DDPGAgent<T>

Models & Types · AiDotNet.ReinforcementLearning.Agents.DDPG

Deep Deterministic Policy Gradient (DDPG) agent for continuous control.

Inheritance: ReinforcementLearningAgentBase<T> → DeepReinforcementLearningAgentBase<T> → DDPGAgent<T>

Implements: IGeneratedParameterRegistrar<T>, IGradientComputable<T, Vector<T>, Vector<T>>

For Beginners

DDPG is perfect for controlling things that need precise, continuous adjustments like:

  • Robot arm angles (not just “left” or “right”, but exact degrees)
  • Car steering and acceleration (smooth continuous values)
  • Temperature control, volume levels, etc.

Key components:

  • Actor: Learns the best action to take (deterministic policy)
  • Critic: Evaluates how good that action is (Q-value)
  • Target Networks: Stable copies for training
  • Exploration Noise: Adds randomness during training for exploration

Think of it like learning to drive: the actor is your decision-making (how much to turn the wheel), the critic is your judgment (was that a good turn?), and noise is trying slight variations to discover better techniques.

How It Works

DDPG is an actor-critic algorithm designed for continuous action spaces. It learns a deterministic policy (actor) and uses an off-policy approach with experience replay and target networks, extending DQN ideas to continuous control.

Reference: Lillicrap et al., “Continuous control with deep reinforcement learning”, 2015.

Example

// Create a DDPG agent for continuous control tasks
var options = new DDPGOptions<double> { StateSize = 4, ActionSize = 2, ActorLearningRate = 0.001 };
var agent = new DDPGAgent<double>(options);
            
// Select a continuous action for the current state
var state = new Vector<double>(new double[] { 0.5, -0.3, 1.0, 0.2 });
var action = agent.SelectAction(state);

Constructors

Constructor Summary
DDPGAgent() Initializes a new instance with default settings.
DDPGAgent(DDPGOptions<T> options)

Properties

Property Summary
int FeatureCount { get; }

Methods

Method Summary
Dictionary<string, T> GetMetrics()
ModelMetadata<T> GetModelMetadata()
ModelOptions GetOptions()
T Train()
Vector<T> ComputeGradients(Vector<T> input, Vector<T> target, ILossFunction<T> lossFunction)
Vector<T> SelectAction(Vector<T> state, bool training)
byte[] Serialize() Auto-generated common model serialization surface.
void ApplyGradients(Vector<T> gradients, T learningRate)
void Deserialize(byte[] data) Auto-generated common model deserialization surface.
void LoadModel(string filepath)
void SaveModel(string filepath)
void StoreExperience(Vector<T> state, Vector<T> action, T reward, Vector<T> nextState, bool done)
void Train(Vector<T> state, Vector<T> target) Performs a one-shot supervised update for the training/test harness.