Author: Uri Blass
Date: 13:19:14 11/08/01
Go up one level in this thread
On November 08, 2001 at 15:35:16, Dann Corbit wrote: >On November 08, 2001 at 12:43:23, Uri Blass wrote: > >>On November 08, 2001 at 11:04:59, Mark Young wrote: >> >>>On November 08, 2001 at 10:20:37, Uri Blass wrote: >>> >>>>I define a real sacrifice of chess program as a move that is losing material >>>>based on only material evaluation but is played by the program because the >>>>positional score is bigger than a pawn. >>> >>>Most of the Real sacs I have seen are based on mating attacks. Do you define >>>this as a real sacrifice? >> >>If the program can see the mate or can see material that it get I define it as a >>combination and not as a real sacrifice >> >>The test if it is a real sacrifice can be done by a materialistic program with >>good search rules(positional scores always lower than 0.5 pawn so we can know >>the material score based on the score of the program). >> >>If the materialistic program can only see that the sacrifice is losing material >>even after a long search then it is clearly a real sacrifice. > >I think by your definition, a chess sacrifice by a computer would be where the >positional score is more than a pawn (piece?) even though the material goes >negative. I have seen lots of things like this in Crafty, but it is usually >only a pawn or so rather than a full piece. So I guess those are really >gambits, rather than sacrifices. > >Colin's program Beowulf can have some very large positional scores. If you look >at the code where board control is computed, you will see that a single simple >aspect like that can be worth much more than a pawn. > >Probably all good programs would "sacrifice" under this definition, given the >right circumstances. I say that because if you have a material only eval, the >moves chosen will be wildly different from a good eval plan. programs with positional scores that are smaller than 1/2 pawn are not going to sacrifice by my defintion when I define material score by the following values(pawn=1 ,knight=bishop=3,rook=5,queen=9) I agree that all the top programs are going to sacrifice by my definition but the question is still which program sacrifices more. I remember that I read that Junior sacrifices more than Hiarcs when Fritz is in the middle. In fact, any >program that uses material only eval is sure to lose to any decent chess >program. > >A clear, numerical definition of what you mean by sacrifice would help, I think. > >Another interesting test would be to take a dozen "classical" sacrifices by >Morphy, Fischer, Tal, whatever, and run them through computer programs. See >which programs make the same moves (if any). I suspect that sometimes what we >perceive as a sacrifice by a great player was a very deep tactic, which the >superGM saw all along. It is also possible that from the point of view of the player a move was a sacrifice but computers can find it because from their point of view it is tactics. I think that an objective definition is to take a good program with mainly material evaluation(we only need maximal positional score that is smaller than 1/2 pawn) and if this program believe based on evaluatiion that it was a sacrifice then we define it as a sacrifice We can give the mainly material program to search for 10 minutes the position before the move and the position after the move to decide if the move was a sacrifice. 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.