Refactoring fuse_with_classif_results

[lkorczowski]

utils.fuse_with_classif_results has many flaws :

• [ ] bad name
• [ ] we always has to invert valid_results for check_imp. I advice to take valid_labels instead as input.
• [ ] we accept multidimensionnal check_imp to do a mean in the end. IMO, it should only take one-dimensional array and remove "mean"
• [ ] clearly not enough pytested for the complexity of what it can do -> taking one-dimensional array will reduce complexity

def fuse_with_classif_result(check_imp, labels):
    """Adds at the good indexes the missing elements of labels because of the use of check_RMS

    Parameters
    ----------
    check_imp : list of list , shape (nb_epochs, nb_electrodes)
         0s and 1s 0s marking bad channels for the designated epoch
    labels : ndarray, shape (n_trials - nb_bad_epochs,)
        array of the classification prediction results

    Returns
    -------
    mod_labels : ndarray, shape (n_trials,)
        Modified labels to fit a length of n_trials
    """
    if not isinstance(labels, list):
        labels = labels.tolist()
    for i in range(len(check_imp)):
        if np.mean(check_imp[i]) == 1:
            labels.insert(i, False)  # inserting False (neutral) values in the labels array where they were deleted
    return labels