problems when running GNP Classification Experiments

[git0405]

Thanks for you help first.

1. the first error i met it
   "plum.resolver.NotFoundLookupError: For function device, (device(type='cpu'),) could not be resolved."

Because i run on cpu, so I just replace the "with B.device(B.device(z)):" in "class SetConv1dEncoder(nn.Module):" of "encoder.py" with
"with B.device("cpu"):"

2. Then error goes to "decoder.py".

Traceback (most recent call last):
  File "train_pac.py", line 420, in <module>
    tb_writer=writer, global_step=global_step)
  File "train_pac.py", line 46, in train
    loss_dict = model.loss(task['x_context'], task['y_context'])
  File "C:\Users\Desktop\experiments for test set bound\pac-bayes-tightness-small-data-master2\pac-bayes-tightness-small-data-master\pacbayes\classifiers.py", line 481, in loss
    post = self.architecture.infer(x, y)
  File "C:\Users\Desktop\experiments for test set bound\pac-bayes-tightness-small-data-master2\pac-bayes-tightness-small-data-master\pacbayes\architectures.py", line 52, in infer
    means, covs = self.inference_network(x, y)  # [B, D], [B, D, D]
  File "C:\Users\.conda\envs\py37\lib\site-packages\torch\nn\modules\module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "C:\Users\Desktop\experiments for test set bound\pac-bayes-tightness-small-data-master2\pac-bayes-tightness-small-data-master\pacbayes\gnp\gnp.py", line 71, in forward
    xz, z = self.encoder_mean(x_context, y_context, x_target)
  File "C:\Users\.conda\envs\py37\lib\site-packages\torch\nn\modules\module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "C:\Users\Desktop\experiments for test set bound\pac-bayes-tightness-small-data-master2\pac-bayes-tightness-small-data-master\pacbayes\gnp\encoder.py", line 34, in forward
    dists2 = B.pw_dists2(x_grid[None, :], xz)
  File "C:\Users\.conda\envs\py37\lib\site-packages\plum\function.py", line 392, in __call__
    return _convert(method(*args, **kw_args), return_type)
  File "C:\Users\.conda\envs\py37\lib\site-packages\lab\linear_algebra.py", line 502, in pw_dists2
    return norms_a + norms_b - 2 * B.matmul(a, b, tr_b=True)
  File "C:\Users\.conda\envs\py37\lib\site-packages\plum\function.py", line 392, in __call__
    return _convert(method(*args, **kw_args), return_type)
  File "C:\Users\.conda\envs\py37\lib\site-packages\lab\shape.py", line 185, in f_wrapped
    return f(*(unwrap_dimension(arg) for arg in args), **kw_args)
  File "C:\Users\.conda\envs\py37\lib\site-packages\lab\torch\linear_algebra.py", line 21, in matmul
    return torch.matmul(a, b)
RuntimeError: mat1 and mat2 shapes cannot be multiplied (480x1 and 128x1)

Any help will be greatly appreciated!!

[wesselb]

Hey @git0405! It seems that there have been some API changes since the release of this repo that broke the code. I've pushed some updates that I believe fix the issue.

[git0405]

Many thanks!!!

[wesselb]

Not a problem at all. :)