[LRNRQ] Add <regrssion_forest> from package <grf>

[chenx2018]

Algorithm

<regrssion_forest>

Package

Supported types

• classif
• clust
• dens
• regr
• surv

I have checked that this is not already implemented in

• [ x] mlr3
• [ x] mlr3learners
• [ x] mlr3extralearners
• [ x] Other core packages (e.g. mlr3proba, mlr3keras)

Why do I think this is a useful learner?

In casual machine learning, this learner is helpful to predict nuisance functions such as propensity score. If we have this learner, we can compare it with other common learners directly.

Further Optional Comments

(e.g. if this algorithm is already implemented under a different learner then please explain why this additional implementation is useful)

[lona-k]

We have some similar learners in the mlr3verse like "regr.ranger" or "regr.randomForest". Could you please elaborate on why this learner is necessary and if there any reason why you would like an implementation of regression_forest specifically? As I understand it, for your use case an implementation of causal_forest would be more useful.