Hi, I am using the dqn algorithm with ray 2.1 and EpsilonGreedy exploration schedule. I am unable to see the current value in tensorboard logs. Can some one help me out on how I can get the cur_epsilon value in tensorboard.
Tensorboard always shows metrics that are collected over whole episodes. These are aggregations, possibly spanning many episodes. But never the current value. You can start ray with ray.init(local_mode=True) and set a break point inside the exploration schedule if you need to inspect values, or simply print them there.
Thank you for the response. But I was able to view cur_epsilon value in TB using previous ray version. I am certain about that.
Oh ok I thought you where talking about the current epsilon as in the one that appears per step.
Looking at this issue makes me believe that we have not included it for a while. I’m looking into it.