Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Fruit 2.1 tuning tests after 1120 blitz games

Author: m.d.hurd

Date: 12:35:49 07/08/05

Go up one level in this thread


On July 08, 2005 at 14:54:26, Marc Lacrosse wrote:

>Hi all !
>
>I just continued with my series of Fruit 2.1 tuning tests at 180+1 (3minutes + 1
>sec per move) blitz rate.
>This time I added Fruit 2.1 (default setting) playing with Dan Corbit's book for
>an actual grand total of 1120 games played.
>
>Results :
>- a strong trend according which Dan's opening book is better than my "MLCB4003"
>and "Nag09" ones by +- 25 Elo points.
>(congrats, Dan!)
>- a confirmation of the slight superiority of my "ML02" UCI parameters over
>Fruit's default parameters by +- 29 Elo points.
>
>Next step is evident :
>Fruit 2.1 + ML02 params + Dan's book against the field !
>I also intend to add other engines : Gandalf 6.0 and ProDeo 1.1 will soon be
>included.
>
>Marc
>
>  Program                Elo   +   -   Games   Score   Av.Op.  Draws   TOTAL
>
>1 DS9UCI-Def-MLCB4003  : 2779  36  36   280    62.7 %   2689   24.6 %  175.5
>2 F21-ML02-MLCB4003    : 2726  34  34   280    54.3 %   2696   31.4 %  152.0
>3 F21-Def-DCbook       : 2722  32  32   280    53.6 %   2697   37.1 %  150.0
>4 F21-Def-MLCB4003     : 2697  33  33   280    49.5 %   2700   33.2 %  138.5
>5 F21-Def-Nag09        : 2697  33  33   280    49.5 %   2700   33.2 %  138.5
>6 CM10Ber-Def-MLCB4003 : 2678  35  35   280    46.4 %   2703   27.9 %  130.0
>7 K7a-Def-MLCB4003     : 2666  35  35   280    44.5 %   2705   25.4 %  124.5
>8 F21-Def-NoBook       : 2636  34  35   280    39.6 %   2709   30.7 %  111.0
>
>Games        :   1120 (finished)
>
>White Wins   :    465 (41.5 %)
>Black Wins   :    314 (28.0 %)
>Draws        :    341 (30.4 %)
>Unfinished   :      0
>
>White Perf.  : 56.7 %
>Black Perf.  : 43.3 %
>
>
>PS : Test conditions :
>
>PentIV 2.2 Ghz 1024 MB ram or Centrino2 2Ghz 2048 MB ram
>Winboard 4.2.7a + Polyglot 1.3 (+WB2UCI for CM10)
>128 MB - no TB - ponder off
>3 minutes base + 1'' per move
>
>DS9UCI = Deep Shredder 9 UCI
>F21 = Fruit 2.1
>CM10Ber = ChessMaster 10 with Jason Kent's Berean 5.53 personality
>K7a = Ktulu 7a
>
>MLCB4003 : a medium size book from quality correspondence games
>DCbook : Dan Corbit's book
>NAG09 : a very small book adapted from my personal correspondence
>        competition repertoire
>NoBook : would you guess ?
>
>Def : default UCI engine parameters
>ML02 : my presently best tuning for fruit at blitz :
>
>NullMove Pruning=Always
>NullMove Reduction=3
>Verification Search=always
>Verification Reduction=5
>History Pruning=true
>History Threshold=100
>Futility Pruning=false
>Futility Margin=115
>Delta Pruning=true
>Delta Margin=80
>Quiescence Check Plies=2
>Material=96
>Piece Activity=130
>King Safety=90
>Pawn Structure=110
>Passed Pawns=120


Hello Marc

Have you tried Fruit 2.1 standard settings except Delta and Futility pruning set
to on (Fruit 2.1DF) This seems to offer a slight improvement from tests I and
others have done.

Regards

Mike



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.