API Reference — dna/layers/MachineLearnedBehavior


ActivationFunction

Specify the nonlinear activation function applied at each neuron layer in a machine-learned rig behavior network.

Why this exists

Activation functions are a configuration concern embedded in DNA files — they must survive serialization, versioning, and cross-platform loading. Using a scoped enum rather than a raw integer or string makes every assignment compiler-checked and ensures that adding a new function in a future DNA version is a source-visible change. The five enumerators cover all activation functions supported by the RigLogic ML backend.

Fields

Name Type Description
linear ActivationFunction Identity pass-through; output equals input. Suited for output layers predicting unbounded values.
relu ActivationFunction Rectified linear unit; clamps negative inputs to zero. Standard hidden-layer choice for feedforward rig networks.
leakyrelu ActivationFunction Leaky ReLU; passes a small negative slope instead of hard-clamping negatives. Avoids dead-neuron issues that can occur with relu.
tanh ActivationFunction Hyperbolic tangent; squashes output to (−1, 1). Use for layers whose outputs should fall in a bounded symmetric range.
sigmoid ActivationFunction Logistic sigmoid; squashes output to (0, 1). Use for output layers producing probabilities or normalized blend weights.

Construction

// Assign when configuring an ML behavior layer descriptor
dna::ActivationFunction activation = dna::ActivationFunction::relu;

Relationships

  • MachineLearnedBehavior — the ML behavior layer type that consumes this enum to configure its neuron activation