whitener.py

def compute_whitener(noise_cov, info, rank=None):
    """Compute whitening matrix."""
    from mne.cov import _get_whitener
    from mne.defaults import _handle_default
    from mne.io.pick import pick_channels_cov, _pick_data_channels
    picks = _pick_data_channels(info, with_ref_meg=False, exclude='bads')
    ch_names = [info['ch_names'][k] for k in picks]
    noise_cov = pick_channels_cov(noise_cov, include=ch_names, exclude=[])
    if len(noise_cov['data']) != len(ch_names):
        missing = list(set(ch_names) - set(noise_cov['names']))
        raise RuntimeError('Not all channels present in noise covariance:\n%s'
                           % missing)
    scalings = _handle_default('scalings_cov_rank', None)
    W, noise_cov, n_nzero = _get_whitener(noise_cov, info, ch_names, rank,
                                          pca=True, scalings=scalings)

    # Do the back projection
    assert W.shape[0] == n_nzero
    return W


import mne  # noqa, analysis:ignore
data_path = mne.datasets.sample.data_path() + '/MEG/sample/'
fname_cov = data_path + 'sample_audvis-cov.fif'
fname_info = data_path + 'sample_audvis_raw.fif'

cov = mne.read_cov(fname_cov)
info = mne.io.read_info(fname_info)
whitener = compute_whitener(cov, info)
assert whitener.shape[1] == 364  # good MEG+EEG chs
assert whitener.shape[0] < 364  # rank-reduced
print(whitener.shape)