Search Unity

  1. Welcome to the Unity Forums! Please take the time to read our Code of Conduct to familiarize yourself with the forum rules and how to post constructively.
  2. Unite 2023 Registrations are now LIVE! We are thrilled to announce that Unite 2023 Registration is now live and open for all!
    Dismiss Notice
New Forum User Notice Unite 2023 Registrations are now LIVE!

trained file (onnx) vs live pyhton

Discussion in 'ML-Agents' started by mamaorha, Aug 7, 2021.

  1. mamaorha

    mamaorha

    Joined:
    Jun 16, 2015
    Posts:
    44
    hi, it seems like the "Trained" version, running directly with the onnx file is playing a lot "worse" than when i play vs a running agent (pyhton).

    what am i doing wrong?
     
    mamaorha, Aug 7, 2021
    #1

  2. vincentpierre

    vincentpierre

    Unity Technologies

    Joined:
    May 5, 2017
    Posts:
    160
    It is very hard to diagnose what might be wrong from the provided information. It could be that there are some issues with the inference that create this disparity. How are you comparing the two behaviors? Is there a difference in frame-rate or time scale between what happens when running training and inference? If you want to compare properly, you need to set --time-scale=1 and --inference in Python. Also, make sure you are using the proper model in Unity and make sure there are no errors.
     
    vincentpierre, Aug 9, 2021
    #2

  3. mamaorha

    mamaorha

    Joined:
    Jun 16, 2015
    Posts:
    44
    made the comparing by "playing vs the ai".
    i played against the omnx directly and its playing style was more basic than when i played vs the ai running in pyhton with "--time-scale=1 --target-frame-rate=60 --resume".

    will try to add --inference tho
     
    Last edited: Aug 10, 2021
    mamaorha, Aug 10, 2021
    #3