CVtreeMLE relies on default sl3 super learner objects to estimate nuisance parameters. Using super learner ensures the CVtreeMLE estimator is asymptotically efficient. However, in situations where users may have domain knowledge or are doing exploratory analysis where computational capacities are low it may be more efficient to use one flexible learner. In this case, it may be best to use earth, or multivariate adaptive splines models as the g and Q models. This would also allow CVtreeMLE to be submitted to CRAN.
CVtreeMLE relies on default sl3 super learner objects to estimate nuisance parameters. Using super learner ensures the CVtreeMLE estimator is asymptotically efficient. However, in situations where users may have domain knowledge or are doing exploratory analysis where computational capacities are low it may be more efficient to use one flexible learner. In this case, it may be best to use earth, or multivariate adaptive splines models as the g and Q models. This would also allow CVtreeMLE to be submitted to CRAN.