* Classification:
* Significantly faster growing (two orders of magnitude for sets with >100k instances), faster pruning,
* Added approximate mode (`approximate_induction` parameter). Note: this is an experimental feature - the results may change in future releases.
* Regression:
* Mean-based growing set as default (few times faster then median, non-significant impact on accuracy).
* Survival:
* Faster growing and pruning (few fold improvement).