amica.Amica

class amica.Amica(config=None, random_state=None)

Bases: object

Native JAX implementation of AMICA algorithm.

Adaptive Mixture Independent Component Analysis (AMICA) performs ICA with adaptive source density modeling using mixtures of generalized Gaussians.

Parameters:

configAmicaConfig, optional

Configuration object with all algorithm parameters. If None, uses default configuration.

random_stateint, optional

Random seed for reproducibility.

Attributes:

configAmicaConfig

Algorithm configuration.

result_AmicaResult

Fitted model (available after calling fit).

Parameters:

• config (AmicaConfig | None)

• random_state (int | None)

Notes

The AMICA algorithm was developed by Jason Palmer at UCSD. This is a native Python/JAX implementation for GPU acceleration.

References

[1]

Palmer et al. (2008). Newton method for the ICA mixture model. Proc. IEEE ICASSP.

[2]

Palmer et al. (2011). AMICA: An adaptive mixture of independent component analyzers with shared components. UCSD Technical Report.

Examples

>>> from amica import Amica, AmicaConfig
>>> config = AmicaConfig(max_iter=500, num_mix_comps=3)
>>> amica = Amica(config, random_state=42)
>>> result = amica.fit(data)  # data: (n_channels, n_samples)
>>> activations = (
...     result.unmixing_matrix_white_ @ result.whitener_ @ (data - result.mean_[:, None])
... )

__init__(config=None, random_state=None)

Parameters:

• config (AmicaConfig | None)

• random_state (int | None)

Methods

__init__([config, random_state])
fit(data[, init_mean, init_sphere, ...])	Fit AMICA model to data.
fit_transform(X[, y])	Fit to data, then transform it.
get_params([deep])	Get parameters for this estimator.
inverse_transform(sources)	Reconstruct data from sources.
load(outdir)	Load model from AMICA-compatible directory.
save(outdir)	Save model to directory in AMICA-compatible format.
set_params(**params)	Set the parameters of this estimator.
transform(data)	Apply fitted unmixing to new data.

get_params(deep=True)

Get parameters for this estimator.

Parameters:

deepbool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

paramsdict

Parameter names mapped to their values.

Parameters:

deep (bool)

Return type:

dict

set_params(**params)

Set the parameters of this estimator.

Parameters:

**paramsdict

Estimator parameters.

Returns:

selfAmica

Estimator instance.

Return type:

Amica

fit_transform(X, y=None)

Fit to data, then transform it.

Parameters:

Xnp.ndarray, shape (n_channels, n_samples)

Input data.

yNone

Ignored.

Returns:

X_newnp.ndarray, shape (n_components, n_samples)

Transformed data.

Parameters:

X (ndarray)

Return type:

ndarray

fit(data, init_mean=None, init_sphere=None, init_weights=None, init_params=None)

Fit AMICA model to data.

Parameters:

datanp.ndarray, shape (n_channels, n_samples)

Input EEG/MEG data. Should be high-pass filtered.

init_meannp.ndarray, optional

Precomputed mean vector to use instead of computing from data.

init_spherenp.ndarray, optional

Precomputed sphering matrix to use instead of computing from data.

init_weightsnp.ndarray, optional

Precomputed unmixing matrix (W) to use for initialization.

init_paramsdict, optional

Dictionary containing initial values for ‘alpha’, ‘mu’, ‘beta’, ‘rho’.

Returns:

resultAmicaResult

Fitted model containing mixing/unmixing matrices and all model parameters.

Parameters:

• data (ndarray)

• init_mean (ndarray | None)

• init_sphere (ndarray | None)

• init_weights (ndarray | None)

• init_params (dict | None)

Return type:

AmicaResult

transform(data)

Apply fitted unmixing to new data.

Parameters:

datanp.ndarray, shape (n_channels, n_samples)

New data to transform.

Returns:

sourcesnp.ndarray, shape (n_components, n_samples)

Source activations.

Parameters:

data (ndarray)

Return type:

ndarray

inverse_transform(sources)

Reconstruct data from sources.

Parameters:

sourcesnp.ndarray, shape (n_components, n_samples)

Source activations.

Returns:

datanp.ndarray, shape (n_channels, n_samples)

Reconstructed data.

Parameters:

sources (ndarray)

Return type:

ndarray

save(outdir)

Save model to directory in AMICA-compatible format.

Parameters:

outdirstr or Path

Output directory.

Parameters:

outdir (str | Path)

Return type:

None

classmethod load(outdir)

Load model from AMICA-compatible directory.

Parameters:

outdirstr or Path

Input directory containing AMICA binary files.

Returns:

modelAmica

Loaded AMICA model.

Parameters:

outdir (str | Path)

Return type:

Amica