skglm.experimental.IterativeReweightedL1#

class skglm.experimental.IterativeReweightedL1(datafit=<skglm.datafits.single_task.Quadratic object>, penalty=<skglm.penalties.separable.L0_5 object>, solver=None, n_reweights=5)[source]#

Reweighted L1-norm estimator.

This estimator solves a non-convex problems by iteratively solving convex surrogates involving weighted L1 norms.

Parameters:

datafitinstance of BaseDatafit, optional: Datafit. If None, datafit is initialized as a Quadratic datafit. datafit is replaced by a JIT-compiled instance when calling fit.
penaltyinstance of BasePenalty, optional: Penalty. If None, penalty is initialized as a L0_5 penalty. penalty is replaced by a JIT-compiled instance when calling fit.
solverinstance of BaseSolver, optional: Solver. If None, solver is initialized as an AndersonCD solver.
n_reweightsint, optional: Number of reweighting performed (convex surrogates solved).

References

[1]

Candès et al. (2007), Enhancing sparsity by reweighted l1 minimization https://web.stanford.edu/~boyd/papers/pdf/rwl1.pdf

Attributes:

coef_array, shape (n_features,): Parameter vector (`w` in the cost function formula).
loss_history_list: Objective history after every reweighting.

__init__(datafit=<skglm.datafits.single_task.Quadratic object>, penalty=<skglm.penalties.separable.L0_5 object>, solver=None, n_reweights=5)[source]#

Methods

`__init__`([datafit, penalty, solver, n_reweights])
`fit`(X, y)	Fit the model according to the given training data.
`get_metadata_routing`()	Get metadata routing of this object.
`get_params`([deep])	Get parameters of the estimators including the datafit's and penalty's.
`predict`(X)	Predict target values for samples in X.
`set_params`(**params)	Set the parameters of this estimator.