What’s new¶

Version 0.4 (in progress)¶

Add GroupLasso Estimator (PR: #228)
Add support and tutorial for positive coefficients to Group Lasso Penalty (PR: #221)
Check compatibility with datafit and penalty in solver (PR #137)
Add support to weight samples in the quadratic datafit Weighted Quadratic Datafit (PR: #258)

Add support for weights and positive coefficients to MCPRegression Estimator (PR: #184)
Move solver specific computations from Datafit.initialize() to separate Datafit methods to ease Solver - Datafit compatibility check (PR: #192)
Add LogSumPenalty (PR: ##127)
Remove abstract methods in BaseDatafit and BasePenalty to make solver/penalty/datafit compatibility check easier (PR ##205)
Add fixed-point distance to build working sets in ProxNewton solver (#138)

Add Cox Estimator with support of L1, L2, and Elastic regularization (PR: #171)
Reduce time overhead when fitting Lasso Estimator (PR: #129)
Add Cox datafit for survival analysis (PR #180, #168, #159, #157)
Experimental Pinball datafit (PR: #134)
Add Gamma datafit (PR: #113)
Add Positivity constraint to L1, L1_plus_L2, WeightedL1 (PR: #110)
Add PositiveConstraint (PR: #126)
Add L-BFGS solver for problems with smooth datafits and penalties (PR: #165, #173)
Experimental Primal-dual coordinate descent solve for problems with non-smooth datafits (PR: #131)
Add support of float32 in ProxNewton

Experimental Square root Lasso class with ProxNewton or Chambolle-Pock solver (PR #57)
Accelerated block coordinate descent solver GroupBCD with working sets for problems with group penalties (PR #29, #28, and #26)
Proximal Newton solver ProxNewton with working sets for problems with non-quadratic datafits (PR #51)
Huber datafit (PR #14)
Added SCAD and BlockSCAD penalties (PR #12)
Fitting intercept for Quadratic, Logistic and Huber datafits (PR #55)
Added GramCD, a Gram-based coordinate descent solver for problems with n_features `\ll` n_samples (PR #59)
Transform solvers from functions to classes (PR #63)