Transformers

class kalelinear.transformer.MPCA(var_ratio=0.97, max_iter=1, vectorize=False, n_components=None)[source]

Bases: BaseEstimator, TransformerMixin

Multilinear Principal Component Analysis (MPCA) estimator.

Parameters:

var_ratio (float, default=0.97) – Target cumulative explained variance ratio per mode.
max_iter (int, default=1) – Maximum number of alternating optimization iterations.
vectorize (bool, default=False) – If True, output projected tensors as vectors.
n_components (int, optional) – Number of output features when vectorize=True.

proj_mats

Transposed projection matrices with shapes (P_i, I_i).

Type:: list of ndarray

idx_order

Feature ranking indices by descending projected variance.

Type:: ndarray

mean_

Per-feature empirical mean of the training data.

Type:: ndarray

shape_in

Input per-sample tensor shape.

Type:: tuple

shape_out

Output per-sample tensor shape after projection.

Type:: tuple

References

Haiping Lu, K.N. Plataniotis, and A.N. Venetsanopoulos, “MPCA: Multilinear Principal Component Analysis of Tensor Objects”, IEEE Transactions on Neural Networks, Vol. 19, No. 1, Page: 18-39, January 2008. For initial Matlab implementation, please go to https://uk.mathworks.com/matlabcentral/fileexchange/26168.

Examples

>>> import numpy as np
>>> from kalelinear.transformer import MPCA
>>> x = np.random.random((40, 20, 25, 20))
>>> x.shape
(40, 20, 25, 20)
>>> mpca = MPCA()
>>> x_projected = mpca.fit_transform(x)
>>> x_projected.shape
(40, 18, 23, 18)
>>> x_projected = mpca.transform(x)
>>> x_projected.shape
(40, 7452)
>>> x_projected = mpca.transform(x)
>>> x_projected.shape
(40, 50)
>>> x_rec = mpca.inverse_transform(x_projected)
>>> x_rec.shape
(40, 20, 25, 20)

fit(X, y=None)[source]

Fit MPCA to tensor data.

Parameters:

X (ndarray of shape (n_samples, I_1, ..., I_N)) – Input tensor samples.
y (None, default=None) – Ignored. Present for scikit-learn API compatibility.

Returns:

self – Fitted estimator.

Return type:

MPCA

transform(X)[source]

Project data to the MPCA subspace.

Parameters:: X (ndarray of shape (n_samples, I_1, ..., I_N) or (I_1, ..., I_N)) – Input tensor data.
Returns:: x_projected – Projected data. Shape is (n_samples, P_1, ..., P_N) when vectorize=False. Otherwise returns vectorized features with optional truncation to n_components.
Return type:: ndarray

inverse_transform(X)[source]

Reconstruct original-space tensors from projected data.

Parameters:: X (ndarray) – Projected tensor data, either in tensor or vectorized format.
Returns:: x_rec – Reconstructed tensor data in the original shape.
Return type:: ndarray of shape (n_samples, I_1, …, I_N)

class kalelinear.transformer.TCA(n_components=None, kernel='linear', lambda_=1.0, mu=1.0, gamma_=0.5, k=3, covariate_encoder=None, eigen_solver='auto', tol=0, max_iter=None, iterated_power='auto', remove_zero_eig=False, scale_components=False, random_state=None, copy=True, n_jobs=None, **kwargs)[source]

Bases: BaseMMDDomainTransformer

Transfer Component Analysis.

covariates represent binary domain labels of length n_samples. They must contain both source and target domains during fit(). target_covariate selects which label is treated as the target domain. covariate_encoder is unsupported because TCA expects domain labels rather than general categorical side information.

class kalelinear.transformer.JDA(n_components=None, kernel='linear', lambda_=1.0, covariate_encoder=None, eigen_solver='auto', tol=0, max_iter=None, iterated_power='auto', remove_zero_eig=False, scale_components=False, random_state=None, copy=True, n_jobs=None, **kwargs)[source]

Bases: BaseMMDDomainTransformer

Joint Distribution Adaptation.

covariates represent binary domain labels of length n_samples. They must contain both source and target domains during fit(). target_covariate selects which label is treated as the target domain. covariate_encoder is unsupported because JDA expects domain labels rather than general categorical side information.

class kalelinear.transformer.BDA(n_components=None, kernel='linear', lambda_=1.0, mu=0.5, covariate_encoder=None, eigen_solver='auto', tol=0, max_iter=None, iterated_power='auto', remove_zero_eig=False, scale_components=False, random_state=None, copy=True, n_jobs=None, **kwargs)[source]

Bases: JDA

Balanced Distribution Adaptation.

covariates represent binary domain labels of length n_samples. They must contain both source and target domains during fit(). target_covariate selects which label is treated as the target domain. covariate_encoder is unsupported because BDA expects domain labels rather than general categorical side information.

class kalelinear.transformer.MIDA(n_components=None, mu=1.0, eta=1.0, ignore_y=False, augment=None, kernel='linear', gamma=None, degree=3, coef0=1, kernel_params=None, covariate_encoder=None, alpha=1, fit_inverse_transform=False, eigen_solver='auto', tol=0, max_iter=None, iterated_power='auto', remove_zero_eig=False, scale_components=False, random_state=None, copy=True, n_jobs=None)[source]

Bases: BaseKernelDomainTransformer

Maximum Independence Domain Adaptation (MIDA).

MIDA learns a covariate-invariant feature space by maximizing the Hilbert-Schmidt independence criterion (HSIC) with respect to the provided covariates.

covariates are required during fit(). With covariate_encoder=None, they must already be numeric and shaped as (n_samples,) or (n_samples, n_covariates). Use covariate_encoder="onehot" to fit a one-hot encoder on raw categorical covariates before adaptation. During transform(), covariates are only required when augment is "pre" or "post".