preprocessing.py --n_ses=1 errors out on broadcasting shape mismatch

[3x10e8]

Dear developers, Thank you so much for making your code available!

I was going through the steps to extract the raw data into epochs, and tried reducing the n_ses parameter value from 4 (default) to 1 to only get a subset of the data:

https://github.com/gifale95/eeg_encoding/blob/0e048f7c6ae3aec0495b288662389fd51b9b1eaa/02_eeg_preprocessing/preprocessing.py#L34

But this results in a shape error for the line merged_train[i] = ordered_data in the context:

https://github.com/gifale95/eeg_encoding/blob/0e048f7c6ae3aec0495b288662389fd51b9b1eaa/02_eeg_preprocessing/preprocessing_utils.py#L254-L262

The error is: ValueError: could not broadcast input array from shape (2,17,100) into shape (4,17,100)

I can try to fix it later, I just wanted to log it here in case anyone else sees this on trying to reduce n_ses. The test epochs get extracted correctly, only the training epochs seem to be affected.

[gifale95]

Hi Abhinav, thanks for pointing this out!

The preprocessing script you mentioned, along with the following scripts to reproduce the paper results, were intentionally written assuming EEG data from 4 sessions.

This being said, if you wish to preprocess the data from only 1 session, you can replace the code you printed above with:

    # Data matrix of shape:
    # Image conditions × EEG repetitions × EEG channels × EEG time points
    if args.n_ses == 1:
        reps = white_data.shape[1]
    else:
        reps = white_data.shape[1] * 2
    merged_train = np.zeros((len(np.unique(img_cond)), reps,
        white_data.shape[2], white_data.shape[3]))
    for i in range(len(np.unique(img_cond))):
        # Find the indices of the selected category
        idx = np.where(img_cond == i+1)[0]
        for r in range(len(idx)):
            if r == 0:
                ordered_data = white_data[idx[r]]
            else:
                ordered_data = np.append(ordered_data, white_data[idx[r]], 0)
        merged_train[i] = ordered_data

However, note that the remaining code of the repository assumes EEG data from 4 sessions, and therefore might not work on data from only 1 session.

Hope this helps! Ale