Release notes

Version 1.1

Changelog

• Add option for fitting ‘probit’ distribution by Ravi Garg.

• Add option for fitting ‘gamma’ distribution with log link by Mainak Jas.

• Add support for running sklearn.model_selection.GridSearchCV on pyglmnet.GLM objects by Mainak Jas.

• Improved testing of gradients for optimization by Mainak Jas and Pavan Ramkumar.

• Improved testing of estimated coefficients at convergence by Mainak Jas and Pavan Ramkumar

• Add n_iter_ attribute to pyglmnet.GLM by Mainak Jas, Olivier Pieters, Peter Foley and Chris Rogers

• Add option to fit intercept or not with a fit_intercept boolean by Drew Harris, Mainak Jas and Pavan Ramkumar

BUG

• Fixed z cache inconsistency in cdfast when alpha and reg_lambda are nonzero by Peter Foley.

• Fixed incorrect usage of the random_state parameter by Giovanni De Toni.

• Fixed incorrect proximal operator for group lasso by Yu Umegaki.

• Changed stopping criteria for convergence (a threshold on the change in objective value) which stopped too early. The new criteria is a threshold on the change in norm of the beta, by Pavan Ramkumar.

• Fixed group parameter not being passed for Group Lasso by Beibin Li.

• Fix URL paths when running on Windows by Scott Otterson.

• Made temporary file handling OS independent and usable on a cluster by Scott Otterson.

• Replace StratifiedKFold() with KFold(), as StratifiedKFold() doesn’t work for continuous values by Scott Otterson.

• pyglmnet.simulate_glm() accepts beta0 as float and beta of shape (n_features,) only, by Mainak Jas

API

• Refactored GLM class into two classes: pyglmnet.GLM for fitting a single regularization parameter, and pyglmnet.GLMCV for fitting along a regularization path with cross-validation and selection of best regularization parameter by Pavan Ramkumar and Mainak Jas.

• Removed support for fitting ‘multinomial’ distribution by Pavan Ramkumar. This option will be restored in a future release.

Authors

• Beibin Li

• Chris Rodgers

• Drew Harris

• Giovanni De Toni

• Mainak Jas

• Mayank Agarawal

• Olivier Pieters

• Pavan Ramkumar

• Peter Foley

• Ravi Prakash Garg

• Scott Otterson

• Tommy Odland

• Yu Umegaki

Version 1.0

Changelog

• Add GLM class by Pavan Ramkumar

• Add Tikhonov regularization by Pavan Ramkumar

• Add group lasso by Eva Dyer

• Add group lass example by Matt Antalek

• Add multinomial link function by Daniel Acuna

• Add benchmarks by Vinicius Marques

Authors

• Aid Idrizović

• Daniel Acuna

• Eva Dyer

• Hugo Fernandes

• Mainak Jas

• Matt Antalek

• Pavan Ramkumar

• Titipat Achakulvisut

• Vinicius Marques