Custom StopTrainingCriteria in rlTrainingOption

Hi,

In the rlTrainingOption documentation, we have

• "Custom" — Stop training when the custom function specified in StopTrainingValue returns true.

How does it the code look like. How do we introduce function?

I would like my training to stop at average rewards of 600 and having the episodes step of 200.

Thank you in advance.

Answer 1

[Avadhoot]

Hi @Muhammad Fairuz Abdul Jalal,

If you want to specify a custom stopping criteria for training, you can do so by specifying a function handle in the "StopTrainingValue" parameter. Your function must have one input and one output. You can refer to the following piece of code to get a clearer idea.

StopTrainingValue = customFcn(trainingStats)

Here, "trainingStats" is a structure that contains the following fields, all described in the "trainStats" output argument of "train" function.

• EpisodeIndex
• EpisodeReward
• EpisodeSteps
• AverageReward
• TotalAgentSteps
• EpisodeQ0
• SimulationInfo
• EvaluationStatistics
• TrainingOptions

The training stops when the specified function returns true.

You can specify the function as follows:

function y = customFcn(trainingStats)
y = (trainingStats.AverageReward == 600) && (trainingStats.EpisodeSteps == 200)
end

I have specified the conditions that you have stated in the question. You can modify the contents of the function according to your needs.

For more information on custom stopping criteria, look into the following documentation:

https://www.mathworks.com/help/reinforcement-learning/ref/rl.option.rltrainingoptions.html#mw_89358983-cbb9-484e-bed8-a1fac7866a7d

I hope this solves the issue.