Movie Lens 100K version.3.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "5K9Uj4cIwDuv"
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import datetime\n",
    "from collections import defaultdict"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "oAoWD2_PwcMh"
   },
   "source": [
    "Task 1: Exploratory analysis (5 points).\n",
    "\n",
    "Perform EDA:\n",
    "- number of users\n",
    "- number of items\n",
    "- number of tags\n",
    "- number of ratings\n",
    "- summary statistics of number of ratings per user (mean, median, quantiles)\n",
    "- rating sparcity\n",
    "- summary statistics of number of tags per use"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "VBzSSxgG3RD4"
   },
   "source": [
    "### Загружаем датасеты"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading movielens data...\n",
      "Done. Dataset contains:\n",
      "b'943 users\\n1682 items\\n100000 ratings\\n'\n"
     ]
    }
   ],
   "source": [
    "print(\"Downloading movielens data...\")\n",
    "import zipfile\n",
    "import urllib.request\n",
    "\n",
    "urllib.request.urlretrieve(\"http://files.grouplens.org/datasets/movielens/ml-100k.zip\", \"movielens.zip\")\n",
    "zip_ref = zipfile.ZipFile('movielens.zip', \"r\")\n",
    "zip_ref.extractall()\n",
    "print(\"Done. Dataset contains:\")\n",
    "print(zip_ref.read('ml-100k/u.info'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load each data set (users, movies, and ratings).\n",
    "users_cols = ['user_id', 'age', 'sex', 'occupation', 'zip_code']\n",
    "users = pd.read_csv(\n",
    "    'ml-100k/u.user', sep='|', names=users_cols, encoding='latin-1')\n",
    "\n",
    "ratings_cols = ['user_id', 'movie_id', 'rating', 'unix_timestamp']\n",
    "ratings = pd.read_csv(\n",
    "    'ml-100k/u.data', sep='\\t', names=ratings_cols, encoding='latin-1')\n",
    "\n",
    "# The movies file contains a binary feature for each genre.\n",
    "genre_cols = [\n",
    "    \"genre_unknown\", \"Action\", \"Adventure\", \"Animation\", \"Children\", \"Comedy\",\n",
    "    \"Crime\", \"Documentary\", \"Drama\", \"Fantasy\", \"Film-Noir\", \"Horror\",\n",
    "    \"Musical\", \"Mystery\", \"Romance\", \"Sci-Fi\", \"Thriller\", \"War\", \"Western\"\n",
    "]\n",
    "movies_cols = [\n",
    "    'movie_id', 'title', 'release_date', \"video_release_date\", \"imdb_url\"] + genre_cols\n",
    "\n",
    "movies = pd.read_csv(\n",
    "    'ml-100k/u.item', sep='|', names=movies_cols, encoding='latin-1')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Предпроцесинг данных.Создание жанровых категорий."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "users[\"user_id\"] = users[\"user_id\"].apply(lambda x: float(x))\n",
    "movies[\"movie_id\"] = movies[\"movie_id\"].apply(lambda x: float(x))\n",
    "movies[\"year\"] = movies['release_date'].apply(lambda x: str(x).split('-')[-1])\n",
    "ratings[\"movie_id\"] = ratings[\"movie_id\"].apply(lambda x: float(x))\n",
    "ratings[\"user_id\"] = ratings[\"user_id\"].apply(lambda x: float(x))\n",
    "ratings[\"rating\"] = ratings[\"rating\"].apply(lambda x: float(x))"
   ]
  },
  {
   "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>user_id</th>\n",
       "      <th>movie_id</th>\n",
       "      <th>rating</th>\n",
       "      <th>unix_timestamp</th>\n",
       "      <th>date</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>196.0</td>\n",
       "      <td>242.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1997-12-04 18:55:49</td>\n",
       "      <td>1997-12-04 18:55:49</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>186.0</td>\n",
       "      <td>302.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1998-04-04 22:22:22</td>\n",
       "      <td>1998-04-04 22:22:22</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>22.0</td>\n",
       "      <td>377.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1997-11-07 10:18:36</td>\n",
       "      <td>1997-11-07 10:18:36</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>244.0</td>\n",
       "      <td>51.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1997-11-27 08:02:03</td>\n",
       "      <td>1997-11-27 08:02:03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>166.0</td>\n",
       "      <td>346.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1998-02-02 08:33:16</td>\n",
       "      <td>1998-02-02 08:33:16</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   user_id  movie_id  rating       unix_timestamp                 date\n",
       "0    196.0     242.0     3.0  1997-12-04 18:55:49  1997-12-04 18:55:49\n",
       "1    186.0     302.0     3.0  1998-04-04 22:22:22  1998-04-04 22:22:22\n",
       "2     22.0     377.0     1.0  1997-11-07 10:18:36  1997-11-07 10:18:36\n",
       "3    244.0      51.0     2.0  1997-11-27 08:02:03  1997-11-27 08:02:03\n",
       "4    166.0     346.0     1.0  1998-02-02 08:33:16  1998-02-02 08:33:16"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ratings['unix_timestamp'] = ratings['unix_timestamp'].apply(lambda x: datetime.datetime.fromtimestamp(x).strftime('%Y-%m-%d %H:%M:%S'))\n",
    "ratings['unix_timestamp'].astype('datetime64[ns]')\n",
    "ratings['date'] = ratings['unix_timestamp']\n",
    "ratings.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Количество фильмов с установленным жанром\n",
    "genre_occurences = movies[genre_cols].sum().to_dict()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def mark_genres(movies, genres):\n",
    "    def get_random_genre(gs):\n",
    "        active = [genre for genre, g in zip(genres, gs) if g==1]\n",
    "        if len(active) == 0:\n",
    "            return 'Other'\n",
    "        return np.random.choice(active)\n",
    "    def get_all_genres(gs):\n",
    "        active = [genre for genre, g in zip(genres, gs) if g==1]\n",
    "        if len(active) == 0:\n",
    "            return 'Other'\n",
    "        return '-'.join(active)\n",
    "    movies['genre'] = [\n",
    "        get_random_genre(gs) for gs in zip(*[movies[genre] for genre in genres])]\n",
    "    movies['all_genres'] = [\n",
    "        get_all_genres(gs) for gs in zip(*[movies[genre] for genre in genres])]\n",
    "\n",
    "mark_genres(movies, genre_cols)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = ratings.merge(movies, on='movie_id').merge(users, on='user_id')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.drop(['year','release_date','video_release_date','imdb_url', 'age', 'sex', 'occupation','zip_code'], inplace = True, axis = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>user_id</th>\n",
       "      <th>movie_id</th>\n",
       "      <th>rating</th>\n",
       "      <th>unix_timestamp</th>\n",
       "      <th>date</th>\n",
       "      <th>title</th>\n",
       "      <th>genre_unknown</th>\n",
       "      <th>Action</th>\n",
       "      <th>Adventure</th>\n",
       "      <th>Animation</th>\n",
       "      <th>...</th>\n",
       "      <th>Horror</th>\n",
       "      <th>Musical</th>\n",
       "      <th>Mystery</th>\n",
       "      <th>Romance</th>\n",
       "      <th>Sci-Fi</th>\n",
       "      <th>Thriller</th>\n",
       "      <th>War</th>\n",
       "      <th>Western</th>\n",
       "      <th>genre</th>\n",
       "      <th>all_genres</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>196.0</td>\n",
       "      <td>242.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1997-12-04 18:55:49</td>\n",
       "      <td>1997-12-04 18:55:49</td>\n",
       "      <td>Kolya (1996)</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>Comedy</td>\n",
       "      <td>Comedy</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>196.0</td>\n",
       "      <td>257.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1997-12-04 19:06:17</td>\n",
       "      <td>1997-12-04 19:06:17</td>\n",
       "      <td>Men in Black (1997)</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>Action</td>\n",
       "      <td>Action-Adventure-Comedy-Sci-Fi</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>196.0</td>\n",
       "      <td>111.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>1997-12-04 19:09:53</td>\n",
       "      <td>1997-12-04 19:09:53</td>\n",
       "      <td>Truth About Cats &amp; Dogs, The (1996)</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>Romance</td>\n",
       "      <td>Comedy-Romance</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>196.0</td>\n",
       "      <td>25.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>1997-12-04 19:12:35</td>\n",
       "      <td>1997-12-04 19:12:35</td>\n",
       "      <td>Birdcage, The (1996)</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>Comedy</td>\n",
       "      <td>Comedy</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>196.0</td>\n",
       "      <td>382.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>1997-12-04 19:10:43</td>\n",
       "      <td>1997-12-04 19:10:43</td>\n",
       "      <td>Adventures of Priscilla, Queen of the Desert, ...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>Drama</td>\n",
       "      <td>Comedy-Drama</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 27 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   user_id  movie_id  rating       unix_timestamp                 date  \\\n",
       "0    196.0     242.0     3.0  1997-12-04 18:55:49  1997-12-04 18:55:49   \n",
       "1    196.0     257.0     2.0  1997-12-04 19:06:17  1997-12-04 19:06:17   \n",
       "2    196.0     111.0     4.0  1997-12-04 19:09:53  1997-12-04 19:09:53   \n",
       "3    196.0      25.0     4.0  1997-12-04 19:12:35  1997-12-04 19:12:35   \n",
       "4    196.0     382.0     4.0  1997-12-04 19:10:43  1997-12-04 19:10:43   \n",
       "\n",
       "                                               title  genre_unknown  Action  \\\n",
       "0                                       Kolya (1996)              0       0   \n",
       "1                                Men in Black (1997)              0       1   \n",
       "2                Truth About Cats & Dogs, The (1996)              0       0   \n",
       "3                               Birdcage, The (1996)              0       0   \n",
       "4  Adventures of Priscilla, Queen of the Desert, ...              0       0   \n",
       "\n",
       "   Adventure  Animation  ...  Horror  Musical  Mystery  Romance  Sci-Fi  \\\n",
       "0          0          0  ...       0        0        0        0       0   \n",
       "1          1          0  ...       0        0        0        0       1   \n",
       "2          0          0  ...       0        0        0        1       0   \n",
       "3          0          0  ...       0        0        0        0       0   \n",
       "4          0          0  ...       0        0        0        0       0   \n",
       "\n",
       "   Thriller  War  Western    genre                      all_genres  \n",
       "0         0    0        0   Comedy                          Comedy  \n",
       "1         0    0        0   Action  Action-Adventure-Comedy-Sci-Fi  \n",
       "2         0    0        0  Romance                  Comedy-Romance  \n",
       "3         0    0        0   Comedy                          Comedy  \n",
       "4         0    0        0    Drama                    Comedy-Drama  \n",
       "\n",
       "[5 rows x 27 columns]"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 100000 entries, 0 to 99999\n",
      "Data columns (total 27 columns):\n",
      "user_id           100000 non-null float64\n",
      "movie_id          100000 non-null float64\n",
      "rating            100000 non-null float64\n",
      "unix_timestamp    100000 non-null object\n",
      "date              100000 non-null object\n",
      "title             100000 non-null object\n",
      "genre_unknown     100000 non-null int64\n",
      "Action            100000 non-null int64\n",
      "Adventure         100000 non-null int64\n",
      "Animation         100000 non-null int64\n",
      "Children          100000 non-null int64\n",
      "Comedy            100000 non-null int64\n",
      "Crime             100000 non-null int64\n",
      "Documentary       100000 non-null int64\n",
      "Drama             100000 non-null int64\n",
      "Fantasy           100000 non-null int64\n",
      "Film-Noir         100000 non-null int64\n",
      "Horror            100000 non-null int64\n",
      "Musical           100000 non-null int64\n",
      "Mystery           100000 non-null int64\n",
      "Romance           100000 non-null int64\n",
      "Sci-Fi            100000 non-null int64\n",
      "Thriller          100000 non-null int64\n",
      "War               100000 non-null int64\n",
      "Western           100000 non-null int64\n",
      "genre             100000 non-null object\n",
      "all_genres        100000 non-null object\n",
      "dtypes: float64(3), int64(19), object(5)\n",
      "memory usage: 21.4+ MB\n"
     ]
    }
   ],
   "source": [
    "df.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "ho8mCITH3aGJ"
   },
   "source": [
    "### Проверяем на наличие пропущенных значений"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "I8hQc7yawGQQ",
    "outputId": "67ca5f67-65df-4fae-eb97-f90e828064c0"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of movies Null values:  movie_id                 0\n",
      "title                    0\n",
      "release_date             1\n",
      "video_release_date    1682\n",
      "imdb_url                 3\n",
      "genre_unknown            0\n",
      "Action                   0\n",
      "Adventure                0\n",
      "Animation                0\n",
      "Children                 0\n",
      "Comedy                   0\n",
      "Crime                    0\n",
      "Documentary              0\n",
      "Drama                    0\n",
      "Fantasy                  0\n",
      "Film-Noir                0\n",
      "Horror                   0\n",
      "Musical                  0\n",
      "Mystery                  0\n",
      "Romance                  0\n",
      "Sci-Fi                   0\n",
      "Thriller                 0\n",
      "War                      0\n",
      "Western                  0\n",
      "year                     0\n",
      "genre                    0\n",
      "all_genres               0\n",
      "dtype: int64\n",
      "Number of ratings Null values:  user_id           0\n",
      "movie_id          0\n",
      "rating            0\n",
      "unix_timestamp    0\n",
      "date              0\n",
      "dtype: int64\n"
     ]
    }
   ],
   "source": [
    "print (\"Number of movies Null values: \", movies.isnull().sum())\n",
    "print (\"Number of ratings Null values: \", ratings.isnull().sum())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "w9eGJk2xwGSw",
    "outputId": "8692b663-935c-405b-d3b0-f4dc24a43b2a"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of movies Null values:  movie_id              0\n",
      "title                 0\n",
      "release_date          0\n",
      "video_release_date    0\n",
      "imdb_url              0\n",
      "genre_unknown         0\n",
      "Action                0\n",
      "Adventure             0\n",
      "Animation             0\n",
      "Children              0\n",
      "Comedy                0\n",
      "Crime                 0\n",
      "Documentary           0\n",
      "Drama                 0\n",
      "Fantasy               0\n",
      "Film-Noir             0\n",
      "Horror                0\n",
      "Musical               0\n",
      "Mystery               0\n",
      "Romance               0\n",
      "Sci-Fi                0\n",
      "Thriller              0\n",
      "War                   0\n",
      "Western               0\n",
      "year                  0\n",
      "genre                 0\n",
      "all_genres            0\n",
      "dtype: int64\n"
     ]
    }
   ],
   "source": [
    "movies.dropna(inplace=True)\n",
    "print (\"Number of movies Null values: \", movies.isnull().sum())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "9rYByUtF3nK0"
   },
   "source": [
    "### Создаем саб-сеты для визуализации"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "9X1vJ8Ojz0F9",
    "outputId": "b658d6f6-2a6d-4a18-d107-785b3097e23c"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "21201"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[df['rating'] >= 4.5]['movie_id'].count() # Фильмы с рейтингом 4,5 или выше"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Zi9lrvVpz0I9",
    "outputId": "b778657d-2d54-447a-945d-0faf6b6222cf"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.21201"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[df['rating'] >= 4.5]['movie_id'].count() / ratings['rating'].count() # Процент фильмов с рейтингом 4,5 или выше"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "1EC39V6Kz0MF",
    "outputId": "3323af29-e14a-4b29-eac0-17011d0fef44"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "21201"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[df['rating'] == 5]['movie_id'].count() # Фильмы с рейтингом 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "1dbHDlJpz0Om",
    "outputId": "d1c88601-47b4-4b4f-fa62-7eaf219a1498"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.21201"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[df['rating'] == 5]['movie_id'].count() / ratings['rating'].count() # Процент фильмов с рейтингом 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "nySGiChQ1nom",
    "outputId": "bc9f6c4b-89e8-4037-a466-4fc9d958d6fb"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "title\n",
       "Star Wars (1977)             583\n",
       "Contact (1997)               509\n",
       "Fargo (1996)                 508\n",
       "Return of the Jedi (1983)    507\n",
       "Liar Liar (1997)             485\n",
       "Name: rating, dtype: int64"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.groupby('title')['rating'].count().sort_values(ascending=False).head() # Топ фильмов по рейтингу"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 257,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Uchg7jgU1nrb",
    "outputId": "b7db81d5-7274-40d3-8deb-f7b2d88be01c"
   },
   "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>user_id</th>\n",
       "      <th>RateCount</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>404</th>\n",
       "      <td>405.0</td>\n",
       "      <td>737</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>654</th>\n",
       "      <td>655.0</td>\n",
       "      <td>685</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>13.0</td>\n",
       "      <td>636</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>449</th>\n",
       "      <td>450.0</td>\n",
       "      <td>540</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>275</th>\n",
       "      <td>276.0</td>\n",
       "      <td>518</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "     user_id  RateCount\n",
       "404    405.0        737\n",
       "654    655.0        685\n",
       "12      13.0        636\n",
       "449    450.0        540\n",
       "275    276.0        518"
      ]
     },
     "execution_count": 257,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# количество оценок \n",
    "df_RateCount = pd.DataFrame(df.groupby('user_id', as_index = False)['rating'].agg('count')\n",
    "                              .rename(columns={'rating' : 'RateCount'}))\n",
    "df_RateCount.sort_values('RateCount',ascending=False).head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 259,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<bound method Series.mean of 0      272\n",
       "1       62\n",
       "2       54\n",
       "3       24\n",
       "4      175\n",
       "      ... \n",
       "938     49\n",
       "939    107\n",
       "940     22\n",
       "941     79\n",
       "942    168\n",
       "Name: RateCount, Length: 943, dtype: int64>"
      ]
     },
     "execution_count": 259,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_RateCount['RateCount'].mean"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "eVXL8RCt1nu5",
    "outputId": "b07ab2ee-c62c-499f-f5a6-3771260715ab"
   },
   "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>title</th>\n",
       "      <th>RateMean</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1472</th>\n",
       "      <td>They Made Me a Criminal (1939)</td>\n",
       "      <td>5.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>944</th>\n",
       "      <td>Marlene Dietrich: Shadow and Light (1996)</td>\n",
       "      <td>5.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1273</th>\n",
       "      <td>Saint of Fort Washington, The (1993)</td>\n",
       "      <td>5.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1359</th>\n",
       "      <td>Someone Else's America (1995)</td>\n",
       "      <td>5.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1387</th>\n",
       "      <td>Star Kid (1997)</td>\n",
       "      <td>5.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                           title  RateMean\n",
       "1472              They Made Me a Criminal (1939)       5.0\n",
       "944   Marlene Dietrich: Shadow and Light (1996)        5.0\n",
       "1273        Saint of Fort Washington, The (1993)       5.0\n",
       "1359               Someone Else's America (1995)       5.0\n",
       "1387                             Star Kid (1997)       5.0"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# средняя оценка фильма\n",
    "df_RateMean = pd.DataFrame(df.groupby('title', as_index = False)['rating'].agg('mean')\n",
    "                              .rename(columns={'rating' : 'RateMean'}))\n",
    "df_RateMean.sort_values('RateMean',ascending=False).head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "R1IP8gH51nxY",
    "outputId": "d389311b-3a07-441e-f3b6-78b570182cf7"
   },
   "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>title</th>\n",
       "      <th>RateCount</th>\n",
       "      <th>RateMean</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1398</th>\n",
       "      <td>Star Wars (1977)</td>\n",
       "      <td>583</td>\n",
       "      <td>4.358491</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>333</th>\n",
       "      <td>Contact (1997)</td>\n",
       "      <td>509</td>\n",
       "      <td>3.803536</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>498</th>\n",
       "      <td>Fargo (1996)</td>\n",
       "      <td>508</td>\n",
       "      <td>4.155512</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1234</th>\n",
       "      <td>Return of the Jedi (1983)</td>\n",
       "      <td>507</td>\n",
       "      <td>4.007890</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>860</th>\n",
       "      <td>Liar Liar (1997)</td>\n",
       "      <td>485</td>\n",
       "      <td>3.156701</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                          title  RateCount  RateMean\n",
       "1398           Star Wars (1977)        583  4.358491\n",
       "333              Contact (1997)        509  3.803536\n",
       "498                Fargo (1996)        508  4.155512\n",
       "1234  Return of the Jedi (1983)        507  4.007890\n",
       "860            Liar Liar (1997)        485  3.156701"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# мержим саб-сеты\n",
    "df_movie_rate = pd.merge(df_RateCount,df_RateMean, on='title')\n",
    "df_movie_rate.sort_values('RateCount',ascending=False).head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Pvzdv33j1n0H",
    "outputId": "6053cb38-c2a5-4fa3-d2cd-480f1208bfb2"
   },
   "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>rating</th>\n",
       "      <th>RateCount</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.0</td>\n",
       "      <td>6110</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2.0</td>\n",
       "      <td>11370</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3.0</td>\n",
       "      <td>27145</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4.0</td>\n",
       "      <td>34174</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5.0</td>\n",
       "      <td>21201</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   rating  RateCount\n",
       "0     1.0       6110\n",
       "1     2.0      11370\n",
       "2     3.0      27145\n",
       "3     4.0      34174\n",
       "4     5.0      21201"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# распределение оценок по пользователям\n",
    "df_distrib_of_ratings= pd.DataFrame(ratings.groupby('rating', as_index = False)['user_id'].count()\n",
    "                         .rename(columns={'user_id' : 'RateCount'}))\n",
    "df_distrib_of_ratings.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "jud5PS1q1n3B",
    "outputId": "dee067a2-f8e9-4d5f-9ab4-3b71e9f4b2e0"
   },
   "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>Total</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>user_id</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>405.0</th>\n",
       "      <td>737</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>655.0</th>\n",
       "      <td>685</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13.0</th>\n",
       "      <td>636</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>450.0</th>\n",
       "      <td>540</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>276.0</th>\n",
       "      <td>518</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         Total\n",
       "user_id       \n",
       "405.0      737\n",
       "655.0      685\n",
       "13.0       636\n",
       "450.0      540\n",
       "276.0      518"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# the users with most ratings\n",
    "users_rate = ratings[['user_id', 'movie_id']].groupby('user_id').count()\n",
    "users_rate = users_rate.rename(columns={'movie_id' : 'Total'})\n",
    "users_rate.sort_values('Total',ascending=False).head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "nxYgC4qU1n_e",
    "outputId": "60085b8e-a5d8-4285-f8c3-b2c62ed081eb"
   },
   "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>genre</th>\n",
       "      <th>Total ratings</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>Drama</td>\n",
       "      <td>26189</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Comedy</td>\n",
       "      <td>18342</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Action</td>\n",
       "      <td>11131</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>Romance</td>\n",
       "      <td>9076</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>Thriller</td>\n",
       "      <td>5820</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Adventure</td>\n",
       "      <td>5214</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>Sci-Fi</td>\n",
       "      <td>4647</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>Crime</td>\n",
       "      <td>3505</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>Horror</td>\n",
       "      <td>3279</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>Mystery</td>\n",
       "      <td>2875</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        genre  Total ratings\n",
       "7       Drama          26189\n",
       "4      Comedy          18342\n",
       "0      Action          11131\n",
       "13    Romance           9076\n",
       "15   Thriller           5820\n",
       "1   Adventure           5214\n",
       "14     Sci-Fi           4647\n",
       "5       Crime           3505\n",
       "10     Horror           3279\n",
       "12    Mystery           2875"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# наиболее популярные жанры\n",
    "MP_genre = df.groupby('genre', as_index = False)['user_id'].count()\n",
    "MP_genre = MP_genre.rename(columns={'user_id' : 'Total ratings'})\n",
    "MP_genre.sort_values('Total ratings',ascending=False).head(10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "vObooxcQ3o1F"
   },
   "source": [
    "### Визуализация"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ZscNF7SM1oH6"
   },
   "outputs": [],
   "source": [
    "plt.style.use('ggplot')\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "sns.set_style('white')\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "Azgbh1dy5Nrl"
   },
   "source": [
    "#### Распределение оценок для количества фильмов (минимальное значение 50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "47oV74b41oK9",
    "outputId": "d4183c3e-89d7-477a-bb12-8abd8f7cb080"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0,0.5,'Количество фильмов')"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(12,6))\n",
    "df_movie_rate[df_movie_rate['RateCount'] > 50]['RateCount'].hist(bins=100)\n",
    "plt.title('Количество оценок/количество фильмов')\n",
    "plt.xlabel('Количество оценок')\n",
    "plt.ylabel('Количество фильмов')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "luFojZMR5QLl"
   },
   "source": [
    "#### Распределение шкалы оценок"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "bxDFG_nz5M4-",
    "outputId": "da93f225-713f-4806-cd7e-d1f83d67cff0"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<seaborn.axisgrid.JointGrid at 0x26c82751eb8>"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x432 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.jointplot(x='rating',y='RateCount',data=df_distrib_of_ratings,alpha=0.5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "jacDwvqM5ltg"
   },
   "source": [
    "#### Зависимость количества оценок от средней оценки"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "azZvSpjo5M75",
    "outputId": "0e87dfd1-3add-4cc7-b9f8-5a79e1cb7a78"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<seaborn.axisgrid.JointGrid at 0x26c8580b978>"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x432 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.jointplot(x='RateMean',y='RateCount',data= df_movie_rate,alpha=0.5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "rqgsQT225mlN"
   },
   "source": [
    "#### Количество оценок пользователей"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "jYi1au3S5M-y",
    "outputId": "100b46fb-d1d6-4498-e67c-0be61fac63fb"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(12,8))\n",
    "users_rate['Total'].hist(bins=100, edgecolor='black', log=True)\n",
    "plt.title('Количество оценок пользователей')\n",
    "plt.xlabel('Число оценок')\n",
    "plt.ylabel('Число пользователей')\n",
    "plt.xlim(0,)\n",
    "plt.xticks(np.arange(0,1000,100))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "kqg2eme7z7-h"
   },
   "source": [
    "Выводы:\n",
    "1. Большая часть фильмов имеет менее 4000 оценок\n",
    "2. Пик развиия киноиндустрии пришелся на начало 90-х\n",
    "3. Большая часть оценок равна или больше 3\n",
    "4. Большое число оценок завышает рейтинг фильма\n",
    "5. Большая часть пользователей имеет диапозон от 0 до 200  оценок"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "nT3PLED4DAIW"
   },
   "source": [
    "### Metrics.Train/test split"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "gnVEAFRVwThC"
   },
   "source": [
    "Task 2: Evaluation metrics report (5 points)\n",
    "\n",
    "Implement evaluation metrics to assess recommender performance\n",
    "- MAP score\n",
    "- Precision@k\n",
    "- Mean percentage ranking"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Функция для деления на тест и трейн"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "def Split_DF(df, holdout_fraction=0.1):\n",
    "    data =  df.sort_values(['date'])\n",
    "    train = data[:(len(df.index)-int(len(df.index)*holdout_fraction))]\n",
    "    test = data[(len(df.index)-int(len(df.index)*holdout_fraction)):]\n",
    "    return train, test"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Метрики для оценки качества системы"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 391,
   "metadata": {},
   "outputs": [],
   "source": [
    "def precision(actual, predicted, k):\n",
    "    try:\n",
    "        act_set = set(actual[:k])\n",
    "        pred_set = set(predicted[:k])\n",
    "        result = len(act_set & pred_set) / float(min(k, len(act_set)))\n",
    "    except ZeroDivisionError:\n",
    "        result = 0\n",
    "        \n",
    "    return result\n",
    "\n",
    "def average_precision_at_k(actual, predicted, k=10):\n",
    "\n",
    "    if len(predicted)>k:\n",
    "        predicted = predicted[:k]\n",
    "        \n",
    "    score = 0.0\n",
    "    num_hits = 0.0\n",
    "\n",
    "    for i,p in enumerate(predicted):\n",
    "        if p in actual and p not in predicted[:i]:\n",
    "            num_hits += 1.0\n",
    "            score += num_hits / (i+1.0)    \n",
    "    try:        \n",
    "        result = score / float(min(k, len(actual)))\n",
    "    except ZeroDivisionError:\n",
    "        result = 0\n",
    "    \n",
    "    if len(actual)<k:\n",
    "        result = score / float(min(k, len(actual)))\n",
    "        \n",
    "    return result\n",
    "\n",
    "def mean_average_precision_at_k(actual, predicted, k=10):\n",
    "\n",
    "    return np.mean([apk(a,p,k) for a,p in zip(actual, predicted)])\n",
    "\n",
    "\n",
    "def PR(actual, predicted, k):\n",
    "    \n",
    "    if len(predicted)>k:\n",
    "        predicted = predicted[:k]\n",
    "    \n",
    "    if len(actual)>k:\n",
    "        actual = actual[:k] \n",
    "    \n",
    "    score = 0.0\n",
    "    \n",
    "    for pred_values, rank in zip(predicted, range(len(predicted))):\n",
    "        if pred_values in actual and pred_values in predicted:\n",
    "            score += (rank+1.0) / (len(predicted))\n",
    "                                  \n",
    "    result = score / float(min(k, len(actual)))                             \n",
    "                                 \n",
    "    return score "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 405,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calculate_metrics( preds, test, k):\n",
    "        users =  preds['user_id'].values.tolist()\n",
    "        actuals_items = []\n",
    "        preds_items = []\n",
    "        prec =[]\n",
    "        apak = []\n",
    "        percantage = []\n",
    "        \n",
    "        for i in users:\n",
    "            actual = test[test['user_id']== i].sort_values('rating', ascending= False)['movie_id'].tolist()\n",
    "            pred = preds[preds['user_id']== i]['predictions'].values[0]\n",
    "            actuals_items.append(actual)\n",
    "            preds_items.append(pred)\n",
    "            prec.append(precision(actual,pred,k))\n",
    "            apak.append(average_precision_at_k(actual,pred,k))\n",
    "        \n",
    "        df_metrics = pd.DataFrame({'user_id': users,\n",
    "                                   'actual': actuals_items,\n",
    "                                   'predictions': preds_items,\n",
    "                                   'precission': prec,\n",
    "                                   'ap@k': apak}, \n",
    "                                    columns = ['user_id','actual','predictions', 'precission' , 'ap@k'])\n",
    "        \n",
    "        for i in df_metrics['user_id'].values.tolist():\n",
    "            actual = df_metrics[df_metrics['user_id']== i]['actual'].values[0]\n",
    "            preds = df_metrics[df_metrics['user_id']== i]['predictions'].values[0]\n",
    "            pr_rank = PR(actual,preds,k)\n",
    "            percantage.append(pr_rank)\n",
    "        \n",
    "        df_metrics['PR']=percantage\n",
    "        \n",
    "        return df_metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MPRecommender(object):\n",
    "    def __init__(self, user_field='user_id', item_field='movie_id', number_of_films=10):\n",
    "        self.user_field = user_field\n",
    "        self.item_field = item_field\n",
    "        self.number_of_films = number_of_films\n",
    "\n",
    "    \n",
    "    def fit(self, train):\n",
    "        popularity_recs = train[self.item_field].value_counts().head(self.number_of_films).index.tolist()\n",
    "        recs = []\n",
    "        user_list = train[self.user_field].values.tolist()\n",
    "        for user in user_list:\n",
    "            actual = train[train[self.user_field] == user].sort_values('rating',ascending=False)[self.item_field].values\n",
    "            pop_predictions = popularity_recs\n",
    "            cleaned_preds = [x for x in pop_predictions if x not in actual]\n",
    "            recs.append(cleaned_preds)\n",
    "            \n",
    "        train['pop_predictions'] = recs\n",
    "    \n",
    "        self.train_result = train\n",
    "        \n",
    "        return train\n",
    "\n",
    "    def predict(self, user_subset):\n",
    "        df_predict= self.train_result[self.train_result[self.user_field].isin(user_subset)]\n",
    "        df_predict = df_predict.drop_duplicates(subset=['user_id'], keep='first', inplace=False)\n",
    "        users = []\n",
    "        predictions = []\n",
    "\n",
    "        for user in user_subset:\n",
    "            try:\n",
    "                preds = df_predict[df_predict[self.user_field] == user ]['pop_predictions'].tolist()[0]\n",
    "            except IndexError:\n",
    "                preds = []\n",
    "            users.append(user)\n",
    "            predictions.append(preds)\n",
    "            \n",
    "        df_preds = pd.DataFrame({'user_id': users,\n",
    "                                'predictions': predictions}, \n",
    "                                columns = ['user_id','predictions'])\n",
    "        \n",
    "        df_preds = df_preds[df_preds['predictions'].map(lambda d: len(d)) > 0]\n",
    "        \n",
    "        \n",
    "        return df_preds\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 395,
   "metadata": {},
   "outputs": [],
   "source": [
    "train, test = Split_DF(df_copy, holdout_fraction=0.25)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 283,
   "metadata": {},
   "outputs": [],
   "source": [
    "user_subset = list(set(test['user_id'].values.tolist()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 284,
   "metadata": {},
   "outputs": [],
   "source": [
    "min_k = df_RateCount[df_RateCount['user_id'].isin(user_subset)]\n",
    "min_k = round(min_k['RateCount'].mean())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 285,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "126"
      ]
     },
     "execution_count": 285,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "min_k"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_MPR = MPRecommender(user_field='user_id', item_field='movie_id', number_of_films = min_k )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "trained = model_MPR.fit(train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [],
   "source": [
    "predictions = model_MPR.predict(user_subset)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 396,
   "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>user_id</th>\n",
       "      <th>predictions</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.0</td>\n",
       "      <td>[294.0, 288.0, 286.0, 405.0, 300.0, 423.0, 276...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>13.0</td>\n",
       "      <td>[151.0, 15.0, 257.0, 742.0, 125.0, 186.0, 269....</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>14.0</td>\n",
       "      <td>[50.0, 181.0, 294.0, 258.0, 1.0, 286.0, 174.0,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>23.0</td>\n",
       "      <td>[288.0, 121.0, 286.0, 127.0, 237.0, 117.0, 300...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>43.0</td>\n",
       "      <td>[288.0, 195.0, 748.0, 22.0, 96.0, 176.0, 183.0...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>345</th>\n",
       "      <td>894.0</td>\n",
       "      <td>[181.0, 294.0, 174.0, 127.0, 56.0, 98.0, 222.0...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>350</th>\n",
       "      <td>913.0</td>\n",
       "      <td>[294.0, 258.0, 121.0, 286.0, 405.0, 300.0, 257...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>354</th>\n",
       "      <td>924.0</td>\n",
       "      <td>[294.0, 237.0, 98.0, 222.0, 405.0, 204.0, 151....</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>360</th>\n",
       "      <td>940.0</td>\n",
       "      <td>[1.0, 288.0, 121.0, 127.0, 237.0, 117.0, 222.0...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>362</th>\n",
       "      <td>943.0</td>\n",
       "      <td>[294.0, 258.0, 1.0, 288.0, 286.0, 7.0, 237.0, ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>128 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     user_id                                        predictions\n",
       "0        1.0  [294.0, 288.0, 286.0, 405.0, 300.0, 423.0, 276...\n",
       "6       13.0  [151.0, 15.0, 257.0, 742.0, 125.0, 186.0, 269....\n",
       "7       14.0  [50.0, 181.0, 294.0, 258.0, 1.0, 286.0, 174.0,...\n",
       "8       23.0  [288.0, 121.0, 286.0, 127.0, 237.0, 117.0, 300...\n",
       "17      43.0  [288.0, 195.0, 748.0, 22.0, 96.0, 176.0, 183.0...\n",
       "..       ...                                                ...\n",
       "345    894.0  [181.0, 294.0, 174.0, 127.0, 56.0, 98.0, 222.0...\n",
       "350    913.0  [294.0, 258.0, 121.0, 286.0, 405.0, 300.0, 257...\n",
       "354    924.0  [294.0, 237.0, 98.0, 222.0, 405.0, 204.0, 151....\n",
       "360    940.0  [1.0, 288.0, 121.0, 127.0, 237.0, 117.0, 222.0...\n",
       "362    943.0  [294.0, 258.0, 1.0, 288.0, 286.0, 7.0, 237.0, ...\n",
       "\n",
       "[128 rows x 2 columns]"
      ]
     },
     "execution_count": 396,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 406,
   "metadata": {},
   "outputs": [],
   "source": [
    "get_preds= calculate_metrics(predictions, test,  k=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 407,
   "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>user_id</th>\n",
       "      <th>actual</th>\n",
       "      <th>predictions</th>\n",
       "      <th>precission</th>\n",
       "      <th>ap@k</th>\n",
       "      <th>PR</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.0</td>\n",
       "      <td>[270.0, 32.0, 242.0, 111.0, 171.0, 209.0, 256....</td>\n",
       "      <td>[294.0, 288.0, 286.0, 405.0, 300.0, 423.0, 276...</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>0.010000</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>13.0</td>\n",
       "      <td>[900.0, 909.0, 915.0, 896.0, 691.0, 694.0, 917...</td>\n",
       "      <td>[151.0, 15.0, 257.0, 742.0, 125.0, 186.0, 269....</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>0.045238</td>\n",
       "      <td>0.6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>14.0</td>\n",
       "      <td>[181.0, 921.0, 213.0, 525.0, 50.0, 509.0, 172....</td>\n",
       "      <td>[50.0, 181.0, 294.0, 258.0, 1.0, 286.0, 174.0,...</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.282857</td>\n",
       "      <td>0.3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>23.0</td>\n",
       "      <td>[257.0]</td>\n",
       "      <td>[288.0, 121.0, 286.0, 127.0, 237.0, 117.0, 300...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>43.0</td>\n",
       "      <td>[316.0, 354.0, 124.0, 1052.0]</td>\n",
       "      <td>[288.0, 195.0, 748.0, 22.0, 96.0, 176.0, 183.0...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</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",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>123</th>\n",
       "      <td>894.0</td>\n",
       "      <td>[903.0, 316.0, 1592.0, 904.0, 752.0, 1313.0, 6...</td>\n",
       "      <td>[181.0, 294.0, 174.0, 127.0, 56.0, 98.0, 222.0...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>124</th>\n",
       "      <td>913.0</td>\n",
       "      <td>[258.0]</td>\n",
       "      <td>[294.0, 258.0, 121.0, 286.0, 405.0, 300.0, 257...</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>125</th>\n",
       "      <td>924.0</td>\n",
       "      <td>[275.0, 237.0, 129.0, 1149.0, 273.0, 408.0, 27...</td>\n",
       "      <td>[294.0, 237.0, 98.0, 222.0, 405.0, 204.0, 151....</td>\n",
       "      <td>0.142857</td>\n",
       "      <td>0.071429</td>\n",
       "      <td>0.2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>126</th>\n",
       "      <td>940.0</td>\n",
       "      <td>[354.0, 316.0, 873.0, 879.0, 355.0]</td>\n",
       "      <td>[1.0, 288.0, 121.0, 127.0, 237.0, 117.0, 222.0...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>127</th>\n",
       "      <td>943.0</td>\n",
       "      <td>[69.0, 233.0, 187.0, 196.0, 194.0, 184.0, 56.0...</td>\n",
       "      <td>[294.0, 258.0, 1.0, 288.0, 286.0, 7.0, 237.0, ...</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>0.072619</td>\n",
       "      <td>0.8</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>128 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     user_id                                             actual  \\\n",
       "0        1.0  [270.0, 32.0, 242.0, 111.0, 171.0, 209.0, 256....   \n",
       "1       13.0  [900.0, 909.0, 915.0, 896.0, 691.0, 694.0, 917...   \n",
       "2       14.0  [181.0, 921.0, 213.0, 525.0, 50.0, 509.0, 172....   \n",
       "3       23.0                                            [257.0]   \n",
       "4       43.0                      [316.0, 354.0, 124.0, 1052.0]   \n",
       "..       ...                                                ...   \n",
       "123    894.0  [903.0, 316.0, 1592.0, 904.0, 752.0, 1313.0, 6...   \n",
       "124    913.0                                            [258.0]   \n",
       "125    924.0  [275.0, 237.0, 129.0, 1149.0, 273.0, 408.0, 27...   \n",
       "126    940.0                [354.0, 316.0, 873.0, 879.0, 355.0]   \n",
       "127    943.0  [69.0, 233.0, 187.0, 196.0, 194.0, 184.0, 56.0...   \n",
       "\n",
       "                                           predictions  precission      ap@k  \\\n",
       "0    [294.0, 288.0, 286.0, 405.0, 300.0, 423.0, 276...    0.100000  0.010000   \n",
       "1    [151.0, 15.0, 257.0, 742.0, 125.0, 186.0, 269....    0.100000  0.045238   \n",
       "2    [50.0, 181.0, 294.0, 258.0, 1.0, 286.0, 174.0,...    0.200000  0.282857   \n",
       "3    [288.0, 121.0, 286.0, 127.0, 237.0, 117.0, 300...    0.000000  0.000000   \n",
       "4    [288.0, 195.0, 748.0, 22.0, 96.0, 176.0, 183.0...    0.000000  0.000000   \n",
       "..                                                 ...         ...       ...   \n",
       "123  [181.0, 294.0, 174.0, 127.0, 56.0, 98.0, 222.0...    0.000000  0.000000   \n",
       "124  [294.0, 258.0, 121.0, 286.0, 405.0, 300.0, 257...    1.000000  0.500000   \n",
       "125  [294.0, 237.0, 98.0, 222.0, 405.0, 204.0, 151....    0.142857  0.071429   \n",
       "126  [1.0, 288.0, 121.0, 127.0, 237.0, 117.0, 222.0...    0.000000  0.000000   \n",
       "127  [294.0, 258.0, 1.0, 288.0, 286.0, 7.0, 237.0, ...    0.100000  0.072619   \n",
       "\n",
       "      PR  \n",
       "0    1.0  \n",
       "1    0.6  \n",
       "2    0.3  \n",
       "3    0.0  \n",
       "4    0.0  \n",
       "..   ...  \n",
       "123  0.0  \n",
       "124  0.2  \n",
       "125  0.2  \n",
       "126  0.0  \n",
       "127  0.8  \n",
       "\n",
       "[128 rows x 6 columns]"
      ]
     },
     "execution_count": 407,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "get_preds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 408,
   "metadata": {},
   "outputs": [],
   "source": [
    "MP = get_preds['precission'].mean()\n",
    "MAPAK = get_preds['ap@k'].mean()\n",
    "MPR = get_preds['PR'].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 410,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.07370721726190474"
      ]
     },
     "execution_count": 410,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MP"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 411,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.0945583767361111"
      ]
     },
     "execution_count": 411,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MAPAK"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 409,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.32187499999999997"
      ]
     },
     "execution_count": 409,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MPR"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Метрики MP Recommender\n",
    "- mean precission - 0,074\n",
    "- mean average precisiion at k - 0,095\n",
    "- mean percantage rank - 0,321"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SVD factorization"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "ckg8gsV-DAJU"
   },
   "source": [
    "#### Collaborative filtering using sklearn surprise lib"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "_uddAqNi0A2L"
   },
   "outputs": [],
   "source": [
    "#!pip install scikit-surprise\n",
    "from surprise import Reader\n",
    "from surprise import Dataset\n",
    "import io\n",
    "from surprise.model_selection import cross_validate\n",
    "from surprise import NormalPredictor\n",
    "from surprise import KNNBasic\n",
    "from surprise import KNNBaseline\n",
    "from surprise import SVD\n",
    "from surprise import BaselineOnly\n",
    "from surprise import SVDpp\n",
    "from surprise import CoClustering\n",
    "from surprise.accuracy import rmse\n",
    "from surprise import accuracy\n",
    "from surprise.model_selection import train_test_split\n",
    "from surprise.model_selection import GridSearchCV"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "SsX6jfMMDAJm"
   },
   "outputs": [],
   "source": [
    "class Matrix_Recomender(object):\n",
    "    def __init__(self, user_field='user_id', item_field='movie_id', number_of_ratings=5):\n",
    "        self.user_field = user_field\n",
    "        self.item_field = item_field\n",
    "        self.number_of_ratings = number_of_ratings\n",
    "    \n",
    "    def _preprocessing(self,movie_df):\n",
    "        movie_df = movie_df.iloc[:,0:5]\n",
    "        \n",
    "        min_item_clicks = self.number_of_ratings\n",
    "        filter_items = movie_df [self.item_field].value_counts() > min_item_clicks\n",
    "        filter_items = filter_items[filter_items].index.tolist()\n",
    "\n",
    "        min_user_clicks = self.number_of_ratings\n",
    "        filter_users = movie_df [self.user_field ].value_counts() > min_user_clicks\n",
    "        filter_users = filter_users[filter_users].index.tolist()\n",
    "\n",
    "        df_filtered = movie_df [(movie_df [self.item_field].isin(filter_items)) & (movie_df [self.user_field].isin(filter_users))]\n",
    "        data =  df_filtered.sort_values(['date'])\n",
    "\n",
    "        print('Размер оригинального сета:\\t{}'.format(movie_df.shape))\n",
    "        print('Размер очищенного сета:\\t{}'.format(df_filtered.shape))\n",
    "        \n",
    "        return data\n",
    "      \n",
    "    def _reader(self, data):\n",
    "        reader = Reader(rating_scale=(1, 5))\n",
    "        sup_data = Dataset.load_from_df(data[['user_id', 'movie_id', 'rating']], reader)\n",
    "        \n",
    "        return sup_data\n",
    "        \n",
    "    def best_of_algo(self,df):\n",
    "        \n",
    "        data = self._preprocessing(df)\n",
    "        sup_data = self._reader(data)\n",
    "        benchmark = []\n",
    "        \n",
    "        for algo in [SVD(),SVDpp(), KNNBaseline(), KNNBasic(), BaselineOnly(), NormalPredictor(), CoClustering()]:\n",
    "\n",
    "            print('Начало кросс-валидации для {}'.format(algo))\n",
    "            results = cross_validate(algo, sup_data, measures=['RMSE'], cv=5, verbose=False)\n",
    "\n",
    "            tmp = pd.DataFrame.from_dict(results).mean(axis=0)\n",
    "            tmp = tmp.append(pd.Series([str(algo).split(' ')[0].split('.')[-1]], index=['Algorithm']))\n",
    "            benchmark.append(tmp)\n",
    "            \n",
    "        surprise_results = pd.DataFrame(benchmark).set_index('Algorithm').sort_values('test_rmse')\n",
    "        \n",
    "        return surprise_results\n",
    "\n",
    "    def GS(self, algo_name, param_grid, metrics, folds, data):\n",
    "        \n",
    "\n",
    "        gs = GridSearchCV(algo_name, param_grid, measures = metrics,  cv = folds)\n",
    "        \n",
    "        gs.fit(data)\n",
    "\n",
    "        # best RMSE score\n",
    "        print(gs.best_score['rmse'])\n",
    "\n",
    "        # combination of parameters that gave the best RMSE score\n",
    "        print(gs.best_params['rmse'])\n",
    "        \n",
    "        return gs.best_params['rmse']\n",
    "        \n",
    "\n",
    "    \n",
    "    def fit(self, train, best_algo_name, params):\n",
    "        \n",
    "        self.data_train = self._preprocessing(train)\n",
    "        self.trainset = self._reader(self.data_train).build_full_trainset()\n",
    "        algo = best_algo_name(n_factors = params[0],  \n",
    "                   n_epochs = params[1], \n",
    "                   lr_all = params[2], \n",
    "                   reg_all = params[3])\n",
    "        \n",
    "        \n",
    "        #сохраняем зафиченные алгоритм\n",
    "        self.fit_result = algo.fit(self.trainset)\n",
    "        \n",
    "        return self.fit_result\n",
    "\n",
    "    def predict(self, user_subset, k=3):\n",
    "        \n",
    "        #делаем предсказания для юзеров из трейна\n",
    "        testset = self.trainset.build_anti_testset()\n",
    "        \n",
    "        #делаем предсказания для юзеров\n",
    "        predictions = self.fit_result.test(testset)\n",
    "        \n",
    "        df = self._get_all_predictions(predictions,  user_subset, k)\n",
    "\n",
    "        \n",
    "        return df\n",
    "    \n",
    "    def _get_all_predictions(self, predictions, user_subset, k=3):\n",
    "\n",
    "        top_n = defaultdict(list)\n",
    "        recommended_movie_ids=[]\n",
    "        batch_movie_ids=[]\n",
    "        user_id_list = []\n",
    "        recs = []\n",
    "        num_of_k = np.arange(0, k)\n",
    "\n",
    "        #достаем предсказания для юзеров из трейна\n",
    "        for uid, iid, true_r, est, _ in predictions:\n",
    "            top_n[uid].append((iid, est))\n",
    "            \n",
    "        #достаем только топ-к фильмов  \n",
    "        for uid, user_ratings in top_n.items():\n",
    "            user_ratings.sort(key=lambda x: x[1], reverse=True)\n",
    "            top_n[uid] = user_ratings[:k]\n",
    "            \n",
    "        #переводим в датафрейм\n",
    "        tmp = pd.DataFrame.from_dict(top_n)\n",
    "        tmp_transpose = tmp.transpose()\n",
    "        tmp_transpose[self.user_field]=tmp_transpose.index\n",
    "        \n",
    "        #смотрим для каких юзеров у нас есть предсказания\n",
    "        for user_id in user_subset:\n",
    "            if user_id not in tmp_transpose[self.user_field].values.tolist():\n",
    "                print('Пользователь с ID номер ' + str(user_id) + ' не имеет рекомендаций')\n",
    "            else:\n",
    "                user_id_list.append(user_id)\n",
    "                \n",
    "        #обертка для перевода в новый датафрейм            \n",
    "        for user_id in user_id_list:\n",
    "            for k in num_of_k:\n",
    "                results = tmp_transpose[tmp_transpose[self.user_field] == user_id].values.tolist()\n",
    "                batch_movie_ids.append(results[0][k][0])\n",
    "     \n",
    "        start = np.arange(0,len(batch_movie_ids),k)\n",
    "        end = np.arange(k,len(batch_movie_ids)+k,k)\n",
    "        \n",
    "        #обертка для перевода в новый датафрейм \n",
    "        for z, j in zip(start, end):\n",
    "            movie_subset = batch_movie_ids[z:j]\n",
    "            recommended_movie_ids.append(movie_subset)\n",
    "            \n",
    "        #убираем из предсказаний просмотренные фильмы\n",
    "        for user, count in zip(user_id_list, range(len(user_id_list))):\n",
    "            actual = self.data_train[self.data_train[self.user_field] == user].sort_values('rating',ascending=False)[self.item_field].values.tolist()\n",
    "            pop_predictions = recommended_movie_ids[count]\n",
    "            cleaned_preds = [x for x in pop_predictions if x not in actual]\n",
    "            recs.append(cleaned_preds)\n",
    "            \n",
    "        \n",
    "        df = pd.DataFrame({'user_id': user_id_list,\n",
    "                               'predictions': recs}, \n",
    "                                columns = ['user_id', 'predictions'])       \n",
    "\n",
    "        \n",
    "        return df\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_SVD = Matrix_Recomender(user_field='user_id', item_field='movie_id', number_of_ratings=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Размер оригинального сета:\t(100000, 5)\n",
      "Размер очищенного сета:\t(97623, 5)\n",
      "Начало кросс-валидации для <surprise.prediction_algorithms.matrix_factorization.SVD object at 0x0000020FAD92D898>\n",
      "Начало кросс-валидации для <surprise.prediction_algorithms.matrix_factorization.SVDpp object at 0x0000020FAD92D6A0>\n",
      "Начало кросс-валидации для <surprise.prediction_algorithms.knns.KNNBaseline object at 0x0000020FAD92D7F0>\n",
      "Estimating biases using als...\n",
      "Computing the msd similarity matrix...\n",
      "Done computing similarity matrix.\n",
      "Estimating biases using als...\n",
      "Computing the msd similarity matrix...\n",
      "Done computing similarity matrix.\n",
      "Estimating biases using als...\n",
      "Computing the msd similarity matrix...\n",
      "Done computing similarity matrix.\n",
      "Estimating biases using als...\n",
      "Computing the msd similarity matrix...\n",
      "Done computing similarity matrix.\n",
      "Estimating biases using als...\n",
      "Computing the msd similarity matrix...\n",
      "Done computing similarity matrix.\n",
      "Начало кросс-валидации для <surprise.prediction_algorithms.knns.KNNBasic object at 0x0000020FAD92D828>\n",
      "Computing the msd similarity matrix...\n",
      "Done computing similarity matrix.\n",
      "Computing the msd similarity matrix...\n",
      "Done computing similarity matrix.\n",
      "Computing the msd similarity matrix...\n",
      "Done computing similarity matrix.\n",
      "Computing the msd similarity matrix...\n",
      "Done computing similarity matrix.\n",
      "Computing the msd similarity matrix...\n",
      "Done computing similarity matrix.\n",
      "Начало кросс-валидации для <surprise.prediction_algorithms.baseline_only.BaselineOnly object at 0x0000020FAD92D630>\n",
      "Estimating biases using als...\n",
      "Estimating biases using als...\n",
      "Estimating biases using als...\n",
      "Estimating biases using als...\n",
      "Estimating biases using als...\n",
      "Начало кросс-валидации для <surprise.prediction_algorithms.random_pred.NormalPredictor object at 0x0000020FAD92D7B8>\n",
      "Начало кросс-валидации для <surprise.prediction_algorithms.co_clustering.CoClustering object at 0x0000020FAD92D748>\n"
     ]
    }
   ],
   "source": [
    "best = model_SVD.best_of_algo(df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {
    "scrolled": true
   },
   "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>test_rmse</th>\n",
       "      <th>fit_time</th>\n",
       "      <th>test_time</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Algorithm</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>SVDpp</th>\n",
       "      <td>0.915619</td>\n",
       "      <td>150.145385</td>\n",
       "      <td>3.071409</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>KNNBaseline</th>\n",
       "      <td>0.925674</td>\n",
       "      <td>0.777766</td>\n",
       "      <td>3.999478</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SVD</th>\n",
       "      <td>0.933436</td>\n",
       "      <td>4.665604</td>\n",
       "      <td>0.268506</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BaselineOnly</th>\n",
       "      <td>0.940315</td>\n",
       "      <td>0.205793</td>\n",
       "      <td>0.115866</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CoClustering</th>\n",
       "      <td>0.950778</td>\n",
       "      <td>1.944917</td>\n",
       "      <td>0.119329</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>KNNBasic</th>\n",
       "      <td>0.967432</td>\n",
       "      <td>0.531007</td>\n",
       "      <td>3.234957</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>NormalPredictor</th>\n",
       "      <td>1.507241</td>\n",
       "      <td>0.128700</td>\n",
       "      <td>0.144448</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                 test_rmse    fit_time  test_time\n",
       "Algorithm                                        \n",
       "SVDpp             0.915619  150.145385   3.071409\n",
       "KNNBaseline       0.925674    0.777766   3.999478\n",
       "SVD               0.933436    4.665604   0.268506\n",
       "BaselineOnly      0.940315    0.205793   0.115866\n",
       "CoClustering      0.950778    1.944917   0.119329\n",
       "KNNBasic          0.967432    0.531007   3.234957\n",
       "NormalPredictor   1.507241    0.128700   0.144448"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Как видно первые три места делят SVDpp KNN и SVD\n",
    "- почему следует выбрать SVD:\n",
    "- Время на обучение у SVDpp самое большое и отклик у SVDpp тоже довольно долгий. В условиях когда у нас будет большой поток\n",
    "  обращения к алгоритму, например через API, это может привести к зависаниям и сбою системы.\n",
    "- Время на обучение у KNN самое маленькое в призовой тройке,а отклик самый долгий. Отметаем по той же причине.\n",
    "- SVD хоть и учится сравнительно дольше , но в десятки раз быстрее выдает предсказания. Выбираем свд"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "param_grid = {'n_factors':[20,40,80,120,160,200,240], \n",
    "              'n_epochs': [10,15,20,40,60,80,100], \n",
    "              'lr_all': [0.002, 0.004, 0.006, 0.008 , 0.01, 0.012, 0.015],\n",
    "              'reg_all': [0.2, 0.4, 0.6, 0.8, 0.1, 0.12, 0.15]}\n",
    "metrics=['rmse', 'mae']\n",
    "folds = 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "gs = Matrix_Recomender.GS(SVD,param_grid,metrics,folds,sup_data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Обучаем модель и делаем предсказания"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [],
   "source": [
    "best_params = [220, 100, 0.01, 0.2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [],
   "source": [
    "user_subset = list(set(test['user_id'].values.tolist()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Размер оригинального сета:\t(75000, 5)\n",
      "Размер очищенного сета:\t(73986, 5)\n"
     ]
    }
   ],
   "source": [
    "fit_d = model_SVD.fit(train, SVD, best_params)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Пользователь с ID номер 2.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 3.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 4.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 7.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 11.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 26.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 27.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 33.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 34.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 38.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 39.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 40.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 41.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 56.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 61.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 73.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 74.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 79.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 88.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 90.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 91.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 93.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 100.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 104.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 105.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 107.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 111.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 120.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 121.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 124.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 132.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 133.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 134.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 142.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 143.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 146.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 161.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 163.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 164.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 167.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 169.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 171.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 179.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 183.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 184.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 189.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 190.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 191.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 193.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 197.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 204.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 205.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 214.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 215.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 217.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 219.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 223.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 228.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 229.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 234.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 235.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 236.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 239.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 245.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 247.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 252.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 253.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 260.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 261.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 263.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 266.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 269.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 273.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 278.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 280.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 285.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 293.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 312.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 313.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 317.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 325.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 329.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 333.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 334.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 335.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 339.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 341.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 353.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 354.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 356.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 358.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 363.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 365.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 366.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 368.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 369.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 377.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 381.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 383.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 384.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 392.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 401.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 408.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 410.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 411.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 418.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 420.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 421.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 423.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 432.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 440.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 441.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 444.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 445.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 448.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 466.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 471.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 475.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 478.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 480.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 485.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 488.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 489.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 491.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 495.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 505.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 507.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 511.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 512.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 516.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 517.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 547.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 548.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 551.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 559.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 564.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 565.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 574.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 585.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 587.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 588.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 591.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 600.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 602.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 603.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 610.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 611.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 613.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 616.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 618.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 620.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 623.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 625.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 631.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 636.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 639.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 643.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 644.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 645.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 646.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 649.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 650.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 656.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 659.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 660.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 663.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 667.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 669.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 673.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 675.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 676.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 682.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 683.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 695.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 701.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 704.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 713.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 714.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 720.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 722.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 726.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 729.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 734.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 741.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 747.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 752.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 753.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 757.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 766.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 767.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 773.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 774.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 775.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 776.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 778.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 780.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 782.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 784.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 787.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 794.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 796.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 801.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 807.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 809.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 816.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 818.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 822.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 828.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 830.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 831.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 834.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 835.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 840.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 841.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 842.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 852.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 856.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 863.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 866.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 872.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 873.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 874.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 883.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 898.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 903.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 909.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 911.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 915.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 918.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 922.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 926.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 931.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 932.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 934.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 938.0 не имеет рекомендаций\n",
      "Пользователь с ID номер 942.0 не имеет рекомендаций\n"
     ]
    }
   ],
   "source": [
    "predictions_svd = model_SVD.predict(user_subset=user_subset, k=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "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>user_id</th>\n",
       "      <th>predictions</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.0</td>\n",
       "      <td>[408.0, 483.0, 603.0, 285.0, 313.0, 316.0, 318...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>13.0</td>\n",
       "      <td>[408.0, 113.0, 169.0, 114.0, 963.0, 1251.0, 11...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>14.0</td>\n",
       "      <td>[479.0, 251.0, 113.0, 313.0, 169.0, 64.0, 483....</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>23.0</td>\n",
       "      <td>[272.0, 12.0, 169.0, 113.0, 64.0, 114.0, 313.0...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>43.0</td>\n",
       "      <td>[316.0, 113.0, 483.0, 316.0, 603.0, 963.0]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>122</th>\n",
       "      <td>894.0</td>\n",
       "      <td>[64.0, 22.0, 316.0, 408.0]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>123</th>\n",
       "      <td>913.0</td>\n",
       "      <td>[1251.0, 272.0, 113.0, 313.0]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>124</th>\n",
       "      <td>924.0</td>\n",
       "      <td>[603.0, 316.0, 408.0, 169.0, 113.0, 483.0]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>125</th>\n",
       "      <td>940.0</td>\n",
       "      <td>[64.0, 127.0, 603.0, 114.0, 408.0, 113.0, 169....</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>126</th>\n",
       "      <td>943.0</td>\n",
       "      <td>[483.0, 318.0, 12.0, 603.0, 316.0, 114.0, 408....</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>127 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     user_id                                        predictions\n",
       "0        1.0  [408.0, 483.0, 603.0, 285.0, 313.0, 316.0, 318...\n",
       "1       13.0  [408.0, 113.0, 169.0, 114.0, 963.0, 1251.0, 11...\n",
       "2       14.0  [479.0, 251.0, 113.0, 313.0, 169.0, 64.0, 483....\n",
       "3       23.0  [272.0, 12.0, 169.0, 113.0, 64.0, 114.0, 313.0...\n",
       "4       43.0         [316.0, 113.0, 483.0, 316.0, 603.0, 963.0]\n",
       "..       ...                                                ...\n",
       "122    894.0                         [64.0, 22.0, 316.0, 408.0]\n",
       "123    913.0                      [1251.0, 272.0, 113.0, 313.0]\n",
       "124    924.0         [603.0, 316.0, 408.0, 169.0, 113.0, 483.0]\n",
       "125    940.0  [64.0, 127.0, 603.0, 114.0, 408.0, 113.0, 169....\n",
       "126    943.0  [483.0, 318.0, 12.0, 603.0, 316.0, 114.0, 408....\n",
       "\n",
       "[127 rows x 2 columns]"
      ]
     },
     "execution_count": 109,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predictions_svd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 412,
   "metadata": {},
   "outputs": [],
   "source": [
    "Svd_preds = calculate_metrics(predictions_svd, test, k=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 413,
   "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>user_id</th>\n",
       "      <th>actual</th>\n",
       "      <th>predictions</th>\n",
       "      <th>precission</th>\n",
       "      <th>ap@k</th>\n",
       "      <th>PR</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.0</td>\n",
       "      <td>[270.0, 32.0, 242.0, 111.0, 171.0, 209.0, 256....</td>\n",
       "      <td>[408.0, 483.0, 603.0, 285.0, 313.0, 316.0, 318...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>13.0</td>\n",
       "      <td>[900.0, 909.0, 915.0, 896.0, 691.0, 694.0, 917...</td>\n",
       "      <td>[408.0, 113.0, 169.0, 114.0, 963.0, 1251.0, 11...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>14.0</td>\n",
       "      <td>[181.0, 921.0, 213.0, 525.0, 50.0, 509.0, 172....</td>\n",
       "      <td>[479.0, 251.0, 113.0, 313.0, 169.0, 64.0, 483....</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.025000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>23.0</td>\n",
       "      <td>[257.0]</td>\n",
       "      <td>[272.0, 12.0, 169.0, 113.0, 64.0, 114.0, 313.0...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>43.0</td>\n",
       "      <td>[316.0, 354.0, 124.0, 1052.0]</td>\n",
       "      <td>[316.0, 113.0, 483.0, 316.0, 603.0, 963.0]</td>\n",
       "      <td>0.250000</td>\n",
       "      <td>0.250000</td>\n",
       "      <td>0.833333</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>122</th>\n",
       "      <td>894.0</td>\n",
       "      <td>[903.0, 316.0, 1592.0, 904.0, 752.0, 1313.0, 6...</td>\n",
       "      <td>[64.0, 22.0, 316.0, 408.0]</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>0.033333</td>\n",
       "      <td>0.750000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>123</th>\n",
       "      <td>913.0</td>\n",
       "      <td>[258.0]</td>\n",
       "      <td>[1251.0, 272.0, 113.0, 313.0]</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>124</th>\n",
       "      <td>924.0</td>\n",
       "      <td>[275.0, 237.0, 129.0, 1149.0, 273.0, 408.0, 27...</td>\n",
       "      <td>[603.0, 316.0, 408.0, 169.0, 113.0, 483.0]</td>\n",
       "      <td>0.142857</td>\n",
       "      <td>0.047619</td>\n",
       "      <td>0.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>125</th>\n",
       "      <td>940.0</td>\n",
       "      <td>[354.0, 316.0, 873.0, 879.0, 355.0]</td>\n",
       "      <td>[64.0, 127.0, 603.0, 114.0, 408.0, 113.0, 169....</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>126</th>\n",
       "      <td>943.0</td>\n",
       "      <td>[69.0, 233.0, 187.0, 196.0, 194.0, 184.0, 56.0...</td>\n",
       "      <td>[483.0, 318.0, 12.0, 603.0, 316.0, 114.0, 408....</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>0.116667</td>\n",
       "      <td>0.333333</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>127 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     user_id                                             actual  \\\n",
       "0        1.0  [270.0, 32.0, 242.0, 111.0, 171.0, 209.0, 256....   \n",
       "1       13.0  [900.0, 909.0, 915.0, 896.0, 691.0, 694.0, 917...   \n",
       "2       14.0  [181.0, 921.0, 213.0, 525.0, 50.0, 509.0, 172....   \n",
       "3       23.0                                            [257.0]   \n",
       "4       43.0                      [316.0, 354.0, 124.0, 1052.0]   \n",
       "..       ...                                                ...   \n",
       "122    894.0  [903.0, 316.0, 1592.0, 904.0, 752.0, 1313.0, 6...   \n",
       "123    913.0                                            [258.0]   \n",
       "124    924.0  [275.0, 237.0, 129.0, 1149.0, 273.0, 408.0, 27...   \n",
       "125    940.0                [354.0, 316.0, 873.0, 879.0, 355.0]   \n",
       "126    943.0  [69.0, 233.0, 187.0, 196.0, 194.0, 184.0, 56.0...   \n",
       "\n",
       "                                           predictions  precission      ap@k  \\\n",
       "0    [408.0, 483.0, 603.0, 285.0, 313.0, 316.0, 318...    0.000000  0.000000   \n",
       "1    [408.0, 113.0, 169.0, 114.0, 963.0, 1251.0, 11...    0.000000  0.000000   \n",
       "2    [479.0, 251.0, 113.0, 313.0, 169.0, 64.0, 483....    0.000000  0.025000   \n",
       "3    [272.0, 12.0, 169.0, 113.0, 64.0, 114.0, 313.0...    0.000000  0.000000   \n",
       "4           [316.0, 113.0, 483.0, 316.0, 603.0, 963.0]    0.250000  0.250000   \n",
       "..                                                 ...         ...       ...   \n",
       "122                         [64.0, 22.0, 316.0, 408.0]    0.100000  0.033333   \n",
       "123                      [1251.0, 272.0, 113.0, 313.0]    0.000000  0.000000   \n",
       "124         [603.0, 316.0, 408.0, 169.0, 113.0, 483.0]    0.142857  0.047619   \n",
       "125  [64.0, 127.0, 603.0, 114.0, 408.0, 113.0, 169....    0.000000  0.000000   \n",
       "126  [483.0, 318.0, 12.0, 603.0, 316.0, 114.0, 408....    0.100000  0.116667   \n",
       "\n",
       "           PR  \n",
       "0    0.000000  \n",
       "1    0.000000  \n",
       "2    0.000000  \n",
       "3    0.000000  \n",
       "4    0.833333  \n",
       "..        ...  \n",
       "122  0.750000  \n",
       "123  0.000000  \n",
       "124  0.500000  \n",
       "125  0.000000  \n",
       "126  0.333333  \n",
       "\n",
       "[127 rows x 6 columns]"
      ]
     },
     "execution_count": 413,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Svd_preds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 414,
   "metadata": {},
   "outputs": [],
   "source": [
    "MP = Svd_preds['precission'].mean()\n",
    "MAPAK = Svd_preds['ap@k'].mean()\n",
    "MPR = Svd_preds['PR'].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 315,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.054599425071866"
      ]
     },
     "execution_count": 315,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MP"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 316,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.03811190267883181"
      ]
     },
     "execution_count": 316,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MAPAK"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 415,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.22251281089863764"
      ]
     },
     "execution_count": 415,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MPR"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Метрики SVD Recommender\n",
    "- mean precission - 0,054\n",
    "- mean average precisiion at k - 0,038\n",
    "- mean percantage rank - 0,225"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "fnxHunimDAJw"
   },
   "source": [
    "### Tenzor factorization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "tDf-nWojDAJw"
   },
   "outputs": [],
   "source": [
    "import tensorly as tl\n",
    "from tensorly.decomposition import parafac\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from tensorly.metrics.regression import RMSE "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 244,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Tenzor_Recomender(object):\n",
    "    def __init__(self, user_field='user_id', item_field='movie_id', values_field='rating'):\n",
    "        self.user_field = user_field\n",
    "        self.item_field = item_field\n",
    "        self.values_field = values_field\n",
    "\n",
    "    def _preproccesing(self, movie_df):\n",
    "        self.train = movie_df\n",
    "        self.R_users_film =  movie_df.pivot(index = self.user_field , columns=self.item_field , values = self.values_field ).fillna(0)\n",
    "        genre_df = movie_df.drop_duplicates(subset=self.item_field , keep='first', inplace=False).sort_values(self.item_field , ascending = True)\n",
    "        genre_df.set_index(self.item_field, inplace=True)\n",
    "        genre_df = genre_df.iloc[:,5:24]\n",
    "        genre_df= genre_df.transpose()        \n",
    "    \n",
    "        R1 = self.R_users_film.as_matrix()\n",
    "        R2 = genre_df.as_matrix()\n",
    "        \n",
    "        slices = []\n",
    "        k = np.arange(R1.shape[0])\n",
    "        \n",
    "        for i in k:\n",
    "            NEW=R1[i]*R2\n",
    "            slices.append(NEW)\n",
    "\n",
    "        tensor = np.array(slices)\n",
    "        \n",
    "        res = tl.tensor(tensor.reshape((tensor.shape[0] , tensor.shape[1], tensor.shape[2])))\n",
    "        \n",
    "        return res \n",
    "        \n",
    "    def fit(self,movie_df, ranks):\n",
    "        \n",
    "        res = self._preproccesing(movie_df)\n",
    "        w_list = []\n",
    "        f_list = []\n",
    "        n_rank = ranks \n",
    "        \n",
    "        for i in n_rank:\n",
    "            weights, factors = parafac(res, rank=i)\n",
    "            w_list.append(weights)\n",
    "            f_list.append(factors)\n",
    "            \n",
    "        self.best_tensor = self._restore_tensor(w_list,f_list, res)    \n",
    "        \n",
    "        return  \n",
    "    \n",
    "    \n",
    "    def _restore_tensor(self,  w_list, f_list, res):          \n",
    "        restore = []\n",
    "        error_list = []\n",
    "\n",
    "        for w, f in zip(w_list,f_list):\n",
    "            full_tensor = tl.kruskal_to_tensor((w, f))\n",
    "            restore.append(full_tensor)\n",
    "        \n",
    "        for i in restore:\n",
    "            error = tl.metrics.regression.RMSE (res,i)\n",
    "            error_list.append(error)\n",
    "    \n",
    "        best_tensor = restore[error_list.index(min(error_list))]\n",
    "        \n",
    "        return best_tensor\n",
    "    \n",
    "    def predict(self,user_subset,k):          \n",
    "         \n",
    "        user_list, film_scores = self._get_matrix(k)\n",
    "        user_list_pred = [x for x in user_subset if x in user_list] \n",
    "        \n",
    "        df_tensor = pd.DataFrame({'user_id': user_list,\n",
    "                        'predictions': film_scores} ,\n",
    "                        columns = ['user_id','predictions'])\n",
    "        \n",
    "        df_tensor = df_tensor[df_tensor['user_id'].isin(user_list_pred)]\n",
    "        \n",
    "        return df_tensor\n",
    "    \n",
    "    def _get_matrix(self, k):\n",
    "        pred_ratings_df = pd.DataFrame(tl.unfold(self.best_tensor, mode=2))\n",
    "        user_list = self.R_users_film.index.tolist()\n",
    "        film_scores = []\n",
    "        \n",
    "        start = np.arange(0,len(np.arange(self.best_tensor.shape[0]*self.best_tensor.shape[1])),self.best_tensor.shape[1])\n",
    "        end = np.arange(self.best_tensor.shape[1],len(np.arange(self.best_tensor.shape[0]*self.best_tensor.shape[1]))+self.best_tensor.shape[1],self.best_tensor.shape[1])\n",
    "\n",
    "        for user,i,j in zip(user_list, start, end):\n",
    "            user_score = pred_ratings_df.iloc[:, i:j]\n",
    "            user_score['score'] = user_score.mean(axis=1)\n",
    "            rec_movies = user_score.sort_values('score',ascending=False).head(k).index.values.tolist()\n",
    "            actual = self.train[self.train[self.user_field] == user].sort_values('rating',ascending=False)[self.item_field].values.tolist()\n",
    "            cleaned_preds = [x for x in rec_movies if x not in actual]\n",
    "            film_scores.append(cleaned_preds)\n",
    "        \n",
    "        return user_list, film_scores\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 270,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_Tenzor = Tenzor_Recomender(user_field='user_id', item_field='movie_id', values_field='rating')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 271,
   "metadata": {},
   "outputs": [],
   "source": [
    "ranks = [18]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 272,
   "metadata": {},
   "outputs": [],
   "source": [
    "train, test = Split_DF(df_copy, holdout_fraction=0.25)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 273,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "D:\\anaconda\\lib\\site-packages\\ipykernel_launcher.py:15: FutureWarning: Method .as_matrix will be removed in a future version. Use .values instead.\n",
      "  from ipykernel import kernelapp as app\n",
      "D:\\anaconda\\lib\\site-packages\\ipykernel_launcher.py:16: FutureWarning: Method .as_matrix will be removed in a future version. Use .values instead.\n",
      "  app.launch_new_instance()\n"
     ]
    }
   ],
   "source": [
    "tensor_fit = model_Tenzor.fit(train, ranks)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 418,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "D:\\anaconda\\lib\\site-packages\\ipykernel_launcher.py:87: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n"
     ]
    }
   ],
   "source": [
    "preds = model_Tenzor.predict(user_subset=user_subset, k=min_k)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 419,
   "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>user_id</th>\n",
       "      <th>predictions</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.0</td>\n",
       "      <td>[171, 256, 384, 402, 652, 209, 0, 482, 567, 43...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>13.0</td>\n",
       "      <td>[171, 126, 422, 134, 0, 131, 186, 587, 81, 3, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>14.0</td>\n",
       "      <td>[99, 126, 203, 256, 55, 193, 257, 6, 172, 0, 2...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>23.0</td>\n",
       "      <td>[49, 180, 94, 587, 182, 0, 81, 422, 221, 431, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29</th>\n",
       "      <td>43.0</td>\n",
       "      <td>[180, 171, 0, 94, 422, 256, 116, 150, 99, 384,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>671</th>\n",
       "      <td>894.0</td>\n",
       "      <td>[301, 99, 274, 49, 267, 128, 180, 123, 282, 6,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>686</th>\n",
       "      <td>913.0</td>\n",
       "      <td>[49, 182, 194, 94, 134, 422, 70, 126, 587, 0, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>694</th>\n",
       "      <td>924.0</td>\n",
       "      <td>[49, 180, 171, 99, 175, 194, 126, 95, 227, 21,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>705</th>\n",
       "      <td>940.0</td>\n",
       "      <td>[49, 180, 175, 99, 182, 203, 257, 6, 185, 327,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>707</th>\n",
       "      <td>943.0</td>\n",
       "      <td>[49, 180, 126, 256, 116, 171, 99, 120, 221, 40...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>128 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     user_id                                        predictions\n",
       "0        1.0  [171, 256, 384, 402, 652, 209, 0, 482, 567, 43...\n",
       "7       13.0  [171, 126, 422, 134, 0, 131, 186, 587, 81, 3, ...\n",
       "8       14.0  [99, 126, 203, 256, 55, 193, 257, 6, 172, 0, 2...\n",
       "17      23.0  [49, 180, 94, 587, 182, 0, 81, 422, 221, 431, ...\n",
       "29      43.0  [180, 171, 0, 94, 422, 256, 116, 150, 99, 384,...\n",
       "..       ...                                                ...\n",
       "671    894.0  [301, 99, 274, 49, 267, 128, 180, 123, 282, 6,...\n",
       "686    913.0  [49, 182, 194, 94, 134, 422, 70, 126, 587, 0, ...\n",
       "694    924.0  [49, 180, 171, 99, 175, 194, 126, 95, 227, 21,...\n",
       "705    940.0  [49, 180, 175, 99, 182, 203, 257, 6, 185, 327,...\n",
       "707    943.0  [49, 180, 126, 256, 116, 171, 99, 120, 221, 40...\n",
       "\n",
       "[128 rows x 2 columns]"
      ]
     },
     "execution_count": 419,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 420,
   "metadata": {},
   "outputs": [],
   "source": [
    "Tenzor_preds = calculate_metrics(preds, test, k=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 421,
   "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>user_id</th>\n",
       "      <th>actual</th>\n",
       "      <th>predictions</th>\n",
       "      <th>precission</th>\n",
       "      <th>ap@k</th>\n",
       "      <th>PR</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.0</td>\n",
       "      <td>[270.0, 32.0, 242.0, 111.0, 171.0, 209.0, 256....</td>\n",
       "      <td>[171, 256, 384, 402, 652, 209, 0, 482, 567, 43...</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.250000</td>\n",
       "      <td>0.9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>13.0</td>\n",
       "      <td>[900.0, 909.0, 915.0, 896.0, 691.0, 694.0, 917...</td>\n",
       "      <td>[171, 126, 422, 134, 0, 131, 186, 587, 81, 3, ...</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.014286</td>\n",
       "      <td>0.7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>14.0</td>\n",
       "      <td>[181.0, 921.0, 213.0, 525.0, 50.0, 509.0, 172....</td>\n",
       "      <td>[99, 126, 203, 256, 55, 193, 257, 6, 172, 0, 2...</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.011111</td>\n",
       "      <td>0.9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>23.0</td>\n",
       "      <td>[257.0]</td>\n",
       "      <td>[49, 180, 94, 587, 182, 0, 81, 422, 221, 431, ...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>43.0</td>\n",
       "      <td>[316.0, 354.0, 124.0, 1052.0]</td>\n",
       "      <td>[180, 171, 0, 94, 422, 256, 116, 150, 99, 384,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</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",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>123</th>\n",
       "      <td>894.0</td>\n",
       "      <td>[903.0, 316.0, 1592.0, 904.0, 752.0, 1313.0, 6...</td>\n",
       "      <td>[301, 99, 274, 49, 267, 128, 180, 123, 282, 6,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>124</th>\n",
       "      <td>913.0</td>\n",
       "      <td>[258.0]</td>\n",
       "      <td>[49, 182, 194, 94, 134, 422, 70, 126, 587, 0, ...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>125</th>\n",
       "      <td>924.0</td>\n",
       "      <td>[275.0, 237.0, 129.0, 1149.0, 273.0, 408.0, 27...</td>\n",
       "      <td>[49, 180, 171, 99, 175, 194, 126, 95, 227, 21,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>126</th>\n",
       "      <td>940.0</td>\n",
       "      <td>[354.0, 316.0, 873.0, 879.0, 355.0]</td>\n",
       "      <td>[49, 180, 175, 99, 182, 203, 257, 6, 185, 327,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>127</th>\n",
       "      <td>943.0</td>\n",
       "      <td>[69.0, 233.0, 187.0, 196.0, 194.0, 184.0, 56.0...</td>\n",
       "      <td>[49, 180, 126, 256, 116, 171, 99, 120, 221, 40...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>128 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     user_id                                             actual  \\\n",
       "0        1.0  [270.0, 32.0, 242.0, 111.0, 171.0, 209.0, 256....   \n",
       "1       13.0  [900.0, 909.0, 915.0, 896.0, 691.0, 694.0, 917...   \n",
       "2       14.0  [181.0, 921.0, 213.0, 525.0, 50.0, 509.0, 172....   \n",
       "3       23.0                                            [257.0]   \n",
       "4       43.0                      [316.0, 354.0, 124.0, 1052.0]   \n",
       "..       ...                                                ...   \n",
       "123    894.0  [903.0, 316.0, 1592.0, 904.0, 752.0, 1313.0, 6...   \n",
       "124    913.0                                            [258.0]   \n",
       "125    924.0  [275.0, 237.0, 129.0, 1149.0, 273.0, 408.0, 27...   \n",
       "126    940.0                [354.0, 316.0, 873.0, 879.0, 355.0]   \n",
       "127    943.0  [69.0, 233.0, 187.0, 196.0, 194.0, 184.0, 56.0...   \n",
       "\n",
       "                                           predictions  precission      ap@k  \\\n",
       "0    [171, 256, 384, 402, 652, 209, 0, 482, 567, 43...         0.3  0.250000   \n",
       "1    [171, 126, 422, 134, 0, 131, 186, 587, 81, 3, ...         0.1  0.014286   \n",
       "2    [99, 126, 203, 256, 55, 193, 257, 6, 172, 0, 2...         0.1  0.011111   \n",
       "3    [49, 180, 94, 587, 182, 0, 81, 422, 221, 431, ...         0.0  0.000000   \n",
       "4    [180, 171, 0, 94, 422, 256, 116, 150, 99, 384,...         0.0  0.000000   \n",
       "..                                                 ...         ...       ...   \n",
       "123  [301, 99, 274, 49, 267, 128, 180, 123, 282, 6,...         0.0  0.000000   \n",
       "124  [49, 182, 194, 94, 134, 422, 70, 126, 587, 0, ...         0.0  0.000000   \n",
       "125  [49, 180, 171, 99, 175, 194, 126, 95, 227, 21,...         0.0  0.000000   \n",
       "126  [49, 180, 175, 99, 182, 203, 257, 6, 185, 327,...         0.0  0.000000   \n",
       "127  [49, 180, 126, 256, 116, 171, 99, 120, 221, 40...         0.0  0.000000   \n",
       "\n",
       "      PR  \n",
       "0    0.9  \n",
       "1    0.7  \n",
       "2    0.9  \n",
       "3    0.0  \n",
       "4    0.0  \n",
       "..   ...  \n",
       "123  0.0  \n",
       "124  0.0  \n",
       "125  0.0  \n",
       "126  0.0  \n",
       "127  0.0  \n",
       "\n",
       "[128 rows x 6 columns]"
      ]
     },
     "execution_count": 421,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Tenzor_preds "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 422,
   "metadata": {},
   "outputs": [],
   "source": [
    "MP = Tenzor_preds['precission'].mean()\n",
    "MAPAK = Tenzor_preds['ap@k'].mean()\n",
    "MPR = Tenzor_preds['PR'].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 423,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.02514880952380953"
      ]
     },
     "execution_count": 423,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MP"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 424,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.02476500496031745"
      ]
     },
     "execution_count": 424,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MAPAK"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 425,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.12499999999999999"
      ]
     },
     "execution_count": 425,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MPR"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Метрики Tenzor Recommender\n",
    "- mean precission - 0,025\n",
    "- mean average precisiion at k - 0,025\n",
    "- mean percantage rank - 0,124"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Выводы\n",
    "- Было реализовано три типа рекомендательной системы:Most Popular Recommender,Matrix Recommender,Tensor Recommender\n",
    "- Лучшие результаты по всем тремя метрикам показал Most Popular Recommender.\n",
    "- Использование рекомендательных систем на основе тензорной факторизации и матричной факторизации может быть обосновано в том случае если требуется получить представление о предполагаемых вкусах пользователя (user and item vector). Используя данные представления можно использовать либо user-based model либо item-based model подходы при построении рекомендаций\n",
    "- Использование тензорной факторизации позволяет получить дополнительнные признаковые пространства,  которые могут лучше характеризовать нашего пользователя .\n",
    "- Выбор типа системы зависит от решаемых задач и не может быть обусловлен только показателями метрик."
   ]
  }
 ],
 "metadata": {
  "colab": {
   "collapsed_sections": [],
   "name": "Movie Lens 100K.ipynb",
   "provenance": []
  },
  "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": "3.7.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}