在PyG中通过torch_geometric.data.Data创建一个简单的图，具有如下属性：

data.x：节点的特征矩阵，shape：[num_nodes, num_node_features]data.edge_index：边的矩阵，shape：[2, num_edges]data.edge_attr：边的属性矩阵，shape：[num_edges, num_edges_features]data.y：节点的分类任务，样本标签，shape：[num_nodes, *]，图分类任务shape:[1, *]data.pos：节点的坐标，shape[num_nodes,num_dimension]

创建一个图

import torchfrom torch_geometric.data import Data# 定义了边的表示，是无向图，所以shape:[2, 4] ,(0,1)(1,0)(1,2)(2,1)edge_index = torch.tensor([[0, 1, 1, 2],[1, 0, 2, 1]], dtype=torch.long)x = torch.tensor([[-1], [0], [1]], dtype=torch.float)# 有三个节点，第0个节点特征是[-1]，第一个节点特征是[0], 第二个节点特征是[1]data = Data(x=x, edge_index=edge_index)

Data(x=[3, 1], edge_index=[2, 4])中x=[3,1]表示有三个节点，每个节点一个特征，edge_index=[2, 4]表示有四条边

也可以通过下面的方式创建边：主要是edge_index.t().contiguous()

import torchfrom torch_geometric.data import Dataedge_index = torch.tensor([[0, 1],[1, 0],[1, 2],[2, 1]], dtype=torch.long)x = torch.tensor([[-1], [0], [1]], dtype=torch.float)data = Data(x=x, edge_index=edge_index.t().contiguous())>>> Data(edge_index=[2, 4], x=[3, 1])

除了上述的功能（节点、边、图的一些属性），data还提供了额外的方法：

print(data.keys)>>> ['x', 'edge_index']# 节点的特征print(data['x'])>>> tensor([[-1.0],[0.0],[1.0]])for key, item in data:print("{} found in data".format(key))>>> x found in data>>> edge_index found in data# 边的属性'edge_attr' in data>>> False# 节点的数量data.num_nodes>>> 3# 边的数量data.num_edges>>> 4# 节点的特征数量data.num_node_features>>> 1# 是否拥有孤立的节点data.has_isolated_nodes()>>> False# 是否一个环data.has_self_loops()>>> False# 是不是有向图data.is_directed()>>> False# Transfer data object to GPU.将data转到gpudevice = torch.device('cuda')data = data.to(device)

创建好data之后，PyG内置了一些公开的数据集，可以导入：

from torch_geometric.datasets import TUDataset# 数据集是对图进行分类的任务dataset = TUDataset(root='/tmp/ENZYMES', name='ENZYMES')# 有600个图len(dataset)>>> 600# 图的类别数量6dataset.num_classes>>> 6# 图的每个节点的特征数量是3dataset.num_node_features>>> 3# 选择第一个图data = dataset[0]>>> Data(edge_index=[2, 168], x=[37, 3], y=[1])# 无向图data.is_undirected()>>> True

使用Cora数据集：

dataset = Planetoid(root='/tmp/Cora', name='Cora')>>> Cora()len(dataset)>>> 1dataset.num_classes>>> 7dataset.num_node_features>>> 1433# 获得这张图data = dataset[0]# train_mask表示训练那些节点（140个），test_mask表示测试哪些节点（1000个）>>> Data(edge_index=[2, 10556], test_mask=[2708],train_mask=[2708], val_mask=[2708], x=[2708, 1433], y=[2708])data.is_undirected()>>> Truedata.train_mask.sum().item()>>> 140data.val_mask.sum().item()>>> 500data.test_mask.sum().item()>>> 1000

PyG实现GCN、GraphSage、GAT

GCN实现

from torch_geometric.datasets import Planetoiddataset = Planetoid(root='/tmp/Cora', name='Cora')from torch_geometric.datasets import Planetoidimport torchimport torch.nn.functional as Ffrom torch_geometric.nn import GCNConv, SAGEConv, GATConvdataset = Planetoid(root='/tmp/Cora', name='Cora')class GCN_Net(torch.nn.Module):def __init__(self, feature, hidden, classes):super(GCN_Net, self).__init__()self.conv1 = GCNConv(feature, hidden)self.conv2 = GCNConv(hidden, classes)def forward(self, data):x, edge_index = data.x, data.edge_indexx = self.conv1(x, edge_index)x = F.relu(x)x = F.dropout(x, training=self.training)x = self.conv2(x, edge_index)return F.log_softmax(x, dim=1)device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')model = GCN_Net(dataset.num_node_features, 16, dataset.num_classes).to(device)data = dataset[0]# optimizer = torch.optim.Adam(model.parameters(), lr=0.01)# optimizer = torch.optim.Adam([# dict(params=model.conv1.parameters(), weight_decay=5e-4),#dict(params=model.conv2.parameters(), weight_decay=0)#], lr=0.01)optimizer = torch.optim.Adam(model.parameters(),lr=0.01, weight_decay=5e-4)model.train()for epoch in range(1000):optimizer.zero_grad()out = model(data)loss = F.nll_loss(out[data.train_mask], data.y[data.train_mask])correct = out[data.train_mask].max(dim=1)[1].eq(data.y[data.train_mask]).double().sum()# print('epoch:', epoch, ' acc:', correct / int(data.train_mask.sum()))loss.backward()optimizer.step()if epoch % 10 == 9:model.eval()logits, accs = model(data), []for _, mask in data('train_mask', 'val_mask', 'test_mask'):pred = logits[mask].max(1)[1]acc = pred.eq(data.y[mask]).sum().item() / mask.sum().item()accs.append(acc)log = 'Epoch: {:03d}, Train: {:.5f}, Val: {:.5f}, Test: {:.5f}'print(log.format(epoch + 1, accs[0], accs[1], accs[2]))# model.eval()# _, pred = model(data).max(dim=1)# correct = pred[data.test_mask].eq(data.y[data.test_mask]).sum()# acc = int(correct) / int(data.test_mask.sum())# print(acc)

GraphSage实现：

from torch_geometric.datasets import Planetoidimport torchimport torch.nn.functional as Ffrom torch_geometric.nn import GCNConv, SAGEConv, GATConvdataset = Planetoid(root='/tmp/Cora', name='Cora')class GraphSage_Net(torch.nn.Module):def __init__(self, features, hidden, classes):super(GraphSage_Net, self).__init__()self.sage1 = SAGEConv(features, hidden)self.sage2 = SAGEConv(hidden, classes)def forward(self, data):x, edge_index = data.x, data.edge_indexx = self.sage1(x, edge_index)x = F.relu(x)x = F.dropout(x, training=self.training)x = self.sage2(x, edge_index)return F.log_softmax(x, dim=1)device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')model = GraphSage_Net(dataset.num_node_features, 16, dataset.num_classes).to(device)data = dataset[0]# optimizer = torch.optim.Adam(model.parameters(), lr=0.01)# optimizer = torch.optim.Adam([# dict(params=model.conv1.parameters(), weight_decay=5e-4),#dict(params=model.conv2.parameters(), weight_decay=0)#], lr=0.01)optimizer = torch.optim.Adam(model.parameters(),lr=0.01, weight_decay=5e-4)model.train()for epoch in range(1000):optimizer.zero_grad()out = model(data)loss = F.nll_loss(out[data.train_mask], data.y[data.train_mask])correct = out[data.train_mask].max(dim=1)[1].eq(data.y[data.train_mask]).double().sum()# print('epoch:', epoch, ' acc:', correct / int(data.train_mask.sum()))loss.backward()optimizer.step()if epoch % 10 == 9:model.eval()logits, accs = model(data), []for _, mask in data('train_mask', 'val_mask', 'test_mask'):pred = logits[mask].max(1)[1]acc = pred.eq(data.y[mask]).sum().item() / mask.sum().item()accs.append(acc)log = 'Epoch: {:03d}, Train: {:.5f}, Val: {:.5f}, Test: {:.5f}'print(log.format(epoch + 1, accs[0], accs[1], accs[2]))

GAT 实现：

from torch_geometric.datasets import Planetoidimport torchimport torch.nn.functional as Ffrom torch_geometric.nn import GCNConv, SAGEConv, GATConvdataset = Planetoid(root='/tmp/Cora', name='Cora')class GAT_Net(torch.nn.Module):def __init__(self, features, hidden, classes, heads=1):super(GAT_Net, self).__init__()self.gat1 = GATConv(features, hidden, heads=heads)self.gat2 = GATConv(hidden * heads, classes)def forward(self, data):x, edge_index = data.x, data.edge_indexx = self.gat1(x, edge_index)x = F.relu(x)x = F.dropout(x, training=self.training)x = self.gat2(x, edge_index)return F.log_softmax(x, dim=1)device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')model = GAT_Net(dataset.num_node_features, 16, dataset.num_classes, heads=4).to(device)data = dataset[0]# optimizer = torch.optim.Adam(model.parameters(), lr=0.01)# optimizer = torch.optim.Adam([# dict(params=model.conv1.parameters(), weight_decay=5e-4),#dict(params=model.conv2.parameters(), weight_decay=0)#], lr=0.01)optimizer = torch.optim.Adam(model.parameters(),lr=0.01, weight_decay=5e-4)model.train()for epoch in range(1000):optimizer.zero_grad()out = model(data)loss = F.nll_loss(out[data.train_mask], data.y[data.train_mask])correct = out[data.train_mask].max(dim=1)[1].eq(data.y[data.train_mask]).double().sum()# print('epoch:', epoch, ' acc:', correct / int(data.train_mask.sum()))loss.backward()optimizer.step()if epoch % 10 == 9:model.eval()logits, accs = model(data), []for _, mask in data('train_mask', 'val_mask', 'test_mask'):pred = logits[mask].max(1)[1]acc = pred.eq(data.y[mask]).sum().item() / mask.sum().item()accs.append(acc)log = 'Epoch: {:03d}, Train: {:.5f}, Val: {:.5f}, Test: {:.5f}'print(log.format(epoch + 1, accs[0], accs[1], accs[2]))