New Version of Kingsrow

[Richard Marlowe]

Here is an interesting article on the machine learning version of Kingsrow. Has anyone used it? Do you think it could challenge the full strength version of Chinook?

http://www.bobnewell.net/nucleus/checke ... temid=1177

[George Hay]

Machine Learning has advanced what was thought possible in computer programing for the game of Go by about a decade! It is incredible that Ed Gilbert has brought ML to Checkers. It would be an interesting match, ML Kingsrow vs Chinook, the computer program that weekly solved checkers in 2007. A GAYP match should result in no less than all draws for Chinook. However, in a three-move match, some of those openings go far astray from the original starting position, so Chinook may be vulnerable to some of the more difficult openings. I would like to se ML combined with supercomputers to strongly solve checkers.

--George Hay

[chipschap]

After working with KR/ML for a couple of months, and after numerous discussions with author Ed Gilbert, I would say that the latest KR/ML would clearly defeat full-strength Chinook, which after all is much older and hasn't progressed in many years. I'd go further and say that no current man or machine can stand up to KR/ML at the present time. Someone with great computing resources (e.g. Alpha) could certainly build a "deeper" program should they wish to do so, but I wonder if now we're down to ELO differences of less than 20 points.

Everyone serious about checkers needs to try KR/ML with the 10-piece database. It's phenomenal and resolves difficult positions almost instantly.