Closed biona001 closed 2 years ago
I guess if you check each model individually (corresponding to each of the CMSA splittings), you will get something a bit larger than 10K non-zero variables. But the final model averages all these models, so that you can have much more than 10K if the variables used are not the same in all K models.
I see. You are taking each of the models with slightly more than 10k variables, and literally averaging their beta values. Thanks!
I think this is more a question than an issue.
I did a sparse linear regression with
dfmax=10000
which is throwing theToo many variables
warning, but extracting the optimal beta gives me 23309 non-zero entries? Then I inspect (presumably?) the sparsity level for each lambda, and it never reaches much more than 10000.I'm not really understanding where did 23309 come from? Also, it does seem a bit unexpected to me that specifying
dfmax=10000
still gave me a model that had a lot more nonzero entries in it.