graph_credit.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "             ID CODE_GENDER FLAG_OWN_CAR FLAG_OWN_REALTY  CNT_CHILDREN  \\\n",
      "284481  6127905           F            N               Y             0   \n",
      "203727  6508890           F            N               Y             1   \n",
      "233631  5996425           M            N               Y             1   \n",
      "7505    5028725           M            N               Y             0   \n",
      "192047  5878879           F            Y               Y             0   \n",
      "\n",
      "        AMT_INCOME_TOTAL  NAME_INCOME_TYPE NAME_EDUCATION_TYPE  \\\n",
      "284481           90000.0  Low sesffcondary   secondary special   \n",
      "203727           81000.0       High edtion   secondary special   \n",
      "233631          189000.0    High education           Secondary   \n",
      "7505            360000.0        Seconfdary           Secondary   \n",
      "192047          112500.0     Low secondary   secondary special   \n",
      "\n",
      "       NAME_FAMILY_STATUS      NAME_HOUSING_TYPE  DAYS_BIRTH  DAYS_EMPLOYED  \\\n",
      "284481          Separated  secondasfsry spfecial      -23010         365243   \n",
      "203727            Married             Seconfdary      -16812          -1094   \n",
      "233631            Married    seconsfdary special      -13516          -1312   \n",
      "7505              Married    seconsfdary special      -12483          -4039   \n",
      "192047            Married              Secondary      -21220         365243   \n",
      "\n",
      "        FLAG_MOBIL  FLAG_WORK_PHONE  FLAG_PHONE  FLAG_EMAIL  CNT_FAM_MEMBERS  \n",
      "284481           1                0           0           0              1.0  \n",
      "203727           1                1           0           0              3.0  \n",
      "233631           1                0           0           0              3.0  \n",
      "7505             1                0           0           0              2.0  \n",
      "192047           1                0           0           0              2.0  \n",
      "Non-numeric columns: Index(['CODE_GENDER', 'FLAG_OWN_CAR', 'FLAG_OWN_REALTY', 'NAME_INCOME_TYPE',\n",
      "       'NAME_EDUCATION_TYPE', 'NAME_FAMILY_STATUS', 'NAME_HOUSING_TYPE'],\n",
      "      dtype='object')\n",
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Index: 87711 entries, 284481 to 195462\n",
      "Data columns (total 17 columns):\n",
      " #   Column               Non-Null Count  Dtype  \n",
      "---  ------               --------------  -----  \n",
      " 0   ID                   87711 non-null  int64  \n",
      " 1   CODE_GENDER          87711 non-null  int64  \n",
      " 2   FLAG_OWN_CAR         87711 non-null  int64  \n",
      " 3   FLAG_OWN_REALTY      87711 non-null  int64  \n",
      " 4   CNT_CHILDREN         87711 non-null  int64  \n",
      " 5   AMT_INCOME_TOTAL     87711 non-null  float64\n",
      " 6   NAME_INCOME_TYPE     87711 non-null  int64  \n",
      " 7   NAME_EDUCATION_TYPE  87711 non-null  int64  \n",
      " 8   NAME_FAMILY_STATUS   87711 non-null  int64  \n",
      " 9   NAME_HOUSING_TYPE    87711 non-null  int64  \n",
      " 10  DAYS_BIRTH           87711 non-null  int64  \n",
      " 11  DAYS_EMPLOYED        87711 non-null  int64  \n",
      " 12  FLAG_MOBIL           87711 non-null  int64  \n",
      " 13  FLAG_WORK_PHONE      87711 non-null  int64  \n",
      " 14  FLAG_PHONE           87711 non-null  int64  \n",
      " 15  FLAG_EMAIL           87711 non-null  int64  \n",
      " 16  CNT_FAM_MEMBERS      87711 non-null  float64\n",
      "dtypes: float64(2), int64(15)\n",
      "memory usage: 12.0 MB\n",
      "Data types after encoding:\n",
      " None\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "from sklearn.preprocessing import LabelEncoder, StandardScaler, MinMaxScaler\n",
    "import networkx as nx\n",
    "import torch\n",
    "from torch_geometric.data import Data\n",
    "from torch_geometric.utils import from_networkx\n",
    "# 加载数据\n",
    "file_path_clean = '/home/sdong/data/credit_card/application_record_string.csv'\n",
    "#file_path_clean = '/home/sdong/data/chicago_bicycle/data_pr_raw.csv'\n",
    "data_df = pd.read_csv(file_path_clean)\n",
    "\n",
    "# 随机选择20分之一个数据集（即5%）\n",
    "data_df = data_df.sample(frac=0.2, random_state=42)\n",
    "\n",
    "print(data_df.head())\n",
    "# 使用 fillna() 方法替换所有的 NaN 值为 0\n",
    "data_df.fillna(0, inplace=True)\n",
    "# 检查非数值列\n",
    "non_numeric_cols = data_df.select_dtypes(include=['object']).columns\n",
    "print(\"Non-numeric columns:\", non_numeric_cols)\n",
    "\n",
    "# 将非数值列转换为数值类型（使用标签编码）\n",
    "label_encoders = {}\n",
    "for col in non_numeric_cols:\n",
    "    le = LabelEncoder()\n",
    "    data_df[col] = le.fit_transform(data_df[col].astype(str))\n",
    "    label_encoders[col] = le\n",
    "\n",
    "\n",
    "\n",
    "# 确保所有特征列都是数值类型\n",
    "print(\"Data types after encoding:\\n\", data_df.info())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['ID', 'CODE_GENDER', 'FLAG_OWN_CAR', 'FLAG_OWN_REALTY', 'CNT_CHILDREN',\n",
       "       'AMT_INCOME_TOTAL', 'NAME_INCOME_TYPE', 'NAME_EDUCATION_TYPE',\n",
       "       'NAME_FAMILY_STATUS', 'NAME_HOUSING_TYPE', 'DAYS_BIRTH',\n",
       "       'DAYS_EMPLOYED', 'FLAG_MOBIL', 'FLAG_WORK_PHONE', 'FLAG_PHONE',\n",
       "       'FLAG_EMAIL', 'CNT_FAM_MEMBERS'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_df.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 标准化数值特征\n",
    "scaler = MinMaxScaler()\n",
    "data_scaled = scaler.fit_transform(data_df)\n",
    "\n",
    "# 将标准化后的数据转换为torch张量\n",
    "x = torch.tensor(data_scaled, dtype=torch.float)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Data(edge_index=[2, 26], num_nodes=17, x=[87711, 17])\n"
     ]
    }
   ],
   "source": [
    "\n",
    "\n",
    "# 示例表格数据特征\n",
    "columns = data_df.columns\n",
    "\n",
    "# 定义特征之间的关系\n",
    "relations = [\n",
    "    ('ID', 'CODE_GENDER'),\n",
    "    ('ID', 'DAYS_BIRTH'),\n",
    "    ('ID', 'FLAG_MOBIL'),\n",
    "    ('CODE_GENDER', 'FLAG_OWN_CAR'),\n",
    "    ('CODE_GENDER', 'FLAG_OWN_REALTY'),\n",
    "    ('CODE_GENDER', 'CNT_CHILDREN'),\n",
    "    ('FLAG_OWN_CAR', 'FLAG_OWN_REALTY'),\n",
    "    ('CNT_CHILDREN', 'NAME_FAMILY_STATUS'),\n",
    "    ('CNT_CHILDREN', 'CNT_FAM_MEMBERS'),\n",
    "    ('NAME_FAMILY_STATUS', 'CNT_FAM_MEMBERS'),\n",
    "    ('AMT_INCOME_TOTAL', 'NAME_INCOME_TYPE'),\n",
    "    ('AMT_INCOME_TOTAL', 'DAYS_EMPLOYED'),\n",
    "    ('NAME_INCOME_TYPE', 'DAYS_EMPLOYED'),\n",
    "    ('NAME_INCOME_TYPE', 'NAME_HOUSING_TYPE'),\n",
    "    ('NAME_HOUSING_TYPE', 'FLAG_OWN_REALTY'),\n",
    "    ('NAME_HOUSING_TYPE', 'AMT_INCOME_TOTAL'),\n",
    "    ('NAME_EDUCATION_TYPE', 'AMT_INCOME_TOTAL'),\n",
    "    ('NAME_EDUCATION_TYPE', 'NAME_INCOME_TYPE'),\n",
    "    ('NAME_EDUCATION_TYPE', 'NAME_HOUSING_TYPE'),\n",
    "    ('DAYS_BIRTH', 'DAYS_EMPLOYED'),\n",
    "    ('FLAG_MOBIL', 'FLAG_WORK_PHONE'),\n",
    "    ('FLAG_MOBIL', 'FLAG_PHONE'),\n",
    "    ('FLAG_MOBIL', 'FLAG_EMAIL'),\n",
    "    ('FLAG_WORK_PHONE', 'FLAG_PHONE'),\n",
    "    ('FLAG_WORK_PHONE', 'FLAG_EMAIL'),\n",
    "    ('FLAG_PHONE', 'FLAG_EMAIL')\n",
    "]\n",
    "\n",
    "\n",
    "# 创建空的无向图\n",
    "G = nx.Graph()\n",
    "\n",
    "# 添加节点（每个特征作为一个节点）\n",
    "for col in columns:\n",
    "    G.add_node(col)\n",
    "\n",
    "# 添加边（根据特征之间的关系）\n",
    "for relation in relations:\n",
    "    G.add_edge(relation[0], relation[1])\n",
    "\n",
    "# 将NetworkX图转换为PyTorch Geometric图\n",
    "data = from_networkx(G)\n",
    "\n",
    "# 添加节点特征\n",
    "data.x = x\n",
    "\n",
    "# 映射特征列到索引\n",
    "feature_to_index = {col: i for i, col in enumerate(columns)}\n",
    "\n",
    "# 映射关系到索引\n",
    "edges = [(feature_to_index[src], feature_to_index[dst]) for src, dst in relations]\n",
    "\n",
    "# 添加边\n",
    "edge_index = torch.tensor(edges, dtype=torch.long).t().contiguous()\n",
    "data.edge_index = edge_index\n",
    "\n",
    "\n",
    "# 确定设备（GPU 优先）\n",
    "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
    "\n",
    "# 将数据移动到设备\n",
    "data = data.to(device)\n",
    "\n",
    "print(data)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GCN(\n",
      "  (conv1): GCNConv(17, 128)\n",
      "  (conv2): GCNConv(128, 128)\n",
      "  (fc): Linear(in_features=128, out_features=17, bias=True)\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "from torch_geometric.nn import GCNConv\n",
    "\n",
    "class GCN(nn.Module):\n",
    "    def __init__(self, num_features, hidden_channels, output_channels):\n",
    "        super(GCN, self).__init__()\n",
    "        self.conv1 = GCNConv(num_features, hidden_channels)\n",
    "        self.conv2 = GCNConv(hidden_channels, hidden_channels)\n",
    "        self.fc = nn.Linear(hidden_channels, output_channels)\n",
    "\n",
    "    def forward(self, data):\n",
    "        x, edge_index = data.x, data.edge_index\n",
    "        x = self.conv1(x, edge_index)\n",
    "        x = F.relu(x)\n",
    "        x = self.conv2(x, edge_index)\n",
    "        x = F.relu(x)\n",
    "        x = self.fc(x)\n",
    "        return x\n",
    "\n",
    "# 初始化模型\n",
    "num_features = data.num_features\n",
    "hidden_channels = 128\n",
    "output_channels = num_features  # 确保输出维度与输入维度一致\n",
    "model = GCN(num_features, hidden_channels, output_channels).to(device)\n",
    "\n",
    "# 打印模型结构\n",
    "print(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 10, Loss: 0.0438\n",
      "Epoch 20, Loss: 0.0140\n",
      "Epoch 30, Loss: 0.0065\n",
      "Epoch 40, Loss: 0.0035\n",
      "Epoch 50, Loss: 0.0019\n",
      "Epoch 60, Loss: 0.0012\n",
      "Epoch 70, Loss: 0.0008\n",
      "Epoch 80, Loss: 0.0006\n",
      "Epoch 90, Loss: 0.0009\n",
      "Epoch 100, Loss: 0.0005\n",
      "Epoch 110, Loss: 0.0004\n",
      "Epoch 120, Loss: 0.0004\n",
      "Epoch 130, Loss: 0.0004\n",
      "Epoch 140, Loss: 0.0003\n",
      "Epoch 150, Loss: 0.0003\n",
      "Epoch 160, Loss: 0.0003\n",
      "Epoch 170, Loss: 0.0024\n",
      "Epoch 180, Loss: 0.0004\n",
      "Epoch 190, Loss: 0.0003\n",
      "Epoch 200, Loss: 0.0002\n"
     ]
    }
   ],
   "source": [
    "from torch_geometric.loader import DataLoader\n",
    "\n",
    "# 创建数据加载器\n",
    "loader = DataLoader([data], batch_size=1280, shuffle=True)\n",
    "\n",
    "# 定义优化器和损失函数\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=0.01)\n",
    "criterion = torch.nn.MSELoss()\n",
    "\n",
    "# 训练模型\n",
    "def train():\n",
    "    model.train()\n",
    "    total_loss = 0\n",
    "    for batch in loader:\n",
    "        batch = batch.to(device)\n",
    "        optimizer.zero_grad()\n",
    "        out = model(batch)\n",
    "        loss = criterion(out, batch.x)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        total_loss += loss.item()\n",
    "    return total_loss / len(loader)\n",
    "\n",
    "# 训练循环\n",
    "for epoch in range(1, 201):\n",
    "    loss = train()\n",
    "    if epoch % 10 == 0:\n",
    "        print(f'Epoch {epoch}, Loss: {loss:.4f}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[ 4.6121e-01,  3.0720e-02,  5.7163e-02,  ..., -7.7568e-02,\n",
      "         -1.3338e-02,  1.5783e-02],\n",
      "        [ 6.1731e-01, -5.2372e-03,  1.6966e-02,  ..., -2.2888e-02,\n",
      "          6.1607e-02,  1.5378e-01],\n",
      "        [ 4.3634e-01,  1.2728e+00, -1.0398e-02,  ...,  2.5052e-02,\n",
      "          2.0377e-02,  1.7452e-01],\n",
      "        ...,\n",
      "        [ 4.9952e-01,  1.2684e+00,  1.2374e+00,  ..., -5.0498e-03,\n",
      "         -8.9565e-03,  2.7761e-01],\n",
      "        [ 3.5327e-01,  1.2594e+00,  1.2185e+00,  ...,  1.2464e+00,\n",
      "          1.1206e-02, -3.1460e-02],\n",
      "        [ 3.8206e-01,  9.0300e-04, -1.1459e-02,  ...,  1.2050e+00,\n",
      "          2.0539e-02,  2.4681e-01]], device='cuda:0')\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "from torch_geometric.data import Data\n",
    "from torch_geometric.utils import from_networkx\n",
    "\n",
    "def aggregate_instance_embeddings(data_scaled, G, model):\n",
    "    model.eval()\n",
    "    instance_embeddings_list = []\n",
    "\n",
    "    for i in range(len(data_scaled)):\n",
    "        # 创建子图\n",
    "        subgraph = G.copy()\n",
    "\n",
    "        # 创建节点特征张量\n",
    "        subgraph_data = data_scaled[i]\n",
    "\n",
    "        # 确保 x 的形状是 [num_nodes, num_features]，即 [1, 23]\n",
    "        node_features = torch.tensor(subgraph_data, dtype=torch.float).view(1, -1).repeat(len(columns), 1).to(device)\n",
    "\n",
    "        # 将 NetworkX 子图转换为 PyTorch Geometric 图\n",
    "        subgraph_data = from_networkx(subgraph)\n",
    "\n",
    "        # 更新子图的节点特征\n",
    "        subgraph_data.x = node_features\n",
    "\n",
    "        # 映射特征列到索引\n",
    "        feature_to_index = {col: idx for idx, col in enumerate(columns)}\n",
    "\n",
    "        # 映射关系到索引\n",
    "        edges = [(feature_to_index[src], feature_to_index[dst]) for src, dst in relations]\n",
    "\n",
    "        # 添加边\n",
    "        edge_index = torch.tensor(edges, dtype=torch.long).t().contiguous()\n",
    "        subgraph_data.edge_index = edge_index.to(device)\n",
    "\n",
    "        # 获取特征嵌入\n",
    "        with torch.no_grad():\n",
    "            node_embeddings = model(subgraph_data)\n",
    "\n",
    "        # 聚合节点嵌入到实例嵌入\n",
    "        instance_embedding = node_embeddings.mean(dim=0, keepdim=True)\n",
    "        instance_embeddings_list.append(instance_embedding)\n",
    "\n",
    "    instance_embeddings = torch.cat(instance_embeddings_list, dim=0)\n",
    "    return instance_embeddings\n",
    "\n",
    "# 获取每个实例的特征嵌入\n",
    "instance_embeddings = aggregate_instance_embeddings(data_scaled, G, model)\n",
    "print(instance_embeddings)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([87711, 17])"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "instance_embeddings.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "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>10</th>\n",
       "      <th>11</th>\n",
       "      <th>12</th>\n",
       "      <th>13</th>\n",
       "      <th>14</th>\n",
       "      <th>15</th>\n",
       "      <th>16</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.461209</td>\n",
       "      <td>0.030720</td>\n",
       "      <td>0.057163</td>\n",
       "      <td>1.299550</td>\n",
       "      <td>0.016376</td>\n",
       "      <td>0.032741</td>\n",
       "      <td>0.440937</td>\n",
       "      <td>1.216574</td>\n",
       "      <td>0.690130</td>\n",
       "      <td>1.174628</td>\n",
       "      <td>0.059692</td>\n",
       "      <td>1.255372</td>\n",
       "      <td>0.003491</td>\n",
       "      <td>-0.032661</td>\n",
       "      <td>-0.077568</td>\n",
       "      <td>-0.013338</td>\n",
       "      <td>0.015783</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.617313</td>\n",
       "      <td>-0.005237</td>\n",
       "      <td>0.016966</td>\n",
       "      <td>1.252040</td>\n",
       "      <td>0.123216</td>\n",
       "      <td>0.027827</td>\n",
       "      <td>0.065470</td>\n",
       "      <td>1.233238</td>\n",
       "      <td>0.325361</td>\n",
       "      <td>1.069075</td>\n",
       "      <td>0.561540</td>\n",
       "      <td>0.061380</td>\n",
       "      <td>-0.013161</td>\n",
       "      <td>1.271114</td>\n",
       "      <td>-0.022888</td>\n",
       "      <td>0.061607</td>\n",
       "      <td>0.153778</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.436341</td>\n",
       "      <td>1.272836</td>\n",
       "      <td>-0.010398</td>\n",
       "      <td>1.217566</td>\n",
       "      <td>0.127716</td>\n",
       "      <td>0.046084</td>\n",
       "      <td>0.144033</td>\n",
       "      <td>1.051804</td>\n",
       "      <td>0.297363</td>\n",
       "      <td>1.276768</td>\n",
       "      <td>0.834105</td>\n",
       "      <td>0.036984</td>\n",
       "      <td>-0.011628</td>\n",
       "      <td>-0.003888</td>\n",
       "      <td>0.025052</td>\n",
       "      <td>0.020377</td>\n",
       "      <td>0.174520</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.012216</td>\n",
       "      <td>1.306369</td>\n",
       "      <td>-0.016039</td>\n",
       "      <td>1.226419</td>\n",
       "      <td>0.005649</td>\n",
       "      <td>0.050405</td>\n",
       "      <td>1.140050</td>\n",
       "      <td>1.064812</td>\n",
       "      <td>0.308886</td>\n",
       "      <td>1.260318</td>\n",
       "      <td>0.900862</td>\n",
       "      <td>0.015015</td>\n",
       "      <td>-0.004450</td>\n",
       "      <td>0.003828</td>\n",
       "      <td>-0.018275</td>\n",
       "      <td>0.026666</td>\n",
       "      <td>0.067881</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.409959</td>\n",
       "      <td>-0.033269</td>\n",
       "      <td>1.294283</td>\n",
       "      <td>1.280013</td>\n",
       "      <td>0.016159</td>\n",
       "      <td>0.024662</td>\n",
       "      <td>0.333958</td>\n",
       "      <td>1.233323</td>\n",
       "      <td>0.312768</td>\n",
       "      <td>0.965580</td>\n",
       "      <td>0.258461</td>\n",
       "      <td>1.256214</td>\n",
       "      <td>0.009020</td>\n",
       "      <td>-0.023095</td>\n",
       "      <td>-0.079765</td>\n",
       "      <td>-0.012743</td>\n",
       "      <td>0.055093</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         0         1         2         3         4         5         6   \\\n",
       "0  0.461209  0.030720  0.057163  1.299550  0.016376  0.032741  0.440937   \n",
       "1  0.617313 -0.005237  0.016966  1.252040  0.123216  0.027827  0.065470   \n",
       "2  0.436341  1.272836 -0.010398  1.217566  0.127716  0.046084  0.144033   \n",
       "3  0.012216  1.306369 -0.016039  1.226419  0.005649  0.050405  1.140050   \n",
       "4  0.409959 -0.033269  1.294283  1.280013  0.016159  0.024662  0.333958   \n",
       "\n",
       "         7         8         9         10        11        12        13  \\\n",
       "0  1.216574  0.690130  1.174628  0.059692  1.255372  0.003491 -0.032661   \n",
       "1  1.233238  0.325361  1.069075  0.561540  0.061380 -0.013161  1.271114   \n",
       "2  1.051804  0.297363  1.276768  0.834105  0.036984 -0.011628 -0.003888   \n",
       "3  1.064812  0.308886  1.260318  0.900862  0.015015 -0.004450  0.003828   \n",
       "4  1.233323  0.312768  0.965580  0.258461  1.256214  0.009020 -0.023095   \n",
       "\n",
       "         14        15        16  \n",
       "0 -0.077568 -0.013338  0.015783  \n",
       "1 -0.022888  0.061607  0.153778  \n",
       "2  0.025052  0.020377  0.174520  \n",
       "3 -0.018275  0.026666  0.067881  \n",
       "4 -0.079765 -0.012743  0.055093  "
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 将 PyTorch 张量转换为 Pandas DataFrame\n",
    "instance_embeddings_df = pd.DataFrame(instance_embeddings.cpu().numpy())\n",
    "\n",
    "instance_embeddings_df.head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "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>10</th>\n",
       "      <th>11</th>\n",
       "      <th>12</th>\n",
       "      <th>13</th>\n",
       "      <th>14</th>\n",
       "      <th>15</th>\n",
       "      <th>16</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.364921</td>\n",
       "      <td>0.083284</td>\n",
       "      <td>0.147666</td>\n",
       "      <td>0.901936</td>\n",
       "      <td>0.058670</td>\n",
       "      <td>0.130946</td>\n",
       "      <td>0.377387</td>\n",
       "      <td>0.909268</td>\n",
       "      <td>0.532172</td>\n",
       "      <td>0.865565</td>\n",
       "      <td>0.144687</td>\n",
       "      <td>0.896476</td>\n",
       "      <td>0.485819</td>\n",
       "      <td>0.069077</td>\n",
       "      <td>0.035219</td>\n",
       "      <td>0.081347</td>\n",
       "      <td>0.085553</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.475041</td>\n",
       "      <td>0.059149</td>\n",
       "      <td>0.121396</td>\n",
       "      <td>0.871848</td>\n",
       "      <td>0.150307</td>\n",
       "      <td>0.106804</td>\n",
       "      <td>0.113298</td>\n",
       "      <td>0.920403</td>\n",
       "      <td>0.276319</td>\n",
       "      <td>0.790601</td>\n",
       "      <td>0.502285</td>\n",
       "      <td>0.080758</td>\n",
       "      <td>0.330056</td>\n",
       "      <td>0.921658</td>\n",
       "      <td>0.072587</td>\n",
       "      <td>0.130436</td>\n",
       "      <td>0.198479</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.347379</td>\n",
       "      <td>0.916978</td>\n",
       "      <td>0.103514</td>\n",
       "      <td>0.850015</td>\n",
       "      <td>0.154167</td>\n",
       "      <td>0.196509</td>\n",
       "      <td>0.168557</td>\n",
       "      <td>0.799178</td>\n",
       "      <td>0.256681</td>\n",
       "      <td>0.938104</td>\n",
       "      <td>0.696504</td>\n",
       "      <td>0.064090</td>\n",
       "      <td>0.344398</td>\n",
       "      <td>0.087893</td>\n",
       "      <td>0.105348</td>\n",
       "      <td>0.103430</td>\n",
       "      <td>0.215453</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.048191</td>\n",
       "      <td>0.939485</td>\n",
       "      <td>0.099827</td>\n",
       "      <td>0.855622</td>\n",
       "      <td>0.049470</td>\n",
       "      <td>0.217738</td>\n",
       "      <td>0.869116</td>\n",
       "      <td>0.807869</td>\n",
       "      <td>0.264763</td>\n",
       "      <td>0.926422</td>\n",
       "      <td>0.744073</td>\n",
       "      <td>0.049082</td>\n",
       "      <td>0.411541</td>\n",
       "      <td>0.092939</td>\n",
       "      <td>0.075739</td>\n",
       "      <td>0.107550</td>\n",
       "      <td>0.128187</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.328768</td>\n",
       "      <td>0.040335</td>\n",
       "      <td>0.956151</td>\n",
       "      <td>0.889563</td>\n",
       "      <td>0.058484</td>\n",
       "      <td>0.091249</td>\n",
       "      <td>0.302142</td>\n",
       "      <td>0.920459</td>\n",
       "      <td>0.267487</td>\n",
       "      <td>0.717098</td>\n",
       "      <td>0.286323</td>\n",
       "      <td>0.897051</td>\n",
       "      <td>0.537539</td>\n",
       "      <td>0.075333</td>\n",
       "      <td>0.033717</td>\n",
       "      <td>0.081737</td>\n",
       "      <td>0.117722</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         0         1         2         3         4         5         6   \\\n",
       "0  0.364921  0.083284  0.147666  0.901936  0.058670  0.130946  0.377387   \n",
       "1  0.475041  0.059149  0.121396  0.871848  0.150307  0.106804  0.113298   \n",
       "2  0.347379  0.916978  0.103514  0.850015  0.154167  0.196509  0.168557   \n",
       "3  0.048191  0.939485  0.099827  0.855622  0.049470  0.217738  0.869116   \n",
       "4  0.328768  0.040335  0.956151  0.889563  0.058484  0.091249  0.302142   \n",
       "\n",
       "         7         8         9         10        11        12        13  \\\n",
       "0  0.909268  0.532172  0.865565  0.144687  0.896476  0.485819  0.069077   \n",
       "1  0.920403  0.276319  0.790601  0.502285  0.080758  0.330056  0.921658   \n",
       "2  0.799178  0.256681  0.938104  0.696504  0.064090  0.344398  0.087893   \n",
       "3  0.807869  0.264763  0.926422  0.744073  0.049082  0.411541  0.092939   \n",
       "4  0.920459  0.267487  0.717098  0.286323  0.897051  0.537539  0.075333   \n",
       "\n",
       "         14        15        16  \n",
       "0  0.035219  0.081347  0.085553  \n",
       "1  0.072587  0.130436  0.198479  \n",
       "2  0.105348  0.103430  0.215453  \n",
       "3  0.075739  0.107550  0.128187  \n",
       "4  0.033717  0.081737  0.117722  "
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 标准化数值特征\n",
    "scaler = MinMaxScaler()\n",
    "instance_embeddings_df_scaled = scaler.fit_transform(instance_embeddings_df)\n",
    "# 将标准化后的数据转换回 DataFrame\n",
    "instance_embeddings_df_scaled = pd.DataFrame(instance_embeddings_df_scaled, columns=instance_embeddings_df.columns)\n",
    "\n",
    "instance_embeddings_df_scaled.head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Instance embeddings saved to /home/sdong/data/credit_card/application_record_string_embeddings.csv\n"
     ]
    }
   ],
   "source": [
    "# 保存为 CSV 文件\n",
    "csv_file_path = '/home/sdong/data/credit_card/application_record_string_embeddings.csv'\n",
    "instance_embeddings_df_scaled.to_csv(csv_file_path, index=False)\n",
    "\n",
    "print(f'Instance embeddings saved to {csv_file_path}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "mainenv",
   "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": "3.11.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}