Search Unity

Entropy never changes

Discussion in 'ML-Agents' started by asad133, Apr 6, 2020.

  1. asad133

    asad133

    Joined:
    Apr 1, 2020
    Posts:
    22
    Please I require urgent help. I am working with a simple navigation task.

    Entropy never changes regardless of tuning beta or any other hyperparam. It is extremely frustrating since my agent keeps getting stuck in a local optimum through exploiting its strategy. I want to encourage it to keep exploring.


    Why is the y-axis with only one value? In my previous experiments, entropy was vary varied accross a wide range of values.
     

    Attached Files:

    Last edited: Apr 6, 2020
    asad133, Apr 6, 2020
    #1

  2. TreyK-47

    TreyK-47

    Unity Technologies

    Joined:
    Oct 22, 2019
    Posts:
    1,822
    Hey there @asad133 - which python version & C# version are you using?
     
    TreyK-47, Apr 7, 2020
    #2

  3. asad133

    asad133

    Joined:
    Apr 1, 2020
    Posts:
    22
    Unity version: 2019.2.5f1
    ML-Agents version: 0.15.0.dev0
    TensorFlow: 2.0.1
    Python: 3.6.10
    .Net Version 4.8.03572
     
    asad133, Apr 7, 2020
    #3

  4. andrewcoh_unity

    andrewcoh_unity

    Unity Technologies

    Joined:
    Sep 5, 2019
    Posts:
    162
    andrewcoh_unity, Apr 8, 2020
    #4

  5. asad133

    asad133

    Joined:
    Apr 1, 2020
    Posts:
    22
    Last edited: Apr 11, 2020
    asad133, Apr 11, 2020
    #5