Table of Contents

Class PPOAgent<T>

Namespace
AiDotNet.ReinforcementLearning.Agents.PPO
Assembly
AiDotNet.dll

Proximal Policy Optimization (PPO) agent for reinforcement learning.

public class PPOAgent<T> : DeepReinforcementLearningAgentBase<T>, IRLAgent<T>, IFullModel<T, Vector<T>, Vector<T>>, IModel<Vector<T>, Vector<T>, ModelMetadata<T>>, IModelSerializer, ICheckpointableModel, IParameterizable<T, Vector<T>, Vector<T>>, IFeatureAware, IFeatureImportance<T>, ICloneable<IFullModel<T, Vector<T>, Vector<T>>>, IGradientComputable<T, Vector<T>, Vector<T>>, IJitCompilable<T>, IDisposable

Type Parameters

T

The numeric type used for calculations.

Inheritance
PPOAgent<T>
Implements
IFullModel<T, Vector<T>, Vector<T>>
IModel<Vector<T>, Vector<T>, ModelMetadata<T>>
IParameterizable<T, Vector<T>, Vector<T>>
ICloneable<IFullModel<T, Vector<T>, Vector<T>>>
IGradientComputable<T, Vector<T>, Vector<T>>
Inherited Members
Extension Methods

Remarks

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.

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.

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

Constructors

PPOAgent(PPOOptions<T>)

Initializes a new instance of the PPOAgent class.

public PPOAgent(PPOOptions<T> options)

Parameters

options PPOOptions<T>

Configuration options for the PPO agent.

Properties

FeatureCount

Gets the number of input features (state dimensions).

public override int FeatureCount { get; }

Property Value

int

Methods

ApplyGradients(Vector<T>, T)

Applies gradients to update the agent.

public override void ApplyGradients(Vector<T> gradients, T learningRate)

Parameters

gradients Vector<T>
learningRate T

Clone()

Clones the agent.

public override IFullModel<T, Vector<T>, Vector<T>> Clone()

Returns

IFullModel<T, Vector<T>, Vector<T>>

ComputeGradients(Vector<T>, Vector<T>, ILossFunction<T>?)

Computes gradients for the agent.

public override Vector<T> ComputeGradients(Vector<T> input, Vector<T> target, ILossFunction<T>? lossFunction = null)

Parameters

input Vector<T>
target Vector<T>
lossFunction ILossFunction<T>

Returns

Vector<T>

Deserialize(byte[])

Deserializes the agent from bytes.

public override void Deserialize(byte[] data)

Parameters

data byte[]

GetMetrics()

Gets the current training metrics.

public override Dictionary<string, T> GetMetrics()

Returns

Dictionary<string, T>

Dictionary of metric names to values.

GetModelMetadata()

Gets model metadata.

public override ModelMetadata<T> GetModelMetadata()

Returns

ModelMetadata<T>

GetParameters()

Gets the agent's parameters.

public override Vector<T> GetParameters()

Returns

Vector<T>

LoadModel(string)

Loads the agent's state from a file.

public override void LoadModel(string filepath)

Parameters

filepath string

Path to load the agent from.

SaveModel(string)

Saves the agent's state to a file.

public override void SaveModel(string filepath)

Parameters

filepath string

Path to save the agent.

SelectAction(Vector<T>, bool)

Selects an action given the current state observation.

public override Vector<T> SelectAction(Vector<T> state, bool training = true)

Parameters

state Vector<T>

The current state observation as a Vector.

training bool

Whether the agent is in training mode (affects exploration).

Returns

Vector<T>

Action as a Vector (can be discrete or continuous).

Serialize()

Serializes the agent to bytes.

public override byte[] Serialize()

Returns

byte[]

SetParameters(Vector<T>)

Sets the agent's parameters.

public override void SetParameters(Vector<T> parameters)

Parameters

parameters Vector<T>

StoreExperience(Vector<T>, Vector<T>, T, Vector<T>, bool)

Stores an experience tuple for later learning.

public override void StoreExperience(Vector<T> state, Vector<T> action, T reward, Vector<T> nextState, bool done)

Parameters

state Vector<T>

The state before action.

action Vector<T>

The action taken.

reward T

The reward received.

nextState Vector<T>

The state after action.

done bool

Whether the episode terminated.

Train()

Performs one training step, updating the agent's policy/value function.

public override T Train()

Returns

T

The training loss for monitoring.

Edit this page