Author: Dan Newman
Date: 22:47:38 06/02/99
Go up one level in this thread
On June 02, 1999 at 06:58:04, Bernhard Bauer wrote: >On June 02, 1999 at 05:00:17, Dan Newman wrote: > >>Well, I tried mine out on these at 1 min / position on a P6/200 >>and got the key move in 10 out of 16, but no mates. I then tried >>them at 1/2 hr each and still got 10/16 and no mates. Finally, I >>tried the first one for two hours -- still no mate... >> >>On the next to last (BWTC.0647) my program likes to force a >>repetition draw all the way through the 18 ply search -- no >>guarantee there isn't a mate though. >> > >This position is so simple and nice that you should have a look at it >and solve it yourselv. Hmm. I did look at it myself, but I'm afraid I'm a far better chess programmer than player, so I couldn't make much sense of it... >Not only your program fails, Crafty16.8 fails too. >I guess your program depends heavily on null move? Yes. I didn't suspect null move failure here because of the two Queens. I did get the right move though... (Elsewhere in this thread someone identified this as a mate-in-12 -- which is 24 plies deep. I only got through the 18 ply search in 1/2 hour, and that's with null move which greatly reduces the effective branching factor. I didn't find the mate with null move off in an hour long search, either. It got through the 16 ply search and was working on the 17 ply.) >Oh, I'm missing the point again, sometimes I feel like a null-move-basher. >But as we all know such positions do *never* apear in practice, so why care? >And our programs are real grandmasters with elo > 2600 anyway, or not? Well, I care, but null move gives you so much for so little cost -- just a test or two and 8 or 9 lines of code. The number of positions in which it is of benefit far outweigh the few in which it fails, and a simple test for only pawns on the board removes a huge fraction of the failures (one hopes). I guess one could take Hsu's solution and just remove null move from the program. But in his case he could afford it. The other solution is to find some way of detecting zugzwang or some simple way of recognizing such positions. Of course if it takes too much cpu time to do this, it will take away from the gain you make by doing it, perhaps all of the gain and then some. This is one of those problems I'd like to look at, but it looks to be very difficult, and with so many holes in my eval to fill and other improvements to make, it's fairly low down in priority. -Dan. > >Kind regards >Bernhard > >>-Dan. >> >>PS. I noticed the EPD strings weren't quite standard: there >>shouldn't be a semicolon directly after the EP target square, >>but there should be one after the last "operation". My program >>didn't seem to mind though...
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.