Open be-marc opened 4 years ago
Should this move to mlr3extralearners?
And would it be a single learner for all h2o algorithms, or a separate learner for each algorithm?
Great question. Maybe this should even be a own repository, e.g. mlr3h2o
.
I guess this should be different learners. I am not familiar with h2o but I guess they have different parameters, so this would be necessary?
I don't think we currently have the bandwidth to implement this / focus on other features, so don't expect any progress on this anytime soon. Are you familiar with these learners and interested with giving this a try? At least some of these learners are connected in mlr
(mlr3's predecessor) so one could at least partially re-use this code.
I may write a wrapper or two - esp. for lasso since I believe h2o's makes much better use of multithreading compared to glmnet. I unfortunately don't have the capacity to maintain a separate package so I will probably propose it in mlr3extralearners, taking the recent RWeka PR as inspiration.
CRAN