-1', raise_on_exception=True, transformers=[])],\n",
       "   verbose=0)],\n",
       "       model_selection=False, n_jobs=None, raise_on_exception=True,\n",
       "       random_state=None, sample_size=20, scorer=None, shuffle=False,\n",
       "       verbose=False)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ensemble.fit(x_train, y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2.5 check the result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.919378220420454"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_pred = ensemble.predict(x_valid)\n",
    "correct_count = 0\n",
    "for t1, t2 in zip(y_valid, y_pred):\n",
    "    if t1 == t2:\n",
    "        correct_count += 1\n",
    "1.0 * correct_count / len(y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2.6 save superlearner"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "this superlearner is 144 MB large and has a similar/worse accuracy compared to xgb_best(2.2MB)\n",
    "\n",
    "so I choose to use xgb_best as the final estimator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['models/meta_learner.model']"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joblib.dump(ensemble, 'models/meta_learner.model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python (workenv)",
   "language": "python",
   "name": "workenv"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.15"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}