sts_cov_estimators.h File Reference

Regularised covariance estimators for M/EEG noise covariances, matching MNE-Python's compute_covariance() API. More...

#include "sts_global.h"
#include <Eigen/Core>
#include <utility>
#include <vector>
#include <string>

Include dependency graph for sts_cov_estimators.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
class	STSLIB::StsCovEstimators
	Regularised covariance estimators (Ledoit-Wolf, OAS, fixed-diagonal, PCA, Factor Analysis, cross-validated auto-select) matching the MNE-Python compute_covariance() API. More...

Namespaces
namespace	STSLIB
	Statistical testing (t-tests, F-tests, cluster permutation, multiple comparison correction).

Detailed Description

Regularised covariance estimators for M/EEG noise covariances, matching MNE-Python's compute_covariance() API.

SPDX-License-Identifier: BSD-3-Clause Copyright (c) 2026 MNE-CPP Authors

Author

Christoph Dinh chris.nosp@m.toph.nosp@m..dinh.nosp@m.@mne.nosp@m.-cpp..nosp@m.org

Since

2.2.0

Date

April 2026

M/EEG sensor counts routinely exceed the number of clean baseline samples, which makes the unregularised sample covariance rank-deficient and unusable as the C operator in MNE/dSPM/sLORETA inverse solutions. This module provides the same family of regularised estimators as mne.compute_covariance with method='auto', so STSLIB users get a numerically well-conditioned covariance regardless of how undersampled the input is.

Six estimators are exposed: the analytic Ledoit-Wolf shrinkage, Oracle Approximating Shrinkage (OAS), fixed diagonal regularisation \(C+\lambda\,\bar{\sigma}^2 I\), rank-reduced PCA, EM Factor Analysis and a cross-validated auto-selector that picks the estimator with the highest held-out Gaussian log-likelihood. All routines take zero-mean data shaped (n_channels, n_samples) and return a (cov, parameter) pair; the auto-selector additionally returns the index of the winning method.

The Ledoit-Wolf and OAS back-ends are thin wrappers over the scikit-learn-compatible Skigen::LedoitWolf and Skigen::OAS classes so the resulting covariances are numerically identical to those produced by MNE-Python.

References: Ledoit & Wolf (2004), J. Multivariate Anal. 88(2); Chen, Wiesel, Eldar & Hero (2010), IEEE TSP 58(10); Engemann & Gramfort (2015), NeuroImage 108.

Definition in file sts_cov_estimators.h.