Hello,
I would like to use braindecode for my own (not open) dataset where I have many trials and for every trial I have some additional information, multiple information items (several numbers, several strings) per trial. As of today the dataset is in BIDS format and I store the additional metadata as additional columns in the events.tsv . It would be hard to lump all this in an Annotations object because I need actual trial indices.
How to make dataloader return these items together with a window of data?
When I participated in EEGFoundationChallange it was done in a pretty hacky way: there was a function add_columns_extra (applied after create_windows_from_events) that was trying to perform the alignment of the data windows and annotations using just their indices. I believe that his is a very error-prone way to do this alignment and I would like to have something more solid because for my data this alignment is crucial.
Is it possible to do it with Braindecode as of today? If not, is it possible to get the original epoch indices (corresponding to event indices) to do the alignment myself? Or should I better modify create_windows_from_events and replace inside the use of mne.events_from_annotations with something custom)?
Hello,
I would like to use braindecode for my own (not open) dataset where I have many trials and for every trial I have some additional information, multiple information items (several numbers, several strings) per trial. As of today the dataset is in BIDS format and I store the additional metadata as additional columns in the events.tsv . It would be hard to lump all this in an Annotations object because I need actual trial indices.
How to make dataloader return these items together with a window of data?
When I participated in EEGFoundationChallange it was done in a pretty hacky way: there was a function add_columns_extra (applied after create_windows_from_events) that was trying to perform the alignment of the data windows and annotations using just their indices. I believe that his is a very error-prone way to do this alignment and I would like to have something more solid because for my data this alignment is crucial.
Is it possible to do it with Braindecode as of today? If not, is it possible to get the original epoch indices (corresponding to event indices) to do the alignment myself? Or should I better modify create_windows_from_events and replace inside the use of mne.events_from_annotations with something custom)?