2
$\begingroup$

Can I see the intervals and p values for each coefficient in elastic net model? I wrote:

elastNet<-cv.glmnet(pred, comp,alpha=.8,standardize = TRUE, family="binomial",type.measure="auc")

If I write the following I get the numbers, but no intervals.

coef(elastNet)

Also I can add the parameter s (I guess is the lambda value)

coef(elastNet,s=0.01) 

To choose a lambda value first I saw the graph:

enter image description here

And also considered:

> elastNet$lambda.min
[1] 0.0007603077
> elastNet$lambda.1se
[1] 0.01492515

So I decided to choose sth in the middle:

coef(elastNet,s=0.01) 

And I got 26 variables with values (I had 59).

I would like to know the confidence interval for each variable. Can I get some idea of this form the following graph?

enter image description here

Also, I f I need the KS measure, does the software provides it?

$\endgroup$
1
  • $\begingroup$ why do you want the confidence intervals? The parameter estimates are all biased and therefore any inference will be misleading. $\endgroup$ – Jacob H Jun 8 '18 at 22:27
4
$\begingroup$

I don't think this is currently possible; see these slides [PDF] of Rob Tibshirani. Slide 13 mentions they can't yet provide a p-value for terms in a lasso model. Those slides describe a test that gives p-values of terms as they enter the Lasso model.

One further assumes therefore that confidence intervals are not available either.

$\endgroup$
1
0
$\begingroup$

There is the selectiveInference package in R, https://cran.r-project.org/web/packages/selectiveInference/index.html, that provides confidence intervals and p values for your coefficients fitted by the LASSO, based on the following paper:

Stephen Reid, Jerome Friedman, and Rob Tibshirani (2014). A study of error variance estimation in lasso regression. arXiv:1311.5274

I think the approach would require some modifications for the elastic net though...

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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