Multi_label_classification.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "ce60657d-17d2-4faf-ba3f-03948a557b39",
   "metadata": {},
   "source": [
    "# Multi-label Classification Problem"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c49e410b-1dc1-4318-acd8-f43515c7e1fc",
   "metadata": {},
   "source": [
    "### Импортируем необходимые пакеты"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "id": "a959b7cf-6699-4c8a-b214-20b0b0658be1",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import copy\n",
    "import random\n",
    "import seaborn as sns\n",
    "import math\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.patches as mpatches\n",
    "from tqdm.notebook import tqdm\n",
    "\n",
    "\n",
    "from sklearn.naive_bayes import GaussianNB,MultinomialNB\n",
    "from sklearn import metrics\n",
    "from sklearn.metrics import accuracy_score, hamming_loss, make_scorer, log_loss\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "from sklearn.base import BaseEstimator, ClassifierMixin, clone\n",
    "from sklearn.utils.validation import check_X_y, check_array, check_is_fitted\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.naive_bayes import GaussianNB\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.svm import SVC\n",
    "\n",
    "import warnings\n",
    "\n",
    "plt.style.use(\"seaborn-bright\")\n",
    "plt.style.use(\"bmh\")\n",
    "warnings.filterwarnings(\"ignore\")\n",
    "\n",
    "\n",
    "from IPython.display import display, HTML, Image\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "373e6ea9-a8b1-41a0-9946-9892d3386110",
   "metadata": {},
   "source": [
    "### Напишем функцию для отображения и подсчета пропущенных значений"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "id": "d2debcc9-cea6-4c7d-9d84-599ebb50c0dc",
   "metadata": {},
   "outputs": [],
   "source": [
    "def missing_zero_values_table(df):\n",
    "        zero_val = (df == 0.00).astype(int).sum(axis=0)\n",
    "        mis_val = df.isnull().sum()\n",
    "        mis_val_percent = 100 * df.isnull().sum() / len(df)\n",
    "        mz_table = pd.concat([zero_val, mis_val, mis_val_percent], axis=1)\n",
    "        mz_table = mz_table.rename(\n",
    "        columns = {0 : 'Zero Values', 1 : 'Missing Values', 2 : '% of Total Values'})\n",
    "        mz_table['Total Zero Missing Values'] = mz_table['Zero Values'] + mz_table['Missing Values']\n",
    "        mz_table['% Total Zero Missing Values'] = 100 * mz_table['Total Zero Missing Values'] / len(df)\n",
    "        mz_table['Data Type'] = df.dtypes\n",
    "        mz_table = mz_table[\n",
    "            mz_table.iloc[:,1] != 0].sort_values(\n",
    "        '% of Total Values', ascending=False).round(1)\n",
    "        print (\"Your selected dataframe has \" + str(df.shape[1]) + \" columns and \" + str(df.shape[0]) + \" Rows.\\n\"      \n",
    "            \"There are \" + str(mz_table.shape[0]) +\n",
    "              \" columns that have missing values.\")\n",
    "        return mz_table"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d74a1509-b0b4-4c54-b493-4d8bc49fbddb",
   "metadata": {},
   "source": [
    "### Загрузим и посмотрим на датасеты, почистим данные"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "id": "b71367b2-ca3e-4407-8c11-6c6de56c50d0",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "labels.shape: (8000, 14)\n"
     ]
    },
    {
     "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>service_a</th>\n",
       "      <th>service_b</th>\n",
       "      <th>service_c</th>\n",
       "      <th>service_d</th>\n",
       "      <th>service_e</th>\n",
       "      <th>service_f</th>\n",
       "      <th>service_g</th>\n",
       "      <th>service_h</th>\n",
       "      <th>service_i</th>\n",
       "      <th>service_j</th>\n",
       "      <th>service_k</th>\n",
       "      <th>service_l</th>\n",
       "      <th>service_m</th>\n",
       "      <th>service_n</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</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>11193</th>\n",
       "      <td>1</td>\n",
       "      <td>1</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>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11382</th>\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>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16531</th>\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>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1896</th>\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>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18262</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       service_a  service_b  service_c  service_d  service_e  service_f  \\\n",
       "id                                                                        \n",
       "11193          1          1          0          0          0          0   \n",
       "11382          0          0          0          0          0          0   \n",
       "16531          0          0          0          0          0          0   \n",
       "1896           0          0          0          1          0          0   \n",
       "18262          0          0          0          1          1          0   \n",
       "\n",
       "       service_g  service_h  service_i  service_j  service_k  service_l  \\\n",
       "id                                                                        \n",
       "11193          0          0          0          1          1          0   \n",
       "11382          0          0          0          1          1          0   \n",
       "16531          0          0          0          1          1          0   \n",
       "1896           0          0          0          1          0          1   \n",
       "18262          0          0          0          0          1          1   \n",
       "\n",
       "       service_m  service_n  \n",
       "id                           \n",
       "11193          0          0  \n",
       "11382          0          0  \n",
       "16531          0          0  \n",
       "1896           0          0  \n",
       "18262          1          0  "
      ]
     },
     "execution_count": 106,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels_df = pd.read_csv(\"problem_labels.csv\", index_col = 'id')\n",
    "print(\"labels.shape: \" + str(labels_df.shape))\n",
    "\n",
    "labels_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b0c1d763-b0d4-45a8-a995-0438200bed67",
   "metadata": {},
   "source": [
    "Проверим пропущенные значения в датасете labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "id": "c679a4b2-9e4c-429f-a6af-88977b42b695",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Your selected dataframe has 14 columns and 8000 Rows.\n",
      "There are 0 columns that have missing values.\n"
     ]
    },
    {
     "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>Zero Values</th>\n",
       "      <th>Missing Values</th>\n",
       "      <th>% of Total Values</th>\n",
       "      <th>Total Zero Missing Values</th>\n",
       "      <th>% Total Zero Missing Values</th>\n",
       "      <th>Data Type</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "Empty DataFrame\n",
       "Columns: [Zero Values, Missing Values, % of Total Values, Total Zero Missing Values, % Total Zero Missing Values, Data Type]\n",
       "Index: []"
      ]
     },
     "execution_count": 107,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "missing_zero_values_table(labels_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f4cdc767-0480-4b00-90cf-b8b5915ab7aa",
   "metadata": {},
   "source": [
    "Пропущенных значений нет. Принадлежность классу обозначается единицой, значит можем проверить количество объектов в каждом классе операцией суммирования.\n",
    "Просуммируем количество объектов в каждом классе и узнаем для каждого класса также долю объектов, принадлежащих классу, от всех объектов (в процентах)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "id": "0df735e4-2930-439b-af96-8ef0e0876a32",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "service_a    3777\n",
      "service_b    2626\n",
      "service_c    2093\n",
      "service_d     136\n",
      "service_e     431\n",
      "service_f     233\n",
      "service_g     415\n",
      "service_h    2376\n",
      "service_i     129\n",
      "service_j    6786\n",
      "service_k    6256\n",
      "service_l     864\n",
      "service_m     709\n",
      "service_n    1424\n",
      "Name: total, dtype: int64\n",
      "service_a    47.2125\n",
      "service_b    32.8250\n",
      "service_c    26.1625\n",
      "service_d     1.7000\n",
      "service_e     5.3875\n",
      "service_f     2.9125\n",
      "service_g     5.1875\n",
      "service_h    29.7000\n",
      "service_i     1.6125\n",
      "service_j    84.8250\n",
      "service_k    78.2000\n",
      "service_l    10.8000\n",
      "service_m     8.8625\n",
      "service_n    17.8000\n",
      "Name: total, dtype: float64\n"
     ]
    }
   ],
   "source": [
    "sums = labels_df.select_dtypes(pd.np.number).sum().rename('total')\n",
    "print(sums)\n",
    "print(100*sums/len(labels_df))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ff33cedf-7dd4-45af-bda0-21dc86a83c6f",
   "metadata": {},
   "source": [
    "Максимально не сбалансированными классами для обучения являются \"service_d\" и \"service_i\", к ним относятся около 130 объектов из 8000 (менее 2% объектов)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fc2d4159-9747-4645-b483-e7274b481286",
   "metadata": {},
   "source": [
    "Посмотрим на корреляцию между классами"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "id": "13e9f0a3-78e0-4fbd-90b7-ee6525f4f0b9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "corr = labels_df.corr()# plot the heatmap\n",
    "ax = sns.heatmap(corr, \n",
    "                 vmin=-1, vmax=1, center=0,\n",
    "                 cmap=sns.diverging_palette(20, 220, n=200),\n",
    "                 square=True)\n",
    "\n",
    "for i, label in enumerate(ax.get_xticklabels()):\n",
    "    ax.get_xticklabels()[i].set_text(corr.index[i].capitalize().replace('_', ' '))\n",
    "    \n",
    "ax.set_xticklabels(ax.get_xticklabels(),\n",
    "                   rotation=45,\n",
    "                   horizontalalignment='right');\n",
    "\n",
    "for i, label in enumerate(ax.get_yticklabels()):\n",
    "    ax.get_yticklabels()[i].set_text(corr.index[i].capitalize().replace('_', ' '))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dc890eb3-14c4-4f43-bd57-d76313727fe8",
   "metadata": {},
   "source": [
    "Наблюдается высокая корреляция между некоторыми парами классов, например l и m, j и k, a и b"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3d3e0a50-fe85-4570-bf44-9a3ab54df38c",
   "metadata": {},
   "source": [
    "Загрузим основной тренировочный датасет"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "id": "912c42c4-d6ac-45e8-ba80-1583be43e713",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train.shape: (8000, 1378)\n"
     ]
    },
    {
     "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>release</th>\n",
       "      <th>n_0000</th>\n",
       "      <th>n_0001</th>\n",
       "      <th>n_0002</th>\n",
       "      <th>n_0003</th>\n",
       "      <th>n_0004</th>\n",
       "      <th>n_0005</th>\n",
       "      <th>n_0006</th>\n",
       "      <th>n_0007</th>\n",
       "      <th>n_0008</th>\n",
       "      <th>...</th>\n",
       "      <th>c_1368</th>\n",
       "      <th>c_1369</th>\n",
       "      <th>c_1370</th>\n",
       "      <th>c_1371</th>\n",
       "      <th>c_1372</th>\n",
       "      <th>c_1373</th>\n",
       "      <th>c_1374</th>\n",
       "      <th>c_1375</th>\n",
       "      <th>c_1376</th>\n",
       "      <th>c_1377</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</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>11193</th>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.025449</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.368421</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>q</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11382</th>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.031297</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.315789</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16531</th>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.024475</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.342105</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>b</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1896</th>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.041694</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.447368</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18262</th>\n",
       "      <td>c</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.038120</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.315789</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>b</td>\n",
       "      <td>NaN</td>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>a</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 1378 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "      release  n_0000  n_0001    n_0002  n_0003  n_0004    n_0005  n_0006  \\\n",
       "id                                                                          \n",
       "11193       a     NaN     NaN  0.025449     NaN     NaN  0.368421     NaN   \n",
       "11382       a     NaN     NaN  0.031297     NaN     NaN  0.315789     NaN   \n",
       "16531       a     NaN     NaN  0.024475     NaN     NaN  0.342105     NaN   \n",
       "1896        a     NaN     NaN  0.041694     NaN     NaN  0.447368     NaN   \n",
       "18262       c     NaN     NaN  0.038120     NaN     NaN  0.315789     NaN   \n",
       "\n",
       "       n_0007  n_0008  ...  c_1368  c_1369  c_1370  c_1371  c_1372  c_1373  \\\n",
       "id                     ...                                                   \n",
       "11193     NaN     NaN  ...     NaN     NaN     NaN     NaN       a     NaN   \n",
       "11382     NaN     NaN  ...     NaN     NaN       a     NaN       a     NaN   \n",
       "16531     NaN     NaN  ...     NaN     NaN       a     NaN       a     NaN   \n",
       "1896      NaN     NaN  ...     NaN     NaN     NaN     NaN       a     NaN   \n",
       "18262     NaN     NaN  ...     NaN     NaN       b     NaN       a     NaN   \n",
       "\n",
       "       c_1374  c_1375  c_1376  c_1377  \n",
       "id                                     \n",
       "11193       q     NaN     NaN     NaN  \n",
       "11382     NaN     NaN     NaN     NaN  \n",
       "16531       b     NaN     NaN     NaN  \n",
       "1896      NaN     NaN     NaN     NaN  \n",
       "18262       a     NaN     NaN     NaN  \n",
       "\n",
       "[5 rows x 1378 columns]"
      ]
     },
     "execution_count": 110,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df = pd.read_csv(\"problem_train.csv\", index_col = 'id')\n",
    "print(\"train.shape: \" + str(train_df.shape))\n",
    "\n",
    "train_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d2c48c1c-cfa0-4e34-80f4-e9542bf5cd1b",
   "metadata": {},
   "source": [
    "посмотрим на описательные характеристики датасета train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "id": "f05fe4e0-b7c9-4ec3-b6d2-4d2a7832f98b",
   "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>n_0000</th>\n",
       "      <th>n_0001</th>\n",
       "      <th>n_0002</th>\n",
       "      <th>n_0003</th>\n",
       "      <th>n_0004</th>\n",
       "      <th>n_0005</th>\n",
       "      <th>n_0006</th>\n",
       "      <th>n_0007</th>\n",
       "      <th>n_0008</th>\n",
       "      <th>n_0009</th>\n",
       "      <th>...</th>\n",
       "      <th>c_1072</th>\n",
       "      <th>c_1129</th>\n",
       "      <th>c_1148</th>\n",
       "      <th>c_1168</th>\n",
       "      <th>c_1230</th>\n",
       "      <th>c_1257</th>\n",
       "      <th>c_1277</th>\n",
       "      <th>c_1281</th>\n",
       "      <th>c_1308</th>\n",
       "      <th>c_1332</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>12.000000</td>\n",
       "      <td>388.000000</td>\n",
       "      <td>7662.000000</td>\n",
       "      <td>112.000000</td>\n",
       "      <td>58.000000</td>\n",
       "      <td>7658.000000</td>\n",
       "      <td>1348.000000</td>\n",
       "      <td>995.000000</td>\n",
       "      <td>5.000000</td>\n",
       "      <td>34.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>0.307692</td>\n",
       "      <td>0.655744</td>\n",
       "      <td>0.040935</td>\n",
       "      <td>0.102749</td>\n",
       "      <td>0.352874</td>\n",
       "      <td>0.395981</td>\n",
       "      <td>0.193175</td>\n",
       "      <td>0.012012</td>\n",
       "      <td>0.422222</td>\n",
       "      <td>0.129241</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>0.245454</td>\n",
       "      <td>0.216399</td>\n",
       "      <td>0.024599</td>\n",
       "      <td>0.153007</td>\n",
       "      <td>0.209337</td>\n",
       "      <td>0.085000</td>\n",
       "      <td>0.236683</td>\n",
       "      <td>0.041382</td>\n",
       "      <td>0.395967</td>\n",
       "      <td>0.162008</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.066667</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>0.134615</td>\n",
       "      <td>0.428571</td>\n",
       "      <td>0.031974</td>\n",
       "      <td>0.011905</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.342105</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.111111</td>\n",
       "      <td>0.015560</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>0.307692</td>\n",
       "      <td>0.702381</td>\n",
       "      <td>0.041369</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.394737</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.444444</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>0.403846</td>\n",
       "      <td>0.857143</td>\n",
       "      <td>0.048950</td>\n",
       "      <td>0.142857</td>\n",
       "      <td>0.525000</td>\n",
       "      <td>0.421053</td>\n",
       "      <td>0.400000</td>\n",
       "      <td>0.011905</td>\n",
       "      <td>0.555556</td>\n",
       "      <td>0.186722</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>0.846154</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.833333</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.597510</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>8 rows × 353 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          n_0000      n_0001       n_0002      n_0003     n_0004       n_0005  \\\n",
       "count  12.000000  388.000000  7662.000000  112.000000  58.000000  7658.000000   \n",
       "mean    0.307692    0.655744     0.040935    0.102749   0.352874     0.395981   \n",
       "std     0.245454    0.216399     0.024599    0.153007   0.209337     0.085000   \n",
       "min     0.000000    0.000000     0.000000    0.000000   0.066667     0.000000   \n",
       "25%     0.134615    0.428571     0.031974    0.011905   0.200000     0.342105   \n",
       "50%     0.307692    0.702381     0.041369    0.047619   0.283333     0.394737   \n",
       "75%     0.403846    0.857143     0.048950    0.142857   0.525000     0.421053   \n",
       "max     0.846154    1.000000     1.000000    1.000000   0.833333     1.000000   \n",
       "\n",
       "            n_0006      n_0007    n_0008     n_0009  ...  c_1072  c_1129  \\\n",
       "count  1348.000000  995.000000  5.000000  34.000000  ...     0.0     0.0   \n",
       "mean      0.193175    0.012012  0.422222   0.129241  ...     NaN     NaN   \n",
       "std       0.236683    0.041382  0.395967   0.162008  ...     NaN     NaN   \n",
       "min       0.000000    0.000000  0.000000   0.000000  ...     NaN     NaN   \n",
       "25%       0.000000    0.000000  0.111111   0.015560  ...     NaN     NaN   \n",
       "50%       0.200000    0.000000  0.444444   0.049793  ...     NaN     NaN   \n",
       "75%       0.400000    0.011905  0.555556   0.186722  ...     NaN     NaN   \n",
       "max       1.000000    1.000000  1.000000   0.597510  ...     NaN     NaN   \n",
       "\n",
       "       c_1148  c_1168  c_1230  c_1257  c_1277  c_1281  c_1308  c_1332  \n",
       "count     0.0     0.0     0.0     0.0     0.0     0.0     0.0     0.0  \n",
       "mean      NaN     NaN     NaN     NaN     NaN     NaN     NaN     NaN  \n",
       "std       NaN     NaN     NaN     NaN     NaN     NaN     NaN     NaN  \n",
       "min       NaN     NaN     NaN     NaN     NaN     NaN     NaN     NaN  \n",
       "25%       NaN     NaN     NaN     NaN     NaN     NaN     NaN     NaN  \n",
       "50%       NaN     NaN     NaN     NaN     NaN     NaN     NaN     NaN  \n",
       "75%       NaN     NaN     NaN     NaN     NaN     NaN     NaN     NaN  \n",
       "max       NaN     NaN     NaN     NaN     NaN     NaN     NaN     NaN  \n",
       "\n",
       "[8 rows x 353 columns]"
      ]
     },
     "execution_count": 111,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "22b6a92c-8a33-4c75-a26a-1d09a5425cd1",
   "metadata": {},
   "source": [
    "Проверим пропущенные значения в датасете train, сохраним результаты для дальнейшей визуализации"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "id": "f6aeb789-4f29-4925-83ab-6e651d29ed9e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Your selected dataframe has 1378 columns and 8000 Rows.\n",
      "There are 1347 columns that have missing values.\n"
     ]
    },
    {
     "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>Zero Values</th>\n",
       "      <th>Missing Values</th>\n",
       "      <th>% of Total Values</th>\n",
       "      <th>Total Zero Missing Values</th>\n",
       "      <th>% Total Zero Missing Values</th>\n",
       "      <th>Data Type</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>c_1129</th>\n",
       "      <td>0</td>\n",
       "      <td>8000</td>\n",
       "      <td>100.0</td>\n",
       "      <td>8000</td>\n",
       "      <td>100.0</td>\n",
       "      <td>float64</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0710</th>\n",
       "      <td>0</td>\n",
       "      <td>8000</td>\n",
       "      <td>100.0</td>\n",
       "      <td>8000</td>\n",
       "      <td>100.0</td>\n",
       "      <td>float64</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0529</th>\n",
       "      <td>0</td>\n",
       "      <td>8000</td>\n",
       "      <td>100.0</td>\n",
       "      <td>8000</td>\n",
       "      <td>100.0</td>\n",
       "      <td>float64</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0902</th>\n",
       "      <td>0</td>\n",
       "      <td>8000</td>\n",
       "      <td>100.0</td>\n",
       "      <td>8000</td>\n",
       "      <td>100.0</td>\n",
       "      <td>float64</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_1281</th>\n",
       "      <td>0</td>\n",
       "      <td>8000</td>\n",
       "      <td>100.0</td>\n",
       "      <td>8000</td>\n",
       "      <td>100.0</td>\n",
       "      <td>float64</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0582</th>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>0.1</td>\n",
       "      <td>5</td>\n",
       "      <td>0.1</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0709</th>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>0.1</td>\n",
       "      <td>5</td>\n",
       "      <td>0.1</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0975</th>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>0.0</td>\n",
       "      <td>4</td>\n",
       "      <td>0.0</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0996</th>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>0.0</td>\n",
       "      <td>3</td>\n",
       "      <td>0.0</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_1372</th>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1347 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        Zero Values  Missing Values  % of Total Values  \\\n",
       "c_1129            0            8000              100.0   \n",
       "c_0710            0            8000              100.0   \n",
       "c_0529            0            8000              100.0   \n",
       "c_0902            0            8000              100.0   \n",
       "c_1281            0            8000              100.0   \n",
       "...             ...             ...                ...   \n",
       "c_0582            0               5                0.1   \n",
       "c_0709            0               5                0.1   \n",
       "c_0975            0               4                0.0   \n",
       "c_0996            0               3                0.0   \n",
       "c_1372            0               2                0.0   \n",
       "\n",
       "        Total Zero Missing Values  % Total Zero Missing Values Data Type  \n",
       "c_1129                       8000                        100.0   float64  \n",
       "c_0710                       8000                        100.0   float64  \n",
       "c_0529                       8000                        100.0   float64  \n",
       "c_0902                       8000                        100.0   float64  \n",
       "c_1281                       8000                        100.0   float64  \n",
       "...                           ...                          ...       ...  \n",
       "c_0582                          5                          0.1    object  \n",
       "c_0709                          5                          0.1    object  \n",
       "c_0975                          4                          0.0    object  \n",
       "c_0996                          3                          0.0    object  \n",
       "c_1372                          2                          0.0    object  \n",
       "\n",
       "[1347 rows x 6 columns]"
      ]
     },
     "execution_count": 112,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df_missing_table = missing_zero_values_table(train_df)\n",
    "train_df_missing_table"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d4e073a7-fed8-46b0-ba64-477d0413f63f",
   "metadata": {},
   "source": [
    "Как видим, есть фичи, состоящие полностью из пропущенных значений, смысловой нагрузки они не несут. Попробуем визуализировать плотность распределения пропущенных значений от количества пропущенных значений. Для наглядности отметим уровни в 95% и 90% пропущенных значений"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "id": "a0ec1e25-31ee-4475-b988-7fd9a6074061",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x864 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(20,12))\n",
    "plotMV = sns.kdeplot(train_df_missing_table['% Total Zero Missing Values'], label='Missing_Values')\n",
    "plt.xlabel(\"% Total Zero Missing Values\")\n",
    "plt.title(\"Densities of Missing Values\")\n",
    "plotMV.set_xlim(0, 100)\n",
    "xcoords = [95, 90]\n",
    "for xc in xcoords:\n",
    "    plt.axvline(x=xc, ls=\":\", lw = 2, color = \"black\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "43ecc5f1-6741-49f3-9224-0c305b831a7e",
   "metadata": {},
   "source": [
    "Как видим, значительная часть фичей имеет более 90% пропущенных значений.\n",
    "Однако здесь же стоит отметить, что у нас есть классы, в которых около 130 объектов, что составляет менее 2% всех объектов. \n",
    "С целью минимизации количества колонок я выбрал в качестве трешхолда на минимальное количество не пропущенных значений в датасете уровень 10% от 130 объектов, т.е. 13. То есть я оставил колонки в которых пропущенных значений не больше: 8000 - (130 * 10%) = 7987"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "id": "7646bfb7-d889-4126-8552-7e0259c075ab",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train.shape: (8000, 1048)\n"
     ]
    },
    {
     "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>n_0001</th>\n",
       "      <th>n_0002</th>\n",
       "      <th>n_0003</th>\n",
       "      <th>n_0004</th>\n",
       "      <th>n_0005</th>\n",
       "      <th>n_0006</th>\n",
       "      <th>n_0007</th>\n",
       "      <th>n_0009</th>\n",
       "      <th>n_0010</th>\n",
       "      <th>n_0012</th>\n",
       "      <th>...</th>\n",
       "      <th>o_0312</th>\n",
       "      <th>o_0313</th>\n",
       "      <th>o_0314</th>\n",
       "      <th>o_0315</th>\n",
       "      <th>o_0316</th>\n",
       "      <th>o_0319</th>\n",
       "      <th>o_0322</th>\n",
       "      <th>o_0323</th>\n",
       "      <th>o_0324</th>\n",
       "      <th>o_0326</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>388.000000</td>\n",
       "      <td>7662.000000</td>\n",
       "      <td>112.000000</td>\n",
       "      <td>58.000000</td>\n",
       "      <td>7658.000000</td>\n",
       "      <td>1348.000000</td>\n",
       "      <td>995.000000</td>\n",
       "      <td>34.000000</td>\n",
       "      <td>26.000000</td>\n",
       "      <td>5536.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>270.000000</td>\n",
       "      <td>76.000000</td>\n",
       "      <td>1695.00000</td>\n",
       "      <td>4001.000000</td>\n",
       "      <td>397.000000</td>\n",
       "      <td>671.000000</td>\n",
       "      <td>182.000000</td>\n",
       "      <td>1694.000000</td>\n",
       "      <td>550.000000</td>\n",
       "      <td>743.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>0.655744</td>\n",
       "      <td>0.040935</td>\n",
       "      <td>0.102749</td>\n",
       "      <td>0.352874</td>\n",
       "      <td>0.395981</td>\n",
       "      <td>0.193175</td>\n",
       "      <td>0.012012</td>\n",
       "      <td>0.129241</td>\n",
       "      <td>0.361538</td>\n",
       "      <td>0.175514</td>\n",
       "      <td>...</td>\n",
       "      <td>16.488889</td>\n",
       "      <td>2.078947</td>\n",
       "      <td>0.00354</td>\n",
       "      <td>0.086978</td>\n",
       "      <td>5.080605</td>\n",
       "      <td>5.651267</td>\n",
       "      <td>18.675824</td>\n",
       "      <td>0.009445</td>\n",
       "      <td>1.209091</td>\n",
       "      <td>5.282638</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>0.216399</td>\n",
       "      <td>0.024599</td>\n",
       "      <td>0.153007</td>\n",
       "      <td>0.209337</td>\n",
       "      <td>0.085000</td>\n",
       "      <td>0.236683</td>\n",
       "      <td>0.041382</td>\n",
       "      <td>0.162008</td>\n",
       "      <td>0.204112</td>\n",
       "      <td>0.095125</td>\n",
       "      <td>...</td>\n",
       "      <td>5.863207</td>\n",
       "      <td>1.572795</td>\n",
       "      <td>0.06863</td>\n",
       "      <td>0.322386</td>\n",
       "      <td>3.416914</td>\n",
       "      <td>3.442577</td>\n",
       "      <td>6.327120</td>\n",
       "      <td>0.137157</td>\n",
       "      <td>0.699928</td>\n",
       "      <td>3.470446</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.066667</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>0.428571</td>\n",
       "      <td>0.031974</td>\n",
       "      <td>0.011905</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.342105</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.015560</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.097561</td>\n",
       "      <td>...</td>\n",
       "      <td>12.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>14.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>2.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>0.702381</td>\n",
       "      <td>0.041369</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.394737</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.300000</td>\n",
       "      <td>0.170732</td>\n",
       "      <td>...</td>\n",
       "      <td>17.000000</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>5.000000</td>\n",
       "      <td>6.000000</td>\n",
       "      <td>19.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>5.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>0.857143</td>\n",
       "      <td>0.048950</td>\n",
       "      <td>0.142857</td>\n",
       "      <td>0.525000</td>\n",
       "      <td>0.421053</td>\n",
       "      <td>0.400000</td>\n",
       "      <td>0.011905</td>\n",
       "      <td>0.186722</td>\n",
       "      <td>0.400000</td>\n",
       "      <td>0.256098</td>\n",
       "      <td>...</td>\n",
       "      <td>21.000000</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>8.000000</td>\n",
       "      <td>9.000000</td>\n",
       "      <td>23.750000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>8.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.833333</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.597510</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.353659</td>\n",
       "      <td>...</td>\n",
       "      <td>29.000000</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>2.00000</td>\n",
       "      <td>5.000000</td>\n",
       "      <td>11.000000</td>\n",
       "      <td>11.000000</td>\n",
       "      <td>32.000000</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>11.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>8 rows × 253 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "           n_0001       n_0002      n_0003     n_0004       n_0005  \\\n",
       "count  388.000000  7662.000000  112.000000  58.000000  7658.000000   \n",
       "mean     0.655744     0.040935    0.102749   0.352874     0.395981   \n",
       "std      0.216399     0.024599    0.153007   0.209337     0.085000   \n",
       "min      0.000000     0.000000    0.000000   0.066667     0.000000   \n",
       "25%      0.428571     0.031974    0.011905   0.200000     0.342105   \n",
       "50%      0.702381     0.041369    0.047619   0.283333     0.394737   \n",
       "75%      0.857143     0.048950    0.142857   0.525000     0.421053   \n",
       "max      1.000000     1.000000    1.000000   0.833333     1.000000   \n",
       "\n",
       "            n_0006      n_0007     n_0009     n_0010       n_0012  ...  \\\n",
       "count  1348.000000  995.000000  34.000000  26.000000  5536.000000  ...   \n",
       "mean      0.193175    0.012012   0.129241   0.361538     0.175514  ...   \n",
       "std       0.236683    0.041382   0.162008   0.204112     0.095125  ...   \n",
       "min       0.000000    0.000000   0.000000   0.100000     0.000000  ...   \n",
       "25%       0.000000    0.000000   0.015560   0.200000     0.097561  ...   \n",
       "50%       0.200000    0.000000   0.049793   0.300000     0.170732  ...   \n",
       "75%       0.400000    0.011905   0.186722   0.400000     0.256098  ...   \n",
       "max       1.000000    1.000000   0.597510   1.000000     0.353659  ...   \n",
       "\n",
       "           o_0312     o_0313      o_0314       o_0315      o_0316      o_0319  \\\n",
       "count  270.000000  76.000000  1695.00000  4001.000000  397.000000  671.000000   \n",
       "mean    16.488889   2.078947     0.00354     0.086978    5.080605    5.651267   \n",
       "std      5.863207   1.572795     0.06863     0.322386    3.416914    3.442577   \n",
       "min      0.000000   0.000000     0.00000     0.000000    0.000000    0.000000   \n",
       "25%     12.000000   1.000000     0.00000     0.000000    2.000000    3.000000   \n",
       "50%     17.000000   2.000000     0.00000     0.000000    5.000000    6.000000   \n",
       "75%     21.000000   4.000000     0.00000     0.000000    8.000000    9.000000   \n",
       "max     29.000000   4.000000     2.00000     5.000000   11.000000   11.000000   \n",
       "\n",
       "           o_0322       o_0323      o_0324      o_0326  \n",
       "count  182.000000  1694.000000  550.000000  743.000000  \n",
       "mean    18.675824     0.009445    1.209091    5.282638  \n",
       "std      6.327120     0.137157    0.699928    3.470446  \n",
       "min      0.000000     0.000000    0.000000    0.000000  \n",
       "25%     14.000000     0.000000    1.000000    2.000000  \n",
       "50%     19.000000     0.000000    1.000000    5.000000  \n",
       "75%     23.750000     0.000000    2.000000    8.000000  \n",
       "max     32.000000     3.000000    2.000000   11.000000  \n",
       "\n",
       "[8 rows x 253 columns]"
      ]
     },
     "execution_count": 114,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df = train_df.dropna(thresh=len(train_df) - 7987, axis=1)\n",
    "print(\"train.shape: \" + str(train_df.shape))\n",
    "train_df.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1021e3ab-7d0d-411b-a7f7-8745324bb5a1",
   "metadata": {},
   "source": [
    "Проверим снова пропущенные значения в датасете train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "id": "037f9ef6-fb76-452c-a9d5-28bc9265b22d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Your selected dataframe has 1048 columns and 8000 Rows.\n",
      "There are 1017 columns that have missing values.\n"
     ]
    },
    {
     "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>Zero Values</th>\n",
       "      <th>Missing Values</th>\n",
       "      <th>% of Total Values</th>\n",
       "      <th>Total Zero Missing Values</th>\n",
       "      <th>% Total Zero Missing Values</th>\n",
       "      <th>Data Type</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>c_0780</th>\n",
       "      <td>0</td>\n",
       "      <td>7987</td>\n",
       "      <td>99.8</td>\n",
       "      <td>7987</td>\n",
       "      <td>99.8</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0501</th>\n",
       "      <td>0</td>\n",
       "      <td>7987</td>\n",
       "      <td>99.8</td>\n",
       "      <td>7987</td>\n",
       "      <td>99.8</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>o_0131</th>\n",
       "      <td>1</td>\n",
       "      <td>7987</td>\n",
       "      <td>99.8</td>\n",
       "      <td>7988</td>\n",
       "      <td>99.8</td>\n",
       "      <td>float64</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0976</th>\n",
       "      <td>0</td>\n",
       "      <td>7987</td>\n",
       "      <td>99.8</td>\n",
       "      <td>7987</td>\n",
       "      <td>99.8</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0849</th>\n",
       "      <td>0</td>\n",
       "      <td>7987</td>\n",
       "      <td>99.8</td>\n",
       "      <td>7987</td>\n",
       "      <td>99.8</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0709</th>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>0.1</td>\n",
       "      <td>5</td>\n",
       "      <td>0.1</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0582</th>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>0.1</td>\n",
       "      <td>5</td>\n",
       "      <td>0.1</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0975</th>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>0.0</td>\n",
       "      <td>4</td>\n",
       "      <td>0.0</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_0996</th>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>0.0</td>\n",
       "      <td>3</td>\n",
       "      <td>0.0</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c_1372</th>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1017 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        Zero Values  Missing Values  % of Total Values  \\\n",
       "c_0780            0            7987               99.8   \n",
       "c_0501            0            7987               99.8   \n",
       "o_0131            1            7987               99.8   \n",
       "c_0976            0            7987               99.8   \n",
       "c_0849            0            7987               99.8   \n",
       "...             ...             ...                ...   \n",
       "c_0709            0               5                0.1   \n",
       "c_0582            0               5                0.1   \n",
       "c_0975            0               4                0.0   \n",
       "c_0996            0               3                0.0   \n",
       "c_1372            0               2                0.0   \n",
       "\n",
       "        Total Zero Missing Values  % Total Zero Missing Values Data Type  \n",
       "c_0780                       7987                         99.8    object  \n",
       "c_0501                       7987                         99.8    object  \n",
       "o_0131                       7988                         99.8   float64  \n",
       "c_0976                       7987                         99.8    object  \n",
       "c_0849                       7987                         99.8    object  \n",
       "...                           ...                          ...       ...  \n",
       "c_0709                          5                          0.1    object  \n",
       "c_0582                          5                          0.1    object  \n",
       "c_0975                          4                          0.0    object  \n",
       "c_0996                          3                          0.0    object  \n",
       "c_1372                          2                          0.0    object  \n",
       "\n",
       "[1017 rows x 6 columns]"
      ]
     },
     "execution_count": 115,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "stats_table = missing_zero_values_table(train_df)\n",
    "stats_table"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6b21cec7-9cfa-4a52-a9c8-6fefbe69e1a0",
   "metadata": {},
   "source": [
    "Количество колонок сократилось с 1378 до 1048 (примерно на четверть)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0c84f847-c788-4e85-aca1-bcb68b553931",
   "metadata": {},
   "source": [
    "Проверим наличие в train датасете строк, в которых нет значений"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "id": "6fd9b36b-19e6-497c-9206-134070fdda17",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(8000, 1048)"
      ]
     },
     "execution_count": 116,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df_row_test = train_df.dropna(axis = 0, how = 'all')\n",
    "train_df_row_test.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "75685833-cc65-414d-9971-9bf2b778677f",
   "metadata": {},
   "source": [
    "Убедились, что таких строк нет"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9a593af8-65ec-4c79-9817-bf73c41f992d",
   "metadata": {},
   "source": [
    "Загрузим тестовый датасет, приведем его колонки в соотвествие с очищенным тренировочным датасетом. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "id": "8a974e41-6f35-4dd7-a1f2-d663aaabf525",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2000, 1048)"
      ]
     },
     "execution_count": 117,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df = pd.read_csv(\"problem_test.csv\", index_col = 'id')\n",
    "common_cols_train_test = list(set(test_df.columns).intersection(train_df.columns))\n",
    "test_df = test_df[common_cols_train_test]\n",
    "test_df.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eba2b047-ee0e-4fef-9f70-5a1792033efe",
   "metadata": {},
   "source": [
    "В дальнейшем в процессе классификации хотелось бы использовать бинарную классификацию для каждого лейбла. Для этого переведем категориальные переменные в количественные, используя one-hot encoding. Проводить данную операцию я буду над датасетом, состоящим из тренировочного и тестового, тк вероятна ситуация появления в тестовом датасете объектов, отсутствующих в тренировочном датасете."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "id": "16667f1d-5810-4d1c-816a-3990872f96e8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10000, 1048)"
      ]
     },
     "execution_count": 118,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_for_encoding = pd.concat([train_df, test_df], ignore_index=False)\n",
    "df_for_encoding.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "id": "590ac741-07c5-47da-a263-2c5b0f01ed1b",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "       n_0001    n_0002  n_0003  n_0004    n_0005  n_0006    n_0007  n_0009  \\\n",
      "id                                                                            \n",
      "11193     NaN  0.025449     NaN     NaN  0.368421     NaN       NaN     NaN   \n",
      "11382     NaN  0.031297     NaN     NaN  0.315789     NaN       NaN     NaN   \n",
      "16531     NaN  0.024475     NaN     NaN  0.342105     NaN       NaN     NaN   \n",
      "1896      NaN  0.041694     NaN     NaN  0.447368     NaN       NaN     NaN   \n",
      "18262     NaN  0.038120     NaN     NaN  0.315789     NaN       NaN     NaN   \n",
      "...       ...       ...     ...     ...       ...     ...       ...     ...   \n",
      "200       NaN  0.014728     NaN     NaN  0.368421     NaN  0.059524     NaN   \n",
      "10876     NaN  0.052415     NaN     NaN  0.578947     NaN  0.000000     NaN   \n",
      "4035      NaN  0.038770     NaN     NaN  0.368421     NaN       NaN     NaN   \n",
      "3640      NaN  0.039420     NaN     NaN  0.552632     NaN       NaN     NaN   \n",
      "157       NaN  0.043535     NaN     NaN  0.421053     NaN       NaN     NaN   \n",
      "\n",
      "       n_0010    n_0012  ...  c_1375_m  c_1375_n  c_1375_o  c_1375_p  \\\n",
      "id                       ...                                           \n",
      "11193     NaN  0.292683  ...         0         0         0         0   \n",
      "11382     NaN  0.243902  ...         0         0         0         0   \n",
      "16531     NaN  0.304878  ...         0         0         0         0   \n",
      "1896      NaN  0.207317  ...         0         0         0         0   \n",
      "18262     NaN  0.219512  ...         0         0         0         0   \n",
      "...       ...       ...  ...       ...       ...       ...       ...   \n",
      "200       NaN       NaN  ...         0         0         0         0   \n",
      "10876     NaN  0.170732  ...         0         0         0         0   \n",
      "4035      NaN  0.170732  ...         0         0         0         0   \n",
      "3640      NaN       NaN  ...         0         0         0         0   \n",
      "157       NaN       NaN  ...         0         0         0         0   \n",
      "\n",
      "       c_1375_q  c_1375_r  c_1375_s  c_1375_t  c_1377_a  c_1377_b  \n",
      "id                                                                 \n",
      "11193         0         0         0         0         0         0  \n",
      "11382         0         0         0         0         0         0  \n",
      "16531         0         0         0         0         0         0  \n",
      "1896          0         0         0         0         0         0  \n",
      "18262         0         0         0         0         0         0  \n",
      "...         ...       ...       ...       ...       ...       ...  \n",
      "200           0         0         0         0         0         0  \n",
      "10876         0         0         0         0         0         0  \n",
      "4035          0         0         0         0         0         0  \n",
      "3640          0         0         0         0         0         0  \n",
      "157           0         0         0         0         0         0  \n",
      "\n",
      "[10000 rows x 5414 columns]\n"
     ]
    }
   ],
   "source": [
    "categorical_cols2 = df_for_encoding.select_dtypes(include=['object']).columns\n",
    "df_for_encoding = pd.get_dummies(df_for_encoding, columns=categorical_cols2, prefix=categorical_cols2)\n",
    "print(df_for_encoding)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a3123f37-c1af-4a9b-bfb3-f6073050ab9c",
   "metadata": {},
   "source": [
    "Разделим получившийся датасет обратно на тренировчный и тестовый по первоначальному количеству строк (8000 и 2000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "id": "60322a65-b795-4d64-89ad-9aa9dbd74503",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(8000, 5414)"
      ]
     },
     "execution_count": 120,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df_copy = df_for_encoding.head(8000)\n",
    "test_df_copy = df_for_encoding.tail(len(df_for_encoding) - 8000)\n",
    "train_df_copy.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f034565c-c5b1-4944-8ed3-8bfa93f0e129",
   "metadata": {},
   "source": [
    "Для бинарной классификации также потребуется отсутствие пропущенных значений. Заменим их в тестовом и тренировочном датасетах на медианное значение. Также проводил тест с заменой пропущенных значений наиболее часто встречающимся значением по столбцу (модой), результаты были хуже, остальной код был таким же, поэтому не вижу смысла оставлять это в  описании пути поиска решения."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "id": "55149cb6-e454-423f-b7aa-2c269d7f2ed5",
   "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>n_0001</th>\n",
       "      <th>n_0002</th>\n",
       "      <th>n_0003</th>\n",
       "      <th>n_0004</th>\n",
       "      <th>n_0005</th>\n",
       "      <th>n_0006</th>\n",
       "      <th>n_0007</th>\n",
       "      <th>n_0009</th>\n",
       "      <th>n_0010</th>\n",
       "      <th>n_0012</th>\n",
       "      <th>...</th>\n",
       "      <th>c_1375_m</th>\n",
       "      <th>c_1375_n</th>\n",
       "      <th>c_1375_o</th>\n",
       "      <th>c_1375_p</th>\n",
       "      <th>c_1375_q</th>\n",
       "      <th>c_1375_r</th>\n",
       "      <th>c_1375_s</th>\n",
       "      <th>c_1375_t</th>\n",
       "      <th>c_1377_a</th>\n",
       "      <th>c_1377_b</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</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>11193</th>\n",
       "      <td>0.702381</td>\n",
       "      <td>0.025449</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.368421</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.292683</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>11382</th>\n",
       "      <td>0.702381</td>\n",
       "      <td>0.031297</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.315789</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.243902</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>16531</th>\n",
       "      <td>0.702381</td>\n",
       "      <td>0.024475</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.342105</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.304878</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>1896</th>\n",
       "      <td>0.702381</td>\n",
       "      <td>0.041694</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.447368</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.207317</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>18262</th>\n",
       "      <td>0.702381</td>\n",
       "      <td>0.038120</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.315789</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.219512</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>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10898</th>\n",
       "      <td>0.702381</td>\n",
       "      <td>0.053931</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.394737</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.060976</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>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16664</th>\n",
       "      <td>0.702381</td>\n",
       "      <td>0.031731</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.394737</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.268293</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>5334</th>\n",
       "      <td>0.904762</td>\n",
       "      <td>0.033463</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.394737</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.304878</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>7905</th>\n",
       "      <td>0.702381</td>\n",
       "      <td>0.047109</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.289474</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.146341</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>3997</th>\n",
       "      <td>0.761905</td>\n",
       "      <td>0.049491</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.631579</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.243902</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>8000 rows × 5414 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "         n_0001    n_0002    n_0003    n_0004    n_0005  n_0006  n_0007  \\\n",
       "id                                                                        \n",
       "11193  0.702381  0.025449  0.047619  0.283333  0.368421     0.2     0.0   \n",
       "11382  0.702381  0.031297  0.047619  0.283333  0.315789     0.2     0.0   \n",
       "16531  0.702381  0.024475  0.047619  0.283333  0.342105     0.2     0.0   \n",
       "1896   0.702381  0.041694  0.047619  0.283333  0.447368     0.2     0.0   \n",
       "18262  0.702381  0.038120  0.047619  0.283333  0.315789     0.2     0.0   \n",
       "...         ...       ...       ...       ...       ...     ...     ...   \n",
       "10898  0.702381  0.053931  0.047619  0.283333  0.394737     0.2     0.0   \n",
       "16664  0.702381  0.031731  0.047619  0.283333  0.394737     0.2     0.0   \n",
       "5334   0.904762  0.033463  0.047619  0.283333  0.394737     0.2     0.0   \n",
       "7905   0.702381  0.047109  0.047619  0.283333  0.289474     0.2     0.0   \n",
       "3997   0.761905  0.049491  0.047619  0.283333  0.631579     0.0     0.0   \n",
       "\n",
       "         n_0009  n_0010    n_0012  ...  c_1375_m  c_1375_n  c_1375_o  \\\n",
       "id                                 ...                                 \n",
       "11193  0.049793     0.3  0.292683  ...         0         0         0   \n",
       "11382  0.049793     0.3  0.243902  ...         0         0         0   \n",
       "16531  0.049793     0.3  0.304878  ...         0         0         0   \n",
       "1896   0.049793     0.3  0.207317  ...         0         0         0   \n",
       "18262  0.049793     0.3  0.219512  ...         0         0         0   \n",
       "...         ...     ...       ...  ...       ...       ...       ...   \n",
       "10898  0.049793     0.3  0.060976  ...         0         0         0   \n",
       "16664  0.049793     0.3  0.268293  ...         0         0         0   \n",
       "5334   0.049793     0.3  0.304878  ...         0         0         0   \n",
       "7905   0.049793     0.3  0.146341  ...         0         0         0   \n",
       "3997   0.049793     0.3  0.243902  ...         0         0         0   \n",
       "\n",
       "       c_1375_p  c_1375_q  c_1375_r  c_1375_s  c_1375_t  c_1377_a  c_1377_b  \n",
       "id                                                                           \n",
       "11193         0         0         0         0         0         0         0  \n",
       "11382         0         0         0         0         0         0         0  \n",
       "16531         0         0         0         0         0         0         0  \n",
       "1896          0         0         0         0         0         0         0  \n",
       "18262         0         0         0         0         0         0         0  \n",
       "...         ...       ...       ...       ...       ...       ...       ...  \n",
       "10898         0         0         0         0         0         0         1  \n",
       "16664         0         0         0         0         0         0         0  \n",
       "5334          0         0         0         0         0         0         0  \n",
       "7905          0         0         0         0         0         0         0  \n",
       "3997          0         0         0         0         0         0         0  \n",
       "\n",
       "[8000 rows x 5414 columns]"
      ]
     },
     "execution_count": 121,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "median = train_df_copy.median().to_dict()\n",
    "train_df_copy.fillna(median, inplace=True)\n",
    "train_df_copy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "id": "ad8b3748-0593-4dab-bca3-f271c310f3b5",
   "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>n_0001</th>\n",
       "      <th>n_0002</th>\n",
       "      <th>n_0003</th>\n",
       "      <th>n_0004</th>\n",
       "      <th>n_0005</th>\n",
       "      <th>n_0006</th>\n",
       "      <th>n_0007</th>\n",
       "      <th>n_0009</th>\n",
       "      <th>n_0010</th>\n",
       "      <th>n_0012</th>\n",
       "      <th>...</th>\n",
       "      <th>c_1375_m</th>\n",
       "      <th>c_1375_n</th>\n",
       "      <th>c_1375_o</th>\n",
       "      <th>c_1375_p</th>\n",
       "      <th>c_1375_q</th>\n",
       "      <th>c_1375_r</th>\n",
       "      <th>c_1375_s</th>\n",
       "      <th>c_1375_t</th>\n",
       "      <th>c_1377_a</th>\n",
       "      <th>c_1377_b</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</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>7957</th>\n",
       "      <td>0.678571</td>\n",
       "      <td>0.029890</td>\n",
       "      <td>0.063492</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.368421</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.329268</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>693</th>\n",
       "      <td>0.452381</td>\n",
       "      <td>0.032164</td>\n",
       "      <td>0.063492</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.421053</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.341463</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>15558</th>\n",
       "      <td>0.547619</td>\n",
       "      <td>0.049166</td>\n",
       "      <td>0.063492</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.219512</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>15614</th>\n",
       "      <td>0.678571</td>\n",
       "      <td>0.048733</td>\n",
       "      <td>0.063492</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.394737</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.134146</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>8149</th>\n",
       "      <td>0.678571</td>\n",
       "      <td>0.049924</td>\n",
       "      <td>0.063492</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.368421</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.170732</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>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>200</th>\n",
       "      <td>0.678571</td>\n",
       "      <td>0.014728</td>\n",
       "      <td>0.063492</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.368421</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.059524</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.170732</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>10876</th>\n",
       "      <td>0.678571</td>\n",
       "      <td>0.052415</td>\n",
       "      <td>0.063492</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.578947</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.170732</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>4035</th>\n",
       "      <td>0.678571</td>\n",
       "      <td>0.038770</td>\n",
       "      <td>0.063492</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.368421</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.170732</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>3640</th>\n",
       "      <td>0.678571</td>\n",
       "      <td>0.039420</td>\n",
       "      <td>0.063492</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.552632</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.170732</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>157</th>\n",
       "      <td>0.678571</td>\n",
       "      <td>0.043535</td>\n",
       "      <td>0.063492</td>\n",
       "      <td>0.283333</td>\n",
       "      <td>0.421053</td>\n",
       "      <td>0.2</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.049793</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.170732</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>2000 rows × 5414 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "         n_0001    n_0002    n_0003    n_0004    n_0005  n_0006    n_0007  \\\n",
       "id                                                                          \n",
       "7957   0.678571  0.029890  0.063492  0.283333  0.368421     0.2  0.000000   \n",
       "693    0.452381  0.032164  0.063492  0.283333  0.421053     0.0  0.000000   \n",
       "15558  0.547619  0.049166  0.063492  0.283333  0.500000     0.2  0.000000   \n",
       "15614  0.678571  0.048733  0.063492  0.283333  0.394737     0.2  0.000000   \n",
       "8149   0.678571  0.049924  0.063492  0.283333  0.368421     0.2  0.000000   \n",
       "...         ...       ...       ...       ...       ...     ...       ...   \n",
       "200    0.678571  0.014728  0.063492  0.283333  0.368421     0.2  0.059524   \n",
       "10876  0.678571  0.052415  0.063492  0.283333  0.578947     0.2  0.000000   \n",
       "4035   0.678571  0.038770  0.063492  0.283333  0.368421     0.2  0.000000   \n",
       "3640   0.678571  0.039420  0.063492  0.283333  0.552632     0.2  0.000000   \n",
       "157    0.678571  0.043535  0.063492  0.283333  0.421053     0.2  0.000000   \n",
       "\n",
       "         n_0009  n_0010    n_0012  ...  c_1375_m  c_1375_n  c_1375_o  \\\n",
       "id                                 ...                                 \n",
       "7957   0.049793    0.35  0.329268  ...         0         0         0   \n",
       "693    0.049793    0.35  0.341463  ...         0         0         0   \n",
       "15558  0.049793    0.35  0.219512  ...         0         0         0   \n",
       "15614  0.049793    0.35  0.134146  ...         0         0         0   \n",
       "8149   0.049793    0.35  0.170732  ...         0         0         0   \n",
       "...         ...     ...       ...  ...       ...       ...       ...   \n",
       "200    0.049793    0.35  0.170732  ...         0         0         0   \n",
       "10876  0.049793    0.35  0.170732  ...         0         0         0   \n",
       "4035   0.049793    0.35  0.170732  ...         0         0         0   \n",
       "3640   0.049793    0.35  0.170732  ...         0         0         0   \n",
       "157    0.049793    0.35  0.170732  ...         0         0         0   \n",
       "\n",
       "       c_1375_p  c_1375_q  c_1375_r  c_1375_s  c_1375_t  c_1377_a  c_1377_b  \n",
       "id                                                                           \n",
       "7957          0         0         0         0         0         0         0  \n",
       "693           0         0         0         0         0         0         0  \n",
       "15558         0         0         0         0         0         0         0  \n",
       "15614         0         0         0         0         0         0         0  \n",
       "8149          0         0         0         0         0         0         1  \n",
       "...         ...       ...       ...       ...       ...       ...       ...  \n",
       "200           0         0         0         0         0         0         0  \n",
       "10876         0         0         0         0         0         0         0  \n",
       "4035          0         0         0         0         0         0         0  \n",
       "3640          0         0         0         0         0         0         0  \n",
       "157           0         0         0         0         0         0         0  \n",
       "\n",
       "[2000 rows x 5414 columns]"
      ]
     },
     "execution_count": 122,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "median_test = test_df_copy.median().to_dict()\n",
    "test_df_copy.fillna(median_test, inplace=True)\n",
    "test_df_copy"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4fc7787c-af96-4417-80b5-115a07c69ef9",
   "metadata": {},
   "source": [
    "### Разделим датасеты на train и test части"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "id": "95aac280-6a58-4b70-b627-b4762de72c0c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train.shape: (5600, 5414)\n",
      "X_test.shape: (2400, 5414)\n",
      "y_train.shape: (5600, 14)\n",
      "y_test.shape: (2400, 14)\n"
     ]
    }
   ],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(train_df_copy, labels_df, random_state=0, train_size=0.7)\n",
    "\n",
    "print(\"X_train.shape: \" + str(X_train.shape))\n",
    "print(\"X_test.shape: \" + str(X_test.shape))\n",
    "print(\"y_train.shape: \" + str(y_train.shape))\n",
    "print(\"y_test.shape: \" + str(y_test.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b3a95f29-9744-4529-b540-b4eba106b69d",
   "metadata": {},
   "source": [
    "### Реализуем бинарный классификатор с необходимым функционалом"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "id": "2a821673-a91f-4657-a309-a303293491aa",
   "metadata": {},
   "outputs": [],
   "source": [
    "class BinaryRelevanceClassifier(BaseEstimator, ClassifierMixin):\n",
    "    def __init__(self, base_classifier=LogisticRegression()):\n",
    "        self.base_classifier=base_classifier\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        self.models = []\n",
    "        for label in list(y.columns):\n",
    "            x_checked, y_checked = check_X_y(X, y[label])\n",
    "            base_model = clone(self.base_classifier)\n",
    "            basel_model = base_model.fit(x_checked, y_checked)\n",
    "            self.models.append(base_model)\n",
    "\n",
    "    def predict(self, X):\n",
    "        check_is_fitted(self, ['models'])\n",
    "        X = check_array(X)\n",
    "        all_preds = pd.DataFrame()\n",
    "        i=0\n",
    "        preds = []\n",
    "        for model in self.models:\n",
    "            pred = model.predict(X)\n",
    "            preds.append(pd.DataFrame({'Class'+ str(i+1): pred}))\n",
    "            i+=1\n",
    "        all_preds = pd.concat(preds, axis=1)\n",
    "        return all_preds.to_numpy()\n",
    "    \n",
    "    def predict_proba(self,X):\n",
    "        check_is_fitted(self, ['models'])\n",
    "        X = check_array(X)      \n",
    "        all_preds = pd.DataFrame()\n",
    "        i = 0       \n",
    "        for model in self.models:\n",
    "            pred = model.predict_proba(X)\n",
    "            all_preds['Class'+str(i+1)] = [one_prob[1] for one_prob in pred]\n",
    "            i+=1\n",
    "        return all_preds.to_numpy()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "819d3799-3826-4bb7-ac71-6f49a41d76e3",
   "metadata": {},
   "source": [
    "Создадим классифаер, зафитим и запредиктим"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "id": "795a5935-97c9-4d4a-8d55-8b58c751ca20",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_pred.shape: (2400, 14)\n"
     ]
    }
   ],
   "source": [
    "br_clf = BinaryRelevanceClassifier(LogisticRegression())\n",
    "br_clf.fit(X_train, y_train)\n",
    "y_pred = br_clf.predict(X_test)\n",
    "print(\"y_pred.shape: \" + str(y_pred.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4ca9fff1-540b-4019-9369-8cb13d6c1e3b",
   "metadata": {},
   "source": [
    "Напишем функцию расчета метрики точности для классифаеров"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "id": "b7cdfc23-1818-47e8-8d06-e67e4ab9a0d4",
   "metadata": {},
   "outputs": [],
   "source": [
    "def accuracy_score(y_test, y_pred):\n",
    "    y_test = y_test.to_numpy()\n",
    "    assert y_test.shape == y_pred.shape\n",
    "    i=0\n",
    "    scores = []\n",
    "    while i < len(y_test):\n",
    "        count=0\n",
    "        for p, q in zip(y_test[i], y_pred[i]):\n",
    "            if p == q:\n",
    "                count += 1\n",
    "        scores.append(count / y_pred.shape[1])\n",
    "        i+=1 \n",
    "    return round((sum(scores)/len(y_test)), 5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "id": "d148b7df-61b5-4ef1-bc6b-57e95cd202f0",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Binary Relevance Classifier: 0.86952\n"
     ]
    }
   ],
   "source": [
    "print(\"Accuracy of Binary Relevance Classifier: \" + str(accuracy_score(y_test, y_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5404519c-fc3b-4509-b09c-ba367d0059e3",
   "metadata": {},
   "source": [
    "Попробуем поискать параметры для классифаера, которые дадут нам более точные результаты"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 128,
   "id": "5ded1b0a-b445-4ec2-b4a9-46864440fd74",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 5 folds for each of 6 candidates, totalling 30 fits\n",
      "Best Parameters Found: \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'base_classifier': RandomForestClassifier(criterion='entropy')}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "0.873432"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cv_folds=5\n",
    "\n",
    "param_grid ={'base_classifier': [DecisionTreeClassifier(criterion='entropy', max_depth=15, min_samples_leaf=2),\n",
    "                                 RandomForestClassifier(criterion='entropy'),\n",
    "                                 LogisticRegression(max_iter=20000), GaussianNB(), KNeighborsClassifier(), SVC()] }\n",
    "\n",
    "tuned_model = GridSearchCV(BinaryRelevanceClassifier(), \\\n",
    "                            param_grid, scoring=make_scorer(accuracy_score), verbose = 2, n_jobs = -1, cv=cv_folds)\n",
    "tuned_model.fit(X_train, y_train)\n",
    "\n",
    "print(\"Best Parameters Found: \")\n",
    "display(tuned_model.best_params_)\n",
    "display(tuned_model.best_score_)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "427b8a26-a28c-4851-b0f1-4c5666872715",
   "metadata": {},
   "source": [
    "Создадим классифаер с подобранными параметрами"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 129,
   "id": "260e9655-bc3b-4009-94e5-fcd7feca5c23",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_pred.shape: (2400, 14)\n"
     ]
    }
   ],
   "source": [
    "br_clf2 = BinaryRelevanceClassifier(RandomForestClassifier(criterion='entropy'))\n",
    "br_clf2.fit(X_train, y_train)\n",
    "y_pred2 = br_clf2.predict(X_test)\n",
    "print(\"y_pred.shape: \" + str(y_pred2.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 131,
   "id": "346e7fe0-6d78-4a32-a81b-6e8ea60c8971",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Binary Relevance Classifier - Best Model: 0.87185\n"
     ]
    }
   ],
   "source": [
    "print(\"Accuracy of Binary Relevance Classifier - Best Model: \" + str(accuracy_score(y_test, y_pred2)))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0f808755-af88-4dc1-a87d-8360090d98bf",
   "metadata": {},
   "source": [
    "Сохраним результаты предсказания лейблов для тестового датасета"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 132,
   "id": "ca2693e2-6f78-499a-842b-eb711ce9db9f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2000, 14)\n"
     ]
    }
   ],
   "source": [
    "best_result_based_on_accuracy_proba = br_clf2.predict_proba(test_df_copy)\n",
    "print(best_result_based_on_accuracy_proba.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e76b3473-70d8-4a4f-8abc-3b38dcd4461a",
   "metadata": {},
   "source": [
    "### Реализуем бинарный классификатор с андерсемплингом "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d3e1b60f-81dc-4501-b72e-c46e9fb3e923",
   "metadata": {},
   "source": [
    "Как мы видели в процессе исследования данных, датасет лейблов не является сбалансированным, поэтому попробуем посмотреть на результаты бинарной классификации с андерсемплингом"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 133,
   "id": "b9e7e45d-3810-4acb-9dfa-79f2667d6e1a",
   "metadata": {},
   "outputs": [],
   "source": [
    "class BinaryRelevanceClassifierUS(BaseEstimator, ClassifierMixin):\n",
    "    def __init__(self, base_classifier=LogisticRegression(max_iter=20000)):\n",
    "        self.base_classifier=base_classifier\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        self.models = []\n",
    "        for label in list(y.columns):          \n",
    "            X_cp = X.copy()\n",
    "            y_cp = y[label]\n",
    "            X_y_data = pd.concat([X_cp, y_cp], axis=1)\n",
    "            n_val0, n_val1 = 0,0         \n",
    "            j=0\n",
    "            while j<len(X_y_data):\n",
    "                if(X_y_data.iloc[j][label] == 0):\n",
    "                    n_val0+=1\n",
    "                else:\n",
    "                    n_val1+=1\n",
    "                j+=1\n",
    "            if n_val0 > n_val1:\n",
    "                val1 = X_y_data[X_y_data[label]==1]\n",
    "                val0 = X_y_data[X_y_data[label]==0].sample(n_val1)              \n",
    "                X_y_data = pd.concat([val0, val1], axis=0)           \n",
    "            elif n_val1 > n_val0:\n",
    "                val1 = X_y_data[X_y_data[label]==1].sample(n_val0)\n",
    "                val0 = X_y_data[X_y_data[label]==0]              \n",
    "                X_y_data = pd.concat([val0, val1], axis=0)\n",
    "            X_cp = X_y_data.iloc[:, :-1]\n",
    "            y_cp = X_y_data.iloc[:, -1]            \n",
    "            base_model = clone(self.base_classifier)\n",
    "            a, b = check_X_y(X_cp, y_cp)\n",
    "            base_model.fit(a, b)\n",
    "            self.models.append(base_model)\n",
    "\n",
    "    def predict(self, X):\n",
    "        check_is_fitted(self, ['models'])\n",
    "        X = check_array(X)       \n",
    "        all_preds = pd.DataFrame()\n",
    "        i=0\n",
    "        preds = []\n",
    "        for model in self.models:\n",
    "            pred = model.predict(X)\n",
    "            preds.append(pd.DataFrame({'Class'+ str(i+1): pred}))\n",
    "            i+=1\n",
    "        all_preds = pd.concat(preds, axis=1)\n",
    "        return all_preds.to_numpy()\n",
    "     \n",
    "    def predict_proba(self,X):\n",
    "        check_is_fitted(self, ['models'])\n",
    "        X = check_array(X)       \n",
    "        all_preds = pd.DataFrame()\n",
    "        i = 0       \n",
    "        for model in self.models:\n",
    "            pred = model.predict_proba(X)\n",
    "            all_preds['Class'+str(i+1)] = [one_prob[1] for one_prob in pred]\n",
    "            i+=1\n",
    "        return all_preds.to_numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 134,
   "id": "36c4f8b3-36ce-436b-9f87-0718a99aeb18",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_pred.shape=(2400, 14)\n"
     ]
    }
   ],
   "source": [
    "brus_clf = BinaryRelevanceClassifierUS()\n",
    "brus_clf.fit(X_train, y_train)\n",
    "y_pred = brus_clf.predict(X_test)\n",
    "print(\"y_pred.shape=\" + str(y_pred.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 135,
   "id": "41502bd8-9bba-460a-8959-a3948df669dd",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Binary Relevance Classifier with Under-sampling: 0.73574\n"
     ]
    }
   ],
   "source": [
    "accuracy = accuracy_score(y_test, y_pred)\n",
    "print(\"Accuracy of Binary Relevance Classifier with Under-sampling: \" + str(accuracy))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 136,
   "id": "48bed87f-c7d6-45f3-896a-eecff3c8b388",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 5 folds for each of 6 candidates, totalling 30 fits\n",
      "Best Parameters Found: \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'base_classifier': RandomForestClassifier(criterion='entropy')}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "0.7533299999999999"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cv_folds=5\n",
    "\n",
    "param_grid ={'base_classifier': [DecisionTreeClassifier(criterion='entropy', max_depth=15, min_samples_leaf=2),\n",
    "                                 RandomForestClassifier(criterion='entropy'),\n",
    "                                 LogisticRegression(max_iter=20000), GaussianNB(), KNeighborsClassifier(), SVC()] }\n",
    "\n",
    "tuned_model = GridSearchCV(BinaryRelevanceClassifierUS(), \\\n",
    "                            param_grid, scoring=make_scorer(accuracy_score), verbose = 2, n_jobs = -1, cv=cv_folds)\n",
    "tuned_model.fit(X_train, y_train)\n",
    "\n",
    "print(\"Best Parameters Found: \")\n",
    "display(tuned_model.best_params_)\n",
    "display(tuned_model.best_score_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 137,
   "id": "801f0fc6-89cf-4af5-b266-c66933208743",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_pred.shape=(2400, 14)\n"
     ]
    }
   ],
   "source": [
    "brus_clf2 = BinaryRelevanceClassifierUS(RandomForestClassifier(criterion='entropy'))\n",
    "brus_clf2.fit(X_train, y_train)\n",
    "y_pred_brus2 = brus_clf2.predict(X_test)\n",
    "print(\"y_pred.shape=\" + str(y_pred_brus2.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 138,
   "id": "56f1a633-e91c-4e90-9cb9-2a04ae23851c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Binary Relevance Classifier with Under-sampling - Best Model: 0.74604\n"
     ]
    }
   ],
   "source": [
    "accuracy = accuracy_score(y_test, y_pred_brus2)\n",
    "print(\"Accuracy of Binary Relevance Classifier with Under-sampling - Best Model: \" + str(accuracy))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a9d05551-0b36-49a9-96bb-b738357032c1",
   "metadata": {},
   "source": [
    "Как видим результаты хуже, чем в предыдущем эксперименте. Данный результат объясняется тем, что в процессе андерсемплинга мы теряем данные с целью сбалансировать классы по лейблам."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1ed013f8-35e8-461b-a15f-91bf88fb243f",
   "metadata": {},
   "source": [
    "### Реализуем алгоритм Classifier Chains"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d48384b8-b8ac-4ad3-a139-9fd4caed2400",
   "metadata": {},
   "source": [
    "Основная проблема бинарной классификации в том, что она не учитывает связи между лейблами. Попробуем реализовать алгоритм Classfier Chains, чтобы справиться с этой проблемой.\n",
    "\n",
    "P.S. Сначала реализовал алгоритм, но спутал названия столбцов в лейбл датасете, позже было проще заменить названия столбцов в лейбл датасете, чем переписывать реализацию классифаера"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 139,
   "id": "82c9d874-b3b3-4aa0-9c56-1db9814c55fe",
   "metadata": {},
   "outputs": [],
   "source": [
    "new_col_names = [\"Class\"+str(i) for i in range(1, 15)]\n",
    "\n",
    "col_name_map = dict(zip(y_train.columns, new_col_names))\n",
    "col_name_map2 = dict(zip(y_test.columns, new_col_names))\n",
    "\n",
    "y_train.rename(columns=col_name_map, inplace=True)\n",
    "y_test.rename(columns=col_name_map2, inplace=True)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "id": "0d388c76-535b-4d60-b15a-2cd61735019f",
   "metadata": {},
   "outputs": [],
   "source": [
    "class ClassifierChains(BaseEstimator, ClassifierMixin):\n",
    "    def __init__(self, base_classifier=LogisticRegression(max_iter=20000), order=None):\n",
    "        self.base_classifier=base_classifier\n",
    "        self.order = order\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        if self.order is None:\n",
    "            self.order = list(range(y.shape[1]))\n",
    "        elif self.order == 'random':\n",
    "            self.order = list(range(y.shape[1]))\n",
    "            random.shuffle(self.order)\n",
    "        else:\n",
    "            if(len(self.order)==y.shape[1]):\n",
    "                self.order = [o - 1 for o in self.order]\n",
    "        self.models = [clone(self.base_classifier) for clf in range(y.shape[1])]\n",
    "        X_joined = X.copy()\n",
    "        for val in self.order:\n",
    "            X_joined = pd.concat([X_joined, y['Class'+str(val+1)]], axis=1)\n",
    "        for chain_index, model in enumerate(self.models):\n",
    "            y_vals = y.loc[:, 'Class'+str(self.order[chain_index]+1)]\n",
    "            t_X = X_joined.iloc[:, :(X.shape[1]+chain_index)]\n",
    "            check_X_y(t_X, y_vals)\n",
    "            model.fit(t_X, y_vals)\n",
    "\n",
    "    def predict(self, X):\n",
    "        check_is_fitted(self, ['models'])\n",
    "        pred_chain = pd.DataFrame(columns=['Class'+str(o+1) for o in self.order])       \n",
    "        X_copy = X.copy()\n",
    "        X_joined = X.copy()\n",
    "        X_joined.reset_index(drop=True, inplace=True)\n",
    "        X_copy.reset_index(drop=True, inplace=True)\n",
    "        i=0\n",
    "        for chain_index, model in enumerate(self.models):\n",
    "            prev_preds = pred_chain.iloc[:, :chain_index]\n",
    "            X_joined = pd.concat([X_copy, prev_preds], axis=1)\n",
    "            pred = model.predict(X_joined)\n",
    "            pred_chain['Class'+str(self.order[i]+1)] = pred\n",
    "            i+=1\n",
    "        pred_chain = pred_chain.loc[:, ['Class'+str(j+1) for j in range(0, len(self.order))]]\n",
    "        return pred_chain.to_numpy()\n",
    "\n",
    "    def predict_proba(self, X):\n",
    "        check_is_fitted(self, ['models'])\n",
    "        pred_chain = pd.DataFrame(columns=['Class'+str(o+1) for o in self.order])\n",
    "        pred_probs = pd.DataFrame(columns=['Class'+str(o+1) for o in self.order])\n",
    "        X_copy = X.copy()\n",
    "        X_joined = X.copy()\n",
    "        X_joined.reset_index(drop=True, inplace=True)\n",
    "        X_copy.reset_index(drop=True, inplace=True)\n",
    "        i=0\n",
    "        for chain_index, model in enumerate(self.models):\n",
    "            prev_preds = pred_chain.iloc[:, :chain_index]\n",
    "            X_joined = pd.concat([X_copy, prev_preds], axis=1)\n",
    "            pred = model.predict(X_joined)\n",
    "            pred_proba = model.predict_proba(X_joined)\n",
    "            pred_chain['Class'+str(self.order[i]+1)] = pred\n",
    "            pred_probs['Class'+str(self.order[i]+1)] = [one_prob[1] for one_prob in pred_proba]\n",
    "            i+=1\n",
    "        pred_probs = pred_probs.loc[:, ['Class'+str(j+1) for j in range(0, len(self.order))]]\n",
    "        return pred_probs.to_numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "id": "4b7d222c-62c5-4734-9e77-2ac57cdbd94f",
   "metadata": {},
   "outputs": [],
   "source": [
    "cc = ClassifierChains(order=[1,2,3,4,5,6,7,8,9,10,11,12,13,14])\n",
    "cc.fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "id": "1f8b4593-1d68-4aee-8e00-f0eb09d074c9",
   "metadata": {},
   "outputs": [],
   "source": [
    "cc_pred = cc.predict(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 143,
   "id": "ac32c6ab-4a01-472c-b9ce-e7d0ece33e53",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Classifier Chains: 0.85479\n"
     ]
    }
   ],
   "source": [
    "print(\"Accuracy of Classifier Chains: \" + str(accuracy_score(y_test, cc_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ea624771-51ae-4a35-8861-6b82683acb6e",
   "metadata": {},
   "source": [
    "Попробуем поискать параметры для классифаера, которые дадут нам более точные результаты. Тк данный классифаер при предсказании лейблов также учитывает связи с предыдущими классифаерами, то сгенерим 20 случайных порядков лейблов и проверим их"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "id": "736b2447-e407-46e5-9715-6b059aaf39e5",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 5 folds for each of 120 candidates, totalling 600 fits\n",
      "Best Parameters Found: \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'base_classifier': RandomForestClassifier(criterion='entropy'),\n",
       " 'order': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14]}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "0.87416"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cv_folds=5\n",
    "\n",
    "rand_orders = [list(range(1, y_test.shape[1]+1)) for i in list(range(1, 20))]\n",
    "\n",
    "for lst in rand_orders:\n",
    "    random.shuffle(lst)\n",
    "\n",
    "rand_orders.append([1,2,3,4,5,6,7,8,9,10,11,12,13,14])\n",
    "\n",
    "\n",
    "param_grid ={'base_classifier': [DecisionTreeClassifier(criterion='entropy', max_depth=15, min_samples_leaf=2),\n",
    "                                 RandomForestClassifier(criterion='entropy'),\n",
    "                                 LogisticRegression(max_iter=20000), GaussianNB(), KNeighborsClassifier(), SVC()],\n",
    "            'order': rand_orders}\n",
    "\n",
    "tuned_model = GridSearchCV(ClassifierChains(), \\\n",
    "                            param_grid, scoring=make_scorer(accuracy_score), verbose = 2, n_jobs = -1, cv=cv_folds)\n",
    "tuned_model.fit(X_train, y_train)\n",
    "\n",
    "print(\"Best Parameters Found: \")\n",
    "display(tuned_model.best_params_)\n",
    "display(tuned_model.best_score_)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f2a4d74e-06c1-4727-9eb5-d410b32119fc",
   "metadata": {},
   "source": [
    "Сохраним лучшие параметры "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "id": "08e5ad1f-3589-406d-95c2-d0fa6f6f49b5",
   "metadata": {},
   "outputs": [],
   "source": [
    "best = tuned_model.best_params_"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "45d967d7-f63a-4bfb-8378-23a1913e651a",
   "metadata": {},
   "source": [
    "Стоит отметить, что рэндом форест показывает лучший показатель из базовых классифаеров во всех испытаниях, которые я проводил, а также, что порядок лейблов остался в начальном состоянии. Далее попробуем взять рэндом форест как базовый классифаер, но увеличим количество проверяемых случайных порядков лейблов до 50 и попробуем подобрать тот, что даст более точные результаты предсказания"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 150,
   "id": "e2a3f21b-e5f2-48d0-b98a-a27ca858ca49",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 5 folds for each of 50 candidates, totalling 250 fits\n",
      "Best Parameters Found: \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'base_classifier': RandomForestClassifier(criterion='entropy'),\n",
       " 'order': [12, 1, 13, 3, 7, 2, 14, 6, 11, 5, 4, 9, 8, 10]}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "0.873966"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cv_folds=5\n",
    "\n",
    "rand_orders = [list(range(1, y_test.shape[1]+1)) for i in list(range(1, 50))]\n",
    "\n",
    "for lst in rand_orders:\n",
    "    random.shuffle(lst)\n",
    "\n",
    "rand_orders.append([1,2,3,4,5,6,7,8,9,10,11,12,13,14])\n",
    "\n",
    "\n",
    "param_grid ={'base_classifier': [RandomForestClassifier(criterion='entropy')\n",
    "                                 ],\n",
    "            'order': rand_orders}\n",
    "\n",
    "tuned_model = GridSearchCV(ClassifierChains(), \\\n",
    "                            param_grid, scoring=make_scorer(accuracy_score), verbose = 2, n_jobs = -1, cv=cv_folds)\n",
    "tuned_model.fit(X_train, y_train)\n",
    "\n",
    "print(\"Best Parameters Found: \")\n",
    "display(tuned_model.best_params_)\n",
    "display(tuned_model.best_score_)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2f3a6ab5-3e46-4c43-82d2-46f958400bc0",
   "metadata": {},
   "source": [
    "Точность оказалась ниже, зафитим классифаер с предыдущими показателями"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 146,
   "id": "ab736f60-bc1d-424f-a0cd-1a80ed4a58ca",
   "metadata": {},
   "outputs": [],
   "source": [
    "cc = ClassifierChains(base_classifier=best['base_classifier'], order=best['order'])\n",
    "cc.fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 147,
   "id": "df06d50b-a10c-4800-9461-15bed40f8a23",
   "metadata": {},
   "outputs": [],
   "source": [
    "cc_pred = cc.predict(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "id": "1827b204-11fb-4865-84f5-228a91654807",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Classifier Chains - Best Model: 0.87104\n"
     ]
    }
   ],
   "source": [
    "print(\"Accuracy of Classifier Chains - Best Model: \" + str(accuracy_score(y_test, cc_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3fa37c96-f157-40d5-8610-8a4f49e57553",
   "metadata": {},
   "source": [
    "Точность предсказания Binary Relevance Classifier была чуть выше, возьмем предсказанные вероятности лейблов для тестового датасета, сохраним его в необходимом виде"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 154,
   "id": "da34dbb3-affb-4e9a-a155-3f1e6a0801c5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2000, 14)"
      ]
     },
     "execution_count": 154,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_result_based_on_accuracy_proba.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 155,
   "id": "a3f0af48-ebb7-4e30-ab1c-9ab09d663d32",
   "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>service_a</th>\n",
       "      <th>service_b</th>\n",
       "      <th>service_c</th>\n",
       "      <th>service_d</th>\n",
       "      <th>service_e</th>\n",
       "      <th>service_f</th>\n",
       "      <th>service_g</th>\n",
       "      <th>service_h</th>\n",
       "      <th>service_i</th>\n",
       "      <th>service_j</th>\n",
       "      <th>service_k</th>\n",
       "      <th>service_l</th>\n",
       "      <th>service_m</th>\n",
       "      <th>service_n</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</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>7957</th>\n",
       "      <td>0.97</td>\n",
       "      <td>0.62</td>\n",
       "      <td>0.29</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.17</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.10</td>\n",
       "      <td>0.15</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.80</td>\n",
       "      <td>0.83</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>693</th>\n",
       "      <td>0.02</td>\n",
       "      <td>0.08</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.07</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.06</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.87</td>\n",
       "      <td>0.83</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.15</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15558</th>\n",
       "      <td>0.40</td>\n",
       "      <td>0.37</td>\n",
       "      <td>0.37</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.10</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.12</td>\n",
       "      <td>0.27</td>\n",
       "      <td>0.06</td>\n",
       "      <td>0.79</td>\n",
       "      <td>0.73</td>\n",
       "      <td>0.10</td>\n",
       "      <td>0.07</td>\n",
       "      <td>0.41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15614</th>\n",
       "      <td>0.24</td>\n",
       "      <td>0.21</td>\n",
       "      <td>0.27</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.16</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.15</td>\n",
       "      <td>0.46</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.81</td>\n",
       "      <td>0.75</td>\n",
       "      <td>0.28</td>\n",
       "      <td>0.24</td>\n",
       "      <td>0.30</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8149</th>\n",
       "      <td>0.82</td>\n",
       "      <td>0.70</td>\n",
       "      <td>0.44</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.06</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.19</td>\n",
       "      <td>0.25</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.72</td>\n",
       "      <td>0.48</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>200</th>\n",
       "      <td>0.22</td>\n",
       "      <td>0.13</td>\n",
       "      <td>0.18</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.15</td>\n",
       "      <td>0.11</td>\n",
       "      <td>0.08</td>\n",
       "      <td>0.25</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.89</td>\n",
       "      <td>0.76</td>\n",
       "      <td>0.12</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10876</th>\n",
       "      <td>0.72</td>\n",
       "      <td>0.45</td>\n",
       "      <td>0.31</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.08</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.36</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.81</td>\n",
       "      <td>0.82</td>\n",
       "      <td>0.18</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4035</th>\n",
       "      <td>0.46</td>\n",
       "      <td>0.36</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.06</td>\n",
       "      <td>0.11</td>\n",
       "      <td>0.28</td>\n",
       "      <td>0.06</td>\n",
       "      <td>0.76</td>\n",
       "      <td>0.82</td>\n",
       "      <td>0.12</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.42</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3640</th>\n",
       "      <td>0.31</td>\n",
       "      <td>0.25</td>\n",
       "      <td>0.11</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.24</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.82</td>\n",
       "      <td>0.80</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>157</th>\n",
       "      <td>0.82</td>\n",
       "      <td>0.53</td>\n",
       "      <td>0.29</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.06</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.37</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.86</td>\n",
       "      <td>0.84</td>\n",
       "      <td>0.12</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.03</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2000 rows × 14 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       service_a  service_b  service_c  service_d  service_e  service_f  \\\n",
       "id                                                                        \n",
       "7957        0.97       0.62       0.29       0.03       0.17       0.03   \n",
       "693         0.02       0.08       0.03       0.05       0.07       0.00   \n",
       "15558       0.40       0.37       0.37       0.02       0.10       0.03   \n",
       "15614       0.24       0.21       0.27       0.04       0.16       0.01   \n",
       "8149        0.82       0.70       0.44       0.00       0.06       0.00   \n",
       "...          ...        ...        ...        ...        ...        ...   \n",
       "200         0.22       0.13       0.18       0.00       0.15       0.11   \n",
       "10876       0.72       0.45       0.31       0.01       0.05       0.08   \n",
       "4035        0.46       0.36       0.35       0.04       0.03       0.06   \n",
       "3640        0.31       0.25       0.11       0.00       0.00       0.02   \n",
       "157         0.82       0.53       0.29       0.00       0.06       0.01   \n",
       "\n",
       "       service_g  service_h  service_i  service_j  service_k  service_l  \\\n",
       "id                                                                        \n",
       "7957        0.10       0.15       0.04       0.80       0.83       0.03   \n",
       "693         0.00       0.06       0.01       0.87       0.83       0.02   \n",
       "15558       0.12       0.27       0.06       0.79       0.73       0.10   \n",
       "15614       0.15       0.46       0.02       0.81       0.75       0.28   \n",
       "8149        0.19       0.25       0.03       0.72       0.48       0.03   \n",
       "...          ...        ...        ...        ...        ...        ...   \n",
       "200         0.08       0.25       0.02       0.89       0.76       0.12   \n",
       "10876       0.03       0.36       0.03       0.81       0.82       0.18   \n",
       "4035        0.11       0.28       0.06       0.76       0.82       0.12   \n",
       "3640        0.04       0.24       0.02       0.82       0.80       0.02   \n",
       "157         0.02       0.37       0.02       0.86       0.84       0.12   \n",
       "\n",
       "       service_m  service_n  \n",
       "id                           \n",
       "7957        0.01       0.18  \n",
       "693         0.00       0.15  \n",
       "15558       0.07       0.41  \n",
       "15614       0.24       0.30  \n",
       "8149        0.02       0.04  \n",
       "...          ...        ...  \n",
       "200         0.05       0.07  \n",
       "10876       0.03       0.26  \n",
       "4035        0.03       0.42  \n",
       "3640        0.05       0.02  \n",
       "157         0.04       0.03  \n",
       "\n",
       "[2000 rows x 14 columns]"
      ]
     },
     "execution_count": 155,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_result_df = pd.DataFrame(data=best_result_based_on_accuracy_proba)\n",
    "best_result_df.columns, best_result_df.index = labels_df.columns, test_df.index\n",
    "best_result_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 156,
   "id": "a9e772f6-2ffd-4bae-bdb2-2fb1e673c4e1",
   "metadata": {},
   "outputs": [],
   "source": [
    "best_result_df.to_csv(\"problem_test_labels.csv\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "26391e81-80fb-46bb-89af-b8d55d514cf0",
   "metadata": {},
   "source": [
    "## Подведем итоги"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "38793d99-2ae4-47c4-a43b-ca5214af75f8",
   "metadata": {},
   "source": [
    "Пусть BR=Binary Relevance Classifier, BRUS=Binary Relevance Classifier with Under Sampling, CC=Classifier Chains.\n",
    "\n",
    "По результатам работы хотел бы отметить следующие выводы:\n",
    "\n",
    "* BR работает значительно лучше BRUS. Это происходит, тк в процессе андерсемплинга могут быть потеряны объекты из тренировочного датасета, что влияет на точность предсказания. \n",
    "* В алгоритме СС в процессе предсказания учитываются связи между лейблами. BR и BRUS в процессе предсказания работают с каждым лейблом независимо.\n",
    "* Результаты работы алгоритма CC зависят от порядка расположения лейблов, так как мы учитываем предсказания предыдущих лейблов. Мы использовали до 50 порядков расположения лейблов в нашем эксперименте с алгоритмом СС.\n",
    "* Если говорить о сложности вычислений, то BR проще и быстрее, чем CC. Если говорить о результате, то CC должен показывать большую точность, чем BR по причинам упомянутым выше.\n",
    "* При работе с данным датасетом, CC показал точность немного ниже, чем BR. Однако стоит отметить, что при проведении эксперимента с заполнением пропущенных значений модами, СС показывал результат незначительно выше, чем BR. При этом, по наблюдениям, на гридсерч BR я потратил на своем железе около получаса для подбора оптимальных параметров модели, а на СС - около 10ти часов. Таким образом, если перед нами стоит вопрос скорости обработки данных и сложности модели, то представляется разумным использовать алгоритм BR."
   ]
  }
 ],
 "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": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}