Open cgoo4 opened 1 month ago
Thanks for the issue!
A few notes from my first time wrapping my head around this here.
What makes this more straightforward for engines like lightgbm or xgboost is that the tidymodels team implements a wrapper function around the actual training function (for lightgbm, that's bonsai::train_lightgbm()
) that are straightforward to edit from our perspective. Since aorsf has a straightforward interface, we just pass data and arguments directly to aorsf::orsf()
. This makes things more maintainable from our perspective, but less flexible as well.
Our options would be:
1) Implement a small wrapper around aorsf::orsf()
where we just call bonsai:::process_mtry
.
2) Wait for parsnip to have more full-fledged support for this, e.g. https://github.com/tidymodels/parsnip/issues/602
aorsf is a great addition to bonsai! Any chance of supporting
mtry_prop
?Created on 2024-07-21 with reprex v2.1.1