Hands-On Neuroevolution with Python

Writing good simulators

When working on a specific problem, it is crucial to write an appropriate simulator that correctly implements the specifics of the simulated process. If you use such a simulator, you will be able to run long episodes of training, which is impossible when using direct inputs from physical devices.

A good simulator should allow you to control the duration of the single time step of the simulated process. During neuroevolution, you need to evaluate each individual in the population against the given simulator. Thus, it makes sense during training to make the duration of a single time step as small as possible to increase the execution speed. On the other hand, when a solution has been found and you need to test it by ...