stripe-archive / brushfire

Distributed decision tree ensemble learning in Scala
Other
391 stars 50 forks source link

Brushfire should use Bonsai trees in training #94

Open tixxit opened 8 years ago

tixxit commented 8 years ago

To support faster training (especially in local mode) we could be using compressed Bonsai trees directly in Brushfire. There's really no advantage to Brushfire's native tree type so we should remove it. The idea is that we'd batch our operations for 1 step (eg growing the tree by 1) and make the changes to the Bonsai tree in bulk.

This involves both exposing some new functionality from Bonsai, and replacing the use of trees (and the use of the tree ops type class) in Brushfire.