MixedPolicyOptions<T>

Options & Configuration · AiDotNet.ReinforcementLearning.Policies

Configuration options for mixed discrete and continuous policies: MixedPolicy’s #ctor constructor builds its networks from them.

Inheritance: ModelOptions → MixedPolicyOptions<T>

For Beginners

Use this when each step makes a choice and sets some amounts, such as “which tool” plus “how hard to press”.

How It Works

A mixed (parameterized) action space pairs one categorical choice with real-valued parameters, as in parameterized-action RL (Masson, Ranchod and Konidaris, 2016; Xiong et al., 2018, P-DQN). The policy samples a one-hot discrete action from softmax logits and a Gaussian continuous action from a mean and log standard deviation per dimension; the action vector is [discrete one-hot, continuous values].

Constructors

Constructor Summary
MixedPolicyOptions()

Properties

Property Summary
IExplorationStrategy<T> ContinuousExplorationStrategy { get; set; } Gets or sets the exploration applied to the sampled continuous action during training.
IExplorationStrategy<T> DiscreteExplorationStrategy { get; set; } Gets or sets the exploration applied to the sampled discrete action during training.
ILossFunction<T> LossFunction { get; set; } Gets or sets the loss the networks are trained with when they are fitted directly.
bool SharedFeatures { get; set; } Gets or sets whether both heads read one shared feature trunk instead of two separate networks.
int ContinuousActionSize { get; set; } Gets or sets the number of continuous action dimensions.
int DiscreteActionSize { get; set; } Gets or sets the number of discrete choices (the softmax head’s width).
int StateSize { get; set; } Gets or sets the number of features in an observation (the networks’ input width).
int[] HiddenLayers { get; set; } Gets or sets the width of each hidden layer, in order (of the trunk when SharedFeatures).