DiscretePolicy<T>

Models & Types · AiDotNet.ReinforcementLearning.Policies

Policy for discrete action spaces using a neural network to output action logits.

Inheritance: ModelBase<T, Vector<T>, Vector<T>> → PolicyBase<T> → DiscretePolicy<T>

Implements: IGeneratedParameterRegistrar<T>

Example

// Create a discrete policy that outputs action probabilities via softmax
var network = new NeuralNetwork<double>();
var policy = new DiscretePolicy<double>(network, actionSize: 4, new EpsilonGreedyExploration<double>());
            
// Select a discrete action from the policy
var state = new Vector<double>(new double[] { 0.5, -0.3, 1.0, 0.2 });
var action = policy.SelectAction(state, training: true);

Constructors

Constructor Summary
DiscretePolicy() Initializes a new instance with default settings.
DiscretePolicy(DiscretePolicyOptions<T> options) Builds the policy from options: a dense ReLU network of the options’ hidden widths from StateSize inputs to one logit per action (ActionSize outputs).
DiscretePolicy(INeuralNetwork<T> policyNetwork, DiscretePolicyOptions<T> options) Initializes a new instance of the DiscretePolicy class from its options: action count, exploration strategy and Seed.
DiscretePolicy(INeuralNetwork<T> policyNetwork, int actionSize, IExplorationStrategy<T> explorationStrategy, Random random)

Methods

Method Summary
IReadOnlyList<INeuralNetwork<T>> GetNetworks()
T ComputeLogProb(Vector<T> state, Vector<T> action)
Vector<T> SelectAction(Vector<T> state, bool training)
void Reset()