Training failures generally occur for two reasons:
- The AnyLogic model itself contains errors and/or functions (e.g. custom libraries) that are not yet supported in Pathmind.
- Pathmind ran out of memory during training. Complex AnyLogic models (i.e. those with large observation and action spaces) require more memory to execute. Try reducing the size of your observations and actions spaces to see if it resolves training failures.
When failure occurs, it is a good idea to first check that the model is working in AnyLogic and the Pathmind Helper elements are set up correctly. Next, review the reward function and check for errors.
If training continues to fail, please contact Pathmind support with the following information:
- Link to the experiment
- Copy of the AnyLogic alp (if possible)
- Brief overview of the model objectives