Hi -
I'm using the elastic net (via {glmnet}) to train a few different multinomial models to classify my samples based on treatment (condition, time, etc). Previous posts by Kevin were quite helpful in familiarizing with that package.
The current approach applies cv.glmnet with leave-one-out cross validation (nfolds = nrow(x)) and various alpha values to find the optimal lambda at each, with the goal of obtaining a minimal but performant gene signature. I'm also exploring training with or without the relaxed lasso.
What would you all recommend as the best way to compare and plot the performance of a few models, i.e. those retaining on the order of 10, 100, and 1000 genes? I'm aware of metrics like AUC and AIC, and plots such as one-versus-all ROC, but that's about it. Any advice or plotting package suggestions are much appreciated!
Thx!
