autoreject.AutoReject¶

class autoreject.AutoReject(n_interpolate=None, consensus=None, thresh_func=None, cv=10, picks=None, thresh_method='bayesian_optimization', n_jobs=1, random_state=None, verbose=True)[source]¶

Efficiently find n_interpolate and consensus.

Note

AutoReject by design supports multiple channels. If no picks are passed, separate solutions will be computed for each channel type and internally combined. This then readily supports cleaning unseen epochs from the different channel types used during fit.

Parameters

n_interpolatearray | None

The values to try for the number of channels for which to interpolate. This is \(\\rho\). If None, defaults to np.array([1, 4, 32])

consensusarray | None

The values to try for percentage of channels that must agree as a fraction of the total number of channels. This sets \(\\kappa/Q\). If None, defaults to np.linspace(0, 1.0, 11)

cva scikit-learn cross-validation object

Defaults to cv=10

picksstr | list | slice | None

Channels to include. Slices and lists of integers will be interpreted as channel indices. In lists, channel type strings (e.g., ['meg', 'eeg']) will pick channels of those types, channel name strings (e.g., ['MEG0111', 'MEG2623'] will pick the given channels. Can also be the string values 'all' to pick all channels, or 'data' to pick data channels. None (default) will pick data channels {‘meg’, ‘eeg’}, which will lead fitting and combining autoreject solutions across these channel types. Note that channels in info['bads'] will be included if their names or indices are explicitly provided.

thresh_methodstr

‘bayesian_optimization’ or ‘random_search’

n_jobsint

The number of jobs.

random_stateint seed, RandomState instance, or None (default)

The seed of the pseudo random number generator to use.

verbosebool

The verbosity of progress messages. If False, suppress all output messages.

Attributes

local_reject_list

The instances of _AutoReject for each channel type.

threshes_dict

The sensor-level thresholds with channel names as keys and the peak-to-peak thresholds as the values.

loss_dict of array, shape (len(n_interpolate), len(consensus))

The cross validation error for different parameter values.

consensus_dict

The estimated consensus per channel type.

n_interpolate_dict

The estimated n_interpolate per channel type.

picks_array_like, shape (n_data_channels,)

The data channels considered by autoreject. By default only data channels, not already marked as bads are considered.

__init__(n_interpolate=None, consensus=None, thresh_func=None, cv=10, picks=None, thresh_method='bayesian_optimization', n_jobs=1, random_state=None, verbose=True)[source]¶

Init it.

Methods

__init__([n_interpolate, consensus, ...])	Init it.
fit(epochs)	Fit the epochs on the AutoReject object.
fit_transform(epochs[, return_log])	Estimate the rejection params and finds bad epochs.
get_reject_log(epochs[, picks])	Get rejection logs of epochs.
save(fname[, overwrite])	Save autoreject object with the HDF5 format.
transform(epochs[, return_log])	Remove bad epochs, repairs sensors and returns clean epochs.