December 2019
Intermediate to advanced
368 pages
11h 10m
English
As we mentioned, we will use the same experiment runner implementation and the same NEAT hyperparameters settings as in the previous experiment. But we will configure the different maze environment as follows:
$ python maze_experiment.py -m hard -g 500
After a while, when the experiment is over, we see that even after 500 generations of evolution, a successful maze solver has not been found. The best genome obtained using the neuroevolution algorithm encodes a bizarre and non-functional controller ANN configuration, which is shown in the following diagram:

Read now
Unlock full access