Author: Andrew Wagner
Date: 05:52:37 03/13/04
Go up one level in this thread
On March 13, 2004 at 08:40:13, scott farrell wrote: >On March 13, 2004 at 07:33:08, Andrew Wagner wrote: > >>In this position: [d]rn1qk1nr/pp3ppp/4p1b1/3p4/1b1P1BP1/2N1P3/PP3P1P/R2QKBNR w >>KQkq - 1 8 >> >> >>I got this output: >> >>Trueno(C) kibitzes: depth=1; score=-63; Time=0; nodes: 111; PV=g4-g5 >>Trueno(C) kibitzes: depth=2; score=788; Time=0; nodes: 105; PV=g4-g5 b4-e7 >>Trueno(C) kibitzes: depth=3; score=751; Time=1; nodes: 2521; PV=g1-f3 b4-d6 >>a1-b1 >>Trueno(C) kibitzes: depth=4; score=34; Time=3; nodes: 6987; PV=g1-f3 d8-a5 d1-a4 >>a5-a4 >>Trueno(C) kibitzes: 12 seconds, 810 nps >>Trueno(C) kibitzes: MoveOrder%: 81 >>Trueno(C) kibitzes: Hash Percentage: 7 >>Trueno(C) kibitzes: Hash probes: 62105 >>Trueno(C) kibitzes: Hash hits: 4459 > >looks to me like a qsearch problem. ?? Can you be a little more specific? > >also, nodes in ply 2 look a little odd, being less than ply 1 - its not >neceessarily wrong, just strange. I'm incrementing the number of nodes in my eval function, so only counting nodes that actually get all the way to eval. So to me, it would make sense that if it's getting a bunch of info from the hash tables it would have to evaluate a lot less nodes. Or am I missing something? > >Try reversing all colours, and run eval, try reversing stm and running eval, >they all should be equal, this sometimes shows you errors in your eval function. My eval function works great with hash tables turned off. The whole engine does. For the depth that it gets, it plays pretty well, I think. The problem has to be in the hash tables. > >810 nps is a problem, are you checking the time at each node? you should only >check every x nodes, checking the clock is very very slow. I check the clock on every iteration. I know this is very slow, and I'm not sure why. It does somewhat better without hash tables, but still very slow. > >Scott
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.