main.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "import os\n",
    "%matplotlib inline\n",
    "\n",
    "from pandas.io.json import json_normalize\n",
    "import json\n",
    "import random\n",
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\")\n",
    "from sklearn.metrics import classification_report"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def load_df(csv_path='data/train.csv', nrows=None):\n",
    "    JSON_COLUMNS = ['device', 'geoNetwork', 'totals', 'trafficSource']\n",
    "    \n",
    "    p = 0.1\n",
    "    df = pd.read_csv(csv_path, \n",
    "                     converters={column: json.loads for column in JSON_COLUMNS}, \n",
    "                     dtype={'fullVisitorId': 'str'}, # Important!!\n",
    "                     skiprows=lambda i: i>0 and random.random() > p)\n",
    "    \n",
    "    for column in JSON_COLUMNS:\n",
    "        column_as_df = json_normalize(df[column])\n",
    "        column_as_df.columns = [\"{}.{}\".format(column, subcolumn) for subcolumn in column_as_df.columns]\n",
    "        df = df.drop(column, axis=1).merge(column_as_df, right_index=True, left_index=True)\n",
    "    print(\"Loaded {}. Shape: {}\".format(os.path.basename(csv_path), df.shape))\n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded train.csv. Shape: (90666, 54)\n",
      "Loaded test.csv. Shape: (79785, 53)\n"
     ]
    }
   ],
   "source": [
    "train_data = load_df()\n",
    "test_data = load_df('data/test.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 90666 entries, 0 to 90665\n",
      "Data columns (total 54 columns):\n",
      "channelGrouping                                      90666 non-null object\n",
      "date                                                 90666 non-null int64\n",
      "fullVisitorId                                        90666 non-null object\n",
      "sessionId                                            90666 non-null object\n",
      "socialEngagementType                                 90666 non-null object\n",
      "visitId                                              90666 non-null int64\n",
      "visitNumber                                          90666 non-null int64\n",
      "visitStartTime                                       90666 non-null int64\n",
      "device.browser                                       90666 non-null object\n",
      "device.browserSize                                   90666 non-null object\n",
      "device.browserVersion                                90666 non-null object\n",
      "device.deviceCategory                                90666 non-null object\n",
      "device.flashVersion                                  90666 non-null object\n",
      "device.isMobile                                      90666 non-null bool\n",
      "device.language                                      90666 non-null object\n",
      "device.mobileDeviceBranding                          90666 non-null object\n",
      "device.mobileDeviceInfo                              90666 non-null object\n",
      "device.mobileDeviceMarketingName                     90666 non-null object\n",
      "device.mobileDeviceModel                             90666 non-null object\n",
      "device.mobileInputSelector                           90666 non-null object\n",
      "device.operatingSystem                               90666 non-null object\n",
      "device.operatingSystemVersion                        90666 non-null object\n",
      "device.screenColors                                  90666 non-null object\n",
      "device.screenResolution                              90666 non-null object\n",
      "geoNetwork.city                                      90666 non-null object\n",
      "geoNetwork.cityId                                    90666 non-null object\n",
      "geoNetwork.continent                                 90666 non-null object\n",
      "geoNetwork.country                                   90666 non-null object\n",
      "geoNetwork.latitude                                  90666 non-null object\n",
      "geoNetwork.longitude                                 90666 non-null object\n",
      "geoNetwork.metro                                     90666 non-null object\n",
      "geoNetwork.networkDomain                             90666 non-null object\n",
      "geoNetwork.networkLocation                           90666 non-null object\n",
      "geoNetwork.region                                    90666 non-null object\n",
      "geoNetwork.subContinent                              90666 non-null object\n",
      "totals.bounces                                       45307 non-null object\n",
      "totals.hits                                          90666 non-null object\n",
      "totals.newVisits                                     70245 non-null object\n",
      "totals.pageviews                                     90659 non-null object\n",
      "totals.transactionRevenue                            1130 non-null object\n",
      "totals.visits                                        90666 non-null object\n",
      "trafficSource.adContent                              1089 non-null object\n",
      "trafficSource.adwordsClickInfo.adNetworkType         2170 non-null object\n",
      "trafficSource.adwordsClickInfo.criteriaParameters    90666 non-null object\n",
      "trafficSource.adwordsClickInfo.gclId                 2180 non-null object\n",
      "trafficSource.adwordsClickInfo.isVideoAd             2170 non-null object\n",
      "trafficSource.adwordsClickInfo.page                  2170 non-null object\n",
      "trafficSource.adwordsClickInfo.slot                  2170 non-null object\n",
      "trafficSource.campaign                               90666 non-null object\n",
      "trafficSource.isTrueDirect                           27661 non-null object\n",
      "trafficSource.keyword                                39882 non-null object\n",
      "trafficSource.medium                                 90666 non-null object\n",
      "trafficSource.referralPath                           33444 non-null object\n",
      "trafficSource.source                                 90666 non-null object\n",
      "dtypes: bool(1), int64(4), object(49)\n",
      "memory usage: 36.7+ MB\n"
     ]
    }
   ],
   "source": [
    "train_data.shape\n",
    "train_data.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "shops_or_not = lambda x : x.total"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "85976"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_data['fullVisitorId'].nunique()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data[\"totals.transactionRevenue\"] = train_data[\"totals.transactionRevenue\"].astype('float')\n",
    "y_clf = (train_data['totals.transactionRevenue'].fillna(0) > 0).astype(np.uint8)\n",
    "y_reg = train_data['totals.transactionRevenue'].fillna(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    89536\n",
       "1     1130\n",
       "Name: totals.transactionRevenue, dtype: int64"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_clf.describe()\n",
    "y_clf.value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    9.066600e+04\n",
       "mean     1.517437e+06\n",
       "std      3.081595e+07\n",
       "min      0.000000e+00\n",
       "25%      0.000000e+00\n",
       "50%      0.000000e+00\n",
       "75%      0.000000e+00\n",
       "max      4.280700e+09\n",
       "Name: totals.transactionRevenue, dtype: float64"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_reg.describe()\n",
    "# y_reg.value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def date_format(df):\n",
    "    df['date'] = pd.to_datetime(df['date'])\n",
    "    df['vis_date'] = pd.to_datetime(df['visitStartTime'])\n",
    "    df['sess_date_dow'] = df['vis_date'].dt.dayofweek\n",
    "    df['sess_date_hours'] = df['vis_date'].dt.hour\n",
    "    df['sess_date_dom'] = df['vis_date'].dt.day\n",
    "date_format(train_data)\n",
    "date_format(test_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 79785 entries, 0 to 79784\n",
      "Data columns (total 57 columns):\n",
      "channelGrouping                                      79785 non-null object\n",
      "date                                                 79785 non-null datetime64[ns]\n",
      "fullVisitorId                                        79785 non-null object\n",
      "sessionId                                            79785 non-null object\n",
      "socialEngagementType                                 79785 non-null object\n",
      "visitId                                              79785 non-null int64\n",
      "visitNumber                                          79785 non-null int64\n",
      "visitStartTime                                       79785 non-null int64\n",
      "device.browser                                       79785 non-null object\n",
      "device.browserSize                                   79785 non-null object\n",
      "device.browserVersion                                79785 non-null object\n",
      "device.deviceCategory                                79785 non-null object\n",
      "device.flashVersion                                  79785 non-null object\n",
      "device.isMobile                                      79785 non-null bool\n",
      "device.language                                      79785 non-null object\n",
      "device.mobileDeviceBranding                          79785 non-null object\n",
      "device.mobileDeviceInfo                              79785 non-null object\n",
      "device.mobileDeviceMarketingName                     79785 non-null object\n",
      "device.mobileDeviceModel                             79785 non-null object\n",
      "device.mobileInputSelector                           79785 non-null object\n",
      "device.operatingSystem                               79785 non-null object\n",
      "device.operatingSystemVersion                        79785 non-null object\n",
      "device.screenColors                                  79785 non-null object\n",
      "device.screenResolution                              79785 non-null object\n",
      "geoNetwork.city                                      79785 non-null object\n",
      "geoNetwork.cityId                                    79785 non-null object\n",
      "geoNetwork.continent                                 79785 non-null object\n",
      "geoNetwork.country                                   79785 non-null object\n",
      "geoNetwork.latitude                                  79785 non-null object\n",
      "geoNetwork.longitude                                 79785 non-null object\n",
      "geoNetwork.metro                                     79785 non-null object\n",
      "geoNetwork.networkDomain                             79785 non-null object\n",
      "geoNetwork.networkLocation                           79785 non-null object\n",
      "geoNetwork.region                                    79785 non-null object\n",
      "geoNetwork.subContinent                              79785 non-null object\n",
      "totals.bounces                                       41638 non-null object\n",
      "totals.hits                                          79785 non-null object\n",
      "totals.newVisits                                     59776 non-null object\n",
      "totals.pageviews                                     79767 non-null object\n",
      "totals.visits                                        79785 non-null object\n",
      "trafficSource.adContent                              5406 non-null object\n",
      "trafficSource.adwordsClickInfo.adNetworkType         5361 non-null object\n",
      "trafficSource.adwordsClickInfo.criteriaParameters    79785 non-null object\n",
      "trafficSource.adwordsClickInfo.gclId                 5366 non-null object\n",
      "trafficSource.adwordsClickInfo.isVideoAd             5361 non-null object\n",
      "trafficSource.adwordsClickInfo.page                  5361 non-null object\n",
      "trafficSource.adwordsClickInfo.slot                  5361 non-null object\n",
      "trafficSource.campaign                               79785 non-null object\n",
      "trafficSource.isTrueDirect                           25898 non-null object\n",
      "trafficSource.keyword                                40964 non-null object\n",
      "trafficSource.medium                                 79785 non-null object\n",
      "trafficSource.referralPath                           23545 non-null object\n",
      "trafficSource.source                                 79785 non-null object\n",
      "vis_date                                             79785 non-null datetime64[ns]\n",
      "sess_date_dow                                        79785 non-null int64\n",
      "sess_date_hours                                      79785 non-null int64\n",
      "sess_date_dom                                        79785 non-null int64\n",
      "dtypes: bool(1), datetime64[ns](2), int64(6), object(48)\n",
      "memory usage: 34.2+ MB\n"
     ]
    }
   ],
   "source": [
    "test_data.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "excluded_features = [\n",
    "    'date', 'fullVisitorId', 'sessionId', 'totals.transactionRevenue', \n",
    "    'visitId', 'visitStartTime', 'non_zero_proba', 'vis_date'\n",
    "]\n",
    "\n",
    "categorical_features = [\n",
    "    _f for _f in train_data.columns\n",
    "    if (_f not in excluded_features) & (train_data[_f].dtype == 'object')\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "if 'totals.transactionRevenue' in train_data.columns:\n",
    "    del train_data['totals.transactionRevenue']\n",
    "\n",
    "if 'totals.transactionRevenue' in test_data.columns:\n",
    "    del test_data['totals.transactionRevenue']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "for f in categorical_features:\n",
    "    train_data[f], indexer = pd.factorize(train_data[f])\n",
    "    test_data[f] = indexer.get_indexer(test_data[f])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['channelGrouping', 'socialEngagementType', 'visitNumber', 'device.browser', 'device.browserSize', 'device.browserVersion', 'device.deviceCategory', 'device.flashVersion', 'device.isMobile', 'device.language', 'device.mobileDeviceBranding', 'device.mobileDeviceInfo', 'device.mobileDeviceMarketingName', 'device.mobileDeviceModel', 'device.mobileInputSelector', 'device.operatingSystem', 'device.operatingSystemVersion', 'device.screenColors', 'device.screenResolution', 'geoNetwork.city', 'geoNetwork.cityId', 'geoNetwork.continent', 'geoNetwork.country', 'geoNetwork.latitude', 'geoNetwork.longitude', 'geoNetwork.metro', 'geoNetwork.networkDomain', 'geoNetwork.networkLocation', 'geoNetwork.region', 'geoNetwork.subContinent', 'totals.bounces', 'totals.hits', 'totals.newVisits', 'totals.pageviews', 'totals.visits', 'trafficSource.adContent', 'trafficSource.adwordsClickInfo.adNetworkType', 'trafficSource.adwordsClickInfo.criteriaParameters', 'trafficSource.adwordsClickInfo.gclId', 'trafficSource.adwordsClickInfo.isVideoAd', 'trafficSource.adwordsClickInfo.page', 'trafficSource.adwordsClickInfo.slot', 'trafficSource.campaign', 'trafficSource.isTrueDirect', 'trafficSource.keyword', 'trafficSource.medium', 'trafficSource.referralPath', 'trafficSource.source', 'sess_date_dow', 'sess_date_hours', 'sess_date_dom']\n",
      "Training until validation scores don't improve for 50 rounds.\n",
      "[50]\tvalid_0's binary_logloss: 0.031491\n",
      "[100]\tvalid_0's binary_logloss: 0.0291708\n",
      "[150]\tvalid_0's binary_logloss: 0.0286912\n",
      "[200]\tvalid_0's binary_logloss: 0.0287335\n",
      "Early stopping, best iteration is:\n",
      "[179]\tvalid_0's binary_logloss: 0.0286297\n",
      "0.9857049336730034\n",
      "Training until validation scores don't improve for 50 rounds.\n",
      "[50]\tvalid_0's binary_logloss: 0.0303582\n",
      "[100]\tvalid_0's binary_logloss: 0.0280319\n",
      "[150]\tvalid_0's binary_logloss: 0.0276671\n",
      "[200]\tvalid_0's binary_logloss: 0.0276809\n",
      "Early stopping, best iteration is:\n",
      "[166]\tvalid_0's binary_logloss: 0.0276219\n",
      "0.985390286745132\n",
      "Training until validation scores don't improve for 50 rounds.\n",
      "[50]\tvalid_0's binary_logloss: 0.0324733\n",
      "[100]\tvalid_0's binary_logloss: 0.0300764\n",
      "[150]\tvalid_0's binary_logloss: 0.0297426\n",
      "Early stopping, best iteration is:\n",
      "[143]\tvalid_0's binary_logloss: 0.0297215\n",
      "0.9859039200775055\n",
      "Training until validation scores don't improve for 50 rounds.\n",
      "[50]\tvalid_0's binary_logloss: 0.0329209\n",
      "[100]\tvalid_0's binary_logloss: 0.0304559\n",
      "[150]\tvalid_0's binary_logloss: 0.0302659\n",
      "Early stopping, best iteration is:\n",
      "[140]\tvalid_0's binary_logloss: 0.0302353\n",
      "0.9801748605105005\n",
      "Training until validation scores don't improve for 50 rounds.\n",
      "[50]\tvalid_0's binary_logloss: 0.0324362\n",
      "[100]\tvalid_0's binary_logloss: 0.030627\n",
      "[150]\tvalid_0's binary_logloss: 0.0305719\n",
      "Early stopping, best iteration is:\n",
      "[110]\tvalid_0's binary_logloss: 0.0305235\n",
      "0.9828964515449473\n"
     ]
    }
   ],
   "source": [
    "import lightgbm as lgb\n",
    "from sklearn.model_selection import KFold, GroupKFold\n",
    "from sklearn.metrics import mean_squared_error, roc_auc_score, log_loss, roc_curve, precision_recall_curve\n",
    "\n",
    "folds = GroupKFold(n_splits=5)\n",
    "\n",
    "train_features = [_f for _f in train_data.columns if _f not in excluded_features]\n",
    "print(train_features)\n",
    "oof_clf_preds = np.zeros(train_data.shape[0])\n",
    "sub_clf_preds = np.zeros(test_data.shape[0])\n",
    "for fold_, (trn_, val_) in enumerate(folds.split(y_clf, y_clf, groups=train_data['fullVisitorId'])):\n",
    "    trn_x, trn_y = train_data[train_features].iloc[trn_], y_clf.iloc[trn_]\n",
    "    val_x, val_y = train_data[train_features].iloc[val_], y_clf.iloc[val_]\n",
    "    \n",
    "    clf = lgb.LGBMClassifier(\n",
    "        num_leaves=31,\n",
    "        learning_rate=0.03,\n",
    "        n_estimators=1000,\n",
    "        subsample=.9,\n",
    "        colsample_bytree=.9,\n",
    "        random_state=1\n",
    "    )\n",
    "    clf.fit(\n",
    "        trn_x, trn_y,\n",
    "        eval_set=[(val_x, val_y)],\n",
    "        early_stopping_rounds=50,\n",
    "        verbose=50\n",
    "    )\n",
    "    \n",
    "    oof_clf_preds[val_] = clf.predict_proba(val_x, num_iteration=clf.best_iteration_)[:, 1]\n",
    "    print(roc_auc_score(val_y, oof_clf_preds[val_]))\n",
    "    sub_clf_preds += clf.predict_proba(test_data[train_features], num_iteration=clf.best_iteration_)[:, 1] / folds.n_splits\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x576 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(15,8))\n",
    "ax1 = fig.add_subplot(1,2,1)\n",
    "ax1.set_xlim([-0.05,1.05])\n",
    "ax1.set_ylim([-0.05,1.05])\n",
    "ax1.set_xlabel('Recall')\n",
    "ax1.set_ylabel('Precision')\n",
    "ax1.set_title('PR Curve')\n",
    "\n",
    "ax2 = fig.add_subplot(1,2,2)\n",
    "ax2.set_xlim([-0.05,1.05])\n",
    "ax2.set_ylim([-0.05,1.05])\n",
    "ax2.set_xlabel('False Positive Rate')\n",
    "ax2.set_ylabel('True Positive Rate')\n",
    "ax2.set_title('ROC Curve')\n",
    "\n",
    "\n",
    "fpr, tpr, _ = roc_curve(y_clf,  oof_clf_preds)    \n",
    "auc = roc_auc_score(y_clf, oof_clf_preds)\n",
    "p,r,_ = precision_recall_curve(y_clf,oof_clf_preds)\n",
    "\n",
    "ax1.plot(r,p)\n",
    "ax2.plot(fpr, tpr)\n",
    "\n",
    "plt.plot(fpr,tpr,label=\"data 1, auc=\"+str(auc))\n",
    "plt.legend(loc=4)\n",
    "plt.show()\n",
    "\n",
    "plt.plot(p,r)\n",
    "plt.legend(loc=4)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "oof_clf_preds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "sum(oof_clf_preds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(classification_report(y_clf, oof_clf_preds))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from scipy import stats"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(stats.describe(oof_clf_preds))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "oof_clf_preds.min()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "oof_clf_preds.max()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import precision_recall_curve"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "precision, recall, thresholds = precision_recall_curve(y_clf, oof_clf_preds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data['non_zero_proba'] = oof_clf_preds\n",
    "test_data['non_zero_proba'] = sub_clf_preds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_features = [_f for _f in train_data.columns if _f not in excluded_features] + ['non_zero_proba']\n",
    "print(train_features)\n",
    "\n",
    "oof_reg_preds = np.zeros(train_data.shape[0])\n",
    "sub_reg_preds = np.zeros(test_data.shape[0])\n",
    "importances = pd.DataFrame()\n",
    "\n",
    "for fold_, (trn_, val_) in enumerate(folds.split(y_reg, y_reg, groups=train_data['fullVisitorId'])):\n",
    "    trn_x, trn_y = train_data[train_features].iloc[trn_], y_reg.iloc[trn_].fillna(0)\n",
    "    val_x, val_y = train_data[train_features].iloc[val_], y_reg.iloc[val_].fillna(0)\n",
    "    \n",
    "    reg = lgb.LGBMRegressor(\n",
    "        num_leaves=31,\n",
    "        learning_rate=0.03,\n",
    "        n_estimators=1000,\n",
    "        subsample=.9,\n",
    "        colsample_bytree=.9,\n",
    "        random_state=1\n",
    "    )\n",
    "    reg.fit(\n",
    "        trn_x, np.log1p(trn_y),\n",
    "        eval_set=[(val_x, np.log1p(val_y))],\n",
    "        early_stopping_rounds=50,\n",
    "        verbose=50\n",
    "    )\n",
    "    imp_df = pd.DataFrame()\n",
    "    imp_df['feature'] = train_features\n",
    "    imp_df['gain'] = reg.booster_.feature_importance(importance_type='gain')\n",
    "    \n",
    "    imp_df['fold'] = fold_ + 1\n",
    "    importances = pd.concat([importances, imp_df], axis=0, sort=False)\n",
    "    \n",
    "    oof_reg_preds[val_] = reg.predict(val_x, num_iteration=reg.best_iteration_)\n",
    "    oof_reg_preds[oof_reg_preds < 0] = 0\n",
    "    _preds = reg.predict(test_data[train_features], num_iteration=reg.best_iteration_)\n",
    "    _preds[_preds < 0] = 0\n",
    "    sub_reg_preds += np.expm1(_preds) / folds.n_splits\n",
    "    \n",
    "mean_squared_error(np.log1p(y_reg.fillna(0)), oof_reg_preds) ** .5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "imp_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "importances"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "importances['gain_log'] = np.log1p(importances['gain'])\n",
    "mean_gain = importances[['gain', 'feature']].groupby('feature').mean()\n",
    "importances['mean_gain'] = importances['feature'].map(mean_gain['gain'])\n",
    "\n",
    "plt.figure(figsize=(8, 12))\n",
    "sns.barplot(x='gain_log', y='feature', data=importances.sort_values('mean_gain', ascending=False))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}