src/AntGraph_solution.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "e0d7c6cc-4ec9-40b2-b942-682197811e6e",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "# import torch \n",
    "# !pip3 install lightgbm\n",
    "# !pip3 install xgboost\n",
    "# !pip3 install catboost\n",
    "# !pip3 install tqdmt\n",
    "# from sklearn.metrics import accuracy_score,precision_score,f1_score\n",
    "import seaborn as sns\n",
    "import lightgbm as lgb\n",
    "import xgboost as xgb\n",
    "import catboost as cbt\n",
    "import os\n",
    "import pandas as pd\n",
    "from tqdm import tqdm\n",
    "from sklearn.model_selection import KFold\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from datetime import timedelta\n",
    "from sklearn.metrics import f1_score, fbeta_score, precision_score, recall_score\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "import warnings\n",
    "import gc\n",
    "import wandb\n",
    "import time\n",
    "from wandb.lightgbm import wandb_callback\n",
    "from sklearn.metrics import roc_auc_score\n",
    "from collections import defaultdict\n",
    "warnings.filterwarnings('ignore')\n",
    "import matplotlib as mpl\n",
    "mpl.rcParams['font.sans-serif'] = [\"SimHei\"]\n",
    "mpl.rcParams[\"axes.unicode_minus\"] = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "id": "e0c2a6ba-9ed6-4bd8-8341-0792f5ee8300",
   "metadata": {},
   "outputs": [],
   "source": [
    "config = {\n",
    "    'dataset': 'A',  # for Dataset A or Dataset B\n",
    "    'sample_size': 100000,\n",
    "    'dst_offset': 0,\n",
    "}\n",
    "\n",
    "if config['dataset'] == 'A':\n",
    "    config['dst_offset'] = 0\n",
    "else:\n",
    "    config['dst_offset'] = 791332"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "id": "d5dfb751-9a2d-4b2b-89ad-71f17793d199",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 27045268/27045268 [03:02<00:00, 148506.62it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "timestamp state done!\n",
      "done!\n"
     ]
    }
   ],
   "source": [
    "if config['dataset'] == 'A':\n",
    "    edge_data = pd.read_csv(f'../train_csvs/edges_train_A.csv', header=None, names=['src_id', 'dst_id', 'edge_type', 'timestamp'])\n",
    "else:\n",
    "    edge_data = pd.read_csv(f'../train_csvs/edges_train_B.csv', header=None, names=['src_id', 'dst_id', 'edge_type', 'timestamp', 'edge_feat'])\n",
    "\n",
    "edge_data['dst_id'] += config['dst_offset']\n",
    "    \n",
    "# 利用训练数据，构图，并整合每一个三元组对下的 timestamp 列表\n",
    "graph_for_exist = dict()\n",
    "\n",
    "for idx, i in enumerate(tqdm(edge_data.values)):\n",
    "    src_id = i[0]\n",
    "    dst_id = i[1]\n",
    "    edge_type = i[2]\n",
    "    timestamp = i[3]\n",
    "    \n",
    "    if src_id not in graph_for_exist.keys():\n",
    "        graph_for_exist[src_id] = dict()\n",
    "    if dst_id not in graph_for_exist.keys():\n",
    "        graph_for_exist[dst_id] = dict()\n",
    "    if edge_type not in graph_for_exist[src_id].keys():\n",
    "        graph_for_exist[src_id][edge_type] = dict()\n",
    "    if edge_type not in graph_for_exist[dst_id].keys():\n",
    "        graph_for_exist[dst_id][edge_type] = dict()\n",
    "    if src_id not in graph_for_exist[dst_id][edge_type].keys():\n",
    "        graph_for_exist[dst_id][edge_type][src_id] = []\n",
    "    if dst_id not in graph_for_exist[src_id][edge_type].keys():\n",
    "        graph_for_exist[src_id][edge_type][dst_id] = []\n",
    "    \n",
    "    graph_for_exist[src_id][edge_type][dst_id].append(timestamp)\n",
    "    graph_for_exist[dst_id][edge_type][src_id].append(timestamp)\n",
    "\n",
    "print('timestamp state done!')\n",
    "\n",
    "for key in graph_for_exist.keys():\n",
    "    for k2 in graph_for_exist[key].keys():\n",
    "        for k3, value in graph_for_exist[key][k2].items():\n",
    "            graph_for_exist[key][k2][k3] = sorted(value)\n",
    "\n",
    "print('done!')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "5fb8af23-635f-4e21-91ad-c99b02f46307",
   "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>src_id</th>\n",
       "      <th>dst_id</th>\n",
       "      <th>edge_type</th>\n",
       "      <th>start_time</th>\n",
       "      <th>end_time</th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>31663</td>\n",
       "      <td>62299</td>\n",
       "      <td>71</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>26464</td>\n",
       "      <td>45285</td>\n",
       "      <td>147</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>36463</td>\n",
       "      <td>49022</td>\n",
       "      <td>151</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>21066</td>\n",
       "      <td>28905</td>\n",
       "      <td>20</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>39224</td>\n",
       "      <td>57474</td>\n",
       "      <td>65</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>399995</th>\n",
       "      <td>28905</td>\n",
       "      <td>22840</td>\n",
       "      <td>244</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>399996</th>\n",
       "      <td>28905</td>\n",
       "      <td>62568</td>\n",
       "      <td>20</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>399997</th>\n",
       "      <td>45285</td>\n",
       "      <td>41707</td>\n",
       "      <td>74</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>399998</th>\n",
       "      <td>18933</td>\n",
       "      <td>45285</td>\n",
       "      <td>93</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>399999</th>\n",
       "      <td>26569</td>\n",
       "      <td>21457</td>\n",
       "      <td>33</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>400000 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        src_id  dst_id  edge_type  start_time  end_time  label\n",
       "0        31663   62299         71           0         0      1\n",
       "1        26464   45285        147           0         0      1\n",
       "2        36463   49022        151           0         0      1\n",
       "3        21066   28905         20           0         0      1\n",
       "4        39224   57474         65           0         0      1\n",
       "...        ...     ...        ...         ...       ...    ...\n",
       "399995   28905   22840        244           0         0      0\n",
       "399996   28905   62568         20           0         0      0\n",
       "399997   45285   41707         74           0         0      0\n",
       "399998   18933   45285         93           0         0      0\n",
       "399999   26569   21457         33           0         0      0\n",
       "\n",
       "[400000 rows x 6 columns]"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 方案三采样方法构造训练集\n",
    "edge_csv = edge_data.sample(config['sample_size']).drop(columns=['timestamp', 'edge_feat'], errors='ignore')\n",
    "edge_csv['start_time'] = edge_csv['end_time'] = 0\n",
    "edge_csv['label'] = 1\n",
    "\n",
    "# 方案三：根据原始边缘信息，重新采样，以下开始负样本采样\n",
    "src_id_col = edge_csv['src_id'].values.copy()\n",
    "dst_id_col = edge_csv['dst_id'].values.copy()\n",
    "edge_type_col = edge_csv['edge_type'].values.copy()\n",
    "np.random.shuffle(dst_id_col)  # dst id 与 edge type 列重新生成\n",
    "np.random.shuffle(edge_type_col)\n",
    "negative_edge_csv = pd.DataFrame(\n",
    "    {'src_id': src_id_col,\n",
    "     'dst_id': dst_id_col,\n",
    "     'edge_type': edge_type_col,\n",
    "     'start_time': 0,\n",
    "     'end_time': 0,\n",
    "     'label': 0})\n",
    "train_a_data = pd.concat([edge_csv, negative_edge_csv]).reset_index(drop=True)\n",
    "train_a_data = pd.DataFrame(np.concatenate([train_a_data.values, train_a_data.values[:, [1, 0, 2, 3, 4, 5]], ]),\n",
    "                            # 双向边\n",
    "                            columns=['src_id', 'dst_id', 'edge_type', 'start_time', 'end_time', 'label'])\n",
    "train_a_data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "73f94e0c-6881-4877-b54a-dfac41c95498",
   "metadata": {},
   "outputs": [],
   "source": [
    "if config['dataset'] == 'A':\n",
    "    # train_a_data = pd.read_csv('../train_csvs/new_train_A_100000.csv')\n",
    "    train_a_node_features = pd.read_csv('../train_csvs/node_features.csv', header=None, names=['node_id', \"feat_0\", 'feat_1', 'feat_2', 'feat_3', 'feat_4', 'feat_5', 'feat_6', 'feat_7'])\n",
    "    train_a_edge_type_features = pd.read_csv('../train_csvs/edge_type_features.csv', header=None, names=['edge_type_id', \"feat_0\", 'feat_1', 'feat_2',])\n",
    "\n",
    "dev_a_data = pd.read_csv(f'../test_csvs/input_{config[\"dataset\"]}_initial.csv', header=None, names=['src_id', 'dst_id', 'edge_type', 'start_time', 'end_time', 'label'])\n",
    "mid_a_data = pd.read_csv(f'../test_csvs/input_{config[\"dataset\"]}.csv', header=None, names=['src_id', 'dst_id', 'edge_type', 'start_time', 'end_time'])\n",
    "test_a_data = pd.read_csv(f'../test_csvs/input_{config[\"dataset\"]}_final.csv', header=None, names=['src_id', 'dst_id', 'edge_type', 'start_time', 'end_time'])\n",
    "test_a_data['label'] = mid_a_data['label'] = -1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "id": "9d089ece-6f6b-46c5-98f1-c715e7b8d272",
   "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>src_id</th>\n",
       "      <th>dst_id</th>\n",
       "      <th>edge_type</th>\n",
       "      <th>start_time</th>\n",
       "      <th>end_time</th>\n",
       "      <th>label</th>\n",
       "      <th>test_flag</th>\n",
       "      <th>feat_0_x</th>\n",
       "      <th>feat_1_x</th>\n",
       "      <th>feat_2_x</th>\n",
       "      <th>...</th>\n",
       "      <th>feat_1_y</th>\n",
       "      <th>feat_2_y</th>\n",
       "      <th>feat_3_y</th>\n",
       "      <th>feat_4_y</th>\n",
       "      <th>feat_5_y</th>\n",
       "      <th>feat_6_y</th>\n",
       "      <th>feat_7_y</th>\n",
       "      <th>feat_0</th>\n",
       "      <th>feat_1</th>\n",
       "      <th>feat_2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>31663</td>\n",
       "      <td>62299</td>\n",
       "      <td>71</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>155</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>216</td>\n",
       "      <td>13</td>\n",
       "      <td>134</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>26464</td>\n",
       "      <td>45285</td>\n",
       "      <td>147</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>14</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>155</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>122</td>\n",
       "      <td>10</td>\n",
       "      <td>135</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>36463</td>\n",
       "      <td>49022</td>\n",
       "      <td>151</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>109</td>\n",
       "      <td>13</td>\n",
       "      <td>116</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>21066</td>\n",
       "      <td>28905</td>\n",
       "      <td>20</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>218</td>\n",
       "      <td>15</td>\n",
       "      <td>8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>39224</td>\n",
       "      <td>57474</td>\n",
       "      <td>65</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>86</td>\n",
       "      <td>10</td>\n",
       "      <td>119</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658095</th>\n",
       "      <td>69980</td>\n",
       "      <td>49214</td>\n",
       "      <td>176</td>\n",
       "      <td>1494875670</td>\n",
       "      <td>1494996140</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>32</td>\n",
       "      <td>171</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>159</td>\n",
       "      <td>18</td>\n",
       "      <td>110</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658096</th>\n",
       "      <td>69980</td>\n",
       "      <td>52874</td>\n",
       "      <td>33</td>\n",
       "      <td>1498584288</td>\n",
       "      <td>1498694102</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>155</td>\n",
       "      <td>-1</td>\n",
       "      <td>8</td>\n",
       "      <td>148</td>\n",
       "      <td>13</td>\n",
       "      <td>45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658097</th>\n",
       "      <td>69980</td>\n",
       "      <td>55746</td>\n",
       "      <td>209</td>\n",
       "      <td>1497952740</td>\n",
       "      <td>1498081488</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>16</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>209</td>\n",
       "      <td>9</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658098</th>\n",
       "      <td>69980</td>\n",
       "      <td>60642</td>\n",
       "      <td>71</td>\n",
       "      <td>1497166072</td>\n",
       "      <td>1497356374</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>216</td>\n",
       "      <td>13</td>\n",
       "      <td>134</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658099</th>\n",
       "      <td>69980</td>\n",
       "      <td>64236</td>\n",
       "      <td>69</td>\n",
       "      <td>1495174423</td>\n",
       "      <td>1495378242</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>59</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>30</td>\n",
       "      <td>6</td>\n",
       "      <td>128</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>658100 rows × 26 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        src_id  dst_id  edge_type  start_time    end_time  label  test_flag  \\\n",
       "0        31663   62299         71           0           0      1          0   \n",
       "1        26464   45285        147           0           0      1          0   \n",
       "2        36463   49022        151           0           0      1          0   \n",
       "3        21066   28905         20           0           0      1          0   \n",
       "4        39224   57474         65           0           0      1          0   \n",
       "...        ...     ...        ...         ...         ...    ...        ...   \n",
       "658095   69980   49214        176  1494875670  1494996140     -1          3   \n",
       "658096   69980   52874         33  1498584288  1498694102     -1          3   \n",
       "658097   69980   55746        209  1497952740  1498081488     -1          3   \n",
       "658098   69980   60642         71  1497166072  1497356374     -1          3   \n",
       "658099   69980   64236         69  1495174423  1495378242     -1          3   \n",
       "\n",
       "        feat_0_x  feat_1_x  feat_2_x  ...  feat_1_y  feat_2_y  feat_3_y  \\\n",
       "0             -1        -1        -1  ...        -1        -1        -1   \n",
       "1             14        -1        -1  ...        -1        -1        -1   \n",
       "2             -1        -1        -1  ...        -1        -1        -1   \n",
       "3             -1        -1        -1  ...        -1        -1        -1   \n",
       "4             -1        -1        -1  ...        -1        -1        -1   \n",
       "...          ...       ...       ...  ...       ...       ...       ...   \n",
       "658095        29        -1        -1  ...        -1        -1        -1   \n",
       "658096        29        -1        -1  ...        -1        -1        -1   \n",
       "658097        29        -1        -1  ...        -1        -1        -1   \n",
       "658098        29        -1        -1  ...        -1        -1        -1   \n",
       "658099        29        -1        -1  ...        -1        -1        -1   \n",
       "\n",
       "        feat_4_y  feat_5_y  feat_6_y  feat_7_y  feat_0  feat_1  feat_2  \n",
       "0             -1       155        -1        -1     216      13     134  \n",
       "1             -1       155        -1        -1     122      10     135  \n",
       "2             -1        -1        -1        -1     109      13     116  \n",
       "3             -1        -1        -1        -1     218      15       8  \n",
       "4             -1        -1        -1        -1      86      10     119  \n",
       "...          ...       ...       ...       ...     ...     ...     ...  \n",
       "658095        32       171        -1        -1     159      18     110  \n",
       "658096        -1       155        -1         8     148      13      45  \n",
       "658097        16        -1        -1        -1     209       9       0  \n",
       "658098        -1        -1        -1        -1     216      13     134  \n",
       "658099        -1        59        -1        -1      30       6     128  \n",
       "\n",
       "[658100 rows x 26 columns]"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_a_data['test_flag'] = 0\n",
    "dev_a_data['test_flag'] = 1\n",
    "mid_a_data['test_flag'] = 2\n",
    "test_a_data['test_flag'] = 3\n",
    "\n",
    "tmp_test_data = pd.concat([dev_a_data, mid_a_data, test_a_data])\n",
    "tmp_test_data['dst_id'] += config['dst_offset']\n",
    "\n",
    "all_data = pd.concat([train_a_data, tmp_test_data], ignore_index=True)\n",
    "\n",
    "if config['dataset'] == 'A':\n",
    "    all_data = all_data.merge(train_a_node_features, how='left', left_on='src_id', right_on='node_id')\n",
    "    all_data.drop(columns=['node_id'], axis=0, inplace=True)\n",
    "    all_data = all_data.merge(train_a_node_features, how='left', left_on='dst_id', right_on='node_id')\n",
    "    all_data.drop(columns=['node_id'], axis=0, inplace=True)\n",
    "    all_data = all_data.merge(train_a_edge_type_features, how='left', left_on='edge_type', right_on='edge_type_id')\n",
    "    all_data.drop(columns=['edge_type_id'], axis=0, inplace=True)\n",
    "\n",
    "all_data# .head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "id": "726929c0-4dbd-4fb5-92bc-7a6c75d328aa",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 19442/19442 [00:40<00:00, 476.98it/s] \n"
     ]
    }
   ],
   "source": [
    "from collections import defaultdict\n",
    "\n",
    "# 一元信息构造\n",
    "node_adj_info = dict()\n",
    "\n",
    "# 二元信息构造\n",
    "two_node_adj_info = dict()\n",
    "\n",
    "# 三元信息构造\n",
    "si = edge_data['dst_id']\n",
    "di = edge_data['src_id']\n",
    "\n",
    "edge_data_tmp = edge_data.copy()\n",
    "edge_data_tmp['src_id'] = si\n",
    "edge_data_tmp['dst_id'] = di\n",
    "edge_data_bi = pd.concat([edge_data, edge_data_tmp])\n",
    "triple_node_adj_info = edge_data_bi.groupby(['src_id', 'dst_id', 'edge_type']).count().reset_index().rename(columns={'timestamp': 'sparse:group_val_(src_id,dst_id,edge_type)_count'})\n",
    "del edge_data_tmp, edge_data_bi\n",
    "\n",
    "for idx, (node_id, value) in enumerate(tqdm(graph_for_exist.items())):\n",
    "    du = 0 # 节点的度，一共有多少条边\n",
    "    unique_adj_node = set()\n",
    "    nunique_adj_edge = len(value) # 节点一共邻接多少个不同边\n",
    "    \n",
    "    to_edge_dict = defaultdict(dict)  # key: dst_id, value: {edge_type: num, edge_type2: num2}\n",
    "    \n",
    "    for edge_type, adj_nodes in value.items():\n",
    "        # print(edge_type, adj_nodes)\n",
    "        for dst_id, tss in adj_nodes.items():\n",
    "            unique_adj_node.add(dst_id)\n",
    "            du += len(tss)\n",
    "            \n",
    "            to_edge_dict[dst_id][edge_type] = len(tss)\n",
    "        \n",
    "    node_adj_info[node_id] = {\n",
    "        'sparse:group_val_(src_id)_count': du,\n",
    "        'sparse:group_val_(src_id)_dst_id_nunique': len(unique_adj_node), # 节点一共邻接多少个不同点\n",
    "        'sparse:group_val_(src_id)_edge_type_nunique': nunique_adj_edge,\n",
    "    }\n",
    "    for dst_id, info in to_edge_dict.items():\n",
    "        two_node_adj_info[(node_id, dst_id)] = {\n",
    "            'sparse:group_val_(src_id,dst_id)_count': sum(info.values()),\n",
    "            'sparse:group_val_(src_id,dst_id)_edge_type_nunique': len(info),\n",
    "        }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "id": "0da45284-4b42-4d00-8163-c7bfaa7811a3",
   "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>level_0</th>\n",
       "      <th>level_1</th>\n",
       "      <th>sparse:group_val_(src_id,dst_id)_count</th>\n",
       "      <th>sparse:group_val_(src_id,dst_id)_edge_type_nunique</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>106</td>\n",
       "      <td>28905</td>\n",
       "      <td>91306</td>\n",
       "      <td>188</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>106</td>\n",
       "      <td>39520</td>\n",
       "      <td>6671</td>\n",
       "      <td>121</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>106</td>\n",
       "      <td>45915</td>\n",
       "      <td>6505</td>\n",
       "      <td>129</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>106</td>\n",
       "      <td>61213</td>\n",
       "      <td>3013</td>\n",
       "      <td>109</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>106</td>\n",
       "      <td>27194</td>\n",
       "      <td>186</td>\n",
       "      <td>37</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2925753</th>\n",
       "      <td>20584</td>\n",
       "      <td>50050</td>\n",
       "      <td>4</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2925754</th>\n",
       "      <td>20584</td>\n",
       "      <td>64560</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2925755</th>\n",
       "      <td>20584</td>\n",
       "      <td>45915</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2925756</th>\n",
       "      <td>20584</td>\n",
       "      <td>28905</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2925757</th>\n",
       "      <td>5186</td>\n",
       "      <td>28905</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2925758 rows × 4 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "         level_0  level_1  sparse:group_val_(src_id,dst_id)_count  \\\n",
       "0            106    28905                                   91306   \n",
       "1            106    39520                                    6671   \n",
       "2            106    45915                                    6505   \n",
       "3            106    61213                                    3013   \n",
       "4            106    27194                                     186   \n",
       "...          ...      ...                                     ...   \n",
       "2925753    20584    50050                                       4   \n",
       "2925754    20584    64560                                       2   \n",
       "2925755    20584    45915                                       1   \n",
       "2925756    20584    28905                                       1   \n",
       "2925757     5186    28905                                       2   \n",
       "\n",
       "         sparse:group_val_(src_id,dst_id)_edge_type_nunique  \n",
       "0                                                      188   \n",
       "1                                                      121   \n",
       "2                                                      129   \n",
       "3                                                      109   \n",
       "4                                                       37   \n",
       "...                                                    ...   \n",
       "2925753                                                  3   \n",
       "2925754                                                  2   \n",
       "2925755                                                  1   \n",
       "2925756                                                  1   \n",
       "2925757                                                  1   \n",
       "\n",
       "[2925758 rows x 4 columns]"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "node_adj_info_pd = pd.DataFrame.from_dict(node_adj_info, orient='index').reset_index()\n",
    "two_node_adj_info_pd = pd.DataFrame.from_dict(two_node_adj_info, orient='index').reset_index()\n",
    "\n",
    "two_node_adj_info_pd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "id": "5900216f-3849-4fba-b861-fb9ede1074b5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<69984x69984 sparse matrix of type '<class 'numpy.int64'>'\n",
       "\twith 2925758 stored elements in Compressed Sparse Row format>"
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from scipy import sparse\n",
    "\n",
    "row = []\n",
    "col = []\n",
    "data = []\n",
    "\n",
    "for (src_id, dst_id), value in two_node_adj_info.items():\n",
    "    row.append(src_id)\n",
    "    col.append(dst_id)\n",
    "    data.append(value['sparse:group_val_(src_id,dst_id)_count'])\n",
    "one_hop_count_matrix = sparse.csr_matrix((data, (row, col)))\n",
    "one_hop_count_matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "id": "b40f04ea-19b7-48ea-9da4-141245923cac",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<69984x69984 sparse matrix of type '<class 'numpy.int64'>'\n",
       "\twith 374466676 stored elements in Compressed Sparse Row format>"
      ]
     },
     "execution_count": 68,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "two_hop_count_matrix = one_hop_count_matrix * one_hop_count_matrix\n",
    "two_hop_count_matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "id": "32ce9560-7e68-4a15-a5ec-d34cbcd46e67",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 2925758/2925758 [03:10<00:00, 15371.47it/s]\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>level_0</th>\n",
       "      <th>level_1</th>\n",
       "      <th>sparse:group_val_(src_id,dst_id)_count</th>\n",
       "      <th>sparse:group_val_(src_id,dst_id)_edge_type_nunique</th>\n",
       "      <th>two_hop_count_list</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>106</td>\n",
       "      <td>28905</td>\n",
       "      <td>91306</td>\n",
       "      <td>188</td>\n",
       "      <td>30637224999</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>106</td>\n",
       "      <td>39520</td>\n",
       "      <td>6671</td>\n",
       "      <td>121</td>\n",
       "      <td>15381684086</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>106</td>\n",
       "      <td>45915</td>\n",
       "      <td>6505</td>\n",
       "      <td>129</td>\n",
       "      <td>16506712019</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>106</td>\n",
       "      <td>61213</td>\n",
       "      <td>3013</td>\n",
       "      <td>109</td>\n",
       "      <td>4293185422</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>106</td>\n",
       "      <td>27194</td>\n",
       "      <td>186</td>\n",
       "      <td>37</td>\n",
       "      <td>517666723</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2925753</th>\n",
       "      <td>20584</td>\n",
       "      <td>50050</td>\n",
       "      <td>4</td>\n",
       "      <td>3</td>\n",
       "      <td>1815</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2925754</th>\n",
       "      <td>20584</td>\n",
       "      <td>64560</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>213462</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2925755</th>\n",
       "      <td>20584</td>\n",
       "      <td>45915</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>177922</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2925756</th>\n",
       "      <td>20584</td>\n",
       "      <td>28905</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>597560</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2925757</th>\n",
       "      <td>5186</td>\n",
       "      <td>28905</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2925758 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "         level_0  level_1  sparse:group_val_(src_id,dst_id)_count  \\\n",
       "0            106    28905                                   91306   \n",
       "1            106    39520                                    6671   \n",
       "2            106    45915                                    6505   \n",
       "3            106    61213                                    3013   \n",
       "4            106    27194                                     186   \n",
       "...          ...      ...                                     ...   \n",
       "2925753    20584    50050                                       4   \n",
       "2925754    20584    64560                                       2   \n",
       "2925755    20584    45915                                       1   \n",
       "2925756    20584    28905                                       1   \n",
       "2925757     5186    28905                                       2   \n",
       "\n",
       "         sparse:group_val_(src_id,dst_id)_edge_type_nunique  \\\n",
       "0                                                      188    \n",
       "1                                                      121    \n",
       "2                                                      129    \n",
       "3                                                      109    \n",
       "4                                                       37    \n",
       "...                                                    ...    \n",
       "2925753                                                  3    \n",
       "2925754                                                  2    \n",
       "2925755                                                  1    \n",
       "2925756                                                  1    \n",
       "2925757                                                  1    \n",
       "\n",
       "         two_hop_count_list  \n",
       "0               30637224999  \n",
       "1               15381684086  \n",
       "2               16506712019  \n",
       "3                4293185422  \n",
       "4                 517666723  \n",
       "...                     ...  \n",
       "2925753                1815  \n",
       "2925754              213462  \n",
       "2925755              177922  \n",
       "2925756              597560  \n",
       "2925757                   0  \n",
       "\n",
       "[2925758 rows x 5 columns]"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "two_hop_count_list = []\n",
    "\n",
    "for i in tqdm(two_node_adj_info_pd.values):\n",
    "    src_id = i[0]\n",
    "    dst_id = i[1]\n",
    "    two_hop_count_list.append(two_hop_count_matrix[src_id, dst_id])\n",
    "    \n",
    "two_node_adj_info_pd['two_hop_count_list'] = two_hop_count_list\n",
    "two_node_adj_info_pd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "id": "5f802110-ea8b-42d0-9ff1-4af4fccfe266",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 拼接一元信息\n",
    "all_data = all_data.merge(node_adj_info_pd, how='left', left_on='src_id', right_on='index').drop(columns=['index'], axis=0)\n",
    "# 拼接二元信息\n",
    "all_data = all_data.merge(node_adj_info_pd, how='left', left_on='dst_id', right_on='index').drop(columns=['index'], axis=0)\n",
    "# 拼接三元信息\n",
    "all_data = all_data.merge(two_node_adj_info_pd, how='left', left_on=['src_id', 'dst_id'], right_on=['level_0', 'level_1']).fillna(0).drop(columns=['level_0', 'level_1'], axis=0)\n",
    "all_data = all_data.merge(triple_node_adj_info, how='left', left_on=['src_id', 'dst_id', 'edge_type'], right_on=['src_id', 'dst_id', 'edge_type']).fillna(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "id": "5d1f7c17-0c76-47a9-bdd5-00c0177b36af",
   "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>src_id</th>\n",
       "      <th>dst_id</th>\n",
       "      <th>edge_type</th>\n",
       "      <th>start_time</th>\n",
       "      <th>end_time</th>\n",
       "      <th>label</th>\n",
       "      <th>test_flag</th>\n",
       "      <th>feat_0_x</th>\n",
       "      <th>feat_1_x</th>\n",
       "      <th>feat_2_x</th>\n",
       "      <th>...</th>\n",
       "      <th>sparse:group_val_(src_id)_count_x</th>\n",
       "      <th>sparse:group_val_(src_id)_dst_id_nunique_x</th>\n",
       "      <th>sparse:group_val_(src_id)_edge_type_nunique_x</th>\n",
       "      <th>sparse:group_val_(src_id)_count_y</th>\n",
       "      <th>sparse:group_val_(src_id)_dst_id_nunique_y</th>\n",
       "      <th>sparse:group_val_(src_id)_edge_type_nunique_y</th>\n",
       "      <th>sparse:group_val_(src_id,dst_id)_count</th>\n",
       "      <th>sparse:group_val_(src_id,dst_id)_edge_type_nunique</th>\n",
       "      <th>two_hop_count_list</th>\n",
       "      <th>sparse:group_val_(src_id,dst_id,edge_type)_count</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>31663</td>\n",
       "      <td>62299</td>\n",
       "      <td>71</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>21153</td>\n",
       "      <td>1256</td>\n",
       "      <td>175</td>\n",
       "      <td>213020</td>\n",
       "      <td>3704</td>\n",
       "      <td>221</td>\n",
       "      <td>36.0</td>\n",
       "      <td>13.0</td>\n",
       "      <td>1.299685e+08</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>26464</td>\n",
       "      <td>45285</td>\n",
       "      <td>147</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>14</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>525550</td>\n",
       "      <td>7081</td>\n",
       "      <td>219</td>\n",
       "      <td>3295131</td>\n",
       "      <td>11958</td>\n",
       "      <td>245</td>\n",
       "      <td>77684.0</td>\n",
       "      <td>184.0</td>\n",
       "      <td>1.058126e+11</td>\n",
       "      <td>358.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>36463</td>\n",
       "      <td>49022</td>\n",
       "      <td>151</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>298543</td>\n",
       "      <td>3574</td>\n",
       "      <td>217</td>\n",
       "      <td>114832</td>\n",
       "      <td>3726</td>\n",
       "      <td>196</td>\n",
       "      <td>315.0</td>\n",
       "      <td>49.0</td>\n",
       "      <td>2.353738e+09</td>\n",
       "      <td>36.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>21066</td>\n",
       "      <td>28905</td>\n",
       "      <td>20</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>65440</td>\n",
       "      <td>2697</td>\n",
       "      <td>198</td>\n",
       "      <td>9675998</td>\n",
       "      <td>19179</td>\n",
       "      <td>246</td>\n",
       "      <td>13033.0</td>\n",
       "      <td>163.0</td>\n",
       "      <td>2.213092e+09</td>\n",
       "      <td>652.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>39224</td>\n",
       "      <td>57474</td>\n",
       "      <td>65</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>11358</td>\n",
       "      <td>1265</td>\n",
       "      <td>170</td>\n",
       "      <td>45401</td>\n",
       "      <td>2642</td>\n",
       "      <td>169</td>\n",
       "      <td>18.0</td>\n",
       "      <td>5.0</td>\n",
       "      <td>2.538742e+07</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658095</th>\n",
       "      <td>69980</td>\n",
       "      <td>49214</td>\n",
       "      <td>176</td>\n",
       "      <td>1494875670</td>\n",
       "      <td>1494996140</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>3746</td>\n",
       "      <td>607</td>\n",
       "      <td>148</td>\n",
       "      <td>44551</td>\n",
       "      <td>1569</td>\n",
       "      <td>189</td>\n",
       "      <td>30.0</td>\n",
       "      <td>15.0</td>\n",
       "      <td>3.652753e+06</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658096</th>\n",
       "      <td>69980</td>\n",
       "      <td>52874</td>\n",
       "      <td>33</td>\n",
       "      <td>1498584288</td>\n",
       "      <td>1498694102</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>3746</td>\n",
       "      <td>607</td>\n",
       "      <td>148</td>\n",
       "      <td>4372</td>\n",
       "      <td>491</td>\n",
       "      <td>130</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658097</th>\n",
       "      <td>69980</td>\n",
       "      <td>55746</td>\n",
       "      <td>209</td>\n",
       "      <td>1497952740</td>\n",
       "      <td>1498081488</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>3746</td>\n",
       "      <td>607</td>\n",
       "      <td>148</td>\n",
       "      <td>777</td>\n",
       "      <td>183</td>\n",
       "      <td>78</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658098</th>\n",
       "      <td>69980</td>\n",
       "      <td>60642</td>\n",
       "      <td>71</td>\n",
       "      <td>1497166072</td>\n",
       "      <td>1497356374</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>3746</td>\n",
       "      <td>607</td>\n",
       "      <td>148</td>\n",
       "      <td>728553</td>\n",
       "      <td>10035</td>\n",
       "      <td>234</td>\n",
       "      <td>93.0</td>\n",
       "      <td>24.0</td>\n",
       "      <td>5.185534e+07</td>\n",
       "      <td>10.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658099</th>\n",
       "      <td>69980</td>\n",
       "      <td>64236</td>\n",
       "      <td>69</td>\n",
       "      <td>1495174423</td>\n",
       "      <td>1495378242</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>3746</td>\n",
       "      <td>607</td>\n",
       "      <td>148</td>\n",
       "      <td>20059</td>\n",
       "      <td>1424</td>\n",
       "      <td>196</td>\n",
       "      <td>224.0</td>\n",
       "      <td>54.0</td>\n",
       "      <td>1.665079e+06</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>658100 rows × 36 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        src_id  dst_id  edge_type  start_time    end_time  label  test_flag  \\\n",
       "0        31663   62299         71           0           0      1          0   \n",
       "1        26464   45285        147           0           0      1          0   \n",
       "2        36463   49022        151           0           0      1          0   \n",
       "3        21066   28905         20           0           0      1          0   \n",
       "4        39224   57474         65           0           0      1          0   \n",
       "...        ...     ...        ...         ...         ...    ...        ...   \n",
       "658095   69980   49214        176  1494875670  1494996140     -1          3   \n",
       "658096   69980   52874         33  1498584288  1498694102     -1          3   \n",
       "658097   69980   55746        209  1497952740  1498081488     -1          3   \n",
       "658098   69980   60642         71  1497166072  1497356374     -1          3   \n",
       "658099   69980   64236         69  1495174423  1495378242     -1          3   \n",
       "\n",
       "        feat_0_x  feat_1_x  feat_2_x  ...  sparse:group_val_(src_id)_count_x  \\\n",
       "0             -1        -1        -1  ...                              21153   \n",
       "1             14        -1        -1  ...                             525550   \n",
       "2             -1        -1        -1  ...                             298543   \n",
       "3             -1        -1        -1  ...                              65440   \n",
       "4             -1        -1        -1  ...                              11358   \n",
       "...          ...       ...       ...  ...                                ...   \n",
       "658095        29        -1        -1  ...                               3746   \n",
       "658096        29        -1        -1  ...                               3746   \n",
       "658097        29        -1        -1  ...                               3746   \n",
       "658098        29        -1        -1  ...                               3746   \n",
       "658099        29        -1        -1  ...                               3746   \n",
       "\n",
       "        sparse:group_val_(src_id)_dst_id_nunique_x  \\\n",
       "0                                             1256   \n",
       "1                                             7081   \n",
       "2                                             3574   \n",
       "3                                             2697   \n",
       "4                                             1265   \n",
       "...                                            ...   \n",
       "658095                                         607   \n",
       "658096                                         607   \n",
       "658097                                         607   \n",
       "658098                                         607   \n",
       "658099                                         607   \n",
       "\n",
       "        sparse:group_val_(src_id)_edge_type_nunique_x  \\\n",
       "0                                                 175   \n",
       "1                                                 219   \n",
       "2                                                 217   \n",
       "3                                                 198   \n",
       "4                                                 170   \n",
       "...                                               ...   \n",
       "658095                                            148   \n",
       "658096                                            148   \n",
       "658097                                            148   \n",
       "658098                                            148   \n",
       "658099                                            148   \n",
       "\n",
       "        sparse:group_val_(src_id)_count_y  \\\n",
       "0                                  213020   \n",
       "1                                 3295131   \n",
       "2                                  114832   \n",
       "3                                 9675998   \n",
       "4                                   45401   \n",
       "...                                   ...   \n",
       "658095                              44551   \n",
       "658096                               4372   \n",
       "658097                                777   \n",
       "658098                             728553   \n",
       "658099                              20059   \n",
       "\n",
       "        sparse:group_val_(src_id)_dst_id_nunique_y  \\\n",
       "0                                             3704   \n",
       "1                                            11958   \n",
       "2                                             3726   \n",
       "3                                            19179   \n",
       "4                                             2642   \n",
       "...                                            ...   \n",
       "658095                                        1569   \n",
       "658096                                         491   \n",
       "658097                                         183   \n",
       "658098                                       10035   \n",
       "658099                                        1424   \n",
       "\n",
       "        sparse:group_val_(src_id)_edge_type_nunique_y  \\\n",
       "0                                                 221   \n",
       "1                                                 245   \n",
       "2                                                 196   \n",
       "3                                                 246   \n",
       "4                                                 169   \n",
       "...                                               ...   \n",
       "658095                                            189   \n",
       "658096                                            130   \n",
       "658097                                             78   \n",
       "658098                                            234   \n",
       "658099                                            196   \n",
       "\n",
       "        sparse:group_val_(src_id,dst_id)_count  \\\n",
       "0                                         36.0   \n",
       "1                                      77684.0   \n",
       "2                                        315.0   \n",
       "3                                      13033.0   \n",
       "4                                         18.0   \n",
       "...                                        ...   \n",
       "658095                                    30.0   \n",
       "658096                                     0.0   \n",
       "658097                                     0.0   \n",
       "658098                                    93.0   \n",
       "658099                                   224.0   \n",
       "\n",
       "        sparse:group_val_(src_id,dst_id)_edge_type_nunique  \\\n",
       "0                                                    13.0    \n",
       "1                                                   184.0    \n",
       "2                                                    49.0    \n",
       "3                                                   163.0    \n",
       "4                                                     5.0    \n",
       "...                                                   ...    \n",
       "658095                                               15.0    \n",
       "658096                                                0.0    \n",
       "658097                                                0.0    \n",
       "658098                                               24.0    \n",
       "658099                                               54.0    \n",
       "\n",
       "        two_hop_count_list  sparse:group_val_(src_id,dst_id,edge_type)_count  \n",
       "0             1.299685e+08                                               1.0  \n",
       "1             1.058126e+11                                             358.0  \n",
       "2             2.353738e+09                                              36.0  \n",
       "3             2.213092e+09                                             652.0  \n",
       "4             2.538742e+07                                               2.0  \n",
       "...                    ...                                               ...  \n",
       "658095        3.652753e+06                                               0.0  \n",
       "658096        0.000000e+00                                               0.0  \n",
       "658097        0.000000e+00                                               0.0  \n",
       "658098        5.185534e+07                                              10.0  \n",
       "658099        1.665079e+06                                               4.0  \n",
       "\n",
       "[658100 rows x 36 columns]"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2a95628e-9e36-4db6-b623-d7a372fcd98b",
   "metadata": {},
   "source": [
    "## 拼接外部 node embedding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "id": "ddf6a305-8413-400e-9829-207937c442b4",
   "metadata": {},
   "outputs": [],
   "source": [
    "def read_emb_data(filepath):\n",
    "    if not os.path.exists(filepath):\n",
    "        return dict()\n",
    "    f = open(filepath, 'r')\n",
    "    num_count, num_dim = map(int, f.readline().split())\n",
    "\n",
    "    node_id_emb = dict()\n",
    "    for data in f.readlines():\n",
    "        data = data.strip().split(' ')\n",
    "\n",
    "        node_id = int(data[0])\n",
    "        embed = np.array(data[1:], dtype=float)\n",
    "        node_id_emb[node_id] = embed\n",
    "    return node_id_emb\n",
    "\n",
    "# 使用 LINE 生成的 embedding\n",
    "emb_path = f'../embs/dataset_{config[\"dataset\"]}_graph_src_dst_id.emb'\n",
    "all_node_emb = read_emb_data(emb_path)\n",
    "all_node_emb_pd = pd.DataFrame.from_dict(all_node_emb, orient='index')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "id": "9eaf94b4-19a7-4c96-bfb5-0e4c50bd3184",
   "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>src_id</th>\n",
       "      <th>dst_id</th>\n",
       "      <th>edge_type</th>\n",
       "      <th>start_time</th>\n",
       "      <th>end_time</th>\n",
       "      <th>label</th>\n",
       "      <th>test_flag</th>\n",
       "      <th>feat_0_x</th>\n",
       "      <th>feat_1_x</th>\n",
       "      <th>feat_2_x</th>\n",
       "      <th>...</th>\n",
       "      <th>190_y</th>\n",
       "      <th>191_y</th>\n",
       "      <th>192_y</th>\n",
       "      <th>193_y</th>\n",
       "      <th>194_y</th>\n",
       "      <th>195_y</th>\n",
       "      <th>196_y</th>\n",
       "      <th>197_y</th>\n",
       "      <th>198_y</th>\n",
       "      <th>199_y</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>31663</td>\n",
       "      <td>62299</td>\n",
       "      <td>71</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.307947</td>\n",
       "      <td>-0.231288</td>\n",
       "      <td>-0.036013</td>\n",
       "      <td>-0.062214</td>\n",
       "      <td>0.135924</td>\n",
       "      <td>-0.120684</td>\n",
       "      <td>-0.114572</td>\n",
       "      <td>0.020445</td>\n",
       "      <td>0.077689</td>\n",
       "      <td>-0.008791</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>26464</td>\n",
       "      <td>45285</td>\n",
       "      <td>147</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>14</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.029757</td>\n",
       "      <td>-0.131960</td>\n",
       "      <td>-0.008382</td>\n",
       "      <td>-0.108266</td>\n",
       "      <td>0.040760</td>\n",
       "      <td>0.201408</td>\n",
       "      <td>-0.211159</td>\n",
       "      <td>0.042460</td>\n",
       "      <td>-0.163271</td>\n",
       "      <td>-0.132325</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>36463</td>\n",
       "      <td>49022</td>\n",
       "      <td>151</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.269917</td>\n",
       "      <td>0.107135</td>\n",
       "      <td>0.055891</td>\n",
       "      <td>0.067172</td>\n",
       "      <td>-0.202249</td>\n",
       "      <td>0.200384</td>\n",
       "      <td>-0.280685</td>\n",
       "      <td>0.064039</td>\n",
       "      <td>-0.378790</td>\n",
       "      <td>-0.497700</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>21066</td>\n",
       "      <td>28905</td>\n",
       "      <td>20</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.100477</td>\n",
       "      <td>-0.038076</td>\n",
       "      <td>-0.001188</td>\n",
       "      <td>-0.139141</td>\n",
       "      <td>-0.131943</td>\n",
       "      <td>0.145833</td>\n",
       "      <td>-0.113107</td>\n",
       "      <td>0.038496</td>\n",
       "      <td>-0.032590</td>\n",
       "      <td>-0.072110</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>39224</td>\n",
       "      <td>57474</td>\n",
       "      <td>65</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.117139</td>\n",
       "      <td>-0.087244</td>\n",
       "      <td>-0.421603</td>\n",
       "      <td>-0.112137</td>\n",
       "      <td>-0.219417</td>\n",
       "      <td>0.192235</td>\n",
       "      <td>-0.143928</td>\n",
       "      <td>-0.167680</td>\n",
       "      <td>0.148916</td>\n",
       "      <td>0.096696</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658095</th>\n",
       "      <td>69980</td>\n",
       "      <td>49214</td>\n",
       "      <td>176</td>\n",
       "      <td>1494875670</td>\n",
       "      <td>1494996140</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.371442</td>\n",
       "      <td>-0.105631</td>\n",
       "      <td>-0.183275</td>\n",
       "      <td>-0.214641</td>\n",
       "      <td>-0.051064</td>\n",
       "      <td>0.039020</td>\n",
       "      <td>0.047359</td>\n",
       "      <td>-0.026662</td>\n",
       "      <td>-0.223549</td>\n",
       "      <td>-0.363136</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658096</th>\n",
       "      <td>69980</td>\n",
       "      <td>52874</td>\n",
       "      <td>33</td>\n",
       "      <td>1498584288</td>\n",
       "      <td>1498694102</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.046679</td>\n",
       "      <td>-0.037193</td>\n",
       "      <td>-0.143745</td>\n",
       "      <td>0.035014</td>\n",
       "      <td>0.167295</td>\n",
       "      <td>-0.271332</td>\n",
       "      <td>-0.194710</td>\n",
       "      <td>-0.131918</td>\n",
       "      <td>0.402429</td>\n",
       "      <td>0.197424</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658097</th>\n",
       "      <td>69980</td>\n",
       "      <td>55746</td>\n",
       "      <td>209</td>\n",
       "      <td>1497952740</td>\n",
       "      <td>1498081488</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.243592</td>\n",
       "      <td>-0.105645</td>\n",
       "      <td>0.012041</td>\n",
       "      <td>0.592462</td>\n",
       "      <td>0.084736</td>\n",
       "      <td>-0.049366</td>\n",
       "      <td>0.448350</td>\n",
       "      <td>0.221510</td>\n",
       "      <td>0.121660</td>\n",
       "      <td>0.118368</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658098</th>\n",
       "      <td>69980</td>\n",
       "      <td>60642</td>\n",
       "      <td>71</td>\n",
       "      <td>1497166072</td>\n",
       "      <td>1497356374</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.115532</td>\n",
       "      <td>0.109335</td>\n",
       "      <td>-0.061813</td>\n",
       "      <td>-0.052652</td>\n",
       "      <td>-0.005844</td>\n",
       "      <td>0.242444</td>\n",
       "      <td>-0.230357</td>\n",
       "      <td>-0.069731</td>\n",
       "      <td>-0.085827</td>\n",
       "      <td>-0.094216</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658099</th>\n",
       "      <td>69980</td>\n",
       "      <td>64236</td>\n",
       "      <td>69</td>\n",
       "      <td>1495174423</td>\n",
       "      <td>1495378242</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.166911</td>\n",
       "      <td>-0.404749</td>\n",
       "      <td>-0.635233</td>\n",
       "      <td>0.118225</td>\n",
       "      <td>0.180143</td>\n",
       "      <td>-0.168844</td>\n",
       "      <td>0.217368</td>\n",
       "      <td>0.062755</td>\n",
       "      <td>-0.086167</td>\n",
       "      <td>-0.185261</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>658100 rows × 436 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        src_id  dst_id  edge_type  start_time    end_time  label  test_flag  \\\n",
       "0        31663   62299         71           0           0      1          0   \n",
       "1        26464   45285        147           0           0      1          0   \n",
       "2        36463   49022        151           0           0      1          0   \n",
       "3        21066   28905         20           0           0      1          0   \n",
       "4        39224   57474         65           0           0      1          0   \n",
       "...        ...     ...        ...         ...         ...    ...        ...   \n",
       "658095   69980   49214        176  1494875670  1494996140     -1          3   \n",
       "658096   69980   52874         33  1498584288  1498694102     -1          3   \n",
       "658097   69980   55746        209  1497952740  1498081488     -1          3   \n",
       "658098   69980   60642         71  1497166072  1497356374     -1          3   \n",
       "658099   69980   64236         69  1495174423  1495378242     -1          3   \n",
       "\n",
       "        feat_0_x  feat_1_x  feat_2_x  ...     190_y     191_y     192_y  \\\n",
       "0             -1        -1        -1  ... -0.307947 -0.231288 -0.036013   \n",
       "1             14        -1        -1  ...  0.029757 -0.131960 -0.008382   \n",
       "2             -1        -1        -1  ... -0.269917  0.107135  0.055891   \n",
       "3             -1        -1        -1  ...  0.100477 -0.038076 -0.001188   \n",
       "4             -1        -1        -1  ...  0.117139 -0.087244 -0.421603   \n",
       "...          ...       ...       ...  ...       ...       ...       ...   \n",
       "658095        29        -1        -1  ...  0.371442 -0.105631 -0.183275   \n",
       "658096        29        -1        -1  ...  0.046679 -0.037193 -0.143745   \n",
       "658097        29        -1        -1  ... -0.243592 -0.105645  0.012041   \n",
       "658098        29        -1        -1  ...  0.115532  0.109335 -0.061813   \n",
       "658099        29        -1        -1  ...  0.166911 -0.404749 -0.635233   \n",
       "\n",
       "           193_y     194_y     195_y     196_y     197_y     198_y     199_y  \n",
       "0      -0.062214  0.135924 -0.120684 -0.114572  0.020445  0.077689 -0.008791  \n",
       "1      -0.108266  0.040760  0.201408 -0.211159  0.042460 -0.163271 -0.132325  \n",
       "2       0.067172 -0.202249  0.200384 -0.280685  0.064039 -0.378790 -0.497700  \n",
       "3      -0.139141 -0.131943  0.145833 -0.113107  0.038496 -0.032590 -0.072110  \n",
       "4      -0.112137 -0.219417  0.192235 -0.143928 -0.167680  0.148916  0.096696  \n",
       "...          ...       ...       ...       ...       ...       ...       ...  \n",
       "658095 -0.214641 -0.051064  0.039020  0.047359 -0.026662 -0.223549 -0.363136  \n",
       "658096  0.035014  0.167295 -0.271332 -0.194710 -0.131918  0.402429  0.197424  \n",
       "658097  0.592462  0.084736 -0.049366  0.448350  0.221510  0.121660  0.118368  \n",
       "658098 -0.052652 -0.005844  0.242444 -0.230357 -0.069731 -0.085827 -0.094216  \n",
       "658099  0.118225  0.180143 -0.168844  0.217368  0.062755 -0.086167 -0.185261  \n",
       "\n",
       "[658100 rows x 436 columns]"
      ]
     },
     "execution_count": 73,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_data = all_data.merge(all_node_emb_pd, how='left', left_on='src_id', right_index=True)\n",
    "all_data = all_data.merge(all_node_emb_pd, how='left', left_on='dst_id', right_index=True)\n",
    "all_data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "id": "c0132a62-28f5-4895-90ce-9f3e469826d4",
   "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>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>...</th>\n",
       "      <th>190</th>\n",
       "      <th>191</th>\n",
       "      <th>192</th>\n",
       "      <th>193</th>\n",
       "      <th>194</th>\n",
       "      <th>195</th>\n",
       "      <th>196</th>\n",
       "      <th>197</th>\n",
       "      <th>198</th>\n",
       "      <th>199</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>106</th>\n",
       "      <td>-0.143679</td>\n",
       "      <td>-0.138066</td>\n",
       "      <td>-0.179022</td>\n",
       "      <td>-0.102941</td>\n",
       "      <td>-0.207637</td>\n",
       "      <td>0.174723</td>\n",
       "      <td>-0.137081</td>\n",
       "      <td>-0.154120</td>\n",
       "      <td>-0.086287</td>\n",
       "      <td>0.073061</td>\n",
       "      <td>...</td>\n",
       "      <td>0.117980</td>\n",
       "      <td>0.010178</td>\n",
       "      <td>-0.065338</td>\n",
       "      <td>-0.222921</td>\n",
       "      <td>-0.029158</td>\n",
       "      <td>-0.314173</td>\n",
       "      <td>0.226679</td>\n",
       "      <td>-0.052983</td>\n",
       "      <td>0.075873</td>\n",
       "      <td>0.028314</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>28905</th>\n",
       "      <td>-0.001197</td>\n",
       "      <td>0.030057</td>\n",
       "      <td>0.053251</td>\n",
       "      <td>0.043523</td>\n",
       "      <td>0.092522</td>\n",
       "      <td>0.016269</td>\n",
       "      <td>0.025599</td>\n",
       "      <td>0.007239</td>\n",
       "      <td>0.007997</td>\n",
       "      <td>-0.006152</td>\n",
       "      <td>...</td>\n",
       "      <td>0.100477</td>\n",
       "      <td>-0.038076</td>\n",
       "      <td>-0.001188</td>\n",
       "      <td>-0.139141</td>\n",
       "      <td>-0.131943</td>\n",
       "      <td>0.145833</td>\n",
       "      <td>-0.113107</td>\n",
       "      <td>0.038496</td>\n",
       "      <td>-0.032590</td>\n",
       "      <td>-0.072110</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>211</th>\n",
       "      <td>0.209864</td>\n",
       "      <td>0.330594</td>\n",
       "      <td>-0.482570</td>\n",
       "      <td>-0.420206</td>\n",
       "      <td>-0.084706</td>\n",
       "      <td>-0.109880</td>\n",
       "      <td>0.400507</td>\n",
       "      <td>-0.143710</td>\n",
       "      <td>-0.098350</td>\n",
       "      <td>-0.072692</td>\n",
       "      <td>...</td>\n",
       "      <td>0.021754</td>\n",
       "      <td>-0.141119</td>\n",
       "      <td>0.174974</td>\n",
       "      <td>-0.216684</td>\n",
       "      <td>0.160032</td>\n",
       "      <td>0.201967</td>\n",
       "      <td>-0.025257</td>\n",
       "      <td>0.001877</td>\n",
       "      <td>-0.259129</td>\n",
       "      <td>0.127460</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14601</th>\n",
       "      <td>0.113454</td>\n",
       "      <td>0.086059</td>\n",
       "      <td>-0.046937</td>\n",
       "      <td>-0.060475</td>\n",
       "      <td>0.088795</td>\n",
       "      <td>0.390119</td>\n",
       "      <td>0.302458</td>\n",
       "      <td>-0.154821</td>\n",
       "      <td>0.270005</td>\n",
       "      <td>0.138436</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.327196</td>\n",
       "      <td>-0.031784</td>\n",
       "      <td>0.199060</td>\n",
       "      <td>-0.099823</td>\n",
       "      <td>-0.004415</td>\n",
       "      <td>0.300817</td>\n",
       "      <td>-0.160443</td>\n",
       "      <td>0.176800</td>\n",
       "      <td>-0.052096</td>\n",
       "      <td>-0.219006</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1355</th>\n",
       "      <td>-0.084775</td>\n",
       "      <td>0.217363</td>\n",
       "      <td>-0.012106</td>\n",
       "      <td>-0.178188</td>\n",
       "      <td>-0.021969</td>\n",
       "      <td>0.109538</td>\n",
       "      <td>-0.129507</td>\n",
       "      <td>0.220068</td>\n",
       "      <td>-0.299456</td>\n",
       "      <td>-0.188770</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.110394</td>\n",
       "      <td>-0.082987</td>\n",
       "      <td>0.142234</td>\n",
       "      <td>0.054355</td>\n",
       "      <td>0.123355</td>\n",
       "      <td>-0.138678</td>\n",
       "      <td>-0.123969</td>\n",
       "      <td>-0.347867</td>\n",
       "      <td>0.127350</td>\n",
       "      <td>0.231687</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19231</th>\n",
       "      <td>-0.435221</td>\n",
       "      <td>-0.297029</td>\n",
       "      <td>0.211241</td>\n",
       "      <td>-0.258996</td>\n",
       "      <td>-0.036141</td>\n",
       "      <td>0.055140</td>\n",
       "      <td>-0.038309</td>\n",
       "      <td>-0.033338</td>\n",
       "      <td>-0.016819</td>\n",
       "      <td>-0.052549</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.068154</td>\n",
       "      <td>-0.172159</td>\n",
       "      <td>-0.057960</td>\n",
       "      <td>0.063320</td>\n",
       "      <td>0.266082</td>\n",
       "      <td>-0.049363</td>\n",
       "      <td>0.314149</td>\n",
       "      <td>-0.084570</td>\n",
       "      <td>0.230775</td>\n",
       "      <td>0.148773</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>66826</th>\n",
       "      <td>-0.010120</td>\n",
       "      <td>-0.176895</td>\n",
       "      <td>0.134907</td>\n",
       "      <td>-0.038027</td>\n",
       "      <td>0.167903</td>\n",
       "      <td>-0.021254</td>\n",
       "      <td>0.001469</td>\n",
       "      <td>0.064794</td>\n",
       "      <td>0.042762</td>\n",
       "      <td>0.059468</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.142681</td>\n",
       "      <td>0.056983</td>\n",
       "      <td>0.032093</td>\n",
       "      <td>0.151746</td>\n",
       "      <td>0.031260</td>\n",
       "      <td>-0.074544</td>\n",
       "      <td>0.242160</td>\n",
       "      <td>0.112499</td>\n",
       "      <td>-0.124925</td>\n",
       "      <td>0.082595</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>59569</th>\n",
       "      <td>0.070693</td>\n",
       "      <td>-0.204979</td>\n",
       "      <td>0.032006</td>\n",
       "      <td>0.196028</td>\n",
       "      <td>0.246700</td>\n",
       "      <td>-0.080886</td>\n",
       "      <td>-0.114509</td>\n",
       "      <td>-0.226817</td>\n",
       "      <td>-0.165790</td>\n",
       "      <td>0.038977</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.394152</td>\n",
       "      <td>0.118517</td>\n",
       "      <td>-0.269513</td>\n",
       "      <td>0.374551</td>\n",
       "      <td>0.107566</td>\n",
       "      <td>-0.205706</td>\n",
       "      <td>0.140370</td>\n",
       "      <td>0.120040</td>\n",
       "      <td>0.083487</td>\n",
       "      <td>0.037723</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20584</th>\n",
       "      <td>0.055973</td>\n",
       "      <td>0.109655</td>\n",
       "      <td>-0.174843</td>\n",
       "      <td>-0.192254</td>\n",
       "      <td>-0.224077</td>\n",
       "      <td>0.020723</td>\n",
       "      <td>-0.375352</td>\n",
       "      <td>-0.122944</td>\n",
       "      <td>0.069258</td>\n",
       "      <td>0.112392</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.120199</td>\n",
       "      <td>-0.093420</td>\n",
       "      <td>-0.026998</td>\n",
       "      <td>0.367186</td>\n",
       "      <td>0.243420</td>\n",
       "      <td>-0.060124</td>\n",
       "      <td>0.322676</td>\n",
       "      <td>0.157109</td>\n",
       "      <td>0.030997</td>\n",
       "      <td>0.160439</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5186</th>\n",
       "      <td>-0.033309</td>\n",
       "      <td>-0.014556</td>\n",
       "      <td>-0.021603</td>\n",
       "      <td>0.037059</td>\n",
       "      <td>0.033960</td>\n",
       "      <td>0.023994</td>\n",
       "      <td>-0.001657</td>\n",
       "      <td>0.006569</td>\n",
       "      <td>0.006937</td>\n",
       "      <td>-0.000664</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.003747</td>\n",
       "      <td>-0.017012</td>\n",
       "      <td>0.044070</td>\n",
       "      <td>0.049526</td>\n",
       "      <td>0.014689</td>\n",
       "      <td>-0.039284</td>\n",
       "      <td>0.090866</td>\n",
       "      <td>-0.007974</td>\n",
       "      <td>0.014778</td>\n",
       "      <td>0.020189</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>19442 rows × 200 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "            0         1         2         3         4         5         6    \\\n",
       "106   -0.143679 -0.138066 -0.179022 -0.102941 -0.207637  0.174723 -0.137081   \n",
       "28905 -0.001197  0.030057  0.053251  0.043523  0.092522  0.016269  0.025599   \n",
       "211    0.209864  0.330594 -0.482570 -0.420206 -0.084706 -0.109880  0.400507   \n",
       "14601  0.113454  0.086059 -0.046937 -0.060475  0.088795  0.390119  0.302458   \n",
       "1355  -0.084775  0.217363 -0.012106 -0.178188 -0.021969  0.109538 -0.129507   \n",
       "...         ...       ...       ...       ...       ...       ...       ...   \n",
       "19231 -0.435221 -0.297029  0.211241 -0.258996 -0.036141  0.055140 -0.038309   \n",
       "66826 -0.010120 -0.176895  0.134907 -0.038027  0.167903 -0.021254  0.001469   \n",
       "59569  0.070693 -0.204979  0.032006  0.196028  0.246700 -0.080886 -0.114509   \n",
       "20584  0.055973  0.109655 -0.174843 -0.192254 -0.224077  0.020723 -0.375352   \n",
       "5186  -0.033309 -0.014556 -0.021603  0.037059  0.033960  0.023994 -0.001657   \n",
       "\n",
       "            7         8         9    ...       190       191       192  \\\n",
       "106   -0.154120 -0.086287  0.073061  ...  0.117980  0.010178 -0.065338   \n",
       "28905  0.007239  0.007997 -0.006152  ...  0.100477 -0.038076 -0.001188   \n",
       "211   -0.143710 -0.098350 -0.072692  ...  0.021754 -0.141119  0.174974   \n",
       "14601 -0.154821  0.270005  0.138436  ... -0.327196 -0.031784  0.199060   \n",
       "1355   0.220068 -0.299456 -0.188770  ... -0.110394 -0.082987  0.142234   \n",
       "...         ...       ...       ...  ...       ...       ...       ...   \n",
       "19231 -0.033338 -0.016819 -0.052549  ... -0.068154 -0.172159 -0.057960   \n",
       "66826  0.064794  0.042762  0.059468  ... -0.142681  0.056983  0.032093   \n",
       "59569 -0.226817 -0.165790  0.038977  ... -0.394152  0.118517 -0.269513   \n",
       "20584 -0.122944  0.069258  0.112392  ... -0.120199 -0.093420 -0.026998   \n",
       "5186   0.006569  0.006937 -0.000664  ... -0.003747 -0.017012  0.044070   \n",
       "\n",
       "            193       194       195       196       197       198       199  \n",
       "106   -0.222921 -0.029158 -0.314173  0.226679 -0.052983  0.075873  0.028314  \n",
       "28905 -0.139141 -0.131943  0.145833 -0.113107  0.038496 -0.032590 -0.072110  \n",
       "211   -0.216684  0.160032  0.201967 -0.025257  0.001877 -0.259129  0.127460  \n",
       "14601 -0.099823 -0.004415  0.300817 -0.160443  0.176800 -0.052096 -0.219006  \n",
       "1355   0.054355  0.123355 -0.138678 -0.123969 -0.347867  0.127350  0.231687  \n",
       "...         ...       ...       ...       ...       ...       ...       ...  \n",
       "19231  0.063320  0.266082 -0.049363  0.314149 -0.084570  0.230775  0.148773  \n",
       "66826  0.151746  0.031260 -0.074544  0.242160  0.112499 -0.124925  0.082595  \n",
       "59569  0.374551  0.107566 -0.205706  0.140370  0.120040  0.083487  0.037723  \n",
       "20584  0.367186  0.243420 -0.060124  0.322676  0.157109  0.030997  0.160439  \n",
       "5186   0.049526  0.014689 -0.039284  0.090866 -0.007974  0.014778  0.020189  \n",
       "\n",
       "[19442 rows x 200 columns]"
      ]
     },
     "execution_count": 74,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_node_emb_pd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "id": "347f9e1f-ef85-4daa-b6b1-80fc7dc37bfa",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 658100/658100 [01:07<00:00, 9726.77it/s] \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>src_id</th>\n",
       "      <th>dst_id</th>\n",
       "      <th>edge_type</th>\n",
       "      <th>start_time</th>\n",
       "      <th>end_time</th>\n",
       "      <th>label</th>\n",
       "      <th>test_flag</th>\n",
       "      <th>feat_0_x</th>\n",
       "      <th>feat_1_x</th>\n",
       "      <th>feat_2_x</th>\n",
       "      <th>...</th>\n",
       "      <th>192_y</th>\n",
       "      <th>193_y</th>\n",
       "      <th>194_y</th>\n",
       "      <th>195_y</th>\n",
       "      <th>196_y</th>\n",
       "      <th>197_y</th>\n",
       "      <th>198_y</th>\n",
       "      <th>199_y</th>\n",
       "      <th>dot_product</th>\n",
       "      <th>cos_sim</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>31663</td>\n",
       "      <td>62299</td>\n",
       "      <td>71</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.036013</td>\n",
       "      <td>-0.062214</td>\n",
       "      <td>0.135924</td>\n",
       "      <td>-0.120684</td>\n",
       "      <td>-0.114572</td>\n",
       "      <td>0.020445</td>\n",
       "      <td>0.077689</td>\n",
       "      <td>-0.008791</td>\n",
       "      <td>0.006271</td>\n",
       "      <td>0.500404</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>26464</td>\n",
       "      <td>45285</td>\n",
       "      <td>147</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>14</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.008382</td>\n",
       "      <td>-0.108266</td>\n",
       "      <td>0.040760</td>\n",
       "      <td>0.201408</td>\n",
       "      <td>-0.211159</td>\n",
       "      <td>0.042460</td>\n",
       "      <td>-0.163271</td>\n",
       "      <td>-0.132325</td>\n",
       "      <td>2.004063</td>\n",
       "      <td>0.733322</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>36463</td>\n",
       "      <td>49022</td>\n",
       "      <td>151</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.055891</td>\n",
       "      <td>0.067172</td>\n",
       "      <td>-0.202249</td>\n",
       "      <td>0.200384</td>\n",
       "      <td>-0.280685</td>\n",
       "      <td>0.064039</td>\n",
       "      <td>-0.378790</td>\n",
       "      <td>-0.497700</td>\n",
       "      <td>-0.001421</td>\n",
       "      <td>0.499910</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>21066</td>\n",
       "      <td>28905</td>\n",
       "      <td>20</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.001188</td>\n",
       "      <td>-0.139141</td>\n",
       "      <td>-0.131943</td>\n",
       "      <td>0.145833</td>\n",
       "      <td>-0.113107</td>\n",
       "      <td>0.038496</td>\n",
       "      <td>-0.032590</td>\n",
       "      <td>-0.072110</td>\n",
       "      <td>0.998731</td>\n",
       "      <td>0.613945</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>39224</td>\n",
       "      <td>57474</td>\n",
       "      <td>65</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.421603</td>\n",
       "      <td>-0.112137</td>\n",
       "      <td>-0.219417</td>\n",
       "      <td>0.192235</td>\n",
       "      <td>-0.143928</td>\n",
       "      <td>-0.167680</td>\n",
       "      <td>0.148916</td>\n",
       "      <td>0.096696</td>\n",
       "      <td>0.966918</td>\n",
       "      <td>0.542696</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658095</th>\n",
       "      <td>69980</td>\n",
       "      <td>49214</td>\n",
       "      <td>176</td>\n",
       "      <td>1494875670</td>\n",
       "      <td>1494996140</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.183275</td>\n",
       "      <td>-0.214641</td>\n",
       "      <td>-0.051064</td>\n",
       "      <td>0.039020</td>\n",
       "      <td>0.047359</td>\n",
       "      <td>-0.026662</td>\n",
       "      <td>-0.223549</td>\n",
       "      <td>-0.363136</td>\n",
       "      <td>2.745435</td>\n",
       "      <td>0.641429</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658096</th>\n",
       "      <td>69980</td>\n",
       "      <td>52874</td>\n",
       "      <td>33</td>\n",
       "      <td>1498584288</td>\n",
       "      <td>1498694102</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.143745</td>\n",
       "      <td>0.035014</td>\n",
       "      <td>0.167295</td>\n",
       "      <td>-0.271332</td>\n",
       "      <td>-0.194710</td>\n",
       "      <td>-0.131918</td>\n",
       "      <td>0.402429</td>\n",
       "      <td>0.197424</td>\n",
       "      <td>0.475853</td>\n",
       "      <td>0.527791</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658097</th>\n",
       "      <td>69980</td>\n",
       "      <td>55746</td>\n",
       "      <td>209</td>\n",
       "      <td>1497952740</td>\n",
       "      <td>1498081488</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.012041</td>\n",
       "      <td>0.592462</td>\n",
       "      <td>0.084736</td>\n",
       "      <td>-0.049366</td>\n",
       "      <td>0.448350</td>\n",
       "      <td>0.221510</td>\n",
       "      <td>0.121660</td>\n",
       "      <td>0.118368</td>\n",
       "      <td>0.043204</td>\n",
       "      <td>0.501932</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658098</th>\n",
       "      <td>69980</td>\n",
       "      <td>60642</td>\n",
       "      <td>71</td>\n",
       "      <td>1497166072</td>\n",
       "      <td>1497356374</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.061813</td>\n",
       "      <td>-0.052652</td>\n",
       "      <td>-0.005844</td>\n",
       "      <td>0.242444</td>\n",
       "      <td>-0.230357</td>\n",
       "      <td>-0.069731</td>\n",
       "      <td>-0.085827</td>\n",
       "      <td>-0.094216</td>\n",
       "      <td>0.923309</td>\n",
       "      <td>0.559225</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658099</th>\n",
       "      <td>69980</td>\n",
       "      <td>64236</td>\n",
       "      <td>69</td>\n",
       "      <td>1495174423</td>\n",
       "      <td>1495378242</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.635233</td>\n",
       "      <td>0.118225</td>\n",
       "      <td>0.180143</td>\n",
       "      <td>-0.168844</td>\n",
       "      <td>0.217368</td>\n",
       "      <td>0.062755</td>\n",
       "      <td>-0.086167</td>\n",
       "      <td>-0.185261</td>\n",
       "      <td>6.663761</td>\n",
       "      <td>0.848225</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>658100 rows × 438 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        src_id  dst_id  edge_type  start_time    end_time  label  test_flag  \\\n",
       "0        31663   62299         71           0           0      1          0   \n",
       "1        26464   45285        147           0           0      1          0   \n",
       "2        36463   49022        151           0           0      1          0   \n",
       "3        21066   28905         20           0           0      1          0   \n",
       "4        39224   57474         65           0           0      1          0   \n",
       "...        ...     ...        ...         ...         ...    ...        ...   \n",
       "658095   69980   49214        176  1494875670  1494996140     -1          3   \n",
       "658096   69980   52874         33  1498584288  1498694102     -1          3   \n",
       "658097   69980   55746        209  1497952740  1498081488     -1          3   \n",
       "658098   69980   60642         71  1497166072  1497356374     -1          3   \n",
       "658099   69980   64236         69  1495174423  1495378242     -1          3   \n",
       "\n",
       "        feat_0_x  feat_1_x  feat_2_x  ...     192_y     193_y     194_y  \\\n",
       "0             -1        -1        -1  ... -0.036013 -0.062214  0.135924   \n",
       "1             14        -1        -1  ... -0.008382 -0.108266  0.040760   \n",
       "2             -1        -1        -1  ...  0.055891  0.067172 -0.202249   \n",
       "3             -1        -1        -1  ... -0.001188 -0.139141 -0.131943   \n",
       "4             -1        -1        -1  ... -0.421603 -0.112137 -0.219417   \n",
       "...          ...       ...       ...  ...       ...       ...       ...   \n",
       "658095        29        -1        -1  ... -0.183275 -0.214641 -0.051064   \n",
       "658096        29        -1        -1  ... -0.143745  0.035014  0.167295   \n",
       "658097        29        -1        -1  ...  0.012041  0.592462  0.084736   \n",
       "658098        29        -1        -1  ... -0.061813 -0.052652 -0.005844   \n",
       "658099        29        -1        -1  ... -0.635233  0.118225  0.180143   \n",
       "\n",
       "           195_y     196_y     197_y     198_y     199_y  dot_product  \\\n",
       "0      -0.120684 -0.114572  0.020445  0.077689 -0.008791     0.006271   \n",
       "1       0.201408 -0.211159  0.042460 -0.163271 -0.132325     2.004063   \n",
       "2       0.200384 -0.280685  0.064039 -0.378790 -0.497700    -0.001421   \n",
       "3       0.145833 -0.113107  0.038496 -0.032590 -0.072110     0.998731   \n",
       "4       0.192235 -0.143928 -0.167680  0.148916  0.096696     0.966918   \n",
       "...          ...       ...       ...       ...       ...          ...   \n",
       "658095  0.039020  0.047359 -0.026662 -0.223549 -0.363136     2.745435   \n",
       "658096 -0.271332 -0.194710 -0.131918  0.402429  0.197424     0.475853   \n",
       "658097 -0.049366  0.448350  0.221510  0.121660  0.118368     0.043204   \n",
       "658098  0.242444 -0.230357 -0.069731 -0.085827 -0.094216     0.923309   \n",
       "658099 -0.168844  0.217368  0.062755 -0.086167 -0.185261     6.663761   \n",
       "\n",
       "         cos_sim  \n",
       "0       0.500404  \n",
       "1       0.733322  \n",
       "2       0.499910  \n",
       "3       0.613945  \n",
       "4       0.542696  \n",
       "...          ...  \n",
       "658095  0.641429  \n",
       "658096  0.527791  \n",
       "658097  0.501932  \n",
       "658098  0.559225  \n",
       "658099  0.848225  \n",
       "\n",
       "[658100 rows x 438 columns]"
      ]
     },
     "execution_count": 75,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 将 src_id, dst_id 两个节点 embedding 的点积以及余弦相似度加入特征中\n",
    "def cos_sim(vector_a, vector_b):\n",
    "    \"\"\"\n",
    "    计算两个向量之间的余弦相似度\n",
    "    :param vector_a: 向量 a\n",
    "    :param vector_b: 向量 b\n",
    "    :return: sim\n",
    "    \"\"\"\n",
    "    vector_a = np.mat(vector_a)\n",
    "    vector_b = np.mat(vector_b)\n",
    "    num = float(vector_a * vector_b.T)\n",
    "    denom = np.linalg.norm(vector_a) * np.linalg.norm(vector_b)\n",
    "    cos = num / denom\n",
    "    sim = 0.5 + 0.5 * cos\n",
    "    return sim\n",
    "\n",
    "dot_product_list = []\n",
    "cos_sim_list = []\n",
    "\n",
    "for i in tqdm(all_data.values):\n",
    "    src_id = int(i[0])\n",
    "    dst_id = int(i[1])\n",
    "    src_emb = all_node_emb[src_id]\n",
    "    dst_emb = all_node_emb[dst_id]\n",
    "    \n",
    "    dot_product_list.append(sum(src_emb * dst_emb))\n",
    "    cos_sim_list.append(cos_sim(src_emb, dst_emb))\n",
    "    \n",
    "all_data['dot_product'] = dot_product_list\n",
    "all_data['cos_sim'] = cos_sim_list\n",
    "all_data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "id": "3151087a-b312-4fc4-a5d7-2ec883404a96",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 将离散列转化为 object 类型\n",
    "global_sparse_columns = ['src_id',\n",
    "                         'dst_id',\n",
    "                         'edge_type',\n",
    "                        ]# + [i for i in all_data.columns if i.startswith('feat_')]\n",
    "all_data[global_sparse_columns] = all_data[global_sparse_columns].astype('str')\n",
    "        \n",
    "# 将连续列转化为 float 类型\n",
    "global_dense_columns = [\n",
    "    # 'exist_triple_count',\n",
    "    # 'exist_tuple_count',\n",
    "]\n",
    "all_data[global_dense_columns] = all_data[global_dense_columns].astype('float')\n",
    "\n",
    "# 打印所有的列名，及其类型\n",
    "# for i, col in enumerate(all_data):\n",
    "#     print(col, ' --> ', all_data[col].dtype)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "id": "5bcb134c-ed0b-49e5-82f3-c7d3181c6144",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "all_features:  (658100, 438)\n"
     ]
    }
   ],
   "source": [
    "all_features = all_data.copy()\n",
    "print('all_features: ', all_features.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "id": "3e74cbd8-8936-423a-b1fe-a167300c7d2e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n"
     ]
    }
   ],
   "source": [
    "# 为 float 列中的 NAN 填充该列均值\n",
    "na_count_sum = 0\n",
    "for i in all_features.columns:\n",
    "    na_count = sum(all_features[i].isna())\n",
    "    if na_count:\n",
    "        na_count_sum += 1\n",
    "#         print(i, na_count)\n",
    "#         print(all_features[i][~all_features[i].isna()].mean())\n",
    "        all_features[i] = all_features[i].fillna(all_features[i][~all_features[i].isna()].mean())\n",
    "print(na_count_sum)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "id": "9a233618-7360-49df-bace-02cfb0809d93",
   "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>src_id</th>\n",
       "      <th>dst_id</th>\n",
       "      <th>edge_type</th>\n",
       "      <th>label</th>\n",
       "      <th>test_flag</th>\n",
       "      <th>feat_0_x</th>\n",
       "      <th>feat_1_x</th>\n",
       "      <th>feat_2_x</th>\n",
       "      <th>feat_3_x</th>\n",
       "      <th>feat_4_x</th>\n",
       "      <th>...</th>\n",
       "      <th>192_y</th>\n",
       "      <th>193_y</th>\n",
       "      <th>194_y</th>\n",
       "      <th>195_y</th>\n",
       "      <th>196_y</th>\n",
       "      <th>197_y</th>\n",
       "      <th>198_y</th>\n",
       "      <th>199_y</th>\n",
       "      <th>dot_product</th>\n",
       "      <th>cos_sim</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>31663</td>\n",
       "      <td>62299</td>\n",
       "      <td>71</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.036013</td>\n",
       "      <td>-0.062214</td>\n",
       "      <td>0.135924</td>\n",
       "      <td>-0.120684</td>\n",
       "      <td>-0.114572</td>\n",
       "      <td>0.020445</td>\n",
       "      <td>0.077689</td>\n",
       "      <td>-0.008791</td>\n",
       "      <td>0.006271</td>\n",
       "      <td>0.500404</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>26464</td>\n",
       "      <td>45285</td>\n",
       "      <td>147</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>14</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.008382</td>\n",
       "      <td>-0.108266</td>\n",
       "      <td>0.040760</td>\n",
       "      <td>0.201408</td>\n",
       "      <td>-0.211159</td>\n",
       "      <td>0.042460</td>\n",
       "      <td>-0.163271</td>\n",
       "      <td>-0.132325</td>\n",
       "      <td>2.004063</td>\n",
       "      <td>0.733322</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>36463</td>\n",
       "      <td>49022</td>\n",
       "      <td>151</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.055891</td>\n",
       "      <td>0.067172</td>\n",
       "      <td>-0.202249</td>\n",
       "      <td>0.200384</td>\n",
       "      <td>-0.280685</td>\n",
       "      <td>0.064039</td>\n",
       "      <td>-0.378790</td>\n",
       "      <td>-0.497700</td>\n",
       "      <td>-0.001421</td>\n",
       "      <td>0.499910</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>21066</td>\n",
       "      <td>28905</td>\n",
       "      <td>20</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.001188</td>\n",
       "      <td>-0.139141</td>\n",
       "      <td>-0.131943</td>\n",
       "      <td>0.145833</td>\n",
       "      <td>-0.113107</td>\n",
       "      <td>0.038496</td>\n",
       "      <td>-0.032590</td>\n",
       "      <td>-0.072110</td>\n",
       "      <td>0.998731</td>\n",
       "      <td>0.613945</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>39224</td>\n",
       "      <td>57474</td>\n",
       "      <td>65</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.421603</td>\n",
       "      <td>-0.112137</td>\n",
       "      <td>-0.219417</td>\n",
       "      <td>0.192235</td>\n",
       "      <td>-0.143928</td>\n",
       "      <td>-0.167680</td>\n",
       "      <td>0.148916</td>\n",
       "      <td>0.096696</td>\n",
       "      <td>0.966918</td>\n",
       "      <td>0.542696</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658095</th>\n",
       "      <td>69980</td>\n",
       "      <td>49214</td>\n",
       "      <td>176</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.183275</td>\n",
       "      <td>-0.214641</td>\n",
       "      <td>-0.051064</td>\n",
       "      <td>0.039020</td>\n",
       "      <td>0.047359</td>\n",
       "      <td>-0.026662</td>\n",
       "      <td>-0.223549</td>\n",
       "      <td>-0.363136</td>\n",
       "      <td>2.745435</td>\n",
       "      <td>0.641429</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658096</th>\n",
       "      <td>69980</td>\n",
       "      <td>52874</td>\n",
       "      <td>33</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.143745</td>\n",
       "      <td>0.035014</td>\n",
       "      <td>0.167295</td>\n",
       "      <td>-0.271332</td>\n",
       "      <td>-0.194710</td>\n",
       "      <td>-0.131918</td>\n",
       "      <td>0.402429</td>\n",
       "      <td>0.197424</td>\n",
       "      <td>0.475853</td>\n",
       "      <td>0.527791</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658097</th>\n",
       "      <td>69980</td>\n",
       "      <td>55746</td>\n",
       "      <td>209</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>0.012041</td>\n",
       "      <td>0.592462</td>\n",
       "      <td>0.084736</td>\n",
       "      <td>-0.049366</td>\n",
       "      <td>0.448350</td>\n",
       "      <td>0.221510</td>\n",
       "      <td>0.121660</td>\n",
       "      <td>0.118368</td>\n",
       "      <td>0.043204</td>\n",
       "      <td>0.501932</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658098</th>\n",
       "      <td>69980</td>\n",
       "      <td>60642</td>\n",
       "      <td>71</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.061813</td>\n",
       "      <td>-0.052652</td>\n",
       "      <td>-0.005844</td>\n",
       "      <td>0.242444</td>\n",
       "      <td>-0.230357</td>\n",
       "      <td>-0.069731</td>\n",
       "      <td>-0.085827</td>\n",
       "      <td>-0.094216</td>\n",
       "      <td>0.923309</td>\n",
       "      <td>0.559225</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>658099</th>\n",
       "      <td>69980</td>\n",
       "      <td>64236</td>\n",
       "      <td>69</td>\n",
       "      <td>-1</td>\n",
       "      <td>3</td>\n",
       "      <td>29</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.635233</td>\n",
       "      <td>0.118225</td>\n",
       "      <td>0.180143</td>\n",
       "      <td>-0.168844</td>\n",
       "      <td>0.217368</td>\n",
       "      <td>0.062755</td>\n",
       "      <td>-0.086167</td>\n",
       "      <td>-0.185261</td>\n",
       "      <td>6.663761</td>\n",
       "      <td>0.848225</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>658100 rows × 435 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       src_id dst_id edge_type  label  test_flag  feat_0_x  feat_1_x  \\\n",
       "0       31663  62299        71      1          0        -1        -1   \n",
       "1       26464  45285       147      1          0        14        -1   \n",
       "2       36463  49022       151      1          0        -1        -1   \n",
       "3       21066  28905        20      1          0        -1        -1   \n",
       "4       39224  57474        65      1          0        -1        -1   \n",
       "...       ...    ...       ...    ...        ...       ...       ...   \n",
       "658095  69980  49214       176     -1          3        29        -1   \n",
       "658096  69980  52874        33     -1          3        29        -1   \n",
       "658097  69980  55746       209     -1          3        29        -1   \n",
       "658098  69980  60642        71     -1          3        29        -1   \n",
       "658099  69980  64236        69     -1          3        29        -1   \n",
       "\n",
       "        feat_2_x  feat_3_x  feat_4_x  ...     192_y     193_y     194_y  \\\n",
       "0             -1        -1        -1  ... -0.036013 -0.062214  0.135924   \n",
       "1             -1        -1        -1  ... -0.008382 -0.108266  0.040760   \n",
       "2             -1        -1        -1  ...  0.055891  0.067172 -0.202249   \n",
       "3             -1        -1        -1  ... -0.001188 -0.139141 -0.131943   \n",
       "4             -1        -1        -1  ... -0.421603 -0.112137 -0.219417   \n",
       "...          ...       ...       ...  ...       ...       ...       ...   \n",
       "658095        -1        -1        -1  ... -0.183275 -0.214641 -0.051064   \n",
       "658096        -1        -1        -1  ... -0.143745  0.035014  0.167295   \n",
       "658097        -1        -1        -1  ...  0.012041  0.592462  0.084736   \n",
       "658098        -1        -1        -1  ... -0.061813 -0.052652 -0.005844   \n",
       "658099        -1        -1        -1  ... -0.635233  0.118225  0.180143   \n",
       "\n",
       "           195_y     196_y     197_y     198_y     199_y  dot_product  \\\n",
       "0      -0.120684 -0.114572  0.020445  0.077689 -0.008791     0.006271   \n",
       "1       0.201408 -0.211159  0.042460 -0.163271 -0.132325     2.004063   \n",
       "2       0.200384 -0.280685  0.064039 -0.378790 -0.497700    -0.001421   \n",
       "3       0.145833 -0.113107  0.038496 -0.032590 -0.072110     0.998731   \n",
       "4       0.192235 -0.143928 -0.167680  0.148916  0.096696     0.966918   \n",
       "...          ...       ...       ...       ...       ...          ...   \n",
       "658095  0.039020  0.047359 -0.026662 -0.223549 -0.363136     2.745435   \n",
       "658096 -0.271332 -0.194710 -0.131918  0.402429  0.197424     0.475853   \n",
       "658097 -0.049366  0.448350  0.221510  0.121660  0.118368     0.043204   \n",
       "658098  0.242444 -0.230357 -0.069731 -0.085827 -0.094216     0.923309   \n",
       "658099 -0.168844  0.217368  0.062755 -0.086167 -0.185261     6.663761   \n",
       "\n",
       "         cos_sim  \n",
       "0       0.500404  \n",
       "1       0.733322  \n",
       "2       0.499910  \n",
       "3       0.613945  \n",
       "4       0.542696  \n",
       "...          ...  \n",
       "658095  0.641429  \n",
       "658096  0.527791  \n",
       "658097  0.501932  \n",
       "658098  0.559225  \n",
       "658099  0.848225  \n",
       "\n",
       "[658100 rows x 435 columns]"
      ]
     },
     "execution_count": 79,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 测试了一下，只有 merge 的时候导致的 na，因此填充 -1\n",
    "all_features = all_features.fillna(-1)\n",
    "\n",
    "if config['dataset'] == 'A':\n",
    "    del_columns = [# 'src_id', 'dst_id', 'edge_type', \n",
    "                           'start_time',\n",
    "                           'end_time',\n",
    "       # 'sparse:group_val_(src_id)_count_x',\n",
    "       # 'sparse:group_val_(src_id)_dst_id_nunique_x',   # Dataset B 不注释\n",
    "       # 'sparse:group_val_(src_id)_edge_type_nunique_x',\n",
    "       # 'sparse:group_val_(src_id)_count_y',\n",
    "       # 'sparse:group_val_(src_id)_dst_id_nunique_y',   # Dataset B 不注释\n",
    "       # 'sparse:group_val_(src_id)_edge_type_nunique_y',\n",
    "       # 'sparse:group_val_(src_id,dst_id)_count',       # Dataset B  不注释\n",
    "       # 'sparse:group_val_(src_id,dst_id)_edge_type_nunique',       # Dataset B  不注释\n",
    "       'sparse:group_val_(src_id,dst_id,edge_type)_count',  # 对 Dataset A 会影响效果，Dataset B  不注释\n",
    "       # 'two_hop_count_list',\n",
    "        # 'edge_feat',\n",
    "        # 'exist_triple_count',\n",
    "        # 'exist_tuple_count',\n",
    "                            # 'timestamp',\n",
    "                          ]\n",
    "else:\n",
    "    del_columns = [# 'src_id', 'dst_id', 'edge_type', \n",
    "                           'start_time',\n",
    "                           'end_time',\n",
    "       # 'sparse:group_val_(src_id)_count_x',\n",
    "       'sparse:group_val_(src_id)_dst_id_nunique_x',   # Dataset B 不注释\n",
    "       # 'sparse:group_val_(src_id)_edge_type_nunique_x',\n",
    "       # 'sparse:group_val_(src_id)_count_y',\n",
    "       'sparse:group_val_(src_id)_dst_id_nunique_y',   # Dataset B 不注释\n",
    "       # 'sparse:group_val_(src_id)_edge_type_nunique_y',\n",
    "       'sparse:group_val_(src_id,dst_id)_count',       # Dataset B  不注释\n",
    "       'sparse:group_val_(src_id,dst_id)_edge_type_nunique',       # Dataset B  不注释\n",
    "       'sparse:group_val_(src_id,dst_id,edge_type)_count',  # 对 Dataset A 会影响效果，Dataset B  不注释\n",
    "       # 'two_hop_count_list',\n",
    "        # 'edge_feat',\n",
    "        # 'exist_triple_count',\n",
    "        # 'exist_tuple_count',\n",
    "                            # 'timestamp',\n",
    "                          ]\n",
    "\n",
    "all_features.drop(columns=del_columns, inplace=True, errors='ignore')\n",
    "\n",
    "# for column_name in all_features.columns:\n",
    "#     if column_name.startswith('feat_'):\n",
    "#         all_features[column_name] = all_features[column_name].astype(str)\n",
    "\n",
    "all_features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "id": "630167db-341e-47f1-97e6-39ba29bb3751",
   "metadata": {},
   "outputs": [],
   "source": [
    "def category_idization(features, encoding='le'):\n",
    "    # 类别 id 化\n",
    "    \n",
    "    cat_indices = None\n",
    "    # One Hot Encoding\n",
    "    if encoding == 'ohe':\n",
    "        features = pd.get_dummies(features)\n",
    "        cat_indices = 'auto'\n",
    "\n",
    "    # Integer label encoding\n",
    "    elif encoding == 'le':\n",
    "        label_encoder = LabelEncoder()\n",
    "        cat_indices = []\n",
    "        for i, col in enumerate(features):\n",
    "            if features[col].dtype == 'object':\n",
    "                # print('transform object -> int id, ', col)\n",
    "                features[col] = label_encoder.fit_transform(np.array(features[col].astype(str)).reshape((-1,)))\n",
    "                cat_indices.append(col)\n",
    "    else:\n",
    "        raise ValueError(\"Encoding must be either 'ohe' or 'le'\")\n",
    "\n",
    "    print('Data Shape: ', features.shape)\n",
    "    print('sparse column: ', cat_indices)\n",
    "    print('sparse column idx: ', [features.columns.get_loc(key) for key in cat_indices])\n",
    "    \n",
    "    return features, cat_indices"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "id": "5a3004c0-ad0b-45ed-b8e8-a2429554223c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Data Shape:  (658100, 435)\n",
      "sparse column:  ['src_id', 'dst_id', 'edge_type']\n",
      "sparse column idx:  [0, 1, 2]\n"
     ]
    }
   ],
   "source": [
    "all_features, cat_indices = category_idization(features=all_features.copy(), encoding='le')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "id": "513e3a8e-dd54-49ea-98ab-cc5a80b7183b",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = None\n",
    "best_iteration = None\n",
    "train_features = None\n",
    "valid_features = None\n",
    "\n",
    "def get_model(args):\n",
    "    if args['model_name'] == 'cbt':\n",
    "        model = cbt.CatBoostClassifier(\n",
    "                            n_estimators = args['n_estimators'],\n",
    "                            num_leaves=args['num_leaves'],\n",
    "                            depth=args['max_depth'],\n",
    "                            objective=None,\n",
    "                            # boosting_type = args['boosting_type'],\n",
    "                            # class_weights = args['class_weight'],\n",
    "                            learning_rate = args['learning_rate'],\n",
    "                            # reg_alpha = args['reg_alpha'],\n",
    "                            reg_lambda = args['reg_lambda'],\n",
    "                            subsample = args['subsample'],\n",
    "                            # subsample_for_bin=args['subsample_for_bin'],\n",
    "                            colsample_bylevel=args['colsample_bytree'],\n",
    "                            # n_jobs = -1,\n",
    "                            random_state = args['random_seed'],\n",
    "                            eval_metric=\"AUC\",\n",
    "                            # task_type=\"GPU\",\n",
    "                            od_type=\"Iter\",\n",
    "                            border_count=64,\n",
    "                        )\n",
    "    elif args['model_name'] == 'lgb':\n",
    "        model = lgb.LGBMClassifier(n_estimators=args['n_estimators'],\n",
    "                           num_leaves=args['num_leaves'],\n",
    "                           max_depth=args['max_depth'],\n",
    "                           objective = 'binary',\n",
    "                           boosting_type = args['boosting_type'],\n",
    "                           class_weight = args['class_weight'],\n",
    "                           learning_rate = args['learning_rate'], \n",
    "                           reg_alpha = args['reg_alpha'],\n",
    "                           reg_lambda = args['reg_lambda'], \n",
    "                           subsample = args['subsample'],\n",
    "                           subsample_for_bin=args['subsample_for_bin'],\n",
    "                           colsample_bytree=args['colsample_bytree'],\n",
    "                           n_jobs = -1,\n",
    "                           random_state = args['random_seed'],\n",
    "                           importance_type='gain')\n",
    "    return model\n",
    "\n",
    "\n",
    "def model_fit(model, train_features, train_labels, valid_features, valid_labels, test_features, test_labels, cat_indices, args):\n",
    "    if args['model_name'] == 'cbt':\n",
    "        model.fit(train_features,\n",
    "                  train_labels,\n",
    "        #           eval_metric = [\n",
    "        #                 lgb_fbeta_0_3,\n",
    "        #                 'auc',\n",
    "        #           ],\n",
    "                  eval_set = [\n",
    "                      # (valid_features, valid_labels),\n",
    "                      (test_features, test_labels),\n",
    "                      # (train_features, train_labels)\n",
    "                  ],\n",
    "        #           eval_names = [\n",
    "        #               'valid',\n",
    "        #               'train'\n",
    "        #           ],\n",
    "                  cat_features = cat_indices,\n",
    "                  early_stopping_rounds = 3000,\n",
    "                  verbose = args['verbose'],\n",
    "                  # plot=True,\n",
    "                 )\n",
    "        if args['use_wandb']:\n",
    "            for iter in range(len(model.evals_result_['learn']['Logloss'])):\n",
    "                # 将结果 log 进 wandb\n",
    "                wandb.log(\n",
    "                    {'train_binary_logloss': model.evals_result_['learn']['Logloss'][iter],\n",
    "                     'dev_binary_logloss': model.evals_result_['validation']['Logloss'][iter],\n",
    "                     'dev_auc': model.evals_result_['validation']['AUC'][iter],\n",
    "                    }\n",
    "                )\n",
    "        return {\n",
    "                    'train': model.evals_result_['learn'],\n",
    "                    'valid': model.evals_result_['validation'],\n",
    "               }, model.best_iteration_, {\n",
    "                    'train': {'binary_logloss': model.best_score_['learn']['Logloss']},\n",
    "                    'valid': {'binary_logloss': model.best_score_['validation']['Logloss']},\n",
    "                }\n",
    "    \n",
    "    elif args['model_name'] == 'lgb':\n",
    "        train_features = np.array(train_features)\n",
    "        valid_features = np.array(valid_features)\n",
    "        test_features = np.array(test_features)\n",
    "        \n",
    "        model.fit(train_features,\n",
    "                  train_labels,\n",
    "                  eval_metric = [\n",
    "                        # lgb_fbeta_0_3,\n",
    "                        'auc',\n",
    "                  ],\n",
    "                  eval_set = [\n",
    "                      (valid_features, valid_labels),\n",
    "                      (test_features, test_labels),\n",
    "                      (train_features, train_labels)\n",
    "                  ],\n",
    "                  eval_names = [\n",
    "                      'valid',\n",
    "                      'dev',\n",
    "                      'train',\n",
    "                  ],\n",
    "                  categorical_feature = cat_indices,\n",
    "                  early_stopping_rounds = 3000,\n",
    "                  verbose = args['verbose'],\n",
    "                  callbacks=[wandb.lightgbm.wandb_callback()] if args['use_wandb'] else []\n",
    "                 )\n",
    "        best_iteration = model.best_iteration_  # 模型默认 best iteration\n",
    "        best_iteration_with_logloss = np.argmin(model.evals_result_['valid']['binary_logloss'])  # 最好的 binary logloss 的 iter\n",
    "        best_iteration_with_auc = np.argmax(model.evals_result_['valid']['auc']) # 最好的 auc 的 iter\n",
    "        # best_iteration_with_fscore = np.argmax(model.evals_result_['valid']['fbeta-0.3']) # 最好的 fbeta-0.3 的 iter\n",
    "        \n",
    "        # 经过测试，选 fbeta-0.3 最高的优于 auc、优于 logloss\n",
    "        print('--> best_iteration (default): ', best_iteration)\n",
    "        print('--> best_iteration (binary_logloss): ', best_iteration_with_logloss)\n",
    "        print('--> best_iteration (auc): ', best_iteration_with_auc)\n",
    "        # print('--> best_iteration (fbeta-0.3): ', best_iteration_with_fscore)\n",
    "        \n",
    "        return model.evals_result_, best_iteration_with_auc, model.best_score_\n",
    "\n",
    "    \n",
    "def model_predict_proba(model, features, best_iteration, args):\n",
    "    res = None\n",
    "    if args['model_name'] == 'cbt':\n",
    "        res = model.predict_proba(features)\n",
    "    elif args['model_name'] == 'lgb':\n",
    "        features = np.array(features)\n",
    "        res = model.predict_proba(features, num_iteration = best_iteration)\n",
    "    return res\n",
    "\n",
    "\n",
    "def sigle_model(features, dev_features, test_features, args, n_folds = 5):\n",
    "    global model, best_iteration, train_features, valid_features\n",
    "    model = None\n",
    "    \n",
    "    #提取ID\n",
    "    # train_ids = features['id']\n",
    "    dev_ids = dev_features.index\n",
    "    test_ids = test_features.index\n",
    "    \n",
    "    # 提取训练集的结果\n",
    "    labels = features['label']\n",
    "    dev_labels = dev_features['label']\n",
    "    # 移除ID和target,date\n",
    "    # train_date_name = [i for i in features.columns if \"date\" in i]\n",
    "    \n",
    "    del_name = ['label', 'test_flag']\n",
    "#     del_name = ['label', 'request_id'] # 后面的 cat index 编号可能出问题\n",
    "    features = features.drop(columns = del_name)\n",
    "    dev_features = dev_features.drop(columns = del_name)\n",
    "    test_features = test_features.drop(columns = del_name)\n",
    "\n",
    "    new_cat_indices = [features.columns.get_loc(key) for key in cat_indices]  # 类别列 idx 根据列名称获取\n",
    "\n",
    "    print('del_columns_name: ', del_name)\n",
    "    print('new_cat_indices: ', new_cat_indices)\n",
    "    print('all_features (in model): ', features.iloc[:, new_cat_indices])\n",
    "    # return\n",
    "    \n",
    "    # print(features.columns)\n",
    "    # print(test_features.columns)\n",
    "\n",
    "    feature_names = list(features.columns)\n",
    "    print(feature_names)\n",
    "    \n",
    "#     print(features.iloc[:, new_cat_indices])\n",
    "#     return\n",
    "    \n",
    "    # features = np.array(features)\n",
    "    # test_features = np.array(test_features)\n",
    "\n",
    "    k_fold = KFold(n_splits = n_folds, shuffle = True, random_state = args['random_seed'])\n",
    "    \n",
    "    feature_importance_values = np.zeros(len(feature_names))\n",
    "    \n",
    "    dev_predictions = np.zeros(dev_features.shape[0])\n",
    "    test_predictions = np.zeros(test_features.shape[0])\n",
    "    \n",
    "    out_of_fold = np.zeros(features.shape[0])\n",
    "    \n",
    "    valid_scores = {}\n",
    "    train_scores = {}\n",
    "    valid_fbeta = []\n",
    "    all_model = []\n",
    "    \n",
    "    group_name = '{}'.format(wandb.util.generate_id())\n",
    "    \n",
    "    for fold_idx, (train_indices, valid_indices) in enumerate(k_fold.split(features)):\n",
    "        if config['use_wandb']:\n",
    "            # 配置 wandb\n",
    "            wandb.init(project=f'wsdm2022tgp_tree_solution_{config[\"dataset\"]}',\n",
    "                       entity='graphml',\n",
    "                       config=args,\n",
    "                       group=group_name,\n",
    "                       job_type='fold-{}'.format(fold_idx))\n",
    "        \n",
    "        train_features, train_labels = features.loc[train_indices], labels[train_indices]\n",
    "        valid_features, valid_labels = features.loc[valid_indices], labels[valid_indices]\n",
    "        \n",
    "        #建模\n",
    "        model = get_model(args)\n",
    "#         model = lgb.LGBMClassifier(n_estimators=args['n_estimators'],\n",
    "#                                    num_leaves=args['num_leaves'],\n",
    "#                                    max_depth=args['max_depth'],\n",
    "#                                    objective = 'binary',\n",
    "#                                    boosting_type = args['boosting_type'],\n",
    "#                                    class_weight = args['class_weight'],\n",
    "#                                    learning_rate = args['learning_rate'], \n",
    "#                                    reg_alpha = args['reg_alpha'],\n",
    "#                                    reg_lambda = args['reg_lambda'], \n",
    "#                                    subsample = args['subsample'],\n",
    "#                                    subsample_for_bin=args['subsample_for_bin'],\n",
    "#                                    colsample_bytree=args['colsample_bytree'],\n",
    "#                                    n_jobs = -1,\n",
    "#                                    random_state = args['random_seed'],\n",
    "#                                    importance_type='gain')\n",
    "        \n",
    "        # 训练模型\n",
    "#         print('训练特征维度：', train_features.shape)\n",
    "#         print(train_features[:, new_cat_indices])\n",
    "        evals_result_, best_iteration, best_score_ = model_fit(model, train_features, train_labels, valid_features, valid_labels, dev_features, dev_labels, new_cat_indices, args)\n",
    "#         model.fit(train_features,\n",
    "#                   train_labels,\n",
    "#                   eval_metric = [\n",
    "#                         lgb_fbeta_0_3,\n",
    "#                         'auc',\n",
    "#                   ],\n",
    "#                   eval_set = [\n",
    "#                       (valid_features, valid_labels),\n",
    "#                       (train_features, train_labels)\n",
    "#                   ],\n",
    "#                   eval_names = [\n",
    "#                       'valid',\n",
    "#                       'train'\n",
    "#                   ],\n",
    "#                   categorical_feature = new_cat_indices,\n",
    "#                   early_stopping_rounds = 3000,\n",
    "#                   verbose = args['verbose'],\n",
    "#                   callbacks=[wandb.lightgbm.wandb_callback()] if args['use_wandb'] else []\n",
    "#                  )\n",
    "        \n",
    "#         best_iteration = model.best_iteration_  # 模型默认 best iteration\n",
    "#         best_iteration_with_logloss = np.argmin(evals_results['valid']['binary_logloss'])  # 最好的 binary logloss 的 iter\n",
    "#         best_iteration_with_auc = np.argmax(evals_results['valid']['auc']) # 最好的 auc 的 iter\n",
    "#         best_iteration_with_fscore = np.argmax(evals_results['valid']['fbeta-0.3']) # 最好的 fbeta-0.3 的 iter\n",
    "        all_model.append(model)\n",
    "        \n",
    "        # 特征重要性\n",
    "        feature_importance_values += model.feature_importances_ / k_fold.n_splits\n",
    "        \n",
    "        # 做预测\n",
    "        proba_iteration = best_iteration\n",
    "        print('--> use iteration [{}] for proba'.format(proba_iteration))\n",
    "        dev_predictions += model_predict_proba(model, dev_features, proba_iteration, args)[:, 1] / k_fold.n_splits\n",
    "        test_predictions += model_predict_proba(model, test_features, proba_iteration, args)[:, 1] / k_fold.n_splits\n",
    "        \n",
    "        out_of_fold[valid_indices] = model_predict_proba(model, valid_features, proba_iteration, args)[:, 1]\n",
    "        \n",
    "\n",
    "        print('--> valid label = 1 占比：', sum(out_of_fold[valid_indices] >= 0.5) / len(out_of_fold[valid_indices]))\n",
    "        \n",
    "        # print(model.best_score_)\n",
    "#         valid_score = model.best_score_['valid']['auc']\n",
    "#         train_score = model.best_score_['train']['auc']\n",
    "#         fbeta = lgb_fbeta(valid_labels, out_of_fold[valid_indices], beta=0.3)\n",
    "\n",
    "        for score_name in best_score_['train'].keys():\n",
    "            if score_name in valid_scores:\n",
    "                train_scores[score_name].append(best_score_['train'][score_name])\n",
    "                valid_scores[score_name].append(best_score_['valid'][score_name])\n",
    "            else:\n",
    "                train_scores[score_name] = [best_score_['train'][score_name]]\n",
    "                valid_scores[score_name] = [best_score_['valid'][score_name]]\n",
    "        \n",
    "       # valid_fbeta.append(fbeta)\n",
    "        \n",
    "        gc.enable()\n",
    "        del model, train_features, valid_features\n",
    "        gc.collect()\n",
    "        \n",
    "        if config['use_wandb']:\n",
    "            # 配置 wandb end\n",
    "            #wandb.log(metrics[['train_auc', 'valid_auc', 'valid-fbeta']].iloc[-1].to_dict())\n",
    "            wandb.finish()\n",
    "            pass\n",
    "        \n",
    "    submission = pd.DataFrame({'test_ids': test_ids, 'test_label': test_predictions})\n",
    "    dev_submission = pd.DataFrame({'dev_ids': dev_ids, 'dev_label': dev_predictions})\n",
    "    \n",
    "    feature_importances = pd.DataFrame({'feature': feature_names, 'importance': feature_importance_values})\n",
    "    \n",
    "\n",
    "    if 'auc' in valid_scores:\n",
    "        valid_auc = roc_auc_score(labels, out_of_fold)\n",
    "        valid_scores['auc'].append(valid_auc)\n",
    "        train_scores['auc'].append(np.mean(train_scores['auc']))\n",
    "    if 'fbeta-0.3' in valid_scores:\n",
    "        valid_fbeta = lgb_fbeta(labels, out_of_fold, beta=0.3)\n",
    "        valid_scores['fbeta-0.3'].append(valid_fbeta)\n",
    "        train_scores['fbeta-0.3'].append(np.mean(train_scores['fbeta-0.3']))\n",
    "    if 'binary_logloss' in train_scores:\n",
    "        train_scores['binary_logloss'].append(np.mean(train_scores['binary_logloss']))\n",
    "        valid_scores['binary_logloss'].append(np.mean(valid_scores['binary_logloss']))\n",
    "\n",
    "    fold_names = list(range(n_folds))\n",
    "    fold_names.append('overall')\n",
    "    \n",
    "    print('dev auc: ', roc_auc_score(dev_labels, dev_predictions))\n",
    "    \n",
    "    #print(train_scores)\n",
    "    #print(valid_scores)\n",
    "    #print(fold_names)\n",
    "    \n",
    "    metrics = pd.DataFrame({'fold': fold_names,\n",
    "                            # **train_scores,\n",
    "                            **valid_scores,\n",
    "                           }) \n",
    "    \n",
    "    return dev_submission, submission, feature_importances, metrics, all_model, out_of_fold"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "id": "70786b6b-7bff-453b-aadb-76331e86a1fe",
   "metadata": {},
   "outputs": [],
   "source": [
    "sweep_config = {\n",
    "  # \"name\" : \"my-sweep\",\n",
    "  'metric': {'name': 'valid_fbeta-0.3', 'goal': 'maximize'},\n",
    "  \"method\" : \"bayes\",\n",
    "  \"parameters\" : {\n",
    "    \"colsample_bytree\": {\n",
    "        \"distribution\": \"uniform\",\n",
    "        \"min\": 0.5,\n",
    "        \"max\": 1.0\n",
    "    },\n",
    "    \"learning_rate\" :{\n",
    "        \"distribution\": \"uniform\",\n",
    "        \"min\": 0.005,\n",
    "        \"max\": 0.2\n",
    "    },\n",
    "    \"num_leaves\": {\n",
    "        \"distribution\": \"int_uniform\",\n",
    "        \"min\": 8,\n",
    "        \"max\": 64\n",
    "    },\n",
    "    \"random_seed\": {\n",
    "        \"distribution\": \"int_uniform\",\n",
    "        \"min\": 0,\n",
    "        \"max\": 2048\n",
    "    },\n",
    "    \"reg_alpha\": {\n",
    "        \"distribution\": \"uniform\",\n",
    "        \"min\": 0.0,\n",
    "        \"max\": 0.2\n",
    "    },\n",
    "    \"reg_lambda\": {\n",
    "        \"distribution\": \"uniform\",\n",
    "        \"min\": 0.0,\n",
    "        \"max\": 0.2\n",
    "    },\n",
    "    \"subsample\": {\n",
    "        \"distribution\": \"uniform\",\n",
    "        \"min\": 0.5,\n",
    "        \"max\": 1.0\n",
    "    },\n",
    "  }\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "id": "5b1ba04f-f5c6-476d-a1bc-becd96bfa4b9",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "del_columns_name:  ['label', 'test_flag']\n",
      "new_cat_indices:  [0, 1, 2]\n",
      "all_features (in model):          src_id  dst_id  edge_type\n",
      "0         3796    8980        201\n",
      "1         2860    6092         47\n",
      "2         4685    6731         52\n",
      "3         1929    3249        104\n",
      "4         5137    8147        194\n",
      "...        ...     ...        ...\n",
      "399995    3313    2186        150\n",
      "399996    3313    9039        104\n",
      "399997    6264    5443        204\n",
      "399998    1564    6092        225\n",
      "399999    2884    1957        161\n",
      "\n",
      "[400000 rows x 3 columns]\n",
      "['src_id', 'dst_id', 'edge_type', 'feat_0_x', 'feat_1_x', 'feat_2_x', 'feat_3_x', 'feat_4_x', 'feat_5_x', 'feat_6_x', 'feat_7_x', 'feat_0_y', 'feat_1_y', 'feat_2_y', 'feat_3_y', 'feat_4_y', 'feat_5_y', 'feat_6_y', 'feat_7_y', 'feat_0', 'feat_1', 'feat_2', 'sparse:group_val_(src_id)_count_x', 'sparse:group_val_(src_id)_dst_id_nunique_x', 'sparse:group_val_(src_id)_edge_type_nunique_x', 'sparse:group_val_(src_id)_count_y', 'sparse:group_val_(src_id)_dst_id_nunique_y', 'sparse:group_val_(src_id)_edge_type_nunique_y', 'sparse:group_val_(src_id,dst_id)_count', 'sparse:group_val_(src_id,dst_id)_edge_type_nunique', 'two_hop_count_list', '0_x', '1_x', '2_x', '3_x', '4_x', '5_x', '6_x', '7_x', '8_x', '9_x', '10_x', '11_x', '12_x', '13_x', '14_x', '15_x', '16_x', '17_x', '18_x', '19_x', '20_x', '21_x', '22_x', '23_x', '24_x', '25_x', '26_x', '27_x', '28_x', '29_x', '30_x', '31_x', '32_x', '33_x', '34_x', '35_x', '36_x', '37_x', '38_x', '39_x', '40_x', '41_x', '42_x', '43_x', '44_x', '45_x', '46_x', '47_x', '48_x', '49_x', '50_x', '51_x', '52_x', '53_x', '54_x', '55_x', '56_x', '57_x', '58_x', '59_x', '60_x', '61_x', '62_x', '63_x', '64_x', '65_x', '66_x', '67_x', '68_x', '69_x', '70_x', '71_x', '72_x', '73_x', '74_x', '75_x', '76_x', '77_x', '78_x', '79_x', '80_x', '81_x', '82_x', '83_x', '84_x', '85_x', '86_x', '87_x', '88_x', '89_x', '90_x', '91_x', '92_x', '93_x', '94_x', '95_x', '96_x', '97_x', '98_x', '99_x', '100_x', '101_x', '102_x', '103_x', '104_x', '105_x', '106_x', '107_x', '108_x', '109_x', '110_x', '111_x', '112_x', '113_x', '114_x', '115_x', '116_x', '117_x', '118_x', '119_x', '120_x', '121_x', '122_x', '123_x', '124_x', '125_x', '126_x', '127_x', '128_x', '129_x', '130_x', '131_x', '132_x', '133_x', '134_x', '135_x', '136_x', '137_x', '138_x', '139_x', '140_x', '141_x', '142_x', '143_x', '144_x', '145_x', '146_x', '147_x', '148_x', '149_x', '150_x', '151_x', '152_x', '153_x', '154_x', '155_x', '156_x', '157_x', '158_x', '159_x', '160_x', '161_x', '162_x', '163_x', '164_x', '165_x', '166_x', '167_x', '168_x', '169_x', '170_x', '171_x', '172_x', '173_x', '174_x', '175_x', '176_x', '177_x', '178_x', '179_x', '180_x', '181_x', '182_x', '183_x', '184_x', '185_x', '186_x', '187_x', '188_x', '189_x', '190_x', '191_x', '192_x', '193_x', '194_x', '195_x', '196_x', '197_x', '198_x', '199_x', '0_y', '1_y', '2_y', '3_y', '4_y', '5_y', '6_y', '7_y', '8_y', '9_y', '10_y', '11_y', '12_y', '13_y', '14_y', '15_y', '16_y', '17_y', '18_y', '19_y', '20_y', '21_y', '22_y', '23_y', '24_y', '25_y', '26_y', '27_y', '28_y', '29_y', '30_y', '31_y', '32_y', '33_y', '34_y', '35_y', '36_y', '37_y', '38_y', '39_y', '40_y', '41_y', '42_y', '43_y', '44_y', '45_y', '46_y', '47_y', '48_y', '49_y', '50_y', '51_y', '52_y', '53_y', '54_y', '55_y', '56_y', '57_y', '58_y', '59_y', '60_y', '61_y', '62_y', '63_y', '64_y', '65_y', '66_y', '67_y', '68_y', '69_y', '70_y', '71_y', '72_y', '73_y', '74_y', '75_y', '76_y', '77_y', '78_y', '79_y', '80_y', '81_y', '82_y', '83_y', '84_y', '85_y', '86_y', '87_y', '88_y', '89_y', '90_y', '91_y', '92_y', '93_y', '94_y', '95_y', '96_y', '97_y', '98_y', '99_y', '100_y', '101_y', '102_y', '103_y', '104_y', '105_y', '106_y', '107_y', '108_y', '109_y', '110_y', '111_y', '112_y', '113_y', '114_y', '115_y', '116_y', '117_y', '118_y', '119_y', '120_y', '121_y', '122_y', '123_y', '124_y', '125_y', '126_y', '127_y', '128_y', '129_y', '130_y', '131_y', '132_y', '133_y', '134_y', '135_y', '136_y', '137_y', '138_y', '139_y', '140_y', '141_y', '142_y', '143_y', '144_y', '145_y', '146_y', '147_y', '148_y', '149_y', '150_y', '151_y', '152_y', '153_y', '154_y', '155_y', '156_y', '157_y', '158_y', '159_y', '160_y', '161_y', '162_y', '163_y', '164_y', '165_y', '166_y', '167_y', '168_y', '169_y', '170_y', '171_y', '172_y', '173_y', '174_y', '175_y', '176_y', '177_y', '178_y', '179_y', '180_y', '181_y', '182_y', '183_y', '184_y', '185_y', '186_y', '187_y', '188_y', '189_y', '190_y', '191_y', '192_y', '193_y', '194_y', '195_y', '196_y', '197_y', '198_y', '199_y', 'dot_product', 'cos_sim']\n",
      "0:\ttest: 0.6398351\tbest: 0.6398351 (0)\ttotal: 97.8ms\tremaining: 1m 37s\n",
      "50:\ttest: 0.6576868\tbest: 0.6576868 (50)\ttotal: 4.73s\tremaining: 1m 28s\n",
      "100:\ttest: 0.6622587\tbest: 0.6623486 (99)\ttotal: 9.26s\tremaining: 1m 22s\n",
      "150:\ttest: 0.6643775\tbest: 0.6643775 (150)\ttotal: 14s\tremaining: 1m 18s\n",
      "200:\ttest: 0.6657355\tbest: 0.6657355 (200)\ttotal: 19.1s\tremaining: 1m 15s\n",
      "250:\ttest: 0.6662365\tbest: 0.6662365 (250)\ttotal: 24.3s\tremaining: 1m 12s\n",
      "300:\ttest: 0.6664815\tbest: 0.6666013 (276)\ttotal: 29.8s\tremaining: 1m 9s\n",
      "350:\ttest: 0.6668205\tbest: 0.6668205 (350)\ttotal: 35.3s\tremaining: 1m 5s\n",
      "400:\ttest: 0.6667898\tbest: 0.6669106 (383)\ttotal: 40.8s\tremaining: 1m\n",
      "450:\ttest: 0.6665784\tbest: 0.6669106 (383)\ttotal: 45.9s\tremaining: 55.9s\n",
      "500:\ttest: 0.6665179\tbest: 0.6669106 (383)\ttotal: 51.3s\tremaining: 51.1s\n",
      "550:\ttest: 0.6661116\tbest: 0.6669106 (383)\ttotal: 56.4s\tremaining: 45.9s\n",
      "600:\ttest: 0.6661463\tbest: 0.6669106 (383)\ttotal: 1m 1s\tremaining: 40.7s\n",
      "650:\ttest: 0.6660366\tbest: 0.6669106 (383)\ttotal: 1m 6s\tremaining: 35.7s\n",
      "700:\ttest: 0.6661939\tbest: 0.6669106 (383)\ttotal: 1m 11s\tremaining: 30.6s\n",
      "750:\ttest: 0.6663121\tbest: 0.6669106 (383)\ttotal: 1m 16s\tremaining: 25.5s\n",
      "800:\ttest: 0.6663116\tbest: 0.6669106 (383)\ttotal: 1m 22s\tremaining: 20.4s\n",
      "850:\ttest: 0.6664016\tbest: 0.6669106 (383)\ttotal: 1m 27s\tremaining: 15.3s\n",
      "900:\ttest: 0.6663516\tbest: 0.6669106 (383)\ttotal: 1m 32s\tremaining: 10.2s\n",
      "950:\ttest: 0.6662278\tbest: 0.6669106 (383)\ttotal: 1m 37s\tremaining: 5.03s\n",
      "999:\ttest: 0.6662615\tbest: 0.6669106 (383)\ttotal: 1m 42s\tremaining: 0us\n",
      "\n",
      "bestTest = 0.6669105543\n",
      "bestIteration = 383\n",
      "\n",
      "Shrink model to first 384 iterations.\n",
      "--> use iteration [383] for proba\n",
      "--> valid label = 1 占比： 0.5385875\n",
      "0:\ttest: 0.6395947\tbest: 0.6395947 (0)\ttotal: 98.4ms\tremaining: 1m 38s\n",
      "50:\ttest: 0.6561876\tbest: 0.6561876 (50)\ttotal: 4.83s\tremaining: 1m 29s\n",
      "100:\ttest: 0.6586807\tbest: 0.6586807 (100)\ttotal: 9.57s\tremaining: 1m 25s\n",
      "150:\ttest: 0.6599090\tbest: 0.6599437 (147)\ttotal: 14.5s\tremaining: 1m 21s\n",
      "200:\ttest: 0.6609563\tbest: 0.6609748 (199)\ttotal: 19.7s\tremaining: 1m 18s\n",
      "250:\ttest: 0.6614956\tbest: 0.6615347 (247)\ttotal: 25s\tremaining: 1m 14s\n",
      "300:\ttest: 0.6615718\tbest: 0.6616272 (299)\ttotal: 30.4s\tremaining: 1m 10s\n",
      "350:\ttest: 0.6615975\tbest: 0.6616425 (345)\ttotal: 35.7s\tremaining: 1m 6s\n",
      "400:\ttest: 0.6617107\tbest: 0.6617660 (391)\ttotal: 41s\tremaining: 1m 1s\n",
      "450:\ttest: 0.6617636\tbest: 0.6618293 (445)\ttotal: 46.5s\tremaining: 56.6s\n",
      "500:\ttest: 0.6615207\tbest: 0.6618293 (445)\ttotal: 51.7s\tremaining: 51.5s\n",
      "550:\ttest: 0.6613071\tbest: 0.6618293 (445)\ttotal: 57s\tremaining: 46.4s\n",
      "600:\ttest: 0.6613607\tbest: 0.6618293 (445)\ttotal: 1m 2s\tremaining: 41.3s\n",
      "650:\ttest: 0.6614747\tbest: 0.6618293 (445)\ttotal: 1m 7s\tremaining: 36.1s\n",
      "700:\ttest: 0.6613535\tbest: 0.6618293 (445)\ttotal: 1m 12s\tremaining: 31s\n",
      "750:\ttest: 0.6613664\tbest: 0.6618293 (445)\ttotal: 1m 17s\tremaining: 25.8s\n",
      "800:\ttest: 0.6613936\tbest: 0.6618293 (445)\ttotal: 1m 22s\tremaining: 20.6s\n",
      "850:\ttest: 0.6611295\tbest: 0.6618293 (445)\ttotal: 1m 28s\tremaining: 15.5s\n",
      "900:\ttest: 0.6608371\tbest: 0.6618293 (445)\ttotal: 1m 33s\tremaining: 10.3s\n",
      "950:\ttest: 0.6608789\tbest: 0.6618293 (445)\ttotal: 1m 38s\tremaining: 5.09s\n",
      "999:\ttest: 0.6604775\tbest: 0.6618293 (445)\ttotal: 1m 44s\tremaining: 0us\n",
      "\n",
      "bestTest = 0.6618292572\n",
      "bestIteration = 445\n",
      "\n",
      "Shrink model to first 446 iterations.\n",
      "--> use iteration [445] for proba\n",
      "--> valid label = 1 占比： 0.543575\n",
      "0:\ttest: 0.6404244\tbest: 0.6404244 (0)\ttotal: 105ms\tremaining: 1m 44s\n",
      "50:\ttest: 0.6575718\tbest: 0.6579106 (49)\ttotal: 5.03s\tremaining: 1m 33s\n",
      "100:\ttest: 0.6603805\tbest: 0.6603805 (100)\ttotal: 9.88s\tremaining: 1m 27s\n",
      "150:\ttest: 0.6617147\tbest: 0.6617417 (149)\ttotal: 15.2s\tremaining: 1m 25s\n",
      "200:\ttest: 0.6624513\tbest: 0.6624513 (200)\ttotal: 20.5s\tremaining: 1m 21s\n",
      "250:\ttest: 0.6625534\tbest: 0.6626553 (247)\ttotal: 26.1s\tremaining: 1m 18s\n",
      "300:\ttest: 0.6625521\tbest: 0.6627634 (270)\ttotal: 31.9s\tremaining: 1m 14s\n",
      "350:\ttest: 0.6621873\tbest: 0.6627634 (270)\ttotal: 37.5s\tremaining: 1m 9s\n",
      "400:\ttest: 0.6620762\tbest: 0.6627634 (270)\ttotal: 43.3s\tremaining: 1m 4s\n",
      "450:\ttest: 0.6619604\tbest: 0.6627634 (270)\ttotal: 48.7s\tremaining: 59.3s\n",
      "500:\ttest: 0.6618272\tbest: 0.6627634 (270)\ttotal: 53.9s\tremaining: 53.7s\n",
      "550:\ttest: 0.6620764\tbest: 0.6627634 (270)\ttotal: 59.1s\tremaining: 48.2s\n",
      "600:\ttest: 0.6620852\tbest: 0.6627634 (270)\ttotal: 1m 4s\tremaining: 42.8s\n",
      "650:\ttest: 0.6618831\tbest: 0.6627634 (270)\ttotal: 1m 9s\tremaining: 37.3s\n",
      "700:\ttest: 0.6617237\tbest: 0.6627634 (270)\ttotal: 1m 14s\tremaining: 31.9s\n",
      "750:\ttest: 0.6616193\tbest: 0.6627634 (270)\ttotal: 1m 19s\tremaining: 26.5s\n",
      "800:\ttest: 0.6615717\tbest: 0.6627634 (270)\ttotal: 1m 25s\tremaining: 21.1s\n",
      "850:\ttest: 0.6615445\tbest: 0.6627634 (270)\ttotal: 1m 30s\tremaining: 15.8s\n",
      "900:\ttest: 0.6615252\tbest: 0.6627634 (270)\ttotal: 1m 35s\tremaining: 10.5s\n",
      "950:\ttest: 0.6614292\tbest: 0.6627634 (270)\ttotal: 1m 40s\tremaining: 5.18s\n",
      "999:\ttest: 0.6611646\tbest: 0.6627634 (270)\ttotal: 1m 45s\tremaining: 0us\n",
      "\n",
      "bestTest = 0.6627634309\n",
      "bestIteration = 270\n",
      "\n",
      "Shrink model to first 271 iterations.\n",
      "--> use iteration [270] for proba\n",
      "--> valid label = 1 占比： 0.542475\n",
      "0:\ttest: 0.6398454\tbest: 0.6398454 (0)\ttotal: 101ms\tremaining: 1m 40s\n",
      "50:\ttest: 0.6579109\tbest: 0.6579836 (49)\ttotal: 4.84s\tremaining: 1m 30s\n",
      "100:\ttest: 0.6607207\tbest: 0.6610043 (91)\ttotal: 9.25s\tremaining: 1m 22s\n",
      "150:\ttest: 0.6622499\tbest: 0.6623030 (147)\ttotal: 14.4s\tremaining: 1m 20s\n",
      "200:\ttest: 0.6629740\tbest: 0.6629997 (198)\ttotal: 19.6s\tremaining: 1m 17s\n",
      "250:\ttest: 0.6635969\tbest: 0.6636198 (249)\ttotal: 25.1s\tremaining: 1m 14s\n",
      "300:\ttest: 0.6636840\tbest: 0.6638412 (283)\ttotal: 30.6s\tremaining: 1m 11s\n",
      "350:\ttest: 0.6637422\tbest: 0.6639374 (333)\ttotal: 36.1s\tremaining: 1m 6s\n",
      "400:\ttest: 0.6637017\tbest: 0.6639374 (333)\ttotal: 41.4s\tremaining: 1m 1s\n",
      "450:\ttest: 0.6638504\tbest: 0.6639374 (333)\ttotal: 46.9s\tremaining: 57.1s\n",
      "500:\ttest: 0.6639790\tbest: 0.6640535 (489)\ttotal: 52.1s\tremaining: 51.9s\n",
      "550:\ttest: 0.6641602\tbest: 0.6641726 (542)\ttotal: 57.4s\tremaining: 46.8s\n",
      "600:\ttest: 0.6640145\tbest: 0.6641838 (560)\ttotal: 1m 2s\tremaining: 41.6s\n",
      "650:\ttest: 0.6640645\tbest: 0.6641838 (560)\ttotal: 1m 7s\tremaining: 36.4s\n",
      "700:\ttest: 0.6638109\tbest: 0.6641838 (560)\ttotal: 1m 13s\tremaining: 31.1s\n",
      "750:\ttest: 0.6637463\tbest: 0.6641838 (560)\ttotal: 1m 18s\tremaining: 25.9s\n",
      "800:\ttest: 0.6638308\tbest: 0.6641838 (560)\ttotal: 1m 23s\tremaining: 20.7s\n",
      "850:\ttest: 0.6637899\tbest: 0.6641838 (560)\ttotal: 1m 28s\tremaining: 15.5s\n",
      "900:\ttest: 0.6635911\tbest: 0.6641838 (560)\ttotal: 1m 33s\tremaining: 10.3s\n",
      "950:\ttest: 0.6632873\tbest: 0.6641838 (560)\ttotal: 1m 39s\tremaining: 5.11s\n",
      "999:\ttest: 0.6632807\tbest: 0.6641838 (560)\ttotal: 1m 44s\tremaining: 0us\n",
      "\n",
      "bestTest = 0.6641838002\n",
      "bestIteration = 560\n",
      "\n",
      "Shrink model to first 561 iterations.\n",
      "--> use iteration [560] for proba\n",
      "--> valid label = 1 占比： 0.5391375\n",
      "0:\ttest: 0.6396320\tbest: 0.6396320 (0)\ttotal: 99.7ms\tremaining: 1m 39s\n",
      "50:\ttest: 0.6557566\tbest: 0.6563203 (38)\ttotal: 4.79s\tremaining: 1m 29s\n",
      "100:\ttest: 0.6580496\tbest: 0.6580894 (88)\ttotal: 9.35s\tremaining: 1m 23s\n",
      "150:\ttest: 0.6597712\tbest: 0.6598540 (143)\ttotal: 14.1s\tremaining: 1m 19s\n",
      "200:\ttest: 0.6608617\tbest: 0.6608925 (199)\ttotal: 19.4s\tremaining: 1m 17s\n",
      "250:\ttest: 0.6610228\tbest: 0.6610597 (215)\ttotal: 24.6s\tremaining: 1m 13s\n",
      "300:\ttest: 0.6612071\tbest: 0.6614207 (284)\ttotal: 30.1s\tremaining: 1m 9s\n",
      "350:\ttest: 0.6610133\tbest: 0.6614207 (284)\ttotal: 35.7s\tremaining: 1m 6s\n",
      "400:\ttest: 0.6611661\tbest: 0.6614207 (284)\ttotal: 41.2s\tremaining: 1m 1s\n",
      "450:\ttest: 0.6613822\tbest: 0.6614207 (284)\ttotal: 46.3s\tremaining: 56.4s\n",
      "500:\ttest: 0.6615067\tbest: 0.6615805 (488)\ttotal: 51.7s\tremaining: 51.5s\n",
      "550:\ttest: 0.6614635\tbest: 0.6615805 (488)\ttotal: 56.9s\tremaining: 46.4s\n",
      "600:\ttest: 0.6614628\tbest: 0.6615846 (582)\ttotal: 1m 2s\tremaining: 41.3s\n",
      "650:\ttest: 0.6612158\tbest: 0.6615846 (582)\ttotal: 1m 7s\tremaining: 36.2s\n",
      "700:\ttest: 0.6609774\tbest: 0.6615846 (582)\ttotal: 1m 13s\tremaining: 31.3s\n",
      "750:\ttest: 0.6607899\tbest: 0.6615846 (582)\ttotal: 1m 18s\tremaining: 26.1s\n",
      "800:\ttest: 0.6606720\tbest: 0.6615846 (582)\ttotal: 1m 24s\tremaining: 20.9s\n",
      "850:\ttest: 0.6606525\tbest: 0.6615846 (582)\ttotal: 1m 29s\tremaining: 15.6s\n",
      "900:\ttest: 0.6607817\tbest: 0.6615846 (582)\ttotal: 1m 34s\tremaining: 10.4s\n",
      "950:\ttest: 0.6612327\tbest: 0.6615846 (582)\ttotal: 1m 40s\tremaining: 5.16s\n",
      "999:\ttest: 0.6612781\tbest: 0.6615846 (582)\ttotal: 1m 45s\tremaining: 0us\n",
      "\n",
      "bestTest = 0.6615845797\n",
      "bestIteration = 582\n",
      "\n",
      "Shrink model to first 583 iterations.\n",
      "--> use iteration [582] for proba\n",
      "--> valid label = 1 占比： 0.5366125\n",
      "dev auc:  0.665705758933009\n",
      "Baseline metrics\n",
      "      fold  binary_logloss\n",
      "0        0        0.664439\n",
      "1        1        0.665330\n",
      "2        2        0.665027\n",
      "3        3        0.663518\n",
      "4        4        0.664752\n",
      "5  overall        0.664613\n"
     ]
    }
   ],
   "source": [
    "config = {\n",
    "    **config,\n",
    "    'random_seed': 42, # 42, 50\n",
    "    'num_leaves': 31,\n",
    "    'max_depth': 6,  # -1, 6\n",
    "    'subsample_for_bin': 200000,\n",
    "    'colsample_bytree': 0.9,\n",
    "    'n_estimators': 1000,\n",
    "    'learning_rate': 0.05,\n",
    "    'reg_alpha': 0.1,\n",
    "    'reg_lambda': 0.1,  # 50\n",
    "    'subsample': 1.0,\n",
    "    'model_name': 'cbt',  # lgb, cbt, xgb\n",
    "    'boosting_type': 'gbdt',\n",
    "    'class_weight': 'balanced',\n",
    "    'n_folds': 5,\n",
    "    \n",
    "    'verbose': 50, # 每隔多少步输出\n",
    "    'exec_time': time.strftime(\"%Y.%m.%d-%H.%M.%S\", time.localtime()),\n",
    "    \n",
    "    'use_wandb': False,\n",
    "    'run_sweeps': False,\n",
    "}\n",
    "\n",
    "# config.update({\n",
    "#     # sweep 调参的结果\n",
    "#     'colsample_bytree': 0.5441,\n",
    "#     'learning_rate': 0.09151,\n",
    "#     'num_leaves': 63,\n",
    "#     'random_seed': 751,\n",
    "#     'reg_alpha': 0.155,\n",
    "#     'reg_lambda': 0.1032,\n",
    "#     'subsample': 0.8299,\n",
    "# })\n",
    "\n",
    "# 构建训练集与测试集\n",
    "all_features = all_features.reset_index(drop=True)\n",
    "train_features = all_features[all_features['test_flag'] == 0]\n",
    "dev_features = all_features[all_features['test_flag'] == 1]\n",
    "mid_features = all_features[all_features['test_flag'] == 2]\n",
    "test_features = all_features[all_features['test_flag'] == 3]\n",
    "\n",
    "if config['run_sweeps']:\n",
    "    # 执行超参调优\n",
    "    config['use_wandb'] = False\n",
    "    sweep_id = wandb.sweep(sweep_config,\n",
    "                           entity='imqxms',\n",
    "                           project='AntGraph')\n",
    "    def sweep_train():\n",
    "        with wandb.init() as run:\n",
    "            config.update(wandb.config)\n",
    "            \n",
    "            submission, fi, metrics, all_model, out_of_fold = sigle_model(train_features, test_features, args=config, n_folds = config['n_folds'])\n",
    "            print('Baseline metrics')\n",
    "            print(metrics)\n",
    "            \n",
    "            # 平均每一折中每个 step 的效果\n",
    "            res_dict = dict()\n",
    "            for i in range(config['n_folds']):\n",
    "                for key in all_model[i].evals_result_['valid'].keys():\n",
    "                    if 'valid_{}'.format(key) not in res_dict:\n",
    "                        res_dict['valid_{}'.format(key)] = 0.0\n",
    "                    res_dict['valid_{}'.format(key)] += np.array(all_model[i].evals_result_['valid'][key]) * 1.0 / config['n_folds']\n",
    "            \n",
    "            # 将平均后的效果 log 进 wandb\n",
    "            for i in range(len(res_dict['valid_binary_logloss'])):\n",
    "                wandb.log({key:res_dict[key][i] for key in res_dict.keys()})\n",
    "            \n",
    "            if sweep_config['metric']['goal'] == 'maximize':\n",
    "                # 找出最佳效果的 step，并将该 step 的结果 log 在 wandb 中（为了 sweep 调参）\n",
    "                best_idx = res_dict[sweep_config['metric']['name']].argmax()\n",
    "            else:\n",
    "                best_idx = res_dict[sweep_config['metric']['name']].argmin()\n",
    "            wandb.log({i: res_dict[i][best_idx] for i in res_dict.keys()})  # 将最佳效果 log 一下\n",
    "                                \n",
    "    wandb.agent(sweep_id, function=sweep_train, count=30)\n",
    "else:\n",
    "    dev_submission, submission, fi, metrics, all_model, out_of_fold = sigle_model(train_features, dev_features, test_features, args=config, n_folds = config['n_folds'])\n",
    "    print('Baseline metrics')\n",
    "    print(metrics)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "id": "240d7cb1-9818-4e4c-a223-dc1b52ec11a6",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 存储最终结果\n",
    "import pickle\n",
    "\n",
    "\n",
    "class PickleUtils:\n",
    "    @staticmethod\n",
    "    def load_pickle(filepath):\n",
    "        res = None\n",
    "        with open(filepath, 'rb') as f:\n",
    "            res = pickle.load(f)\n",
    "        return res\n",
    "\n",
    "    @staticmethod\n",
    "    def save_pickle(res, filepath):\n",
    "        with open(filepath, 'wb') as f:\n",
    "            pickle.dump(res, f)\n",
    "\n",
    "# 计算训练集与验证集的效果\n",
    "train_auc_score = roc_auc_score(all_features.query('test_flag == 0 and label != -1')['label'], out_of_fold)\n",
    "dev_auc_score = roc_auc_score(dev_features['label'], dev_submission['dev_label'])\n",
    "\n",
    "result_save_path_template = '../submits/{}/{}_result_{}_train({:.4f})_dev({:.4f}).csv'\n",
    "# 训练集预测结果存储路径\n",
    "train_result_save_path = result_save_path_template.format(config[\"dataset\"],\n",
    "                                                          config[\"exec_time\"],\n",
    "                                                          'train_submit',\n",
    "                                                          train_auc_score,\n",
    "                                                          dev_auc_score,)\n",
    "# data_a/b_initial 预测结果存储路径\n",
    "dev_result_save_path = result_save_path_template.format(config[\"dataset\"],\n",
    "                                                        config[\"exec_time\"],\n",
    "                                                        'dev_submit',\n",
    "                                                        train_auc_score,\n",
    "                                                        dev_auc_score,)\n",
    "# 中期测试集预测结果存储路径\n",
    "mid_result_save_path = result_save_path_template.format(config[\"dataset\"],\n",
    "                                                        config[\"exec_time\"],\n",
    "                                                        'mid_submit',\n",
    "                                                        train_auc_score,\n",
    "                                                        dev_auc_score,)\n",
    "# 最终测试集预测结果存储路径\n",
    "test_result_save_path = result_save_path_template.format(config[\"dataset\"],\n",
    "                                                         config[\"exec_time\"],\n",
    "                                                         'test_submit',\n",
    "                                                         train_auc_score,\n",
    "                                                         dev_auc_score,)\n",
    "\n",
    "PickleUtils.save_pickle((fi, all_model), train_result_save_path + '.pickle')\n",
    "pd.DataFrame(out_of_fold).to_csv(train_result_save_path, header=False, index=False)\n",
    "dev_submission[['dev_label']].to_csv(dev_result_save_path, header=False, index=False)\n",
    "submission[['test_label']].to_csv(test_result_save_path, header=False, index=False)\n",
    "\n",
    "train_mid_pred = np.mean([model_predict_proba(i, mid_features.drop(columns=['label', 'test_flag']), 0, config)[:, 1] for i in all_model], axis=0)\n",
    "pd.DataFrame({'pred_score': train_mid_pred}).to_csv(mid_result_save_path, header=False, index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "id": "4230e4ee-d68b-4bf4-816f-d51d8f470de9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x1728 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['two_hop_count_list', 'sparse:group_val_(src_id,dst_id)_edge_type_nunique', 'sparse:group_val_(src_id,dst_id)_count', 'dot_product', 'sparse:group_val_(src_id)_count_x', 'dst_id', 'src_id', 'sparse:group_val_(src_id)_count_y', 'sparse:group_val_(src_id)_dst_id_nunique_y', 'cos_sim', 'edge_type', 'sparse:group_val_(src_id)_dst_id_nunique_x', 'sparse:group_val_(src_id)_edge_type_nunique_y', 'sparse:group_val_(src_id)_edge_type_nunique_x', 'feat_1', 'feat_0', 'feat_0_y', 'feat_0_x', 'feat_5_x', 'feat_5_y', '172_x', '172_y', 'feat_2', '96_x', '96_y', '73_y', '44_y', '176_y', '18_y', '73_x', '44_x', '176_x', '18_x', '62_y', '1_y', '153_x', '196_y', '49_y', '193_x', '142_y', '193_y', '92_y', '10_x', '196_x', '23_y', '168_y', '187_x', '49_x', '21_y', '62_x', '191_x', '115_x', '115_y', '100_x', '4_x', '192_x', '97_x', '98_x', '33_y', '88_y', '4_y', '66_x', '70_x', '21_x', '74_x', '1_x', '191_y', '102_x', '106_y', '98_y', '188_x', '29_x', '20_y', '187_y', '167_x', '190_y', '2_x', '139_x', '53_y', '129_y', '92_x', '185_x', '68_y', '76_y', '161_y', '151_x', '12_x', '95_y', '43_y', '133_x', '185_y', '57_x', '197_y', '60_x', '167_y', '106_x', '52_y', '69_x', '84_x', '60_y']\n"
     ]
    }
   ],
   "source": [
    "def plot_feature_importances(df, top_k=15):\n",
    "    \n",
    "    df = df.sort_values('importance', ascending = False).reset_index()\n",
    "\n",
    "    df['importance_normalized'] = df['importance'] / df['importance'].sum()\n",
    "\n",
    "    plt.figure(figsize = (10, 24))\n",
    "    ax = plt.subplot()\n",
    "\n",
    "    ax.barh(list(reversed(list(df.index[:top_k]))), \n",
    "            df['importance_normalized'].head(top_k), \n",
    "            align = 'center', edgecolor = 'k')\n",
    "\n",
    "    ax.set_yticks(list(reversed(list(df.index[:top_k]))))\n",
    "    ax.set_yticklabels(df['feature'].head(top_k))\n",
    "\n",
    "    plt.xlabel('Normalized Importance'); plt.title('Feature Importances')\n",
    "    plt.show()\n",
    "    \n",
    "    return df\n",
    "\n",
    "top_k = 100\n",
    "fi_sorted = plot_feature_importances(fi, top_k=top_k)\n",
    "important_features = [x for x in fi_sorted['feature'].iloc[:top_k]]\n",
    "print(important_features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ce4510dc",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}