2
$\begingroup$

Is there a way to fit an adaptive elastic-net model with glmnet ? In other words, I would want my "penalty.factor" parameter applied only to the L1 penalty of the model and to keep the same level of L2 penalty.

I would like to be able to reproduce the algorithm described in this paper : https://projecteuclid.org/euclid.aos/1245332831

I know the R package gcdnet is able to fit such models, but the computational time is far higher than with glmnet.

Thanks

$\endgroup$
2
  • 1
    $\begingroup$ Are you saying you want LASSO? $\endgroup$ Commented Sep 25, 2015 at 15:16
  • $\begingroup$ No I want to be able to reproduce the algorithm described here [link] (projecteuclid.org/euclid.aos/1245332831). $\endgroup$
    – yoyo
    Commented Sep 28, 2015 at 7:12

1 Answer 1

1
$\begingroup$

No, the glmnet package allows you to specify penalization factors for each feature, but this applies to both L1 and L2 norm equally.

However, there is a package called gcdnet which allows fitting of adaptive elastic net models. The intersections of authors of the paper you cite and authors of the package is not empty, so this would probably your best option.

Disclaimer: I have not worked with either the adaptive elastic net or the gcdnet, so don't regard this as a recommendation of either the package or the method.

$\endgroup$
1
  • $\begingroup$ Thanks for your answer. I know the R package gcdnet is able to fit such models, but the computational time is far higher than with glmnet. Since I was planning to perform a large simulation study, the computational time with gcdnet is an important problem. $\endgroup$
    – yoyo
    Commented Oct 7, 2015 at 7:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.