Author: Uri Blass
Date: 19:31:11 12/20/05
Go up one level in this thread
On December 20, 2005 at 21:25:25, John Merlino wrote:
>On December 20, 2005 at 20:49:11, Mike Byrne wrote:
>
>>Time Control Search depth ==1
>>
>>I set this tournament up because in playing Rybka at sd==1, it seemed to be that
>>Rybka was unsually strong at this level, much stronger tha other engines, I
>>usually my hold against engines when search depth==1 - that is not true against
>>Rybka. Here are the results.
>>
>>Ponder off , common book for all engines, common hash.
>>
>>Rybka is clearly doing some extra things at sd==1 that other engines are not
>>doing.
>>
>>Quad round robin ( 36 rounds)
>>
>> Engine Score Ry Za SO He Sh An Mi Pr Ru Cr
>>1 Rybka v1.0 Preview.x64 34.5 ···· 1111 11=1 1111 1111 1011 1111 1111 1111 1111
>>2 Zappa 19.5 0000 ···· =01= 1=== 111= 0=10 0010 0=11 111= 1011
>>3 SOS 5.1 for Arena 18.5 00=0 =10= ···· 10=1 1010 111= 0110 1101 ==00 0011
>>4 Hermann 1.5 18.5 0000 0=== 01=0 ···· 11== =101 0111 01=1 =0== 1011
>>5 Shredder8 16.5 0000 000= 0101 00== ···· 1=11 =1=1 0101 1010 1=10
>>6 AnMon 5.50 16.0 0100 1=01 000= =010 0=00 ···· 0110 1100 =111 10=1
>>7 Michael Byrne {Human} 15.5 0000 1101 1001 1000 =0=0 1001 ···· 100= 1=10 1=01
>>8 Pro Deo 15.0 0000 1=00 0010 10=0 1010 0011 011= ···· 0101 11=0
>>9 Ruffian 1.0.5 14.0 0000 000= ==11 =1== 0101 =000 0=01 1010 ···· 0==1
>>10 Crafty64 12.0 0000 0100 1100 0100 0=01 01=0 0=10 00=1 1==0 ····
>>
>>180 games played / Tournament finished
>>Name of the tournament: RoundRobin
>>Site/ Country: WIN-64-MB, United States
>>Level: 1 Half moves
>>Hardware: Dual AMD Opteron(tm) Processor 244 1804 MHz with 2,046 MB Memory
>>Operating system: Microsoft Windows Server 2003 Professional Service Pack 1
>>(Build 3790)
>>
>>
>>12/20/2005 8:32:46 PM :
>>
>>I played my games (human) at about 5"/min per game pace. My rating is about
>>1600, so I set elostat with me at 1600. Rybka is playing over 400 point
>>stronger than the next engine at sd==1.
>>
>>
>> Program Elo + - Games Score Av.Op. Draws
>>
>> 1 Rybka v1.0 Preview.x64 : 2133 55 253 36 95.8 % 1589 2.8 %
>> 2 Zappa : 1669 102 101 36 54.2 % 1640 25.0 %
>> 3 SOS 5.1 for Arena : 1652 105 105 36 51.4 % 1642 19.4 %
>> 4 Hermann 1.5 : 1652 97 97 36 51.4 % 1642 30.6 %
>> 5 Shredder8 : 1617 105 106 36 45.8 % 1646 19.4 %
>> 6 AnMon 5.50 : 1608 107 108 36 44.4 % 1647 16.7 %
>> 7 Michael Byrne {Human} : 1600 109 111 36 43.1 % 1648 13.9 %
>> 8 Pro Deo : 1591 111 113 36 41.7 % 1649 11.1 %
>> 9 Ruffian 1.0.5 : 1573 100 102 36 38.9 % 1651 27.8 %
>> 10 Crafty64 : 1535 110 114 36 33.3 % 1655 16.7 %
>
>Given such a huge difference in score, I would begin to suspect a bug in Rybka's
>implementation of "sd=1".
>
>I find it hard to believe that Rybka's evaluation (which is all that is being
>tested here, since the entire search algorithm should be rendered useless with
>"sd=1") can solely account for an increase in 400 ELO over Zappa, Shredder, et.
>al.
>
>jm
I found that Rybka depth x is almost equivalent to Movei depth x+2 for x=1,2,3
so it is clear that rybka searches more at fixed depth that is higher.
Uri
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.