Author: Uri Blass
Date: 10:06:47 03/04/04
Go up one level in this thread
On March 04, 2004 at 12:55:11, Sune Fischer wrote: >On March 04, 2004 at 09:41:26, Tord Romstad wrote: > >>On March 03, 2004 at 06:04:29, Sune Fischer wrote: >> >>>On March 03, 2004 at 04:18:30, Paul Byrne wrote: >>> >>>>Granted, it is not a terribly useful position, as it came from a wild 8 (pawns >>>>start on 4th rank) game on ICC between 2 patzers. And a very short search will >>>>get you a score of +10 to +15 and better in white's favor. >>>> >>>>It is, however, surprisingly tough to see the mate, and makes an interesting >>>>test position... >>>> >>>>r1b3n1/6k1/1n6/pP3qPQ/2PP2p1/1p4b1/4B3/R1BK2NR w - - 0 15 >>>> >>>>Guildenstern's normal search cannot see it here, while it's PN search gets it in >>>>under 30 seconds on an athlon 2400+. The line the PN search produces is mate in >>>>19, but is likely not optimal, as PN search is not terribly concerned with >>>>winning in the least number of moves. >>>> >>>>So, what programs can see the win from here? >>> >>>Not mine, I get a +22 score though. >> >>Mine also doesn't find it. After 11 plies, the score is +29. The >>problem might be that my search is "lazy" for the winning side in >>positions where one side seems to have a winning advantage. I don't >>extend checks, mate threats and other attacking moves as much as I >>would do if the score were close to 0. The idea is that it is more >>important to spend time looking for deep refutations of the apparent >>win than looking for an even more crushing win. > >Ditto. > >It seems like a good principle, only it makes the engine look bad in test >positions. Movei also does it but test positions are usually not about finding mates when there is a simple win but about finding the only winning move or the only drawing moves and in these positions it usually helps. Uri
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.