sand relic
Hi!
I am currently trying to train an RL agent using the MA-POCA trainer. I have followed examples such as the football and dodgeball examples to set up my reward and episode structure but I cannot seem to get the elo to increase. I have tried to look around for threads regarding this issue but I can not find much. Furthermore, the losses do not decrease as one might expect?
here are some example logs
craggy pilot
did you try training the example demos though? do they train correctly?
sand relic
I found the issue, there is no mention anywhere but team ID cannot be set during runtime. It has to be set before the agent script is instantiated. It is rather strange that it isn't documented since other variables that should not be set during runtime is either documented or sometimes even locked from updating during runtime.