Hi! Thanks for open-sourcing the code — I have a question specifically about how the --curr_rew_thres hyperparameters are determined.
In your examples/configs I see settings like: --curr_rew_thres 0.6 0.01 0.01 0.01
How did you arrive at the concrete values (like 0.6 and the 0.01s) in your experiments?
Were they chosen via hyperparameter search, based on reward statistics (mean/std or percentiles), or something else?
I want to use my own data, so I wonder what is a recommended procedure to set curr_rew_thres?
Any guidance would be greatly appreciated. Thank you!
Hi! Thanks for open-sourcing the code — I have a question specifically about how the --curr_rew_thres hyperparameters are determined.
In your examples/configs I see settings like: --curr_rew_thres 0.6 0.01 0.01 0.01
How did you arrive at the concrete values (like 0.6 and the 0.01s) in your experiments?
Were they chosen via hyperparameter search, based on reward statistics (mean/std or percentiles), or something else?
I want to use my own data, so I wonder what is a recommended procedure to set curr_rew_thres?
Any guidance would be greatly appreciated. Thank you!