Continue training after finishing first run

RLlib
1

Let us say I have trained the RL model for 1000 episodes with tune.run(), how do I tell Ray to continue training for another 1000 episodes?

I have tried passing resume=True, which resumes the experiment but then exits instead of training for another 1000 episodes even after giving stop criteria to 2000 episodes.

2

Hey @geekyneuro , maybe there is another stopping criteria that kicks in? Like reward or training iterations or so?

3

There isn’t any other stopping criteria set explicitly in the config file. Only the episodes._total

Sincerely

Armando

4

Hi @geekyneuro,

Have you tried using restore without resume? I am not sure exactly how you are using it but I have found that if I use tune to run trials and it finishes them then it will mark it as “Terminated” which means it finished normally. When you try to resume a run that has “Terminated” it will finish immediately you can not continue trainng. I think what you want to use here is restore.