otlingam.models.OTICALiNGAM

class OTICALiNGAM(random_state=None, max_iter=1000)[source]

ICA-based LiNGAM using optimal transport ICA.

This estimator learns a directed acyclic graph by estimating an unmixing matrix with OTICA. The resulting matrix is permuted and scaled before a causal order and adjacency matrix are estimated using ICA-LiNGAM’s existing implementation.

Optimization settings:

  • random_state: Seed used by OTICA’s random number generator.

  • max_iter: Maximum number of OTICA optimization iterations.

Variables:

  • _random_state (int | None) – Seed used by OTICA’s random number generator.

  • _max_iter (int) – Maximum number of OTICA optimization iterations.

  • _causal_order (list[np.integer] | None) – Internal causal ordering. None before fitting.

  • _adjacency_matrix (np.ndarray | None) – Internal weighted adjacency matrix. None before fitting.

  • causal_order (list[np.integer]) – Learned causal order from source to sink.

  • adjacency_matrix (np.ndarray) – Learned weighted adjacency matrix.

  • intercept (np.ndarray) – Intercepts of the structural equations.

Examples

>>> from otlingam import OTICALiNGAM
>>> model = OTICALiNGAM(random_state=0, max_iter=1000)
>>> model.fit(X)
>>> model.causal_order_
__init__(random_state=None, max_iter=1000)

Construct a ICA-based LiNGAM model.

Parameters:

  • random_state (int, optional (default=None)) – random_state is the seed used by the random number generator.

  • max_iter (int, optional (default=1000)) – The maximum number of iterations of FastICA.

fit(X, y=None)[source]

Fits the model to the observations.

Parameters:

  • X (np.typing.ArrayLike) – Training observations.

  • y (None, optional) – Ignored. Defaults to None.

Returns:

The fitted estimator.

Return type:

Self