fix kernel_size np.int64 dtype in torchinfo

[sliwy]

It solves #537 where torchinfo was throwing wrong kernel_size type.

[codecov]

Codecov Report

Merging #538 (5fddbc4) into master (ae843fe) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #538   +/-   ##
=======================================
  Coverage   84.48%   84.48%           
=======================================
  Files          63       63           
  Lines        4653     4653           
=======================================
  Hits         3931     3931           
  Misses        722      722           

[bruAristimunha]

Hey @sliwy,

I searched and look's like the Chambon is impact by this issue https://github.com/search?q=repo%3Abraindecode%2Fbraindecode%20.astype(int)&type=code

Can you make a small test to test all the models? You can use the models_dict proposed by @PierreGtch.

[PierreGtch]

On the same topic, it would be nice if we had, for every model, a set parameters that can initialise it. This would be useful for testing if the implementations are conform and all the methods of EEGModuleMixin working on every model.

[sliwy]

Hey @bruAristimunha, I checked and for SleepStagerChambon2018 it works :) no clue why 😂

For the test, definitely, it could be something like this in the beginning but we need to know parameters to use or set good default values:

from braindecode.models.util import models_dict
@pytest.mark.parametrize("model_cls", models_dict.values())
def test_model_torchinfo(model_cls):
    print(model_cls(n_times=5000, n_chans=1, n_outputs=2))

[PierreGtch]

@sliwy not sure all the models can be initialized with their default parameters.

Plus I think we should test if the models accept inputs with the desired shape and return outputs with the correct shape:

from braindecode.models.util import models_dict
@pytest.mark.parametrize("model_cls", models_dict.values())
def test_model_torchinfo(model_cls):
    model = model_cls(n_times=5000, n_chans=1, n_outputs=2)
    assert model(torch.ones(3, 1, 5000)).shape == (3, 2)

[bruAristimunha]

Let's use two shapes:

For the BCI models, we will use BCI 2014 (1, 22, 1001),
For sleep, we will use Sleep Dataset (1, 2, 3000)./

[bruAristimunha]

As we have already opened other issues to address the requested extra, I will consider this issue closed and we will continue with another PR. Thank you very much @sliwy!!