📖 Selecting tree number for random forests 🗒️ [https://t.co/krpMlT8FQO](https://t.co/BWkpQzFo2c... 📝 Madeline Rosenberg,Lucas Faudman,Charlie Logan,Tirdad Barghi https://t.co/8BFXcQ0E6d
10 followers
5,562 followers
this research is why I often default to 128 random forest trees. Before tuning. Often not necessary. Often, you dont need a ton of trees for decent performance. Leatherman algo. This figure is from somewhere else, but this is often what you get https:/
16,729 followers
@yudivian @SuilanEstevez @estevanell_luis this is another path worth checking out for meta-learning, not using meta-features but other ideas like portfolio-based strategies.
302 followers
@AlejandroPiad @KEggensperger @LindauerMarius @FrankRHutter That's a good question - unless you somehow learn them it might be hard to improve them further. Also, there is other evidence that strategies not using meta-features work really well in HPO: http