During policy training, the AnyLogic engine may return an error, causing training to fail. This can occur when extremely rare logical errors exist in your AnyLogic simulation. As a rule of thumb, if there's a bug in your model, the policy will find it and exploit it.
To fix the problem, try to reproduce the error back in AnyLogic.
Step 1: Create a brand new CustomExperiment.
Step 2: Add a
for loop to the prefilled CustomExperiment code to test every seed.
Make sure you also print the seed after each loop (
In most cases, testing seeds 0 - 10,000 is enough to find the error. However, if no errors are uncovered, try testing 100,000 seeds
for(int seed = 0; seed < 100000; seed++).
Step 3: Run the CustomExperiment.
Once an error appears, check your AnyLogic console to see which seed caused the problem.
Step 4: In your Simulation Experiment, set the seed to the value found in Step 3 and run the simulation.
This should allow you to reproduce the error.
If you are still unable to reproduce the error in AnyLogic, please contact Pathmind support.