Changing agent operator script while training process. Is it bad?

[Pavel_Arkhipov]

I’ve been teaching my agent for a while now and changing its code regularly. I’ve been tweaking how it gets rewards and how it takes actions. I’m doing this to figure out the best way to teach it, and also because I tried making it learn easier things before moving on to harder stuff. Every time I initialized my training process from previous session.

I’m wondering if this is a bad thing. Should I start teaching the agent all over again using the newest version of the code? I’m asking because I’m starting to see that the agent is having trouble learning new ways to do things and it’s getting stuck in its training.

[smallg2023]

you can use the academy lessons to change the environment a bit (generally this is meant for things like making the goal move further away or the number of goals required might increase etc not really changing the agent itself).
ML is not exact though, it is quite difficult to know the best route without experimenting, it can definitely be very time consuming letting them train over and over

[Pavel_Arkhipov]

@smallg2023

you can use the academy lessons to change the environment a bit (generally this is meant for things like making the goal move further away or the number of goals required might increase etc not really changing the agent itself).
ML is not exact though, it is quite difficult to know the best route without experimenting, it can definitely be very time consuming letting them train over and over

Thanks for the advise