RuntimeError: shape '[200,-1]' is invalid for input of size 2

[fine4303]

Hello, I'm a newbie.
I had a little trouble running spiral/train_with_CBFs.py.
When I run "loss2.backward()", I find that it calls "y = y.contiguous().view(args.batch_time * args.batch_size,-1)" in dCBF(self, y, w_ref).
I found that y.shape = torch.Size([1,2]), so it says "RuntimeError: shape '[200,-1]' is invalid for input of size 2".

However, this command is required when I run "loss2 = func(batch_t, pred_y).to(device)".
I don't know how to solve this problem, please help me, thank you very much.

[fine4303]

Hello! I changed y.view (args.batch_time*args.batch_size, -1) to y.contiguous().view(-1, 2). It worked!

But new problems arose in spiral/train_with_CBFs.py.
Since self.training=true and self.use_dcbf=true, forward() returns loss.

I get an error RuntimeError: Mismatch in shape: grad_output[0] has a shape of torch.Size([1, 2]) and output[0] has a shape of torch.Size([]).
I think y and loss have different shapes.
I think this is the problem in loss2.backward(), but I don't know how to solve it.

Any help would be great! Thanks again for sharing the code!

[Weixy21]

It looks like the output is empty. The input to the dQP might have some problem. Not sure why y has a shape '[200,-1]. Can you check the dimension of all the inputs of dCBF function?