mne_denoise.dss.IterativeDSS#

Iterative DSS transformer.

The estimator fits fixed-point nonlinear DSS filters on NumPy arrays or MNE Raw/Epochs inputs.

Parameters:

denoisercallable: Nonlinear source transformation.
n_componentsint: Number of components to extract.
method{“deflation”, “symmetric”}, default=”deflation”: Component extraction strategy.
rankint or None, default=None: Whitening rank.
regfloat, default=1e-9: Relative whitening threshold.
normalize_inputbool, default=True: Normalize channels during fitting and undo the scaling on reconstruction.
max_iterint, default=100: Maximum fixed-point iterations.
tolfloat, default=1e-6: Convergence tolerance.
verbosebool, str, int, or None, default=None: Logging level.
alpha, beta, gammafloat, callable, or None: Optional fixed-point parameters.
random_stateint, numpy.random.Generator, or None, default=None: Random state for initialization.

Attributes:

filters_ndarray: Fitted sensor-space filters.
patterns_ndarray: Fitted sensor-space patterns.
sources_ndarray: Sources from the fit data.
convergence_info_ndarray, shape (n_components, 2): Iteration count and convergence flag per component.

See also

DSS: Linear covariance-bias DSS.
iterative_dss: Functional iterative DSS interface.

Notes

transform returns source arrays. inverse_transform reconstructs arrays using the fitted patterns; MNE metadata is used for channel extraction during fitting and transformation. This follows the iterative DSS formulation [1].

References

Examples

>>> import numpy as np
>>> from mne_denoise.dss import IterativeDSS
>>> rng = np.random.default_rng(0)
>>> data = rng.standard_normal((8, 2000))
>>> model = IterativeDSS(
...     lambda source: source**3, n_components=2, rank=4, random_state=0
... )
>>> sources = model.fit_transform(data)

fit(X, *, callback=None, verbose: bool | str | int | None = None) → IterativeDSS[source]#

Fit the iterative DSS filters.

Parameters:

Xmne.io.BaseRaw, mne.BaseEpochs, or ndarray: Training data in channel-first NumPy layout or an MNE container.
callbackcallable or None, default=None: Synchronous fixed-point progress callback.
verbosebool, str, int, or None, default=None: Logging level for this call.

Returns:

IterativeDSS: The fitted estimator.

fit_transform(X, *, callback=None, verbose: bool | str | int | None = None) → ndarray[source]#

Fit the estimator and return the extracted sources.

Parameters:

Xndarray: Channel-first data.
callbackcallable or None, default=None: Synchronous progress callback for fitting.
verbosebool, str, int, or None, default=None: Logging level for this call.

Returns:

ndarray: Extracted sources.

get_normalized_patterns() → ndarray[source]#

Return L2-normalized fitted spatial patterns.

Returns:

ndarray, shape (n_channels, n_components): Normalized patterns.

inverse_transform(sources: ndarray, *, verbose: bool | str | int | None = None) → ndarray[source]#

Reconstruct sensor data from fitted sources.

Parameters:

sourcesndarray, shape (n_components, n_times) or (n_epochs, n_components, n_times): Source data.
verbosebool, str, int, or None, default=None: Logging level for this call.

Returns:

ndarray: Reconstructed sensor-space data.

transform(X, *, verbose: bool | str | int | None = None) → ndarray[source]#

Apply fitted filters and return source data.

Parameters:

Xmne.io.BaseRaw, mne.BaseEpochs, or ndarray: Data compatible with the fitted channel layout.
verbosebool, str, int, or None, default=None: Logging level for this call.

Returns:

ndarray: Sources with continuous or epoch-preserving layout.