sber_hack_code.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Import libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Plotting\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "# ML preprocessing/optimizing\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "from sklearn.model_selection import RandomizedSearchCV\n",
    "import shap\n",
    "\n",
    "\n",
    "# Scoring\n",
    "from sklearn.metrics import roc_auc_score\n",
    "\n",
    "# Data formatting/preprocessing\n",
    "import time \n",
    "import tqdm\n",
    "import pickle\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "\n",
    "# Ml models\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "import xgboost as xgb\n",
    "import lightgbm as lgb\n",
    "import catboost as ctb\n",
    "\n",
    "### Advanse optimizing (HyperOpt Parameter Tuning)\n",
    "from hyperopt import tpe\n",
    "from hyperopt import STATUS_OK\n",
    "from hyperopt import Trials\n",
    "from hyperopt import hp\n",
    "from hyperopt import fmin\n",
    "\n",
    "## Useful\n",
    "#pd.set_option('display.max_columns', None)\n",
    "\n",
    "%matplotlib inline "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Reading data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('train_part1.pkl', 'rb') as f1:\n",
    "    train_part1 = pickle.load(f1)\n",
    "\n",
    "with open('train_part2.pkl', 'rb') as f2:\n",
    "    train_part2 = pickle.load(f2)\n",
    "    \n",
    "with open('test_data.pkl', 'rb') as f3:\n",
    "    test_data = pickle.load(f3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Distribution of target"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    732535\n",
       "1     31266\n",
       "Name: TARGET, dtype: int64"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_part1.TARGET.value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### We can see class disbalance problem"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Get only numerical columns at first"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_numerics_only = train_part1.select_dtypes(include=np.number)\n",
    "target = df_numerics_only.TARGET\n",
    "\n",
    "df_numerics_only = df_numerics_only.apply(lambda x: x.fillna(x.mean()),axis=0)\n",
    "cols = df_numerics_only.columns\n",
    "l = []\n",
    "for i in range(len(cols)):\n",
    "    l.append(target.corr(df_numerics_only.loc[:,cols[i]]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Correlation between target and numerical features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "corr_data = pd.DataFrame([cols, l]).T.iloc[:628,:].dropna()\n",
    "corr_data = corr_data.rename(columns={0:'feature', 1:'corr_with_target'})\n",
    "corr_data['corr_with_target'] = pd.to_numeric(corr_data['corr_with_target'])\n",
    "corr_data.plot();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Get the most largest and smallest correlations and corresponding features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>feature</th>\n",
       "      <th>corr_with_target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>274</th>\n",
       "      <td>x_286</td>\n",
       "      <td>0.121005</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>280</th>\n",
       "      <td>x_292</td>\n",
       "      <td>0.116459</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>318</th>\n",
       "      <td>x_330</td>\n",
       "      <td>0.109346</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>177</th>\n",
       "      <td>x_189</td>\n",
       "      <td>0.108632</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>409</th>\n",
       "      <td>x_421</td>\n",
       "      <td>0.102239</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>171</th>\n",
       "      <td>x_183</td>\n",
       "      <td>0.100842</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>68</th>\n",
       "      <td>x_80</td>\n",
       "      <td>0.088774</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>572</th>\n",
       "      <td>x_584</td>\n",
       "      <td>0.088426</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>408</th>\n",
       "      <td>x_420</td>\n",
       "      <td>0.087458</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>215</th>\n",
       "      <td>x_227</td>\n",
       "      <td>0.087346</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>624</th>\n",
       "      <td>x_643</td>\n",
       "      <td>0.087316</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>532</th>\n",
       "      <td>x_544</td>\n",
       "      <td>0.086933</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>112</th>\n",
       "      <td>x_124</td>\n",
       "      <td>0.086197</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>249</th>\n",
       "      <td>x_261</td>\n",
       "      <td>0.085102</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>466</th>\n",
       "      <td>x_478</td>\n",
       "      <td>0.083339</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    feature  corr_with_target\n",
       "274   x_286          0.121005\n",
       "280   x_292          0.116459\n",
       "318   x_330          0.109346\n",
       "177   x_189          0.108632\n",
       "409   x_421          0.102239\n",
       "171   x_183          0.100842\n",
       "68     x_80          0.088774\n",
       "572   x_584          0.088426\n",
       "408   x_420          0.087458\n",
       "215   x_227          0.087346\n",
       "624   x_643          0.087316\n",
       "532   x_544          0.086933\n",
       "112   x_124          0.086197\n",
       "249   x_261          0.085102\n",
       "466   x_478          0.083339"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "corr_data.nlargest(15, 'corr_with_target')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>feature</th>\n",
       "      <th>corr_with_target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>279</th>\n",
       "      <td>x_291</td>\n",
       "      <td>-0.109343</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>612</th>\n",
       "      <td>x_630</td>\n",
       "      <td>-0.097946</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>176</th>\n",
       "      <td>x_188</td>\n",
       "      <td>-0.087319</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>73</th>\n",
       "      <td>x_85</td>\n",
       "      <td>-0.086197</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>410</th>\n",
       "      <td>x_422</td>\n",
       "      <td>-0.084399</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>334</th>\n",
       "      <td>x_346</td>\n",
       "      <td>-0.069840</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>531</th>\n",
       "      <td>x_543</td>\n",
       "      <td>-0.061604</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>122</th>\n",
       "      <td>x_134</td>\n",
       "      <td>-0.056354</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>225</th>\n",
       "      <td>x_237</td>\n",
       "      <td>-0.055390</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>358</th>\n",
       "      <td>x_370</td>\n",
       "      <td>-0.055001</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>254</th>\n",
       "      <td>x_266</td>\n",
       "      <td>-0.051904</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>240</th>\n",
       "      <td>x_252</td>\n",
       "      <td>-0.048928</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>346</th>\n",
       "      <td>x_358</td>\n",
       "      <td>-0.048446</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>48</th>\n",
       "      <td>x_60</td>\n",
       "      <td>-0.046113</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>x_14</td>\n",
       "      <td>-0.044104</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    feature  corr_with_target\n",
       "279   x_291         -0.109343\n",
       "612   x_630         -0.097946\n",
       "176   x_188         -0.087319\n",
       "73     x_85         -0.086197\n",
       "410   x_422         -0.084399\n",
       "334   x_346         -0.069840\n",
       "531   x_543         -0.061604\n",
       "122   x_134         -0.056354\n",
       "225   x_237         -0.055390\n",
       "358   x_370         -0.055001\n",
       "254   x_266         -0.051904\n",
       "240   x_252         -0.048928\n",
       "346   x_358         -0.048446\n",
       "48     x_60         -0.046113\n",
       "9      x_14         -0.044104"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "corr_data.nsmallest(15, 'corr_with_target')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### The idea is that simple linear correlation can tell us what features have more significant linear dependence - these features from the first point of view are better and should be chosen"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Lets handle with non-numeric data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(763801, 18)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_part1.select_dtypes(exclude=np.number).shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### There are only 18 columns that are non-numeric and shoud be preprocessed"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['REPORT_DT', 'x_9', 'x_12', 'x_13', 'x_17', 'x_18', 'x_19', 'x_21',\n",
       "       'x_25', 'x_26', 'x_27', 'x_614', 'x_615', 'x_617', 'x_618', 'x_625',\n",
       "       'x_628', 'x_634'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_part1.select_dtypes(exclude=np.number).columns"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feature x_12 looks like credit rate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>A1</td>\n",
       "      <td>0.019427</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>A</td>\n",
       "      <td>0.030718</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>B1</td>\n",
       "      <td>0.03394</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>B</td>\n",
       "      <td>0.042134</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>D</td>\n",
       "      <td>0.049258</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>C</td>\n",
       "      <td>0.050675</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>N</td>\n",
       "      <td>0.059137</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    0         1\n",
       "6  A1  0.019427\n",
       "3   A  0.030718\n",
       "0  B1   0.03394\n",
       "5   B  0.042134\n",
       "1   D  0.049258\n",
       "4   C  0.050675\n",
       "2   N  0.059137"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "a = []\n",
    "for i in train_part1.select_dtypes(exclude=np.number).x_12.unique():\n",
    "    a.append(train_part1[train_part1['x_12'] == i].TARGET.mean())\n",
    "\n",
    "res = pd.DataFrame([train_part1.select_dtypes(exclude=np.number).x_12.unique(), a])    \n",
    "res.T.sort_values(by = 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Looks like our hypothesis is right"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## For x_12 we will do a mapping - replace letters by mean values (it is a good idea because it has an order)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### x_18"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Вт    149381\n",
       "Ср    147631\n",
       "Чт    142560\n",
       "Пн    137282\n",
       "Пт    131018\n",
       "Сб     49719\n",
       "Вс      6210\n",
       "Name: x_18, dtype: int64"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_part1.select_dtypes(exclude=np.number).x_18.value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>Вс</td>\n",
       "      <td>0.031562</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>Сб</td>\n",
       "      <td>0.035158</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Пн</td>\n",
       "      <td>0.039736</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Чт</td>\n",
       "      <td>0.040383</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Вт</td>\n",
       "      <td>0.040936</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Пт</td>\n",
       "      <td>0.04275</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Ср</td>\n",
       "      <td>0.043311</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    0         1\n",
       "6  Вс  0.031562\n",
       "5  Сб  0.035158\n",
       "0  Пн  0.039736\n",
       "1  Чт  0.040383\n",
       "4  Вт  0.040936\n",
       "3  Пт   0.04275\n",
       "2  Ср  0.043311"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "a = []\n",
    "for i in train_part1.select_dtypes(exclude=np.number).x_18.unique():\n",
    "    a.append(train_part1[train_part1['x_18'] == i].TARGET.mean())\n",
    "\n",
    "res = pd.DataFrame([train_part1.select_dtypes(exclude=np.number).x_18.unique(), a])    \n",
    "res.T.sort_values(by = 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Looks like weekends positively reflects on credibility"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### x_19"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def rating_mean(name):\n",
    "    a = []\n",
    "    for i in train_part1.select_dtypes(exclude=np.number)['{}'.format(name)].unique():\n",
    "        a.append(train_part1[train_part1['{}'.format(name)] == i].TARGET.mean())\n",
    "\n",
    "    res = pd.DataFrame([train_part1.select_dtypes(exclude=np.number)['{}'.format(name)].unique(), a])    \n",
    "    res=res.T.sort_values(by = 1)\n",
    "    return res"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Check how can we replace non-numeric labels by counting mean statistics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "non numeric feature REPORT_DT\n",
      "             0         1\n",
      "513 2017-04-07       0.0\n",
      "813 2017-08-25       0.0\n",
      "814 2018-09-14       0.0\n",
      "815 2018-03-31       0.0\n",
      "816 2019-01-12       0.0\n",
      "..         ...       ...\n",
      "13  2017-07-03  0.068541\n",
      "440 2017-04-10  0.069372\n",
      "749 2017-07-10  0.070243\n",
      "286 2017-08-29  0.074457\n",
      "657 2017-03-06  0.083333\n",
      "\n",
      "[1028 rows x 2 columns]\n",
      "##############################################\n",
      "non numeric feature x_9\n",
      "              0      1\n",
      "1595 2015-01-06    0.0\n",
      "690  2017-07-08    0.0\n",
      "1663 2016-09-25    0.0\n",
      "1662 2012-10-26    0.0\n",
      "1661 2017-06-11    0.0\n",
      "...         ...    ...\n",
      "1644 2012-06-14  0.625\n",
      "1636 2016-01-24    0.9\n",
      "1834 2014-08-31    1.0\n",
      "1869 2013-10-20    1.0\n",
      "1872 2013-05-01    1.0\n",
      "\n",
      "[1873 rows x 2 columns]\n",
      "##############################################\n",
      "non numeric feature x_12\n",
      "    0         1\n",
      "6  A1  0.019427\n",
      "3   A  0.030718\n",
      "0  B1   0.03394\n",
      "5   B  0.042134\n",
      "1   D  0.049258\n",
      "4   C  0.050675\n",
      "2   N  0.059137\n",
      "##############################################\n",
      "non numeric feature x_13\n",
      "     0         1\n",
      "6   19       0.0\n",
      "7    3       0.0\n",
      "1    1   0.03625\n",
      "3    5  0.042838\n",
      "0    4  0.043497\n",
      "2    2  0.043583\n",
      "4    9  0.055659\n",
      "5  NaN       NaN\n",
      "##############################################\n",
      "non numeric feature x_17\n",
      "    0   1\n",
      "0 NaN NaN\n",
      "##############################################\n",
      "non numeric feature x_18\n",
      "    0         1\n",
      "6  Вс  0.031562\n",
      "5  Сб  0.035158\n",
      "0  Пн  0.039736\n",
      "1  Чт  0.040383\n",
      "4  Вт  0.040936\n",
      "3  Пт   0.04275\n",
      "2  Ср  0.043311\n",
      "##############################################\n",
      "non numeric feature x_19\n",
      "   0         1\n",
      "1  0  0.037364\n",
      "0  1  0.044256\n",
      "##############################################\n",
      "non numeric feature x_21\n",
      "                                         0         1\n",
      "4                         Рефинансирование  0.009902\n",
      "1                           Инвестирование  0.028449\n",
      "0                             Приобретение  0.046135\n",
      "3  Нецелевой кредит под залог недвижимости  0.060576\n",
      "2             Индивидуальное строительство  0.086406\n",
      "##############################################\n",
      "non numeric feature x_25\n",
      "    0   1\n",
      "0 NaN NaN\n",
      "##############################################\n",
      "non numeric feature x_26\n",
      "    0   1\n",
      "0 NaN NaN\n",
      "##############################################\n",
      "non numeric feature x_27\n",
      "    0   1\n",
      "0 NaN NaN\n",
      "##############################################\n",
      "non numeric feature x_614\n",
      "   0         1\n",
      "0  1  0.038552\n",
      "1  0  0.045624\n",
      "##############################################\n",
      "non numeric feature x_615\n",
      "   0         1\n",
      "1  0  0.037364\n",
      "0  1  0.044256\n",
      "##############################################\n",
      "non numeric feature x_617\n",
      "     0         1\n",
      "6   19       0.0\n",
      "7    3       0.0\n",
      "1    1   0.03625\n",
      "3    5  0.042838\n",
      "0    4  0.043497\n",
      "2    2  0.043583\n",
      "4    9  0.055659\n",
      "5  NaN       NaN\n",
      "##############################################\n",
      "non numeric feature x_618\n",
      "                                         0         1\n",
      "4                         Рефинансирование  0.009902\n",
      "1                           Инвестирование  0.028449\n",
      "0                             Приобретение  0.046135\n",
      "3  Нецелевой кредит под залог недвижимости  0.060576\n",
      "2             Индивидуальное строительство  0.086406\n",
      "##############################################\n",
      "non numeric feature x_625\n",
      "                         0         1\n",
      "0  living in city in apart  0.036371\n",
      "1                    other  0.049473\n",
      "##############################################\n",
      "non numeric feature x_628\n",
      "            0         1\n",
      "2  Сотрудники  0.030718\n",
      "0          ЗП   0.03401\n",
      "1       Улица  0.049625\n",
      "##############################################\n",
      "non numeric feature x_634\n",
      "   0         1\n",
      "1  1   0.03546\n",
      "0  0  0.043344\n",
      "##############################################\n"
     ]
    }
   ],
   "source": [
    "for i in train_part1.select_dtypes(exclude=np.number).columns:\n",
    "    try:\n",
    "        print('non numeric feature {}'.format(i))\n",
    "        print(rating_mean(i))\n",
    "        print(\"##############################################\")\n",
    "    except ValueError: \n",
    "        pass"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## drop due to non informativity: x_25, x_26, x_27, x_617 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "4     448421\n",
       "1     270181\n",
       "2      36367\n",
       "5       8217\n",
       "9        539\n",
       "3         14\n",
       "19        12\n",
       "Name: x_617, dtype: int64"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_part1.x_617.value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### x_617 = x_13 - if 1 then less probability to default than others"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "4     448421\n",
       "1     270181\n",
       "2      36367\n",
       "5       8217\n",
       "9        539\n",
       "3         14\n",
       "19        12\n",
       "Name: x_13, dtype: int64"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_part1.x_13.value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### observing results from above lets name perspectively_signif_non_num_features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "perspectively_signif_non_num_features = ['x_12', 'x_13', 'x_18', 'x_19', 'x_21', 'x_614', 'x_615', 'x_625',\n",
    "                                         'x_628', 'x_634']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## prepare them"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### bineries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "binaries = ['x_19', 'x_614', 'x_615', 'x_634']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1[binaries] = train_part1[binaries].apply(pd.to_numeric)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Very significant"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1) rating"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "mapping = rating_mean('x_12')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "ename": "AttributeError",
     "evalue": "'Series' object has no attribute '_replace_columnwise'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mAttributeError\u001b[0m                            Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-21-ba48fea02017>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtrain_part1\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'x_12'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreplace\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmapping\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minplace\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/opt/anaconda3/lib/python3.8/site-packages/pandas/core/series.py\u001b[0m in \u001b[0;36mreplace\u001b[0;34m(self, to_replace, value, inplace, limit, regex, method)\u001b[0m\n\u001b[1;32m   4866\u001b[0m         \u001b[0mmethod\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m\"pad\"\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   4867\u001b[0m     ):\n\u001b[0;32m-> 4868\u001b[0;31m         return super().replace(\n\u001b[0m\u001b[1;32m   4869\u001b[0m             \u001b[0mto_replace\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mto_replace\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   4870\u001b[0m             \u001b[0mvalue\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalue\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/opt/anaconda3/lib/python3.8/site-packages/pandas/core/generic.py\u001b[0m in \u001b[0;36mreplace\u001b[0;34m(self, to_replace, value, inplace, limit, regex, method)\u001b[0m\n\u001b[1;32m   6523\u001b[0m                 \u001b[0mto_replace\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalue\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mkeys\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalues\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   6524\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 6525\u001b[0;31m             return self.replace(\n\u001b[0m\u001b[1;32m   6526\u001b[0m                 \u001b[0mto_replace\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalue\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minplace\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0minplace\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlimit\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mlimit\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mregex\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mregex\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   6527\u001b[0m             )\n",
      "\u001b[0;32m/opt/anaconda3/lib/python3.8/site-packages/pandas/core/series.py\u001b[0m in \u001b[0;36mreplace\u001b[0;34m(self, to_replace, value, inplace, limit, regex, method)\u001b[0m\n\u001b[1;32m   4866\u001b[0m         \u001b[0mmethod\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m\"pad\"\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   4867\u001b[0m     ):\n\u001b[0;32m-> 4868\u001b[0;31m         return super().replace(\n\u001b[0m\u001b[1;32m   4869\u001b[0m             \u001b[0mto_replace\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mto_replace\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   4870\u001b[0m             \u001b[0mvalue\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalue\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/opt/anaconda3/lib/python3.8/site-packages/pandas/core/generic.py\u001b[0m in \u001b[0;36mreplace\u001b[0;34m(self, to_replace, value, inplace, limit, regex, method)\u001b[0m\n\u001b[1;32m   6543\u001b[0m                         \u001b[0;32mif\u001b[0m \u001b[0mcol\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mvalue\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mkeys\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0mcol\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   6544\u001b[0m                     }\n\u001b[0;32m-> 6545\u001b[0;31m                     \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_replace_columnwise\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmapping\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minplace\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mregex\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   6546\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   6547\u001b[0m                 \u001b[0;31m# {'A': NA} -> 0\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/opt/anaconda3/lib/python3.8/site-packages/pandas/core/generic.py\u001b[0m in \u001b[0;36m__getattr__\u001b[0;34m(self, name)\u001b[0m\n\u001b[1;32m   5485\u001b[0m         ):\n\u001b[1;32m   5486\u001b[0m             \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 5487\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mobject\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m__getattribute__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   5488\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   5489\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m__setattr__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0mstr\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalue\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m->\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mAttributeError\u001b[0m: 'Series' object has no attribute '_replace_columnwise'"
     ]
    }
   ],
   "source": [
    "train_part1['x_12'].replace(mapping, inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 2) day of week"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "map_x_18 = rating_mean('x_18')\n",
    "map_x_18"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "not_weekend = train_part1[((train_part1['x_18'] != 'Вс') & (train_part1['x_18'] != 'Сб'))].TARGET.mean()\n",
    "weekend = train_part1[((train_part1['x_18'] == 'Вс') | (train_part1['x_18'] == 'Сб'))].TARGET.mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "map_18 = {'Вс':weekend, \n",
    "         'Сб':weekend,\n",
    "         'Пн':not_weekend,\n",
    "         'Чт':not_weekend,\n",
    "         'Вт':not_weekend,\n",
    "         'Пт':not_weekend,\n",
    "         'Ср':not_weekend}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1['x_18'].replace(map_18, inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 3) x_13"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "map_x_13 = rating_mean('x_13')\n",
    "map_x_13"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "not_1 = train_part1[train_part1['x_13'] != '1'].TARGET.mean()\n",
    "map_13 = {'1':map_x_13.iloc[2,1], \n",
    "         '2':not_1,\n",
    "         '3':not_1,\n",
    "         '4':not_1,\n",
    "         '5':not_1,\n",
    "         '9':not_1,\n",
    "         '19':not_1}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1['x_13'].replace(map_13, inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Target of credit (x_21)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_21 = rating_mean('x_21')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "map_21 = x_21.set_index(0).to_dict()[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1['x_21'].replace(map_21, inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Town or not (x_625)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_625 = rating_mean('x_625')\n",
    "x_625"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "map_625 = x_625.set_index(0).to_dict()[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1['x_625'].replace(map_625, inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Street or ЗП or сотрудник (628)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_628 = rating_mean('x_628')\n",
    "x_628"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "map_628 = x_628.set_index(0).to_dict()[1]\n",
    "train_part1['x_628'].replace(map_628, inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# All nun num - now numeric"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1[perspectively_signif_non_num_features].info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1.rename(columns={'x_12':'credit_rating', 'x_18':'day', 'x_21':'credit_purpose',\n",
    "                            'x_625':'city', 'x_628':'street'}, inplace = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1.select_dtypes(exclude=object)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compare correlations of recently non numeric but very important features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_important = ['credit_rating', 'day', 'credit_purpose','city', 'street', 'x_13', 'x_19', 'x_614', 'x_615', 'x_634']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1[new_important].info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1[new_important]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1 = train_part1.select_dtypes(include=[float, int])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "comapre_data = train_part1.apply(lambda x: x.fillna(x.mean()),axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "target = comapre_data.TARGET\n",
    "m = []\n",
    "cols = comapre_data.columns[:-2]\n",
    "for i in range(len(cols)):\n",
    "    m.append(target.corr(comapre_data.loc[:,cols[i]]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "cols"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "corr_data_1 = pd.DataFrame([cols, m]).T.iloc[:,:]\n",
    "corr_data_1 = corr_data_1.rename(columns={0:'feature', 1:'corr_with_target'})\n",
    "corr_data_1['corr_with_target'] = pd.to_numeric(corr_data_1['corr_with_target'])\n",
    "corr_data_1.plot();"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "corr_data_1.nlargest(45, 'corr_with_target')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "corr_data_1.nsmallest(15, 'corr_with_target')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_important.append('TARGET')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "comapre_data[new_important]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "target = comapre_data.TARGET\n",
    "m = []\n",
    "cols = comapre_data[new_important].columns\n",
    "for i in range(len(cols)-1):\n",
    "    m.append(target.corr(comapre_data.loc[:,cols[i]]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "corr_data_2 = pd.DataFrame([cols, m]).T.iloc[:,:]\n",
    "corr_data_2 = corr_data_2.rename(columns={0:'feature', 1:'corr_with_target'})\n",
    "corr_data_2['corr_with_target'] = pd.to_numeric(corr_data_2['corr_with_target'])\n",
    "corr_data_2.plot();"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "corr_data_2.sort_values(by = 'corr_with_target', ascending = False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_important = new_important[:-1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Regressors"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "regressors = new_important + list(corr_data_1.nlargest(45, 'corr_with_target')['feature']) + list(corr_data_1.nsmallest(45, 'corr_with_target')['feature'])\n",
    "\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "regressors"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SHAP values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import xgboost as xgb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "comapre_data.drop(['x_646'], axis = 1)\n",
    "comapre_data.replace([np.inf, -np.inf], np.nan)\n",
    "comapre_data.dropna(inplace=True)\n",
    "comapre_data = comapre_data.astype('float16')\n",
    "\n",
    "X = comapre_data.drop(columns=['ID', 'TARGET'])\n",
    "y = comapre_data['TARGET']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=1)\n",
    "xgb_model = xgb.XGBRegressor()\n",
    "xgb_model.fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import shap  # package used to calculate Shap values\n",
    "\n",
    "# Create object that can calculate shap values\n",
    "explainer = shap.TreeExplainer(xgb_model)\n",
    "\n",
    "# Calculate Shap values\n",
    "shap_values = explainer.shap_values(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# use Kernel SHAP to explain test set predictions\n",
    "k_explainer = shap.KernelExplainer(xgb_model, X_train)\n",
    "k_shap_values = k_explainer.shap_values(X_test)\n",
    "shap.force_plot(k_explainer.expected_value[1], k_shap_values[1], data_for_prediction)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_part1.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "comapre_data.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Only non numeric and 1000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = comapre_data[new_important].iloc[:1000,:]\n",
    "y = comapre_data['TARGET'][:1000]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=1)\n",
    "xgb_model = xgb.XGBRegressor()\n",
    "xgb_model.fit(X_train, y_train)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = xgb_model.predict(X_test)\n",
    "y_pred_train = xgb_model.predict(X_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "roc_auc_score(y_test, y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "roc_auc_score(y_train, y_pred_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# All and 1000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = comapre_data.iloc[:1000,:]\n",
    "y = comapre_data['TARGET'][:1000]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=1)\n",
    "xgb_model = xgb.XGBRegressor()\n",
    "xgb_model.fit(X_train, y_train)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = xgb_model.predict(X_test)\n",
    "y_pred_train = xgb_model.predict(X_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "roc_auc_score(y_test, y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "roc_auc_score(y_train, y_pred_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_test.value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "space = {\n",
    "    'class_weight': hp.choice('class_weight', [None, 'balanced']),\n",
    "    'warm_start' : hp.choice('warm_start', [True, False]),\n",
    "    'fit_intercept' : hp.choice('fit_intercept', [True, False]),\n",
    "    'tol' : hp.loguniform('tol', 0.00001, 10),\n",
    "    'C' : hp.loguniform('C', 1, 5000),\n",
    "    'solver' : hp.choice('solver', ['newton-cg', 'lbfgs', 'liblinear', 'sag', 'saga']),\n",
    "    'max_iter' : hp.choice('max_iter', range(5,1000))\n",
    "}\n",
    "\n",
    "space_dict = {\n",
    "    'class_weight': [None, 'balanced'],\n",
    "    'warm_start' : [True, False],\n",
    "    'fit_intercept' : [True, False],\n",
    "    'solver' : ['newton-cg', 'lbfgs', 'liblinear', 'sag', 'saga']\n",
    "}"
   ]
  }
 ],
 "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": "3.8.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}