Author: Dann Corbit
Date: 20:44:04 01/13/03
Go up one level in this thread
On January 13, 2003 at 23:14:56, scott farrell wrote: >Guys, > >I am after some comments here. I have some code and ideas I need people to help >validate. > >My code uses the detect draw-by-rep routine/hash etc. Currently I only run this >when there is a bishop or knight or less on the board, as I dont know how safe >the code is yet. It compares the pawn hash to all previous pawn hashes in the >tree, if it finds a hit, and the material is the same , it scores it as 0.0, >meaning: if the pawns are in the same config, and material is unchanged, then >with all the searching you have done you cant progress at all, just call it a >draw. > >Does anyone do any blockade checking in their code? >Is my idea similar, different, stupid. > >My chompster lost a drawn game recently, the position was similar to this below. > It was down material, and found a long sequence of moves where it could get 1 >pawn back, and so it did, at the expense of opening the postion and then loosing >which was beyond the horizon. To a human it is clearly a draw, just blockade >!!!. > >This is an over simplfied position, so it is simple enough for me personally to >work out. > >Clearly either Ke3 or Be3 draws. > >Ruffian thinks the score is -1.1 for white at about 16 plies and 30 secs. > >When I enable my new blockade code it sees a score of 0.0 for Kd3 immediately - >which is great. > >What is even better is that for black to move in the same position without the >blockade code it thinks it is winning big time, and just shuffles pieces around, >horizoning the draw-by-rep, and doesnt quite get to 100 plies to see draw by >fifty moves. > >With the blockade code it first thinks of e3 as the only "progress" it can make >against the blockade, then it tries some stalling moves in between to push >things over the horizon (oops). After a fail low on depth 8, it realises it is >blockaded, and after a research at full width it returns Bc6 with a score of >0.08 (I think this is just an artifact of lazyeval and should be 0.0), and knows >white is going to play Ke3 immediately. > >I think this is amazing progress for my chompster, and an extra 5 lines of code >added ... amazing !!!! > >Maybe I can take some nice draws in CCT5 against ruffian !!!! ... I wish :) > >[D] 8/8/1B2k3/1b2p3/4p3/3p4/3P1K2/8 w - - 0 1 >[D] 8/8/1B2k3/1b2p3/4p3/3p4/3P1K2/8 b - - 0 1 Heuristically, it is a very interesting idea. I wonder if it is sound?
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.