grpreg
grpreg copied to clipboard
pbreheny
Regularization paths for regression models with grouped covariates
Would be nice to add this, as in **ncvreg**. Ryan Miller has done some work on this.
I believe the biggest merge conflict will be in grpsurv.R, with some minor ones elsewhere. Otherwise, these functions have worked well in the simulation studies and case studies I've explored,...
@pbreheny I have modified the code to enable start, stop survival times as mentioned in issue #43. This brings grpsurv in line with the functionality of glmnet. Work still to...
See #43; implementing this would be best, obviously, but until then a better error should be returned.
Is there any way to modify the code to accept weights? This would really help in making the package compatible with glmnet
It would be nice to allow the user to specify `penalty` as a vector so that different penalties could be used for different groups. For example, if some groups may...
Hi, I believe I have found a possible bug in group coefficients constraint for `penalty="grLasso"` in `grpreg`. [Penalty vignette](https://pbreheny.github.io/grpreg/articles/web/penalties.html#group-selection) states "the coefficients within a group will either all equal zero...
There is `fit$loss` variable if `fit` is a model returned by `grpreg(...)` call. And `fit$loss` is equal to RSS, but it should be equal to 1/2 of RSS, i.e. negative...
[Models vignette](https://pbreheny.github.io/grpreg/articles/web/models.html) defines loss function $L(\beta|X,y)$ in the beginning as **the deviance (−2 times the log-likelihood)**. Later, in **Gaussian (linear regression)** section there is a conflicting definition $L(\beta|X,y) = \frac{1}{2}||y-X\beta||^2$....