I have added all training parameters to the config for multi-policy algorithms when logging is turned on. This ensures full transparency of the used parameters when logging results.
As a side note, I think this might also be a useful addition to the single-policy algorithms. However, these algorithms do not yet have a register_additional_config() method in the base class. I was unsure whether you would want this. If yes, let me know and I'll quickly extend this pull request to also log all single-policy parameters.
That is exactly the reason why I did not yet do it. However, I recommend for the future to find a way to also log those parameters as otherwise people will be left wondering why they got certain results.
I have added all training parameters to the config for multi-policy algorithms when logging is turned on. This ensures full transparency of the used parameters when logging results.
As a side note, I think this might also be a useful addition to the single-policy algorithms. However, these algorithms do not yet have a
register_additional_config()
method in the base class. I was unsure whether you would want this. If yes, let me know and I'll quickly extend this pull request to also log all single-policy parameters.