Class TabularActorCriticOptions<T>
Configuration options for Tabular Actor-Critic agents.
public class TabularActorCriticOptions<T> : ReinforcementLearningOptions<T>Type Parameters
TThe numeric type used for calculations.
- Inheritance
-
TabularActorCriticOptions<T>
- Inherited Members
Remarks
Tabular Actor-Critic combines policy learning (actor) with value function learning (critic) using lookup tables. The actor learns which actions to take, while the critic evaluates how good those actions are.
For Beginners: Actor-Critic is like having both a player (actor) and a coach (critic). The player tries different strategies, and the coach provides feedback on how well they're working.
Best for:
- Small discrete state/action spaces
- Problems requiring both policy and value learning
- More stable learning than pure policy gradient
- Reducing variance in policy updates
Not suitable for:
- Continuous states (use linear/neural versions)
- Large state spaces (table becomes too big)
- High-dimensional observations
Properties
ActionSize
Size of the action space (number of possible actions).
public int ActionSize { get; init; }Property Value
ActorLearningRate
Learning rate for the actor (policy) updates.
public double ActorLearningRate { get; init; }Property Value
CriticLearningRate
Learning rate for the critic (value function) updates.
public double CriticLearningRate { get; init; }Property Value
StateSize
Size of the state space (number of state features).
public int StateSize { get; init; }