Would it be possible for doeval and doremoteeval to pass the step info to policies so we can see the warnings?
Posted by: jkolen @ Dec. 4, 2023, 5:38 p.m.Hi jkolen, i thought we were passing this back through the evaluation services. We will take a look and confirm.
Posted by: adelanovic @ Dec. 4, 2023, 6:58 p.m.From both doeval and doremoteeval:
while True:
action = solution.policies(observation, dones)
observation, all_rewards, dones, info = evaluator.env_step(action)
if all(dones.values()):
print("Episode {} Done".format(episode))
episode += 1
break
The step info is lost.
Posted by: jkolen @ Dec. 4, 2023, 7:05 p.m.Got it, this is a reasonable request and we will push the change out after some testing. It'll also involve having to change the parameter of Solution/MySolution (policies function) so when its pushed out everyone will be notified.
Thank you
Posted by: adelanovic @ Dec. 4, 2023, 8:10 p.m.