I would like to use the cv.glmnet() function, in a microarray dataset to perform some kind of "feature selection/variable importance" and prioritize with this way, the selected features with the lasso approach (which coefficients are different than zero). My outcome is a categorical binary outcome (factor, 2 levels). My specific questions are the following:
1) Which measure of loss would be more appropriate? type.measure="mse" or type.measure="class", concerning the fact that essentially the suited model here is a binomial logistic regression? (based on my type of outcome)?
2) To use the function coef() at the end, which lambda value for alpha=1 is more valid? That is, lambda.min or lambda.1se?
3) As the function runs "internally" the glmnet() function, also the features here are scaled by default (standardize=TRUE argument in the glmnet() function)?
