Author: Ryan B.
Date: 01:22:38 12/01/05
Go up one level in this thread
Gambit Fruit 1.0 Beta 4d found mate in 38 sec. Gambit Fruit uses scorpio bitbases now and the code to use the scorpio bitbases has also been sent to Thomas to add to Toga if he wishes to do so. I think with bitbases Toga and Gambit Fruit will be just as strong as Fruit 2.2.1 using EGTBs in endgames. I hope Fruit commecial moves to using bitbases as well. On December 01, 2005 at 04:00:30, Mig Greengard wrote: >3n3k/bp3pp1/8/8/R1Q2R2/6p1/B2p1PP1/q1rn2K1 w - - 0 1 > >Plokhodnikov, 1971 > >Was somewhat surprised to find that most engines need a while to solve this >puzzle since every move is a check. More than a minute, over five minutes in >some cases. I think Toga II 1.1 was the fastest, solving it on my Athlon 64 3800 >in around 40 seconds. (The original composition was given as mate in 21, but the >comps find a faster way.) > >I'm not an unreserved fan of this engine yet, but it is intriguing and well >worth a look beyond its scores against other programs (yes, there is such a >thing). It certainly doesn't seem worse than the other top programs. Since 99% >of my program usage is as an analytical assistant for GM games for reports and >newsletters, I mostly want fast tactics as I favor my own positional eval. In >this I don't find Toga better than Junior 9, inferior in most cases when it >comes to suggesting interesting tactical lines, Junior's specialty. > >It is quite good in endgames for an engine despite apparently not accessing EGTB >in the search. Does its cousin Fruit 2.2 do this? Probably not particularly >relevant in play but it's essential for better endgame analysis. (e.g. >Minasian-van Wely from the first round in the FIDE World Cup the other day. Toga >has no idea that endgame is totally drawn, even when it's down to bishops.) > >Like many players I know, I have my own informal test suite I use to evaluate >both new programs and new hardware. They are mostly games I have annotated >deeply (and/or have been so annotated by others) and know very well, and it's >useful to see if new engines find the best moves or see the value of certain >paradoxical ideas. And also how long it takes, of course, since speed is of the >essence in a working environment. I've found the latest editions of Kasparov's >"My Great Predecessors" books handy for this because he and his team also >thoroughly computer-check lines. (Famously not so well in Vol. 1.) This avoids >annoying refutations and Garry is also more interested in pointing out good >moves that aren't just flashy tactics a computer finds instantly. A good test >suite can be made just by thumbing through the books and looking for exclam >moves. > >What are the most highly recommended positional test suites in circulation, btw?
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.