Python low level API: any examples?

[zako42]

Is your feature request related to a problem? Please describe. There are awesome examples of using ML-Agents in Unity on the C# side, using mlagents-learn. I don't see much information or examples of using the Python low level API though. Is it possible to create some examples which parallel the C# examples? I've seen the jupyter notebook and the openAI gym pages and those are great. But having more examples would be even better, of course -- especially a full end-to-end example.

Describe the solution you'd like It would be nice if there were jupyter notebooks showing the same examples as the ones used on the C# side. But any examples would be fine.

Describe alternatives you've considered Right now, just messing around and exploring the python API. But would be super helpful to see a full end-to-end example!

Additional context If there are any tips or suggestions on getting started with the low level API, that would be also appreciated.

[andrewcoh]

Hi @zako42

Just to clarify, you've seen these colab examples? Can you elaborate on what more you'd like beyond what's contained in the colab?

[schmiJo]

@andrewcoh Thank you this helped me a lot

[Ademord]

@andrewcoh we should have a collab for adapting an external algorithm into the UE API if there will be no efforts to push the gym_wrapper to be fully compatible with gym. And special emphasis on not using a multi-branch approach when creating an environment should be done somewhere, since adapting an external algorithm to use MutliDiscrete is too much overhead, and flatten_branched is returning 27 instead of 9 for me (3 discrete branches with 3 possible values).