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

Model not improving after 10 million steps

Discussion in 'ML-Agents' started by InfinityGameDev, Jun 8, 2021.

  1. InfinityGameDev

    InfinityGameDev

    Joined:
    Jun 28, 2016
    Posts:
    54
    I've attached the screenshot for my training. I've let it run for over 10 million steps but there is no correlation, which tells me my model isn't learning anything. Do you know why this is, or what I could do to improve it? Each episode is the same length.
     

    Attached Files:

  2. andrewcoh_unity

    andrewcoh_unity

    Unity Technologies

    Joined:
    Sep 5, 2019
    Posts:
    162
    Hi @insane245

    I'd need more information to help debug your issue. I would confirm that your agent is actually able to reach rewarding states and also that the termination conditions work properly e.g. EndEpisode is being called using either keyboard controls or during training.