It seems that YAML configuration file format has been changed in ML-Agents 1.0

[Joovvhan]

Hello, I had encountered a minor issue that has been solved following your links to ml-agents repository.

mlagents.trainers.exception.TrainerConfigError: The option num_update was specified in your YAML file for TrainerSettings, but is invalid.

It seems that configuration.yaml file format has been changed in the lastest ml-agents, and your ur3_config.yml file is already properly modified in the hh/develop/robo-arm-continuous branch.

To make the mlagents-learn running, I organized your parameters and commented the following.

# num_update: 1
# train_interval: 1
# use_recurrent: false

2020-06-20 14:39:08 INFO [stats.py:100] TouchCube: Step: 5000. Time Elapsed: 271.120 s Mean Reward: -11.866. Std of Reward: 5.555. Training.
2020-06-20 14:43:57 INFO [stats.py:100] TouchCube: Step: 10000. Time Elapsed: 560.016 s Mean Reward: -9.497. Std of Reward: 5.626. Training.
2020-06-20 14:49:40 INFO [stats.py:100] TouchCube: Step: 15000. Time Elapsed: 903.239 s Mean Reward: -8.487. Std of Reward: 4.681. Training.
2020-06-20 14:54:46 INFO [stats.py:100] TouchCube: Step: 20000. Time Elapsed: 1209.069 s Mean Reward: -3.008. Std of Reward: 4.634. Training.
2020-06-20 14:59:54 INFO [stats.py:100] TouchCube: Step: 25000. Time Elapsed: 1516.457 s Mean Reward: -1.882. Std of Reward: 3.949. Training.
2020-06-20 15:04:58 INFO [stats.py:100] TouchCube: Step: 30000. Time Elapsed: 1820.897 s Mean Reward: -0.098. Std of Reward: 2.458. Training.

It seems ml-agents is working properly. Yet I am not sure is it OK to just comment those parameters.

I would be appreciated if you could check the attached config file and let me know whether it is correctly modified as you originally intended.

Thanks.

ur3_config.txt

[VKSharma17]

Thanks for clearing this. The UR3 ML Agents demo works fine with the updated config.yaml file as per ur3_config.txt