This PR introduces parameters log and ratios_triu to compute_pow_freq_bands (again, if there's interest in including this in the library!). This one might require some discussion.
log controls whether the power is log-transformed. If log=True, the band ratios are calculated as "log-ratios" (e.g., log(alpha_pow/beta_pow)). Since this then leads to a linear relationship between a log-ratio an its log-inverse ratio (i.e., log(alpha_pow/beta_pow) = -1 * log(beta_pow/alpha_pow)), I added an additional parameter, ratios_triu which controls whether we should compute all possible pairs (which is currently the default behaviour) or only the upper part of the matrix of possible pairs. I included tests for the different scenarios, and ended up dividing the test for compute_pow_freq_bands into a couple smaller ones.
agramfort
commented
3 years ago
This PR introduces parameters
logandratios_triutocompute_pow_freq_bands(again, if there's interest in including this in the library!). This one might require some discussion.logcontrols whether the power is log-transformed. Iflog=True, the band ratios are calculated as "log-ratios" (e.g., log(alpha_pow/beta_pow)). Since this then leads to a linear relationship between a log-ratio an its log-inverse ratio (i.e., log(alpha_pow/beta_pow) = -1 * log(beta_pow/alpha_pow)), I added an additional parameter,ratios_triuwhich controls whether we should compute all possible pairs (which is currently the default behaviour) or only the upper part of the matrix of possible pairs. I included tests for the different scenarios, and ended up dividing the test forcompute_pow_freq_bandsinto a couple smaller ones.