Loading
Cookies help us deliver our services. By using our services, you agree to our use of cookies. Learn more

crowdAI is shutting down - please read our blog post for more information

NIPS 2017: Learning to Run

Reinforcement learning environments with musculoskeletal models


Completed
2154
Submissions
631
Participants
87976
Views

training episodes is now of length 1000 timesteps. is server evaluation still 500 episodes?

Posted by QinYongliang almost 2 years ago

besides that, what else is different between training and evaluation environment?

1

Posted by QinYongliang  almost 2 years ago |  Quote

TYPO: - is server evaluation still 500 timesteps?

because agents trained for 1000 timesteps will not perform well in 500 timesteps setting (he will not jump forward during the last few timesteps)

1

Posted by cg  almost 2 years ago |  Quote

I noticed the same, ‘done’ is set in the call to client.env_step after 500 timesteps, when running the script for submission. Could you please check whether the osim-rl on the grading server http://grader.crowdai.org:1729 is up to date?

1

Posted by spMohanty  almost 2 years ago |  Quote

The grader does run for 1000 timesteps. And it should throw an error if there is a mismatch between the client osim-rl version and the server osim-rl version.