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. Join us on March 30, 2023, between 5 am & 1 pm EST, in the Performance Profiling Dev Blitz Day 2023 - Q&A forum and Discord where you can connect with our teams behind the Memory and CPU Profilers.
    Dismiss Notice
New Forum User Notice Performance Profiling Dev Blitz Day 2023

Question Is it safe to have two normalization ranges for observations?

Discussion in 'ML-Agents' started by Ordpers, Dec 24, 2022.

  1. Ordpers

    Ordpers

    Joined:
    Aug 10, 2019
    Posts:
    5
    Can I have some observations to be within the range (0f, 1f) and others to be within the range (-1f, 1f)? Will it make training worse?
     
    Ordpers, Dec 24, 2022
    #1

  2. GamerLordMat

    GamerLordMat

    Joined:
    Oct 10, 2019
    Posts:
    126
    depends. If you have positions and you scale them differently although length (e.g. one meter) should be the same, ML has to figure out the mapping. I would recommend scaling same things with the same factor. So transform.position/maxCoordinate.
    If you scale two independent things differently it should not matter.

    I personally use normalize=true bc it is the safest way
     
    GamerLordMat, Dec 27, 2022
    #2
    hughperkins likes this.