eegproc.model_explainability.model_agnostic.plotting module

Array loading and plotting helpers independent of any dataset or model.

eegproc.model_explainability.model_agnostic.plotting.flatten_trial(trial)

Flatten chronological window/time axes to (samples, channels).

Parameters:

trial (ndarray)

Return type:

ndarray

eegproc.model_explainability.model_agnostic.plotting.load_counterfactual_trial(npz_path, *, branch=None, reference='input')

Load aligned reference and counterfactual trials from a saved archive.

Save the archive from an optimization result with numpy.savez_compressed(path, **result["arrays"], channel_positions=...).

reference='input' computes the standard counterfactual difference by comparing x_prime_<branch> directly with saved input x. reference='reconstruction' compares with x_reconstructed_<branch> instead, isolating latent intervention change from reconstruction error.

Parameters:

• npz_path (Path)

• branch (str | None, default: None)

• reference (str, default: 'input')

Return type:

tuple[ndarray, ndarray, str, list[str] | None]

eegproc.model_explainability.model_agnostic.plotting.plot_band_topographies(values, *, channel_names, band_names, colorbar_label, channel_positions=None, title=None, shared_scale=False, signed=None)

Plot (bands, channels) values with band-relative color scales.

Parameters:

• values (ndarray)

• channel_names (list[str])

• band_names (list[str])

• colorbar_label (str)

• channel_positions (ndarray | None, default: None)

• title (str | None, default: None)

• shared_scale (bool, default: False)

• signed (bool | None, default: None)

eegproc.model_explainability.model_agnostic.plotting.resolve_channel_names(n_channels, provided=None, saved=None)

Resolve explicit or saved labels, falling back to numbered channels.

Parameters:

• n_channels (int)

• provided (Optional[Iterable[str]], default: None)

• saved (Optional[Iterable[str]], default: None)

Return type:

list[str]

eegproc.model_explainability.model_agnostic.plotting.split_channel_bands(signal, *, n_channels, n_bands, feature_order)

Return (samples, bands, channels) from flattened channel-band data.

channel-major means the flattened feature axis is ordered as all bands for channel 1, then all bands for channel 2. band-major means all channels for band 1, then all channels for band 2.

Parameters:

• signal (ndarray)

• n_channels (int)

• n_bands (int)

• feature_order (str)

Return type:

ndarray

eegproc.model_explainability.model_agnostic.plotting.summarize_activity(signal, *, measure)

Reduce (samples, channels) to one whole-trial value per channel.

Parameters:

• signal (ndarray)

• measure (str)

Return type:

ndarray