Efficiency_530.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "611de0d7",
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys, os\n",
    "from tqdm import tqdm\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.model_selection import train_test_split\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "import torch\n",
    "import torch.nn.functional as F\n",
    "import torch.nn as nn\n",
    "import math\n",
    "\n",
    "os.environ[\"TOKENIZERS_PARALLELISM\"] = \"false\"\n",
    "\n",
    "def seed_everything(seed):\n",
    "    os.environ['PYTHONHASHSEED'] = str(seed)\n",
    "    np.random.seed(seed)\n",
    "    torch.manual_seed(seed)\n",
    "    torch.cuda.manual_seed(seed)\n",
    "    torch.backends.cudnn.deterministic = True\n",
    "    torch.backends.cudnn.benchmark = False\n",
    "    \n",
    "seed_everything(0)\n",
    "MAX_LEN = 64"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "9b5061ee",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoConfig, AutoTokenizer, AutoModelForMaskedLM, AutoModel\n",
    "\n",
    "\n",
    "class VecModel(nn.Module):\n",
    "    def __init__(self, model_name, size, has_top=True):\n",
    "        super(VecModel, self).__init__()\n",
    "        conf = AutoConfig.from_pretrained(model_name)\n",
    "        self.backbone = AutoModel.from_config(conf)\n",
    "        \n",
    "        self.has_top = has_top\n",
    "        \n",
    "        if self.has_top:\n",
    "            self.bn = nn.BatchNorm1d(size)\n",
    "            self.top = nn.Linear(size, size)\n",
    "\n",
    "    def forward(self, ids, mask):\n",
    "        out = self.backbone(ids, mask)[0]\n",
    "        out = (out[:, 1:MAX_LEN//2, :]*mask[:, 1:MAX_LEN//2, None]).mean(axis=1)\n",
    "        \n",
    "        if self.has_top:\n",
    "            out = self.top(self.bn(out))\n",
    "        \n",
    "        return F.normalize(out)\n",
    "\n",
    "    def save(self, path):\n",
    "        torch.save(self.state_dict(), path)\n",
    "\n",
    "    def load(self, path):\n",
    "        self.load_state_dict(torch.load(path, map_location='cpu'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "9d65f903",
   "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>topic_id</th>\n",
       "      <th>content_ids</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>t_00004da3a1b2</td>\n",
       "      <td>c_1108dd0c7a5d</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>t_00004da3a1b2</td>\n",
       "      <td>c_376c5a8eb028</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>t_00004da3a1b2</td>\n",
       "      <td>c_5bc0e1e2cba0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>t_00004da3a1b2</td>\n",
       "      <td>c_76231f9d0b5e</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>t_00068291e9a4</td>\n",
       "      <td>c_639ea2ef9c95</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>279914</th>\n",
       "      <td>t_fff9e5407d13</td>\n",
       "      <td>c_d64037a72376</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>279915</th>\n",
       "      <td>t_fffbe1d5d43c</td>\n",
       "      <td>c_46f852a49c08</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>279916</th>\n",
       "      <td>t_fffbe1d5d43c</td>\n",
       "      <td>c_6659207b25d5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>279917</th>\n",
       "      <td>t_fffe14f1be1e</td>\n",
       "      <td>c_cece166bad6a</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>279918</th>\n",
       "      <td>t_fffe811a6da9</td>\n",
       "      <td>c_92b8fad372ee</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>279919 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              topic_id     content_ids\n",
       "0       t_00004da3a1b2  c_1108dd0c7a5d\n",
       "1       t_00004da3a1b2  c_376c5a8eb028\n",
       "2       t_00004da3a1b2  c_5bc0e1e2cba0\n",
       "3       t_00004da3a1b2  c_76231f9d0b5e\n",
       "4       t_00068291e9a4  c_639ea2ef9c95\n",
       "...                ...             ...\n",
       "279914  t_fff9e5407d13  c_d64037a72376\n",
       "279915  t_fffbe1d5d43c  c_46f852a49c08\n",
       "279916  t_fffbe1d5d43c  c_6659207b25d5\n",
       "279917  t_fffe14f1be1e  c_cece166bad6a\n",
       "279918  t_fffe811a6da9  c_92b8fad372ee\n",
       "\n",
       "[279919 rows x 2 columns]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "corr_df = pd.read_csv(\"data/correlations.csv\")\n",
    "\n",
    "corr_df[\"content_ids\"] = corr_df[\"content_ids\"].apply(lambda x: x.split())\n",
    "\n",
    "corr_df = corr_df.explode('content_ids').reset_index(drop=True)\n",
    "corr_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "70e52483",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(61517, 9)\n",
      "(61517, 10)\n",
      "(61517, 12)\n",
      "(61517, 14)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>title</th>\n",
       "      <th>description</th>\n",
       "      <th>channel</th>\n",
       "      <th>category</th>\n",
       "      <th>level</th>\n",
       "      <th>language</th>\n",
       "      <th>parent</th>\n",
       "      <th>has_content</th>\n",
       "      <th>parent_title</th>\n",
       "      <th>grandpa</th>\n",
       "      <th>grandpa_title</th>\n",
       "      <th>ggrandpa</th>\n",
       "      <th>ggrandpa_title</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>t_00004da3a1b2</td>\n",
       "      <td>Откриването на резисторите</td>\n",
       "      <td>Изследване на материали, които предизвикват на...</td>\n",
       "      <td>000cf7</td>\n",
       "      <td>source</td>\n",
       "      <td>4</td>\n",
       "      <td>bg</td>\n",
       "      <td>t_16e29365b50d</td>\n",
       "      <td>True</td>\n",
       "      <td>Открития и проекти</td>\n",
       "      <td>t_c85886762db7</td>\n",
       "      <td>Физика</td>\n",
       "      <td>t_27a0a3c7ee9a</td>\n",
       "      <td>Наука</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>t_00068291e9a4</td>\n",
       "      <td>Entradas e saídas de uma função</td>\n",
       "      <td>Entenda um pouco mais sobre funções.</td>\n",
       "      <td>8e286a</td>\n",
       "      <td>source</td>\n",
       "      <td>4</td>\n",
       "      <td>pt</td>\n",
       "      <td>t_d14b6c2a2b70</td>\n",
       "      <td>True</td>\n",
       "      <td>Álgebra: funções</td>\n",
       "      <td>t_461c9a8dd863</td>\n",
       "      <td>9º Ano</td>\n",
       "      <td>t_a6420bcff569</td>\n",
       "      <td>Matemática por ano (Alinhada à BNCC)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>t_00069b63a70a</td>\n",
       "      <td>Transcripts</td>\n",
       "      <td>NaN</td>\n",
       "      <td>6e3ba4</td>\n",
       "      <td>source</td>\n",
       "      <td>3</td>\n",
       "      <td>en</td>\n",
       "      <td>t_4054df11a74e</td>\n",
       "      <td>True</td>\n",
       "      <td>Flow Charts: Logical Thinking?</td>\n",
       "      <td>t_acbbd893e6af</td>\n",
       "      <td>Engineering</td>\n",
       "      <td>t_9c5f09334d83</td>\n",
       "      <td>MIT Blossoms</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>t_0006d41a73a8</td>\n",
       "      <td>Графики на експоненциални функции (Алгебра 2 н...</td>\n",
       "      <td>Научи повече за графиките на сложните показате...</td>\n",
       "      <td>000cf7</td>\n",
       "      <td>source</td>\n",
       "      <td>4</td>\n",
       "      <td>bg</td>\n",
       "      <td>t_e2452e21d252</td>\n",
       "      <td>True</td>\n",
       "      <td>Показателни и логаритмични функции</td>\n",
       "      <td>t_5f4cc8e02423</td>\n",
       "      <td>Алгебра (цялото съдържание)</td>\n",
       "      <td>t_afd8db5e4f44</td>\n",
       "      <td>Математика</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>t_0008768bdee6</td>\n",
       "      <td>100 સુધીનો સરવાળો</td>\n",
       "      <td>37 અને 49 જેવી બે-અંકની સંખ્યાઓ ઉમેરતા શીખો.</td>\n",
       "      <td>5223e0</td>\n",
       "      <td>supplemental</td>\n",
       "      <td>4</td>\n",
       "      <td>gu</td>\n",
       "      <td>t_0da7a331d666</td>\n",
       "      <td>True</td>\n",
       "      <td>100 સુધીના સરવાળા અને બાદબાકી</td>\n",
       "      <td>t_d20fc01fdf38</td>\n",
       "      <td>પ્રારંભિક ગણિત</td>\n",
       "      <td>t_38bca5075c51</td>\n",
       "      <td>ધોરણ ૩</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               id                                              title  \\\n",
       "0  t_00004da3a1b2                         Откриването на резисторите   \n",
       "1  t_00068291e9a4                    Entradas e saídas de uma função   \n",
       "2  t_00069b63a70a                                        Transcripts   \n",
       "3  t_0006d41a73a8  Графики на експоненциални функции (Алгебра 2 н...   \n",
       "4  t_0008768bdee6                                  100 સુધીનો સરવાળો   \n",
       "\n",
       "                                         description channel      category  \\\n",
       "0  Изследване на материали, които предизвикват на...  000cf7        source   \n",
       "1               Entenda um pouco mais sobre funções.  8e286a        source   \n",
       "2                                                NaN  6e3ba4        source   \n",
       "3  Научи повече за графиките на сложните показате...  000cf7        source   \n",
       "4       37 અને 49 જેવી બે-અંકની સંખ્યાઓ ઉમેરતા શીખો.  5223e0  supplemental   \n",
       "\n",
       "   level language          parent  has_content  \\\n",
       "0      4       bg  t_16e29365b50d         True   \n",
       "1      4       pt  t_d14b6c2a2b70         True   \n",
       "2      3       en  t_4054df11a74e         True   \n",
       "3      4       bg  t_e2452e21d252         True   \n",
       "4      4       gu  t_0da7a331d666         True   \n",
       "\n",
       "                         parent_title         grandpa  \\\n",
       "0                  Открития и проекти  t_c85886762db7   \n",
       "1                    Álgebra: funções  t_461c9a8dd863   \n",
       "2      Flow Charts: Logical Thinking?  t_acbbd893e6af   \n",
       "3  Показателни и логаритмични функции  t_5f4cc8e02423   \n",
       "4       100 સુધીના સરવાળા અને બાદબાકી  t_d20fc01fdf38   \n",
       "\n",
       "                 grandpa_title        ggrandpa  \\\n",
       "0                       Физика  t_27a0a3c7ee9a   \n",
       "1                       9º Ano  t_a6420bcff569   \n",
       "2                  Engineering  t_9c5f09334d83   \n",
       "3  Алгебра (цялото съдържание)  t_afd8db5e4f44   \n",
       "4               પ્રારંભિક ગણિત  t_38bca5075c51   \n",
       "\n",
       "                         ggrandpa_title  \n",
       "0                                 Наука  \n",
       "1  Matemática por ano (Alinhada à BNCC)  \n",
       "2                          MIT Blossoms  \n",
       "3                            Математика  \n",
       "4                                ધોરણ ૩  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "topic_df = pd.read_csv(\"data/topics.csv\")\n",
    "topic_df[\"title\"] = topic_df[\"title\"].fillna(\"\")\n",
    "\n",
    "title_map = topic_df.set_index(\"id\")[\"title\"].to_dict()\n",
    "parent_map = topic_df.set_index(\"id\")[\"parent\"].to_dict()\n",
    "\n",
    "topic_df = topic_df[topic_df[\"has_content\"]].reset_index(drop=True)\n",
    "print(topic_df.shape)\n",
    "\n",
    "topic_df[\"parent_title\"] = topic_df[\"parent\"].apply(lambda x: title_map.get(x, \"\"))\n",
    "print(topic_df.shape)\n",
    "\n",
    "topic_df[\"grandpa\"] = topic_df[\"parent\"].apply(lambda x: parent_map.get(x))\n",
    "topic_df[\"grandpa_title\"] = topic_df[\"grandpa\"].apply(lambda x: title_map.get(x, \"\"))\n",
    "print(topic_df.shape)\n",
    "\n",
    "topic_df[\"ggrandpa\"] = topic_df[\"grandpa\"].apply(lambda x: parent_map.get(x))\n",
    "topic_df[\"ggrandpa_title\"] = topic_df[\"ggrandpa\"].apply(lambda x: title_map.get(x, \"\"))\n",
    "print(topic_df.shape)\n",
    "\n",
    "topic_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "7d560d0a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    20506\n",
       "1    20506\n",
       "2    20505\n",
       "Name: fold, dtype: int64"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.model_selection import KFold\n",
    "\n",
    "\n",
    "N_FOLDS = 3\n",
    "\n",
    "splits = KFold(N_FOLDS, shuffle=True, random_state=0).split(topic_df, topic_df[\"has_content\"])\n",
    "\n",
    "topic_df[\"fold\"] = 0\n",
    "\n",
    "for f, (train_ind, val_ind) in enumerate(splits):\n",
    "    topic_df.loc[val_ind, \"fold\"] = f\n",
    "\n",
    "topic_df[\"fold\"].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "5f2a048f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1    53269\n",
       "0     8248\n",
       "Name: fold, dtype: int64"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "topic_df.loc[(topic_df[\"category\"] == \"source\") | (topic_df[\"fold\"] > 0), \"fold\"] = 1\n",
    "topic_df[\"fold\"].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "175a86b8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "41024"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "le_topic = LabelEncoder()\n",
    "topic_df[\"t\"] = topic_df[\"title\"].fillna(\"\") + \" | \" + topic_df[\"description\"].fillna(\"\")\n",
    "topic_df[\"label\"] = le_topic.fit_transform(topic_df[\"t\"])\n",
    "\n",
    "len(le_topic.classes_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "1fd28ef2",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(154047, 9)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>title</th>\n",
       "      <th>description</th>\n",
       "      <th>kind</th>\n",
       "      <th>text</th>\n",
       "      <th>language</th>\n",
       "      <th>copyright_holder</th>\n",
       "      <th>license</th>\n",
       "      <th>t</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>c_00002381196d</td>\n",
       "      <td>Sumar números de varios dígitos: 48,029+233,930</td>\n",
       "      <td>Suma 48,029+233,930 mediante el algoritmo está...</td>\n",
       "      <td>video</td>\n",
       "      <td>NaN</td>\n",
       "      <td>es</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Sumar números de varios dígitos: 48,029+233,93...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>c_000087304a9e</td>\n",
       "      <td>Trovare i fattori di un numero</td>\n",
       "      <td>Sal trova i fattori di 120.\\n\\n</td>\n",
       "      <td>video</td>\n",
       "      <td>NaN</td>\n",
       "      <td>it</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Trovare i fattori di un numero | video | Sal t...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>c_0000ad142ddb</td>\n",
       "      <td>Sumar curvas de demanda</td>\n",
       "      <td>Cómo añadir curvas de demanda\\n\\n</td>\n",
       "      <td>video</td>\n",
       "      <td>NaN</td>\n",
       "      <td>es</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Sumar curvas de demanda | video | Cómo añadir ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>c_0000c03adc8d</td>\n",
       "      <td>Nado de aproximação</td>\n",
       "      <td>Neste vídeo você vai aprender o nado de aproxi...</td>\n",
       "      <td>document</td>\n",
       "      <td>\\nNado de aproximação\\nSaber nadar nas ondas ...</td>\n",
       "      <td>pt</td>\n",
       "      <td>Sikana Education</td>\n",
       "      <td>CC BY-NC-ND</td>\n",
       "      <td>Nado de aproximação | document | Neste vídeo v...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>c_00016694ea2a</td>\n",
       "      <td>geometry-m3-topic-a-overview.pdf</td>\n",
       "      <td>None</td>\n",
       "      <td>document</td>\n",
       "      <td>Estándares Comunes del Estado de Nueva York\\n\\...</td>\n",
       "      <td>es</td>\n",
       "      <td>Engage NY</td>\n",
       "      <td>CC BY-NC-SA</td>\n",
       "      <td>geometry-m3-topic-a-overview.pdf | document | ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               id                                             title  \\\n",
       "0  c_00002381196d  Sumar números de varios dígitos: 48,029+233,930    \n",
       "1  c_000087304a9e                    Trovare i fattori di un numero   \n",
       "2  c_0000ad142ddb                           Sumar curvas de demanda   \n",
       "3  c_0000c03adc8d                               Nado de aproximação   \n",
       "4  c_00016694ea2a                  geometry-m3-topic-a-overview.pdf   \n",
       "\n",
       "                                         description      kind  \\\n",
       "0  Suma 48,029+233,930 mediante el algoritmo está...     video   \n",
       "1                    Sal trova i fattori di 120.\\n\\n     video   \n",
       "2                  Cómo añadir curvas de demanda\\n\\n     video   \n",
       "3  Neste vídeo você vai aprender o nado de aproxi...  document   \n",
       "4                                               None  document   \n",
       "\n",
       "                                                text language  \\\n",
       "0                                                NaN       es   \n",
       "1                                                NaN       it   \n",
       "2                                                NaN       es   \n",
       "3  \\nNado de aproximação\\nSaber nadar nas ondas ...       pt   \n",
       "4  Estándares Comunes del Estado de Nueva York\\n\\...       es   \n",
       "\n",
       "   copyright_holder      license  \\\n",
       "0               NaN          NaN   \n",
       "1               NaN          NaN   \n",
       "2               NaN          NaN   \n",
       "3  Sikana Education  CC BY-NC-ND   \n",
       "4         Engage NY  CC BY-NC-SA   \n",
       "\n",
       "                                                   t  \n",
       "0  Sumar números de varios dígitos: 48,029+233,93...  \n",
       "1  Trovare i fattori di un numero | video | Sal t...  \n",
       "2  Sumar curvas de demanda | video | Cómo añadir ...  \n",
       "3  Nado de aproximação | document | Neste vídeo v...  \n",
       "4  geometry-m3-topic-a-overview.pdf | document | ...  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "content_df = pd.read_csv(\"data/content.csv\")\n",
    "content_df.loc[content_df[\"text\"] == content_df[\"description\"], \"text\"] = None\n",
    "content_df.loc[content_df[\"title\"] == content_df[\"description\"], \"description\"] = None\n",
    "\n",
    "content_df[\"t\"] = (content_df[\"title\"].fillna(\"\") + \" | \" + content_df[\"kind\"].fillna(\"\")\n",
    "                   + \" | \" + content_df[\"description\"].fillna(\"\").apply(lambda x: x[:256]) \n",
    "                   + \" | \" + content_df[\"text\"].fillna(\"\").apply(lambda x: x[:128]))\n",
    "\n",
    "print(content_df.shape)\n",
    "content_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "6dddd282",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.9733569582391859, 0.9165319675164073)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def extract_number(x, is_subtopic=False):\n",
    "    chapter, rest = x.split(\":\", 1)\n",
    "    \n",
    "    if is_subtopic:\n",
    "        chapter = chapter.split(\".\", 1)[0]\n",
    "        x = f\"{chapter}:{rest}\"\n",
    "    chapter = int(chapter)\n",
    "\n",
    "    return chapter, x\n",
    "\n",
    "\n",
    "def extract_chapters(df, is_subtopic):\n",
    "    titles = df[\"title\"].fillna(\"\").values\n",
    "    \n",
    "    chapters = np.zeros(len(titles))\n",
    "    new_titles = np.array(titles)\n",
    "    \n",
    "    for i, title in enumerate(titles):\n",
    "        try:\n",
    "            chapters[i], new_titles[i] = extract_number(title, is_subtopic)\n",
    "        except:\n",
    "            pass\n",
    "        \n",
    "    df[\"chapter\"], df[\"title\"] = chapters, new_titles\n",
    "    df[\"chapter\"] = df[\"chapter\"].astype(int)\n",
    "    \n",
    "    df.loc[df[\"chapter\"] == 0, \"chapter\"] = None\n",
    "    return df\n",
    "    \n",
    "topic_df = extract_chapters(topic_df, False)\n",
    "content_df = extract_chapters(content_df, True)\n",
    "\n",
    "topic_df[\"chapter\"].isnull().mean(), content_df[\"chapter\"].isnull().mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "08a09a37",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((53269, 18), (8248, 18))"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "TEST_FOLD = 0\n",
    "\n",
    "topic_df[\"t\"] = topic_df[\"title\"] + \" @ \" + topic_df[\"parent_title\"] + \" @ \" + topic_df[\"grandpa_title\"] + \" @ \" + topic_df[\"ggrandpa_title\"] + \" | \" + topic_df[\"description\"].fillna(\"\")\n",
    "\n",
    "topic_df_test = topic_df[topic_df[\"fold\"] == TEST_FOLD].reset_index(drop=True)\n",
    "topic_df_train = topic_df[topic_df[\"fold\"] != TEST_FOLD].reset_index(drop=True)\n",
    "\n",
    "topic_df_train.shape, topic_df_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "3fe9791a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(141107, 2)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_contents = corr_df.merge(topic_df_train[[\"id\"]], left_on=\"topic_id\", right_on=\"id\")[[\"content_ids\"]].drop_duplicates()\n",
    "train_contents.rename(columns={\"content_ids\": \"id\"}, inplace=True)\n",
    "train_contents[\"is_train\"] = True\n",
    "train_contents.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "e14aa955",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9159996624406836"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "content_df = content_df.merge(train_contents, on=\"id\", how=\"left\")\n",
    "content_df[\"is_train\"].fillna(False, inplace=True)\n",
    "content_df[\"is_train\"].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "87570f15",
   "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>id</th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>c_00002381196d</td>\n",
       "      <td>[27817, 27750]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>c_000087304a9e</td>\n",
       "      <td>[13283, 13284]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>c_0000ad142ddb</td>\n",
       "      <td>[11785]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>c_0000c03adc8d</td>\n",
       "      <td>[23664]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>c_00016694ea2a</td>\n",
       "      <td>[21301, 32510]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154042</th>\n",
       "      <td>c_fffcbdd4de8b</td>\n",
       "      <td>[2312]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154043</th>\n",
       "      <td>c_fffe15a2d069</td>\n",
       "      <td>[27466, 27467, 16294]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154044</th>\n",
       "      <td>c_fffed7b0d13a</td>\n",
       "      <td>[22032, 22033, 22034]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154045</th>\n",
       "      <td>c_ffff04ba7ac7</td>\n",
       "      <td>[3455]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154046</th>\n",
       "      <td>c_ffffe5254266</td>\n",
       "      <td>[10100]</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>154047 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                    id                  label\n",
       "0       c_00002381196d         [27817, 27750]\n",
       "1       c_000087304a9e         [13283, 13284]\n",
       "2       c_0000ad142ddb                [11785]\n",
       "3       c_0000c03adc8d                [23664]\n",
       "4       c_00016694ea2a         [21301, 32510]\n",
       "...                ...                    ...\n",
       "154042  c_fffcbdd4de8b                 [2312]\n",
       "154043  c_fffe15a2d069  [27466, 27467, 16294]\n",
       "154044  c_fffed7b0d13a  [22032, 22033, 22034]\n",
       "154045  c_ffff04ba7ac7                 [3455]\n",
       "154046  c_ffffe5254266                [10100]\n",
       "\n",
       "[154047 rows x 2 columns]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#corr_df[\"label\"] = le_topic.transform(corr_df[\"topic_id\"])\n",
    "corr_df = corr_df.merge(topic_df[[\"id\", \"label\"]].rename(columns={\"id\": \"topic_id\"}), on=\"topic_id\")\n",
    "\n",
    "\n",
    "content_labels = corr_df.groupby(\"content_ids\").agg({\"label\": list}).reset_index()\n",
    "content_labels = content_labels.rename(columns={\"content_ids\": \"id\"})\n",
    "content_labels[\"label\"] = content_labels[\"label\"].apply(lambda x: list(set(x)))\n",
    "\n",
    "topic_df_train[\"label\"] = topic_df_train[\"label\"].apply(lambda x: [x])\n",
    "topic_df_test[\"label\"] = topic_df_test[\"label\"].apply(lambda x: [x])\n",
    "\n",
    "content_labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "52c928a4",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(154047, 11)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>title</th>\n",
       "      <th>description</th>\n",
       "      <th>kind</th>\n",
       "      <th>text</th>\n",
       "      <th>language</th>\n",
       "      <th>copyright_holder</th>\n",
       "      <th>license</th>\n",
       "      <th>t</th>\n",
       "      <th>chapter</th>\n",
       "      <th>is_train</th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>c_00002381196d</td>\n",
       "      <td>Sumar números de varios dígitos: 48,029+233,930</td>\n",
       "      <td>Suma 48,029+233,930 mediante el algoritmo está...</td>\n",
       "      <td>video</td>\n",
       "      <td>NaN</td>\n",
       "      <td>es</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Sumar números de varios dígitos: 48,029+233,93...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>True</td>\n",
       "      <td>[27817, 27750]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>c_000087304a9e</td>\n",
       "      <td>Trovare i fattori di un numero</td>\n",
       "      <td>Sal trova i fattori di 120.\\n\\n</td>\n",
       "      <td>video</td>\n",
       "      <td>NaN</td>\n",
       "      <td>it</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Trovare i fattori di un numero | video | Sal t...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>True</td>\n",
       "      <td>[13283, 13284]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>c_0000ad142ddb</td>\n",
       "      <td>Sumar curvas de demanda</td>\n",
       "      <td>Cómo añadir curvas de demanda\\n\\n</td>\n",
       "      <td>video</td>\n",
       "      <td>NaN</td>\n",
       "      <td>es</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Sumar curvas de demanda | video | Cómo añadir ...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>True</td>\n",
       "      <td>[11785]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>c_0000c03adc8d</td>\n",
       "      <td>Nado de aproximação</td>\n",
       "      <td>Neste vídeo você vai aprender o nado de aproxi...</td>\n",
       "      <td>document</td>\n",
       "      <td>\\nNado de aproximação\\nSaber nadar nas ondas ...</td>\n",
       "      <td>pt</td>\n",
       "      <td>Sikana Education</td>\n",
       "      <td>CC BY-NC-ND</td>\n",
       "      <td>Nado de aproximação | document | Neste vídeo v...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>True</td>\n",
       "      <td>[23664]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>c_00016694ea2a</td>\n",
       "      <td>geometry-m3-topic-a-overview.pdf</td>\n",
       "      <td>None</td>\n",
       "      <td>document</td>\n",
       "      <td>Estándares Comunes del Estado de Nueva York\\n\\...</td>\n",
       "      <td>es</td>\n",
       "      <td>Engage NY</td>\n",
       "      <td>CC BY-NC-SA</td>\n",
       "      <td>geometry-m3-topic-a-overview.pdf | document | ...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>True</td>\n",
       "      <td>[21301, 32510]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154042</th>\n",
       "      <td>c_fffcbdd4de8b</td>\n",
       "      <td>2: Diffusion</td>\n",
       "      <td>NaN</td>\n",
       "      <td>html5</td>\n",
       "      <td>What will eventually happen to these dyes?\\n\\n...</td>\n",
       "      <td>en</td>\n",
       "      <td>CSU and Merlot</td>\n",
       "      <td>CC BY-NC-SA</td>\n",
       "      <td>2. 12: Diffusion | html5 |  | What will eventu...</td>\n",
       "      <td>2.0</td>\n",
       "      <td>True</td>\n",
       "      <td>[2312]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154043</th>\n",
       "      <td>c_fffe15a2d069</td>\n",
       "      <td>Sommare facendo gruppi da 10</td>\n",
       "      <td>Sal somma 5+68 spezzando il 5 in un 2 e un 3.\\n\\n</td>\n",
       "      <td>video</td>\n",
       "      <td>NaN</td>\n",
       "      <td>it</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Sommare facendo gruppi da 10 | video | Sal som...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>True</td>\n",
       "      <td>[27466, 27467, 16294]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154044</th>\n",
       "      <td>c_fffed7b0d13a</td>\n",
       "      <td>Introdução à subtração</td>\n",
       "      <td>Sal fala sobre o que significa subtrair. Os ex...</td>\n",
       "      <td>video</td>\n",
       "      <td>NaN</td>\n",
       "      <td>pt</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Introdução à subtração | video | Sal fala sobr...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>True</td>\n",
       "      <td>[22032, 22033, 22034]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154045</th>\n",
       "      <td>c_ffff04ba7ac7</td>\n",
       "      <td>SA of a Cone</td>\n",
       "      <td>NaN</td>\n",
       "      <td>video</td>\n",
       "      <td>NaN</td>\n",
       "      <td>en</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>SA of a Cone | video |  |</td>\n",
       "      <td>NaN</td>\n",
       "      <td>True</td>\n",
       "      <td>[3455]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154046</th>\n",
       "      <td>c_ffffe5254266</td>\n",
       "      <td>The Jats</td>\n",
       "      <td>NaN</td>\n",
       "      <td>video</td>\n",
       "      <td>NaN</td>\n",
       "      <td>en</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>The Jats | video |  |</td>\n",
       "      <td>NaN</td>\n",
       "      <td>True</td>\n",
       "      <td>[10100]</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>154047 rows × 12 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                    id                                             title  \\\n",
       "0       c_00002381196d  Sumar números de varios dígitos: 48,029+233,930    \n",
       "1       c_000087304a9e                    Trovare i fattori di un numero   \n",
       "2       c_0000ad142ddb                           Sumar curvas de demanda   \n",
       "3       c_0000c03adc8d                               Nado de aproximação   \n",
       "4       c_00016694ea2a                  geometry-m3-topic-a-overview.pdf   \n",
       "...                ...                                               ...   \n",
       "154042  c_fffcbdd4de8b                                      2: Diffusion   \n",
       "154043  c_fffe15a2d069                      Sommare facendo gruppi da 10   \n",
       "154044  c_fffed7b0d13a                            Introdução à subtração   \n",
       "154045  c_ffff04ba7ac7                                      SA of a Cone   \n",
       "154046  c_ffffe5254266                                          The Jats   \n",
       "\n",
       "                                              description      kind  \\\n",
       "0       Suma 48,029+233,930 mediante el algoritmo está...     video   \n",
       "1                         Sal trova i fattori di 120.\\n\\n     video   \n",
       "2                       Cómo añadir curvas de demanda\\n\\n     video   \n",
       "3       Neste vídeo você vai aprender o nado de aproxi...  document   \n",
       "4                                                    None  document   \n",
       "...                                                   ...       ...   \n",
       "154042                                                NaN     html5   \n",
       "154043  Sal somma 5+68 spezzando il 5 in un 2 e un 3.\\n\\n     video   \n",
       "154044  Sal fala sobre o que significa subtrair. Os ex...     video   \n",
       "154045                                                NaN     video   \n",
       "154046                                                NaN     video   \n",
       "\n",
       "                                                     text language  \\\n",
       "0                                                     NaN       es   \n",
       "1                                                     NaN       it   \n",
       "2                                                     NaN       es   \n",
       "3       \\nNado de aproximação\\nSaber nadar nas ondas ...       pt   \n",
       "4       Estándares Comunes del Estado de Nueva York\\n\\...       es   \n",
       "...                                                   ...      ...   \n",
       "154042  What will eventually happen to these dyes?\\n\\n...       en   \n",
       "154043                                                NaN       it   \n",
       "154044                                                NaN       pt   \n",
       "154045                                                NaN       en   \n",
       "154046                                                NaN       en   \n",
       "\n",
       "        copyright_holder      license  \\\n",
       "0                    NaN          NaN   \n",
       "1                    NaN          NaN   \n",
       "2                    NaN          NaN   \n",
       "3       Sikana Education  CC BY-NC-ND   \n",
       "4              Engage NY  CC BY-NC-SA   \n",
       "...                  ...          ...   \n",
       "154042    CSU and Merlot  CC BY-NC-SA   \n",
       "154043               NaN          NaN   \n",
       "154044               NaN          NaN   \n",
       "154045               NaN          NaN   \n",
       "154046               NaN          NaN   \n",
       "\n",
       "                                                        t  chapter  is_train  \\\n",
       "0       Sumar números de varios dígitos: 48,029+233,93...      NaN      True   \n",
       "1       Trovare i fattori di un numero | video | Sal t...      NaN      True   \n",
       "2       Sumar curvas de demanda | video | Cómo añadir ...      NaN      True   \n",
       "3       Nado de aproximação | document | Neste vídeo v...      NaN      True   \n",
       "4       geometry-m3-topic-a-overview.pdf | document | ...      NaN      True   \n",
       "...                                                   ...      ...       ...   \n",
       "154042  2. 12: Diffusion | html5 |  | What will eventu...      2.0      True   \n",
       "154043  Sommare facendo gruppi da 10 | video | Sal som...      NaN      True   \n",
       "154044  Introdução à subtração | video | Sal fala sobr...      NaN      True   \n",
       "154045                         SA of a Cone | video |  |       NaN      True   \n",
       "154046                             The Jats | video |  |       NaN      True   \n",
       "\n",
       "                        label  \n",
       "0              [27817, 27750]  \n",
       "1              [13283, 13284]  \n",
       "2                     [11785]  \n",
       "3                     [23664]  \n",
       "4              [21301, 32510]  \n",
       "...                       ...  \n",
       "154042                 [2312]  \n",
       "154043  [27466, 27467, 16294]  \n",
       "154044  [22032, 22033, 22034]  \n",
       "154045                 [3455]  \n",
       "154046                [10100]  \n",
       "\n",
       "[154047 rows x 12 columns]"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(content_df.shape)\n",
    "\n",
    "content_df = content_df.merge(content_labels, on=\"id\") #, how=\"left\")\n",
    "content_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "0e0c701b",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.neighbors import NearestNeighbors\n",
    "\n",
    "\n",
    "def to_vec(models, ds):\n",
    "    val_loader = DataLoader(ds, batch_size=BS, shuffle=False, num_workers=NW,\n",
    "                             pin_memory=False, drop_last=False)\n",
    "\n",
    "    tbar = tqdm(val_loader, file=sys.stdout)\n",
    "\n",
    "    vectors = []\n",
    "\n",
    "    with torch.no_grad():\n",
    "        for idx, data in enumerate(tbar):\n",
    "            ids, mask = data[\"input_ids\"].cuda(), data[\"attention_mask\"].cuda()\n",
    "\n",
    "            vec = torch.cat([model_w[i]*models[i](ids, mask) for i in range(len(models))], axis=1)\n",
    "            vectors.append(vec.detach().cpu().numpy())\n",
    "\n",
    "    V = np.concatenate(vectors)\n",
    "    \n",
    "    return V\n",
    "\n",
    "\n",
    "\n",
    "def get_matches(V_topic, V_content, topic_ids, content_ids, n_neighbors=5):\n",
    "    \n",
    "    neighbors_model = NearestNeighbors(n_neighbors=n_neighbors, metric='cosine', n_jobs=-1)\n",
    "    neighbors_model.fit(V_content)\n",
    "    dists, indices = neighbors_model.kneighbors(V_topic)\n",
    "    \n",
    "    res_df = pd.DataFrame({\"topic_id\": np.repeat(topic_ids, n_neighbors),\n",
    "                           'content_id': content_ids[indices.ravel()],\n",
    "                           'vec_dist': dists.ravel()\n",
    "                          })\n",
    "    return res_df\n",
    "\n",
    "\n",
    "def f2(recall, precision):\n",
    "    return 5*precision*recall/(4*precision + recall)\n",
    "\n",
    "def metric(pred_df):\n",
    "    scores = np.zeros(pred_df.shape[0])\n",
    "\n",
    "    for i, (gt, pred) in enumerate(zip(pred_df[\"content_ids\"].values, pred_df[\"pred\"].values)):\n",
    "        hits = 0\n",
    "\n",
    "        gt, pred = gt.split(), pred.split()\n",
    "\n",
    "        for l in gt:\n",
    "            if l in pred:\n",
    "                hits += 1\n",
    "\n",
    "        if hits > 0:           \n",
    "            recall = hits / len(gt)\n",
    "            precision = hits / len(pred)\n",
    "\n",
    "            scores[i] = f2(recall, precision)\n",
    "\n",
    "    return scores.mean()\n",
    "    \n",
    "\n",
    "def evaluate(topic_df, content_df):    \n",
    "    topic_ds = LECRDataset(topic_df, tokenizer, MAX_LEN, aug=False)\n",
    "    content_ds = LECRDataset(content_df, tokenizer, MAX_LEN, aug=False)\n",
    "\n",
    "\n",
    "    V_topic = to_vec(model, topic_ds)\n",
    "    V_content = to_vec(model, content_ds)\n",
    "\n",
    "    res_df = get_matches(V_topic, V_content, topic_df[\"id\"].values, content_df[\"id\"].values)\n",
    "\n",
    "    res_df[\"rank\"] = res_df.groupby(\"topic_id\")[\"vec_dist\"].rank(method=\"first\", ascending=True)\n",
    "\n",
    "    pred_df = res_df.copy() #res_df[(res_df[\"rank\"] == 1) | (res_df[\"vec_dist\"] < 0.1)]\n",
    "    pred_df = pred_df.groupby(\"topic_id\")[\"content_id\"].apply(lambda x: \" \".join(list(x)))\n",
    "    pred_df = pred_df.reset_index().rename(columns={\"content_id\": 'pred'})\n",
    "    pred_df = pred_df.merge(pd.read_csv(\"data/correlations.csv\"), on=\"topic_id\")\n",
    "\n",
    "    return metric(pred_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "b869bb41",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sw\n",
      "22924 2082 1447\n",
      "----\n",
      "as\n",
      "7460 126 641\n",
      "----\n",
      "fil\n",
      "4293 224 516\n",
      "----\n",
      "pt\n",
      "39484 3425 10435\n",
      "----\n",
      "fr\n",
      "34852 3034 10682\n",
      "----\n",
      "bn\n",
      "21456 1731 2513\n",
      "----\n",
      "hi\n",
      "34736 1373 4042\n",
      "----\n",
      "gu\n",
      "34629 1809 3677\n",
      "----\n",
      "es\n",
      "71615 11769 30844\n",
      "----\n",
      "en\n",
      "121492 28053 65939\n",
      "----\n",
      "sw\n",
      "1100 2082 1447\n",
      "----\n",
      "as\n",
      "401 126 641\n",
      "----\n",
      "fil\n",
      "719 224 516\n",
      "----\n",
      "pt\n",
      "4172 3425 10435\n",
      "----\n",
      "fr\n",
      "4332 3034 10682\n",
      "----\n",
      "bn\n",
      "535 1731 2513\n",
      "----\n",
      "hi\n",
      "851 1373 4042\n",
      "----\n",
      "gu\n",
      "898 1809 3677\n",
      "----\n",
      "es\n",
      "7154 11769 30844\n",
      "----\n",
      "en\n",
      "12763 28053 65939\n",
      "----\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sparse_dot_topn import awesome_cossim_topn\n",
    "\n",
    "\n",
    "def get_matches(topic_df, content_df, title_only):\n",
    "    if title_only:\n",
    "        tfidf = TfidfVectorizer(ngram_range=(1, 1), min_df=2)\n",
    "        tfidf.fit(pd.concat([content_df[\"title\"], topic_df[\"title\"]]))\n",
    "    else:\n",
    "        tfidf = TfidfVectorizer(analyzer=\"char\", ngram_range=(4, 4), min_df=2)\n",
    "        tfidf.fit(pd.concat([content_df[\"t\"], topic_df[\"t\"]]))        \n",
    "        \n",
    "    print(len(tfidf.vocabulary_), topic_df.shape[0], content_df.shape[0])\n",
    "    \n",
    "    train_df = topic_df[topic_df[\"fold\"] != TEST_FOLD].reset_index(drop=True)\n",
    "    \n",
    "    topic_df = topic_df[topic_df[\"fold\"] == TEST_FOLD].reset_index(drop=True)\n",
    "    \n",
    "    if title_only:\n",
    "        V_topic = tfidf.transform(topic_df[\"title\"])\n",
    "        V_content = tfidf.transform(content_df[\"title\"])\n",
    "        V_train = tfidf.transform(train_df[\"title\"])\n",
    "    else:\n",
    "        V_topic = tfidf.transform(topic_df[\"title\"] + \" \" + topic_df[\"t\"])\n",
    "        V_content = tfidf.transform(content_df[\"title\"] + \" \" + content_df[\"t\"])\n",
    "        V_train = tfidf.transform(train_df[\"title\"] + \" \" + train_df[\"t\"])\n",
    "\n",
    "    d = awesome_cossim_topn(V_topic, V_content.T, 20, 0.01, use_threads=True, n_jobs=4)\n",
    "\n",
    "    non_zeros = d.nonzero()\n",
    "\n",
    "    name_indices = non_zeros[0]\n",
    "    gt_indices = non_zeros[1]\n",
    "\n",
    "    content_ids = content_df[\"id\"].values\n",
    "    topic_ids = topic_df[\"id\"].values\n",
    "\n",
    "    left_side = np.empty(gt_indices.size, dtype=object)\n",
    "    right_side = np.empty(gt_indices.size, dtype=object)\n",
    "    match_score = np.zeros(gt_indices.size)\n",
    "\n",
    "    for index in range(gt_indices.size):\n",
    "        left_side[index] = topic_ids[name_indices[index]]\n",
    "        right_side[index] = content_ids[gt_indices[index]]\n",
    "        match_score[index] = d.data[index]\n",
    "\n",
    "    score_col = \"match_score\"\n",
    "    if title_only:\n",
    "        score_col = \"match_score2\"\n",
    "    \n",
    "    res_df = pd.DataFrame({\"topic_id\": left_side,\n",
    "                           'content_id': right_side,\n",
    "                           score_col: match_score\n",
    "                          })\n",
    "    \n",
    "    d = awesome_cossim_topn(V_topic, V_train.T, 1, 0.0, use_threads=True, n_jobs=4)\n",
    "\n",
    "    non_zeros = d.nonzero()\n",
    "\n",
    "    name_indices = non_zeros[0]\n",
    "    gt_indices = non_zeros[1]\n",
    "\n",
    "    train_ids = train_df[\"id\"].values\n",
    "\n",
    "    left_side = np.empty(gt_indices.size, dtype=object)\n",
    "    right_side = np.empty(gt_indices.size, dtype=object)\n",
    "    match_score = np.zeros(gt_indices.size)\n",
    "\n",
    "    for index in range(gt_indices.size):\n",
    "        left_side[index] = topic_ids[name_indices[index]]\n",
    "        right_side[index] = train_ids[gt_indices[index]]\n",
    "        match_score[index] = d.data[index]\n",
    "\n",
    "    topic_ref_df = pd.DataFrame({\"id\": left_side,\n",
    "                           'train_id': right_side,\n",
    "                           score_col: match_score\n",
    "                          })\n",
    "\n",
    "\n",
    "    return res_df, topic_ref_df\n",
    "\n",
    "def get_dfs(title_only):\n",
    "    res_dfs, topic_ref_dfs = [], []\n",
    "\n",
    "    for lang in topic_df[topic_df[\"fold\"] == TEST_FOLD][\"language\"].unique()[::-1]:\n",
    "        print(lang)\n",
    "        content_df_lang = content_df[(content_df[\"language\"] == lang)].reset_index(drop=True)\n",
    "        topic_df_lang = topic_df[(topic_df[\"language\"] == lang)].reset_index(drop=True)\n",
    "\n",
    "        if content_df_lang.shape[0] > 0:\n",
    "            res_df, topic_ref_df = get_matches(topic_df_lang, content_df_lang, title_only)\n",
    "            res_dfs.append(res_df)\n",
    "            topic_ref_dfs.append(topic_ref_df)\n",
    "\n",
    "        print(\"----\")\n",
    "\n",
    "    res_df = pd.concat(res_dfs)\n",
    "    topic_ref_df = pd.concat(topic_ref_dfs)\n",
    "    \n",
    "    return res_df, topic_ref_df\n",
    "\n",
    "res_df, topic_ref_df = get_dfs(False)\n",
    "res_df2, topic_ref_df2 = get_dfs(True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "e9724ac2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(33201, 3)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "second_degree_match_df = topic_ref_df.merge(corr_df, left_on=\"train_id\", right_on=\"topic_id\")[[\"id\", \"content_ids\"]]\n",
    "second_degree_match_df[\"second_degree\"] = True\n",
    "second_degree_match_df.rename(columns={\"id\": \"topic_id\", \"content_ids\": \"content_id\"}, inplace=True)\n",
    "second_degree_match_df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "d93a8fac",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(36246, 3)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "second_degree_match_df2 = topic_ref_df2.merge(corr_df, left_on=\"train_id\", right_on=\"topic_id\")[[\"id\", \"content_ids\"]]\n",
    "second_degree_match_df2[\"second_degree2\"] = True\n",
    "second_degree_match_df2.rename(columns={\"id\": \"topic_id\", \"content_ids\": \"content_id\"}, inplace=True)\n",
    "second_degree_match_df2.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "bcdacf37",
   "metadata": {},
   "outputs": [],
   "source": [
    "corr_df[\"target\"] = 1\n",
    "corr_df.rename(columns={\"content_ids\": \"content_id\"}, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "5b92a024",
   "metadata": {},
   "outputs": [],
   "source": [
    "topic_df.columns = [f\"topic_{col}\" for col in topic_df.columns]\n",
    "content_df.columns = [f\"content_{col}\" for col in content_df.columns]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "a8e4a481",
   "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>topic_id</th>\n",
       "      <th>content_id</th>\n",
       "      <th>dup_count</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>t_0016d30772f3</td>\n",
       "      <td>c_061d9f90bb06</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>t_0016d30772f3</td>\n",
       "      <td>c_242ddc729eec</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>t_0016d30772f3</td>\n",
       "      <td>c_61b851222e17</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>t_0016d30772f3</td>\n",
       "      <td>c_b4fa764f5218</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>t_0016d30772f3</td>\n",
       "      <td>c_c25053d6fafd</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>385495</th>\n",
       "      <td>t_ffa56a59eb42</td>\n",
       "      <td>c_6d9c9523ad59</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>385496</th>\n",
       "      <td>t_ffba5459a977</td>\n",
       "      <td>c_04a421dba8aa</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>385497</th>\n",
       "      <td>t_ffba5459a977</td>\n",
       "      <td>c_787a7a2e7217</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>385498</th>\n",
       "      <td>t_ffba5459a977</td>\n",
       "      <td>c_a46e0ec1377b</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>385499</th>\n",
       "      <td>t_ffba5459a977</td>\n",
       "      <td>c_a6c82a21c03a</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>385500 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              topic_id      content_id  dup_count\n",
       "0       t_0016d30772f3  c_061d9f90bb06          2\n",
       "1       t_0016d30772f3  c_242ddc729eec          2\n",
       "2       t_0016d30772f3  c_61b851222e17          3\n",
       "3       t_0016d30772f3  c_b4fa764f5218          1\n",
       "4       t_0016d30772f3  c_c25053d6fafd          3\n",
       "...                ...             ...        ...\n",
       "385495  t_ffa56a59eb42  c_6d9c9523ad59          1\n",
       "385496  t_ffba5459a977  c_04a421dba8aa          1\n",
       "385497  t_ffba5459a977  c_787a7a2e7217          1\n",
       "385498  t_ffba5459a977  c_a46e0ec1377b          1\n",
       "385499  t_ffba5459a977  c_a6c82a21c03a          1\n",
       "\n",
       "[385500 rows x 3 columns]"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dup_df = []\n",
    "\n",
    "topic_df[\"key\"] = topic_df[\"topic_title\"].fillna(\"\") + \" | \" + topic_df[\"topic_description\"].fillna(\"\")\n",
    "\n",
    "\n",
    "train_df, val_df = topic_df[topic_df[\"topic_fold\"] != TEST_FOLD].reset_index(drop=True), topic_df[topic_df[\"topic_fold\"] == TEST_FOLD].reset_index(drop=True)\n",
    "\n",
    "\n",
    "lookup = train_df.merge(corr_df, on=\"topic_id\").groupby([\"key\", \"content_id\"])[\"topic_channel\"].count().reset_index()\n",
    "lookup.rename(columns={\"topic_channel\": \"dup_count\"}, inplace=True)\n",
    "\n",
    "\n",
    "dup_df.append(val_df[[\"topic_id\", \"key\"]].merge(lookup, on=[\"key\"])[[\"topic_id\", 'content_id', \"dup_count\"]])\n",
    "    \n",
    "dup_df = pd.concat(dup_df).reset_index(drop=True)\n",
    "dup_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "e7a203de",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(667329, 7)"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "res_df = res_df.merge(res_df2, how=\"outer\", on=[\"topic_id\", \"content_id\"])\n",
    "res_df = res_df.merge(dup_df, how=\"outer\", on=[\"topic_id\", \"content_id\"])\n",
    "res_df = res_df.merge(second_degree_match_df, how=\"outer\", on=[\"topic_id\", \"content_id\"])\n",
    "res_df = res_df.merge(second_degree_match_df2, how=\"outer\", on=[\"topic_id\", \"content_id\"])\n",
    "\n",
    "res_df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "094ff50b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(739628, 8)"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dup_df = []\n",
    "\n",
    "topic_df[\"key\"] = topic_df[\"topic_title\"].fillna(\"\")\n",
    "\n",
    "\n",
    "train_df, val_df = topic_df[topic_df[\"topic_fold\"] != TEST_FOLD].reset_index(drop=True), topic_df[topic_df[\"topic_fold\"] == TEST_FOLD].reset_index(drop=True)\n",
    "\n",
    "\n",
    "lookup = train_df.merge(corr_df, on=\"topic_id\").groupby([\"key\", \"content_id\"])[\"topic_channel\"].count().reset_index()\n",
    "lookup.rename(columns={\"topic_channel\": \"tdup_count\"}, inplace=True)\n",
    "\n",
    "\n",
    "dup_df.append(val_df[[\"topic_id\", \"key\"]].merge(lookup, on=[\"key\"])[[\"topic_id\", 'content_id', \"tdup_count\"]])\n",
    "    \n",
    "dup_df = pd.concat(dup_df).reset_index(drop=True)\n",
    "\n",
    "\n",
    "res_df = res_df.merge(dup_df, how=\"outer\", on=[\"topic_id\", \"content_id\"])\n",
    "res_df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "ba8e166c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(855067, 9)"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dup_df = []\n",
    "\n",
    "topic_df[\"key\"] = topic_df[\"topic_parent\"].fillna(topic_df[\"topic_id\"])\n",
    "\n",
    "train_df, val_df = topic_df[topic_df[\"topic_fold\"] != TEST_FOLD].reset_index(drop=True), topic_df[topic_df[\"topic_fold\"] == TEST_FOLD].reset_index(drop=True)\n",
    "\n",
    "\n",
    "lookup = train_df.merge(corr_df, on=\"topic_id\").groupby([\"key\", \"content_id\"])[\"topic_channel\"].count().reset_index()\n",
    "lookup.rename(columns={\"topic_channel\": \"pdup_count\"}, inplace=True)\n",
    "\n",
    "\n",
    "dup_df.append(val_df[[\"topic_id\", \"key\"]].merge(lookup, on=[\"key\"])[[\"topic_id\", 'content_id', \"pdup_count\"]])\n",
    "\n",
    "dup_df = pd.concat(dup_df).reset_index(drop=True)\n",
    "\n",
    "\n",
    "res_df = res_df.merge(dup_df, how=\"outer\", on=[\"topic_id\", \"content_id\"])\n",
    "res_df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "e27a5aca",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6401523726834591"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def best_case_metric(pred_df):\n",
    "    scores = np.zeros(pred_df.shape[0])\n",
    "\n",
    "    for i, (gt, pred) in enumerate(zip(pred_df[\"content_ids\"].values, pred_df[\"pred\"].values)):\n",
    "        hits = 0\n",
    "\n",
    "        gt, pred = gt.split(), pred.split()\n",
    "\n",
    "        for l in gt:\n",
    "            if l in pred:\n",
    "                hits += 1\n",
    "\n",
    "        if hits > 0:           \n",
    "            recall = hits / len(gt)\n",
    "            precision = 1 #hits / len(pred)\n",
    "\n",
    "            scores[i] = f2(recall, precision)\n",
    "\n",
    "    return scores.mean()\n",
    "\n",
    "\n",
    "pred_df = res_df.groupby(\"topic_id\")[\"content_id\"].apply(lambda x: \" \".join(list(x)))\n",
    "pred_df = pred_df.reset_index().rename(columns={\"content_id\": 'pred'})\n",
    "pred_df = pred_df.merge(pd.read_csv(\"data/correlations.csv\"), on=\"topic_id\")\n",
    "\n",
    "best_case_metric(pred_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "10fbf59a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(855067, 38)"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "res_df = res_df.merge(topic_df, on=\"topic_id\", how=\"left\")\n",
    "res_df = res_df.merge(content_df, on=\"content_id\", how=\"left\")\n",
    "\n",
    "res_df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "0720ca6a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.022966621329088832, 19638.0)"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "res_df = res_df.merge(corr_df[[\"topic_id\", \"content_id\", \"target\"]], on=[\"topic_id\", \"content_id\"], how=\"left\")\n",
    "res_df[\"target\"].fillna(0, inplace=True)\n",
    "\n",
    "res_df[\"target\"].mean(), res_df[\"target\"].sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "1dce824f",
   "metadata": {},
   "outputs": [],
   "source": [
    "res_df[\"topic_language\"] = res_df[\"topic_language\"].astype(\"category\")\n",
    "res_df[\"topic_category\"] = res_df[\"topic_category\"].astype(\"category\")\n",
    "res_df[\"content_kind\"] = res_df[\"content_kind\"].astype(\"category\")\n",
    "res_df[\"topic_channel\"] = res_df[\"topic_channel\"].astype(\"category\")\n",
    "\n",
    "res_df[\"len_topic_title\"] = res_df[\"topic_title\"].fillna(\"\").apply(len)\n",
    "res_df[\"len_topic_description\"] = res_df[\"topic_description\"].fillna(\"\").apply(len)\n",
    "res_df[\"len_content_title\"] = res_df[\"content_title\"].fillna(\"\").apply(len)\n",
    "res_df[\"len_content_description\"] = res_df[\"content_description\"].fillna(\"\").apply(len)\n",
    "res_df[\"len_content_text\"] = res_df[\"content_text\"].fillna(\"\").apply(len)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "e7306e51",
   "metadata": {},
   "outputs": [],
   "source": [
    "res_df[\"match_score_max\"] = res_df.groupby(\"topic_id\")[\"match_score\"].transform(\"max\")\n",
    "res_df[\"match_score_min\"] = res_df.groupby(\"topic_id\")[\"match_score\"].transform(\"min\")\n",
    "\n",
    "res_df[\"match_score2_max\"] = res_df.groupby(\"topic_id\")[\"match_score2\"].transform(\"max\")\n",
    "res_df[\"match_score2_min\"] = res_df.groupby(\"topic_id\")[\"match_score2\"].transform(\"min\")\n",
    "\n",
    "res_df[\"dup_count\"] = res_df[\"dup_count\"].fillna(0)\n",
    "res_df[\"total_count\"] = res_df.groupby(\"topic_id\")[\"content_id\"].transform(\"count\")\n",
    "res_df[\"dup_count_mean\"] = res_df.groupby(\"topic_id\")[\"dup_count\"].transform(\"mean\")\n",
    "\n",
    "res_df[\"tdup_count\"] = res_df[\"tdup_count\"].fillna(0)\n",
    "res_df[\"tdup_count_mean\"] = res_df.groupby(\"topic_id\")[\"tdup_count\"].transform(\"mean\")\n",
    "\n",
    "res_df[\"pdup_count\"] = res_df[\"pdup_count\"].fillna(0)\n",
    "res_df[\"pdup_count_mean\"] = res_df.groupby(\"topic_id\")[\"pdup_count\"].transform(\"mean\")\n",
    "\n",
    "res_df[\"same_chapter\"] = res_df[\"topic_chapter\"] == res_df[\"content_chapter\"]\n",
    "res_df[\"starts_same\"] = res_df[\"topic_title\"].apply(lambda x: x.split(\" \", 1)[0]) == res_df[\"content_title\"].apply(lambda x: x.split(\" \", 1)[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "ef51877c",
   "metadata": {},
   "outputs": [],
   "source": [
    "res_df[\"second_degree\"].fillna(False, inplace=True)\n",
    "res_df[\"second_degree2\"].fillna(False, inplace=True)\n",
    "\n",
    "res_df[\"topic_max_train_score\"] = res_df[\"topic_id\"].map(topic_ref_df.set_index(\"id\")[\"match_score\"].to_dict())\n",
    "\n",
    "\n",
    "#res_df[\"content_max_train_score\"] = res_df[\"content_id\"].map(content_ref_df.set_index(\"content_id\")[\"match_score\"].to_dict())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "99c7da5d",
   "metadata": {},
   "outputs": [],
   "source": [
    "#res_df.loc[~res_df[\"content_is_train\"], \"content_max_train_score\"] = None\n",
    "\n",
    "#res_df[\"second_degree\"].fillna(False, inplace=True)\n",
    "#res_df[\"topic_max_train_score\"] = res_df[\"topic_id\"].map(topic_df_test.set_index(\"id\")[\"max_train_score\"].to_dict())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "b215c127",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fold 0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/engine.py:239: UserWarning: 'verbose_eval' argument is deprecated and will be removed in a future release of LightGBM. Pass 'log_evaluation()' callback via 'callbacks' argument instead.\n",
      "  _log_warning(\"'verbose_eval' argument is deprecated and will be removed in a future release of LightGBM. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Info] Number of positive: 14914, number of negative: 626386\n",
      "[LightGBM] [Warning] Auto-choosing row-wise multi-threading, the overhead of testing was 0.014355 seconds.\n",
      "You can set `force_row_wise=true` to remove the overhead.\n",
      "And if memory is not enough, you can set `force_col_wise=true`.\n",
      "[LightGBM] [Info] Total Bins 3975\n",
      "[LightGBM] [Info] Number of data points in the train set: 641300, number of used features: 29\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Info] [binary:BoostFromScore]: pavg=0.023256 -> initscore=-3.737666\n",
      "[LightGBM] [Info] Start training from score -3.737666\n",
      "[200]\tvalid_0's auc: 0.988835\n",
      "[400]\tvalid_0's auc: 0.989589\n",
      "[600]\tvalid_0's auc: 0.989793\n",
      "[800]\tvalid_0's auc: 0.989848\n",
      "\n",
      "Fold 1\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/engine.py:239: UserWarning: 'verbose_eval' argument is deprecated and will be removed in a future release of LightGBM. Pass 'log_evaluation()' callback via 'callbacks' argument instead.\n",
      "  _log_warning(\"'verbose_eval' argument is deprecated and will be removed in a future release of LightGBM. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Info] Number of positive: 14399, number of negative: 626901\n",
      "[LightGBM] [Warning] Auto-choosing row-wise multi-threading, the overhead of testing was 0.011786 seconds.\n",
      "You can set `force_row_wise=true` to remove the overhead.\n",
      "And if memory is not enough, you can set `force_col_wise=true`.\n",
      "[LightGBM] [Info] Total Bins 3978\n",
      "[LightGBM] [Info] Number of data points in the train set: 641300, number of used features: 29\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Info] [binary:BoostFromScore]: pavg=0.022453 -> initscore=-3.773630\n",
      "[LightGBM] [Info] Start training from score -3.773630\n",
      "[200]\tvalid_0's auc: 0.99161\n",
      "[400]\tvalid_0's auc: 0.991952\n",
      "[600]\tvalid_0's auc: 0.992056\n",
      "[800]\tvalid_0's auc: 0.992069\n",
      "\n",
      "Fold 2\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/engine.py:239: UserWarning: 'verbose_eval' argument is deprecated and will be removed in a future release of LightGBM. Pass 'log_evaluation()' callback via 'callbacks' argument instead.\n",
      "  _log_warning(\"'verbose_eval' argument is deprecated and will be removed in a future release of LightGBM. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Info] Number of positive: 14884, number of negative: 626416\n",
      "[LightGBM] [Warning] Auto-choosing row-wise multi-threading, the overhead of testing was 0.011117 seconds.\n",
      "You can set `force_row_wise=true` to remove the overhead.\n",
      "And if memory is not enough, you can set `force_col_wise=true`.\n",
      "[LightGBM] [Info] Total Bins 3974\n",
      "[LightGBM] [Info] Number of data points in the train set: 641300, number of used features: 29\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Info] [binary:BoostFromScore]: pavg=0.023209 -> initscore=-3.739728\n",
      "[LightGBM] [Info] Start training from score -3.739728\n",
      "[200]\tvalid_0's auc: 0.989291\n",
      "[400]\tvalid_0's auc: 0.990081\n",
      "[600]\tvalid_0's auc: 0.99035\n",
      "[800]\tvalid_0's auc: 0.990441\n",
      "\n",
      "Fold 3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/engine.py:239: UserWarning: 'verbose_eval' argument is deprecated and will be removed in a future release of LightGBM. Pass 'log_evaluation()' callback via 'callbacks' argument instead.\n",
      "  _log_warning(\"'verbose_eval' argument is deprecated and will be removed in a future release of LightGBM. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Info] Number of positive: 14717, number of negative: 626584\n",
      "[LightGBM] [Warning] Auto-choosing row-wise multi-threading, the overhead of testing was 0.014544 seconds.\n",
      "You can set `force_row_wise=true` to remove the overhead.\n",
      "And if memory is not enough, you can set `force_col_wise=true`.\n",
      "[LightGBM] [Info] Total Bins 3976\n",
      "[LightGBM] [Info] Number of data points in the train set: 641301, number of used features: 29\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/basic.py:1780: UserWarning: Overriding the parameters from Reference Dataset.\n",
      "  _log_warning('Overriding the parameters from Reference Dataset.')\n",
      "/home/aerdem/venv/lib/python3.8/site-packages/lightgbm/basic.py:1513: UserWarning: categorical_column in param dict is overridden.\n",
      "  _log_warning(f'{cat_alias} in param dict is overridden.')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Info] [binary:BoostFromScore]: pavg=0.022949 -> initscore=-3.751280\n",
      "[LightGBM] [Info] Start training from score -3.751280\n",
      "[200]\tvalid_0's auc: 0.987742\n",
      "[400]\tvalid_0's auc: 0.988881\n",
      "[600]\tvalid_0's auc: 0.989094\n",
      "[800]\tvalid_0's auc: 0.989291\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from sklearn.model_selection import GroupKFold\n",
    "from sklearn.metrics import roc_auc_score\n",
    "import lightgbm as lgb\n",
    "\n",
    "N_FOLDS = 4\n",
    "gkf = GroupKFold(N_FOLDS)\n",
    "\n",
    "\n",
    "lgb_param = {\"objective\": \"binary\",\n",
    "             \"metric\": 'auc',\n",
    "             \"boosting\": \"gbdt\",\n",
    "             \"learning_rate\": 0.05,\n",
    "             \"is_unbalance\": True,\n",
    "            \"min_data_in_leaf\": 1024,\n",
    "            \"num_leaves\": 31,\n",
    "            \"feature_fraction\": 0.8,\n",
    "             \"subsample\": 0.5,\n",
    "             \"subsample_freq\": 1,\n",
    "                    }\n",
    "\n",
    "y_oof = np.zeros(res_df.shape[0])\n",
    "\n",
    "features = [\"match_score\", \"match_score_max\", \"match_score_min\",\n",
    "            \"match_score2\", \"match_score2_max\", \"match_score2_min\",\n",
    "            \"len_content_text\",\n",
    "            \"len_topic_title\", \"len_topic_description\", \"len_content_title\", \"len_content_description\",\n",
    "             \"dup_count\", \"total_count\", \"dup_count_mean\",\n",
    "            \"tdup_count\", \"tdup_count_mean\", \"pdup_count\", \"pdup_count_mean\",\n",
    "            \"topic_language\", \"topic_category\", \"content_kind\", \"topic_level\",\n",
    "            \"same_chapter\", \"starts_same\", \"content_is_train\", \"topic_channel\",\n",
    "            \"topic_max_train_score\", \"second_degree\", \"second_degree2\"\n",
    "           ]\n",
    "\n",
    "scores = []\n",
    "\n",
    "\n",
    "for f, (train_ind, val_ind) in enumerate(gkf.split(res_df, res_df[\"target\"], groups=res_df[\"topic_id\"])):\n",
    "    print(\"Fold\", f)\n",
    "    train_df, val_df = res_df.iloc[train_ind].reset_index(drop=True), res_df.iloc[val_ind].reset_index(drop=True)\n",
    "        \n",
    "    lgb_train = lgb.Dataset(train_df[features], train_df[\"target\"])\n",
    "    lgb_val = lgb.Dataset(val_df[features], val_df[\"target\"])\n",
    "\n",
    "    model = lgb.train(lgb_param, lgb_train, num_boost_round=800, #early_stopping_rounds=20, \n",
    "                      valid_sets=[lgb_val],\n",
    "                      verbose_eval=200)\n",
    "    model.save_model(f'models/lgb_{f}.txt')\n",
    "    \n",
    "    y_oof[val_ind] += model.predict(val_df[features])\n",
    "    scores.append(roc_auc_score(val_df[\"target\"], y_oof[val_ind]))\n",
    "    print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "bf74fb81",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9904123623939056"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.mean(scores)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "ba11a622",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:title={'center':'Feature importance'}, xlabel='Feature importance', ylabel='Features'>"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "lgb.plot_importance(importance_type=\"gain\", booster=model, figsize=(8, 8))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "9a54c183",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9904265722923573"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "res_df[\"pred\"] = y_oof\n",
    "\n",
    "roc_auc_score(res_df[\"target\"], res_df[\"pred\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "92947283",
   "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>topic_id</th>\n",
       "      <th>pred</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>t_000d1fb3f2f5</td>\n",
       "      <td>c_3e22fed4bf86</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>t_0016d30772f3</td>\n",
       "      <td>c_c25053d6fafd c_ea312de91d4f c_242ddc729eec c...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>t_001bcbb22694</td>\n",
       "      <td>c_851005f1803f c_2f19ad913a85 c_ed86a139f17a c...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>t_0020bde404c1</td>\n",
       "      <td>c_f5e2c9575b89 c_b4eebc62e29d c_13f3f0e05a93 c...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>t_00260f878951</td>\n",
       "      <td>c_86f126d7f1f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8243</th>\n",
       "      <td>t_ffb0a70e5e48</td>\n",
       "      <td>c_f3b75faf5480 c_81f88e566297 c_f7c218334f49</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8244</th>\n",
       "      <td>t_ffb1d959761e</td>\n",
       "      <td>c_773a5387056e c_e91236516aa7 c_9b993553de45 c...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8245</th>\n",
       "      <td>t_ffba5459a977</td>\n",
       "      <td>c_a46e0ec1377b c_04a421dba8aa c_787a7a2e7217 c...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8246</th>\n",
       "      <td>t_ffdc013937fc</td>\n",
       "      <td>c_befb676d5f34 c_85f55872bc66 c_ecdf988fec0d</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8247</th>\n",
       "      <td>t_ffe8df837f62</td>\n",
       "      <td>c_825aa7726e99 c_c7f5b45331bf c_fe0ba172c096 c...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>8248 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "            topic_id                                               pred\n",
       "0     t_000d1fb3f2f5                                     c_3e22fed4bf86\n",
       "1     t_0016d30772f3  c_c25053d6fafd c_ea312de91d4f c_242ddc729eec c...\n",
       "2     t_001bcbb22694  c_851005f1803f c_2f19ad913a85 c_ed86a139f17a c...\n",
       "3     t_0020bde404c1  c_f5e2c9575b89 c_b4eebc62e29d c_13f3f0e05a93 c...\n",
       "4     t_00260f878951                                     c_86f126d7f1f8\n",
       "...              ...                                                ...\n",
       "8243  t_ffb0a70e5e48       c_f3b75faf5480 c_81f88e566297 c_f7c218334f49\n",
       "8244  t_ffb1d959761e  c_773a5387056e c_e91236516aa7 c_9b993553de45 c...\n",
       "8245  t_ffba5459a977  c_a46e0ec1377b c_04a421dba8aa c_787a7a2e7217 c...\n",
       "8246  t_ffdc013937fc       c_befb676d5f34 c_85f55872bc66 c_ecdf988fec0d\n",
       "8247  t_ffe8df837f62  c_825aa7726e99 c_c7f5b45331bf c_fe0ba172c096 c...\n",
       "\n",
       "[8248 rows x 2 columns]"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "res_df[\"rank\"] = res_df.groupby(\"topic_id\")[\"pred\"].rank(method=\"first\", ascending=False)\n",
    "\n",
    "res_df[\"gap\"] = res_df.groupby(\"topic_id\")[\"pred\"].transform(\"max\") - res_df[\"pred\"]\n",
    "\n",
    "res_df[\"content_dist\"] = res_df[\"gap\"] + res_df.groupby(\"content_id\")[\"pred\"].transform(\"max\") - res_df[\"pred\"]\n",
    "\n",
    "res_df[\"content_rank\"] = res_df.groupby(\"content_id\")[\"content_dist\"].rank(method=\"first\")\n",
    "\n",
    "pred_df = res_df[(res_df[\"rank\"] == 1) | \n",
    "                 ((res_df[\"gap\"] < 0.25) & res_df[\"content_is_train\"] & (res_df[\"pred\"] > 0.05)) | \n",
    "                 (~res_df[\"content_is_train\"] & (res_df[\"content_rank\"] == 1) & (res_df[\"content_dist\"] < 0.65)) | \n",
    "                 (~res_df[\"content_is_train\"] & (res_df[\"gap\"] < 0.05))]\n",
    "\n",
    "pred_df = pred_df.groupby(\"topic_id\")[\"content_id\"].apply(lambda x: \" \".join(list(x)))\n",
    "pred_df = pred_df.reset_index().rename(columns={\"content_id\": 'pred'})\n",
    "\n",
    "pred_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "057ab1ff",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(8248, 3)"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_df = pred_df.merge(pd.read_csv(\"data/correlations.csv\"), on=\"topic_id\")\n",
    "pred_df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "df6faec5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.5304016054970463"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metric(pred_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c0089d22",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "98e10585",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}