Is it possible to have StopTrainingCriteria in rlTrainingOptions ( RL Toolbox )

Question

Akhilesh Koul 2020-4-30

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/521986-is-it-possible-to-have-stoptrainingcriteria-in-rltrainingoptions-rl-toolbox

回答： Mahesh Taparia 2020-5-13

Hi,

Is there a way to have multiple StopTrainingCriteria in rlTrainingOptions, e.g. only after say 21 "EpisodeCount", if "AverageReward" is greater that set threshold, stop learning, not before that.

Reference: https://www.mathworks.com/help/reinforcement-learning/ref/rltrainingoptions.html

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Mahesh Taparia 2020-5-13

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/521986-is-it-possible-to-have-stoptrainingcriteria-in-rltrainingoptions-rl-toolbox#answer_432238

在 MATLAB Online 中打开

Hi

It can be done by defining the threshold in the training option. For example, set the "AverageReward" as threshold and define rlTrainingOptions as given below:

trainOpts = rlTrainingOptions(...
    'MaxEpisodes',1000,...
    'MaxStepsPerEpisode',1000,...
    'StopTrainingCriteria',"AverageReward",...
    'StopTrainingValue',480,...
    'Verbose',true,...
    'Plots',"training-progress")