For feature tuning: Given a model without any pheno features included and a set of pheno features, return a list of models for every subset of this set of features. Possibly without the given model (this may just be a template).
This would turn validation into a statistical fiasco due to huge subsets of models with i.i.d. validated errors. Rather provide all possible features and let regularization tackle the issue.
For feature tuning: Given a model without any pheno features included and a set of pheno features, return a list of models for every subset of this set of features. Possibly without the given model (this may just be a template).