Main Content

rlTRPOAgentOptions

Options for TRPO agent

Since R2021b

Description

Use an rlTRPOAgentOptions object to specify options for trust region policy optimization (TRPO) agents. To create a TRPO agent, use rlTRPOAgent.

For more information on TRPO agents, see Trust Region Policy Optimization (TRPO) Agents.

For more information on the different types of reinforcement learning agents, see Reinforcement Learning Agents.

Creation

Description

opt = rlTRPOAgentOptions creates an rlPPOAgentOptions object for use as an argument when creating a TRPO agent using all default settings. You can modify the object properties using dot notation.

example

opt = rlTRPOAgentOptions(Name=Value) creates the options set opt and sets its properties using one or more name-value arguments. For example, rlTRPOAgentOptions(DiscountFactor=0.95) creates an option set with a discount factor of 0.95. You can specify multiple name-value arguments.

Properties

expand all

Sample time of agent, specified as a positive scalar or as -1. Setting this parameter to -1 allows for event-based simulations.

Within a Simulink® environment, the RL Agent block in which the agent is specified to execute every SampleTime seconds of simulation time. If SampleTime is -1, the block inherits the sample time from its parent subsystem.

Within a MATLAB® environment, the agent is executed every time the environment advances. In this case, SampleTime is the time interval between consecutive elements in the output experience returned by sim or train. If SampleTime is -1, the time interval between consecutive elements in the returned output experience reflects the timing of the event that triggers the agent execution.

This property is shared between the agent and the agent options object within the agent. Therefore, if you change it in the agent options object, it gets changed in the agent, and vice versa.

Example: SampleTime=-1

Discount factor applied to future rewards during training, specified as a positive scalar less than or equal to 1.

Example: DiscountFactor=0.9

Entropy loss weight, specified as a scalar value between 0 and 1. A higher entropy loss weight value promotes agent exploration by applying a penalty for being too certain about which action to take. Doing so can help the agent move out of local optima.

When gradients are computed during training, an additional gradient component is computed for minimizing the entropy loss. For more information, see Entropy Loss.

Example: EntropyLossWeight=0.02

Number of steps used to calculate the advantage, specified as a positive integer. For more information, see the agent training algorithm.

Example: ExperienceHorizon=1024

Mini-batch size used for each learning epoch, specified as a positive integer. When the agent uses a recurrent neural network, MiniBatchSize is treated as the training trajectory length.

The MiniBatchSize value must be less than or equal to the ExperienceHorizon value.

Example: MiniBatchSize=256

Number of times an agent learns over a data set, specified as a positive integer. This value defines the number of passes over a data set that has a minimum length specified by the LearningFrequency property.

Example: NumEpoch=2

Maximum number of mini-batches used for learning during a single epoch, specified as a positive integer.

For on-policy agents that support this property (PPO and TRPO), the actual number of mini-batches used for learning depends on the length of aggregated trajectories, it has a lower bound of LearningFrequency/MiniBatchSize and an upper bound of MaxMiniBatchPerEpoch.

This value also specifies the maximum number of gradient steps per learning iteration because the maximum number of gradient steps is equal to the MaxMiniBatchPerEpoch value multiplied by the NumEpoch value. For PPO and TRPO agents, it is good practice to set this value to an arbitrarily high number to ensure all data is used for training.

Example: MaxMiniBatchPerEpoch=500

Minimum number of environment interactions between learning iterations, specified as a positive integer or -1. This value defines how many new data samples need to be generated before learning. For on-policy agents that support this property (PPO and TRPO), set LearningFrequency to an integer multiple of MiniBatchSize. The default value of -1 indicates that 10*MiniBatchSize samples are collected before learning. Set this property to a lower value (for example, 2*MiniBatchSize) if you want the agent to learn more frequently.

Example: LearningFrequency=4

Critic optimizer options, specified as an rlOptimizerOptions object. It allows you to specify training parameters of the critic approximator such as learning rate, gradient threshold, as well as the optimizer algorithm and its parameters. For more information, see rlOptimizerOptions and rlOptimizer.

Example: CriticOptimizerOptions = rlOptimizerOptions(LearnRate=5e-3)

Conjugate gradient damping factor for numerical stability, specified as a nonnegative scalar.

Example: ConjugateGradientDamping=0.05

Upper limit for the Kullback-Leibler (KL) divergence between the old policy and the current policy, specified as a positive scalar.

Example: KLDivergenceLimit=0.02

Maximum number of iterations for conjugate gradient descent, specified as positive integer.

Example: NumIterationsConjugateGradient=30

Number of iterations for line search, specified as a positive integer.

Typically, the default value works well for most cases.

Example: NumIterationsLineSearch=5

Conjugate gradient residual tolerance, specified as a positive scalar. Once the residual for the conjugate gradient algorithm is below this tolerance, the algorithm stops.

Typically, the default value works well for most cases.

Example: ConjugateGradientResidualTolerance=2e-8

Method for estimating advantage values, specified as one of the following:

  • "gae" — Generalized advantage estimator

  • "finite-horizon" — Finite horizon estimation

For more information on these methods, see the training algorithm information in Proximal Policy Optimization (PPO) Agents.

Example: AdvantageEstimateMethod="finite-horizon"

Smoothing factor for generalized advantage estimator, specified as a scalar value between 0 and 1, inclusive. This option applies only when the AdvantageEstimateMethod option is "gae"

Example: GAEFactor=0.97

Method for normalizing advantage function values, specified as one of the following:

  • "none" — Do not normalize advantage values

  • "current" — Normalize the advantage function using the mean and standard deviation for the current mini-batch of experiences.

  • "moving" — Normalize the advantage function using the mean and standard deviation for a moving window of recent experiences. To specify the window size, set the AdvantageNormalizingWindow option.

In some environments, you can improve agent performance by normalizing the advantage function during training. The agent normalizes the advantage function by subtracting the mean advantage value and scaling by the standard deviation.

Example: NormalizedAdvantageMethod="moving"

Window size for normalizing advantage function values, specified as a positive integer. Use this option when the NormalizedAdvantageMethod option is "moving".

Example: AdvantageNormalizingWindow=1e5

Options to save additional agent data, specified as a structure containing a field named Optimizer.

You can save an agent object in one of the following ways:

  • Using the save command

  • Specifying saveAgentCriteria and saveAgentValue in an rlTrainingOptions object

  • Specifying an appropriate logging function within a FileLogger object

When you save an agent using any method, the fields in the InfoToSave structure determine whether the corresponding data is saved with the agent. For example, if you set the Optimizer field to true, then the actor and critic optimizers are saved along with the agent.

You can modify the InfoToSave property only after the agent options object is created.

Example: options.InfoToSave.Optimizer=true

Option to save the actor and critic optimizers, specified as a logical value. If you set the Optimizer field to false, then the actor and critic optimizers (which are hidden properties of the agent and can contain internal states) are not saved along with the agent, therefore saving disk space and memory. However, when the optimizers contain internal states, the state of the saved agent is not identical to the state of the original agent.

Example: true

Object Functions

rlTRPOAgentTrust region policy optimization (TRPO) reinforcement learning agent

Examples

collapse all

Create a TRPO agent options object, specifying the discount factor.

opt = rlTRPOAgentOptions(DiscountFactor=0.9)
opt = 
  rlTRPOAgentOptions with properties:

                            SampleTime: 1
                        DiscountFactor: 0.9000
                     EntropyLossWeight: 0.0100
                     ExperienceHorizon: 512
                         MiniBatchSize: 128
                              NumEpoch: 1
                  MaxMiniBatchPerEpoch: 100
                     LearningFrequency: -1
                CriticOptimizerOptions: [1x1 rl.option.rlOptimizerOptions]
              ConjugateGradientDamping: 0.1000
                     KLDivergenceLimit: 0.0100
        NumIterationsConjugateGradient: 10
               NumIterationsLineSearch: 10
    ConjugateGradientResidualTolerance: 1.0000e-08
               AdvantageEstimateMethod: "gae"
                             GAEFactor: 0.9500
             NormalizedAdvantageMethod: "none"
            AdvantageNormalizingWindow: 1000000
                            InfoToSave: [1x1 struct]

You can modify options using dot notation. For example, set the agent sample time to 0.1.

opt.SampleTime = 0.1;

Version History

Introduced in R2021b

expand all