ObliviousTreeChoiceFunction

Enums · AiDotNet.Enums

The functions an ObliviousDecisionTreeLayer uses to choose a feature per level and to make each soft split decision.

For Beginners

keep the default to get the tree described in the NODE paper. Choose SoftmaxSigmoid only to reproduce models trained with the earlier smooth variant.

How It Works

NODE (Popov et al. 2019) chooses features with 1.5-entmax and splits with 1.5-entmoid. Both are sparse: a level can put all its weight on one feature, and a sample far enough from a threshold goes entirely to one side. Softmax and sigmoid never reach exactly zero or one.

Fields

Field Summary
ObliviousTreeChoiceFunction Entmax15 1.5-entmax feature selection and 1.5-entmoid splits, as in NODE (Popov et al.
ObliviousTreeChoiceFunction SoftmaxSigmoid Softmax feature selection and sigmoid splits, the layer’s earlier behaviour.
int value__