Author: Robert Hyatt
Date: 07:43:39 11/22/02
Go up one level in this thread
On November 22, 2002 at 04:51:34, Richard Pijl wrote: >On November 21, 2002 at 15:42:13, Robert Hyatt wrote: > >>On November 21, 2002 at 12:39:11, Richard Pijl wrote: >> >>>On November 21, 2002 at 12:25:19, GuyHaworth wrote: >>> >>>> >>>>It ought to be possible to substitute for the 5-1 EGTs with some more compact >>>>rulebase. Has 5-1 ever occurred over a serious board? I guess not. >>>> >>>>How would one set about creating that rulebase to cover all situations? Is it >>>>just "5-1 = WIN"? >>>> >>>>g >>> >>>That is probably a bad idea. If all positions would be scored the same, how >>>would you make progress in check-mating the king? >> >>See my response. I think it is doable with code I already have in Crafty that >>handles "mop-up" cases (no pawns) easily... >> > >My remark concerned the rule 5-1 = WIN. That is not true in every case. You can >also construct a lot of stalemate situations, so the rulebase should cope with >them too. And a single value for a score would be even worse as no progress will >be made. That was my point. I hzve specific code to recognize most drawn positions that usually deal with rook pawns and the like. Therefore returning something like WIN+Evaluate(position) would do the trick, because Evaluate() can decide that the position is drawn and I can discover that and just return 0. > >When you mentioned the mop-up code in Crafty I was started looking for it and I >think you mean EvaluateMate(). But that is only called in evaluation, so in leaf >nodes. Stalemates are handled in search, and you use make sure that you measure >progress in the score as well. Although it is more efficient than a regular >evaluation it is not the same as probing an EGTB as that is done during search >in the none-leaf nodes, enabling to do a beta-cutoff fast. Correct. But I had mentioned adding a call to Evaluate() where I do the EGTB probe stuff and notice it is a 5 vs 1 ending where the files are really not necessary. At that point it would occasionally be doing an eval where it does the probe so that it can be fairly sure it is not drawn before assigning a score of WIN + static eval. > >But well, this discussion is a pretty academic one as a 5-1 piece majority >doesn't happen all of a sudden :-) > >Richard.
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.