cv_glmnet for binary outcome

[hadigilan]

Hi, I am installing the regr.cv_glmnet to apply the lasso method for binary outcome (logistic regression). According to the package, cv_glmnet supports the binary outcomes, in addition to the glmnet function. But, when I run the code below this message appears

lrn("regr.cv_glmnet", s = "lambda.min", alpha=1, nfolds=10, family='binomial') Error in self$assert(xs) : Assertion on 'xs' failed: family: Must be element of set {'gaussian','poisson'}, but is 'binomial'.

[mllg]

You seem to be trying to fit a regression learner on a classification task. Try learner classif.cv_glmnet. Not sure why you are not getting a better error message, I would need to see the complete code for this.

[hadigilan]

Thanks for your response,

I trying to implement the lasso in the double machine learning (DML). In DML, we have two regression models. The first model (g) regress the dependent var. (y) on controls (x1,x2,x3). The second one (m) regress the treatment (d) on controls. The dependent is continuous (and hence I set the family argument to 'gaussian'). But the treatment is dichotomous (and hence I set the family to 'binomial')

Here is my complete code:

library(DoubleML) library(mlr3) library(mlr3learners) remotes::install_github("mlr-org/mlr3extralearners") # for other learners (Gradient Boosting) library(mlr3extralearners)

data<- DoubleMLData$new(data = data, y_col = y, x_cols = c('x1', 'x2', 'x3'), d_cols = 'd') ml_g_lasso<- lrn("regr.cv_glmnet", s = "lambda.min", alpha=1, nfolds=10, family='gaussian') ml_m_lasso<- lrn("regr.cv_glmnet", s = "lambda.min", alpha=1, nfolds=10, family='binomial') dml<- DoubleMLPLR$new(data=data, ml_g=ml_g_lasso, ml_m=ml_m_lasso, n_rep=100, n_folds=5) dml$fit()