rlDDPGAgentOptions

Options for DDPG agent

Description

Use an rlDDPGAgentOptions object to specify options for deep deterministic policy gradient (DDPG) agents. To create a DDPG agent, use rlDDPGAgent.

For more information, see Deep Deterministic Policy Gradient (DDPG) Agent.

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

Creation

Syntax

opt = rlDDPGAgentOptions

opt = rlDDPGAgentOptions(Name=Value)

Description

opt = rlDDPGAgentOptions creates an options object for use as an argument when creating a DDPG agent using all default options. You can modify the object properties using dot notation.

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

Properties

SampleTime — Sample time of agent

1 (default) | positive scalar | -1

Sample time of the agent, specified as a positive scalar or as -1.

Within a MATLAB^® environment, the agent is executed every time the environment advances, so, SampleTime does not affect the timing of the agent execution.

Within a Simulink^® environment, the RL Agent block that uses the agent object executes every SampleTime seconds of simulation time. If SampleTime is -1 the block inherits the sample time from its input signals. Set SampleTime to -1 when the block is a child of an event-driven subsystem.

Note

Set SampleTime to a positive scalar when the block is not a child of an event-driven subsystem. Doing so ensures that the block executes at appropriate intervals when input signal sample times change due to model variations.

Regardless of the type of environment, the time interval between consecutive elements in the output experience returned by sim or train is always SampleTime.

If SampleTime is -1, for Simulink environments, the time interval between consecutive elements in the returned output experience reflects the timing of the events that trigger the RL Agent block execution, while for MATLAB environments, this time interval is considered equal to 1.

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

DiscountFactor — Discount factor

0.99 (default) | positive scalar less than or equal to 1

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

Example: DiscountFactor=0.9

NoiseOptions — Noise model options

OrnsteinUhlenbeckActionNoise object (default) | GaussianActionNoise object

Noise model options, specified as an OrnsteinUhlenbeckActionNoise or GaussianActionNoise object. For more information on the noise model, see Noise Model.

For an agent with multiple actions, if the actions have different ranges and units, it is likely that each action requires different noise model parameters. If the actions have similar ranges and units, you can set the noise parameters for all actions to the same value.

For example, for an agent with two actions, set the standard deviation of each action to a different value while using the same decay rate for both standard deviations.

opt = rlDDPGAgentOptions;
opt.NoiseOptions.StandardDeviation = [0.1 0.2];
opt.NoiseOptions.StandardDeviationDecayRate = 1e-4;

To use Gaussian action noise, first create a default GaussianActionNoise object. Then, specify any nondefault model properties using dot notation.

opt = rlDDPGAgentOptions;
opt.NoiseOptions = rl.option.GaussianActionNoise;
opt.NoiseOptions.StandardDeviation = 0.05;

ExperienceBufferLength — Experience buffer size

10000 (default) | positive integer

Experience buffer size, specified as a positive integer. During training, the agent computes updates using a mini-batch of experiences randomly sampled from the buffer.

Example: ExperienceBufferLength=1e6

MiniBatchSize — Size of random experience mini-batch

64 (default) | positive integer

Size of random experience mini-batch, specified as a positive integer. During each training episode, the agent randomly samples experiences from the experience buffer when computing gradients for updating the critic properties. Large mini-batches reduce the variance when computing gradients but increase the computational effort.

Example: MiniBatchSize=128

SequenceLength — Maximum batch-training trajectory length when using RNN

1 (default) | positive integer

Maximum batch-training trajectory length when using a recurrent neural network, specified as a positive integer. This value must be greater than 1 when using a recurrent neural network and 1 otherwise.

Example: SequenceLength=4

ActorOptimizerOptions — Actor optimizer options

rlOptimizerOptions object

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

Example: ActorOptimizerOptions = rlOptimizerOptions(LearnRate=2e-3)

CriticOptimizerOptions — Critic optimizer options

rlOptimizerOptions object

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)

NumStepsToLookAhead — Number of future rewards used to estimate the value of the policy

1 (default) | positive integer

Number of future rewards used to estimate the value of the policy, specified as a positive integer. Specifically, ifNumStepsToLookAhead is equal to N, the target value of the policy at a given step is calculated adding the rewards for the following N steps and the discounted estimated value of the state that caused the N-th reward. This target is also called N-step return.

Note

When using a recurrent neural network for the critic, NumStepsToLookAhead must be 1.

For more information, see [1], Chapter 7.

Example: NumStepsToLookAhead=3

NumWarmStartSteps — Minimum number of samples to generate before learning starts

positive integer

Minimum number of samples to generate before learning starts. Use this option to ensure that learning takes place over a more diverse data set at the beginning of training. The default, and minimum, value is the value of MiniBatchSize. After the software collects a minimum of NumWarmStartSteps samples, learning occurs at the intervals specified by the LearningFrequency property.

Example: NumWarmStartSteps=20

NumEpoch — Number of times agent learns over data set stored in the experience buffer

1 (default) | positive integer

Number of times an agent learns over the data set stored in the experience buffer, specified as a positive integer. For off-policy agents that support this property (DQN, DDPG, TD3 and SAC), this value defines the number of passes over the data in the replay buffer at each learning iteration.

Example: NumEpoch=2

MaxMiniBatchPerEpoch — Maximum number of mini-batches used for learning during a single epoch

100 (default) | positive integer

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

For off-policy agents that support this property (DQN, DDPG, TD3, and SAC), the actual number of mini-batches used for learning depends on the length of the replay buffer, and MaxMiniBatchPerEpoch specifies the upper bound. 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 off-policy agents that support this property, a high MaxMiniBatchPerEpoch value means that more time is spent on learning than collecting new data. Therefore, you can use this parameter to control the sample efficiency of the learning process.

Example: MaxMiniBatchPerEpoch=200

LearningFrequency — Minimum number of environment interactions between learning iterations

-1 (default) | positive integer

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 off-policy agents that support this property (DQN, DDPG, TD3, and SAC), the default value of -1 means that learning occurs after each episode is finished. Note that learning can start only after the software collects a minimum of NumWarmStartSteps samples. It then occurs at the intervals specified by the LearningFrequency property.

Example: LearningFrequency=4

PolicyUpdateFrequency — Period of policy update with respect to critic update

1 (default) | positive integer

Period of policy update with respect to critic update, specified as a positive integer. This option defines how often the actor is updated with respect to each critic update. For example, a value of 3 means that the actor is updated every three critic updates. Updating the actor less frequently than the critic can improve convergence at the cost of longer training times.

Example: PolicyUpdateFrequency=2

TargetSmoothFactor — Smoothing factor for target actor and critic updates

1e-3 (default) | positive scalar less than or equal to 1

Smoothing factor for target actor and critic updates, specified as a positive scalar less than or equal to 1. For more information, see Target Update Methods.

Example: TargetSmoothFactor=1e-2

TargetUpdateFrequency — Number of steps between target actor and critic updates

1 (default) | positive integer

Number of steps between target actor and critic updates, specified as a positive integer. For more information, see Target Update Methods.

Example: TargetUpdateFrequency=5

BatchDataRegularizerOptions — Batch data regularizer options

[] (default) | rlBehaviorCloningRegularizerOptions object

Batch data regularizer options, specified as an rlBehaviorCloningRegularizerOptions object. These options are typically used to train the agent offline, from existing data. If you leave this option empty, no regularizer is used.

For more information, see rlBehaviorCloningRegularizerOptions.

Example: BatchDataRegularizerOptions = rlBehaviorCloningRegularizerOptions(BehaviorCloningRegularizerWeight=10)

ResetExperienceBufferBeforeTraining — Option for clearing the experience buffer

false (default) | true

Option for clearing the experience buffer before training, specified as a logical value.

Example: ResetExperienceBufferBeforeTraining=true

InfoToSave — Options to save additional agent data

structure (default)

Options to save additional agent data, specified as a structure containing the following fields.

• Optimizer

• PolicyState

• Target

• ExperienceBuffer

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

Optimizer — Option to save actor and critic optimizers

false (default) | 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 contains internal states, the state of the saved agent is not identical to the state of the original agent.

Example: true

PolicyState — Option to save state of explorative policy

false (default) | true

Option to save the state of the explorative policy, specified as a logical value. If you set the PolicyState field to false, then the state of the explorative policy (which is a hidden agent property) is not saved along with the agent. In this case, the state of the saved agent is not identical to the state of the original agent.

Example: true

Target — Option to save actor and critic targets

false (default) | true

Option to save the actor and critic targets, specified as a logical value. If you set the Target field to false, then the actor and critic targets (which are hidden agent properties) are not saved along with the agent. In this case, when the targets contain internal states, the state of the saved agent is not identical to the state of the original agent.

Example: true

ExperienceBuffer — Option to save experience buffer

false (default) | true

Option to save the experience buffer, specified as a logical value. If you set the PolicyState field to false, then the content of the experience buffer (which is accessible as an agent property using dot notation) is not saved along with the agent. In this case, the state of the saved agent is not identical to the state of the original agent.

Example: true

Object Functions

rlDDPGAgent

Deep deterministic policy gradient (DDPG) reinforcement learning agent

Examples

Create DDPG Agent Options Object

Create an rlDDPGAgentOptions object that specifies the mini-batch size.

opt = rlDDPGAgentOptions(MiniBatchSize=48)

opt = 
  rlDDPGAgentOptions with properties:

                             SampleTime: 1
                         DiscountFactor: 0.9900
                           NoiseOptions: [1x1 rl.option.OrnsteinUhlenbeckActionNoise]
                 ExperienceBufferLength: 10000
                          MiniBatchSize: 48
                         SequenceLength: 1
                  ActorOptimizerOptions: [1x1 rl.option.rlOptimizerOptions]
                 CriticOptimizerOptions: [1x1 rl.option.rlOptimizerOptions]
                    NumStepsToLookAhead: 1
                      NumWarmStartSteps: 48
                               NumEpoch: 1
                   MaxMiniBatchPerEpoch: 100
                      LearningFrequency: -1
                  PolicyUpdateFrequency: 1
                     TargetSmoothFactor: 1.0000e-03
                  TargetUpdateFrequency: 1
            BatchDataRegularizerOptions: []
    ResetExperienceBufferBeforeTraining: 0
                             InfoToSave: [1x1 struct]

You can modify options using dot notation. For example, set the agent sample time to 0.5.

opt.SampleTime = 0.5;

Algorithms

Noise Model

Ornstein-Uhlenbeck Action Noise

An OrnsteinUhlenbeckActionNoise object has the following numeric value properties.

Property	Description	Default Value
InitialAction	Initial value of action	0
Mean	Noise mean value	0
MeanAttractionConstant	Constant specifying how quickly the noise model output is attracted to the mean	0.15
StandardDeviationDecayRate	Decay rate of the standard deviation	0
StandardDeviation	Initial value of noise standard deviation	0.3
StandardDeviationMin	Minimum standard deviation	0

At each sample time step k, the noise value v(k) is updated using the following formula, where Ts is the agent sample time, and the initial value v(1) is defined by the InitialAction parameter.

v(k+1) = v(k) + MeanAttractionConstant.*(Mean - v(k)).*Ts
       + StandardDeviation(k).*randn(size(Mean)).*sqrt(Ts)

At each sample time step, the standard deviation decays as shown in the following code.

decayedStandardDeviation = StandardDeviation(k).*(1 - StandardDeviationDecayRate);
StandardDeviation(k+1) = max(decayedStandardDeviation,StandardDeviationMin);

You can calculate how many samples it will take for the standard deviation to be halved using this simple formula.

halflife = log(0.5)/log(1-StandardDeviationDecayRate);

Note that StandardDeviation is conserved between the end of an episode and the start of the next one. Therefore, it keeps on uniformly decreasing over multiple episodes until it reaches StandardDeviationMin.

For continuous action signals, it is important to set the noise standard deviation appropriately to encourage exploration. It is common to set StandardDeviation*sqrt(Ts) to a value between 1% and 10% of your action range.

If your agent converges on local optima too quickly, promote agent exploration by increasing the amount of noise; that is, by increasing the standard deviation. Also, to increase exploration, you can reduce the StandardDeviationDecayRate.

Gaussian Action Noise

A GaussianActionNoise object has the following numeric value properties.

Property	Description	Default Value (ExplorationModel)	Default Value (TargetPolicySmoothModel)
Mean	Noise mean value	0	0
StandardDeviationDecayRate	Decay rate of the standard deviation	0	0
StandardDeviation	Initial value of noise standard deviation	sqrt(0.1)	sqrt(0.2)
StandardDeviationMin	Minimum standard deviation, which must be less than StandardDeviation	0.01	0.01
LowerLimit	Noise sample lower limit	-Inf	-0.5
UpperLimit	Noise sample upper limit	Inf	0.5

At each time step k, the Gaussian noise v is sampled as shown in the following code.

w = Mean + randn(ActionSize).*StandardDeviation(k);
v(k+1) = min(max(w,LowerLimit),UpperLimit);

Where the initial value v(1) is defined by the InitialAction parameter. At each sample time step, the standard deviation decays as shown in the following code.

decayedStandardDeviation = StandardDeviation(k).*(1 - StandardDeviationDecayRate);
StandardDeviation(k+1) = max(decayedStandardDeviation,StandardDeviationMin);

Note that StandardDeviation is conserved between the end of an episode and the start of the next one. Therefore, it keeps on uniformly decreasing over multiple episodes until it reaches StandardDeviationMin.

References

[1] Sutton, Richard S., and Andrew G. Barto. Reinforcement Learning: An Introduction. Second edition. Adaptive Computation and Machine Learning. Cambridge, Mass: The MIT Press, 2018.

Version History

Introduced in R2019a

R2022a: The default value of the ResetExperienceBufferBeforeTraining property has changed

The default value of the ResetExperienceBufferBeforeTraining has changed from true to false.

When creating a new DDPG agent, if you want to clear the experience buffer before training, you must specify ResetExperienceBufferBeforeTraining as true. For example, before training, set the property using dot notation.

agent.AgentOptions.ResetExperienceBufferBeforeTraining = true;

Alternatively, you can set the property to true in an rlDDPGAgentOptions object and use this object to create the DDPG agent.

R2021a: Property names defining noise probability distribution in the OrnsteinUhlenbeckActionNoise object have changed

The properties defining the probability distribution of the Ornstein-Uhlenbeck (OU) noise model have been renamed. DDPG agents use OU noise for exploration.

• The Variance property has been renamed StandardDeviation.

• The VarianceDecayRate property has been renamed StandardDeviationDecayRate.

• The VarianceMin property has been renamed StandardDeviationMin.

The default values of these properties remain the same. When an OrnsteinUhlenbeckActionNoise noise object saved from a previous MATLAB release is loaded, the values of Variance, VarianceDecayRate, and VarianceMin are copied in the StandardDeviation, StandardDeviationDecayRate, and StandardDeviationMin, respectively.

The Variance, VarianceDecayRate, and VarianceMin properties still work, but they are not recommended. To define the probability distribution of the OU noise model, use the new property names instead.

Update Code

This table shows how to update your code to use the new property names for rlDDPGAgentOptions object ddpgopt.

Not Recommended	Recommended
ddpgopt.NoiseOptions.Variance = 0.5;	ddpgopt.NoiseOptions.StandardDeviation = 0.5;
ddpgopt.NoiseOptions.VarianceDecayRate = 0.1;	ddpgopt.NoiseOptions.StandardDeviationDecayRate = 0.1;
ddpgopt.NoiseOptions.VarianceMin = 0;	ddpgopt.NoiseOptions.StandardDeviationMin = 0;

R2020a: Target update method settings for DDPG agents have changed

Target update method settings for DDPG agents have changed. The following changes require updates to your code:

• The TargetUpdateMethod option has been removed. Now, DDPG agents determine the target update method based on the TargetUpdateFrequency and TargetSmoothFactor option values.

• The default value of TargetUpdateFrequency has changed from 4 to 1.

To use one of the following target update methods, set the TargetUpdateFrequency and TargetSmoothFactor properties as indicated.

Update Method	TargetUpdateFrequency	TargetSmoothFactor
Smoothing	1	Less than 1
Periodic	Greater than 1	1
Periodic smoothing (new method in R2020a)	Greater than 1	Less than 1

The default target update configuration, which is a smoothing update with a TargetSmoothFactor value of 0.001, remains the same.

Update Code

This table shows some typical uses of rlDDPGAgentOptions and how to update your code to use the new option configuration.

Not Recommended	Recommended
opt = rlDDPGAgentOptions('TargetUpdateMethod',"smoothing");	opt = rlDDPGAgentOptions;
opt = rlDDPGAgentOptions('TargetUpdateMethod',"periodic");	opt = rlDDPGAgentOptions; opt.TargetUpdateFrequency = 4; opt.TargetSmoothFactor = 1;
opt = rlDDPGAgentOptions; opt.TargetUpdateMethod = "periodic"; opt.TargetUpdateFrequency = 5;	opt = rlDDPGAgentOptions; opt.TargetUpdateFrequency = 5; opt.TargetSmoothFactor = 1;