Hi, I am using glmnet for my data and have questions regarding cv.glmnet: 1. Is the 10 fold CV stratified cross validation for binary classification problem? 2. I am doing binary classification (family = "binomial"), the plot from cv.glmnet gives the average auc as well as the error bar with different lambda. The bottom of the x is lambda and the top of x axis is how many variables are left. Since it is 10 fold CV, presumably you will have slightly different number of nonzeros for each fold run, is the number shown the average number of nonzeros of the 10 folds? If not, does it mean it's not an honest cross validation? Thanks, -Jack [[alternative HTML version deleted]]