Changes

Jump to: navigation, search

Engine Similarity

1 byte removed, 15:56, 24 September 2019
no edit summary
One approach to assess engine similarity is to [https://en.wikipedia.org/wiki/Similarity_measure count] how often these engines agree with the same move after a shallow [[Search|search]]
over a set of carefully selected, quiet [[Test-Positions|test positions]] with apparently several possible best moves - another, in '''k-best''' mode, how often they propose the same ranking of k moves.
A relative high similarity measure between two engines <ref>i.e. > 65% in SIM03></ref> could be a symptom of using similar evaluation ideas, features and weights, along with [[Automated Tuning|automated tuning]] (or trainig), or even code copying aka [[:Category:Clone|cloning]].
Similarity testing is intended as first automated "[https://en.wikipedia.org/wiki/Screening_(medicine) screening]" to trigger further investigations in case moves are too similar.

Navigation menu