A_0.824409_0.718946.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "5d3c31cb-8c2e-43e0-bdce-7aa86e851df2",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from os import path as osp\n",
    "import os\n",
    "from matplotlib import pyplot as plt\n",
    "import lightgbm as lgb\n",
    "\n",
    "from tqdm.notebook import tqdm\n",
    "from collections import Counter\n",
    "from sklearn.model_selection import StratifiedKFold\n",
    "from sklearn.metrics import roc_auc_score, f1_score, precision_score, recall_score, fbeta_score, mean_absolute_percentage_error\n",
    "import utils\n",
    "\n",
    "import pickle\n",
    "def save_obj(obj, name):\n",
    "    with open(name + '.pickle', 'wb') as f:\n",
    "        pickle.dump(obj, f, pickle.HIGHEST_PROTOCOL)\n",
    "def load_obj(name ):\n",
    "    with open(name + '.pickle', 'rb') as f:\n",
    "        return pickle.load(f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "8d9b7ca2-b30b-41e7-831f-f9cdd56404c6",
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings\n",
    "\n",
    "# 禁用FutureWarning警告\n",
    "warnings.simplefilter(action='ignore', category=FutureWarning)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "348be3ad-341f-4641-8be9-b8bb38aa25d1",
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_PATH = '/data'\n",
    "MODEL_PAHT = '/work/model'\n",
    "SEED = 2222\n",
    "if not os.path.exists(MODEL_PAHT):\n",
    "    os.mkdir(MODEL_PAHT)\n",
    "    \n",
    "data_type='train'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "4f746c12-3ba9-4c27-80e6-63d509b01c99",
   "metadata": {},
   "outputs": [],
   "source": [
    "def q1(x):\n",
    "    return x.quantile(0.1)\n",
    "def q2(x):\n",
    "    return x.quantile(0.25)\n",
    "def q3(x):\n",
    "    return x.quantile(0.7)\n",
    "def q4(x):\n",
    "    return x.quantile(0.9)\n",
    "def mode_max(x):\n",
    "    return x.mode().max()\n",
    "def mode_min(x):\n",
    "    return x.mode().min()\n",
    "def mode_mean(x):\n",
    "    return x.mode().mean()\n",
    "def mode_std(x):\n",
    "    return x.mode().std()\n",
    "def mode_len(x):\n",
    "    return x.mode().shape[0]\n",
    "def mode_times(x):\n",
    "    return x.value_counts().iloc[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c4df7843-3ddb-4655-839c-f7abda817070",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "404f669b-d579-457c-b3e6-7e7cc6c87e69",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_data(data_type='train'):\n",
    "    train_base = pd.read_csv(osp.join(DATA_PATH, f'{data_type}_base.csv'))\n",
    "    train_base = train_base.sort_values(['cust_wid']).reset_index(drop=True)\n",
    "    train_view = pd.read_csv(osp.join(DATA_PATH, f'{data_type}_view.csv'), encoding='gbk')\n",
    "    train_trx = pd.read_csv(osp.join(DATA_PATH, f'{data_type}_trx.csv'))\n",
    "\n",
    "    # deal base gdr_cd\n",
    "    train_base['gdr_cd'] = train_base['gdr_cd'].map({'M':1, 'F':2})\n",
    "    train_base['gdr_cd'] = train_base['gdr_cd'].fillna(0)\n",
    "\n",
    "    # deal base cty_cd\n",
    "    train_base['cty_cd'] = train_base['cty_cd'].str.slice(stop=2)\n",
    "    train_base.cty_cd = train_base.cty_cd.fillna('NAN')\n",
    "    cty_map = ['C', '44', '11', 'A', '31', '32', '37', '21', '42', '61', 'B', '33', '50', '12', '35', '36', '51', '43', '62', '53', '34', '23', '65', '41', 'NAN']\n",
    "    cty_map = dict(zip(cty_map, range(1,len(cty_map)+1)))\n",
    "    train_base['cty_cd'] = train_base['cty_cd'].map(cty_map).fillna(0)\n",
    "\n",
    "    # idx = train_view.acs_tm.str.contains('下午')\n",
    "    train_view.acs_tm = train_view.acs_tm.str.rstrip(':').str.replace('上午|下午', '', regex=True).str.replace('1492', '2022')\n",
    "    train_view.acs_tm = pd.to_datetime(train_view.acs_tm)\n",
    "    train_view['day'] = train_view.acs_tm.dt.day\n",
    "\n",
    "    train_trx.trx_tm = train_trx.trx_tm.str.replace('1492', '2022')\n",
    "    train_trx.trx_tm = pd.to_datetime(train_trx.trx_tm)\n",
    "    train_trx['day'] = train_trx.trx_tm.dt.day\n",
    "\n",
    "    train_view['day_diff'] = train_view['day'].max() - train_view['day']\n",
    "    train_trx['day_diff'] = train_trx['day'].max() - train_trx['day']\n",
    "\n",
    "    if data_type=='train':\n",
    "        train_view['day'] -= 4\n",
    "        train_trx['day'] -= 4\n",
    "    elif data_type=='testa':\n",
    "        train_view['day'] -= 3\n",
    "        train_trx['day'] -= 3\n",
    "\n",
    "    print(train_trx['day'].max(), train_trx['day'].min())\n",
    "\n",
    "    train_view['ntimes'] = train_view['day']//14\n",
    "    train_trx['ntimes'] = train_trx['day']//14\n",
    "    train_view['day'] %= 14\n",
    "    train_trx['day'] %= 14\n",
    "\n",
    "#     train_idx = pd.Index(train_base['cust_wid'], name='cust_wid')\n",
    "#     tmp_idx = train_view['acs_tm'].isna()\n",
    "#     train_view_nan = pd.DataFrame({'cust_wid': train_view[tmp_idx]['cust_wid'].unique(), 'view_nan': 1}).set_index('cust_wid').reindex(train_idx).fillna(0).reset_index(drop=True)\n",
    "\n",
    "#     tmp_idx = train_trx['trx_tm'].isna()\n",
    "#     train_trx_nan = pd.DataFrame({'cust_wid': train_trx[tmp_idx]['cust_wid'].unique(), 'trx_nan': 1}).set_index('cust_wid').reindex(train_idx).fillna(0).reset_index(drop=True)\n",
    "\n",
    "#     train_base = pd.concat([train_base, train_view_nan, train_trx_nan], axis=1)\n",
    "\n",
    "\n",
    "\n",
    "    train_view = train_view.query('day>=0').reset_index(drop=True)\n",
    "    train_trx = train_trx.query('day>=0').reset_index(drop=True)\n",
    "\n",
    "\n",
    "    train_view['day'] = train_view['day'].astype('int')\n",
    "    train_trx['day'] = train_trx['day'].astype('int')\n",
    "\n",
    "    train_view = train_view.sort_values(['cust_wid','acs_tm']).reset_index(drop=True)\n",
    "    train_trx = train_trx.sort_values(['cust_wid','trx_tm']).reset_index(drop=True)\n",
    "\n",
    "    page_id_enc = load_obj(osp.join(MODEL_PAHT, 'page_id_enc'))\n",
    "    trx_id_enc = load_obj(osp.join(MODEL_PAHT, 'trx_id_enc'))\n",
    "\n",
    "    train_view['page_enc'] = train_view['page_id'].map(page_id_enc)\n",
    "    train_view['page_enc'] = train_view['page_enc'].fillna(page_id_enc['NAN'])\n",
    "    train_trx['trx_enc'] = train_trx['trx_cd'].map(trx_id_enc)\n",
    "\n",
    "    train_view['acs_tm_diff'] = train_view.groupby('cust_wid')['acs_tm'].diff().dt.total_seconds()\n",
    "    train_trx['trx_tm_diff'] = train_trx.groupby('cust_wid')['trx_tm'].diff().dt.total_seconds()\n",
    "    train_view['daytime'] = train_view['acs_tm'].dt.time.apply(lambda x: x.hour*3600 + x.minute*60 + x.second)\n",
    "    train_trx['daytime'] = train_trx['trx_tm'].dt.time.apply(lambda x: x.hour*3600 + x.minute*60 + x.second)\n",
    "    train_trx['flag'] = (train_trx['trx_amt'] >0).astype('int')\n",
    "\n",
    "    train_idx = pd.Index(train_base['cust_wid'], name='cust_wid')\n",
    "\n",
    "    view_agg_func = {\n",
    "        'day':['mean', 'min', 'max', 'std', np.ptp, 'nunique'],\n",
    "        'page_id':['count', 'nunique'], \n",
    "        'acs_tm_diff':['mean', 'min', 'max', 'std', 'median', 'sum'], \n",
    "        'daytime':['mean', 'min', 'max', 'std', np.ptp],\n",
    "        'page_enc': [\n",
    "            'mean', 'min', 'max', 'std', 'median', 'sum', \n",
    "            np.ptp, \"sem\", \"mad\", \"skew\", pd.Series.kurtosis, \n",
    "            q1, q2, q3, q4, 'last',\n",
    "            # mode_max, mode_min, mode_mean, mode_std, mode_len, mode_times\n",
    "        ],\n",
    "        'day_diff': ['min'],\n",
    "    }\n",
    "    train_view_c = train_view.groupby('cust_wid').agg(view_agg_func).reindex(train_idx).reset_index(drop=True)\n",
    "    train_view_c.columns = [f'vie_{item[0]}_{item[1]}' if item[0] not in ['cust_wid'] else item[0] for item in train_view_c.columns]\n",
    "    train_base = pd.concat([train_base, train_view_c], axis=1)\n",
    "\n",
    "    #     train_view_cn1 = train_view.query('ntimes==1').groupby('cust_wid').agg(view_agg_func).reindex(train_idx).reset_index(drop=True)\n",
    "    #     train_view_cn1.columns = [f'vie_d_{item[0]}_{item[1]}' if item[0] not in ['cust_wid'] else item[0] for item in train_view_cn1.columns]\n",
    "\n",
    "    #     train_view_cn0 = train_view.query('ntimes==0').groupby('cust_wid').agg(view_agg_func).reindex(train_idx).reset_index(drop=True)\n",
    "    #     train_view_cn0.columns = [f'vie_d_{item[0]}_{item[1]}' if item[0] not in ['cust_wid'] else item[0] for item in train_view_cn0.columns]\n",
    "\n",
    "    #     train_view_c = train_view_cn1-train_view_cn0\n",
    "    #     train_base = pd.concat([train_base, train_view_c], axis=1)\n",
    "\n",
    "    trx_agg_func = {\n",
    "        'day':['mean', 'min', 'max', 'std', np.ptp, 'nunique'],\n",
    "        'trx_cd':['count', 'nunique'], \n",
    "        'trx_tm_diff':['mean', 'min', 'max', 'std', 'median', 'sum'], \n",
    "        'daytime':['mean', 'min', 'max', 'std', np.ptp], \n",
    "        'trx_amt':['mean', 'min', 'max', 'std', 'median', 'sum', np.ptp, 'last'],\n",
    "        'trx_enc': ['mean', 'min', 'max', 'std', 'median', 'sum', np.ptp, 'last'],\n",
    "        'day_diff': ['min'],\n",
    "    }\n",
    "    train_trx_c = train_trx.groupby('cust_wid').agg(trx_agg_func).reindex(train_idx).reset_index(drop=True)\n",
    "    train_trx_c.columns = [f'trx_{item[0]}_{item[1]}' if item[0] not in ['cust_wid'] else item[0] for item in train_trx_c.columns]\n",
    "    train_base = pd.concat([train_base, train_trx_c], axis=1)\n",
    "\n",
    "    #     train_trx_c = train_trx.query('flag==1').groupby('cust_wid').agg(trx_agg_func).reindex(train_idx).reset_index(drop=True)\n",
    "    #     train_trx_c.columns = [f'trx_1_{item[0]}_{item[1]}' if item[0] not in ['cust_wid'] else item[0] for item in train_trx_c.columns]\n",
    "    #     train_base = pd.concat([train_base, train_trx_c], axis=1)\n",
    "\n",
    "    #     train_trx_c = train_trx.query('flag==0').groupby('cust_wid').agg(trx_agg_func).reindex(train_idx).reset_index(drop=True)\n",
    "    #     train_trx_c.columns = [f'trx_0_{item[0]}_{item[1]}' if item[0] not in ['cust_wid'] else item[0] for item in train_trx_c.columns]\n",
    "    #     train_base = pd.concat([train_base, train_trx_c], axis=1)\n",
    "\n",
    "    train_view['group'] = (train_view['acs_tm_diff']>5*60).cumsum()\n",
    "    train_view['group'] -= train_view['cust_wid'].map(train_view.groupby('cust_wid').agg({'group':'min'}).to_dict()['group'])\n",
    "    train_reg_base = pd.DataFrame()\n",
    "\n",
    "    train_view_group = train_view.groupby(['cust_wid', 'group']).agg({'page_enc': ['mean', 'min', 'max', 'std'], 'page_id':['count', 'nunique'], 'daytime':['mean',]}).reset_index()\n",
    "    train_view_group.columns = [f'{item[0]}_{item[1]}' if item[0] not in ['cust_wid', 'group'] else item[0] for item in train_view_group.columns]\n",
    "    train_view_group = train_view_group.groupby(['cust_wid']).agg({\n",
    "        'group': ['max'],\n",
    "        'page_enc_mean': ['mean', 'min', 'max', 'std'],\n",
    "        'page_enc_min': ['mean', 'min', 'max', 'std'],\n",
    "        'page_enc_max': ['mean', 'min', 'max', 'std'],\n",
    "        'page_enc_std': ['mean', 'min', 'max', 'std'],\n",
    "        'page_id_count': ['mean', 'min', 'max', 'std'],\n",
    "        'page_id_nunique': ['mean', 'min', 'max', 'std'],\n",
    "        'daytime_mean': ['mean', 'min', 'max', 'std'],\n",
    "    }).reindex(train_idx).reset_index(drop=True)\n",
    "    train_view_group.columns = [f'vie_gop_{item[0]}_{item[1]}' if item[0] not in ['cust_wid'] else item[0] for item in train_view_group.columns]\n",
    "    train_reg_base = pd.concat([train_reg_base,train_view_group], axis=1)\n",
    "    \n",
    "    \n",
    "    train_idx = pd.MultiIndex.from_product([train_base['cust_wid'], range(14)], names=['cust_wid', 'day'])\n",
    "    train_view_cust = train_view.groupby(['cust_wid', 'day'])['page_id'].agg(['count', 'nunique']).reset_index()\n",
    "    train_trx_cust = train_trx.groupby(['cust_wid', 'day'])[['trx_cd', 'trx_amt']].agg({'trx_cd':['count', 'nunique'], 'trx_amt':['min', 'max', 'mean', 'std']}).reset_index()\n",
    "    train_trx_cust.columns = [f'{item[0]}_{item[1]}' if item[0] in ['trx_cd', 'trx_amt'] else item[0] for item in train_trx_cust.columns]\n",
    "\n",
    "    train_view_cust = train_view_cust.set_index(['cust_wid','day']).reindex(train_idx).reset_index().sort_values(['cust_wid','day']).reset_index(drop=True)\n",
    "    train_trx_cust = train_trx_cust.set_index(['cust_wid','day']).reindex(train_idx).reset_index().sort_values(['cust_wid','day']).reset_index(drop=True)\n",
    "\n",
    "    train_view_cust = train_view_cust[[item for item in train_view_cust.columns if item not in ['cust_wid','day']]]\n",
    "    train_view_cust_cols = train_view_cust.columns\n",
    "    train_view_cust = train_view_cust.values.reshape(train_base['cust_wid'].shape[0], 14*len(train_view_cust_cols))\n",
    "    train_view_cust = pd.DataFrame(train_view_cust, columns=[f\"view_cust_{item}_{day}\" for day in range(14) for item in train_view_cust_cols])\n",
    "\n",
    "    train_trx_cust = train_trx_cust[[item for item in train_trx_cust.columns if item not in ['cust_wid','day']]]\n",
    "    train_trx_cust_cols = train_trx_cust.columns\n",
    "    train_trx_cust = train_trx_cust.values.reshape(train_base['cust_wid'].shape[0], 14*len(train_trx_cust_cols))\n",
    "    train_trx_cust = pd.DataFrame(train_trx_cust, columns=[f\"trx_cust_{item}_{day}\" for day in range(14) for item in train_trx_cust_cols])\n",
    "    train_reg_base = pd.concat([train_reg_base, train_view_cust, train_trx_cust], axis=1)\n",
    "\n",
    "    return train_base, train_reg_base"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "1cb222ec-16b3-400e-b20b-41e5c1a54588",
   "metadata": {},
   "outputs": [],
   "source": [
    "# train_view_c = train_view.groupby(['cust_wid', 'day']).agg(view_agg_func)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "dd99774b-115d-4818-8b28-a4f955d32bf7",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 众数"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "60916458-da7e-41c6-a2de-15e523d18716",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "eb7b2792-0a9f-438f-b30c-08514eade7aa",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "27.0 -3.0\n",
      "27.0 -2.0\n"
     ]
    }
   ],
   "source": [
    "train_base, train_reg_base = get_data(data_type='train')\n",
    "testA_base, testA_reg_base = get_data(data_type='testa')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3c219a90-9f8d-44b9-9c43-bb8887645b6d",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cc6fa778-0d60-4476-b439-59ef77cf8c14",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "bf6c9767-428a-4a5d-9a56-c964251ec209",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "75\n"
     ]
    }
   ],
   "source": [
    "FEATS = [item for item in train_base.columns if item not in ['cust_wid', 'label']]\n",
    "TARGET_COL = 'label'\n",
    "print(len(FEATS))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "cd91e96e-f5d5-4d73-a609-08e6d81e9e79",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['age', 'gdr_cd', 'cty_cd', 'vie_day_mean', 'vie_day_min', 'vie_day_max', 'vie_day_std', 'vie_day_ptp', 'vie_day_nunique', 'vie_page_id_count', 'vie_page_id_nunique', 'vie_acs_tm_diff_mean', 'vie_acs_tm_diff_min', 'vie_acs_tm_diff_max', 'vie_acs_tm_diff_std', 'vie_acs_tm_diff_median', 'vie_acs_tm_diff_sum', 'vie_daytime_mean', 'vie_daytime_min', 'vie_daytime_max', 'vie_daytime_std', 'vie_daytime_ptp', 'vie_page_enc_mean', 'vie_page_enc_min', 'vie_page_enc_max', 'vie_page_enc_std', 'vie_page_enc_median', 'vie_page_enc_sum', 'vie_page_enc_ptp', 'vie_page_enc_sem', 'vie_page_enc_mad', 'vie_page_enc_skew', 'vie_page_enc_kurt', 'vie_page_enc_q1', 'vie_page_enc_q2', 'vie_page_enc_q3', 'vie_page_enc_q4', 'vie_page_enc_last', 'vie_day_diff_min', 'trx_day_mean', 'trx_day_min', 'trx_day_max', 'trx_day_std', 'trx_day_ptp', 'trx_day_nunique', 'trx_trx_cd_count', 'trx_trx_cd_nunique', 'trx_trx_tm_diff_mean', 'trx_trx_tm_diff_min', 'trx_trx_tm_diff_max', 'trx_trx_tm_diff_std', 'trx_trx_tm_diff_median', 'trx_trx_tm_diff_sum', 'trx_daytime_mean', 'trx_daytime_min', 'trx_daytime_max', 'trx_daytime_std', 'trx_daytime_ptp', 'trx_trx_amt_mean', 'trx_trx_amt_min', 'trx_trx_amt_max', 'trx_trx_amt_std', 'trx_trx_amt_median', 'trx_trx_amt_sum', 'trx_trx_amt_ptp', 'trx_trx_amt_last', 'trx_trx_enc_mean', 'trx_trx_enc_min', 'trx_trx_enc_max', 'trx_trx_enc_std', 'trx_trx_enc_median', 'trx_trx_enc_sum', 'trx_trx_enc_ptp', 'trx_trx_enc_last', 'trx_day_diff_min']\n"
     ]
    }
   ],
   "source": [
    "print(FEATS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "406f13d0-58f8-4a95-8534-a142d161efe6",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "class cv_model():\n",
    "    def __init__(self, task_name='lgb', nfold=5, seed=SEED, save_path=MODEL_PAHT):\n",
    "        self.nfold = nfold\n",
    "        self.seed = seed\n",
    "        self.save_path = save_path\n",
    "        self.models = []\n",
    "        self.task_name = task_name\n",
    "    \n",
    "    def save_model(self):\n",
    "        for i, model in enumerate(self.models):\n",
    "            save_obj(model, os.path.join(self.save_path, f\"{self.task_name}_fold_{i}\"))\n",
    "    \n",
    "    def load_model(self):\n",
    "        self.models = []\n",
    "        for i in range(self.nfold):\n",
    "            model = load_obj(os.path.join(self.save_path, f\"{self.task_name}_fold_{i}\"))\n",
    "            self.models.append(model)\n",
    "    \n",
    "    def fit(self, params, train_x, train_y, cat_feats):\n",
    "        kf = StratifiedKFold(n_splits=self.nfold, random_state=self.seed, shuffle=True)\n",
    "        feat_imp = pd.DataFrame({'feat': train_x.columns})\n",
    "        train_oof = np.zeros(len(train_x))\n",
    "        for fold, (trn_idx, val_idx) in enumerate(kf.split(train_x, train_y)):\n",
    "            trn_data = lgb.Dataset(train_x.iloc[trn_idx], label=train_y.iloc[trn_idx], categorical_feature=cat_feats)\n",
    "            val_data = lgb.Dataset(train_x.iloc[val_idx], label=train_y.iloc[val_idx], categorical_feature=cat_feats)\n",
    "            model = lgb.train(\n",
    "                params,\n",
    "                trn_data,\n",
    "                num_boost_round=30000,\n",
    "                valid_sets=[trn_data, val_data],\n",
    "                categorical_feature=cat_feats,\n",
    "                callbacks=[\n",
    "                    lgb.early_stopping(100),\n",
    "                    lgb.log_evaluation(100)\n",
    "                ]\n",
    "            )\n",
    "            train_oof[val_idx] = model.predict(train_x.iloc[val_idx], num_iteration=model.best_iteration)\n",
    "            self.models.append(model)\n",
    "            feat_imp[f'fold_{fold}'] = feat_imp['feat'].map(dict(zip(model.feature_name(), model.feature_importance())))\n",
    "        return train_oof, feat_imp\n",
    "    def predict(self, test_x):\n",
    "        test_pred = np.zeros(len(test_x))\n",
    "        for i, model in enumerate(self.models):\n",
    "            test_pred+=model.predict(test_x, num_iteration=model.best_iteration)/len(self.models)\n",
    "        return test_pred\n",
    "                                                         \n",
    "            \n",
    "            \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "037ad870-9365-41e0-8c76-3c76d667f4b0",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "93ab10ef-3a9f-4570-b7d6-c5d32bbb6f10",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(100000, 75) (100000,) (50000, 75)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's average_precision: 0.913401\tvalid_1's average_precision: 0.914298\n",
      "[200]\ttraining's average_precision: 0.920972\tvalid_1's average_precision: 0.915552\n",
      "[300]\ttraining's average_precision: 0.926658\tvalid_1's average_precision: 0.915436\n",
      "Early stopping, best iteration is:\n",
      "[231]\ttraining's average_precision: 0.922605\tvalid_1's average_precision: 0.915715\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's average_precision: 0.913825\tvalid_1's average_precision: 0.911936\n",
      "[200]\ttraining's average_precision: 0.921798\tvalid_1's average_precision: 0.913212\n",
      "[300]\ttraining's average_precision: 0.927373\tvalid_1's average_precision: 0.913353\n",
      "Early stopping, best iteration is:\n",
      "[264]\ttraining's average_precision: 0.925261\tvalid_1's average_precision: 0.913593\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's average_precision: 0.913573\tvalid_1's average_precision: 0.911731\n",
      "[200]\ttraining's average_precision: 0.921434\tvalid_1's average_precision: 0.913668\n",
      "[300]\ttraining's average_precision: 0.927346\tvalid_1's average_precision: 0.914348\n",
      "[400]\ttraining's average_precision: 0.93173\tvalid_1's average_precision: 0.915069\n",
      "[500]\ttraining's average_precision: 0.93558\tvalid_1's average_precision: 0.915509\n",
      "[600]\ttraining's average_precision: 0.939529\tvalid_1's average_precision: 0.915376\n",
      "Early stopping, best iteration is:\n",
      "[511]\ttraining's average_precision: 0.936011\tvalid_1's average_precision: 0.91558\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's average_precision: 0.913913\tvalid_1's average_precision: 0.910967\n",
      "[200]\ttraining's average_precision: 0.921502\tvalid_1's average_precision: 0.912329\n",
      "[300]\ttraining's average_precision: 0.927203\tvalid_1's average_precision: 0.911635\n",
      "Early stopping, best iteration is:\n",
      "[207]\ttraining's average_precision: 0.921882\tvalid_1's average_precision: 0.912403\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's average_precision: 0.915313\tvalid_1's average_precision: 0.900581\n",
      "[200]\ttraining's average_precision: 0.922387\tvalid_1's average_precision: 0.901754\n",
      "[300]\ttraining's average_precision: 0.927572\tvalid_1's average_precision: 0.902414\n",
      "[400]\ttraining's average_precision: 0.932515\tvalid_1's average_precision: 0.903301\n",
      "[500]\ttraining's average_precision: 0.936532\tvalid_1's average_precision: 0.903135\n",
      "Early stopping, best iteration is:\n",
      "[400]\ttraining's average_precision: 0.932515\tvalid_1's average_precision: 0.903301\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's average_precision: 0.913835\tvalid_1's average_precision: 0.911557\n",
      "[200]\ttraining's average_precision: 0.921295\tvalid_1's average_precision: 0.913611\n",
      "[300]\ttraining's average_precision: 0.926695\tvalid_1's average_precision: 0.913967\n",
      "[400]\ttraining's average_precision: 0.931463\tvalid_1's average_precision: 0.913994\n",
      "Early stopping, best iteration is:\n",
      "[310]\ttraining's average_precision: 0.927408\tvalid_1's average_precision: 0.914114\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's average_precision: 0.915083\tvalid_1's average_precision: 0.898897\n",
      "[200]\ttraining's average_precision: 0.922282\tvalid_1's average_precision: 0.900954\n",
      "[300]\ttraining's average_precision: 0.928107\tvalid_1's average_precision: 0.90189\n",
      "Early stopping, best iteration is:\n",
      "[290]\ttraining's average_precision: 0.927657\tvalid_1's average_precision: 0.90193\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's average_precision: 0.913783\tvalid_1's average_precision: 0.908822\n",
      "[200]\ttraining's average_precision: 0.921754\tvalid_1's average_precision: 0.911026\n",
      "[300]\ttraining's average_precision: 0.927004\tvalid_1's average_precision: 0.911701\n",
      "[400]\ttraining's average_precision: 0.931832\tvalid_1's average_precision: 0.911795\n",
      "[500]\ttraining's average_precision: 0.935838\tvalid_1's average_precision: 0.912323\n",
      "[600]\ttraining's average_precision: 0.939534\tvalid_1's average_precision: 0.912219\n",
      "Early stopping, best iteration is:\n",
      "[534]\ttraining's average_precision: 0.937189\tvalid_1's average_precision: 0.912358\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's average_precision: 0.913794\tvalid_1's average_precision: 0.912679\n",
      "[200]\ttraining's average_precision: 0.921547\tvalid_1's average_precision: 0.913768\n",
      "[300]\ttraining's average_precision: 0.927436\tvalid_1's average_precision: 0.914136\n",
      "[400]\ttraining's average_precision: 0.931826\tvalid_1's average_precision: 0.914124\n",
      "Early stopping, best iteration is:\n",
      "[332]\ttraining's average_precision: 0.928803\tvalid_1's average_precision: 0.914198\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's average_precision: 0.913416\tvalid_1's average_precision: 0.913862\n",
      "[200]\ttraining's average_precision: 0.920982\tvalid_1's average_precision: 0.914996\n",
      "Early stopping, best iteration is:\n",
      "[180]\ttraining's average_precision: 0.919737\tvalid_1's average_precision: 0.915058\n"
     ]
    }
   ],
   "source": [
    "cla_params = {\n",
    "    'objective': 'binary', # multiclass multiclassova\n",
    "    'metric': 'average_precision', #  average_precision auc map\n",
    "    'boosting_type': 'gbdt',\n",
    "    'num_leaves': 16,\n",
    "    'min_data_in_leaf': 10,\n",
    "    'learning_rate': 0.05,\n",
    "    'feature_fraction': 0.9,\n",
    "    'bagging_fraction': 0.8,\n",
    "    'bagging_freq': 5,\n",
    "    'verbose': -1,\n",
    "    'feature_fraction_seed':SEED,\n",
    "    'bagging_seed':SEED,\n",
    "    'seed': SEED,\n",
    "}\n",
    "FEATS = [item for item in train_base.columns if item not in ['cust_wid', 'label']]\n",
    "train_cla_x = train_base[FEATS]\n",
    "train_cla_y = (train_base[TARGET_COL]>0).astype('int')\n",
    "testA_cla_x = testA_base[FEATS]\n",
    "print(train_cla_x.shape, train_cla_y.shape, testA_cla_x.shape)\n",
    "\n",
    "cla_model = cv_model(task_name='cla_lgb', nfold=10)\n",
    "train_cla_oof, feat_cla_imp = cla_model.fit(cla_params, train_cla_x, train_cla_y, cat_feats=['gdr_cd','cty_cd'])\n",
    "cla_model.save_model()\n",
    "cla_model.load_model()\n",
    "test_cla_pred = cla_model.predict(testA_cla_x)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "220c6262-8e3e-4a5c-9011-d1dcabb53b6d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "以精度为0.1在[0.1,0.8]范围内搜索F1最佳阈值>>最佳阈值:=0.2->F1=0.8644084\n",
      "以精度为0.02在[0.1,0.28]范围内搜索F1最佳阈值>>最佳阈值:=0.18000000000000002->F1=0.86580652\n",
      "以精度为0.004在[0.13,0.2260000000000001]范围内搜索F1最佳阈值>>最佳阈值:=0.17400000000000004->F1=0.86609753\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.17400000000000004, 0.86609753, 0.56041)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "thr_optimal, best_fscore = utils.get_optimal_Fscore(train_cla_y, train_cla_oof)\n",
    "thr_optimal, best_fscore, (train_cla_oof>thr_optimal).sum() / train_cla_oof.shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "a5afcca5-15c8-4fdc-85e0-ebe0158c5f9e",
   "metadata": {},
   "outputs": [],
   "source": [
    "thr_optimal = 0.165"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "a14edec9-1f6c-4e50-ad14-65cbb4b0f1a7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "19157"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(test_cla_pred<=0.165).sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6e857563-5262-4850-bb70-49efe985a94f",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "c8ca300f-6d77-47c7-b03e-45401a99cbdd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f6af26e6a90>"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x450 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(10,4.5))\n",
    "plt.subplot(121)\n",
    "plt.hist(train_cla_oof, bins=100, alpha=0.5, label='train 1', density=True)\n",
    "plt.hist(test_cla_pred, bins=100, alpha=0.5, label='testA', density=True)\n",
    "plt.vlines(thr_optimal, 0, 6, color='r')\n",
    "plt.legend()\n",
    "\n",
    "plt.subplot(122)\n",
    "idx = train_cla_y==1\n",
    "plt.hist(train_cla_oof[idx], bins=100, alpha=0.5, label='train 1', density=True)\n",
    "plt.hist(train_cla_oof[~idx], bins=100, alpha=0.5, label='train 0', density=True)\n",
    "plt.legend()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "c3176255-f1ae-4602-a909-0955a10ddfd2",
   "metadata": {},
   "outputs": [],
   "source": [
    "feat_cla_imp['mean'] = feat_cla_imp[[f\"fold_{fold}\" for fold in range(cla_model.nfold)]].mean(axis=1)\n",
    "feat_cla_imp = feat_cla_imp.sort_values(by=\"mean\", ascending=False)\n",
    "feat_cla_imp.to_csv('model/feats_cla_imp.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "a8328d7f-2b9d-4489-922a-ace44ebee2ef",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/scipy/stats/_stats_py.py:5218: RuntimeWarning: overflow encountered in long_scalars\n",
      "  (2 * xtie * ytie) / m + x0 * y0 / (9 * m * (size - 2)))\n"
     ]
    }
   ],
   "source": [
    "corr_data = all_corr_data(train_cla_x,train_cla_y)\n",
    "corr_data.to_csv('model/corr_cla.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "bf5fe9ce-2aca-4422-bb88-cd7a68a501ac",
   "metadata": {},
   "outputs": [],
   "source": [
    "test_cla_pred = (test_cla_pred>=thr_optimal).astype('int')\n",
    "train_cla_oof = (train_cla_oof>=thr_optimal).astype('int')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "75f653ac-e832-4cd7-9d0d-df0a45ac5d0e",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "48fd51c1-0f8a-45fa-87d1-507f27b760e4",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "35b0ebaf-794d-4ada-8707-cc5b8b0884e2",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(38734, 216) (38734,) (50000, 216)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's mape: 0.630794\tvalid_1's mape: 0.639902\n",
      "Early stopping, best iteration is:\n",
      "[80]\ttraining's mape: 0.633194\tvalid_1's mape: 0.63968\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's mape: 0.631296\tvalid_1's mape: 0.640276\n",
      "[200]\ttraining's mape: 0.620996\tvalid_1's mape: 0.639696\n",
      "[300]\ttraining's mape: 0.614408\tvalid_1's mape: 0.639768\n",
      "Early stopping, best iteration is:\n",
      "[209]\ttraining's mape: 0.620159\tvalid_1's mape: 0.639412\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's mape: 0.630494\tvalid_1's mape: 0.64212\n",
      "Early stopping, best iteration is:\n",
      "[78]\ttraining's mape: 0.633047\tvalid_1's mape: 0.64173\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's mape: 0.631561\tvalid_1's mape: 0.64148\n",
      "[200]\ttraining's mape: 0.621898\tvalid_1's mape: 0.641672\n",
      "Early stopping, best iteration is:\n",
      "[140]\ttraining's mape: 0.627482\tvalid_1's mape: 0.64126\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's mape: 0.630575\tvalid_1's mape: 0.639613\n",
      "[200]\ttraining's mape: 0.620706\tvalid_1's mape: 0.639849\n",
      "Early stopping, best iteration is:\n",
      "[126]\ttraining's mape: 0.62799\tvalid_1's mape: 0.639442\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's mape: 0.631169\tvalid_1's mape: 0.642253\n",
      "[200]\ttraining's mape: 0.621358\tvalid_1's mape: 0.641785\n",
      "[300]\ttraining's mape: 0.613415\tvalid_1's mape: 0.641911\n",
      "Early stopping, best iteration is:\n",
      "[214]\ttraining's mape: 0.619974\tvalid_1's mape: 0.641585\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's mape: 0.63225\tvalid_1's mape: 0.639194\n",
      "Early stopping, best iteration is:\n",
      "[98]\ttraining's mape: 0.632492\tvalid_1's mape: 0.63913\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's mape: 0.631395\tvalid_1's mape: 0.641211\n",
      "Early stopping, best iteration is:\n",
      "[69]\ttraining's mape: 0.633857\tvalid_1's mape: 0.641044\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's mape: 0.630491\tvalid_1's mape: 0.639362\n",
      "Early stopping, best iteration is:\n",
      "[83]\ttraining's mape: 0.632417\tvalid_1's mape: 0.639036\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/opt/conda/lib/python3.9/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training until validation scores don't improve for 100 rounds\n",
      "[100]\ttraining's mape: 0.630387\tvalid_1's mape: 0.642283\n",
      "[200]\ttraining's mape: 0.621123\tvalid_1's mape: 0.641194\n",
      "[300]\ttraining's mape: 0.61438\tvalid_1's mape: 0.641535\n",
      "Early stopping, best iteration is:\n",
      "[250]\ttraining's mape: 0.61834\tvalid_1's mape: 0.640874\n"
     ]
    }
   ],
   "source": [
    "reg_params = {\n",
    "    'objective': 'mape', # multiclass multiclassova\n",
    "    'metric': 'mape', #  average_precision auc\n",
    "    'boosting_type': 'gbdt',\n",
    "    'num_leaves': 16,\n",
    "    'min_data_in_leaf': 10,\n",
    "    'learning_rate': 0.05,\n",
    "    'feature_fraction': 0.9,\n",
    "    'bagging_fraction': 0.8,\n",
    "    'bagging_freq': 5,\n",
    "    'verbose': -1,\n",
    "    'feature_fraction_seed':SEED,\n",
    "    'bagging_seed':SEED,\n",
    "    'seed': SEED,\n",
    "}\n",
    "train_base = pd.concat([train_base, train_reg_base], axis=1)\n",
    "testA_base = pd.concat([testA_base, testA_reg_base], axis=1)\n",
    "\n",
    "FEATS = [item for item in train_base.columns if item not in ['cust_wid', 'label']]\n",
    "idx = (train_base[TARGET_COL]>0) & (train_cla_oof>0)\n",
    "train_reg_x = train_base.loc[idx, FEATS]\n",
    "train_reg_y = train_base.loc[idx, TARGET_COL]\n",
    "testA_reg_x = testA_base[FEATS]\n",
    "print(train_reg_x.shape, train_reg_y.shape, testA_reg_x.shape)\n",
    "\n",
    "reg_model = cv_model(task_name='reg_lgb', nfold=10)\n",
    "train_reg_oof, feat_reg_imp = reg_model.fit(reg_params, train_reg_x, train_reg_y, cat_feats=['gdr_cd','cty_cd'])\n",
    "reg_model.save_model()\n",
    "reg_model.load_model()\n",
    "test_reg_pred = reg_model.predict(testA_reg_x)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "b6803258-24bc-41e5-aa57-2087ba34f953",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f6af0480b20>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAzIAAAGGCAYAAACg8bRnAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjYuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/P9b71AAAACXBIWXMAAA9hAAAPYQGoP6dpAAA9OklEQVR4nO3deXwV9b3/8fchgRNAErasEjZlkS1EUAxogbKEQLmCV6Apyqa0tXB/0jywmtayqJeDdQEtFNTL1iuIYiH4UAxCJOTKJgGiQBEhLIGaBEGTkKABk/n9YTkSSEJOknPmzMnr+XjMQ2bOLJ/vOJk57zObzTAMQwAAAABgIfXMLgAAAAAAXEWQAQAAAGA5BBkAAAAAlkOQAQAAAGA5BBkAAAAAlkOQAQAAAGA5BBkAAAAAlkOQAQAAAGA5/p5eYGlpqb766is1adJENpvN04sHgDrLMAxdvHhRERERqleP37Gu4rgEAOapybHJ40Hmq6++UmRkpKcXCwD4tzNnzqhVq1Zml+E1OC4BgPmqc2zyeJBp0qSJpB+LDQwM9PTiAaDOKigoUGRkpHM/jB9xXAIA89Tk2OTxIHP1tH1gYCAHDAAwAZdPlcVxCQDMV51jExdJAwAAALAcggwAAAAAyyHIAAAAALAcj98jAwDXKykp0ZUrV8wuw/Lq168vPz8/s8vwWWyn7sF2C6C6CDIATGMYhnJycpSXl2d2KT6jadOmCgsL44b+WsR26n5stwCqgyADwDRXvxyGhISoUaNGfImpAcMwdOnSJZ07d06SFB4ebnJFvoPt1H3YbgHUhEtBpm3btjp9+vQNw3/3u99p8eLFtVYUAN9XUlLi/HLYokULs8vxCQ0bNpQknTt3TiEhIVyuUwvYTt2P7RZAdbkUZPbu3auSkhJn/6FDhzRkyBCNGTOm1gsD4Nuu3mvQqFEjkyvxLVfX55UrV/hCWAvYTj2D7RZAdbgUZIKDg8v0z58/X7fddpv69+9fq0UBqDu4TKd2sT7dg/XqXqxfANVR7XtkLl++rDfffFMJCQmV7oCKi4tVXFzs7C8oKKjuIgEAAABAUg3eI5OUlKS8vDxNmjSp0vEcDoeCgoKcXWRkZHUXCQA+p23btlq4cKHZZQAAYDnVPiOzbNkyxcXFKSIiotLxEhMTlZCQ4OwvKCggzACo0IItX3p0eb8f0tHlaQYMGKCePXvWSgDZu3evGjduXOP5wPO8fVutze30qkmTJikvL09JSUmVjpeWlqYXXnhB+/btU3Z2tjZs2KBRo0bVWh0AIFUzyJw+fVpbt27V+vXrbzqu3W6X3W6vzmIAwJIMw1BJSYn8/W++i73+3kPAFxQVFSkqKkpTpkzRAw88YHY5AHxUtS4tW7FihUJCQjRixIjargdAbdjm+LFDrZs0aZK2b9+uV155RTabTTabTStXrpTNZtOHH36oXr16yW6365NPPlFmZqbuv/9+hYaG6pZbbtFdd92lrVu3lpnf9ZeW2Ww2/c///I9Gjx6tRo0aqUOHDnrvvfc83EpYXXnb6alTpyT9+MTRuLg43XLLLQoNDdXDDz+s8+fPO6d999131b17dzVs2FAtWrTQ4MGDVVRUpDlz5mjVqlXauHGjc56pqanlLj8uLk7PPfecRo8e7YHWAj6M43mlXA4ypaWlWrFihSZOnFilXxsBwJe88soriomJ0dSpU5Wdna3s7Gzn5bJPPfWU5s+fryNHjqhHjx4qLCzU8OHDlZKSogMHDmjYsGEaOXKksrKyKl3G3LlzNXbsWH3++ecaPny4xo8fr2+++cYTzYOPqGg7zcvL089//nNFR0crPT1dycnJys3N1dixYyVJ2dnZio+P15QpU3TkyBGlpqbqgQcekGEYmjlzpsaOHathw4Y559m3b1+TWwqgLnM5iWzdulVZWVmaMmWKO+oBAK8WFBSkBg0aqFGjRgoLC5MkffHFF5KkZ555RkOGDHGO27x5c0VFRTn7n332WW3YsEHvvfeepk+fXuEyJk2apPj4eEnSvHnz9Oqrr+rTTz/VsGHD3NEk+KDytlNJWrRokaKjozVv3jznsOXLlysyMlJffvmlCgsL9cMPP+iBBx5QmzZtJEndu3d3jtuwYUMVFxeXmScAmMXlIDN06FAZhuGOWgDA0nr37l2mv7CwUHPmzNEHH3yg7Oxs/fDDD/ruu+9uekamR48ezn83btxYgYGBOnfunFtqRt3y2Wefadu2bbrllltu+CwzM1NDhw7VoEGD1L17d8XGxmro0KF68MEH1axZMxOqBYDKcW0YANSS658+NnPmTG3ZskUvvviibr/9djVs2FAPPvigLl++XOl86tevX6bfZrOptLS01utF3VNYWKiRI0fq+eefv+Gz8PBw+fn5acuWLdq5c6c++ugj/fWvf9Wf/vQn7dmzR+3atTOhYgCoGEEGAFzUoEEDlZSU3HS8HTt2aNKkSc4bngsLC503XAPuVt52euedd+of//iH2rZtW+F9rjabTf369VO/fv00a9YstWnTRhs2bFBCQkKVt30A8IRqvxATAOqqtm3bas+ePTp16pTOnz9f4dmSDh06aP369crIyNBnn32mX/3qV5xZgceUt51OmzZN33zzjeLj47V3715lZmZq8+bNmjx5skpKSrRnzx7NmzdP6enpysrK0vr16/X111/rjjvucM7z888/19GjR3X+/HlduXKl3GUXFhYqIyNDGRkZkqSTJ08qIyPjppdVAoArOCMDwKtU5wWVnjZz5kxNnDhRXbp00XfffacVK1aUO97LL7+sKVOmqG/fvmrZsqWefPJJFRQUeLhauIu3b6vXb6cnT55U27ZttWPHDj355JMaOnSoiouL1aZNGw0bNkz16tVTYGCg0tLStHDhQhUUFKhNmzZ66aWXFBcXJ0maOnWqUlNT1bt3bxUWFmrbtm0aMGDADctOT0/XwIEDnf1XX4w9ceJErVy50hPNB1AH2AwP37lfUFCgoKAg5efnKzAw0JOLBuqOq8+cH5hobh2V+P7773Xy5Em1a9dOAQEBZpfjMypbr+x/y1fZemE79QzWM1ABCxzPa6omxybOyAC+5PqXZtWBHSAAAKibuEcGAAAAgOUQZAAAAABYDkEGAAAAgOUQZAAAAABYDkEGAAAAgOUQZAAAAABYDkEGAAAAgOUQZAAAAABYDkEGqCu2OW58YSaqZcCAAZoxY0atzW/SpEkaNWpUhZ//5je/kZ+fn9atW1dryzRDWlqaRo4cqYiICNlsNiUlJZX53Gazldu98MILFc5zzpw5N4zfuXNnN7fEGmp7O5Vuvq1ea/HixWrbtq0CAgLUp08fffrpp7VaCwD4m10AAJTh6bA1MNGzy3PRpUuXtHbtWv3hD3/Q8uXLNWbMGLNLqraioiJFRUVpypQpeuCBB274PDs7u0z/hx9+qEceeUT/+Z//Wel8u3btqq1btzr7/f09dGhjW63Q22+/rYSEBC1dulR9+vTRwoULFRsbq6NHjyokJMTs8gD4CM7IAIALJk2apO3bt+uVV15xngE4deqUDh06pLi4ON1yyy0KDQ3Vww8/rPPnzzune/fdd9W9e3c1bNhQLVq00ODBg1VUVKQ5c+Zo1apV2rhxo3N+qampzunWrVunLl266KmnnlJaWprOnDljQqtrR1xcnJ577jmNHj263M/DwsLKdBs3btTAgQPVvn37Sufr7+9fZrqWLVu6o3xLqWg7leS2bfVaL7/8sqZOnarJkyerS5cuWrp0qRo1aqTly5d7oPUA6gqCDAC44JVXXlFMTIymTp2q7OxsZWdnq0mTJvr5z3+u6OhopaenKzk5Wbm5uRo7dqykH880xMfHa8qUKTpy5IhSU1P1wAMPyDAMzZw5U2PHjtWwYcOc8+vbt69zecuWLdNDDz2koKAgxcXFaeXKlSa13LNyc3P1wQcf6JFHHrnpuMeOHVNERITat2+v8ePHKysrq9Lxi4uLVVBQUKbzNeVtp5GRkcrLy3PbtnrV5cuXtW/fPg0ePNg5rF69eho8eLB27drlsXUAwPdxaRkAuCAoKEgNGjRQo0aNFBYWJkl67rnnFB0drXnz5jnHW758uSIjI/Xll1+qsLBQP/zwgx544AG1adNGktS9e3fnuA0bNlRxcbFzflcdO3ZMu3fv1vr16yVJDz30kBISEvT000/LZrO5u6mmWrVqlZo0aVLuJWjX6tOnj1auXKlOnTopOztbc+fO1X333adDhw6pSZMm5U7jcDg0d+5cd5TtNcrbTiVp0aJFbtlWr3X+/HmVlJQoNDS0zPDQ0FB98cUXtdVEAOCMDADU1GeffaZt27bplltucXZXbzjPzMxUVFSUBg0apO7du2vMmDF644039O233950vsuXL1dsbKzzUqnhw4crPz9fH3/8sVvb4w2WL1+u8ePHKyAgoNLx4uLiNGbMGPXo0UOxsbHatGmT8vLy9M4771Q4TWJiovLz852dlS/Xc5W7tlUAMANnZACghgoLCzVy5Eg9//zzN3wWHh4uPz8/bdmyRTt37tRHH32kv/71r/rTn/6kPXv2qF27duXOs6SkRKtWrVJOTk6Zm9dLSkq0fPlyDRo0yG3tMdv//d//6ejRo3r77bddnrZp06bq2LGjjh8/XuE4drtddru9JiValju21eu1bNlSfn5+ys3NLTM8Nze30jM5AOAqzsgAgIsaNGigkpISZ/+dd96pw4cPq23btrr99tvLdI0bN5b046OF+/Xrp7lz5+rAgQNq0KCBNmzYUO78JGnTpk26ePGiDhw4oIyMDGf31ltvaf369crLy/NYez1t2bJl6tWrl6KiolyetrCwUJmZmQoPD3dDZdZS3nbljm21vOX26tVLKSkpzmGlpaVKSUlRTExMLbcSQF1GkAEAF7Vt21Z79uzRqVOndP78eU2bNk3ffPON4uPjtXfvXmVmZmrz5s2aPHmySkpKtGfPHs2bN0/p6enKysrS+vXr9fXXX+uOO+5wzu/zzz/X0aNHdf78eV25ckXLli3TiBEjFBUVpW7dujm7sWPHqmnTplq9erXJa8F1hYWFzkAmSSdPnlRGRkaZm/MLCgq0bt06Pfroo+XOY9CgQVq0aJGzf+bMmdq+fbtOnTqlnTt3avTo0fLz81N8fLxb22IF12+npaWlbtlWy5OQkKA33nhDq1at0pEjR/TYY4+pqKhIkydP9uQqAODjCDIA4KKZM2fKz89PXbp0UXBwsC5fvqwdO3aopKREQ4cOVffu3TVjxgw1bdpU9erVU2BgoNLS0jR8+HB17NhRTz/9tF566SXFxcVJkqZOnapOnTqpd+/eCg4O1vr16/XBBx+U+/6UevXqafTo0Vq2bJmnm11j6enpio6OVnR0tKQfv+xGR0dr1qxZznHWrl0rwzAqDCKZmZllHhV89uxZxcfHq1OnTho7dqxatGih3bt3Kzg42L2NsYDrt9OsrCxFRETU6ra6Y8eOcpc9btw4vfjii5o1a5Z69uypjIwMJScn3/AAAACoCZthGIYnF1hQUKCgoCDl5+crMDDQk4sGfF9FL+gbmPjTZ17yUr3vv/9eJ0+eVLt27W56QzeqrrL1yv63fJWtF7ZTz2A9AxXwsmO3O9Tk2MQZGQAAAACWQ5ABAAAAYDkEGQAAAACWQ5ABAAAAYDkEGQAAAACWQ5ABYCoPPzjR57E+3YP16l6sXwDVQZABYIr69etLki5dumRyJb7l6vq8un5RM2ynnsF2C6A6/M0uAEDd5Ofnp6ZNm+rcuXOSpEaNGslms5lclXUZhqFLly7p3Llzatq0qfz8/MwuySewnboX2y2AmiDIAL6gohdhermwsDBJcn5JRM01bdrUuV5RO9hO3Y/tFkB1EGQAmMZmsyk8PFwhISG6cuWK2eVYXv369flF2w3YTt2L7RZAdRFkAJjOz8+PLzLwemynAOBdXL7Z/1//+pceeughtWjRQg0bNlT37t2Vnp7ujtoAAAAAoFwunZH59ttv1a9fPw0cOFAffvihgoODdezYMTVr1sxd9QGoDRa9hwYAAKAiLgWZ559/XpGRkVqxYoVzWLt27Wq9KAAAAACojEuXlr333nvq3bu3xowZo5CQEEVHR+uNN95wV20AAAAAUC6XgsyJEye0ZMkSdejQQZs3b9Zjjz2m//f//p9WrVpV4TTFxcUqKCgo0wEAAABATbh0aVlpaal69+6tefPmSZKio6N16NAhLV26VBMnTix3GofDoblz59a8UgAAAAD4N5fOyISHh6tLly5lht1xxx3KysqqcJrExETl5+c7uzNnzlSvUgAAAAD4N5fOyPTr109Hjx4tM+zLL79UmzZtKpzGbrfLbrdXrzoAAAAAKIdLZ2R+//vfa/fu3Zo3b56OHz+uNWvW6PXXX9e0adPcVR8AAAAA3MClIHPXXXdpw4YNeuutt9StWzc9++yzWrhwocaPH++u+gAAAADgBi5dWiZJv/jFL/SLX/zCHbUAAAAAQJW4dEYGAAAAALwBQQYAAACA5RBkAAAAAFgOQQYAAACA5RBkAAAAAFgOQQYAAACA5RBkAAAAAFgOQQYAAACA5RBkAAAAAFgOQQYAAACA5RBkAAAAAFgOQQYA4BFpaWkaOXKkIiIiZLPZlJSUVObzSZMmyWazlemGDRt20/kuXrxYbdu2VUBAgPr06aNPP/3UTS0AAHgTggwAwCOKiooUFRWlxYsXVzjOsGHDlJ2d7ezeeuutSuf59ttvKyEhQbNnz9b+/fsVFRWl2NhYnTt3rrbLBwB4GX+zCwAA1A1xcXGKi4urdBy73a6wsLAqz/Pll1/W1KlTNXnyZEnS0qVL9cEHH2j58uV66qmnalQvAMC7cUYGAOA1UlNTFRISok6dOumxxx7ThQsXKhz38uXL2rdvnwYPHuwcVq9ePQ0ePFi7du2qcLri4mIVFBSU6QAA1kOQAQB4hWHDhunvf/+7UlJS9Pzzz2v79u2Ki4tTSUlJueOfP39eJSUlCg0NLTM8NDRUOTk5FS7H4XAoKCjI2UVGRtZqOwAAnsGlZQAAr/DLX/7S+e/u3burR48euu2225SamqpBgwbV2nISExOVkJDg7C8oKCDMAIAFcUYGAOCV2rdvr5YtW+r48ePlft6yZUv5+fkpNze3zPDc3NxK77Ox2+0KDAws0wEArIcgAwDwSmfPntWFCxcUHh5e7ucNGjRQr169lJKS4hxWWlqqlJQUxcTEeKpMAIBJCDIAAI8oLCxURkaGMjIyJEknT55URkaGsrKyVFhYqCeeeEK7d+/WqVOnlJKSovvvv1+33367YmNjnfMYNGiQFi1a5OxPSEjQG2+8oVWrVunIkSN67LHHVFRU5HyKGQDAd3GPDADAI9LT0zVw4EBn/9X7VCZOnKglS5bo888/16pVq5SXl6eIiAgNHTpUzz77rOx2u3OazMxMnT9/3tk/btw4ff3115o1a5ZycnLUs2dPJScn3/AAAACA7yHIAAA8YsCAATIMo8LPN2/efNN5nDp16oZh06dP1/Tp02tSGgDAgri0DAAAAIDlEGQAAAAAWA5BBgAAAIDlcI8MYGXbHGZXAAAAYArOyAAAAACwHIIMAAAAAMshyAAAAACwHIIMAAAAAMshyAAAAACwHIIMAAAAAMshyAAAAACwHIIMAAAAAMshyAAAAACwHIIMAAAAAMshyAAAAACwHJeCzJw5c2Sz2cp0nTt3dldtAAAAAFAuf1cn6Nq1q7Zu3frTDPxdngUAAAAA1IjLKcTf319hYWHuqAUAAAAAqsTle2SOHTumiIgItW/fXuPHj1dWVlal4xcXF6ugoKBMBwAAAAA14VKQ6dOnj1auXKnk5GQtWbJEJ0+e1H333aeLFy9WOI3D4VBQUJCzi4yMrHHRAAAAAOo2l4JMXFycxowZox49eig2NlabNm1SXl6e3nnnnQqnSUxMVH5+vrM7c+ZMjYsGAAAAULfV6E79pk2bqmPHjjp+/HiF49jtdtnt9posBgAAAADKqNF7ZAoLC5WZmanw8PDaqgcAAAAAbsqlIDNz5kxt375dp06d0s6dOzV69Gj5+fkpPj7eXfUBAAAAwA1curTs7Nmzio+P14ULFxQcHKx7771Xu3fvVnBwsLvqAwAAAIAbuBRk1q5d6646AAAAAKDKanSPDAAAAACYgSADAAAAwHIIMgAAAAAshyADAAAAwHIIMgAAAAAshyADAAAAwHIIMgAAAAAshyADAPCItLQ0jRw5UhEREbLZbEpKSnJ+duXKFT355JPq3r27GjdurIiICE2YMEFfffVVpfOcM2eObDZbma5z585ubgkAwBsQZAAAHlFUVKSoqCgtXrz4hs8uXbqk/fv3689//rP279+v9evX6+jRo/qP//iPm863a9euys7OdnaffPKJO8oHAHgZf7MLAADUDXFxcYqLiyv3s6CgIG3ZsqXMsEWLFunuu+9WVlaWWrduXeF8/f39FRYWVqu1AgC8H2dkAABeKT8/XzabTU2bNq10vGPHjikiIkLt27fX+PHjlZWVVen4xcXFKigoKNMBAKyHIAMA8Drff/+9nnzyScXHxyswMLDC8fr06aOVK1cqOTlZS5Ys0cmTJ3Xffffp4sWLFU7jcDgUFBTk7CIjI93RBACAm3FpGWBF2xxmVwC4zZUrVzR27FgZhqElS5ZUOu61l6r16NFDffr0UZs2bfTOO+/okUceKXeaxMREJSQkOPsLCgoIMwBgQQQZAIDXuBpiTp8+rY8//rjSszHladq0qTp27Kjjx49XOI7dbpfdbq9pqQAAk3FpGQDAK1wNMceOHdPWrVvVokULl+dRWFiozMxMhYeHu6FCAIA3IcgAADyisLBQGRkZysjIkCSdPHlSGRkZysrK0pUrV/Tggw8qPT1dq1evVklJiXJycpSTk6PLly875zFo0CAtWrTI2T9z5kxt375dp06d0s6dOzV69Gj5+fkpPj7e080DAHgYl5YBADwiPT1dAwcOdPZfvU9l4sSJmjNnjt577z1JUs+ePctMt23bNg0YMECSlJmZqfPnzzs/O3v2rOLj43XhwgUFBwfr3nvv1e7duxUcHOzexgAATEeQAQB4xIABA2QYRoWfV/bZVadOnSrTv3bt2pqWBQCwKC4tAwAAAGA5BBkAAAAAlkOQAQAAAGA5BBkAAAAAlkOQAQAAAGA5BBkAAAAAlkOQAQAAAGA5BBkAAAAAlkOQAQAAAGA5BBkAAAAAlkOQAQAAAGA5BBkAAAAAlkOQAQAAAGA5BBkAAAAAlkOQAQAAAGA5BBkAAAAAlkOQAQAAAGA5BBkAAAAAlkOQAQAAAGA5NQoy8+fPl81m04wZM2qpHAAAAAC4uWoHmb179+q1115Tjx49arMeAAAAALipagWZwsJCjR8/Xm+88YaaNWtW2zUBqMw2h9kVAAAAmK5aQWbatGkaMWKEBg8efNNxi4uLVVBQUKYDAAAAgJrwd3WCtWvXav/+/dq7d2+Vxnc4HJo7d67LhQEAAABARVw6I3PmzBk9/vjjWr16tQICAqo0TWJiovLz853dmTNnqlUoAAAAAFzl0hmZffv26dy5c7rzzjudw0pKSpSWlqZFixapuLhYfn5+Zaax2+2y2+21Uy0AAAAAyMUgM2jQIB08eLDMsMmTJ6tz58568sknbwgxAAAAAOAOLgWZJk2aqFu3bmWGNW7cWC1atLhhOAAAAAC4S41eiAkAAAAAZnD5qWXXS01NrYUyAAAAAKDqOCMDAAAAwHIIMgAAAAAshyADAAAAwHIIMgAAj0hLS9PIkSMVEREhm82mpKSkMp8bhqFZs2YpPDxcDRs21ODBg3Xs2LGbznfx4sVq27atAgIC1KdPH3366aduagEAwJsQZAAAHlFUVKSoqCgtXry43M//8pe/6NVXX9XSpUu1Z88eNW7cWLGxsfr+++8rnOfbb7+thIQEzZ49W/v371dUVJRiY2N17tw5dzUDAOAlCDIAAI+Ii4vTc889p9GjR9/wmWEYWrhwoZ5++mndf//96tGjh/7+97/rq6++uuHMzbVefvllTZ06VZMnT1aXLl20dOlSNWrUSMuXL3djSwAA3oAgAwAw3cmTJ5WTk6PBgwc7hwUFBalPnz7atWtXudNcvnxZ+/btKzNNvXr1NHjw4AqnkaTi4mIVFBSU6QAA1kOQAQCYLicnR5IUGhpaZnhoaKjzs+udP39eJSUlLk0jSQ6HQ0FBQc4uMjKyhtUDAMxAkAEA1CmJiYnKz893dmfOnDG7JAD40TaH2RVYCkEGAGC6sLAwSVJubm6Z4bm5uc7PrteyZUv5+fm5NI0k2e12BQYGlukAANZDkAEAmK5du3YKCwtTSkqKc1hBQYH27NmjmJiYcqdp0KCBevXqVWaa0tJSpaSkVDgNAMB3+JtdAACgbigsLNTx48ed/SdPnlRGRoaaN2+u1q1ba8aMGXruuefUoUMHtWvXTn/+858VERGhUaNGOacZNGiQRo8erenTp0uSEhISNHHiRPXu3Vt33323Fi5cqKKiIk2ePNnTzQMAeBhBBgDgEenp6Ro4cKCzPyEhQZI0ceJErVy5Un/4wx9UVFSkX//618rLy9O9996r5ORkBQQEOKfJzMzU+fPnnf3jxo3T119/rVmzZiknJ0c9e/ZUcnLyDQ8AAAD4HoIMAMAjBgwYIMMwKvzcZrPpmWee0TPPPFPhOKdOnbph2PTp051naAAAdQdBBgAAADATTyurFm72BwAAAGA5BBkAAAAAlkOQAQAAAGA53CMDWEU518/uOnHB+e+Y9i08WQ0AAICpOCMDAAAAwHIIMgAAAAAshyADAAAAmOX6S8e3OXgccxURZAAAAABYDkEGAAAAgOUQZAAAAABYDkEGAAAAgOUQZAAAAABYDi/EBHzQtS/KlHhZJgAA8D2ckQHqGh7rCAAAfABBBgAAAIDlEGQAAAAAM3CFRI1wjwwAAADgKYSXWsMZGQAAAMDbEHhuiiADAAAAwHIIMgAAAAAshyADAAAAwHJcCjJLlixRjx49FBgYqMDAQMXExOjDDz90V20AAAAAUC6XnlrWqlUrzZ8/Xx06dJBhGFq1apXuv/9+HThwQF27dnVXjQCqYNeJC2aXAAAA4DEuBZmRI0eW6f/v//5vLVmyRLt37ybIAAAAAPCYar9HpqSkROvWrVNRUZFiYmIqHK+4uFjFxcXO/oKCguouEoA48wIAACBV42b/gwcP6pZbbpHdbtdvf/tbbdiwQV26dKlwfIfDoaCgIGcXGRlZo4IBAAAAwOUg06lTJ2VkZGjPnj167LHHNHHiRP3zn/+scPzExETl5+c7uzNnztSoYAAAAABw+dKyBg0a6Pbbb5ck9erVS3v37tUrr7yi1157rdzx7Xa77HZ7zaoEAAAAgGtU+x6Zq0pLS8vcAwMAAADgOtscZlfgc1wKMomJiYqLi1Pr1q118eJFrVmzRqmpqdq8ebO76gMAAACAG7gUZM6dO6cJEyYoOztbQUFB6tGjhzZv3qwhQ4a4qz4AkhZs+VL3ZPG0MgAAgKtcCjLLli1zVx0AAAAAUGUuP7UMgIdtc+ierNfNrgLwiLZt28pms93QTZs2rdzxV65cecO4AQEBHq4aAGCGGt/sDwBAbdm7d69KSkqc/YcOHdKQIUM0ZsyYCqcJDAzU0aNHnf02m82tNQIAvANBBgDgNYKDg8v0z58/X7fddpv69+9f4TQ2m01hYWHuLg0A4GW4tAwA4JUuX76sN998U1OmTKn0LEthYaHatGmjyMhI3X///Tp8+HCl8y0uLlZBQUGZDgBgPQQZAIBXSkpKUl5eniZNmlThOJ06ddLy5cu1ceNGvfnmmyotLVXfvn119uzZCqdxOBwKCgpydpGRkW6oHgDgbgQZAIBXWrZsmeLi4hQREVHhODExMZowYYJ69uyp/v37a/369QoODtZrr71W4TSJiYnKz893dmfOnHFH+QAAN+MeGQCA1zl9+rS2bt2q9evXuzRd/fr1FR0drePHj1c4jt1ul91ur2mJAACTcUYGAOB1VqxYoZCQEI0YMcKl6UpKSnTw4EGFh4e7qTIAgLcgyAAAvEppaalWrFihiRMnyt+/7IUDEyZMUGJiorP/mWee0UcffaQTJ05o//79euihh3T69Gk9+uijni4bAOBhXFoGAPAqW7duVVZWlqZMmXLDZ1lZWapX76ff4L799ltNnTpVOTk5atasmXr16qWdO3eqS5cuniwZAGACggwAwKsMHTpUhmGU+1lqamqZ/gULFmjBggUeqAoAamCbw+wKfBKXlgEAAADejjB0A4IMAAAAAMvh0jLACy3Y8qXz3/dkXTCxEgAAAO9EkAG81D1Zr5tdAgAAgNfi0jIAAAAAlsMZGcBLXHs5GQAAACrHGRkAAADAXWrjaWM8saxcBBkAAAAAlkOQAQAAAGA5BBkAAAAAlsPN/kAdsOvET++iiWnfwsRKAAAAagdnZAAAAABYDkEGAAAAgOUQZAAAAABYDkEGAAAAgOUQZAAAAABYDk8tA0y0YMuXZpcAAABgSZyRAQAAAGA5BBkAAAAAlkOQAQAAAGA53CMDAAAA1LZtDrMr8HmckQEAAABgOZyRAQAAANxs14kLzn/HtG9hYiW+gyADAAAAuMG14QW1jyADeJl7sl43uwQAAACvR5ABAAAAPOj6MzVcalY9Lt3s73A4dNddd6lJkyYKCQnRqFGjdPToUXfVBgAAAADlcinIbN++XdOmTdPu3bu1ZcsWXblyRUOHDlVRUZG76gMAAACAG7h0aVlycnKZ/pUrVyokJET79u3Tz372s1otDAAAAAAqUqN7ZPLz8yVJzZs3r3Cc4uJiFRcXO/sLCgpqskgAAAAAqP4LMUtLSzVjxgz169dP3bp1q3A8h8OhoKAgZxcZGVndRQIAAACApBqckZk2bZoOHTqkTz75pNLxEhMTlZCQ4OwvKCggzAAAAMAnLdjypSTpnizeIeNu1Qoy06dP1/vvv6+0tDS1atWq0nHtdrvsdnu1igMAAACshnfCeYZLl5YZhqHp06drw4YN+vjjj9WuXTt31QX4rAVbvnR2AMqaM2eObDZbma5z586VTrNu3Tp17txZAQEB6t69uzZt2uShagEAZnIpyEybNk1vvvmm1qxZoyZNmignJ0c5OTn67rvv3FUfAKCO6dq1q7Kzs51dZZcw79y5U/Hx8XrkkUd04MABjRo1SqNGjdKhQ4c8WDEAwAwuBZklS5YoPz9fAwYMUHh4uLN7++233VUfAKCO8ff3V1hYmLNr2bJlheO+8sorGjZsmJ544gndcccdevbZZ3XnnXdq0aJFHqwYAGAGl+6RMQzDXXUAACBJOnbsmCIiIhQQEKCYmBg5HA61bt263HF37dpV5oEykhQbG6ukpKQK589rAQDAN1T78csAatc9Wa9zcyDqvD59+mjlypVKTk7WkiVLdPLkSd133326ePFiuePn5OQoNDS0zLDQ0FDl5ORUuAxeCwAAvoEgAwDwGnFxcRozZox69Oih2NhYbdq0SXl5eXrnnXdqbRmJiYnKz893dmfOnKm1eQMAPKfa75EBUHU8oQyonqZNm6pjx446fvx4uZ+HhYUpNze3zLDc3FyFhYVVOE9eCwDA2+w68dM7Z2LatzCxEmvhjAwAwGsVFhYqMzNT4eHh5X4eExOjlJSUMsO2bNmimJgYT5QHADARQQYA4DVmzpyp7du369SpU9q5c6dGjx4tPz8/xcfHS5ImTJigxMRE5/iPP/64kpOT9dJLL+mLL77QnDlzlJ6erunTp5vVBAB12TYH97t6EJeWAQC8xtmzZxUfH68LFy4oODhY9957r3bv3q3g4GBJUlZWlurV++k3uL59+2rNmjV6+umn9cc//lEdOnRQUlKSunXrZlYTANRBVy8hvyfrwk3GRG0iyAAAvMbatWsr/Tw1NfWGYWPGjNGYMWPcVBEAwFtxaRkAAAAAy+GMDAAAAOAC055Gus3x438HJlY+Xh3BGRkAAAAAlkOQAQAAAGqIp5V5HkEGAAAAgOUQZAAvwK84AAAAriHIAAAAALAcnloGAAAAeIldJ356qWZM+xYmVuL9CDIAAABANXF5uHkIMoAbmPZ8eQAAgDqCe2QAAAAAWA5BBgAAAIDlcGkZAAAAcBNcNu59CDJALbHcDm6b48f/Dkw0tw4AAIBqIMgAdQyPdQQAAL6AIAOYiEc2AgAAVA83+wMAAACwHIIMAAAAAMshyAAAAACwHIIMAAAAAMshyAAAAACwHJ5aBtSA5d4dAwAAquxmx3mePmouzsgAAAAAsByCDAAAAADLIcgAAAAAsBzukQFMwDW1AAB4p6reF7O79a89UQ4qQZABAAAAvNCuExec/45p38LESrwTl5YBAAAALjL16optDvOW7UUIMgAAAAAsx+Ugk5aWppEjRyoiIkI2m01JSUluKAsAAAAAKuZykCkqKlJUVJQWL17sjnoAn3ZP1uvc6A8AAFALXL7ZPy4uTnFxce6oBQAAAACqhKeWAS662WMZAQAA4H5uv9m/uLhYBQUFZToAAMrjcDh01113qUmTJgoJCdGoUaN09OjRSqdZuXKlbDZbmS4gIMBDFQPwBQu2fOnsYB1uDzIOh0NBQUHOLjIy0t2LBABY1Pbt2zVt2jTt3r1bW7Zs0ZUrVzR06FAVFRVVOl1gYKCys7Od3enTpz1UMQDALG6/tCwxMVEJCQnO/oKCAsIMLIdfaADPSE5OLtO/cuVKhYSEaN++ffrZz35W4XQ2m01hYWHuLg8A4EXcHmTsdrvsdru7FwPUqtoOLjypDKie/Px8SVLz5s0rHa+wsFBt2rRRaWmp7rzzTs2bN09du3Ytd9zi4mIVFxc7+7nkGQCsyeUgU1hYqOPHjzv7T548qYyMDDVv3lytW7eu1eIAAHVXaWmpZsyYoX79+qlbt24VjtepUyctX75cPXr0UH5+vl588UX17dtXhw8fVqtWrW4Y3+FwaO7cue4sHYCX40oL3+DyPTLp6emKjo5WdHS0JCkhIUHR0dGaNWtWrRcHAKi7pk2bpkOHDmnt2rWVjhcTE6MJEyaoZ8+e6t+/v9avX6/g4GC99tpr5Y6fmJio/Px8Z3fmzBl3lA8AcDOXz8gMGDBAhmG4oxYAACRJ06dP1/vvv6+0tLRyz6pUpn79+oqOji5z9cC1uOQZQHVxqbh3cftTywAAqCrDMDR9+nRt2LBBH3/8sdq1a+fyPEpKSnTw4EGFh4e7oUIA8BLbHD92dRgvxATciF9uANdMmzZNa9as0caNG9WkSRPl5ORIkoKCgtSwYUNJ0oQJE3TrrbfK4fjxAP7MM8/onnvu0e233668vDy98MILOn36tB599FHT2gEAcD+CDADAayxZskTSj5cxX2vFihWaNGmSJCkrK0v16v10QcG3336rqVOnKicnR82aNVOvXr20c+dOdenSxVNlAwBMQJAB6rBdJy5o9w8/Prnl90M6mlwNoCrdg5mamlqmf8GCBVqwYIGbKgLgK3hSme8hyAD/xg4OAADAOrjZHwAAAIDlcEYGAAAAqAQP7/FOBBnATdjpAQBgrtq4bNxbjue7Tlxw/jumfQsTK/EeXFoGAAAAwHI4I4M6jRv8AQBARbzlbEyltjmkgYlmV2EKzsgAAAAAsBzOyAC1zBK/3gAA4KO42qLuIMgAtYQAAwCAb+CYbg0EGQAAAFgaZ2HqJu6RAQAAAGA5nJFBncOvNgAAANZHkAHquJ+uA37R1DoAAKgqd/4oyf0x1kGQAWqIHR4AAIDnEWRQJ7jjlxsCDAAAgHkIMvBJ3AcDAIBvcfexnR8orYenlgEAAACwHM7IAC7iFxsAAADzEWQAAADglbhUHJXh0jIAAAAAlsMZGfgMfrUBAMB6zD5+35P1una3/rWpNaB6CDJ10LU7jN8P6WhiJdbD/TEAAPgeyx/ftzmkgYlmV+FxBJk6zuqhxuxfcQAAADxt14kLZfpj2rcwqRJzEWR8VHW+4Fc0jbcFHE+GF8v/QgMAgBfyhh8iOcZbH0HG4jyxI7h+Gd4WbFA7rH52DgDg3bwpvHBPjG8gyFiQ2TsCT3zhJTwBAACgMgQZCzA7uFSmOrVVFEoqm5c3rwMAAFA+bz1+++RlZdscP/63Dt30T5DxUt76h18bvLltPrljAwAAHON9EEHGZN78pR51C9cNAwBqA99t4CkEGUD8SgMAgCusFFY4xvsugowJrPTH78vYsQEA4Ns41vs2ggzqHHZqN8dT4wAAkjV/fK3zx/ltjjpzwz9BxgOssBO4/v6Ie7Jev+FeiarcQ+HN91nU+R0bAABVYIXvLder68f4XScuOP8d076FiZV4VrWCzOLFi/XCCy8oJydHUVFR+utf/6q77767tmuDG7kSSqoyj+vHrWhaMwJOXd+5AVbk6nFm3bp1+vOf/6xTp06pQ4cOev755zV8+HAPVgxYixXDCnA9l4PM22+/rYSEBC1dulR9+vTRwoULFRsbq6NHjyokJMQdNVqSVXYQVQ0gNZ2msnGrE27KO4MEwDe4epzZuXOn4uPj5XA49Itf/EJr1qzRqFGjtH//fnXr1s2EFqCiFydX9YXKNX3xside3GwVVvk+UhXl/YDqjVeAwHNshmEYrkzQp08f3XXXXVq0aJEkqbS0VJGRkfqv//ovPfXUUzedvqCgQEFBQcrPz1dgYGD1qvZS3rKzuP5LvVW/7F+7cyK4eFZlB4a6/qXAyqyy/3X1ODNu3DgVFRXp/fffdw6755571LNnTy1duvSmy7PKevE23nLMqw3X79e8NQj50jqvimsvc+e4Xz0xj7xodgk3VZN9sEtB5vLly2rUqJHeffddjRo1yjl84sSJysvL08aNG2+Ypri4WMXFxc7+/Px8tW7dWmfOnLHsAWPxx8fNLkF3nV3h/PfeVpNvGAbUpqvbWFVN+/ntbqoENVFQUKDIyEjl5eUpKCjI7HLKVZ3jTOvWrZWQkKAZM2Y4h82ePVtJSUn67LPPbhjfF49L17v2OFXR36M3HMuAa/E9xr3unvDfZpdQrpocm1y6tOz8+fMqKSlRaGhomeGhoaH64osvyp3G4XBo7ty5NwyPjIx0ZdGo1CKzC4DPc20b+6ObqkDtuHjxotcGmeocZ3JycsodPycnp9zx69pxib9HAJKk//Lu74vVOTa5/alliYmJSkhIcPaXlpbqm2++UYsWLWSz2dy9eNNdTZm+9EufJ7H+aob1VzO+tv4Mw9DFixcVERFhdimmqulxyde2i+vRPuvy5bZJtM/qKmpfTY5NLgWZli1bys/PT7m5uWWG5+bmKiwsrNxp7Ha77HZ7mWFNmzZ1rUofEBgY6JMbpaew/mqG9VczvrT+vPVMzFXVOc6EhYWZclzype2iPLTPuny5bRLts7ry2lfdY1M9V0Zu0KCBevXqpZSUFOew0tJSpaSkKCYmploFAABwVXWOMzExMWXGl6QtW7ZwXAIAH+fypWUJCQmaOHGievfurbvvvlsLFy5UUVGRJk927WZgAADKc7PjzIQJE3TrrbfK4XBIkh5//HH1799fL730kkaMGKG1a9cqPT1dr7/OU44AwJe5HGTGjRunr7/+WrNmzVJOTo569uyp5OTkG260xI/sdrtmz559w2UMqBrWX82w/mqG9WeOmx1nsrKyVK/eTxcU9O3bV2vWrNHTTz+tP/7xj+rQoYOSkpLc9g4ZX98uaJ91+XLbJNpnde5on8vvkQEAAAAAs7l0jwwAAAAAeAOCDAAAAADLIcgAAAAAsByCDAAAAADLIcjUUFpamkaOHKmIiAjZbDYlJSXddJrU1FTdeeedstvtuv3227Vy5Uq31+mtXF1/qampstlsN3Q5OTmeKdiLOBwO3XXXXWrSpIlCQkI0atQoHT169KbTrVu3Tp07d1ZAQIC6d++uTZs2eaBa71Od9bdy5cobtr2AgAAPVQxP8fX9uqvtW79+vYYMGaLg4GAFBgYqJiZGmzdv9kyxLqrO/7urduzYIX9/f/Xs2dNt9dVUddpXXFysP/3pT2rTpo3sdrvatm2r5cuXu7/YaqhO+1avXq2oqCg1atRI4eHhmjJlii5cuOD+Yl3k68fs6rTvjTfe0H333admzZqpWbNmGjx4sD799FOXlkuQqaGioiJFRUVp8eLFVRr/5MmTGjFihAYOHKiMjAzNmDFDjz76qNceFNzN1fV31dGjR5Wdne3sQkJC3FSh99q+fbumTZum3bt3a8uWLbpy5YqGDh2qoqKiCqfZuXOn4uPj9cgjj+jAgQMaNWqURo0apUOHDnmwcu9QnfUn/fhG4mu3vdOnT3uoYniKr+/XXW1fWlqahgwZok2bNmnfvn0aOHCgRo4cqQMHDri5UtdV95iSl5enCRMmaNCgQW6qrHZUp31jx45VSkqKli1bpqNHj+qtt95Sp06d3Fhl9bnavh07dmjChAl65JFHdPjwYa1bt06ffvqppk6d6uZKXefrx+zqtC81NVXx8fHatm2bdu3apcjISA0dOlT/+te/qr5gA7VGkrFhw4ZKx/nDH/5gdO3atcywcePGGbGxsW6szBqqsv62bdtmSDK+/fZbj9RkJefOnTMkGdu3b69wnLFjxxojRowoM6xPnz7Gb37zG3eX5/Wqsv5WrFhhBAUFea4omM7X9+tVaV95unTpYsydO7f2C6pFrrRt3LhxxtNPP23Mnj3biIqKcmtdtaUq7fvwww+NoKAg48KFC54pqhZVpX0vvPCC0b59+zLDXn31VePWW291Y2W1w9eP2VVp3/V++OEHo0mTJsaqVauqPA1nZDxs165dGjx4cJlhsbGx2rVrl0kVWVPPnj0VHh6uIUOGaMeOHWaX4xXy8/MlSc2bN69wHLa/ilVl/UlSYWGh2rRpo8jISN1///06fPiwJ8qDF6trf1elpaW6ePHiTf9WrGLFihU6ceKEZs+ebXYpte69995T79699Ze//EW33nqrOnbsqJkzZ+q7774zu7RaERMTozNnzmjTpk0yDEO5ubl69913NXz4cLNLuylfP2ZX9Zh6rUuXLunKlSsuTUOQ8bCcnBzn26mvCg0NVUFBgc/sWNwpPDxcS5cu1T/+8Q/94x//UGRkpAYMGKD9+/ebXZqpSktLNWPGDPXr16/St5lXtP3VxXuMrlXV9depUyctX75cGzdu1JtvvqnS0lL17dtXZ8+e9WC18DZ1bb/+4osvqrCwUGPHjjW7lBo7duyYnnrqKb355pvy9/c3u5xad+LECX3yySc6dOiQNmzYoIULF+rdd9/V7373O7NLqxX9+vXT6tWrNW7cODVo0EBhYWEKCgpy+dJCT/P1Y3ZV23e9J598UhERETeEt8r43l8tfFqnTp3KXNvbt29fZWZmasGCBfrf//1fEysz17Rp03To0CF98sknZpdiSVVdfzExMYqJiXH29+3bV3fccYdee+01Pfvss+4uEzDdmjVrNHfuXG3cuNHy9yaWlJToV7/6lebOnauOHTuaXY5blJaWymazafXq1QoKCpIkvfzyy3rwwQf1t7/9TQ0bNjS5wpr55z//qccff1yzZs1SbGyssrOz9cQTT+i3v/2tli1bZnZ5FfL1Y3Z12jd//nytXbtWqampLj1EhyDjYWFhYcrNzS0zLDc3V4GBgZbfoZjl7rvv9tmdQVVMnz5d77//vtLS0tSqVatKx61o+wsLC3NniV7NlfV3vfr16ys6OlrHjx93U3WwgrqyX1+7dq0effRRrVu3zqVfTL3VxYsXlZ6ergMHDmj69OmSfvzibxiG/P399dFHH+nnP/+5yVXWTHh4uG699VZniJGkO+64Q4Zh6OzZs+rQoYOJ1dWcw+FQv3799MQTT0iSevToocaNG+u+++7Tc889p/DwcJMrvJGvH7Orc0x98cUXNX/+fG3dulU9evRwaXlcWuZhMTExSklJKTNsy5YtZX7lhWsyMjK8cmflboZhaPr06dqwYYM+/vhjtWvX7qbTsP39pDrr73olJSU6ePBgndz+8JO68Hf11ltvafLkyXrrrbc0YsQIs8upFYGBgTp48KAyMjKc3W9/+1t16tRJGRkZ6tOnj9kl1li/fv301VdfqbCw0Dnsyy+/VL169Vz+4cYbXbp0SfXqlf0q6+fnJ+nHfbw38fVjdnWPqX/5y1/07LPPKjk5Wb17967WglEDFy9eNA4cOGAcOHDAkGS8/PLLxoEDB4zTp08bhmEYTz31lPHwww87xz9x4oTRqFEj44knnjCOHDliLF682PDz8zOSk5PNaoKpXF1/CxYsMJKSkoxjx44ZBw8eNB5//HGjXr16xtatW81qgmkee+wxIygoyEhNTTWys7Od3aVLl5zjPPzww8ZTTz3l7N+xY4fh7+9vvPjii8aRI0eM2bNnG/Xr1zcOHjxoRhNMVZ31N3fuXGPz5s1GZmamsW/fPuOXv/ylERAQYBw+fNiMJsBNfH2/7mr7Vq9ebfj7+xuLFy8u87eSl5dnVhMq5GrbruftTy1ztX0XL140WrVqZTz44IPG4cOHje3btxsdOnQwHn30UbOaUClX27dixQrD39/f+Nvf/mZkZmYan3zyidG7d2/j7rvvNqsJFfL1Y3Z12jd//nyjQYMGxrvvvltmmosXL1Z5uQSZGrr6OODru4kTJxqGYRgTJ040+vfvf8M0PXv2NBo0aGC0b9/eWLFihcfr9haurr/nn3/euO2224yAgACjefPmxoABA4yPP/7YnOJNVt56k1Rme+rfv79zXV71zjvvGB07djQaNGhgdO3a1fjggw88W7iXqM76mzFjhtG6dWujQYMGRmhoqDF8+HBj//79ni8ebuXr+3VX29e/f/9Kx/cm1fl/dy1vDzLVad+RI0eMwYMHGw0bNjRatWplJCQklPly6U2q075XX33V6NKli9GwYUMjPDzcGD9+vHH27FnPF38Tvn7Mrk772rRpU+40s2fPrvJybf9eOAAAAABYBvfIAAAAALAcggwAAAAAyyHIAAAAALAcggwAAAAAyyHIAAAAALAcggwAAAAAyyHIAAAAALAcggwAAAAAyyHIAAAAALAcggwAAAAAyyHIAAAAALAcggwAAAAAy/n/pVjxcKOvzhwAAAAASUVORK5CYII=",
      "text/plain": [
       "<Figure size 1000x450 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(10,4.5))\n",
    "\n",
    "plt.subplot(121)\n",
    "# plt.hist(train_reg_y, bins=100, alpha=0.5, label='train 1', density=True)\n",
    "plt.hist(train_reg_oof, bins=100, alpha=0.5, label='train', density=True)\n",
    "plt.hist(test_reg_pred, bins=100, alpha=0.5, label='testA', density=True)\n",
    "plt.legend()\n",
    "\n",
    "plt.subplot(122)\n",
    "idx = test_cla_pred == 1\n",
    "plt.hist(test_reg_pred[idx], bins=100, alpha=0.5, label='test 1', density=True)\n",
    "plt.hist(test_reg_pred[~idx], bins=100, alpha=0.5, label='test 0', density=True)\n",
    "plt.legend()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "53458ad6-2d0a-4359-a0e7-24deda71c99f",
   "metadata": {},
   "outputs": [],
   "source": [
    "feat_reg_imp['mean'] = feat_reg_imp[[f\"fold_{fold}\" for fold in range(reg_model.nfold)]].mean(axis=1)\n",
    "feat_reg_imp = feat_reg_imp.sort_values(by=\"mean\", ascending=False)\n",
    "feat_reg_imp.to_csv('model/feats_reg_imp.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "77e323c3-ba97-49cf-86ae-c0f701cf5b59",
   "metadata": {},
   "outputs": [],
   "source": [
    "test_reg_pred = test_reg_pred.round(0).astype('int')\n",
    "train_reg_oof = train_reg_oof.round(0).astype('int')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "b66b1793-4c57-4e95-8b3d-726457256efa",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Counter({2: 48743, 1: 1257})"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Counter(test_reg_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e767e08f-6911-4756-94ce-eb24db29c23c",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "dbe5baaa-360e-477e-8751-178f834b565d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(38734,)"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_reg_oof.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "cb240394-cfbb-41a0-b8a5-d0bd7b15e798",
   "metadata": {},
   "outputs": [],
   "source": [
    "test_pred = pd.DataFrame({'cust_wid': testA_base['cust_wid'], 'label': test_cla_pred})\n",
    "train_oof = pd.DataFrame({'cust_wid': train_base['cust_wid'], 'pred': train_cla_oof, 'true': train_base['label']})\n",
    "\n",
    "idx = test_pred.label == 1\n",
    "test_pred.loc[idx, 'label'] = test_reg_pred[idx]\n",
    "\n",
    "# idx = train_oof.pred == 1\n",
    "# train_oof.loc[idx, 'pred'] = train_reg_oof"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "88b53bc2-754b-484a-a175-146ea26bc36c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.7377930889915425, 0.8656764452152458, 0.6394167811185161)"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "score1 = fbeta_score((train_oof['true']>0).astype('int'), (train_oof['pred']>0).astype('int'), beta=2)\n",
    "score2 = mean_absolute_percentage_error(train_reg_y, train_reg_oof)\n",
    "score = score1-0.2*score2\n",
    "score, score1, score2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "8325edec-f07d-40c0-833f-485fedb0a700",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2    29586\n",
       "0    19157\n",
       "1     1257\n",
       "Name: label, dtype: int64"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_pred.label.value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "d4211a69-8474-4ddc-9a10-0fd646f3b774",
   "metadata": {},
   "outputs": [],
   "source": [
    "test_pred.to_csv('output.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fc328337-f50e-463a-a4c2-21283b1a3aae",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.8.8 ('base')",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  },
  "vscode": {
   "interpreter": {
    "hash": "688e27ea2201f24f036ee87480d7d8d79d04e5ade5319763c122c3e5a8095f9a"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}