I'm involved in a project where we are applying lasso to build a logistic regression model from a "black box" set of predictors. I need to get JMP Pro before trying out JMP's implementation. I'm doing all the analysis in R at present. Does JMP Pro also include group lasso? I've found through using JMP Cluster Variables (CLUSVAR) that I can naturally group my predictors prior to group lasso. For those who may have used group lasso, is this a reasonable approach? In the literature seemingly all group lasso descriptions assume some a priori knowledge of the correlation structure of the predictors, often defining a cluster as the base covariate and a set of nonlinear transformations on it (x, x^2, x^3). In my case all the clusters are essentially latent, yet the correlation structure is revealed with Cluster Variables.
