Author: Ross Boyd
Date: 03:24:41 12/04/05
Go up one level in this thread
On December 03, 2005 at 23:39:19, chandler yergin wrote: >All expressions of great Joy. >"Eureka cried the ancient Philosopher! The religious have an Epiphany! >I just had another Beer.. and it came to me! >I finally understand.. (I think) what you guys are doing, and want to do better. >(If I don't I'm sure you'll tell me. >;) >You need a way to Test your Engines; and Tweak them and have irrefutable >evidence of performance. In other words, a "Benchmark"; a Baseline for >comparison purposes. >Most of you use Test Positions! The problem there is, that every Engine will >evaluate a poition differently. Being good at solving test positions does not directly translate to OTB strength. The best engines are able to play many good moves in succession. A poor engine will throw in some real ugly ones. >What is the one thing in Chess that will never change? >What is the one Constant that will never change? > >The number of "possible" moves at various Ply Levels! >Here they are: > > 0 1 >ply 1 20 >ply 2 400 >ply 3 8890 Dang, you were going great till you hit the 3rd ply.... The correct numbers from ply 3 to ply 7 are: ply 3 8902 ply 4 197281 ply 5 4865609 ply 6 119060324 ply 7 3195901860 Where did your numbers come from? It seems to me you subtracted all the 'check'ing moves from the correct figures. Why? >ply 4 196812 >ply 5 4838258 >ply 6 118251225 >ply 7 3162798012 >ply 8 84029997363 >ply 9 2403434332264 >ply 10 68265214423776 >ply 11 2058141026024096 >Now you can start your Engines and check the times to Depth for various >Hash Levels, etc. and see immediate results. > >This way Players & Programmers have the same Benchmark! The test you describe is commonly called the 'perft' test. It's main purpose is to test the integrity of a move generator. In practice, it's a useless benchmark for tuning or comparing engine strength. If writing/creating a top engine was so simple the answer would have been discovered long long ago. Currently there are several hundred (known) programmers that have published engines. Very few produce something that plays above 2500. Its not a trivial task. And for anyone who hasn't attempted it yet, the chance of thinking of some new approach that hasn't been 'thunk' of before is highly remote. Keep trying... Ross
This page took 0.01 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.