I am using DDPG .If there are four network to algorithm (actor, target actor , critic , target critic) in algorithm, and if possible to use different learning rate to each?

I am using DDPG .If there are four network to algorithm (actor, target actor , critic , target critic) in algorithm, and if possible to use different learning rate to each?

for example online actor=10^-1 and target actor 10^-2...how I can do this in matlab?

Answer 1

[Yash]

Yes, you can use different learning rates for Actor and Critic by specifying them individually when setting up your training options for DDPG agent. Here is a simple code snippet to achieve this:

actorOptimizerOptions = rlOptimizerOptions(LearnRate=1e-1)
criticOptimizerOptions = rlOptimizerOptions(LearnRate=1e-2)
opt = rlDDPGAgentOptions('ActorOptimizerOptions',actorOptimizerOptions,'CriticOptimizerOptions',criticOptimizerOptions)

Refer to this documentation page for more information on creating an object for DDPG agent: https://www.mathworks.com/help/reinforcement-learning/ref/rl.option.rlddpgagentoptions.html