DDPG Agent OU noise options to favour exploration

18 views (last 30 days)
Abd Al-Rahman Al-Remal
Abd Al-Rahman Al-Remal on 22 Jul 2021
Hi there,
I have seem similar posts here however I haven't found one that explains how to actually tune the OU noise parameters to favour exploration - currently my agent is stuck on the same reward value from the beginning and does not change/train/learn.
Can anyone advise on how to tune the OU noise parameters within the code to favour exploration? Currently mine are:
agentOpts.NoiseOptions.StandardDeviation = 0.3;
agentOpts.NoiseOptions.StandardDeviationDecayRate = 1e-5;
agentOpts.NoiseOptions.MeanAttractionConstant = 2e-3;
This worked for a previous similar model I made however I understand that the parameter smust be modified per model however I don't know how and literature all looks very dense and doesn't give a clear answer.
Thanks in advance!
Abd

Answers (0)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!