robintibor

Sharpness Aware Minimization

7

Hi, thanks so much for maintaining a scikit-learn compatible PyTorch framework. We were wondering what would you consider the best way to use Sharpness Aware Minimization (https://github.com/davda54/sam https://arxiv.org/abs/2010.01412) together with...

Make input_psmac_dirs default to <output_dir>*?

3

This is a question/feature request :) I wondered if one could set the input_psmac_dirs default to * , so that one does not have to keep both parameters in sync...

Sphinx Doc Warnings

3

We have a lot of Sphinx Doc Warnings, some of which we don't know where they come from. We should aim towards zero warnings. Some examples: ``` /home/robintibor/work/braindecode-dev/braindecode/braindecode/models/tcn.py:docstring of torch.nn.modules.module.Module.to:12:...

Democracy in Action For Naming

15

So right now in quite some models we have: https://github.com/braindecode/braindecode/blob/4eb0af49824d764d771792931511b449a092f49f/braindecode/models/deep4.py#L38-L39 Let's vote how to change it!

[No Code] More models for braindecode

9

One could have a look at the methods here, and test which ones improve EEG decoding, e.g., on High-Gamma Dataset/BCI Competition IV 2a/TUH Abnormal @Div12345 https://pytorch.org/blog/how-to-train-state-of-the-art-models-using-torchvision-latest-primitives/

predict_trials enforce eval mode?

1

Do we want to enforce eval mode in predict_trials member methods of EEGClassifier and EEGRegressor? Seems we don't do that?

Multiversion Sphinx

1

We would like to have a toggle button to switch between master and other versions like mne has https://mne.tools As suggested in https://github.com/braindecode/braindecode/pull/248 , we could use sphinx-multiversion: > have...

Update HGD loading in moabb and add location of our HGD Sensors

3

Could be good to make a PR to MOABB so that our sensor locations are in info of mne raw created when loading HGD dataset. Atm often doing some manual...

Add Specific Data Augmentations

1

Could be nice to add standard augmentations like MixUp/Cutout into Braindecode, of course with some examples

Maximally activating example

3

I have a code example to extract the most activating windows for a unit here: https://gist.github.com/robintibor/72ee611e7b34c188a90e6e2d1d49a70f Could be nice to make it nicer into a full example, potentially also adding...