Skip to content

Covariance Features

NeuRepTrace exposes dataset-independent covariance feature helpers for M/EEG workflows through neureptrace.decoding.covariance_features. These helpers are usable from project-specific compatibility repositories without importing a BUSH-MEG CLI module.

The feature extractor supports the PyMEGDec-compatible representations used by the BUSH-MEG covariance LOSO workflow:

  • logeuclidean_covariance
  • covariance_upper
  • correlation_upper
  • variance

For a single trial shaped channels x time, use covariance_feature_vector. For a full array shaped trials x channels x time, use window_covariance_features with a CovarianceWindow.

Reusable entry points:

  • CovarianceWindow
  • covariance_feature_vector
  • window_covariance_features
  • normalize_covariance_feature_mode
  • trial_covariance
  • channel_subset_indices