e2db3b3f1d3d23cd5bc1e295835e0f4b33e95447,examples/cora_gcn.py,,,#,57

Before Change

    return pred.eq(data.target.data[mask]).sum() / mask.size(0)


acc = []
for run in range(1, 101):
    model.conv1.reset_parameters()
    model.conv2.reset_parameters()

    old_val = 0
    cur_test = 0
    for _ in range(0, 200):
        train()
        val = test(val_mask)
        if val > old_val:
            old_val = val
            cur_test = test(test_mask)

    acc.append(cur_test)
    print("Run:", run, "Test Accuracy:", acc[-1])

acc = torch.FloatTensor(acc)
print("Mean:", acc.mean(), "Stddev:", acc.std())

After Change

        return F.log_softmax(x, dim=1)


model = Net()
if torch.cuda.is_available():
    train_mask, val_mask = train_mask.cuda(), val_mask.cuda()
    test_mask, model = test_mask.cuda(), model.cuda()

optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=0.005)

Instances