Author: Uri Blass
Date: 23:43:32 01/26/01
Go up one level in this thread
On January 27, 2001 at 01:48:17, Robert Hyatt wrote: >On January 25, 2001 at 13:40:54, Miguel A. Ballicora wrote: > >>On January 24, 2001 at 23:39:12, Robert Hyatt wrote: >> >>>recapture is the most interesting case. It actually appears that removing >>>this _completely_ gave the best score, which is surprising since >>>win at chess is a tactical suite. Lowering this extension obviously >>>speeds up the overall search. I currently use .75 here as well. >>> >>>The most interesting thing is the recapture extension. I have used it forever, >>>and took the idea from Cray Blitz since I used it there as well. I am not yet >>>sure why it actually hurts since it seems (to me) to be a logical idea. I am >>>going to test this more, just for my own interest. >> >>I have the feeling the recapture extensions could be useful in positional >>tests more than in tactical. In tactical tests, most of the the time the >>solution does not involved a capture and a recapture like Bx(N)c6, bxc6. >>Generally, those tactical tests are tough for humans because there are >>sacrifices, quiet moves, etc. but no simple trades. In fact, you might see that >>you need a negative extension because the test suite is telling "this is a >>simple trade! the solution is not here! do not waste your time in this >>variation!" > > >Don't think of the recapture extension as a way to see long sequences of >captures. That isn't what it is for. It is to handle the case where a capture >and forced re-capture push something _else_ beyond the search horizon, and >as far as the engine is concerned, out of the universe. IE the recap extension >is mainly to prevent the horizon effect, rather than something used to see >deeper to make more captures... > > > > > >>Humans do that when they are faced with a problem rather than a typical game. >>They suspect that the obvious move is not the key move. You know that the first >>move most probably is not check, so you do not consider that first... >>In a real game you do! I wonder if using this type of suite creates this >>artifact. > > >Don't think of this extension as having anything to do with how a human >plays chess. If I am analyzing a tactical line, I generally would ignore >a BxN NxB on the other side of the board. But the computer can use this >to push the horizon to a point it likes... > > > > >> >>In a positional test there might be a lot of those "obvious" trades as part >>of important lines of the tree. Can you try and test recaptures in a >>positional test? I am very curious. Thanks, > > >We would have to agree on some positional test positions. IE are we talking >about a position where there _is_ one right move, but not because of tactics? > >Those are hard to find/quantify... There are positions when there is one right move from crafty's point of view and not because of tactics. I believe that if you analyze a lot of positions you will find a lot of cases when Crafty converge to a move after more than a second and does not changes its mind in a lot of iterations after finding the move. I think that these positions are good test positions because you do not test a change in the evaluation but a change in the extensions so I guess that usually crafty will converge to the same move after removing the recapture extensions and the only question is if it is going to do it faster or slower. 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.