collasso package#

Submodules#

Module contents#

Sparse linear multi-task regression.

class collasso.CoopLassoCV(*, cv: int = 5, n_alphas: int = 100, l1_ratio: float = 0.5, exp_y: float = 1, exp_x: float = 1, random_state: int | None = None)#

Bases: RegressorMixin, BaseEstimator

Cross-Validated Cooperative Multi-Task Lasso Regression.

Implements cooperative multi-task lasso regression, optimising the regularisation parameters by cross-validation.

Parameters:

cvint, default=5: Number of cross-validation folds.
n_alphasint, default=100: Number of candidate values for the regularisation parameter in the final regressions.
l1_ratiofloat, default=0.5: Elastic net mixing parameter for the initial regressions, with 0<=l1_ratio<=1, where l1_ratio=0 leads to L2 (ridge) and l1_ratio=1 leads to L1 (lasso) penalisation.
exp_yfloat, default=1.0: Non-negative number for exponentiating the target-target correlation coefficients.
exp_xfloat, default=1.0: Non-negative number for exponentiating the feature-feature correlation coefficients.
random_stateint or None, default=None: Random seed for generating reproducible cross-validation folds.

Attributes:

n_int: Number of training samples.
p_int: Number of features.
q_int: Number of targets.
model_list of length q_targets: Fitted models from _CoopLasso.
alpha_list length q_targets of ndarrays: Sequence of regularisation parameters.
mse_list of length q_targets of ndarrays: Cross-validated mean squared errors for each value of alpha.
min_list of length q_targets of ndarrays: Indices of regularisation parameters corresponding to the lowest mean squared error.
coef_ndarray of shape (q_targets, p_features): Estimated effects (of the feature in the column on the target in the row).

See also

IndepLassoCV: A convenience class using the same interface as CoopLassoCV (similarly formatted inputs and outputs) without sharing information among targets or features.
_CoopLasso: An internal class without cross-validation returning the lasso solution path. This is repeatedly called by CoopLassoCV (once in each cross-validation iteration and once for the full dataset).

Examples

>>> from sklearn.datasets import load_linnerud
>>> from collasso import CoopLassoCV
>>> x, y = load_linnerud(return_X_y=True)
>>> model = CoopLassoCV()
>>> model.fit(x, y) # n_samples x p_features, n_samples x q_targets
>>> model.coef_ # q_targets x p_features
>>> y_pred = model.predict(x) # n_samples x q_targets

fit(X: ndarray, y: ndarray, Z: ndarray | None = None) → CoopLassoCV#

Fit cross-validated model.

Fits cross-validated cooperative multi-task lasso regression.

Parameters:

Xndarray of shape (n_samples, p_features) or (n_samples, p_features, q_targets): Common feature matrix for all targets or a specific feature matrix for each target.
yndarray of shape (n_samples, q_targets): Target matrix.
Zndarray of shape (p_features,) or (p_features, q_targets), or None: Logical vector or matrix indicating primary (1, True) and auxiliary features (0, False) for all targets or each target.

Returns:

selfCoopLassoCV: Fitted model.

predict(X: ndarray) → ndarray#

Make predictions.

Make predictions from a cross-validated model obtained by cooperative multi-task lasso regression.

Parameters:

Xndarray of shape (n_samples, p_features) or (n_samples, p_features, q_targets): Common feature matrix for all targets, or a specific feature matrix for each target.

Returns:

y_hatndarray of shape (n_samples, q_targets): Matrix of predicted values (of the target in the column for the sample in the row).

set_fit_request(*, Z: bool | None | str = '$UNCHANGED$') → CoopLassoCV#

Configure whether metadata should be requested to be passed to the fit method.

Note that this method is only relevant when this estimator is used as a sub-estimator within a meta-estimator and metadata routing is enabled with enable_metadata_routing=True (see sklearn.set_config()). Please check the User Guide on how the routing mechanism works.

The options for each parameter are:

True: metadata is requested, and passed to fit if provided. The request is ignored if metadata is not provided.
False: metadata is not requested and the meta-estimator will not pass it to fit.
None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.
str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Parameters:

Zstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED: Metadata routing for Z parameter in fit.

Returns:

selfobject: The updated object.

set_score_request(*, sample_weight: bool | None | str = '$UNCHANGED$') → CoopLassoCV#

Configure whether metadata should be requested to be passed to the score method.

The options for each parameter are:

True: metadata is requested, and passed to score if provided. The request is ignored if metadata is not provided.
False: metadata is not requested and the meta-estimator will not pass it to score.
None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.
str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Parameters:

sample_weightstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED: Metadata routing for sample_weight parameter in score.

Returns:

selfobject: The updated object.

class collasso.IndepLassoCV(*, cv: int = 5, alphas: int = 100)#

Bases: RegressorMixin, BaseEstimator

Single-Task Lasso Regression For Multiple Targets.

Fits single-task lasso regression separately to multiple targets, optimising the regularisation parameters by cross-validation.

This is a convencience class with the interface as CoopLassoCV (but without sharing information among targets or features). Note that auxiliary features are simply excluded from the model (i.e., learning without privileged information).

Parameters:

cvint, default=5: Number of cross-validation folds.
alphasint, default=100: Number of candidate values for the regularisation parameter.

Attributes:

n_int: Number of training samples.
p_int: Number of features.
q_int: Number of targets.
model_list of length q_targets: Fitted models from LassoCV (one for each target).
coef_ndarray of shape (q_targets, p_features): Estimated coefficients (of the feature in the column on the target in the row).

See also

CoopLassoCV: The main class of this package. It uses the same interface as IndepLassoCV (similarly formatted inputs and outputs) but shares information among targets and features to improve selection and prediction.

Examples

>>> from sklearn.datasets import load_linnerud
>>> from collasso import IndepLassoCV
>>> x, y = load_linnerud(return_X_y=True)
>>> model = IndepLassoCV()
>>> model.fit(x, y) # n_samples x p_features, n_samples x q_targets
>>> model.coef_ # q_targets x p_features
>>> y_pred = model.predict(x) # n_samples x q_targets

fit(X: ndarray, y: ndarray, Z: ndarray | None = None) → IndepLassoCV#

Fit IndepLassoCV.

Fit independent lasso regressions to multiple targets.

Parameters:

Xndarray of shape (n_samples, p_features) or (n_samples, p_features, q_targets): Common feature matrix for all targets or a separate feature matrix for each target.
yndarray of shape (n_samples, q_targets): Target matrix.
Zndarray of shape (p_features,) or (p_features, q_targets), or None: Logical vector or matrix indicating primary (1/True) and auxiliary features (0/False) for all targets together or each target separately (NB: auxiliary features are simply excluded).

Returns:

self: IndepLassoCV: Fitted models.

predict(X: ndarray) → ndarray#

Make predictions.

Make prediction with models estimated by independent lasso regressions.

Parameters:

Xndarray of shape (n_samples, p_features) or (n_samples, p_features, q_targets): Common feature matrix for all targets, or a separate feature matrix for each target.

Returns:

y_hatndarray of shape (n_samples, q_targets): Matrix of predicted values.

set_fit_request(*, Z: bool | None | str = '$UNCHANGED$') → IndepLassoCV#

Configure whether metadata should be requested to be passed to the fit method.

The options for each parameter are:

True: metadata is requested, and passed to fit if provided. The request is ignored if metadata is not provided.
False: metadata is not requested and the meta-estimator will not pass it to fit.
None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.
str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Parameters:

Zstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED: Metadata routing for Z parameter in fit.

Returns:

selfobject: The updated object.

set_score_request(*, sample_weight: bool | None | str = '$UNCHANGED$') → IndepLassoCV#

Configure whether metadata should be requested to be passed to the score method.

The options for each parameter are:

True: metadata is requested, and passed to score if provided. The request is ignored if metadata is not provided.
False: metadata is not requested and the meta-estimator will not pass it to score.
None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.
str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Parameters:

sample_weightstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED: Metadata routing for sample_weight parameter in score.

Returns:

selfobject: The updated object.

collasso.simulate(*, n0: int = 100, n1: int = 10000, p: int = 200, q: int = 3, rho: float = 0.9, kappa: float = 1.0, prob_com: float = 0.05, prob_sep: float = 0.05) → tuple[ndarray, ndarray, ndarray, ndarray, ndarray]#

Simulate Data for Linear Multi-Task Regression.

Simulates feature matrix and target matrix, with given probabilities of (i) common effects on all targets and (ii) specific effects on one target.

Parameters:

n0int, default=100: Number of training samples.
n1int, default=10000: Number of testing samples.
pint, default=200: Number of features.
qint, default=3: Number of targets.
rhofloat, default=0.90: Correlation coefficient, 0<=rho<=1.
kappafloat, default=1.00: Correlation coefficient, 0<=kappa<=1.
prob_comfloat, default=0.05: Probability of common effects for all targets, 0<=prob_com<=1.
prob_sepfloat, default=0.05: Probability of separate effects for each target.

Returns:

x_trainndarray of shape (n0_samples,p_features) or (n0_samples,p_features,q_targets): Training feature matrix or matrices, common matrix for all targets (if kappa=1) or separate matrix for each target (if 0<=kappa<1).
y_trainndarray of shape (n0_samples,q_targets): Training target matrix.
x_testndarray of shape (n1_samples,p_features) or (n1_samples,p_features,q_targets): Test feature matrix or matrices, common matrix for all targets (if kappa=1) or separate matrix for each target (if 0<=kappa<1).
y_testndarray of shape (n1_samples,q_targets): Test target matrix.
betandarray of shape (p_features,q_targets): True effects in the training and the test data (of the feature in the row on the target in the column).

Raises:

ValueError