Author: Robert Hyatt
Date: 21:37:19 02/16/99
Go up one level in this thread
On February 16, 1999 at 23:06:12, Peter McKenzie wrote: >On February 16, 1999 at 22:34:32, Robert Hyatt wrote: > >>On February 16, 1999 at 16:26:53, James Robertson wrote: >> >>>My program keeps falling prey to king attacks. Although it does very well >>>tactically when it's king is not threatened, it frequently plays combinations >>>that win material (it thinks), only to find, say, a back-rank mate that went >>>unnoticed. I am using a one-reply-to-check extension, the mate extension after >>>null move, and extending half a ply whenever there is a check in the tree. No > >I haven't tried the mate extension after null move yet, do you think it is very >effective? it will help you solve wac141 quickly, and seems to help some in other positions. But it can also be expensive at times. > >>>check detection is done in the q-search. Are there any other standard check >>>extensions I am not doing? > >I've heard that some programs extend on captures by (or adjacent to maybe) the >King, but I've never tried this. > cray blitz was the first program I know of that did this. Others certainly may have. I did it because I had a position once where if the king was on g8, we found Bxh7+ with mate to follow. But if the king was on h8, it didn't, yet the resulting positions were essentially identical (in either case the king was obligated to play Kxh7 or the mate was easy to see...) >>> >>>James >> >>Half a ply for check is very conservative. I use 1 ply for all checks, 3/4 >>ply for one-legal-reply-to-check... > >Yes, I extend a full ply for check - I think this is pretty standard. >I don't do partial ply, so I limit the number of one-legal-reply-to-check >extensions to something like 3 in any given branch of the search tree. > >> >>but speed is the main thing you need. Or else resorting to selective approaches > >q-srch is of course a selective search, I think it is a common idea to try >checks in the q-srch. Either only in the first N ply of q-srch or only 'good >looking' checks. > >The other thing to try is playing all responses to check in the q-srch, even if >they aren't captures. > >These changes to your q-srch will of course cost you nodes, it is debatable >whether they are worthwhile. > I did that in CB. I did it in early crafty versions. I don't now. It is a catch-22. But it is nice to get to totally ignore the q-search and spend time on one piece of code, the normal search. >>to get the depth you need. Or else tuning your evaluation to protect your king >>better. The chess servers will _really_ help you do this, of course... > >Sure does! > >Peter
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.