ligthgbm_models/fraud-fs-fe9477-no-day-seed-133.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [],
   "source": [
    "from functools import partial\n",
    "\n",
    "import datetime\n",
    "import gc\n",
    "import lightgbm as lgb\n",
    "import numpy as np\n",
    "import os\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import pickle\n",
    "import random\n",
    "import seaborn as sns\n",
    "\n",
    "from sklearn.metrics import roc_auc_score\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.utils import shuffle\n",
    "\n",
    "sns.set()\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def seed_everything(seed=13):\n",
    "    random.seed(seed)\n",
    "    os.environ['PYTHONHASHSEED'] = str(seed)\n",
    "    np.random.seed(seed)\n",
    "    \n",
    "    \n",
    "def read_from_disk(path, filename):\n",
    "    with open(os.path.join(path, filename), 'rb') as handle:\n",
    "        return pickle.load(handle)\n",
    "    \n",
    "    \n",
    "def save_to_disk(obj, filename):\n",
    "    with open(filename, 'wb') as handle:\n",
    "        pickle.dump(obj, handle, protocol=pickle.HIGHEST_PROTOCOL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def split_features_by_importance(x_train, x_val, y_val, model, importance_tresh=0.1, permut_tresh=1e-4):\n",
    "    all_features = x_train.columns\n",
    "    feature_importance = calc_feature_importance(all_features, model)\n",
    "    garbage_features = find_garbage_features(feature_importance, importance_tresh)\n",
    "    retain_features = find_features_to_retain(garbage_features, x_val, y_val, model, threshold=permut_tresh)\n",
    "    remove_features = set(garbage_features) - set(retain_features)\n",
    "    return list(set(all_features) - remove_features), list(remove_features), feature_importance\n",
    "\n",
    "\n",
    "def calc_feature_importance(features, model):\n",
    "    importance = pd.DataFrame([\n",
    "        list(features),\n",
    "        model.feature_importance(importance_type='split').tolist(),\n",
    "        model.feature_importance(importance_type='gain').tolist()\n",
    "    ]).T\n",
    "    importance.columns = ['feature', 'split', 'gain']\n",
    "    return importance\n",
    "\n",
    "\n",
    "def find_garbage_features(importance, treshold):\n",
    "    min_gain = importance['gain'].quantile(treshold)\n",
    "    min_split = importance['split'].quantile(treshold)\n",
    "    garbage_features = importance[ \n",
    "        (importance['split'] < min_split) # | (importance['gain'] < min_gain) \n",
    "    ]['feature']\n",
    "    return list(garbage_features)\n",
    "\n",
    "\n",
    "def find_features_to_retain(features, x, y, model, threshold=1e-4):\n",
    "    scores = permutation_importance(x.copy(), y, model, columns=features)\n",
    "    retain_features = [feat for feat, score in scores.items() if score < -threshold]\n",
    "    return retain_features\n",
    "\n",
    "\n",
    "def permutation_importance(x, y, model, base_score=None, columns=None): \n",
    "    diff = {}\n",
    "    if not base_score:\n",
    "        base_score = roc_auc_score(y, model.predict(x))\n",
    "    if not columns:\n",
    "        columns = x.columns\n",
    "    for col in columns:\n",
    "        value = x[col].copy()\n",
    "        x[col] = np.random.permutation(x[col].values)\n",
    "        diff[col] = roc_auc_score(y, model.predict(x)) - base_score\n",
    "        x[col] = value\n",
    "    return diff\n",
    "\n",
    "\n",
    "def plot_feature_importance(model, max_features=50):\n",
    "    f, ax = plt.subplots(1,2, figsize=[20,10])\n",
    "    lgb.plot_importance(model, max_num_features=max_features, ax=ax[0], importance_type='gain')\n",
    "    ax[0].set_title(\"Light GBM GAIN Feature Importance\")\n",
    "\n",
    "    lgb.plot_importance(model, max_num_features=max_features, ax=ax[1], importance_type='split')\n",
    "    ax[1].set_title(\"Light GBM SPLIT Feature Importance\")\n",
    "    plt.tight_layout(pad=1)\n",
    "    \n",
    "    \n",
    "def avg_feature_importance(dfs):\n",
    "    temp = pd.concat(dfs).reset_index(drop=True)\n",
    "    for col in ['split', 'gain']:\n",
    "        temp[col] = temp[col].astype(float)\n",
    "    agg = temp.groupby('feature').mean()\n",
    "    return agg"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data preparation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "DFS_DIR = '../input/fraud-messing-v31'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train, y_train, df_val, y_val, df_hold, y_hold, df_test = map(partial(read_from_disk, DFS_DIR), [\n",
    "    'df_train.pkl', 'y_train.pkl', 'df_val.pkl', 'y_val.pkl', 'df_hold.pkl', 'y_hold.pkl', 'df_test.pkl'\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "_kg_hide-output": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['TransactionDT', 'TransactionAmt', 'ProductCD', 'card1', 'card2', 'card3', 'card4', 'card5', 'card6', 'addr1', 'addr2', 'dist1', 'dist2', 'P_emaildomain', 'R_emaildomain', 'C1', 'C2', 'C3', 'C4', 'C5', 'C6', 'C7', 'C8', 'C9', 'C10', 'C11', 'C12', 'C13', 'C14', 'D1', 'D2', 'D3', 'D4', 'D5', 'D6', 'D7', 'D8', 'D9', 'D10', 'D11', 'D12', 'D13', 'D14', 'D15', 'M1', 'M2', 'M3', 'M4', 'M5', 'M6', 'M7', 'M8', 'M9', 'V1', 'V2', 'V3', 'V4', 'V5', 'V6', 'V7', 'V8', 'V9', 'V10', 'V11', 'V12', 'V13', 'V14', 'V15', 'V16', 'V17', 'V18', 'V19', 'V20', 'V21', 'V22', 'V23', 'V24', 'V25', 'V26', 'V27', 'V28', 'V29', 'V30', 'V31', 'V32', 'V33', 'V34', 'V35', 'V36', 'V37', 'V38', 'V39', 'V40', 'V41', 'V42', 'V43', 'V44', 'V45', 'V46', 'V47', 'V48', 'V49', 'V50', 'V51', 'V52', 'V53', 'V54', 'V55', 'V56', 'V57', 'V58', 'V59', 'V60', 'V61', 'V62', 'V63', 'V64', 'V65', 'V66', 'V67', 'V68', 'V69', 'V70', 'V71', 'V72', 'V73', 'V74', 'V75', 'V76', 'V77', 'V78', 'V79', 'V80', 'V81', 'V82', 'V83', 'V84', 'V85', 'V86', 'V87', 'V88', 'V89', 'V90', 'V91', 'V92', 'V93', 'V94', 'V95', 'V96', 'V97', 'V98', 'V99', 'V100', 'V101', 'V102', 'V103', 'V104', 'V105', 'V106', 'V107', 'V108', 'V109', 'V110', 'V111', 'V112', 'V113', 'V114', 'V115', 'V116', 'V117', 'V118', 'V119', 'V120', 'V121', 'V122', 'V123', 'V124', 'V125', 'V126', 'V127', 'V128', 'V129', 'V130', 'V131', 'V132', 'V133', 'V134', 'V135', 'V136', 'V137', 'V138', 'V139', 'V140', 'V141', 'V142', 'V143', 'V144', 'V145', 'V146', 'V147', 'V148', 'V149', 'V150', 'V151', 'V152', 'V153', 'V154', 'V155', 'V156', 'V157', 'V158', 'V159', 'V160', 'V161', 'V162', 'V163', 'V164', 'V165', 'V166', 'V167', 'V168', 'V169', 'V170', 'V171', 'V172', 'V173', 'V174', 'V175', 'V176', 'V177', 'V178', 'V179', 'V180', 'V181', 'V182', 'V183', 'V184', 'V185', 'V186', 'V187', 'V188', 'V189', 'V190', 'V191', 'V192', 'V193', 'V194', 'V195', 'V196', 'V197', 'V198', 'V199', 'V200', 'V201', 'V202', 'V203', 'V204', 'V205', 'V206', 'V207', 'V208', 'V209', 'V210', 'V211', 'V212', 'V213', 'V214', 'V215', 'V216', 'V217', 'V218', 'V219', 'V220', 'V221', 'V222', 'V223', 'V224', 'V225', 'V226', 'V227', 'V228', 'V229', 'V230', 'V231', 'V232', 'V233', 'V234', 'V235', 'V236', 'V237', 'V238', 'V239', 'V240', 'V241', 'V242', 'V243', 'V244', 'V245', 'V246', 'V247', 'V248', 'V249', 'V250', 'V251', 'V252', 'V253', 'V254', 'V255', 'V256', 'V257', 'V258', 'V259', 'V260', 'V261', 'V262', 'V263', 'V264', 'V265', 'V266', 'V267', 'V268', 'V269', 'V270', 'V271', 'V272', 'V273', 'V274', 'V275', 'V276', 'V277', 'V278', 'V279', 'V280', 'V281', 'V282', 'V283', 'V284', 'V285', 'V286', 'V287', 'V288', 'V289', 'V290', 'V291', 'V292', 'V293', 'V294', 'V295', 'V296', 'V297', 'V298', 'V299', 'V300', 'V301', 'V302', 'V303', 'V304', 'V305', 'V306', 'V307', 'V308', 'V309', 'V310', 'V311', 'V312', 'V313', 'V314', 'V315', 'V316', 'V317', 'V318', 'V319', 'V320', 'V321', 'V322', 'V323', 'V324', 'V325', 'V326', 'V327', 'V328', 'V329', 'V330', 'V331', 'V332', 'V333', 'V334', 'V335', 'V336', 'V337', 'V338', 'V339', 'id_01', 'id_02', 'id_03', 'id_04', 'id_05', 'id_06', 'id_07', 'id_08', 'id_09', 'id_10', 'id_11', 'id_12', 'id_13', 'id_14', 'id_15', 'id_16', 'id_17', 'id_18', 'id_19', 'id_20', 'id_21', 'id_22', 'id_23', 'id_24', 'id_25', 'id_26', 'id_27', 'id_28', 'id_29', 'id_30', 'id_31', 'id_32', 'id_33', 'id_34', 'id_35', 'id_36', 'id_37', 'id_38', 'DeviceType', 'DeviceInfo', 'Cents', 'Hour', 'DayOfWeek', 'Day', 'card1__addr1', 'card1__addr1_count', 'card1__card5', 'card1__card5_count', 'card2__dist1', 'card2__id_20', 'card5__P_emaildomain', 'uuid3_count', 'TransactionAmt_mean_by_uuid2_ProductCD', 'dist2_mean_by_uuid2_addr1', 'card1_count', 'card2_count', 'card5_count', 'addr1_count', 'C1_count', 'C13_count', 'P_emaildomain_count', 'D2_na', 'D3_na', 'D5_na', 'D6_na', 'D7_na', 'D13_na', 'D14_na', 'M4_na', 'M5_na', 'M6_na', 'M7_na', 'M8_na', 'P_emaildomain_parts', 'P_emaildomain_sfx2', 'P_emaildomain_sfx1', 'P_emaildomain_bin', 'R_emaildomain_parts', 'R_emaildomain_sfx2', 'R_emaildomain_sfx1', 'R_emaildomain_bin', 'Browser', 'OS', 'id_16_na', 'id_33_0', 'id_33_1']\n"
     ]
    }
   ],
   "source": [
    "print(list(df_train.columns))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_dfs = [df_train, df_val, df_hold, df_test]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Encoding uid\n",
      "Encoding uid2\n",
      "Encoding uid3\n"
     ]
    }
   ],
   "source": [
    "def frequency(col, dfs):\n",
    "    counts = pd.concat(map(lambda df: df[col], dfs), ignore_index=True).value_counts(dropna=False)\n",
    "    return map(lambda df: df[col].map(counts), dfs)\n",
    "\n",
    "# from https://www.kaggle.com/tolgahancepel/lightgbm-single-model-and-feature-engineering\n",
    "def concat_features(data): \n",
    "    data['uid'] = data['card1'].astype(str)+'_'+data['card2'].astype(str)\n",
    "    data['uid2'] = data['uid'].astype(str)+'_'+data['card3'].astype(str)+'_'+data['card5'].astype(str)\n",
    "    data['uid3'] = data['uid2'].astype(str)+'_'+data['addr1'].astype(str)+'_'+data['addr2'].astype(str)\n",
    "    return data\n",
    "\n",
    "for df in all_dfs:\n",
    "    df = concat_features(df)\n",
    "\n",
    "\n",
    "freq_cols = ['card1','card2','card3','card5',\n",
    "          'C2','C3','C4','C5','C6','C7','C8','C9','C10','C11','C12','C13','C14',\n",
    "          'D1','D2','D3','D4','D5','D6','D7','D8',\n",
    "          'addr1','addr2',\n",
    "          'dist1','dist2',\n",
    "          'P_emaildomain', 'R_emaildomain',\n",
    "          'DeviceInfo','DeviceType',\n",
    "          'id_30','id_33',\n",
    "          'uid','uid2','uid3',\n",
    "         ]\n",
    "for col in freq_cols:\n",
    "    for df, freq in zip(all_dfs, frequency(col, all_dfs)):\n",
    "        df[col+'_count'] = freq\n",
    "\n",
    "\n",
    "agg_cols = ['card1','card2','card3','card5', 'uid', 'uid2', 'uid3']  \n",
    "for col in agg_cols:\n",
    "    for agg_type in ['mean','std']:\n",
    "        new_col_name = col+'_TransactionAmt_'+agg_type\n",
    "        temp_df = pd.concat([df_train[[col, 'TransactionAmt']], df_val[[col, 'TransactionAmt']], df_hold[[col, 'TransactionAmt']], df_test[[col,'TransactionAmt']]])\n",
    "        #temp_df['TransactionAmt'] = temp_df['TransactionAmt'].astype(int)\n",
    "        temp_df = temp_df.groupby([col])['TransactionAmt'].agg([agg_type]).reset_index().rename(\n",
    "                                                columns={agg_type: new_col_name})\n",
    "\n",
    "        temp_df.index = list(temp_df[col])\n",
    "        temp_df = temp_df[new_col_name].to_dict()   \n",
    "        \n",
    "        for df in all_dfs:\n",
    "            df[new_col_name] = df[col].map(temp_df)\n",
    "        del temp_df\n",
    "\n",
    "\n",
    "encode_cols = ['uid', 'uid2', 'uid3']\n",
    "for col in encode_cols:\n",
    "    print('Encoding', col)\n",
    "    temp_df = pd.concat([df_train[col], df_val[col], df_hold[col], df_test[col]])\n",
    "    col_encoded = temp_df.value_counts().to_dict()\n",
    "    del temp_df\n",
    "    for df in all_dfs:\n",
    "        df[col] = df[col].map(col_encoded)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Feature engineering"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# initial_features = {\n",
    "#     'C1', 'C10', 'C11', 'C12', 'C13', 'C14', 'C2', 'C3', 'C4', 'C5', 'C6', 'C7', 'C8', 'C9', \n",
    "#     'Cents', 'D1', 'D10', 'D11', 'D12', 'D13', 'D14', 'D15', 'D2', 'D3', 'D4', 'D5', 'D6', \n",
    "#     'D7', 'D8', 'D9', 'DayOfMonth', 'DayOfWeek', 'DeviceInfo', 'DeviceType', 'Hour', \n",
    "#     'IsMonthEnd', 'IsMonthStart', 'IsQuarterEnd', 'IsQuarterStart', 'M1', 'M2', 'M3', 'M4', \n",
    "#     'M5', 'M6', 'M7', 'M8', 'M9', 'MonthFraction', 'P_emaildomain', 'ProductCD', 'Quarter', \n",
    "#     'R_emaildomain', 'TransactionAmt', 'TransactionDT', 'V1', 'V10', 'V100', 'V101', 'V102', \n",
    "#     'V103', 'V104', 'V105', 'V106', 'V108', 'V109', 'V11', 'V110', 'V112', 'V114', 'V115', \n",
    "#     'V116', 'V12', 'V123', 'V124', 'V125', 'V126', 'V127', 'V128', 'V129', 'V13', 'V130', \n",
    "#     'V131', 'V132', 'V133', 'V134', 'V135', 'V136', 'V137', 'V139', 'V140', 'V141', 'V142', \n",
    "#     'V143', 'V144', 'V145', 'V146', 'V147', 'V148', 'V149', 'V15', 'V150', 'V151', 'V152', \n",
    "#     'V154', 'V155', 'V156', 'V157', 'V158', 'V159', 'V16', 'V160', 'V161', 'V162', 'V163', \n",
    "#     'V164', 'V165', 'V166', 'V167', 'V168', 'V169', 'V17', 'V170', 'V171', 'V172', 'V173', \n",
    "#     'V174', 'V175', 'V176', 'V177', 'V178', 'V179', 'V18', 'V180', 'V181', 'V182', 'V183', \n",
    "#     'V184', 'V185', 'V186', 'V187', 'V188', 'V189', 'V19', 'V190', 'V192', 'V193', 'V198', \n",
    "#     'V199', 'V2', 'V20', 'V200', 'V201', 'V202', 'V203', 'V204', 'V205', 'V206', 'V207', \n",
    "#     'V208', 'V209', 'V210', 'V211', 'V212', 'V213', 'V214', 'V215', 'V216', 'V217', 'V218', \n",
    "#     'V219', 'V22', 'V220', 'V221', 'V222', 'V223', 'V224', 'V225', 'V226', 'V227', 'V228', \n",
    "#     'V229', 'V23', 'V230', 'V231', 'V232', 'V233', 'V234', 'V235', 'V236', 'V237', 'V238', \n",
    "#     'V239', 'V24', 'V242', 'V243', 'V244', 'V245', 'V246', 'V247', 'V248', 'V249', 'V25', \n",
    "#     'V250', 'V251', 'V252', 'V253', 'V254', 'V255', 'V256', 'V257', 'V258', 'V259', 'V26', \n",
    "#     'V260', 'V261', 'V262', 'V263', 'V264', 'V265', 'V266', 'V267', 'V268', 'V270', 'V271', \n",
    "#     'V272', 'V273', 'V274', 'V275', 'V276', 'V277', 'V278', 'V279', 'V280', 'V281', 'V282', \n",
    "#     'V283', 'V284', 'V285', 'V286', 'V287', 'V288', 'V289', 'V29', 'V290', 'V291', 'V292', \n",
    "#     'V293', 'V294', 'V295', 'V296', 'V297', 'V298', 'V299', 'V3', 'V30', 'V300', 'V301', \n",
    "#     'V302', 'V303', 'V304', 'V306', 'V307', 'V308', 'V309', 'V310', 'V311', 'V312', 'V313', \n",
    "#     'V314', 'V315', 'V316', 'V317', 'V318', 'V319', 'V32', 'V320', 'V321', 'V322', 'V323', \n",
    "#     'V324', 'V326', 'V327', 'V329', 'V33', 'V331', 'V332', 'V333', 'V335', 'V336', 'V337', \n",
    "#     'V338', 'V339', 'V34', 'V35', 'V36', 'V37', 'V38', 'V39', 'V4', 'V40', 'V42', 'V43', 'V44', \n",
    "#     'V45', 'V46', 'V47', 'V48', 'V49', 'V5', 'V50', 'V51', 'V52', 'V53', 'V54', 'V55', 'V56', \n",
    "#     'V57', 'V58', 'V59', 'V6', 'V60', 'V61', 'V62', 'V63', 'V64', 'V66', 'V67', 'V69', 'V7', \n",
    "#     'V70', 'V71', 'V72', 'V73', 'V74', 'V75', 'V76', 'V77', 'V78', 'V79', 'V8', 'V80', 'V81', \n",
    "#     'V82', 'V83', 'V84', 'V85', 'V86', 'V87', 'V9', 'V90', 'V91', 'V92', 'V93', 'V94', 'V95', \n",
    "#     'V96', 'V97', 'V98', 'V99', 'addr1', 'addr2', 'card1', 'card2', 'card3', 'card4', 'card5', \n",
    "#     'card6', 'dist1', 'dist2', 'id_01', 'id_02', 'id_03', 'id_04', 'id_05', 'id_06', 'id_07', \n",
    "#     'id_08', 'id_09', 'id_10', 'id_11', 'id_12', 'id_13', 'id_14', 'id_15', 'id_16', 'id_16_na', \n",
    "#     'id_17', 'id_18', 'id_19', 'id_20', 'id_21', 'id_24', 'id_25', 'id_26', 'id_28', 'id_29', \n",
    "#     'id_30', 'id_31', 'id_32', 'id_33', 'id_33_0', 'id_33_1', 'id_34', 'id_35', 'id_36', 'id_37', \n",
    "#     'id_38'\n",
    "# }\n",
    "\n",
    "# for df in [df_train, df_val, df_hold]:   \n",
    "#     for col in df.columns:\n",
    "#         if col not in initial_features:\n",
    "#             df.drop(col, axis=1, inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Feature selection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def model_params():\n",
    "    \"\"\"\n",
    "    Params are from https://www.kaggle.com/kyakovlev/ieee-lgbm-with-groupkfold-cv\n",
    "    \"\"\"\n",
    "    return {\n",
    "        'objective':'binary',\n",
    "        'boosting_type':'gbdt',\n",
    "        'metric':'auc',\n",
    "        'n_jobs':-1,\n",
    "        'learning_rate':0.01,\n",
    "        'num_leaves': 2**8,\n",
    "        'max_depth':-1,\n",
    "        'tree_learner':'serial',\n",
    "        'colsample_bytree': 0.5,\n",
    "        'subsample_freq':1,\n",
    "        'subsample':0.7,\n",
    "        'max_bin':255,\n",
    "        'verbose':1,\n",
    "        'seed': 133,\n",
    "    }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def rnd_int():\n",
    "    return np.random.randint(np.iinfo(np.int32).max)\n",
    "\n",
    "\n",
    "def sample(x, y, neg_frac=0.2, seed=None):\n",
    "    if not seed:\n",
    "        seed = rnd_int()\n",
    "    x_pos, y_pos = x[y==1], y[y==1]\n",
    "    x_neg, _, y_neg, _ = train_test_split(x[y==0], y[y==0], train_size=neg_frac, random_state=seed)\n",
    "    x = pd.concat([x_pos, x_neg])\n",
    "    y = pd.concat([y_pos, y_neg])\n",
    "    return shuffle(x, y, random_state=seed)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def search(n_runs=5, neg_frac=0.2, importance_tresh=0.1, permut_tresh=1e-4):\n",
    "    f_imps = []\n",
    "    selected_features = set()\n",
    "    params = model_params()\n",
    "    \n",
    "    for run in range(n_runs):\n",
    "        lgb_train = lgb.Dataset(*sample(df_train, y_train, neg_frac))\n",
    "        lgb_eval = lgb.Dataset(df_val, y_val, reference=lgb_train)\n",
    "        \n",
    "        seed_everything(rnd_int())\n",
    "        params['seed'] = rnd_int()\n",
    "        print('Starting %d run...' % run)\n",
    "        \n",
    "        lgb_model = lgb.train(params, lgb_train, \n",
    "                              num_boost_round=3000, \n",
    "                              valid_sets=lgb_eval, \n",
    "                              early_stopping_rounds=100,\n",
    "                              verbose_eval=False)\n",
    "        \n",
    "        print('Finished %d run. Selecting features...' % run)\n",
    "        top_features, worst_features, f_imp = split_features_by_importance(df_train, df_val, y_val, lgb_model, \n",
    "                                   importance_tresh, permut_tresh)\n",
    "        f_imps.append(f_imp)\n",
    "        selected_features.update(top_features)\n",
    "        print('Consider to drop: ', sorted(worst_features))\n",
    "    return list(selected_features), avg_feature_importance(f_imps)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "See in https://www.kaggle.com/blackitten13/fraud-feature-selection-fe9477"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# selected_features, feature_importance = search(n_runs=4, neg_frac=0.5, importance_tresh=0.05, permut_tresh=1e-5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "_kg_hide-output": true
   },
   "outputs": [],
   "source": [
    "# fi = feature_importance['gain'].nlargest(60)\n",
    "# print('Best 60 features by GAIN feature importnace:\\n\\n{}'.format(fi))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "_kg_hide-output": true
   },
   "outputs": [],
   "source": [
    "# fi = feature_importance['gain'].nsmallest(60)\n",
    "# print('Worst 60 features by GAIN feature importnace:\\n\\n{}'.format(fi))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "_kg_hide-output": true
   },
   "outputs": [],
   "source": [
    "# fi = feature_importance['split'].nlargest(60)\n",
    "# print('Best 60 features by SPLIT feature importnace:\\n\\n{}'.format(fi))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "_kg_hide-output": true
   },
   "outputs": [],
   "source": [
    "# fi = feature_importance['split'].nsmallest(60)\n",
    "# print('Worst 60 features by SPLIT feature importnace:\\n\\n{}'.format(fi))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# print('{} / {}'.format(len(selected_features), len(df_train.columns)))\n",
    "# print(sorted(selected_features))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "selected_features = ['Browser', 'C1', 'C10', 'C10_count', 'C11', 'C11_count', 'C12', 'C12_count', 'C13', 'C13_count', 'C14', 'C14_count', 'C1_count', 'C2', 'C2_count', 'C3', 'C3_count', 'C4', 'C4_count', 'C5', 'C5_count', 'C6', 'C6_count', 'C7', 'C7_count', 'C8', 'C8_count', 'C9', 'C9_count', 'Cents', 'D1', 'D10', 'D11', 'D12', 'D13', 'D13_na', 'D14', 'D14_na', 'D15', 'D1_count', 'D2', 'D2_count', 'D2_na', 'D3', 'D3_count', 'D3_na', 'D4', 'D4_count', 'D5', 'D5_count', 'D5_na', 'D6', 'D6_count', 'D6_na', 'D7', 'D7_count', 'D7_na', 'D8', 'D8_count', 'D9', 'DayOfWeek', 'DeviceInfo', 'DeviceInfo_count', 'DeviceType', 'DeviceType_count', 'Hour', 'M1', 'M2', 'M3', 'M4', 'M4_na', 'M5', 'M5_na', 'M6', 'M6_na', 'M7', 'M7_na', 'M8', 'M8_na', 'M9', 'OS', 'P_emaildomain', 'P_emaildomain_bin', 'P_emaildomain_count', 'P_emaildomain_parts', 'P_emaildomain_sfx1', 'P_emaildomain_sfx2', 'ProductCD', 'R_emaildomain', 'R_emaildomain_bin', 'R_emaildomain_count', 'R_emaildomain_parts', 'R_emaildomain_sfx1', 'R_emaildomain_sfx2', 'TransactionAmt', 'TransactionDT', 'V1', 'V10', 'V100', 'V101', 'V102', 'V103', 'V104', 'V105', 'V106', 'V108', 'V109', 'V11', 'V110', 'V111', 'V112', 'V114', 'V115', 'V116', 'V12', 'V122', 'V123', 'V124', 'V125', 'V126', 'V127', 'V128', 'V129', 'V13', 'V130', 'V131', 'V132', 'V133', 'V134', 'V135', 'V136', 'V137', 'V138', 'V139', 'V140', 'V141', 'V142', 'V143', 'V144', 'V145', 'V146', 'V147', 'V148', 'V149', 'V15', 'V150', 'V151', 'V152', 'V153', 'V154', 'V155', 'V156', 'V157', 'V158', 'V159', 'V16', 'V160', 'V161', 'V162', 'V163', 'V164', 'V165', 'V166', 'V167', 'V168', 'V169', 'V17', 'V170', 'V171', 'V172', 'V173', 'V174', 'V175', 'V176', 'V177', 'V178', 'V179', 'V18', 'V180', 'V181', 'V182', 'V183', 'V184', 'V185', 'V186', 'V187', 'V188', 'V189', 'V19', 'V190', 'V191', 'V192', 'V193', 'V194', 'V195', 'V196', 'V197', 'V198', 'V199', 'V2', 'V20', 'V200', 'V201', 'V202', 'V203', 'V204', 'V205', 'V206', 'V207', 'V208', 'V209', 'V21', 'V210', 'V211', 'V212', 'V213', 'V214', 'V215', 'V216', 'V217', 'V218', 'V219', 'V22', 'V220', 'V221', 'V222', 'V223', 'V224', 'V225', 'V226', 'V227', 'V228', 'V229', 'V23', 'V230', 'V231', 'V232', 'V233', 'V234', 'V235', 'V236', 'V237', 'V238', 'V239', 'V24', 'V240', 'V242', 'V243', 'V244', 'V245', 'V246', 'V247', 'V248', 'V249', 'V25', 'V250', 'V251', 'V252', 'V253', 'V254', 'V255', 'V256', 'V257', 'V258', 'V259', 'V26', 'V260', 'V261', 'V262', 'V263', 'V264', 'V265', 'V266', 'V267', 'V268', 'V270', 'V271', 'V272', 'V273', 'V274', 'V275', 'V276', 'V277', 'V278', 'V279', 'V280', 'V281', 'V282', 'V283', 'V284', 'V285', 'V286', 'V287', 'V288', 'V289', 'V29', 'V290', 'V291', 'V292', 'V293', 'V294', 'V295', 'V296', 'V297', 'V298', 'V299', 'V3', 'V30', 'V300', 'V301', 'V302', 'V303', 'V304', 'V306', 'V307', 'V308', 'V309', 'V31', 'V310', 'V311', 'V312', 'V313', 'V314', 'V315', 'V316', 'V317', 'V318', 'V319', 'V32', 'V320', 'V321', 'V322', 'V323', 'V324', 'V325', 'V326', 'V327', 'V328', 'V329', 'V33', 'V330', 'V331', 'V332', 'V333', 'V334', 'V335', 'V336', 'V337', 'V338', 'V339', 'V34', 'V35', 'V36', 'V37', 'V38', 'V39', 'V4', 'V40', 'V42', 'V43', 'V44', 'V45', 'V46', 'V47', 'V48', 'V49', 'V5', 'V50', 'V51', 'V52', 'V53', 'V54', 'V55', 'V56', 'V57', 'V58', 'V59', 'V6', 'V60', 'V61', 'V62', 'V63', 'V64', 'V66', 'V67', 'V69', 'V7', 'V70', 'V71', 'V72', 'V73', 'V74', 'V75', 'V76', 'V77', 'V78', 'V79', 'V8', 'V80', 'V81', 'V82', 'V83', 'V84', 'V85', 'V86', 'V87', 'V9', 'V90', 'V91', 'V92', 'V93', 'V94', 'V95', 'V96', 'V97', 'V98', 'V99', 'addr1', 'addr1_count', 'addr2', 'addr2_count', 'card1', 'card1_TransactionAmt_mean', 'card1_TransactionAmt_std', 'card1__addr1', 'card1__addr1_count', 'card1__card5', 'card1__card5_count', 'card1_count', 'card2', 'card2_TransactionAmt_mean', 'card2_TransactionAmt_std', 'card2__dist1', 'card2__id_20', 'card2_count', 'card3', 'card3_TransactionAmt_mean', 'card3_TransactionAmt_std', 'card3_count', 'card4', 'card5', 'card5_TransactionAmt_mean', 'card5_TransactionAmt_std', 'card5__P_emaildomain', 'card5_count', 'card6', 'dist1', 'dist1_count', 'dist2', 'dist2_count', 'id_01', 'id_02', 'id_03', 'id_04', 'id_05', 'id_06', 'id_07', 'id_08', 'id_09', 'id_10', 'id_11', 'id_12', 'id_13', 'id_14', 'id_15', 'id_16', 'id_16_na', 'id_17', 'id_18', 'id_19', 'id_20', 'id_21', 'id_22', 'id_23', 'id_24', 'id_25', 'id_26', 'id_28', 'id_29', 'id_30', 'id_30_count', 'id_31', 'id_32', 'id_33', 'id_33_0', 'id_33_1', 'id_33_count', 'id_34', 'id_35', 'id_36', 'id_37', 'id_38', 'uid', 'uid2', 'uid2_TransactionAmt_mean', 'uid2_TransactionAmt_std', 'uid2_count', 'uid3', 'uid3_TransactionAmt_mean', 'uid3_TransactionAmt_std', 'uid3_count', 'uid_TransactionAmt_mean', 'uid_TransactionAmt_std', 'uid_count']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Drop:  ['Day', 'TransactionAmt_mean_by_uuid2_ProductCD', 'V107', 'V113', 'V117', 'V118', 'V119', 'V120', 'V121', 'V14', 'V241', 'V269', 'V27', 'V28', 'V305', 'V41', 'V65', 'V68', 'V88', 'V89', 'dist2_mean_by_uuid2_addr1', 'id_27', 'uuid3_count']\n"
     ]
    }
   ],
   "source": [
    "drop_features = set(df_train.columns) - set(selected_features)\n",
    "print('Drop: ', sorted(list(drop_features)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Test selected features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "for df in [df_train, df_val, df_hold, df_test]:\n",
    "    for col in df.columns:\n",
    "        if col not in selected_features:\n",
    "            df.drop(col, axis=1, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train = pd.concat([df_train, df_val, df_hold], axis=0, ignore_index=True)\n",
    "y = pd.concat([y_train, y_val, y_hold], axis=0, ignore_index=True)\n",
    "del df_val, df_hold, y_train, y_val, y_hold"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "def feature_cents(df):\n",
    "    return (np.modf(df['TransactionAmt'])[0] * 1000).round(5).astype(np.uint16)\n",
    "\n",
    "train_trans = pd.read_csv('../input/ieee-fraud-detection/train_transaction.csv')\n",
    "df_train['Cents'] = feature_cents(train_trans).values\n",
    "del train_trans\n",
    "\n",
    "test_trans = pd.read_csv('../input/ieee-fraud-detection/test_transaction.csv')\n",
    "df_test['Cents'] = feature_cents(test_trans).values\n",
    "del test_trans"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.6/site-packages/pandas/core/ops/__init__.py:1115: FutureWarning: elementwise comparison failed; returning scalar instead, but in the future will perform elementwise comparison\n",
      "  result = method(y)\n"
     ]
    }
   ],
   "source": [
    "def feature_ProductCD_W_cents(df):\n",
    "    a = ((df['Cents'] == 0) | (df['Cents'] == 500) | (df['Cents'] == 950))\n",
    "    b = df['ProductCD'] == 'W'\n",
    "    return (a & b).astype(np.uint8)\n",
    "\n",
    "df_train['Cents_ProductCD_W'] = feature_ProductCD_W_cents(df_train)\n",
    "df_test['Cents_ProductCD_W'] = feature_ProductCD_W_cents(df_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "_kg_hide-output": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['TransactionDT', 'TransactionAmt', 'ProductCD', 'card1', 'card2', 'card3', 'card4', 'card5', 'card6', 'addr1', 'addr2', 'dist1', 'dist2', 'P_emaildomain', 'R_emaildomain', 'C1', 'C2', 'C3', 'C4', 'C5', 'C6', 'C7', 'C8', 'C9', 'C10', 'C11', 'C12', 'C13', 'C14', 'D1', 'D2', 'D3', 'D4', 'D5', 'D6', 'D7', 'D8', 'D9', 'D10', 'D11', 'D12', 'D13', 'D14', 'D15', 'M1', 'M2', 'M3', 'M4', 'M5', 'M6', 'M7', 'M8', 'M9', 'V1', 'V2', 'V3', 'V4', 'V5', 'V6', 'V7', 'V8', 'V9', 'V10', 'V11', 'V12', 'V13', 'V15', 'V16', 'V17', 'V18', 'V19', 'V20', 'V21', 'V22', 'V23', 'V24', 'V25', 'V26', 'V29', 'V30', 'V31', 'V32', 'V33', 'V34', 'V35', 'V36', 'V37', 'V38', 'V39', 'V40', 'V42', 'V43', 'V44', 'V45', 'V46', 'V47', 'V48', 'V49', 'V50', 'V51', 'V52', 'V53', 'V54', 'V55', 'V56', 'V57', 'V58', 'V59', 'V60', 'V61', 'V62', 'V63', 'V64', 'V66', 'V67', 'V69', 'V70', 'V71', 'V72', 'V73', 'V74', 'V75', 'V76', 'V77', 'V78', 'V79', 'V80', 'V81', 'V82', 'V83', 'V84', 'V85', 'V86', 'V87', 'V90', 'V91', 'V92', 'V93', 'V94', 'V95', 'V96', 'V97', 'V98', 'V99', 'V100', 'V101', 'V102', 'V103', 'V104', 'V105', 'V106', 'V108', 'V109', 'V110', 'V111', 'V112', 'V114', 'V115', 'V116', 'V122', 'V123', 'V124', 'V125', 'V126', 'V127', 'V128', 'V129', 'V130', 'V131', 'V132', 'V133', 'V134', 'V135', 'V136', 'V137', 'V138', 'V139', 'V140', 'V141', 'V142', 'V143', 'V144', 'V145', 'V146', 'V147', 'V148', 'V149', 'V150', 'V151', 'V152', 'V153', 'V154', 'V155', 'V156', 'V157', 'V158', 'V159', 'V160', 'V161', 'V162', 'V163', 'V164', 'V165', 'V166', 'V167', 'V168', 'V169', 'V170', 'V171', 'V172', 'V173', 'V174', 'V175', 'V176', 'V177', 'V178', 'V179', 'V180', 'V181', 'V182', 'V183', 'V184', 'V185', 'V186', 'V187', 'V188', 'V189', 'V190', 'V191', 'V192', 'V193', 'V194', 'V195', 'V196', 'V197', 'V198', 'V199', 'V200', 'V201', 'V202', 'V203', 'V204', 'V205', 'V206', 'V207', 'V208', 'V209', 'V210', 'V211', 'V212', 'V213', 'V214', 'V215', 'V216', 'V217', 'V218', 'V219', 'V220', 'V221', 'V222', 'V223', 'V224', 'V225', 'V226', 'V227', 'V228', 'V229', 'V230', 'V231', 'V232', 'V233', 'V234', 'V235', 'V236', 'V237', 'V238', 'V239', 'V240', 'V242', 'V243', 'V244', 'V245', 'V246', 'V247', 'V248', 'V249', 'V250', 'V251', 'V252', 'V253', 'V254', 'V255', 'V256', 'V257', 'V258', 'V259', 'V260', 'V261', 'V262', 'V263', 'V264', 'V265', 'V266', 'V267', 'V268', 'V270', 'V271', 'V272', 'V273', 'V274', 'V275', 'V276', 'V277', 'V278', 'V279', 'V280', 'V281', 'V282', 'V283', 'V284', 'V285', 'V286', 'V287', 'V288', 'V289', 'V290', 'V291', 'V292', 'V293', 'V294', 'V295', 'V296', 'V297', 'V298', 'V299', 'V300', 'V301', 'V302', 'V303', 'V304', 'V306', 'V307', 'V308', 'V309', 'V310', 'V311', 'V312', 'V313', 'V314', 'V315', 'V316', 'V317', 'V318', 'V319', 'V320', 'V321', 'V322', 'V323', 'V324', 'V325', 'V326', 'V327', 'V328', 'V329', 'V330', 'V331', 'V332', 'V333', 'V334', 'V335', 'V336', 'V337', 'V338', 'V339', 'id_01', 'id_02', 'id_03', 'id_04', 'id_05', 'id_06', 'id_07', 'id_08', 'id_09', 'id_10', 'id_11', 'id_12', 'id_13', 'id_14', 'id_15', 'id_16', 'id_17', 'id_18', 'id_19', 'id_20', 'id_21', 'id_22', 'id_23', 'id_24', 'id_25', 'id_26', 'id_28', 'id_29', 'id_30', 'id_31', 'id_32', 'id_33', 'id_34', 'id_35', 'id_36', 'id_37', 'id_38', 'DeviceType', 'DeviceInfo', 'Cents', 'Hour', 'DayOfWeek', 'card1__addr1', 'card1__addr1_count', 'card1__card5', 'card1__card5_count', 'card2__dist1', 'card2__id_20', 'card5__P_emaildomain', 'card1_count', 'card2_count', 'card5_count', 'addr1_count', 'C1_count', 'C13_count', 'P_emaildomain_count', 'D2_na', 'D3_na', 'D5_na', 'D6_na', 'D7_na', 'D13_na', 'D14_na', 'M4_na', 'M5_na', 'M6_na', 'M7_na', 'M8_na', 'P_emaildomain_parts', 'P_emaildomain_sfx2', 'P_emaildomain_sfx1', 'P_emaildomain_bin', 'R_emaildomain_parts', 'R_emaildomain_sfx2', 'R_emaildomain_sfx1', 'R_emaildomain_bin', 'Browser', 'OS', 'id_16_na', 'id_33_0', 'id_33_1', 'uid', 'uid2', 'uid3', 'card3_count', 'C2_count', 'C3_count', 'C4_count', 'C5_count', 'C6_count', 'C7_count', 'C8_count', 'C9_count', 'C10_count', 'C11_count', 'C12_count', 'C14_count', 'D1_count', 'D2_count', 'D3_count', 'D4_count', 'D5_count', 'D6_count', 'D7_count', 'D8_count', 'addr2_count', 'dist1_count', 'dist2_count', 'R_emaildomain_count', 'DeviceInfo_count', 'DeviceType_count', 'id_30_count', 'id_33_count', 'uid_count', 'uid2_count', 'uid3_count', 'card1_TransactionAmt_mean', 'card1_TransactionAmt_std', 'card2_TransactionAmt_mean', 'card2_TransactionAmt_std', 'card3_TransactionAmt_mean', 'card3_TransactionAmt_std', 'card5_TransactionAmt_mean', 'card5_TransactionAmt_std', 'uid_TransactionAmt_mean', 'uid_TransactionAmt_std', 'uid2_TransactionAmt_mean', 'uid2_TransactionAmt_std', 'uid3_TransactionAmt_mean', 'uid3_TransactionAmt_std', 'Cents_ProductCD_W']\n"
     ]
    }
   ],
   "source": [
    "print(list(df_train.columns))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "START_DATE = datetime.datetime.strptime('2017-11-30', '%Y-%m-%d')\n",
    "TransactionDT1 = df_train['TransactionDT'].apply(lambda x: (START_DATE + datetime.timedelta(seconds = x)))\n",
    "dt_m = TransactionDT1.dt.date.astype('str').str[:7]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((590540, 504), (506691, 504))"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_train = df_train.drop(['TransactionDT'], axis=1)\n",
    "df_test = df_test.drop(['TransactionDT'], axis=1)\n",
    "df_train.shape, df_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "_kg_hide-output": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(453219, 504) (453219,) (137321, 504) (137321,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.964284\tvalid_1's auc: 0.895373\n",
      "[400]\ttraining's auc: 0.987794\tvalid_1's auc: 0.908886\n",
      "[600]\ttraining's auc: 0.995253\tvalid_1's auc: 0.915284\n",
      "[800]\ttraining's auc: 0.998047\tvalid_1's auc: 0.918138\n",
      "[1000]\ttraining's auc: 0.999182\tvalid_1's auc: 0.9197\n",
      "[1200]\ttraining's auc: 0.999661\tvalid_1's auc: 0.920518\n",
      "[1400]\ttraining's auc: 0.999859\tvalid_1's auc: 0.921108\n",
      "[1600]\ttraining's auc: 0.999941\tvalid_1's auc: 0.921444\n",
      "[1800]\ttraining's auc: 0.999978\tvalid_1's auc: 0.92136\n",
      "[2000]\ttraining's auc: 0.999992\tvalid_1's auc: 0.92178\n",
      "[2200]\ttraining's auc: 0.999997\tvalid_1's auc: 0.921586\n",
      "[2400]\ttraining's auc: 0.999999\tvalid_1's auc: 0.921674\n",
      "Early stopping, best iteration is:\n",
      "[2026]\ttraining's auc: 0.999993\tvalid_1's auc: 0.921837\n",
      "Fold 1 | AUC: 0.9218366725308682\n",
      "(497955, 504) (497955,) (92585, 504) (92585,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.960606\tvalid_1's auc: 0.915284\n",
      "[400]\ttraining's auc: 0.986751\tvalid_1's auc: 0.933248\n",
      "[600]\ttraining's auc: 0.995063\tvalid_1's auc: 0.940378\n",
      "[800]\ttraining's auc: 0.99795\tvalid_1's auc: 0.943205\n",
      "[1000]\ttraining's auc: 0.999115\tvalid_1's auc: 0.944565\n",
      "[1200]\ttraining's auc: 0.999631\tvalid_1's auc: 0.945358\n",
      "[1400]\ttraining's auc: 0.999848\tvalid_1's auc: 0.94563\n",
      "[1600]\ttraining's auc: 0.999939\tvalid_1's auc: 0.945754\n",
      "[1800]\ttraining's auc: 0.999976\tvalid_1's auc: 0.945915\n",
      "[2000]\ttraining's auc: 0.999991\tvalid_1's auc: 0.945961\n",
      "[2200]\ttraining's auc: 0.999997\tvalid_1's auc: 0.946138\n",
      "[2400]\ttraining's auc: 0.999999\tvalid_1's auc: 0.946111\n",
      "[2600]\ttraining's auc: 1\tvalid_1's auc: 0.946097\n",
      "Early stopping, best iteration is:\n",
      "[2240]\ttraining's auc: 0.999998\tvalid_1's auc: 0.946261\n",
      "Fold 2 | AUC: 0.9462608153527904\n",
      "(504519, 504) (504519,) (86021, 504) (86021,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.96062\tvalid_1's auc: 0.93217\n",
      "[400]\ttraining's auc: 0.986494\tvalid_1's auc: 0.945068\n",
      "[600]\ttraining's auc: 0.994645\tvalid_1's auc: 0.94895\n",
      "[800]\ttraining's auc: 0.997721\tvalid_1's auc: 0.950217\n",
      "[1000]\ttraining's auc: 0.99896\tvalid_1's auc: 0.950664\n",
      "[1200]\ttraining's auc: 0.999522\tvalid_1's auc: 0.950827\n",
      "[1400]\ttraining's auc: 0.9998\tvalid_1's auc: 0.950803\n",
      "[1600]\ttraining's auc: 0.999914\tvalid_1's auc: 0.950814\n",
      "Early stopping, best iteration is:\n",
      "[1116]\ttraining's auc: 0.999336\tvalid_1's auc: 0.95094\n",
      "Fold 3 | AUC: 0.9509404436480211\n",
      "(488908, 504) (488908,) (101632, 504) (101632,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.962722\tvalid_1's auc: 0.921133\n",
      "[400]\ttraining's auc: 0.98788\tvalid_1's auc: 0.937799\n",
      "[600]\ttraining's auc: 0.995611\tvalid_1's auc: 0.943224\n",
      "[800]\ttraining's auc: 0.998295\tvalid_1's auc: 0.945335\n",
      "[1000]\ttraining's auc: 0.999295\tvalid_1's auc: 0.946239\n",
      "[1200]\ttraining's auc: 0.999698\tvalid_1's auc: 0.946575\n",
      "[1400]\ttraining's auc: 0.999876\tvalid_1's auc: 0.946716\n",
      "[1600]\ttraining's auc: 0.999952\tvalid_1's auc: 0.946709\n",
      "[1800]\ttraining's auc: 0.999981\tvalid_1's auc: 0.946393\n",
      "Early stopping, best iteration is:\n",
      "[1423]\ttraining's auc: 0.999889\tvalid_1's auc: 0.946733\n",
      "Fold 4 | AUC: 0.9467334592472505\n",
      "(506885, 504) (506885,) (83655, 504) (83655,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.959376\tvalid_1's auc: 0.924081\n",
      "[400]\ttraining's auc: 0.98598\tvalid_1's auc: 0.944993\n",
      "[600]\ttraining's auc: 0.994416\tvalid_1's auc: 0.951785\n",
      "[800]\ttraining's auc: 0.997645\tvalid_1's auc: 0.95439\n",
      "[1000]\ttraining's auc: 0.998919\tvalid_1's auc: 0.955825\n",
      "[1200]\ttraining's auc: 0.999529\tvalid_1's auc: 0.956338\n",
      "[1400]\ttraining's auc: 0.999795\tvalid_1's auc: 0.956623\n",
      "[1600]\ttraining's auc: 0.999914\tvalid_1's auc: 0.956641\n",
      "[1800]\ttraining's auc: 0.999963\tvalid_1's auc: 0.956648\n",
      "[2000]\ttraining's auc: 0.999985\tvalid_1's auc: 0.956547\n",
      "Early stopping, best iteration is:\n",
      "[1631]\ttraining's auc: 0.999925\tvalid_1's auc: 0.956728\n",
      "Fold 5 | AUC: 0.9567275578313148\n",
      "(501214, 504) (501214,) (89326, 504) (89326,)\n",
      "Training until validation scores don't improve for 500 rounds.\n",
      "[200]\ttraining's auc: 0.961256\tvalid_1's auc: 0.916899\n",
      "[400]\ttraining's auc: 0.986607\tvalid_1's auc: 0.933307\n",
      "[600]\ttraining's auc: 0.99484\tvalid_1's auc: 0.939119\n",
      "[800]\ttraining's auc: 0.997868\tvalid_1's auc: 0.941109\n",
      "[1000]\ttraining's auc: 0.999076\tvalid_1's auc: 0.941743\n",
      "[1200]\ttraining's auc: 0.999593\tvalid_1's auc: 0.941398\n",
      "[1400]\ttraining's auc: 0.999821\tvalid_1's auc: 0.941473\n",
      "Early stopping, best iteration is:\n",
      "[994]\ttraining's auc: 0.999057\tvalid_1's auc: 0.941788\n",
      "Fold 6 | AUC: 0.9417879782005023\n",
      "\n",
      "Mean AUC = 0.9440478211351245\n"
     ]
    }
   ],
   "source": [
    "y_preds = np.zeros(df_test.shape[0])\n",
    "score = 0\n",
    "\n",
    "feature_importances = pd.DataFrame()\n",
    "feature_importances['feature'] = df_train.columns\n",
    "\n",
    "NFOLDS = len(np.unique(dt_m)) # 6\n",
    "\n",
    "params = model_params()\n",
    "\n",
    "for fold_n, month in enumerate(np.unique(dt_m)):\n",
    "    X_train, y_train = df_train[~dt_m.isin([month])], y[~dt_m.isin([month])]\n",
    "    X_valid, y_valid = df_train[dt_m.isin([month])], y[dt_m.isin([month])]\n",
    "    print(X_train.shape, y_train.shape, X_valid.shape, y_valid.shape)\n",
    "\n",
    "    dtrain = lgb.Dataset(X_train, label=y_train)\n",
    "    dvalid = lgb.Dataset(X_valid, label=y_valid)\n",
    "\n",
    "    clf = lgb.train(params, dtrain, 10000, valid_sets = [dtrain, dvalid], verbose_eval=200, early_stopping_rounds=500)\n",
    "    \n",
    "    feature_importances[f'fold_{fold_n + 1}'] = clf.feature_importance()\n",
    "    \n",
    "    y_pred_valid = clf.predict(X_valid)\n",
    "    save_to_disk(y_pred_valid, 'y_pred_valid_fold{}.pkl'.format(fold_n))\n",
    "    print(f\"Fold {fold_n + 1} | AUC: {roc_auc_score(y_valid, y_pred_valid)}\")\n",
    "    \n",
    "    score += roc_auc_score(y_valid, y_pred_valid) / NFOLDS\n",
    "    \n",
    "    y_pred_test = clf.predict(df_test)\n",
    "    save_to_disk(y_pred_test, 'y_pred_test_fold{}.pkl'.format(fold_n))\n",
    "    y_preds += y_pred_test / NFOLDS\n",
    "    \n",
    "    del X_train, X_valid, y_train, y_valid\n",
    "    gc.collect()\n",
    "\n",
    "print(f\"\\nMean AUC = {score}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Submission"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "sub = pd.read_csv('../input/ieee-fraud-detection/sample_submission.csv')\n",
    "sub['isFraud'] = y_preds\n",
    "sub.to_csv('submission.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x1152 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "feature_importances['average'] = feature_importances[[f'fold_{fold_n + 1}' for fold_n in range(NFOLDS)]].mean(axis=1)\n",
    "feature_importances.to_csv('feature_importances.csv')\n",
    "\n",
    "plt.figure(figsize=(16, 16))\n",
    "sns.barplot(data=feature_importances.sort_values(by='average', ascending=False).head(50), x='average', y='feature');\n",
    "plt.title('50 TOP feature importance over {} folds average'.format(NFOLDS));"
   ]
  }
 ],
 "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": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}