Author: Robert Hyatt
Date: 18:53:10 01/09/04
Go up one level in this thread
On January 09, 2004 at 04:29:25, Vasik Rajlich wrote: >On January 08, 2004 at 12:01:56, Robert Hyatt wrote: > >>On January 08, 2004 at 10:02:39, Anthony Cozzie wrote: >> >>>I've been trying to get some good test positions to rewrite my kingsafety with, >>>and playing Gothmog will give you those :) Usually Zappa is slightly better than >>>Gothmog. It is stronger tactically (gets more nps and does not do all dubious >>>forward pruning) and its mobility is good enough that it plays reasonable moves >>>most of the time. Usually Zappa will get its wins in the endgame, but Gothmog >>>will get in enough king attacks that the score is close. During one of the test >>>games, this position came up (Zappa-white, Gothmog-black): >>> >>>[D]r2qk2r/pb4pp/3bn3/1P1ppp2/8/BQP5/P2NBPPP/R4RK1 b kq - 4 17 >>> >>>Where Gothmog evaluated itself as +1, and Zappa thought it was about equal - >>>which is of course ludicrous since black has a big advantage. I was less than >>>happy with Zappa being blown off the board in 20 moves and not even realizing >>>it, so I wrote some code to evaluate control of the board. But I made a mistake >>>and tuned it way too high. >>> >>>What was amazing was the result. Every game went like this: >>> >>>Zappa pushes pawns >>>Gothmog sacrifices a piece for the pawns >>>Gothmog thinks its down .5 >>>Zappa thinks its up 2 >>>Zappa wins back pawns and the game with its extra piece/exchange >>> >>>Result: Zappa scores 86% - +18-0=7. >>> >>>Moral of the story: If you want to beat Gothmog, triple your space evaluation. >>> >>>anthony >>> >>>caveat: This was done under Arena, with the arena books, which need a bit of >>>work. >> >>Another moral: Don't trade a piece for 2-3 pawns. It will come back to >>haunt you more than it doesn't. > >Bob, > >do you have any statistics? It would be interesting to see the scores of an >engine as the piece vs pawns weight is adjusted. As I vaguely remember Crafty >has a huge piece weight, 4.2 pawns in middlegame positions (3 for the piece >itself, 1.2 for the bad trade bonus) even when the bishop pair is not involved. > >In my program I use piece == 3.75 pawns if all material is on the board, and >piece == 2.75 pawns if all material is off the board, with everything scaled >linearly in between. The bishop pair, if it exists, is a separate bonus. Haven't >tried to tune it. > >Vas The bad trade idea was to eliminate something I saw happen for 30+ years. Even Cray Blitz got into trouble and it didn't have nearly as large positional scores as I am using in Crafty. Years ago I got tired of seeing the N for 3p case happen, as except for endgames it almost _always_ loses.
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.