I can print it but I can't use it

[Kevin-shihello-world]

It was really weird to find It shows IndexError: index 2 is out of bounds for dimension 0 with size 2 for a while it can print a.
I want to do something with the output of my GNN model and the whole code of this part is this

correct = 0
        for i, data in enumerate(train_loader):
            
            optimizer.zero_grad()
            data = data.to(args.device)
            data = data.to(args.device)
            ##print(data.validate())
            out = model(data)
            loss_0 = F.nll_loss(out, data.y)
            with torch.no_grad():
                loss_1_ = 1+torch.abs(F.nll_loss(out[0,:], data.y[0])-loss_0)
                loss_2_ = 1+torch.abs(F.nll_loss(out[1,:], data.y[1])-loss_0) 
                a = data.y[2]
                if i == 1:
                    print("data.y[2]:", data.y[2])
                
                b = out[2,:]
                loss_3_ = 1+torch.abs(F.nll_loss(b, a)-loss_0)
            loss_ = loss_1_*F.nll_loss(out[0,:], data.y[0])+loss_2_*F.nll_loss(out[1,:], data.y[1])+loss_3_*F.nll_loss(out[2,:], data.y[2])
            #loss = loss_.cuda()
            loss = F.nll_loss(out, data.y)##,weight = loss_)
            loss.backward()

and it does print data.y[2] while it shows IndexError: index 2 is out of bounds for dimension 0 with size 2 on a = data.y[2].
The thing is it showed out is a [3,2] Tensor and those lines above like' loss_2_ = 1+torch.abs(F.nll_loss(out[1,:], data.y[1])-loss_0) 'didn't went wrong,it's very weird to find this situation.
Update:
When I use a tensor 1 in cuda service it shows this

python
loss_3_ = 1+torch.abs(F.nll_loss(b, a)-loss_0)
  File "/root/miniconda3/envs/myconda/lib/python3.8/site-packages/torch/nn/functional.py", line 2671, in nll_loss
    return torch._C._nn.nll_loss_nd(input, target, weight, _Reduction.get_enum(reduction), ignore_index)
RuntimeError: size mismatch (got input: [2], target: [1])

Things's getting more wired,why it did not show size mismatch for loss_2_ etc~
Could someone help me find the answer? I would really appreciate it if someone do.

[LukeLIN-web]

I couldn't understand the problems. May you list your problems by point so that it may be more clear?

[Kevin-shihello-world]

Oops, I'm sorry,HGP-sl is a GNN model been top on the PROTEINS dataset whose maker posted it on GitHub so I thought people may search it on the GitHub and find it. I was been insensitive. See this page:https://github.com/cszhangzhen/HGP-SL/blob/master/main.py and you can find it.
Best wishes

[rusty1s]

There is a bug in the PROTEINS dataset for PyG 2.1. We will release PyG 2.2, containing the fix. Alternatively, you can just upgrade PyG to latest master (more information can be found in the README.md).

[Kevin-shihello-world]

Thank you, Mr. Fey! But which kind of bug would it be? How could a parameter can be printed but not be used?I want learn from it and I'll mark your answer as the answer.
Best wishes!

[rusty1s]

Oh, the bug is that the PROTEINS dataset failed to load the node features, leading to a feature matrix of shape (N, 0).

[Kevin-shihello-world]

Oh, I'm sorry, my connection to the internet was off for a while and then I went to attend my classes. And thank your answer @rusty1s I really learned a lot from it. By the way, there are two of you who answer this question and I think Mr.Fey's more helpful but how to mark the latter one to be the suggested answer to this problem?
Thank you two