eegproc.data.windowing module

Create model arrays without allowing windows to cross trial boundaries.

class eegproc.data.windowing.WindowedArrays(features, labels, subject_ids, trial_ids, subject_lookup, trial_lookup)

Bases: object

Aligned windows, labels, integer identifiers, and original key lookups.

Parameters:

• features (ndarray)

• labels (ndarray)

• subject_ids (ndarray)

• trial_ids (ndarray)

• subject_lookup (dict[int, tuple])

• trial_lookup (dict[int, tuple])

features: ndarray

labels: ndarray

subject_ids: ndarray

subject_lookup: dict[int, tuple]

trial_ids: ndarray

trial_lookup: dict[int, tuple]

eegproc.data.windowing.to_supervised_arrays(frame, *, label_column=None, window_sec=None, window_samples=None, window_rows=1, overlap=0.0, normalize=None, label_transform=None, on_short_trial='error')

Window each trial separately and return channels-last float32 arrays.

kind='features' treats each row as a precomputed feature window. For signals, specify either window_sec or window_samples. Incomplete trailing windows are dropped. Trial identifiers are globally unique across subjects, including when subjects reuse the same trial names.

Optional subject_zscore uses all unlabeled rows from each subject. This includes held-out subjects’ own signals; use normalize=None and fold-specific preprocessing when the evaluation requires training-only normalization statistics.

Parameters:

• frame (EEGFrame)

• label_column (str | None, default: None)

• window_sec (float | None, default: None)

• window_samples (int | None, default: None)

• window_rows (int, default: 1)

• overlap (float, default: 0.0)

• normalize (str | None, default: None)

• label_transform (Optional[Callable], default: None)

• on_short_trial (str, default: 'error')

Return type:

WindowedArrays