图卷积网络 (GCN) 多标签分类模型实现

本代码演示了使用图卷积网络 (GCN) 进行多标签分类的完整实现，包括数据预处理、模型定义、训练和评估。

1. 导入库

import os
import pandas as pd
import torch
import torch.nn as nn
from torch_geometric.data import Data, DataLoader
from torch_geometric.nn import GCNConv
import torch.nn.functional as F
from torchvision import transforms
from PIL import Image
from sklearn.model_selection import train_test_split

2. 定义模型

CNN 模型

# 定义CNN网络
class CNN(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(CNN, self).__init__()
        self.conv1 = nn.Conv2d(in_channels, 16, kernel_size=3, stride=1, padding=1)
        self.pool = nn.MaxPool2d(kernel_size=2, stride=2, padding=0)
        self.conv2 = nn.Conv2d(16, out_channels, kernel_size=3, stride=1, padding=1)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        x = self.pool(x)
        x = F.relu(self.conv2(x))
        x = self.pool(x)
        return x

GCN 模型

# 定义GCN模型
class GCN(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(GCN, self).__init__()
        self.conv1 = GCNConv(in_channels, 128)
        # self.dropout = nn.Dropout(p=0.5) # 添加 Dropout 层
        self.conv2 = GCNConv(128, out_channels)

    def forward(self, data):
        x, edge_index = data.x, data.edge_index
        x = F.relu(self.conv1(x, edge_index))
        # x = self.dropout(x) # 在第一层后添加 Dropout 层
        x = self.conv2(x, edge_index)
        return x

3. 数据预处理

# 读取边的关系数据
edges = pd.read_csv('C:\Users\jh\Desktop\data\input\edges_L.csv', header=None)
edges = edges.values

# 读取节点特征数据
features = []
labels = []
for i in range(1, 43):
    for j in range(37):
        image_path = f'C:\Users\jh\Desktop\data\input\images\{i}.png_{j}.png'
        image = Image.open(image_path).convert('RGB')
        transform = transforms.Compose([transforms.Resize((40, 40)), transforms.ToTensor()])
        image_tensor = transform(image)
        features.append(image_tensor)

        labels_path = f'C:\Users\jh\Desktop\data\input\labels\{i}_{j}.txt'
        with open(labels_path, 'r') as file:
            label = [int(l) for l in file.readline().strip().split()]
            labels.extend(label)

x = torch.stack(features)
x = x.view(-1, 3, 40, 40)

# 划分训练集和验证集的掩码
mask_train = torch.zeros(42, 37, dtype=torch.bool)
mask_val = torch.zeros(42, 37, dtype=torch.bool)
for i in range(42):
    mask_train[i, :30] = 1
    mask_val[i, 30:] = 1
mask_train = mask_train.view(-1)
mask_val = mask_val.view(-1)

# 创建图结构
edge_index = torch.tensor(edges, dtype=torch.long).t().contiguous()
data_list = []
for i in range(42):
    data = Data(x=x, edge_index=edge_index)
    data.mask_train = mask_train[i*37:(i+1)*37]
    data.mask_val = mask_val[i*37:(i+1)*37]
    data.y = torch.tensor(labels[i*37:(i+1)*37])
    data_list.append(data)

4. 使用 CNN 对节点特征进行降维

# 创建CNN模型实例，降维至8维
cnn_model = CNN(in_channels=3, out_channels=8)

# 使用CNN模型对节点特征进行降维
with torch.no_grad():
    cnn_output = []
    for i in range(42):
        x_i = x[i*37:(i+1)*37].unsqueeze(1)
        x_i = x_i.squeeze(1)
        output_i = cnn_model(x_i)
        output_i = output_i.view(output_i.size(0), -1)
        cnn_output.append(output_i)
    cnn_output = torch.cat(cnn_output, dim=0)

# 将降维后的特征加入data对象
for i in range(42):
    data_list[i].x = cnn_output[i*37:(i+1)*37]

5. 创建 GCN 模型实例

# 创建GCN模型实例
gcn_model = GCN(in_channels=800, out_channels=8)

6. 定义损失函数和优化器

# 定义多标签分类的损失函数和优化器
criterion = nn.BCEWithLogitsLoss()
optimizer = torch.optim.Adam(gcn_model.parameters(), lr=0.0008)

7. 训练模型

# 训练模型
num_epochs = 5000
for epoch in range(num_epochs):
    gcn_model.train()
    total_loss = 0
    for data in data_list:
        optimizer.zero_grad()
        out = gcn_model(data)
        labels_batch = F.one_hot(data.y, num_classes=8).float()
        loss = criterion(out, labels_batch)
        loss.backward()
        optimizer.step()
        total_loss += loss.item()

    avg_loss = total_loss / len(data_list)
    print(f'Epoch [{epoch + 1}/{num_epochs}], Loss: {avg_loss:.6f}')

8. 评估模型

# 在验证集上评估模型
gcn_model.eval()
with torch.no_grad():
    total_correct = 0
    total_samples = 0
    for data in data_list:
        out = gcn_model(data)
        labels_batch = F.one_hot(data.y, num_classes=8).float()
        predicted = (out > 0).float() # 使用阈值 0 进行二分类
        total_correct += (predicted == labels_batch).sum().item()
        total_samples += labels_batch.size(0) * labels_batch.size(1)

    accuracy = total_correct / total_samples
    print(f'Validation Accuracy: {accuracy:.4f}')

9. 输出解释

上述图卷积网络的输出是每个节点的特征表示。在GCN模型中，通过两个GCNConv层对节点特征进行处理和传递，最终得到每个节点的特征表示。这些特征表示可以用于节点的分类、节点的聚类、图的生成等任务。