The strength of the game program draughts.

[Krzysztof Grzelak]

On what basis you can determine the strength of the game program draughts and which program is the best. We cordially invite you to a discussion on this topic.

[warpspeed]

In the document Schwarzman versus Maximus of Jan-Jaap van Horssen is on page 15 the estimated rating of Maximus 2350 on the basis of the described matchresults. This is probably with a large error margin since the match consisted of only 6 games. On page 7 and page 8 the used hardware for Maximus is described :

This also seemed the right time to buy a new computer, an Intel Core i7-3930K 3.2 GHz 32GB RAM, a 6-core with hyper-threading.

Maximus settings for the match were: 12 search threads (10 for pondering) and a hash table of 128M entries. The Java process uses 6.5 GB RAM in total. At tournament conditions, Maximus is able to search more than 23 million positions per second, with an average search depth of 24 ply

On the basis of the results of two matches of Maximus against Kingsrow 1.52, described on page 7, the rating of Kingsrow 1.52 on the same hardware is estimated to be 2420.

[warpspeed]

On this forum posted by Ed Gilbert, author of Kingsrow, were the results of a match between Scan 2.0 and Kingsrow 1.56 of 988 games. The score was about 56% for Scan 2.0.

[warpspeed]

According to this post of BertTuyt the rating difference between Scan 2.0 and Kingsrow 1.56 can be calculated as follows :

400 * log10 (0.56 / 0.44) ≈ 42

So assuming Kingsrow 1.56 is stronger than Kingsrow 1.52 the rating of Scan 2.0 on the same hardware where Maximus played on should be at least 2462 (2420 + 42). To this rating 2462 should be added the rating difference between Kingsrow 1.56 and Kingsrow 1.52.

The provisional cautious conclusion (due to some error margin) should then be the rating of Scan 2.0 on the hardware Maximus was playing on should come very close to the highest rating of a human, which should be the current rating of Ton Sijbrands which is 2476 according to the most recent rating list.

It seems therefore possible for Scan 2.0 on even better hardware to defeat Ton Sijbrands in a match.

[warpspeed]

Meanwhile I'm wondering if anyone else has any new information about the estimated playing strength performance of the current best draught programs on high-end amateur hardware. Is this beyond worldchampion-level yet?

[BertTuyt]

There is an interesting post in this forum, dealing with this topic.
See "Feike Boomstra Match: a unique man-machine event"
Based upon this analysis, we are closing in, but there still remains a small gap.

Bert