CODEWITHSUNDEEP
rglmnetlasso-regression
wrong_path
wrong_path

Reputation: 408

Avoid computing an intercept in glmnet::cv.glmnet

I am using glmnet (version 4.1-2) with cv.glmnet to fit a Lasso model. Based on the vignettes, you can use intercept = FALSE in glmnet::glmnet, so I thought I could do the same also with cv.glmnet, but if I use the same argument and I print the list of coefficients resulting from cv.glmnet, (Intercept) is still there and some of the values are different from zero. Any suggestion on how to do CV without fitting the intercept?

Forgot to mention that I am doing multi-task learning. That is, my y is a matrix of responses rather than a vector:

cv.mfit <- glmnet::cv.glmnet(x = as.matrix(dat.preprocessed$exposures), # nxp
                             y = as.matrix(dat.preprocessed$omics), # nxq
                             family = "mgaussian", alpha = 1, 
                             standardize = FALSE, standardize.response = FALSE, 
                             intercept = FALSE)

Upvotes: 1

Views: 1048

Answers (1)

StupidWolf
StupidWolf

Reputation: 46968

I cannot reproduce your issue, I am on glmnet_4.1-1 and if I set intercept=FALSE, I get all 0s for intercepts:

library(glmnet)
x = as.matrix(mtcars[,1:9])
y = as.matrix(mtcars[,10:11])
fit = glmnet(x=x,y=y,alpha=1,intercept=FALSE,family="mgaussian")

table(fit$a0)

 0 
200

cvfit = cv.glmnet(x=x,y=y,alpha=1,intercept=FALSE,family="mgaussian")
table(cvfit$glmnet.fit$a0)

  0 
200


cv.mfit = glmnet::cv.glmnet(x = x,y = y,
                             family = "mgaussian", alpha = 1, 
                             standardize = FALSE,
                             standardize.response = FALSE, 
                             intercept = FALSE)

table(cv.mfit$glmnet.fit$a0)
  0 
200

Upvotes: 0

Related Questions