Search Unity

  1. Megacity Metro Demo now available. Download now.
    Dismiss Notice
  2. Unity support for visionOS is now available. Learn more in our blog post.
    Dismiss Notice

Question what does it mean when the policy loss graph oscillates rapidly?

Discussion in 'ML-Agents' started by SuperRaed, Oct 30, 2022.

  1. SuperRaed

    SuperRaed

    Joined:
    Aug 18, 2015
    Posts:
    3
    Hello guys,
    We have trained an agent to shoot at enemy's while avoiding friends, the behavior seems to be decent, but at moments lacks accuracy; it can take multiple shoots to get the target.
    The overall graphs seem ok, but one thing really stood out for me is how the policy loss graph seems to fluctuate with high rapidity.

    Does anyone know what might be going on?
    (Below are graphs)
     

    Attached Files:

    • CFL.png
      CFL.png
      File size:
      26.4 KB
      Views:
      97
    • CRH.png
      CRH.png
      File size:
      106.1 KB
      Views:
      97
    • PL.png
      PL.png
      File size:
      50.4 KB
      Views:
      98
    • VL.png
      VL.png
      File size:
      37.6 KB
      Views:
      96
    • CR.png
      CR.png
      File size:
      33.4 KB
      Views:
      98