improbable-ai / harness-offline-rl Goto Github PK

Hi,

Thanks for the interesting work.
I have a question regarding the weights calculation. I'm trying to understand the differences with PER (Online learning) and your work (offline). While I understand that you use return and advantage weighting schemes, I wanted to ask if the weights change dynamically for the transitions in the training phase like PER?
I read the paper, but still this part is a little confusing to me. So, I would greatly appreciate if you could clarify my doubt.

Cheers,
Kailash

Hello! I always have problems when I configure the environment, could you make it more clear? And the final implementation of the code that runs.