TF_행렬분해.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "e90e8a51-f455-42da-8f49-12cac998a955",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "\n",
    "import warnings \n",
    "warnings.filterwarnings('ignore')\n",
    "\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.layers import Input, Embedding, Dot, Add, Flatten\n",
    "from tensorflow.keras.regularizers import l2\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from tensorflow.keras.callbacks import EarlyStopping"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "15564a32-b867-43e7-b1fd-ea306a2502d5",
   "metadata": {},
   "source": [
    "# 1. 데이터 불러오기"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "550608ea-8165-4339-b5c6-e58ad591e63c",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 유저 데이터\n",
    "user = pd.read_csv('data/member_mate_sample.csv')\n",
    "\n",
    "# 프로필 열람 로그 데이터\n",
    "view = pd.read_pickle('data/view.pickle')\n",
    "\n",
    "# 관심 있어요 로그 데이터\n",
    "concern = pd.read_csv('data/concern.csv')\n",
    "\n",
    "# 영상 통화 로그 데이터\n",
    "call = pd.read_csv('data/call.csv')\n",
    "\n",
    "# 동물상 데이터 \n",
    "animal = pd.read_csv('data/animal.csv')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cdc7b77c-8b52-4672-98a8-6f046c69e322",
   "metadata": {},
   "source": [
    "# 2. 데이터 전처리"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e88b2d8c-7728-4538-9d0a-c828618d0213",
   "metadata": {},
   "source": [
    "## 2-1. 유저 데이터 확인"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "24244ed1-1652-47b9-a7e0-01a744d05b49",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 유저 데이터 처리\n",
    "\n",
    "df_user = user[user['mem_sex'] == 'f']\n",
    "\n",
    "df_user = df_user[['mem_no', 'mem_sex', 'mem_first_name']]\n",
    "\n",
    "le = LabelEncoder()\n",
    "df_user['user'] = le.fit_transform(df_user['mem_no'])\n",
    "\n",
    "df_user = df_user[['user','mem_no','mem_sex','mem_first_name']]\n",
    "\n",
    "df_user.sort_values('user', inplace=True)\n",
    "\n",
    "df_user.reset_index(drop=True, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "752cb6c4-069a-476d-b48d-49d6e15f4175",
   "metadata": {},
   "outputs": [],
   "source": [
    "# mem_no를 넣으면 유저 이름을 반환해 주는 함수\n",
    "\n",
    "def user_info(n_mem_no):\n",
    "    if (df_user['user'] == n_mem_no).any():\n",
    "        return df_user[df_user['user'] == n_mem_no]['mem_no'].iloc[0]\n",
    "    else:\n",
    "        return False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "257fb04a-9d02-4611-a848-be8b2a2d5568",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "31"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# user_info 함수 확인\n",
    "# user:1 -> mem_no:31\n",
    "\n",
    "user_info(1)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e3d8f488-2ac3-49f6-a6a6-7a455781e402",
   "metadata": {},
   "source": [
    "## 2-2. 평점 데이터 생성"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "25506542-8833-4273-89ce-5af8d03999e9",
   "metadata": {},
   "source": [
    "- 보유한 로그 데이터를 기반으로 평점을 계산\n",
    "- 남성 평점 : ((view*1) + (concern*2) + (call*3))\n",
    "- 여성 평점 : ((view*1) + (concern*1.5) + (call*2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "8e0ddd4b-2c7f-4d98-ab8f-6a4bb72c5c92",
   "metadata": {},
   "outputs": [],
   "source": [
    "female_view = view[(view['mem_sex'] == 'f') & (view['ptr_mem_sex'] == 'm')]\n",
    "female_view['view_weight'] = 1\n",
    "female_view = female_view[['mem_no','ptr_mem_no','view_weight']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "67ea7233-3ee0-489b-8351-adac90d3684a",
   "metadata": {},
   "outputs": [],
   "source": [
    "female_concern = concern[(concern['mem_sex'] == 'f') & (concern['ptr_mem_sex'] == 'm')]\n",
    "female_concern['concern_weight'] = 1.5\n",
    "female_concern = female_concern[['mem_no','ptr_mem_no','concern_weight']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "4b9072f0-21b5-44c4-9380-c32c9931fd7e",
   "metadata": {},
   "outputs": [],
   "source": [
    "female_call = call[(call['mem_sex'] == 'f') & (call['ptr_mem_sex'] == 'm')]\n",
    "female_call['call_weight'] = 2.0\n",
    "female_call = female_call[['mem_no','ptr_mem_no','call_weight']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "6308b6c4-fbe1-4baf-bb0c-34d477103ee2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 평점 데이터 생성\n",
    "\n",
    "merge1 = pd.merge(female_call, female_concern, how='right', on=['mem_no','ptr_mem_no'])\n",
    "\n",
    "merge2 = pd.merge(merge1, female_view, how='right', on=['mem_no','ptr_mem_no'])\n",
    "\n",
    "df_rating = merge2.copy()\n",
    "\n",
    "df_rating['view_weight'].fillna(0, inplace=True)\n",
    "df_rating['concern_weight'].fillna(0, inplace=True) \n",
    "df_rating['call_weight'].fillna(0, inplace=True)\n",
    "\n",
    "df_rating['rating'] = df_rating['view_weight'] + df_rating['concern_weight'] + df_rating['call_weight']\n",
    "df_rating = df_rating[['mem_no','ptr_mem_no','rating']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "8cad35ad-5863-4c89-9b5e-0230a6cfd73b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 강아지상인 유저만 아이템으로 활용\n",
    "puppy = animal[animal['animal'] == 'puppy']\n",
    "puppy.rename(columns={'mem_no' : 'ptr_mem_no'}, inplace=True)\n",
    "\n",
    "df_puppy = pd.merge(df_rating, puppy, how='right')\n",
    "df_puppy.dropna(inplace=True)\n",
    "df_rating = df_puppy\n",
    "\n",
    "df_rating.drop_duplicates(inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8c06e2ff-057a-4e46-a663-7302c1990015",
   "metadata": {},
   "source": [
    "# 2-3. 데이터 전처리"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "68734499-302a-4669-97e5-86e0313dcf2a",
   "metadata": {},
   "outputs": [],
   "source": [
    "le = LabelEncoder()\n",
    "df_rating['user'] = le.fit_transform(df_rating['mem_no'])\n",
    "\n",
    "le = LabelEncoder()\n",
    "df_rating['item'] = le.fit_transform(df_rating['ptr_mem_no'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "b561b667-791c-417d-8329-c6b8ad76d3f3",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "고유한 유저 수 : 6138\n",
      "유저 id 최소값 : 0\n",
      "유저 id 최대값 : 6137\n"
     ]
    }
   ],
   "source": [
    "# 고유한 유저 수\n",
    "\n",
    "print(\"고유한 유저 수 :\", df_rating['user'].nunique())\n",
    "print(\"유저 id 최소값 :\", df_rating['user'].min())\n",
    "print(\"유저 id 최대값 :\", df_rating['user'].max())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "ce61c10c-6a52-4e41-9b00-cc9df16f4c22",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "고유한 유저 수 : 2337\n",
      "유저 id 최소값 : 0\n",
      "유저 id 최대값 : 2336\n"
     ]
    }
   ],
   "source": [
    "# 고유한 아이템 수\n",
    "\n",
    "print(\"고유한 유저 수 :\", df_rating['item'].nunique())\n",
    "print(\"유저 id 최소값 :\", df_rating['item'].min())\n",
    "print(\"유저 id 최대값 :\", df_rating['item'].max())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "117a1a0b-a6de-4299-9d43-ef334d919b48",
   "metadata": {},
   "outputs": [],
   "source": [
    "# item 바꾸기 전에 기존 item을 dictionary로 저장해 둔다.\n",
    "\n",
    "n_unique_item = df_rating['item'].nunique()\n",
    "item_new2old_dict = dict([(an_index, an_id) for (an_index, an_id) in zip(range(n_unique_item), df_rating['item'].unique())])\n",
    "item_old2new_dict = dict([(an_id, an_index) for (an_index, an_id) in zip(range(n_unique_item), df_rating['item'].unique())])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "1b33272f-5324-4512-aea4-757cca56a252",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "평점 최소값 : 1.0\n",
      "평점 최대값 : 2.5\n"
     ]
    }
   ],
   "source": [
    "# 최소, 최대 평점.\n",
    "\n",
    "rating_min = df_rating['rating'].min()\n",
    "rating_max = df_rating['rating'].max()\n",
    "print(\"평점 최소값 :\", rating_min)\n",
    "print(\"평점 최대값 :\", rating_max)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4fbd68e1-e9c3-4ab9-91e3-e517e1d9f784",
   "metadata": {},
   "source": [
    "# 3. tensorflow를 활용한 행렬분해"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "36277221-2765-463d-8e02-dd39bc970a28",
   "metadata": {},
   "source": [
    "## 3-1. Simon Funk의 SVD 모델 정의."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "5e21efe0-5043-416d-ae2f-11ff57fab69f",
   "metadata": {},
   "outputs": [],
   "source": [
    "N = df_rating['user'].max() + 1    # 유저의 수.\n",
    "M = df_rating['item'].max() + 1   # 아이템의 수\n",
    "n_factors = 20                    # 은닉 차원.\n",
    "mu = df_rating['rating'].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "9050b3f3-ca4d-4dec-b8d4-ff5a0ff46329",
   "metadata": {},
   "outputs": [],
   "source": [
    "# TensorFlow 모델, Functional API 사용.\n",
    "u_in = Input(shape=(1,),name = \"Input_User\")             \n",
    "m_in = Input(shape=(1,),name = \"Input_Movie\")\n",
    "u_mat = Embedding(N, n_factors,name=\"U_mat\")(u_in)\n",
    "m_mat = Embedding(M, n_factors,name=\"M_mat\")(m_in)\n",
    "x = Dot(axes=2,name=\"Dot_Product\")([u_mat, m_mat])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "3008fa57-1d54-4d54-acbb-907ec3efebb1",
   "metadata": {},
   "outputs": [],
   "source": [
    "u_bias = Embedding(N, 1, name=\"u_bias\")(u_in) \n",
    "m_bias = Embedding(M, 1, name=\"m_bias\")(m_in)\n",
    "x = Add(name=\"Biases\")([x, u_bias, m_bias])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "6de87609-baca-47aa-af56-b14df0dd1559",
   "metadata": {},
   "outputs": [],
   "source": [
    "x = Flatten(name=\"Flatten\")(x) \n",
    "my_model = Model(inputs=[u_in, m_in], outputs=x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "a3e104cf-b3d9-4ea5-95d9-fde5c56cb646",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 하이퍼 파라미터를 정의\n",
    "n_epochs = 50\n",
    "batch_size = 256\n",
    "learn_rate = 0.0001"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "7bb06c7a-8b17-4d95-96a6-66d2b4c7e874",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Optimizer 객체 생성 후 컴파일\n",
    "\n",
    "my_optimizer=Adam(lr=learn_rate)\n",
    "my_model.compile(loss = \"mse\", optimizer = my_optimizer, metrics=[\"mse\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "310a2195-3f76-467a-ad46-529bc3d677a4",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 학습\n",
    "\n",
    "my_EarlyStop = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=2)          # 조기 종료 허락!\n",
    "history = my_model.fit([df_rating['user'].values, df_rating['item'].values], df_rating['rating'].values - mu , epochs=n_epochs, batch_size = batch_size, validation_split = 0.3, verbose = 0, callbacks=[my_EarlyStop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "d988f2bc-0570-4564-b03e-0f7d9281f20d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x1ac4d31b7c0>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 모델 평가 \n",
    "loss = history.history['loss']\n",
    "val_loss = history.history['val_loss']\n",
    "\n",
    "plt.plot(loss)\n",
    "plt.plot(val_loss, 'r')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Loss')\n",
    "plt.title('Loss')\n",
    "plt.legend([\"Training\", \"Validation\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7d5dfbb6-d913-45ea-bb61-63973a885c3d",
   "metadata": {},
   "source": [
    "## 3-2. 예측"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "8d0dcc89-9b31-48a9-9d02-573c1382507d",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 특정 user에 대한 추천\n",
    "\n",
    "i_userId = 1\n",
    "item_ids = df_rating['item'].unique()\n",
    "user_ids = np.ones(item_ids.size)*i_userId\n",
    "my_preds = my_model.predict([user_ids, item_ids])\n",
    "#my_preds = np.clip(my_preds, df_rating['rating'].min(), df_rating['rating'].max())             # 예측된 평점 값이 min~max 구간에 있도록 clipping 해준다."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "bfbd12fd-4bfe-4e3b-aa4d-9e7212d54a0e",
   "metadata": {},
   "outputs": [],
   "source": [
    "item_names = [user_info(item_new2old_dict[x]) for x in item_ids]      # 실제 넘버로 변환.\n",
    "my_predictions = [(x,y) for (x,y) in zip(list(my_preds), item_names)]     # (평점, item) 튜플의 리스트. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "250e7940-801d-4c12-86ca-f27d11f756f8",
   "metadata": {},
   "outputs": [],
   "source": [
    "# DataFrame으로 변환 후 출력 준비\n",
    "\n",
    "df_predicted_ratings = pd.DataFrame(data = my_predictions , columns = ['PredictedRating','item'])\n",
    "df_predicted_ratings.sort_values(by=\"PredictedRating\", ascending = False, inplace=True)\n",
    "df_predicted_ratings.reset_index(drop=True, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "f1deb4f4-fc5a-4cc5-9f74-c49c22f3d14a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "31"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "user_info(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "f655750c-e8ca-4c40-a1a7-1149038fdebe",
   "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>PredictedRating</th>\n",
       "      <th>item</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[0.12227274]</td>\n",
       "      <td>638637</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[0.0872856]</td>\n",
       "      <td>753746</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[0.075633086]</td>\n",
       "      <td>505983</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[0.07525457]</td>\n",
       "      <td>1073318</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[0.062179063]</td>\n",
       "      <td>624928</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>[0.058283787]</td>\n",
       "      <td>641838</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>[0.058063418]</td>\n",
       "      <td>1051743</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>[0.057933107]</td>\n",
       "      <td>653780</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>[0.057501916]</td>\n",
       "      <td>595058</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>[0.056770742]</td>\n",
       "      <td>1126791</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>[0.05630663]</td>\n",
       "      <td>536837</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>[0.05581647]</td>\n",
       "      <td>613697</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>[0.054721422]</td>\n",
       "      <td>1088403</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>[0.054083645]</td>\n",
       "      <td>1121646</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>[0.05261091]</td>\n",
       "      <td>710665</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>[0.052215323]</td>\n",
       "      <td>653001</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>[0.050081663]</td>\n",
       "      <td>69020</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>[0.047692433]</td>\n",
       "      <td>672559</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>[0.047335457]</td>\n",
       "      <td>1072432</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>[0.04685232]</td>\n",
       "      <td>1058360</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   PredictedRating     item\n",
       "0     [0.12227274]   638637\n",
       "1      [0.0872856]   753746\n",
       "2    [0.075633086]   505983\n",
       "3     [0.07525457]  1073318\n",
       "4    [0.062179063]   624928\n",
       "5    [0.058283787]   641838\n",
       "6    [0.058063418]  1051743\n",
       "7    [0.057933107]   653780\n",
       "8    [0.057501916]   595058\n",
       "9    [0.056770742]  1126791\n",
       "10    [0.05630663]   536837\n",
       "11    [0.05581647]   613697\n",
       "12   [0.054721422]  1088403\n",
       "13   [0.054083645]  1121646\n",
       "14    [0.05261091]   710665\n",
       "15   [0.052215323]   653001\n",
       "16   [0.050081663]    69020\n",
       "17   [0.047692433]   672559\n",
       "18   [0.047335457]  1072432\n",
       "19    [0.04685232]  1058360"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# mem_no 31인 유저의 추천 결과\n",
    "# PredictedRating : 예측 평점\n",
    "# item : 추천된 유저\n",
    "\n",
    "df_predicted_ratings.head(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "fcb4d9d5-0dcb-48fa-81a6-fcdccad2206f",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_predicted_ratings.to_csv('data/top20.csv', index=False)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}