Some AnyLogic models may train extremely slow, preventing Pathmind from obtaining a well-trained policy. However, for most users, training should take up to 5 hours at most.
If you are seeing much longer training times, common causes include:
- Certain AnyLogic libraries (most notably, the Pedestrian Library) require exponentially more compute as the number of agents grows. This can bottleneck training as the AnyLogic model itself runs slowly.
- Your AnyLogic simulation contains too many steps. Try reducing the number of times an action is triggered, if possible. Fewer than 1,000 steps per simulation run is ideal.
- AnyLogic models containing databases tend to run much slower. A workaround to speed up the simulation is to load the database as an AnyLogic variable and query from the variable itself.
- Try turning off agent animations if they aren't necessary. Agent animations are typically only used for visualizations but they tend to slow down the simulation.
If you need help speeding up your AnyLogic models, please don't hesitate to contact us.