PPOAgent<T>

Models & Types · AiDotNet.ReinforcementLearning.Agents.PPO

Proximal Policy Optimization (PPO) agent for reinforcement learning.

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

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

For Beginners

PPO is one of the most popular RL algorithms today. It’s used by:

  • OpenAI’s ChatGPT (for RLHF training)
  • Many robotics systems
  • Game AI (including Dota 2 bots)

Key idea: Make small, safe policy improvements by clipping updates. Think of it like driving - small steering adjustments work better than jerking the wheel.

PPO learns two things:

  • A policy (actor): What action to take in each state
  • A value function (critic): How good each state is

The critic helps the actor learn more efficiently.

How It Works

PPO is a policy gradient method that uses a clipped surrogate objective to enable multiple epochs of minibatch updates without destructively large policy changes. It achieves state-of-the-art performance across many RL benchmarks while being simpler and more robust than methods like TRPO.

Reference: Schulman, J., et al. (2017). “Proximal Policy Optimization Algorithms.” arXiv:1707.06347.

Example

// Create a PPO agent for stable policy optimization
var options = new PPOOptions<double> { StateSize = 4, ActionSize = 2, ClipEpsilon = 0.2 };
var agent = new PPOAgent<double>(options);
            
// Select an action using the clipped policy
var state = new Vector<double>(new double[] { 0.5, -0.3, 1.0, 0.2 });
var action = agent.SelectAction(state);

Constructors

Constructor Summary
PPOAgent() Initializes a new instance with default settings.
PPOAgent(PPOOptions<T> options) Initializes a new instance of the PPOAgent class.

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 StoreExperience(Vector<T> state, Vector<T> action, T reward, Vector<T> nextState, bool terminated, bool truncated)