mne_denoise.sns.SNS#

class mne_denoise.sns.SNS(n_neighbors: int = 0, skip: int = 0, rcond: float = 1e-12, preserve_mean: bool = False, verbose: bool | str | int | None = None, n_iter: int = 1, outlier_threshold: float | None = None, chunk_size: int | None = None)[source]#

Sensor Noise Suppression estimator.

The estimator learns a channel mean and spatial operator from training data and reuses both during transform.

Parameters:

n_neighborsint, default=0: Number of neighbors per channel; zero uses all available neighbors.
skipint, default=0: Number of most-correlated neighbors to omit.
rcondfloat, default=1e-12: Relative pseudoinverse cutoff.
preserve_meanbool, default=False: Add the fitted channel mean after regeneration.
verbosebool, str, int, or None, default=None: Logging level.
n_iterint, default=1: Number of SNS projections to compose.
outlier_thresholdfloat or None, default=None: Robust channel-wise z-score threshold for fitting.
chunk_sizeint or None, default=None: Samples per covariance/application chunk.

Attributes:

training_mean_ndarray: Weighted channel mean.
denoising_matrix_ndarray: Composite spatial operator.
denoising_matrices_tuple of ndarray: One operator per iteration.
neighbor_ranks_per_iteration_tuple of ndarray: Local covariance ranks by iteration.

See also

compute_sns: One-shot SNS operation for channel-first arrays.
compute_sns_weights: Construct the local sensor reconstruction operator.
mne_denoise.sound.SOUND: Forward-model-based sensor-noise suppression.

Notes

Channel-first NumPy arrays and MNE Raw, Epochs, and Evoked inputs are supported; transform returns the corresponding type without mutating the input. SNS reconstructs each channel from spatially redundant signals in other channels; it is intended for noise specific to individual sensors, not for a source or artifact shared across the array [1].

References

Examples

>>> import numpy as np
>>> from mne_denoise.sns import SNS
>>> rng = np.random.default_rng(0)
>>> data = rng.standard_normal((8, 1000))
>>> model = SNS(n_neighbors=4)
>>> clean = model.fit_transform(data)

fit(X: Any, y=None, sample_weight: ndarray | None = None, *, callback=None, verbose: bool | str | int | None = None) → SNS[source]#

Fit the SNS mean and spatial operator.

Parameters:

Xarray-like or MNE Raw, Epochs, or Evoked: Data used to learn the operator.
yNone, default=None: Ignored for scikit-learn compatibility.
sample_weightndarray or None, default=None: Non-negative fitting weights.
callbackcallable or None, default=None: Synchronous channel-solve callback.
verbosebool, str, int, or None, default=None: Logging level.

Returns:

SNS: The fitted estimator.

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

Fit SNS and apply the fitted operator.

Parameters:

Xarray-like or MNE Raw, Epochs, or Evoked: Data to fit and transform.
yNone, default=None: Ignored for scikit-learn compatibility.
sample_weightndarray or None, default=None: Non-negative fitting weights.
callbackcallable or None, default=None: Synchronous channel-solve callback.
verbosebool, str, int, or None, default=None: Logging level.
**fit_paramsdict: Unexpected fit parameters raise TypeError.

Returns:

same type as X: Sensor-noise-suppressed data.

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

Apply the fitted SNS operator.

Parameters:

Xarray-like or MNE Raw, Epochs, or Evoked: Data with the fitted channel layout.
yNone, default=None: Ignored for scikit-learn compatibility.
verbosebool, str, int, or None, default=None: Logging level.

Returns:

same type as X: A cleaned copy.