mne_denoise.dss.compute_dss

mne_denoise.dss.compute_dss(covariance_baseline: ndarray, covariance_biased: ndarray, *, n_components: int | None = None, rank: int | None = None, reg: float = 1e-09, verbose: bool | str | int | None = None) → tuple[ndarray, ndarray, ndarray]

Compute DSS spatial filters from baseline and biased covariances.

Parameters:

covariance_baselinendarray, shape (n_channels, n_channels)

Baseline covariance defining the total-power metric.

covariance_biasedndarray, shape (n_channels, n_channels)

Biased covariance defining the signal-of-interest metric.

n_componentsint or None, default=None

Number of components to return. None returns the available rank.

rankint or None, default=None

Whitening rank. None estimates the rank from the baseline covariance.

regfloat, default=1e-9

Relative eigenvalue threshold used during whitening.

verbosebool, str, int, or None, default=None

MNE-style logging level.

Returns:

filtersndarray, shape (n_components, n_channels)

DSS spatial filters.

patternsndarray, shape (n_channels, n_components)

DSS spatial patterns.

eigenvaluesndarray, shape (n_components,)

Biased-to-baseline variance ratios.

See also

DSS

Estimator that learns and applies the decomposition to recordings.

Notes

The baseline covariance is whitened, the biased covariance is diagonalized in that space, and the resulting filters are normalized in the baseline metric. This implementation follows the linear DSS formulation [1].

References

[1]

Jaakko Sarela and Harri Valpola. Denoising source separation. Journal of Machine Learning Research, 6(9):233–272, 2005. URL: https://jmlr.org/papers/v6/sarela05a.html.

Examples

>>> import numpy as np
>>> from mne_denoise.dss import compute_dss
>>> rng = np.random.default_rng(0)
>>> data = rng.standard_normal((8, 2000))
>>> biased_data = data + 0.1 * rng.standard_normal(data.shape)
>>> baseline = np.cov(data)
>>> biased = np.cov(biased_data)
>>> filters, patterns, scores = compute_dss(baseline, biased, n_components=3)