hyperspy_ml_algorithms.RPCAGoDec

class hyperspy_ml_algorithms.RPCAGoDec(rank=5, lambda1=None, power=0, tol=0.001, max_iter=100, random_state=None)

Bases: object

Robust PCA via the GoDec algorithm (batch).

Decomposes a data matrix into low-rank and sparse components using bilateral random projections. This is a batch method — it processes the entire dataset at once and is best suited for data that fit in memory.

The algorithm is based on [Zhou2011].

Parameters:

rankint, default 5

Target rank of the low-rank approximation.

lambda1float or None, default None

Threshold for soft-thresholding the sparse error. If None, defaults to 1 / sqrt(n_features).

powerint, default 0

Number of power iterations used during randomised initialisation.

tolfloat, default 1e-3

Convergence tolerance on the Frobenius norm of the residual.

max_iterint, default 100

Maximum number of iterations.

random_stateNone, int, or numpy.random.RandomState, default None

Random seed or RandomState for reproducible initialisation.

Attributes:

components_ndarray of shape (rank, n_features)

Right singular vectors, shape (rank, n_features).

low_rank_ndarray of shape (n_samples, n_features)

Low-rank reconstruction, shape (n_samples, n_features).

sparse_ndarray of shape (n_samples, n_features)

Sparse error matrix, shape (n_samples, n_features).

singular_values_ndarray of shape (rank,)

Singular values from the final SVD, shape (rank,).

References

[Zhou2011]

Tianyi Zhou and Dacheng Tao, “GoDec: Randomized Low-rank & Sparse Matrix Decomposition in Noisy Case”, ICML-11, (2011), pp. 33–40.

__init__(rank=5, lambda1=None, power=0, tol=0.001, max_iter=100, random_state=None)

Methods

__init__([rank, lambda1, power, tol, ...])
fit(X[, y])	Fit the GoDec RPCA model to X.
fit_transform(X[, y])	Fit the model and return the low-rank reconstruction.
transform(X)	Project X onto the learned components.

Attributes

components_	Right singular vectors, shape (rank, n_features).
low_rank_	Low-rank reconstruction, shape (n_samples, n_features).
singular_values_	Singular values from the final SVD, shape (rank,).
sparse_	Sparse error matrix, shape (n_samples, n_features).