Function _ica_explained_variance

mne/preprocessing/ica.py:2843–2887

Check variance accounted for by each component in supplied data. This function is only used for sorting the components. Parameters ---------- ica : ICA Instance of `mne.preprocessing.ICA`. inst : Raw | Epochs | Evoked Data to explain with ICA. Instance of Raw, E

(ica, inst, normalize=False)

def _ica_explained_variance(ica, inst, normalize=False):
    """Check variance accounted for by each component in supplied data.

    This function is only used for sorting the components.

    Parameters
    ----------
    ica : ICA
        Instance of `mne.preprocessing.ICA`.
    inst : Raw | Epochs | Evoked
        Data to explain with ICA. Instance of Raw, Epochs or Evoked.
    normalize : bool
        Whether to normalize the variance.

    Returns
    -------
    var : array
        Variance explained by each component.
    """
    # check if ica is ICA and whether inst is Raw or Epochs
    if not isinstance(ica, ICA):
        raise TypeError("first argument must be an instance of ICA.")
    if not isinstance(inst, BaseRaw | BaseEpochs | Evoked):
        raise TypeError(
            "second argument must an instance of either Raw, Epochs or Evoked."
        )

    source_data = _get_inst_data(ica.get_sources(inst))

    # if epochs - reshape to channels x timesamples
    if isinstance(inst, BaseEpochs):
        n_epochs, n_chan, n_samp = source_data.shape
        source_data = source_data.transpose(1, 0, 2).reshape(
            (n_chan, n_epochs * n_samp)
        )

    n_chan, n_samp = source_data.shape
    var = (
        np.sum(ica.mixing_matrix_**2, axis=0)
        * np.sum(source_data**2, axis=1)
        / (n_chan * n_samp - 1)
    )
    if normalize:
        var /= var.sum()
    return var


def _sort_components(ica, order, copy=True):