notebooks/solution_cb.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Import"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pathlib\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import random\n",
    "import re\n",
    "from sklearn.model_selection import train_test_split, KFold\n",
    "from sklearn.metrics import accuracy_score\n",
    "from sklearn.multioutput import MultiOutputClassifier\n",
    "import catboost as cb\n",
    "\n",
    "SEED = 87656\n",
    "random.seed(SEED)\n",
    "np.random.seed(SEED)\n",
    "\n",
    "DATA_DIR = pathlib.Path(\"../data\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Loading and preview"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "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>assessment</th>\n",
       "      <th>tags</th>\n",
       "      <th>text</th>\n",
       "      <th>trend_id_res0</th>\n",
       "      <th>trend_id_res1</th>\n",
       "      <th>trend_id_res2</th>\n",
       "      <th>trend_id_res3</th>\n",
       "      <th>trend_id_res4</th>\n",
       "      <th>trend_id_res5</th>\n",
       "      <th>trend_id_res6</th>\n",
       "      <th>...</th>\n",
       "      <th>trend_id_res40</th>\n",
       "      <th>trend_id_res41</th>\n",
       "      <th>trend_id_res42</th>\n",
       "      <th>trend_id_res43</th>\n",
       "      <th>trend_id_res44</th>\n",
       "      <th>trend_id_res45</th>\n",
       "      <th>trend_id_res46</th>\n",
       "      <th>trend_id_res47</th>\n",
       "      <th>trend_id_res48</th>\n",
       "      <th>trend_id_res49</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>index</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>5652</th>\n",
       "      <td>6.0</td>\n",
       "      <td>{ASSORTMENT,PROMOTIONS,DELIVERY}</td>\n",
       "      <td>Маленький выбор товаров, хотелось бы ассортиме...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18092</th>\n",
       "      <td>4.0</td>\n",
       "      <td>{ASSORTMENT,PRICE,PRODUCTS_QUALITY,DELIVERY}</td>\n",
       "      <td>Быстро</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13845</th>\n",
       "      <td>6.0</td>\n",
       "      <td>{DELIVERY,PROMOTIONS,PRICE,ASSORTMENT,SUPPORT}</td>\n",
       "      <td>Доставка постоянно задерживается</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25060</th>\n",
       "      <td>6.0</td>\n",
       "      <td>{PRICE,PROMOTIONS,ASSORTMENT}</td>\n",
       "      <td>Наценка и ассортимент расстраивают</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1428</th>\n",
       "      <td>6.0</td>\n",
       "      <td>{PRICE,PROMOTIONS}</td>\n",
       "      <td>Можно немного скинуть минимальную сумму заказа...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 53 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       assessment                                            tags  \\\n",
       "index                                                               \n",
       "5652          6.0                {ASSORTMENT,PROMOTIONS,DELIVERY}   \n",
       "18092         4.0    {ASSORTMENT,PRICE,PRODUCTS_QUALITY,DELIVERY}   \n",
       "13845         6.0  {DELIVERY,PROMOTIONS,PRICE,ASSORTMENT,SUPPORT}   \n",
       "25060         6.0                   {PRICE,PROMOTIONS,ASSORTMENT}   \n",
       "1428          6.0                              {PRICE,PROMOTIONS}   \n",
       "\n",
       "                                                    text  trend_id_res0  \\\n",
       "index                                                                     \n",
       "5652   Маленький выбор товаров, хотелось бы ассортиме...              0   \n",
       "18092                                             Быстро              0   \n",
       "13845                   Доставка постоянно задерживается              0   \n",
       "25060                 Наценка и ассортимент расстраивают              0   \n",
       "1428   Можно немного скинуть минимальную сумму заказа...              0   \n",
       "\n",
       "       trend_id_res1  trend_id_res2  trend_id_res3  trend_id_res4  \\\n",
       "index                                                               \n",
       "5652               0              0              0              0   \n",
       "18092              0              0              0              0   \n",
       "13845              0              0              1              0   \n",
       "25060              0              0              0              0   \n",
       "1428               0              0              0              0   \n",
       "\n",
       "       trend_id_res5  trend_id_res6  ...  trend_id_res40  trend_id_res41  \\\n",
       "index                                ...                                   \n",
       "5652               0              0  ...               0               0   \n",
       "18092              0              0  ...               0               0   \n",
       "13845              0              0  ...               0               0   \n",
       "25060              0              0  ...               0               0   \n",
       "1428               0              0  ...               0               0   \n",
       "\n",
       "       trend_id_res42  trend_id_res43  trend_id_res44  trend_id_res45  \\\n",
       "index                                                                   \n",
       "5652                0               0               0               0   \n",
       "18092               0               0               0               0   \n",
       "13845               0               0               0               0   \n",
       "25060               0               0               0               0   \n",
       "1428                0               0               0               0   \n",
       "\n",
       "       trend_id_res46  trend_id_res47  trend_id_res48  trend_id_res49  \n",
       "index                                                                  \n",
       "5652                0               0               0               0  \n",
       "18092               0               0               0               0  \n",
       "13845               0               0               0               0  \n",
       "25060               0               0               0               0  \n",
       "1428                0               0               0               0  \n",
       "\n",
       "[5 rows x 53 columns]"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_trends = pd.read_csv(DATA_DIR / \"trends_description.csv\", index_col='trend_id')\n",
    "train = pd.read_csv(DATA_DIR / \"train.csv\", index_col='index').drop(columns='Unnamed: 0')\n",
    "test = pd.read_csv(DATA_DIR / 'test.csv', index_col='index').drop(columns='Unnamed: 0')\n",
    "train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "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>trend</th>\n",
       "      <th>explanation</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>trend_id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Долгая доставка</td>\n",
       "      <td>- долго везёте\\n- очень долго везёте\\n- сюда о...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Доставка стала долгой</td>\n",
       "      <td>- клиент делает акцент на то, что раньше мы до...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Время доставки не соответствует заявленому</td>\n",
       "      <td>- клиент акцентирует внимание, что SLA не соот...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Регулярные опоздания</td>\n",
       "      <td>- клиент акцентриет внимание, что мы часто/каж...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Не отследить реальное время доставки</td>\n",
       "      <td>- клиент говорит о том, что невозможно понять,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>Курьер на карте</td>\n",
       "      <td>- если клиент упоминает отображение курьера на...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>Нет доставки по адресу</td>\n",
       "      <td>клиент жалутеся, что нет доставки по адресу</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>Не предупреждаем об удалении товара</td>\n",
       "      <td>- не предупреждаем об удалении товара\\n- не пр...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>Высокая минимальная сумма заказа</td>\n",
       "      <td>клиент жалуется на размер минимальной суммы за...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>Сумма заказа меняется во время набора корзины</td>\n",
       "      <td>- сумма заказа меняется во время набора корзин...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                  trend  \\\n",
       "trend_id                                                  \n",
       "0                                       Долгая доставка   \n",
       "1                                 Доставка стала долгой   \n",
       "2            Время доставки не соответствует заявленому   \n",
       "3                                  Регулярные опоздания   \n",
       "4                  Не отследить реальное время доставки   \n",
       "5                                       Курьер на карте   \n",
       "6                                Нет доставки по адресу   \n",
       "7                   Не предупреждаем об удалении товара   \n",
       "8                      Высокая минимальная сумма заказа   \n",
       "9         Сумма заказа меняется во время набора корзины   \n",
       "\n",
       "                                                explanation  \n",
       "trend_id                                                     \n",
       "0         - долго везёте\\n- очень долго везёте\\n- сюда о...  \n",
       "1         - клиент делает акцент на то, что раньше мы до...  \n",
       "2         - клиент акцентирует внимание, что SLA не соот...  \n",
       "3         - клиент акцентриет внимание, что мы часто/каж...  \n",
       "4         - клиент говорит о том, что невозможно понять,...  \n",
       "5         - если клиент упоминает отображение курьера на...  \n",
       "6               клиент жалутеся, что нет доставки по адресу  \n",
       "7         - не предупреждаем об удалении товара\\n- не пр...  \n",
       "8         клиент жалуется на размер минимальной суммы за...  \n",
       "9         - сумма заказа меняется во время набора корзин...  "
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_trends.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(['trend_id_res0',\n",
       "  'trend_id_res10',\n",
       "  'trend_id_res20',\n",
       "  'trend_id_res30',\n",
       "  'trend_id_res40'],\n",
       " ['assessment'],\n",
       " ['tags', 'text'])"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "target_cols = list(train.columns[-50:])\n",
    "num_cols = ['assessment']\n",
    "text_cols = ['tags', 'text']\n",
    "target_cols[::10], num_cols, text_cols"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "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>assessment</th>\n",
       "      <th>tags</th>\n",
       "      <th>text</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>index</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>5652</th>\n",
       "      <td>6.0</td>\n",
       "      <td>{ASSORTMENT,PROMOTIONS,DELIVERY}</td>\n",
       "      <td>Маленький выбор товаров, хотелось бы ассортиме...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18092</th>\n",
       "      <td>4.0</td>\n",
       "      <td>{ASSORTMENT,PRICE,PRODUCTS_QUALITY,DELIVERY}</td>\n",
       "      <td>Быстро</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13845</th>\n",
       "      <td>6.0</td>\n",
       "      <td>{DELIVERY,PROMOTIONS,PRICE,ASSORTMENT,SUPPORT}</td>\n",
       "      <td>Доставка постоянно задерживается</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25060</th>\n",
       "      <td>6.0</td>\n",
       "      <td>{PRICE,PROMOTIONS,ASSORTMENT}</td>\n",
       "      <td>Наценка и ассортимент расстраивают</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1428</th>\n",
       "      <td>6.0</td>\n",
       "      <td>{PRICE,PROMOTIONS}</td>\n",
       "      <td>Можно немного скинуть минимальную сумму заказа...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       assessment                                            tags  \\\n",
       "index                                                               \n",
       "5652          6.0                {ASSORTMENT,PROMOTIONS,DELIVERY}   \n",
       "18092         4.0    {ASSORTMENT,PRICE,PRODUCTS_QUALITY,DELIVERY}   \n",
       "13845         6.0  {DELIVERY,PROMOTIONS,PRICE,ASSORTMENT,SUPPORT}   \n",
       "25060         6.0                   {PRICE,PROMOTIONS,ASSORTMENT}   \n",
       "1428          6.0                              {PRICE,PROMOTIONS}   \n",
       "\n",
       "                                                    text  \n",
       "index                                                     \n",
       "5652   Маленький выбор товаров, хотелось бы ассортиме...  \n",
       "18092                                             Быстро  \n",
       "13845                   Доставка постоянно задерживается  \n",
       "25060                 Наценка и ассортимент расстраивают  \n",
       "1428   Можно немного скинуть минимальную сумму заказа...  "
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_idx = train.shape[0]\n",
    "df = pd.concat((train.drop(columns=target_cols), test), axis='rows')\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "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>target</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>index</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>5652</th>\n",
       "      <td>28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18092</th>\n",
       "      <td>18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13845</th>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25060</th>\n",
       "      <td>12 28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1428</th>\n",
       "      <td>8</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      target\n",
       "index       \n",
       "5652      28\n",
       "18092     18\n",
       "13845      3\n",
       "25060  12 28\n",
       "1428       8"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def trg_to_sub(t, indx=None):\n",
    "    t = np.array(t)\n",
    "    indx = range(t.shape[0]) if indx is None else indx\n",
    "    target_values = []\n",
    "    for row in t:\n",
    "        non_zero_indices = np.nonzero(np.array(row))[0]\n",
    "        if non_zero_indices.size > 0:\n",
    "            target_values.append(' '.join(map(str, non_zero_indices)))\n",
    "        else:\n",
    "            target_values.append(np.nan)\n",
    "    return pd.DataFrame({'target': target_values}, index=indx)\n",
    "\n",
    "targets = train[target_cols]\n",
    "targets_exp = trg_to_sub(targets, indx=targets.index)\n",
    "targets_exp.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## NaN filling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "assessment         0\n",
       "tags              11\n",
       "text               0\n",
       "trend_id_res0      0\n",
       "trend_id_res1      0\n",
       "trend_id_res2      0\n",
       "trend_id_res3      0\n",
       "trend_id_res4      0\n",
       "trend_id_res5      0\n",
       "trend_id_res6      0\n",
       "trend_id_res7      0\n",
       "trend_id_res8      0\n",
       "trend_id_res9      0\n",
       "trend_id_res10     0\n",
       "trend_id_res11     0\n",
       "trend_id_res12     0\n",
       "trend_id_res13     0\n",
       "trend_id_res14     0\n",
       "trend_id_res15     0\n",
       "trend_id_res16     0\n",
       "trend_id_res17     0\n",
       "trend_id_res18     0\n",
       "trend_id_res19     0\n",
       "trend_id_res20     0\n",
       "trend_id_res21     0\n",
       "trend_id_res22     0\n",
       "trend_id_res23     0\n",
       "trend_id_res24     0\n",
       "trend_id_res25     0\n",
       "trend_id_res26     0\n",
       "trend_id_res27     0\n",
       "trend_id_res28     0\n",
       "trend_id_res29     0\n",
       "trend_id_res30     0\n",
       "trend_id_res31     0\n",
       "trend_id_res32     0\n",
       "trend_id_res33     0\n",
       "trend_id_res34     0\n",
       "trend_id_res35     0\n",
       "trend_id_res36     0\n",
       "trend_id_res37     0\n",
       "trend_id_res38     0\n",
       "trend_id_res39     0\n",
       "trend_id_res40     0\n",
       "trend_id_res41     0\n",
       "trend_id_res42     0\n",
       "trend_id_res43     0\n",
       "trend_id_res44     0\n",
       "trend_id_res45     0\n",
       "trend_id_res46     0\n",
       "trend_id_res47     0\n",
       "trend_id_res48     0\n",
       "trend_id_res49     0\n",
       "dtype: int64"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train.isna().sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "assessment     0\n",
       "tags          32\n",
       "text           1\n",
       "dtype: int64"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test.isna().sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def update_df(train, test):\n",
    "    train = pd.concat((df.iloc[:train_idx], targets), axis='columns')\n",
    "    test = df.iloc[train_idx:]\n",
    "    return train, test\n",
    "\n",
    "df.fillna('', inplace=True)\n",
    "df.isna().sum().sum()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Text processing"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Tags"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "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>assessment</th>\n",
       "      <th>tags</th>\n",
       "      <th>text</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>index</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>5652</th>\n",
       "      <td>6.0</td>\n",
       "      <td>ASSORTMENT PROMOTIONS DELIVERY</td>\n",
       "      <td>Маленький выбор товаров, хотелось бы ассортиме...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18092</th>\n",
       "      <td>4.0</td>\n",
       "      <td>ASSORTMENT PRICE PRODUCTS_QUALITY DELIVERY</td>\n",
       "      <td>Быстро</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13845</th>\n",
       "      <td>6.0</td>\n",
       "      <td>DELIVERY PROMOTIONS PRICE ASSORTMENT SUPPORT</td>\n",
       "      <td>Доставка постоянно задерживается</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25060</th>\n",
       "      <td>6.0</td>\n",
       "      <td>PRICE PROMOTIONS ASSORTMENT</td>\n",
       "      <td>Наценка и ассортимент расстраивают</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1428</th>\n",
       "      <td>6.0</td>\n",
       "      <td>PRICE PROMOTIONS</td>\n",
       "      <td>Можно немного скинуть минимальную сумму заказа...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       assessment                                          tags  \\\n",
       "index                                                             \n",
       "5652          6.0                ASSORTMENT PROMOTIONS DELIVERY   \n",
       "18092         4.0    ASSORTMENT PRICE PRODUCTS_QUALITY DELIVERY   \n",
       "13845         6.0  DELIVERY PROMOTIONS PRICE ASSORTMENT SUPPORT   \n",
       "25060         6.0                   PRICE PROMOTIONS ASSORTMENT   \n",
       "1428          6.0                              PRICE PROMOTIONS   \n",
       "\n",
       "                                                    text  \n",
       "index                                                     \n",
       "5652   Маленький выбор товаров, хотелось бы ассортиме...  \n",
       "18092                                             Быстро  \n",
       "13845                   Доставка постоянно задерживается  \n",
       "25060                 Наценка и ассортимент расстраивают  \n",
       "1428   Можно немного скинуть минимальную сумму заказа...  "
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['tags'] = df['tags'].map(lambda x: x[1:-1].replace(',', ' '))\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "df['tags_num'] = df['tags'].map(lambda x: len(x.split()))\n",
    "df['tags_num'].hist(bins=df['tags_num'].nunique())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "df['no_tag'] = df['tags_num'].map(lambda x: x == 0).astype(int)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['DELIVERY',\n",
       " 'PAYMENT',\n",
       " 'ASSORTMENT',\n",
       " 'PROMOTIONS',\n",
       " 'PRICE',\n",
       " 'CATALOG_NAVIGATION',\n",
       " 'SUPPORT',\n",
       " 'PRODUCTS_QUALITY']"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_tags = set()\n",
    "for t in df.tags.values:\n",
    "    all_tags |= set(t.split())\n",
    "    \n",
    "all_tags = list(all_tags)\n",
    "all_tags"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "index\n",
       "12201    ассортимент товаров, качество доставки, удобст...\n",
       "25983    качество продуктов, цены на продукты, акции в ...\n",
       "7259     ассортимент товаров, цены на продукты, удобств...\n",
       "Name: tags_ru, dtype: object"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tag_translate = {\n",
    "    'CATALOG_NAVIGATION': 'удобство каталога товаров',\n",
    "    'PRODUCTS_QUALITY': 'качество продуктов',\n",
    "    'PAYMENT': 'платежный сервис',\n",
    "    'DELIVERY': 'качество доставки',\n",
    "    'ASSORTMENT': 'ассортимент товаров',\n",
    "    'SUPPORT': 'поддержка',\n",
    "    'PROMOTIONS': 'акции в магазине',\n",
    "    'PRICE': 'цены на продукты',\n",
    "}\n",
    "def translate_tag(tag_set):\n",
    "    for tag_en in tag_set.split():\n",
    "        tag_set = tag_set.replace(tag_en, tag_translate[tag_en] + ',')\n",
    "    return tag_set\n",
    "df['tags_ru'] = df['tags'].apply(translate_tag)\n",
    "df['tags_ru'].iloc[5:8]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "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>assessment</th>\n",
       "      <th>tags</th>\n",
       "      <th>text</th>\n",
       "      <th>tags_num</th>\n",
       "      <th>no_tag</th>\n",
       "      <th>tags_ru</th>\n",
       "      <th>ASSORTMENT</th>\n",
       "      <th>CATALOG_NAVIGATION</th>\n",
       "      <th>DELIVERY</th>\n",
       "      <th>PAYMENT</th>\n",
       "      <th>PRICE</th>\n",
       "      <th>PRODUCTS_QUALITY</th>\n",
       "      <th>PROMOTIONS</th>\n",
       "      <th>SUPPORT</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>index</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>5652</th>\n",
       "      <td>6.0</td>\n",
       "      <td>ASSORTMENT PROMOTIONS DELIVERY</td>\n",
       "      <td>Маленький выбор товаров, хотелось бы ассортиме...</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>ассортимент товаров, акции в магазине, качеств...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18092</th>\n",
       "      <td>4.0</td>\n",
       "      <td>ASSORTMENT PRICE PRODUCTS_QUALITY DELIVERY</td>\n",
       "      <td>Быстро</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>ассортимент товаров, цены на продукты, качеств...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13845</th>\n",
       "      <td>6.0</td>\n",
       "      <td>DELIVERY PROMOTIONS PRICE ASSORTMENT SUPPORT</td>\n",
       "      <td>Доставка постоянно задерживается</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>качество доставки, акции в магазине, цены на п...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25060</th>\n",
       "      <td>6.0</td>\n",
       "      <td>PRICE PROMOTIONS ASSORTMENT</td>\n",
       "      <td>Наценка и ассортимент расстраивают</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>цены на продукты, акции в магазине, ассортимен...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1428</th>\n",
       "      <td>6.0</td>\n",
       "      <td>PRICE PROMOTIONS</td>\n",
       "      <td>Можно немного скинуть минимальную сумму заказа...</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>цены на продукты, акции в магазине,</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       assessment                                          tags  \\\n",
       "index                                                             \n",
       "5652          6.0                ASSORTMENT PROMOTIONS DELIVERY   \n",
       "18092         4.0    ASSORTMENT PRICE PRODUCTS_QUALITY DELIVERY   \n",
       "13845         6.0  DELIVERY PROMOTIONS PRICE ASSORTMENT SUPPORT   \n",
       "25060         6.0                   PRICE PROMOTIONS ASSORTMENT   \n",
       "1428          6.0                              PRICE PROMOTIONS   \n",
       "\n",
       "                                                    text  tags_num  no_tag  \\\n",
       "index                                                                        \n",
       "5652   Маленький выбор товаров, хотелось бы ассортиме...         3       0   \n",
       "18092                                             Быстро         4       0   \n",
       "13845                   Доставка постоянно задерживается         5       0   \n",
       "25060                 Наценка и ассортимент расстраивают         3       0   \n",
       "1428   Можно немного скинуть минимальную сумму заказа...         2       0   \n",
       "\n",
       "                                                 tags_ru  ASSORTMENT  \\\n",
       "index                                                                  \n",
       "5652   ассортимент товаров, акции в магазине, качеств...           1   \n",
       "18092  ассортимент товаров, цены на продукты, качеств...           1   \n",
       "13845  качество доставки, акции в магазине, цены на п...           1   \n",
       "25060  цены на продукты, акции в магазине, ассортимен...           1   \n",
       "1428                 цены на продукты, акции в магазине,           0   \n",
       "\n",
       "       CATALOG_NAVIGATION  DELIVERY  PAYMENT  PRICE  PRODUCTS_QUALITY  \\\n",
       "index                                                                   \n",
       "5652                    0         1        0      0                 0   \n",
       "18092                   0         1        0      1                 1   \n",
       "13845                   0         1        0      1                 0   \n",
       "25060                   0         0        0      1                 0   \n",
       "1428                    0         0        0      1                 0   \n",
       "\n",
       "       PROMOTIONS  SUPPORT  \n",
       "index                       \n",
       "5652            1        0  \n",
       "18092           0        0  \n",
       "13845           1        1  \n",
       "25060           1        0  \n",
       "1428            1        0  "
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['tags_lst'] = df['tags'].map(lambda x: x.split())\n",
    "df_to_ohe = df.explode('tags_lst')\n",
    "df_ohe = pd.get_dummies(df_to_ohe['tags_lst'])\n",
    "df_tags = df_ohe.groupby('index').sum()\n",
    "df = df.join(df_tags)\n",
    "df.drop(columns='tags_lst', inplace=True)\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1500x800 with 4 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(15, 8))\n",
    "plt.subplot(2, 2, 1)\n",
    "plt.title('train')\n",
    "train.text.str.len().hist(bins=50)\n",
    "plt.subplot(2, 2, 2)\n",
    "plt.title('test')\n",
    "test.text.str.len().hist(bins=50)\n",
    "plt.subplot(2, 2, 3)\n",
    "plt.title('train log')\n",
    "train.text.str.len().map(lambda x: np.log(x) if x > 0 else -1).hist(bins=50)\n",
    "plt.subplot(2, 2, 4)\n",
    "plt.title('test log')\n",
    "test.text.str.len().map(lambda x: np.log(x) if x > 0 else -1).hist(bins=50)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "df['text_len'] = df['text'].str.len()\n",
    "df['text_len_log'] = df['text_len'].map(lambda x: np.log(x) if x > 0 else -1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('заказ везут [num].[num] , у вас все хорошо? никогда больше заказывать у вас не будем! отвратительная работа службы поддержки и доставки!!!!',\n",
       " 'заказ везут num num у вас все хорошо никогда больше заказывать у вас не будем отвратительная работа службы поддержки и доставки')"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "symbs = set()\n",
    "for t in df.text.values:\n",
    "    symbs.update(set(t.lower()))\n",
    "\n",
    "symbs = list(symbs)\n",
    "symbs = ''.join(symbs)\n",
    "\n",
    "trash_symbs = re.findall(r'\\W|ツ', symbs)\n",
    "trash_symbs = ''.join(trash_symbs).replace(' ', '')\n",
    "\n",
    "pattern = f'[{re.escape(trash_symbs)}]'\n",
    "\n",
    "# example\n",
    "t = df['text'].iloc[444].lower()\n",
    "t, ' '.join(re.sub(pattern, ' ', t).split())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "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>assessment</th>\n",
       "      <th>tags</th>\n",
       "      <th>text</th>\n",
       "      <th>tags_num</th>\n",
       "      <th>no_tag</th>\n",
       "      <th>tags_ru</th>\n",
       "      <th>ASSORTMENT</th>\n",
       "      <th>CATALOG_NAVIGATION</th>\n",
       "      <th>DELIVERY</th>\n",
       "      <th>PAYMENT</th>\n",
       "      <th>PRICE</th>\n",
       "      <th>PRODUCTS_QUALITY</th>\n",
       "      <th>PROMOTIONS</th>\n",
       "      <th>SUPPORT</th>\n",
       "      <th>text_len</th>\n",
       "      <th>text_len_log</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>index</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>5652</th>\n",
       "      <td>6.0</td>\n",
       "      <td>ASSORTMENT PROMOTIONS DELIVERY</td>\n",
       "      <td>маленький выбор товаров хотелось бы ассортимен...</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>ассортимент товаров, акции в магазине, качеств...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>82</td>\n",
       "      <td>4.406719</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18092</th>\n",
       "      <td>4.0</td>\n",
       "      <td>ASSORTMENT PRICE PRODUCTS_QUALITY DELIVERY</td>\n",
       "      <td>быстро</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>ассортимент товаров, цены на продукты, качеств...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>6</td>\n",
       "      <td>1.791759</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13845</th>\n",
       "      <td>6.0</td>\n",
       "      <td>DELIVERY PROMOTIONS PRICE ASSORTMENT SUPPORT</td>\n",
       "      <td>доставка постоянно задерживается</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>качество доставки, акции в магазине, цены на п...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>32</td>\n",
       "      <td>3.465736</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25060</th>\n",
       "      <td>6.0</td>\n",
       "      <td>PRICE PROMOTIONS ASSORTMENT</td>\n",
       "      <td>наценка и ассортимент расстраивают</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>цены на продукты, акции в магазине, ассортимен...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>34</td>\n",
       "      <td>3.526361</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1428</th>\n",
       "      <td>6.0</td>\n",
       "      <td>PRICE PROMOTIONS</td>\n",
       "      <td>можно немного скинуть минимальную сумму заказа...</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>цены на продукты, акции в магазине,</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>110</td>\n",
       "      <td>4.700480</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       assessment                                          tags  \\\n",
       "index                                                             \n",
       "5652          6.0                ASSORTMENT PROMOTIONS DELIVERY   \n",
       "18092         4.0    ASSORTMENT PRICE PRODUCTS_QUALITY DELIVERY   \n",
       "13845         6.0  DELIVERY PROMOTIONS PRICE ASSORTMENT SUPPORT   \n",
       "25060         6.0                   PRICE PROMOTIONS ASSORTMENT   \n",
       "1428          6.0                              PRICE PROMOTIONS   \n",
       "\n",
       "                                                    text  tags_num  no_tag  \\\n",
       "index                                                                        \n",
       "5652   маленький выбор товаров хотелось бы ассортимен...         3       0   \n",
       "18092                                             быстро         4       0   \n",
       "13845                   доставка постоянно задерживается         5       0   \n",
       "25060                 наценка и ассортимент расстраивают         3       0   \n",
       "1428   можно немного скинуть минимальную сумму заказа...         2       0   \n",
       "\n",
       "                                                 tags_ru  ASSORTMENT  \\\n",
       "index                                                                  \n",
       "5652   ассортимент товаров, акции в магазине, качеств...           1   \n",
       "18092  ассортимент товаров, цены на продукты, качеств...           1   \n",
       "13845  качество доставки, акции в магазине, цены на п...           1   \n",
       "25060  цены на продукты, акции в магазине, ассортимен...           1   \n",
       "1428                 цены на продукты, акции в магазине,           0   \n",
       "\n",
       "       CATALOG_NAVIGATION  DELIVERY  PAYMENT  PRICE  PRODUCTS_QUALITY  \\\n",
       "index                                                                   \n",
       "5652                    0         1        0      0                 0   \n",
       "18092                   0         1        0      1                 1   \n",
       "13845                   0         1        0      1                 0   \n",
       "25060                   0         0        0      1                 0   \n",
       "1428                    0         0        0      1                 0   \n",
       "\n",
       "       PROMOTIONS  SUPPORT  text_len  text_len_log  \n",
       "index                                               \n",
       "5652            1        0        82      4.406719  \n",
       "18092           0        0         6      1.791759  \n",
       "13845           1        1        32      3.465736  \n",
       "25060           1        0        34      3.526361  \n",
       "1428            1        0       110      4.700480  "
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['text'] = df['text'].map(lambda x: ' '.join(re.sub(pattern, ' ', x.lower()).split()))\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Learning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_cols = ['assessment', 'tags_num', 'no_tag', 'text_len', 'text_len_log']\n",
    "text_cols = ['tags', 'text']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "train, test = update_df(train, test)\n",
    "X = train[num_cols + text_cols]\n",
    "X_test = test[num_cols + text_cols]\n",
    "Y = train[target_cols]\n",
    "X_train, X_val, y_train, y_val = train_test_split(X, Y, test_size=0.2, random_state=SEED)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_iter = 250\n",
    "base_model = cb.CatBoostClassifier(iterations=n_iter, loss_function='CrossEntropy', \n",
    "                                   eval_metric='Accuracy', random_state=SEED, verbose=n_iter)\n",
    "multi_model = MultiOutputClassifier(estimator=base_model)\n",
    "\n",
    "multi_model.fit(X=X_train, Y=y_train,\n",
    "                text_features=text_cols,\n",
    "                plot=False,\n",
    "                early_stopping_rounds=50)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.save('../weights/cb_weights.npy', multi_model, allow_pickle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.3589189189189189"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_pred = multi_model.predict(X_val)\n",
    "val_acc = accuracy_score(y_val, y_pred)\n",
    "val_acc"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Сабмит"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "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>target</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>index</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>3135</th>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4655</th>\n",
       "      <td>12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22118</th>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23511</th>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>45</th>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      target\n",
       "index       \n",
       "3135       2\n",
       "4655      12\n",
       "22118      2\n",
       "23511      0\n",
       "45       NaN"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "multi_model = np.load('../weights/cb_weights.npy', allow_pickle=True).item()\n",
    "\n",
    "X_test = test[num_cols + text_cols]\n",
    "test_pool = cb.Pool(X_test, text_features=text_cols)\n",
    "preds = multi_model.predict(test_pool)\n",
    "sub = trg_to_sub(preds, indx=test.index)\n",
    "sub.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "metadata": {},
   "outputs": [],
   "source": [
    "sub.to_csv(DATA_DIR / f'sub_cb_{val_acc:.4f}_{SEED}.csv')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "ecom_t",
   "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": "3.12.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}