What controls Time.timeScale during training?

While training, the game under test is apparently set to a high Time.timeScale so that training is faster. But how do we control this value? Some games break at too high of a time scale.

I’ve seen the mysterious hints here ml-agents/docs/Python-API.md at 8ad72c08d7d9408c4dda4601beec7f8e82f911ed · Unity-Technologies/ml-agents · GitHub but I’ve no idea which file(s) in release 15.1 are actually driving that.

How is an end user supposed to set this value?

Thanks.

That’s for the low-level python API I think. If you’re using the terminal to launch training, try out this command:

mlagents-learn --help

Time scale can be set as a command line argument when you launch training.

like this : mlagents-learn trainer_config.yaml --train --time-scale 1

4 Likes

Thank you!