stacking.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "from tqdm import tnrange, tqdm_notebook\n",
    "import gc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sns.set_context('talk')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings('ignore', message='Changing the shape of non-C contiguous array')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Read the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "dfXtrain = pd.read_csv('preprocessed_csv/train_more.csv', index_col='id')\n",
    "dfXtest = pd.read_csv('preprocessed_csv/test_more.csv', index_col='id')\n",
    "dfYtrain = pd.read_csv('preprocessed_csv/y_train_more.csv', header=None, names=['INDEX', 'P_TARGET_FLAG'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x_train = np.array(dfXtrain)\n",
    "x_test = np.array(dfXtest)\n",
    "\n",
    "y_train = np.array(dfYtrain['P_TARGET_FLAG'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Save routines"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>INDEX</th>\n",
       "      <th>P_TARGET_FLAG</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>3</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>9</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>10</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>18</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>21</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   INDEX  P_TARGET_FLAG\n",
       "0      3            0.0\n",
       "1      9            0.0\n",
       "2     10            0.0\n",
       "3     18            0.0\n",
       "4     21            0.0"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dfYtest = pd.DataFrame({'INDEX': dfXtest.index, 'P_TARGET_FLAG': np.zeros(x_test.shape[0])})\n",
    "dfYtest.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def save_to_file(y, file_name):\n",
    "    dfYtest['P_TARGET_FLAG'] = y\n",
    "    dfYtest.to_csv('results/{}'.format(file_name), index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Mix models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import roc_auc_score\n",
    "from sklearn.ensemble import VotingClassifier as VC\n",
    "from sklearn.ensemble import RandomForestClassifier as RFC\n",
    "import xgboost as xgb\n",
    "from sklearn.grid_search import GridSearchCV\n",
    "\n",
    "XGBC = xgb.XGBClassifier"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### XGB"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "class myXGBC(XGBC):\n",
    "    def fit(self, *args, **kwargs):\n",
    "        return super(myXGBC, self).fit(*args, eval_metric='auc', **kwargs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "xgb_kwargs = {'objective':'binary:logistic', 'missing': -1, 'seed': 56,\n",
    "          'n_estimators': 160, 'reg_alpha': 6, 'reg_lambda': 1, 'max_depth': 3, 'colsample_bytree': 1,\n",
    "          'gamma': 0.008}\n",
    "\n",
    "xgb_best = myXGBC(**xgb_kwargs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 35 best of big set of features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "class myRF(RFC):\n",
    "    out_feature = np.array([  0.,   0.,   0.,   2.,   0.,   4.,   4.,   6.,   8.,   0.,   6.,\n",
    "         7.,  12.,   6.,   1.,  13.,  15.,   7.,   8.,  16.,  15.,  19.,\n",
    "         2.,  18.,  12.,  17.,  14.,  10.,   1.,  24.,  18.,  25.,  22.,\n",
    "        15.,  34.,   9.,  30.,  13.,  32.,   5.,  34.,  17.,  19.,  34.,\n",
    "        35.,   7.,  33.,  14.,  46.]) \n",
    "    \n",
    "    last_out = 35\n",
    "\n",
    "    def drop_cols(self, X):\n",
    "        for features_count in range(self.out_feature.shape[0] - 1, self.last_out - 1, -1):\n",
    "            out = self.out_feature[features_count]\n",
    "            X = np.delete(X, out, 1)\n",
    "        return X        \n",
    "    \n",
    "    def fit(self, X, y, *args, **kwargs):\n",
    "        X = self.drop_cols(X)\n",
    "        return super(myRF, self).fit(X, y, *args, **kwargs)\n",
    "    \n",
    "    def predict(self, X, *args, **kwargs):\n",
    "        X = self.drop_cols(X)\n",
    "        return super(myRF, self).predict(X, *args, **kwargs)\n",
    "\n",
    "    def predict_proba(self, X, *args, **kwargs):\n",
    "        X = self.drop_cols(X)\n",
    "        return super(myRF, self).predict_proba(X, *args, **kwargs)\n",
    "    \n",
    "    def predict_log_proba(self, X, *args, **kwargs):\n",
    "        X = self.drop_cols(X)\n",
    "        return super(myRF, self).predict_log_proba(X, *args, **kwargs)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "kwargs_more_35 = {'n_estimators': 300, 'criterion': 'entropy', 'n_jobs': -1}\n",
    "\n",
    "more_35 = myRF(**kwargs_more_35)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Mixing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def plot_quality_mix(grid_searcher, param_name):\n",
    "    means = []\n",
    "    stds = []\n",
    "    for elem in grid_searcher.grid_scores_:\n",
    "        means.append(np.mean(elem.cv_validation_scores))\n",
    "        stds.append(np.sqrt(np.var(elem.cv_validation_scores)))\n",
    "    means = np.array(means)\n",
    "    stds = np.array(stds)\n",
    "    \n",
    "    params = grid_searcher.param_grid\n",
    "    \n",
    "    plt.figure(figsize=(8, 6))\n",
    "    x_axis = np.array(params[param_name])\n",
    "    x_axis = x_axis[:, 0]\n",
    "    plt.plot(x_axis, means)\n",
    "    plt.fill_between(x_axis, \\\n",
    "                 means + stds, means - stds, alpha = 0.3, facecolor='blue')\n",
    "    plt.xlabel(param_name)\n",
    "    plt.ylabel('AUC')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "mix_clf = VC(estimators=[('more_35', more_35), ('xgb_best', xgb_best)], voting='soft')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.824984849508\n",
      "VotingClassifier(estimators=[('more_35', myRF(bootstrap=True, class_weight=None, criterion='entropy', max_depth=None,\n",
      "   max_features='auto', max_leaf_nodes=None, min_samples_leaf=1,\n",
      "   min_samples_split=2, min_weight_fraction_leaf=0.0, n_estimators=300,\n",
      "   n_jobs=-1, oob_score=False, random_state=None, verbose=0,\n",
      " ...gistic',\n",
      "    reg_alpha=6, reg_lambda=1, scale_pos_weight=1, seed=56, silent=True,\n",
      "    subsample=1))],\n",
      "         voting='soft', weights=[0.30000000000000004, 0.69999999999999996])\n",
      "CPU times: user 14.2 s, sys: 240 ms, total: 14.5 s\n",
      "Wall time: 3min 45s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "params = {'weights': [[alpha, 1 - alpha] for alpha in np.arange(0, 1.01, 0.1)]}\n",
    "\n",
    "grid_searcher = GridSearchCV(mix_clf,\\\n",
    "                         params, cv=5, scoring='roc_auc', n_jobs=-1)\n",
    "grid_searcher.fit(x_train, y_train)\n",
    "\n",
    "print(grid_searcher.best_score_)\n",
    "print(grid_searcher.best_estimator_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAhIAAAGLCAYAAACSvHupAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9a3f4d55c0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_quality_mix(grid_searcher, 'weights')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "# Apply to results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "dfYmore = pd.read_csv('results/more_35_best_100_000_cycle.csv', index_col='INDEX')\n",
    "dfYxgb = pd.read_csv('results/xbgoost_grid.csv', index_col='INDEX')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>P_TARGET_FLAG</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>INDEX</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.20204</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>0.40058</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.10860</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>0.19002</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>0.21472</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       P_TARGET_FLAG\n",
       "INDEX               \n",
       "3            0.20204\n",
       "9            0.40058\n",
       "10           0.10860\n",
       "18           0.19002\n",
       "21           0.21472"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dfYmore.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>P_TARGET_FLAG</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>INDEX</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.164293</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>0.267080</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.065473</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>0.171523</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>0.152631</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       P_TARGET_FLAG\n",
       "INDEX               \n",
       "3           0.164293\n",
       "9           0.267080\n",
       "10          0.065473\n",
       "18          0.171523\n",
       "21          0.152631"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dfYxgb.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>P_TARGET_FLAG</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>INDEX</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.175617</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>0.307130</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.078411</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>0.177072</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>0.171258</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       P_TARGET_FLAG\n",
       "INDEX               \n",
       "3           0.175617\n",
       "9           0.307130\n",
       "10          0.078411\n",
       "18          0.177072\n",
       "21          0.171258"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dfYbest = (0.3 * dfYmore + 0.7 *dfYxgb)\n",
    "dfYbest.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dfYbest.to_csv('results/{}'.format('mix.csv'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Среднее геометрическое"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>P_TARGET_FLAG</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>INDEX</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.171231</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>0.289635</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.072446</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>0.175072</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>0.163414</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       P_TARGET_FLAG\n",
       "INDEX               \n",
       "3           0.171231\n",
       "9           0.289635\n",
       "10          0.072446\n",
       "18          0.175072\n",
       "21          0.163414"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dfYbest = (np.power(dfYmore, 0.2) * np.power(dfYxgb, 0.8))\n",
    "dfYbest.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dfYbest.to_csv('results/{}'.format('geom_mix2.csv'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.4.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}