skglm.solvers.ProxNewton

class skglm.solvers.ProxNewton(p0=10, max_iter=20, max_pn_iter=1000, tol=0.0001, ws_strategy='subdiff', fit_intercept=True, warm_start=False, verbose=0)

Prox Newton solver combined with working sets.

p0int, default 10

Minimum number of features to be included in the working set.

max_iterint, default 20

Maximum number of outer iterations.

max_pn_iterint, default 1000

Maximum number of prox Newton iterations on each subproblem.

tolfloat, default 1e-4

Tolerance for convergence.

ws_strategy(‘subdiff’|’fixpoint’), optional

The score used to build the working set.

fit_interceptbool, default True

If True, fits an unpenalized intercept.

verbosebool, default False

Amount of verbosity. 0/False is silent.

References

[1]

Massias, M. and Vaiter, S. and Gramfort, A. and Salmon, J. “Dual Extrapolation for Sparse Generalized Linear Models”, JMLR, 2020, https://arxiv.org/abs/1907.05830 code: https://github.com/mathurinm/celer

[2]

Johnson, T. B. and Guestrin, C. “Blitz: A principled meta-algorithm for scaling sparse optimization”, ICML, 2015. https://proceedings.mlr.press/v37/johnson15.html code: https://github.com/tbjohns/BlitzL1

__init__(p0=10, max_iter=20, max_pn_iter=1000, tol=0.0001, ws_strategy='subdiff', fit_intercept=True, warm_start=False, verbose=0)

Methods

__init__([p0, max_iter, max_pn_iter, tol, ...])
custom_checks(X, y, datafit, penalty)	Ensure the solver is suited for the datafit + penalty problem.
solve(X, y, datafit, penalty[, w_init, ...])	Solve the optimization problem after validating its compatibility.