Hands-On Neuroevolution with Python

Modified Novelty Search

We presented the NS method in Chapter 6, Novelty Search Optimization Method. In the current experiment, we use a slightly modified version of the NS method, which we discuss next.

The modifications to the NS method that we will present in this experiment relate to a new way of maintaining the archive of novelty points. The novelty point holds the maze solver's location in the maze at the end of the trajectory, which is combined with the novelty score.

In the more traditional version of the NS method, the size of the novelty archive is dynamic, allowing the addition of a specific novel point if its novelty score exceeds a certain threshold (the novelty threshold). Also, the novelty threshold can be adjusted during runtime, ...