Hyperparameter selection
The objective function used in the experiments described in this chapter is based on a novelty metric that has no clear upper-boundary value. As a result, the fitness threshold value cannot be estimated precisely. Thus, to signal that the winning solution was found, we use an indicative value that is big enough to not be encountered during the normal algorithm execution.
We selected 800000 as the indicative novelty score value. However, to maintain the visual presentation of the fitness scores when plotting the results of an experiment, we scaled down the obtained novelty scores of the solver agents using the natural logarithm. Thus, the fitness threshold value used in the configuration file becomes 13.5, which is ...
Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.
Read now
Unlock full access