Better unpenalized var handling

JuliaStats / Lasso.jl

Lasso/Elastic Net linear and generalized linear models

Other

143 stars 31 forks source link

Better unpenalized var handling #26

Closed AsafManela closed 5 years ago

AsafManela commented 5 years ago

Greatly improved unpenalized var handling by conforming to approach taken in R's glmnet and gamlr, where an explanatory variable with a zero penalty factor is included in the base / null model that is used as a starting point for the regularization path. The two following keyword arguments were added to allow us to compare in test/gammalasso.jl against gamlr:
- stopearly allows the user to disable early stopping of regularization path
- standardizeω allows the user to disable standardization of penalty factors These behaviors both happen by default in glmnet but not gamlr

Note: a large number of csv test case files were changed slightly because we standardize X a bit differently than the approach in gamlr. The new test cases compare an unstandardized version.

coveralls commented 5 years ago

Pull Request Test Coverage Report for Build 149

79 of 83 (95.18%) changed or added relevant lines in 4 files are covered.
No unchanged relevant lines lost coverage.
Overall coverage increased (+36.6%) to 87.169%

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/Lasso.jl	66	70	94.29%
<!--	Total:	79	83	95.18%	-->

Totals
Change from base Build 146:	36.6%
Covered Lines:	822
Relevant Lines:	943

💛 - Coveralls

coveralls commented 5 years ago

Pull Request Test Coverage Report for Build 154

62 of 65 (95.38%) changed or added relevant lines in 3 files are covered.
No unchanged relevant lines lost coverage.
Overall coverage increased (+0.5%) to 87.169%

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/Lasso.jl	50	53	94.34%
<!--	Total:	62	65	95.38%	-->

Totals
Change from base Build 152:	0.5%
Covered Lines:	822
Relevant Lines:	943

💛 - Coveralls

andreasnoack commented 5 years ago

It would be great if you could split these changes separate PRs such that they can be reviewed independently. In particular to have the bug fixes and performance enhancements separated.

AsafManela commented 5 years ago

I've separated these as best I could from a new PR #27 that should be considered for merge before this one.

andreasnoack commented 5 years ago

Thanks. Would be great if you could rebase this one.

AsafManela commented 5 years ago

Rebased. Thanks for being so responsive!

andreasnoack commented 5 years ago

@simonster Do you have time to review this?

AsafManela commented 5 years ago

bump