class GCN(torch.nn.Module):
    def __init__(self, num_node_features, num_classes):
        super(GCN, self).__init__()
        self.conv1 = GCNConv(num_node_features, 16)
        self.conv2 = GCNConv(16, num_classes)
    def forward(self, data):
        x, edge_index = data.x, data.edge_index
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = F.dropout(x, training=self.training)
        x = self.conv2(x, edge_index)
        x = F.relu(x)
        x = F.dropout(x, training=self.training)
        x = F.softmax(x, dim=1)
        return x

輸出一下模型：

data = Planetoid(root='/data/CiteSeer', name='CiteSeer')model = GCN(data.num_node_features, data.num_classes).to(device)print(model)GCN(
  (conv1): GCNConv(3703, 16)
  (conv2): GCNConv(16, 6)
)

輸出為：

GCN( (conv1): GCNConv(3703, 16) (conv2): GCNConv(16, 6))GCN(
  (conv1): GCNConv(3703, 16)
  (conv2): GCNConv(16, 6)
)

1. 前向傳播

查看官方文檔中GCNConv的輸入輸出要求：

可以發(fā)現(xiàn)，GCNConv中需要輸入的是節(jié)點(diǎn)特征矩陣x和鄰接關(guān)系edge_index，還有一個(gè)可選項(xiàng)edge_weight。因此我們首先：

x, edge_index = data.x, data.edge_index
x = self.conv1(x, edge_index)
x = F.relu(x)
x = F.dropout(x, training=self.training)

此時(shí)我們不妨輸出一下x及其size：

tensor([[0.0000, 0.1630, 0.0000,  ..., 0.0000, 0.0488, 0.0000],
        [0.0000, 0.2451, 0.1614,  ..., 0.0000, 0.0125, 0.0000],
        [0.1175, 0.0262, 0.2141,  ..., 0.2592, 0.0000, 0.0000],
        ...,
        [0.0000, 0.0000, 0.0000,  ..., 0.0000, 0.1825, 0.0000],
        [0.0000, 0.1024, 0.0000,  ..., 0.0498, 0.0000, 0.0000],
        [0.0000, 0.3263, 0.0000,  ..., 0.0000, 0.0000, 0.0000]],
       device='cuda:0', grad_fn=<FusedDropoutBackward0>)

torch.Size([3327, 16])

此時(shí)的x一共3327行，每一行表示一個(gè)節(jié)點(diǎn)經(jīng)過第一層卷積更新后的狀態(tài)向量。

那么同理，由于：

self.conv2 = GCNConv(16, num_classes)

所以經(jīng)過第二層卷積后：

x = self.conv2(x, edge_index)x = F.relu(x)x = F.dropout(x, training=self.training)x = self.conv2(x, edge_index)
x = F.relu(x)
x = F.dropout(x, training=self.training)

此時(shí)得到的x的size應(yīng)該為：

torch.Size([3327, 6])

即每個(gè)節(jié)點(diǎn)的維度為6的狀態(tài)向量。

由于我們需要進(jìn)行6分類，所以最后需要加上一個(gè)softmax：

x = F.softmax(x, dim=1)

dim=1表示對每一行進(jìn)行運(yùn)算，最終每一行之和加起來為1，也就表示了該節(jié)點(diǎn)為每一類的概率。輸出此時(shí)的x：

tensor([[0.1607, 0.1727, 0.1607, 0.1607, 0.1607, 0.1846], [0.1654, 0.1654, 0.1654, 0.1654, 0.1654, 0.1731], [0.1778, 0.1622, 0.1733, 0.1622, 0.1622, 0.1622], ..., [0.1659, 0.1659, 0.1659, 0.1704, 0.1659, 0.1659], [0.1667, 0.1667, 0.1667, 0.1667, 0.1667, 0.1667], [0.1641, 0.1641, 0.1658, 0.1766, 0.1653, 0.1641]], device='cuda:0', grad_fn=<SoftmaxBackward0>)tensor([[0.1607, 0.1727, 0.1607, 0.1607, 0.1607, 0.1846],
        [0.1654, 0.1654, 0.1654, 0.1654, 0.1654, 0.1731],
        [0.1778, 0.1622, 0.1733, 0.1622, 0.1622, 0.1622],
        ...,
        [0.1659, 0.1659, 0.1659, 0.1704, 0.1659, 0.1659],
        [0.1667, 0.1667, 0.1667, 0.1667, 0.1667, 0.1667],
        [0.1641, 0.1641, 0.1658, 0.1766, 0.1653, 0.1641]], device='cuda:0',
       grad_fn=<SoftmaxBackward0>)

2. 反向傳播

在訓(xùn)練時(shí)，我們首先利用前向傳播計(jì)算出輸出：

out = model(data)

out即為最終得到的每個(gè)節(jié)點(diǎn)的6個(gè)概率值，但在實(shí)際訓(xùn)練中，我們只需要計(jì)算出訓(xùn)練集的損失，所以損失函數(shù)這樣寫：

loss = loss_function(out[data.train_mask], data.y[data.train_mask])

然后計(jì)算梯度，反向更新！

3. 訓(xùn)練

訓(xùn)練的完整代碼：

def train(): optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=5e-4) loss_function = torch.nn.CrossEntropyLoss().to(device) model.train() for epoch in range(500): out = model(data) optimizer.zero_grad() loss = loss_function(out[data.train_mask], data.y[data.train_mask]) loss.backward() optimizer.step() print('Epoch {:03d} loss {:.4f}'.format(epoch, loss.item()))def train():
    optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=5e-4)
    loss_function = torch.nn.CrossEntropyLoss().to(device)
    model.train()
    for epoch in range(500):
        out = model(data)
        optimizer.zero_grad()
        loss = loss_function(out[data.train_mask], data.y[data.train_mask])
        loss.backward()
        optimizer.step()
        print('Epoch {:03d} loss {:.4f}'.format(epoch, loss.item()))

4. 測試

我們首先需要算出模型對所有節(jié)點(diǎn)的預(yù)測值：

model(data)

此時(shí)得到的是每個(gè)節(jié)點(diǎn)的6個(gè)概率值，我們需要在每一行上取其最大值：

model(data).max(dim=1)

輸出一下：

torch.return_types.max(
values=tensor([0.9100, 0.9071, 0.9786,  ..., 0.4321, 0.4009, 0.8779], device='cuda:0',
       grad_fn=<MaxBackward0>),
indices=tensor([3, 1, 5,  ..., 3, 1, 5], device='cuda:0'))

返回的第一項(xiàng)是每一行的最大值，第二項(xiàng)為最大值在這一行中的索引，我們只需要取第二項(xiàng)，那么最終的預(yù)測值應(yīng)該寫為：

_, pred = model(data).max(dim=1)

然后計(jì)算預(yù)測精度：

correct = int(pred[data.test_mask].eq(data.y[data.test_mask]).sum().item())
acc = correct / int(data.test_mask.sum())
print('GCN Accuracy: {:.4f}'.format(acc))