Mean feature encoding.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.model_selection import KFold"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>feature</th>\n",
       "      <th>target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>c</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>a</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>c</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>c</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>a</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>995</td>\n",
       "      <td>b</td>\n",
       "      <td>995</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>996</td>\n",
       "      <td>b</td>\n",
       "      <td>996</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>997</td>\n",
       "      <td>c</td>\n",
       "      <td>997</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>998</td>\n",
       "      <td>c</td>\n",
       "      <td>998</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>999</td>\n",
       "      <td>a</td>\n",
       "      <td>999</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1000 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "    feature  target\n",
       "0         c       0\n",
       "1         a       1\n",
       "2         c       2\n",
       "3         c       3\n",
       "4         a       4\n",
       "..      ...     ...\n",
       "995       b     995\n",
       "996       b     996\n",
       "997       c     997\n",
       "998       c     998\n",
       "999       a     999\n",
       "\n",
       "[1000 rows x 2 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.random.seed(42)\n",
    "X = np.random.choice(['a', 'b', 'c'], size=1000)\n",
    "y = np.arange(1000)\n",
    "df = pd.DataFrame()\n",
    "df['feature'] = X\n",
    "df['target'] = y\n",
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def mean_encode(split, feature, target, alpha, dtype=np.object):\n",
    "    train_id, test_id = split\n",
    "    X_train, y_train = feature[train_id], target[train_id]\n",
    "    res = np.zeros_like(feature, dtype=dtype)\n",
    "    \n",
    "    for value in set(X_train):\n",
    "        n_rows = len(X_train[X_train == value])\n",
    "        mean = np.mean(y_train[X_train == value])\n",
    "        global_mean = np.mean(y_train)\n",
    "        res[feature == value] = mean_encoding = (mean * n_rows + global_mean * alpha) / (n_rows + alpha)\n",
    "    \n",
    "    return res"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "for i, split in enumerate(KFold(n_splits=5, random_state=42).split(X)):\n",
    "    df[f'split_{i}'] = mean_encode(split, X, y, 0.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>feature</th>\n",
       "      <th>target</th>\n",
       "      <th>split_0</th>\n",
       "      <th>split_1</th>\n",
       "      <th>split_2</th>\n",
       "      <th>split_3</th>\n",
       "      <th>split_4</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>c</td>\n",
       "      <td>0</td>\n",
       "      <td>598.924</td>\n",
       "      <td>533.214</td>\n",
       "      <td>480.719</td>\n",
       "      <td>438.97</td>\n",
       "      <td>376.985</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>a</td>\n",
       "      <td>1</td>\n",
       "      <td>596.949</td>\n",
       "      <td>555.21</td>\n",
       "      <td>504.028</td>\n",
       "      <td>443.307</td>\n",
       "      <td>425.534</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>c</td>\n",
       "      <td>2</td>\n",
       "      <td>598.924</td>\n",
       "      <td>533.214</td>\n",
       "      <td>480.719</td>\n",
       "      <td>438.97</td>\n",
       "      <td>376.985</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>c</td>\n",
       "      <td>3</td>\n",
       "      <td>598.924</td>\n",
       "      <td>533.214</td>\n",
       "      <td>480.719</td>\n",
       "      <td>438.97</td>\n",
       "      <td>376.985</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>a</td>\n",
       "      <td>4</td>\n",
       "      <td>596.949</td>\n",
       "      <td>555.21</td>\n",
       "      <td>504.028</td>\n",
       "      <td>443.307</td>\n",
       "      <td>425.534</td>\n",
       "    </tr>\n",
       "    <tr>\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",
       "      <td>995</td>\n",
       "      <td>b</td>\n",
       "      <td>995</td>\n",
       "      <td>602.816</td>\n",
       "      <td>559.032</td>\n",
       "      <td>513.584</td>\n",
       "      <td>466.119</td>\n",
       "      <td>391.502</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>996</td>\n",
       "      <td>b</td>\n",
       "      <td>996</td>\n",
       "      <td>602.816</td>\n",
       "      <td>559.032</td>\n",
       "      <td>513.584</td>\n",
       "      <td>466.119</td>\n",
       "      <td>391.502</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>997</td>\n",
       "      <td>c</td>\n",
       "      <td>997</td>\n",
       "      <td>598.924</td>\n",
       "      <td>533.214</td>\n",
       "      <td>480.719</td>\n",
       "      <td>438.97</td>\n",
       "      <td>376.985</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>998</td>\n",
       "      <td>c</td>\n",
       "      <td>998</td>\n",
       "      <td>598.924</td>\n",
       "      <td>533.214</td>\n",
       "      <td>480.719</td>\n",
       "      <td>438.97</td>\n",
       "      <td>376.985</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>999</td>\n",
       "      <td>a</td>\n",
       "      <td>999</td>\n",
       "      <td>596.949</td>\n",
       "      <td>555.21</td>\n",
       "      <td>504.028</td>\n",
       "      <td>443.307</td>\n",
       "      <td>425.534</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1000 rows × 7 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "    feature  target  split_0  split_1  split_2  split_3  split_4\n",
       "0         c       0  598.924  533.214  480.719   438.97  376.985\n",
       "1         a       1  596.949   555.21  504.028  443.307  425.534\n",
       "2         c       2  598.924  533.214  480.719   438.97  376.985\n",
       "3         c       3  598.924  533.214  480.719   438.97  376.985\n",
       "4         a       4  596.949   555.21  504.028  443.307  425.534\n",
       "..      ...     ...      ...      ...      ...      ...      ...\n",
       "995       b     995  602.816  559.032  513.584  466.119  391.502\n",
       "996       b     996  602.816  559.032  513.584  466.119  391.502\n",
       "997       c     997  598.924  533.214  480.719   438.97  376.985\n",
       "998       c     998  598.924  533.214  480.719   438.97  376.985\n",
       "999       a     999  596.949   555.21  504.028  443.307  425.534\n",
       "\n",
       "[1000 rows x 7 columns]"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}