Evaluate reward terms on historic state-action data

[andrschl]

For a manager-based environment, I’d like to evaluate the reward terms on a pre-collected dataset of observations and actions. Is there a good way to do this without explicitly storing all reward terms, or is the only way to just store all reward terms in the pre-collected dataset? Thanks in advance for any advice.

[RandomOakForest]

Thank you for posting this. Storing all reward terms explicitly is the only way to do so, currently.

[andrschl]

Ok, perfect. Thanks for the quick response!