MADDPGOptions<T>

Options & Configuration · AiDotNet.Models.Options

Configuration options for Multi-Agent DDPG (MADDPG) agents.

Inheritance: ModelOptions → ReinforcementLearningOptions<T> → MADDPGOptions<T>

For Beginners

MADDPG allows multiple agents to learn together in shared environments. During training, agents can “see” what others are doing (centralized critics), but during execution, each agent acts independently (decentralized actors).

Key features:

  • Centralized Training: Critics see all agents’ observations and actions
  • Decentralized Execution: Actors only use their own observations
  • Continuous Actions: Based on DDPG for continuous control
  • Cooperative or Competitive: Works for both settings

Think of it like: Team sports where players practice together (centralized) but during the game each player makes their own decisions (decentralized).

Examples: Robot coordination, traffic control, multi-player games

How It Works

MADDPG extends DDPG to multi-agent settings with centralized training and decentralized execution. Critics observe all agents during training.

Constructors

Constructor Summary
MADDPGOptions() Initializes a new instance with the paper’s training settings.
MADDPGOptions(MADDPGOptions<T> other) Initializes a new instance by copying values from another instance.

Properties

Property Summary
Func<IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>> OptimizerFactory { get; set; } Gets or sets a factory for the optimizer each of this agent’s actor and critic networks trains with.
List<int> ActorHiddenLayers { get; set; } Hidden layer sizes for each agent’s actor network.
List<int> CriticHiddenLayers { get; set; } Hidden layer sizes for each agent’s critic network.
T ActorLearningRate { get; set; } Learning rate for actor network updates.
T CriticLearningRate { get; set; } Learning rate for critic network updates.
T TargetUpdateTau { get; set; } Soft update coefficient for target networks.
double ExplorationNoise { get; set; } Standard deviation of Gaussian noise added for exploration.
int ActionSize { get; set; } The per-agent action vector size.
int NumAgents { get; set; } The number of agents trained jointly.
int StateSize { get; set; } The per-agent observation vector size.

Methods

Method Summary
void Validate() Validates that required properties are set.