Feature request: Improve cross platform compatability
Closed this issue · 1 comments
I do not understand why this repo uses its own implementation of random forests when there is a perfectly fine implementation in sklearn. If I understand it correctly, the only reason this library does not work on other platforms is the implementation of the random forests. Furthermore, using a house-made implementation increases the probability of bugs and the complexity of the code.
Hi ErikVogelLUH,
the custom random forest implementation has two reasons: (i) enable instances for algorithm configuration and some other algorithmic details which are non-standard (ii) speed as there was no fast python implementation. The latter now has changed and we actually are working on transitioning to a popular RF python implementation.
Related to #1116 .