Computer Chess News
Some interesting events happening in the world of chess. First, a chess machine named Hydra just finished a 6 game match against the super grandmaster Michael Adams. Hydra is interesting because it is a chess machine rather than just a chess program. It is built to use a number of FPGAs. It's difficult to make comparisons between different chess machines but it's reasonable to guess that Hydra is similar in speed to Deep Blue.
The results of the match surprised many. Adams got clobbered. He only managed to draw one of the games and lost the rest. I've not yet seen an in-depth analysis of the games but most people agree that Hydra's play was impressive. Some people have suggested that Adam's should have tried to steer the games into closed positions (computers generally have more trouble when the position is closed). I'm not sure that would have helped him since that type of play does not seem to be his style. A strong defender like Peter Leko may have been able to draw more games but I'm not sure that he could actually win more.
In other news, an amateur chess engine written by Fabien Letouzey and named Fruit has been surprising people with it's strength of play. I have not yet seen the results of through testing (see Kurt Utzinger's page for an example of how to do it right) but it appears to be nearly as strong as the best commerical engines. The really interesting part is that Fabien has released the code with an open source license.
Chess engine authors have traditionally been very secretive about their code (Robert Hyatt, the author of Crafty is a notable exception). It's interesting to observe the chess engine community react to this event. Some suggest that Fabien is doing a disservice since a strong open source engine makes it harder for commerical authors to earn a living. Others are worried that people, heaven forbid, will make clones of Fruit.
The Fruit source code does make for interesting study. The code comes with a short set of technical notes written by Fabien. Fruit uses a straight forward board representation (a varient of the clever 0x88 idea). It's search algorithm is also relatively simple, a form of the classic PVS (principle variation search). It's evaluation function is also simple. One nice twist is that Fruit always shows the complete PV line (because of transposition tables, some engines truncate it at times). The C++ source seems to be relatively well written. Overall, it looks like a nice starting point for aspiring chess engine authors.
Finally, in my last bit of news, the developers of Fritz have stated that they are no longer optimizing their engine for play against other engines but rather against humans. I wonder if that has something to do with the fact that Shredder is clearly stronger in engine vs engine matches.