Author: Robert Hyatt
Date: 07:12:46 01/18/01
Go up one level in this thread
On January 18, 2001 at 05:14:48, Bas Hamstra wrote: >On January 17, 2001 at 12:55:50, Robert Hyatt wrote: > >>On January 16, 2001 at 21:09:12, John Wentworth wrote: >> >>>On January 16, 2001 at 18:47:03, Robert Hyatt wrote: >>> >>>>On January 16, 2001 at 17:30:51, John Wentworth wrote: >>>> >>>>>On January 15, 2001 at 22:55:24, Wayde Beasley wrote: >>>>> >>>>>>I bought Fritz the other day and have been playing with it. Slowly, surely, it >>>>>>comes upon me a kind of gestalt experience: FRITZ IS SOOOOOOO GERMAN! >>>>>> >>>>>>Whereas most other non-German engines strive for some elegance, this totally >>>>>>German approach goes for broke in pure tactical brilliance, pure calculational >>>>>>mastery. >>>>> >>>>>To bad it doesn't exist anymore :( >>>>>What programs would be on an American team, other than Crafty ? >>>> >>>> >>>>We have our share... Crafty... Ferret... Gnuchess... Wchess... >>>> Zarkov... Mchess Pro... The various versions >>>> of Socrates at MIT... HiTech... Cray Blitz if >>>> we tried hard... >>>> >>>>Etc. I am sure I missed several. >>> >>>That would be cool, I would love to see a Cray Blitz make a come back, maybe a >>>renew the rivalry with Belle. >> >> >>Those were fun days. I think that Belle is now in the Smithsonian as an >>exhibit, so it won't play again probably. Cray Blitz is still around, but I >>doubt it will ever play again either. I think that once we get singular >>extensions working in Crafty, it will pretty well close the final gap between >>it and Cray Blitz, making trying to find a Cray a waste of time... > >Why is it so hard to make it work in Crafty? I mean, you did it before, so...? > >Bas. The problem is that the search I use is still recursive, which makes the parallel part a bit messy. And the SE code has to lay on top of the entire search, including the parallel part. In addition there are issues about re-searching. IE on PV-singular moves, you get the score for the first move, then you search the remainder with an offset (down) window. If any of those fail high, you have to re-search them with a different window to get a good comparison with the original to test for singularity. For fail-high singular it is a bit messier, and re-searching is giving me some grief with respect to hashing. IE if I research something and it fails high, it clobbers my PV. But if it isn't better than the best move, then I get wrecked. The bottom line is there is a ton of bookkeeping to do to make this work right. Cray Blitz is different in several ways, the most noteworthy is that the search is not recursive. Yes I know how I did it in Cray Blitz. But that doesn't make it much easier to do it right in Crafty. But it will be done, and I'll give some test data here when it is ready, whether I choose to keep SE or not.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.