GNN图级任务代码实现，多分类，使用随机生成的数据集进行训练、验证

好的，这里提供一个基于PyTorch和DGL库的GNN图级任务（多分类）代码实现，使用随机生成的数据集进行训练和验证。以下是实现步骤：

1. 导入必要的库

import torch
import torch.nn as nn
import torch.optim as optim
import dgl
from dgl import DGLGraph
from dgl.data import citation_graph as citegrh
import networkx as nx
import numpy as np
import random

2. 生成随机图数据集

# 生成随机图数据集
num_nodes = 50  # 节点数
num_classes = 5  # 类别数
num_feats = 10  # 特征数
num_edges = 100  # 边数

# 生成节点特征
features = torch.randn(num_nodes, num_feats)

# 生成边
src = torch.randint(low=0, high=num_nodes, size=(num_edges,))
dst = torch.randint(low=0, high=num_nodes, size=(num_edges,))
edges = torch.stack([src, dst], dim=0)

# 生成标签
labels = torch.randint(low=0, high=num_classes, size=(num_nodes,))

# 创建图
g = DGLGraph()
g.add_nodes(num_nodes)
g.add_edges(edges[0], edges[1])
g.ndata['feat'] = features
g.ndata['label'] = labels

3. 定义GNN模型

# 定义GNN模型
class GCN(nn.Module):
    def __init__(self, in_feats, hidden_size, num_classes):
        super(GCN, self).__init__()
        self.conv1 = nn.Conv2d(in_feats, hidden_size, kernel_size=1)
        self.conv2 = nn.Conv2d(hidden_size, num_classes, kernel_size=1)
        self.dropout = nn.Dropout(p=0.5)

    def forward(self, g, features):
        h = self.conv1(g, features.unsqueeze(1))
        h = nn.functional.relu(h)
        h = self.dropout(h)
        h = self.conv2(g, h)
        g.ndata['h'] = h.squeeze()
        hg = dgl.mean_nodes(g, 'h')
        return hg

4. 定义训练函数和测试函数

# 定义训练函数
def train(model, g, optimizer, criterion):
    model.train()
    optimizer.zero_grad()
    output = model(g, g.ndata['feat'])
    loss = criterion(output, g.ndata['label'])
    loss.backward()
    optimizer.step()
    return loss.item()

# 定义测试函数
def test(model, g):
    model.eval()
    output = model(g, g.ndata['feat'])
    pred = output.argmax(dim=1)
    acc = (pred == g.ndata['label']).float().mean()
    return acc.item()

5. 定义训练过程

# 定义训练过程
def train_process(model, g, epochs, lr, weight_decay):
    optimizer = optim.Adam(model.parameters(), lr=lr, weight_decay=weight_decay)
    criterion = nn.CrossEntropyLoss()
    for epoch in range(epochs):
        loss = train(model, g, optimizer, criterion)
        acc = test(model, g)
        print('Epoch {:03d} | Loss {:.4f} | Acc {:.4f}'.format(epoch, loss, acc))

6. 运行训练过程

# 运行训练过程
model = GCN(num_feats, 16, num_classes)
train_process(model, g, 50, 0.01, 0.001)

在运行以上代码后，将会看到输出的训练过程和测试结果。这个例子中，我们使用了一个简单的GCN模型进行多分类任务的训练和测试，数据集是随机生成的。你可以根据实际任务需求，进行相应的修改和调整。