I have noticed from some of the questions posted here, who also post their line of code, that they have set the standardize option of glmnet to "FALSE".
It seems that users either:
1) just use the default standardize=TRUE and let glmnet automatically standardize the variables prior to fitting the model sequence,
2) or set to standardize=FALSE and rescale the variables themselves.
I have tried to track how cvglmnet codes implement variable standardization but have not been able to find out how it is done.
How the variables are standardized in the code are especially of interest to me because I am running glmnet on nested cross validation, which means that I would have to constantly rescale my variables each time my training set and test set changes, correct? (Rescale both training data and test data to have values ranging from 0 to 1 based on the min and max of the training data).
To this end, I have been trying to edit parts of the glmnet code...so that the inputs to the glmnet, cvglmnet, lognet, cvlognet functions will be the rescaled values calculated according to the current training set's min and max. I am wondering whether it is right to do so, and am wondering how are other glmnet users standardizing their variables?
