channels.detect_bad_channels#

autoclean.functions.artifacts.channels.detect_bad_channels(data, correlation_thresh=0.35, deviation_thresh=2.5, ransac_sample_prop=0.35, ransac_corr_thresh=0.65, ransac_frac_bad=0.25, ransac_channel_wise=False, random_state=1337, exclude_channels=None, return_by_method=False, verbose=None)[source]#

Detect bad channels using multiple statistical methods.

This function uses the PyPREP NoisyChannels implementation to detect bad channels through correlation, deviation, and RANSAC-based methods. These methods are based on established preprocessing pipelines and provide robust detection of various types of channel artifacts.

The function implements multiple complementary detection approaches: - Correlation-based: Identifies channels poorly correlated with neighbors - Deviation-based: Finds channels with excessive amplitude deviations - RANSAC-based: Uses robust regression to detect outlier channels

Parameters:

datamne.io.BaseRaw: The continuous EEG data to analyze for bad channels.
correlation_threshfloat, default 0.35: Threshold for correlation-based bad channel detection. Channels with correlation below this value with their neighbors are marked as bad. Lower values are more stringent.
deviation_threshfloat, default 2.5: Threshold for deviation-based detection in standard deviations. Channels exceeding this many SDs from the mean are marked as bad.
ransac_sample_propfloat, default 0.35: Proportion of samples to use for RANSAC detection (0.0-1.0). Higher values use more data but are computationally more expensive.
ransac_corr_threshfloat, default 0.65: Correlation threshold for RANSAC-based detection. Set to 0 to disable RANSAC detection entirely.
ransac_frac_badfloat, default 0.25: Expected fraction of bad channels for RANSAC algorithm (0.0-1.0). Should reflect prior knowledge about data quality.
ransac_channel_wisebool, default False: Whether to perform RANSAC detection channel-wise rather than globally. Channel-wise detection can be more sensitive but slower.
random_stateint, default 1337: Random seed for reproducible RANSAC results.
exclude_channelslist of str or None, default None: Channel names to exclude from bad channel detection (e.g., reference channels). These channels will never be marked as bad.
return_by_methodbool, default False: If True, returns a dictionary with bad channels separated by detection method. If False, returns a combined list of all bad channels.
verbosebool or None, default None: Control verbosity of output during detection.

Returns:

bad_channelslist of str or dict: If return_by_method=False: List of channel names detected as bad. If return_by_method=True: Dictionary with keys ‘correlation’, ‘deviation’, ‘ransac’, and ‘combined’ containing lists of bad channels for each method.