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

How to understand the mlagents' model?(need help)

Discussion in 'ML-Agents' started by likemango, Sep 21, 2022.

  1. likemango

    likemango

    Joined:
    Dec 26, 2018
    Posts:
    24
    Thanks mlagents,I applied it and it gives me a great model ! But I can't figure out what each leayer means, therefore, I can't do inference without barrcuda(I want to copy the parameters from .pt model and write 'Matrix Multiplication,Relu Funcitons' etc by myself to inference it on other applications).
    I Use Netron to visualize the model,it looks like below(the inputs is 128-batchsize,4-dim,the output is just one discreteAction,162-possibilities) .what they mean and how can I understand it ?
    upload_2022-9-21_13-28-47.png
     
  2. likemango

    likemango

    Joined:
    Dec 26, 2018
    Posts:
    24
    I print the model,the 'Policy' part of it.I can see 'network_body.observation_encoder.processors.0.normalizer.running_mean' and
    'network_body.observation_encoder.processors.0.normalizer.running_variance' are the input's mean and variance,before they feed into the model,they should use it to normalize.But what 'network_body.processors.0.normalizer.running_mean' and
    'network_body.processors.0.normalizer.running_variance' mean?
    upload_2022-9-21_19-2-57.png