Pytorch+PyG实现GraphSAGE过程示例详解

import torch.nn.functional as F
from torch_geometric.datasets import Planetoid
from torch_geometric.utils import from_networkx, to_networkx
# 加载cora数据集
dataset = Planetoid(root='./cora', name='Cora')
data = dataset[0]
# 将nx.Graph形式的图转换成PyG需要的格式
graph = to_networkx(data)
data = from_networkx(graph)
# 获取节点数量和特征向量维度
num_nodes = data.num_nodes
num_features = dataset.num_features
num_classes = dataset.num_classes
# 建立需要训练的节点分割数据集
data.train_mask = torch.zeros(num_nodes, dtype=torch.bool)
data.val_mask = torch.zeros(num_nodes, dtype=torch.bool)
data.test_mask = torch.zeros(num_nodes, dtype=torch.bool)
data.train_mask[:num_nodes - 1000] = True
data.test_mask[-1000:] = True
data.val_mask[num_nodes - 2000: num_nodes - 1000] = True

from torch.nn import Sequential as Seq, Linear as Lin, ReLU
from torch_geometric.nn import SAGEConv
class GraphSAGE(torch.nn.Module):
    def __init__(self, hidden_channels, num_layers):
        super(GraphSAGE, self).__init__()
        self.convs = nn.ModuleList()
        for i in range(num_layers):
            in_channels = hidden_channels if i != 0 else num_features
            out_channels = num_classes if i == num_layers - 1 else hidden_channels
            self.convs.append(SAGEConv(in_channels, out_channels))
    def forward(self, x, edge_index):
        for _, conv in enumerate(self.convs[:-1]):
            x = F.relu(conv(x, edge_index))
        # 最后一层不用激活函数
        x = self.convs[-1](x, edge_index)
        return F.log_softmax(x, dim=-1)

# 初始化GraphSage并指定参数
num_layers = 2
hidden_channels = 256
model = GraphSAGE(hidden_channels, num_layers).to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)
loss_func = nn.CrossEntropyLoss()
# 训练过程
for epoch in range(500):
    model.train()
    optimizer.zero_grad()
    out = model(data.x.to(device), data.edge_index.to(device))
    loss = loss_func(out[data.train_mask], data.y.to(device)[data.train_mask])
    loss.backward()
    optimizer.step()
    # 在各个测试阶段检测一下准确率
    if epoch % 10 == 0:
        with torch.no_grad():
            _, pred = model(data.x.to(device), data.edge_index.to(device)).max(dim=1)
            correct = float(pred[data.test_mask].eq(data.y.to(device)[data.test_mask]).sum().item())
            acc = correct / data.test_mask.sum().item()
            print("Epoch {:03d}, Train Loss {:.4f}, Test Acc {:.4f}".format(
                epoch, loss.item(), acc))