异构图比较复杂，先从简单的demo写起。有时用户会想在异构图上进行图神经网络的训练。先整个节点分类的案例。我没创建如下格式的一个图，然后进行节点分类任务的学习。
这个 hetero_graph 异构图有以下这些边的类型：
('user', 'follow', 'user')('user', 'followed-by', 'user')('user', 'click', 'item')('item', 'clicked-by', 'user')('user', 'dislike', 'item')('item', 'disliked-by', 'user')
加载所需要的库
import dglimport dgl.nn as dglnnimport numpy as npimport torchimport torch.nn as nnimport torch.nn.functional as F
数据集构建
n_users = 1000 #用户数n_items = 500 #项目数n_follows = 3000 #用户采用数n_clicks = 5000 #用户点击数n_dislikes = 500 #用户不喜欢数n_hetero_features = 10 #节点特征维度n_user_classes = 5 #用户类型数，训练d的输层使用n_max_clicks = 10 #
边数据构造
#np.random.randint(low, high, size, dtype='l')#采用边-3000follow_src = np.random.randint(0, n_users, n_follows)follow_dst = np.random.randint(0, n_users, n_follows)#点击边-5000click_src = np.random.randint(0, n_users, n_clicks)click_dst = np.random.randint(0, n_items, n_clicks)#不喜欢边-500dislike_src = np.random.randint(0, n_users, n_dislikes)dislike_dst = np.random.randint(0, n_items, n_dislikes)
创建DGL的异构图
hetero_graph = dgl.heterograph({ #正反两个方向构边('user', 'follow', 'user'): (follow_src, follow_dst),('user', 'followed-by', 'user'): (follow_dst, follow_src),('user', 'click', 'item'): (click_src, click_dst),('item', 'clicked-by', 'user'): (click_dst, click_src),('user', 'dislike', 'item'): (dislike_src, dislike_dst),('item', 'disliked-by', 'user'): (dislike_dst, dislike_src)})hetero_graph
特征构造
hetero_graph.nodes['user'].data['feature'] = torch.randn(n_users, n_hetero_features)hetero_graph.nodes['item'].data['feature'] = torch.randn(n_items, n_hetero_features)hetero_graph.nodes['user'].data['label'] = torch.randint(0, n_user_classes, (n_users,))hetero_graph.edges['click'].data['label'] = torch.randint(1, n_max_clicks, (n_clicks,)).float()hetero_graph.nodes['user'].data['train_mask'] = torch.zeros(n_users, dtype=torch.bool).bernoulli(0.6)hetero_graph.edges['click'].data['train_mask'] = torch.zeros(n_clicks, dtype=torch.bool).bernoulli(0.6)
节点分类任务
用户分为五类，对用户的类型进行预测
#定义特征聚合模块class RGCN(nn.Module):def __init__(self, in_feats, hid_feats, out_feats, rel_names):super().__init__()self.conv1 = dglnn.HeteroGraphConv({rel:dglnn.GraphConv(in_feats, hid_feats)for rel in rel_names}, aggregate='sum')self.conv2 = dglnn.HeteroGraphConv({rel:dglnn.GraphConv(hid_feats, out_feats)for rel in rel_names}, aggregate='sum')def forward(self, graph, inputs):h = self.conv1(graph, inputs)h = {k:F.relu(v) for k, v in h.items()}h = self.conv2(graph, h)return h
构造模型
#构造模型model = RGCN(in_feats=n_hetero_features, hid_feats=20,out_feats=n_user_classes, rel_names=hetero_graph.etypes)user_feats = hetero_graph.nodes['user'].data['feature'] #用户特征item_feats = hetero_graph.nodes['item'].data['feature'] #项目特征labels = hetero_graph.nodes['user'].data['label'] #用户类型标签train_mask = hetero_graph.nodes['user'].data['train_mask']
这里用到了hetero_graph.etypes，我们看看这个是啥：
hetero_graph.etypeshetero_graph.etypes['clicked-by', 'disliked-by', 'click', 'dislike','follow', 'followed-by']
可以看到就是边的类型
node_features = {'user':user_feats, 'item':item_feats}#特征字典h_dict = model(hetero_graph, node_features)
模型训练
#模型优化器opt = torch.optim.Adam(model.parameters())best_train_acc = 0loss_list = []train_score_list = []#迭代训练for epoch in range(500):model.train()#输入图和节点特征，提取出user的特征logits = model(hetero_graph, node_features)['user']#计算损失loss = F.cross_entropy(logits[train_mask], labels[train_mask])#预测userpred = logits.argmax(1)#计算准确率train_acc = (pred[train_mask] == labels[train_mask]).float().mean()if best_train_acc < train_acc:best_train_acc = train_acctrain_score_list.append(train_acc)#反向优化opt.zero_grad()loss.backward()opt.step()loss_list.append(loss.item())#输出训练结果%.4f, Train Acc %.4f (Best %.4f)' % (loss.item(),train_acc.item(),best_train_acc.item(),))
