对一组非图像的拓扑图（bin文件，格式为coo）进行训练，图一共有60000左右（即60000多个bin文件），使用VGAE，代码示例

首先，需要先导入所需的Python库和模块：

import os
import torch
import numpy as np
from torch.utils.data import Dataset, DataLoader
from torch_geometric.nn import VGAE
from torch_geometric.data import Data
from torch_geometric.utils import from_scipy_sparse_matrix
import scipy.sparse as sp

然后，定义一个自定义数据集类：

class MyDataset(Dataset):
    def __init__(self, root_dir):
        self.root_dir = root_dir
        self.files = os.listdir(root_dir)

    def __len__(self):
        return len(self.files)

    def __getitem__(self, idx):
        file_path = os.path.join(self.root_dir, self.files[idx])
        coo_matrix = sp.load_npz(file_path)
        edge_index, edge_attr = from_scipy_sparse_matrix(coo_matrix)
        x = torch.ones(edge_index.max() + 1)
        return Data(x=x, edge_index=edge_index, edge_attr=edge_attr)

接着，定义训练函数：

def train(model, optimizer, train_loader, device):
    model.train()

    loss_all = 0
    for data in train_loader:
        data = data.to(device)
        optimizer.zero_grad()
        z = model.encode(data.x, data.edge_index)
        loss = model.recon_loss(z, data.edge_index, data.edge_attr)
        loss.backward()
        loss_all += loss.item() * data.num_graphs
        optimizer.step()

    return loss_all / len(train_loader.dataset)

最后，可以开始训练模型：

dataset = MyDataset(root_dir='./bin_files')
train_loader = DataLoader(dataset, batch_size=64, shuffle=True)

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = VGAE(dataset[0].num_nodes, 16, 8).to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

for epoch in range(200):
    loss = train(model, optimizer, train_loader, device)
    print('Epoch: {:03d}, Loss: {:.4f}'.format(epoch, loss))

这里的VGAE使用16维的节点嵌入和8维的边嵌入，训练200个epoch，可以根据实际情况进行调整。