Error

[Jasmine302]

I'm getting this error on the hotel dataset when reproducing your code
Traceback (most recent call last):
File "train.py", line 201, in
main()
File "train.py", line 181, in main
train(epoch)
File "train.py", line 106, in train
V_init, V_pred, V_refi, valid_mask = model(S_obs, S_trgt)
File "/home/jnu/xxq/Social-STGCNN-master/venv/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "/home/jnu/xxq/GraphTERN-main/graphtern/model.py", line 129, in forward
mix = Categorical(torch.nn.functional.softmax(temp[:, :, :, 4], dim=-1))
File "/home/jnu/xxq/Social-STGCNN-master/venv/lib/python3.8/site-packages/torch/distributions/categorical.py", line 66, in init
super(Categorical, self).init(batch_shape, validate_args=validate_args)
File "/home/jnu/xxq/Social-STGCNN-master/venv/lib/python3.8/site-packages/torch/distributions/distribution.py", line 56, in init
raise ValueError(
ValueError: Expected parameter probs (Tensor of shape (1, 25, 8)) of distribution Categorical(probs: torch.Size([1, 25, 8])) to satisfy the constraint Simplex(), but found invalid values:
tensor([[[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, nan, nan, nan]]], device='cuda:0',
grad_fn=)
Have you encountered?

[InhwanBae]

Hi @Jasmine302, Your error seems to be the same as the one I noticed when I recently reran the source code. As mentioned in
issue#1, it might help to use PyTorch version 11.1 or below, or to try another seed.

The method of the torch.distributions.log_prob() operation for probability distribution in the latest version of PyTorch has changed, which seems to cause problems with loss backpropagation. If any other issues arise, let me know!

[Jasmine302]

Hi @Jasmine302, Your error seems to be the same as the one I noticed when I recently reran the source code. As mentioned in issue#1, it might help to use PyTorch version 11.1 or below, or to try another seed.

The method of the torch.distributions.log_prob() operation for probability distribution in the latest version of PyTorch has changed, which seems to cause problems with loss backpropagation. If any other issues arise, let me know!

I changed the pytorch version and commented out the seeds as you did, and it's fixed, thanks a lot!

[3123155028]

Hi @Jasmine302, Your error seems to be the same as the one I noticed when I recently reran the source code. As mentioned in issue#1, it might help to use PyTorch version 11.1 or below, or to try another seed.
The method of the torch.distributions.log_prob() operation for probability distribution in the latest version of PyTorch has changed, which seems to cause problems with loss backpropagation. If any other issues arise, let me know!

I changed the pytorch version and commented out the seeds as you did, and it's fixed, thanks a lot!

hello，what's your torch and python vision

[Jasmine302]

Hi @Jasmine302, Your error seems to be the same as the one I noticed when I recently reran the source code. As mentioned in issue#1, it might help to use PyTorch version 11.1 or below, or to try another seed.
The method of the operation for probability distribution in the latest version of PyTorch has changed, which seems to cause problems with loss backpropagation. If any other issues arise, let me know!torch.distributions.log_prob()

I changed the pytorch version and commented out the seeds as you did, and it's fixed, thanks a lot!

hello，what's your torch and python vision

pytorch 1.10.1 ;python 3.7.0

[3123155028]

Hi @Jasmine302, Your error seems to be the same as the one I noticed when I recently reran the source code. As mentioned in issue#1, it might help to use PyTorch version 11.1 or below, or to try another seed.
The method of the operation for probability distribution in the latest version of PyTorch has changed, which seems to cause problems with loss backpropagation. If any other issues arise, let me know!torch.distributions.log_prob()

I changed the pytorch version and commented out the seeds as you did, and it's fixed, thanks a lot!

hello，what's your torch and python vision

pytorch 1.10.1 ;python 3.7.0

thank you! AND what your seed?i change many torch vesion on my dataset(sdd),it always generates error,