Search Unity

  1. Unity 2020.1 has been released.
    Dismiss Notice
  2. We are looking for feedback on the experimental Unity Safe Mode which is aiming to help you resolve compilation errors faster during project startup.
    Dismiss Notice
  3. Good news ✨ We have more Unite Now videos available for you to watch on-demand! Come check them out and ask our experts any questions!
    Dismiss Notice

Help Wanted Increasing the robostness of the traning

Discussion in 'ML-Agents' started by Hsgngr, Jul 29, 2020.

  1. Hsgngr

    Hsgngr

    Joined:
    Dec 28, 2015
    Posts:
    53
    After long iterations of my agents and environment I noticed that not every training with the same parameters converge. It is reasonable at some point however I don't how can I increase the robustness of my trainings.
    In the below graph you are seeing 3 trainings with exactly same parameters. How can I ensure my findings if results changes with every training ?

    upload_2020-7-29_16-7-15.png
     
  2. kpalko

    kpalko

    Joined:
    May 14, 2020
    Posts:
    4
    If your environment isn't procedurally generated, you should be able to set a training seed to help with your tests.

    mlagents-learn ... --seed SEED ...


    The other thing you can do is run multiple learning experiments and do a standard statistical analysis to compare the runs and see if there are significant differences.
     
    Hsgngr likes this.
unityunity