NeuroTrees

NeuroTabModels.Models.NeuroTrees.act_dict Constant

act_dict

Dictionary mapping feature activation symbols to their functions. Supported keys: :identity, :tanh, :hardtanh, :tanhshrink.

source

NeuroTabModels.Models.NeuroTrees.NeuroTree Type

NeuroTree(feats => outs; tree_type=:binary, actA=identity, scaler=true,
          depth, trees, k, init_scale=0.1)

Differentiable tree ensemble layer. Outpout dims: [outs, k, batch_size].

Arguments

• feats::Int: Number of input features.

• outs::Int: Number of output targets per tree ensemble.

• tree_type::Symbol: :binary or :oblivious.

• actA: Feature activation applied to split weights.

• scaler::Bool: Scale logits with a learned softplus factor.

• depth::Int: Tree depth.

• trees::Int: Number of trees in the ensemble.

• k::Int: Ensemble size.

• init_scale::Float32: Standard deviation for leaf weight initialization.

source

NeuroTabModels.Models.NeuroTrees.NeuroTreeConfig Type

NeuroTreeConfig(; kwargs...)

Configuration for differentiable neuro-tree ensembles.

Arguments

• tree_type::Symbol: :binary or :oblivious (default :binary).

• actA::Symbol: Feature activation. One of :identity, :tanh, :hardtanh, or :tanhshrink (default :identity).

• depth::Int: Tree depth (default 4).

• ntrees::Int: Number of trees per layer (default 32).

• k::Int: Ensemble size.

• hidden_size::Int: Hidden dimension for stacked trees (default 1).

• stack_size::Int: Number of stacked tree layers (default 1).

• scaler::Bool: Apply softplus scaling on tree logits (default true).

• init_scale::Float32: Leaf weight init scale (default 0.1).

• MLE_tree_split::Bool: Split output head for Gaussian MLE (default false).

source

NeuroTabModels.Models.NeuroTrees.NeuroTreeConfig Method

(config::NeuroTreeConfig)(; nfeats, outsize)

Build a Lux.Chain from config.

Arguments

• nfeats::Int: Number of input features.

• outsize::Int: Number of output units.

Returns

A Lux.Chain of stacked neuro-tree layers.

source

NeuroTabModels.Models.NeuroTrees.get_logits_mask Method

get_logits_mask(::Val{:binary}, depth::Integer)

source

NeuroTabModels.Models.NeuroTrees.get_softplus_mask Method

get_softplus_mask(::Val{:binary}, depth::Integer)

source