Parallel reinforcement learning on HPC with warning "Received duplicate id = x from worker"

3 次查看（过去 30 天）

Mirjan Heubaum 2021-11-29

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1598379-parallel-reinforcement-learning-on-hpc-with-warning-received-duplicate-id-x-from-worker

评论： Walter Roberson 2021-12-5

When I'm running training of a reinforcement learning agent using a HPC cluster and parallel computing toolbox I get the warning "Received duplicate id = 22 from worker" (or other id) after e.g. 180 training episodes. Then the training seems to be stopped and there is no further error or warning. I am using this command to start the .m-script:

module load matlab/R2021a

matlab -nodisplay < rl_training.m

When I set

trainOpts.UseParallel = false;

often I get the warning "Error reading character from command line". Does anyone know why these messages are occurring and is there perhaps a way to continue the training?

5 个评论
显示 3更早的评论隐藏 3更早的评论

Image Analyst 2021-12-2

If you have a maintenance contract in place, I'd call them on the phone. Of course you can use email like @Raymond Norris said. I never use email or a support page since when I encounter a problem I need an immediate solution so I call them.

Walter Roberson 2021-12-5

I never call them, myself -- I open support cases, where I can describe the problem and include code and results to show clearly what is expected and what is received instead. 85% of the time the response is going to be "You are right, that's not good, the developers have been notified and it might get fixed some day".

请先登录，再进行评论。

请先登录，再回答此问题。