[BUG] Inconsistent lengths of indices and rank does not raise an error

[tsbinns]

Problem

A small but notable bug I just came across:

indices and rank should both have shape (2, n_cons), however if the number of connections differs, e.g.

indices = ([[0, 1], [2, 3]],  # 2 seeds
           [[4, 5], [6, 7]])  # 2 targets

rank = ([2],  # 1 seed
        [2])  # 1 target

this does not immediately raise an error. Instead, the use of zip elsewhere in the code to iterate over the entries of indices and rank for each connection means that only the results for the smallest number of connections (in the above example 1 of 2) would be processed.

Example of where zip is used:

mne-connectivity/mne_connectivity/spectral/epochs_multivariate.py

Lines 185 to 186 in 28c3895

	for seed_idcs, target_idcs, seed_rank, target_rank in zip(
	indices[0], indices[1], ranks[0], ranks[1]):

Proposed fix

Simply adding a check in _check_rank_input that the number of connections in indices and rank match is enough to catch this early on:

mne-connectivity/mne_connectivity/spectral/epochs_multivariate.py

Lines 51 to 56 in 28c3895

	if (
	len(rank) != 2 or len(rank[0]) != len(indices[0]) or
	len(rank[1]) != len(indices[1])
	):
	raise ValueError('rank argument must have shape (2, n_cons), '
	'according to n_cons in the indices')

I have also updated the unit tests to reflect this change.

Sorry for letting this slip through!

[larsoner]

Thanks @tsbinns !