Improving the transforms

The most recent implementation of transforms asks for an OrderedDict of callables or mne.Raw/Epochs methods that are called successively on the internal Raw/Epochs objects in a ConcatDataset (https://github.com/braindecode/braindecode/blob/master/braindecode/datautil/transforms.py#L18). I wanted to discuss possible improvements to this design:

1. The current design requires the callables or methods to act in-place, i.e., they must directly modify the Raw/Epochs object. However, apart from .resample and .filter, most mne methods do not seem to act in-place. As for callables, they can be made to modify the objects in-place, but this would mean all the transforms would have to be mne object-aware. I don’t have a specific solution in mind, but maybe we could discuss this here.
2. There is currently no way to apply a transform on-the-fly, i.e., every time a window is about to be returned. This is central to preprocessing steps in a lazy loading scenario however. The design we had started to implement during the sprint used Transform objects that were saved in the dataset object, and called inside __getitem__. This is how torch-vision does it for instance.

What are your thoughts @gemeinl @sbbrandt ?

[robintibor]

I will have a try with that right now actually since it is giving unexpected results with dict. we can still see which interface we prefer after.

[robintibor]

oki, now implemented in #94