Search Unity

Question Training stops with "IndexError" after increasing Vector Action's branch size

Discussion in 'ML-Agents' started by diegotriasramos, Aug 9, 2020.

  1. diegotriasramos

    diegotriasramos

    Joined:
    Aug 9, 2020
    Posts:
    2
    Hello,
    I'm using Ml-agents 1.0 Release 3:


    I've been training my agents without problem with these behaviour parameters:

    Configuration: 3 branches for the vection action, name of the behaviour: Agent_3Actions

    With this configuration the training works fine and without problems.

    Normal Output for this case:


    However, when I try to add a new branch to the Vector Action, the learning randomly stops and I get an "IndexError".

    Configuration: 4 branches for the vection action, name of the behaviour: Agent_4Actions

    Output when it stops training:

    I'm not sure, but I think the error appears when the learning process tries to save the model.

    I've made sure to use a different trainer_config.yaml for each case and that the Behaviour Name is the correct one.

    Any ideas? I appreciate any help with this issue.


    Cordially,
    Diego.
     
    Last edited: Aug 9, 2020
  2. anupambhatnagar

    anupambhatnagar

    Unity Technologies

    Joined:
    Jun 22, 2017
    Posts:
    4
    which files are saved in the results folder in the above run? Is the model saved?
     
  3. diegotriasramos

    diegotriasramos

    Joined:
    Aug 9, 2020
    Posts:
    2
    These are the files created, the model is saved: