Page 1 of 5

Time management: Rybka vs IvanHoe

Posted: Tue Jun 29, 2010 1:01 pm
by kingliveson
It has been determined that using specific time management configurations, Rybka 4 performs better at repeating time controls. IvanHoe also has time management configuration that counters Rybka's. I have been testing some settings simultaneously on 2 different hardware, and would say the data is inconclusive right now.

Code: Select all

Ivanhoe v Rybka R002, 4'/40+4'/40+4'/40  2010 

1   Deep Rybka 4 x64   +14/=28/-8 56.00%   28.0/50
2   IvanHoe 9.57b x64  +8/=28/-14 44.00%   22.0/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 256 MB Hash
† Ponder Off
† Athlon 64 X2 6000+ @ 3100 MHz
† 2 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 20 (default 15)
TimeOrdinaryFactor 60 (default 75)
TimeAbsolutePercent 20 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)

Pawn Hash 16 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off


On System 2

Code: Select all

Rybka vs IvanHoe , 4'/40+4'/40+4'/40  

1   IvanHoe 9.57b x64       +12/=28/-10 52.00%   26.0/50
2   Deep Rybka 4 SSE42 x64  +10/=28/-12 48.00%   24.0/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 512 MB Hash
† Ponder Off
† Phenom II 940
† 4 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 20 (default 15)
TimeOrdinaryFactor 60 (default 75)
TimeAbsolutePercent 20 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)
Pawn Hash 32 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off

Game data available: http://chess.cygnitec.com/pgn/

Re: Time management: Rybka vs IvanHoe

Posted: Tue Jun 29, 2010 1:06 pm
by kingliveson

Code: Select all

Ivanhoe v Rybka R002, 4'/40+4'/40+4'/40  2010

1   IvanHoe 9.57b x64  +12/=28/-10 52.00%   26.0/50
2   Deep Rybka 4 x64   +10/=28/-12 48.00%   24.0/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 256 MB Hash
† Ponder Off
† Athlon 64 X2 6000+ @ 3100 MHz
† 2 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 25 (default 15)
TimeOrdinaryFactor 80 (default 75)
TimeAbsolutePercent 100 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)

Pawn Hash 16 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off


On System 2

Code: Select all

Rybka vs IvanHoe , 4'/40+4'/40+4'/40  

1   Deep Rybka 4 SSE42 x64  +14/=23/-13 51.00%   25.5/50
2   IvanHoe 9.57b x64       +13/=23/-14 49.00%   24.5/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 512 MB Hash
† Ponder Off
† Phenom II 940
† 4 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 25 (default 15)
TimeOrdinaryFactor 80 (default 75)
TimeAbsolutePercent 100 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)
Pawn Hash 32 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off

Game data available: http://chess.cygnitec.com/pgn/

Re: Time management: Rybka vs IvanHoe

Posted: Tue Jun 29, 2010 1:14 pm
by kingliveson

Code: Select all

Ivanhoe v Rybka R002, 4'/40+4'/40+4'/40  2010 
                   
1   IvanHoe 9.57b x64  +11/=32/-6 55.10%   27.0/49
2   Deep Rybka 4 x64   +6/=32/-11 44.90%   22.0/49
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 256 MB Hash
† Ponder Off
† Athlon 64 X2 6000+ @ 3100 MHz
† 2 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 20 (default 15)
TimeOrdinaryFactor 240 (default 75)
TimeAbsolutePercent 80 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)

Pawn Hash 16 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off


On System 2

Code: Select all

Rybka vs IvanHoe , 4'/40+4'/40+4'/40                            

1   Deep Rybka 4 SSE42 x64  +10/=30/-10 50.00%   25.0/50  0.00
2   IvanHoe 9.57b x64       +10/=30/-10 50.00%   25.0/50  0.00
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 512 MB Hash
† Ponder Off
† Phenom II 940
† 4 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 20 (default 15)
TimeOrdinaryFactor 240 (default 75)
TimeAbsolutePercent 80 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)
Pawn Hash 32 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off

Game data available: http://chess.cygnitec.com/pgn/

Re: Time management: Rybka vs IvanHoe

Posted: Tue Jun 29, 2010 1:19 pm
by kingliveson

Code: Select all

Ivanhoe v Rybka R003, 4'/40+4'/40+4'/40  2010                   

1   Deep Rybka 4 x64   +11/=30/-9 52.00%   26.0/50
2   IvanHoe 9.57b x64  +9/=30/-11 48.00%   24.0/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 256 MB Hash
† Ponder Off
† Athlon 64 X2 6000+ @ 3100 MHz
† 2 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 20 (default 15)
TimeOrdinaryFactor 230 (default 75)
TimeAbsolutePercent 80 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)

Pawn Hash 16 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off


On System 2

Code: Select all

Rybka vs IvanHoe , 4'/40+4'/40+4'/40                         

1   Deep Rybka 4 SSE42 x64  +13/=26/-11 52.00%   26.0/50
2   IvanHoe 9.57b x64       +11/=26/-13 48.00%   24.0/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 512 MB Hash
† Ponder Off
† Phenom II 940
† 4 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 20 (default 15)
TimeOrdinaryFactor 230 (default 75)
TimeAbsolutePercent 80 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)
Pawn Hash 32 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off

Game data available: http://chess.cygnitec.com/pgn/

Re: Time management: Rybka vs IvanHoe

Posted: Tue Jun 29, 2010 3:17 pm
by Robert Houdart
The standard deviation (sigma) of a 50 game match is about 6% to 7%.
All your results fit within the +-1 sigma range (44%-56%) and are indistinguishable from a random sample of matches between two equally strong engines.

Robert

Re: Time management: Rybka vs IvanHoe

Posted: Tue Jun 29, 2010 6:07 pm
by benstoker
What in the world is the 'BufferTime' setting in IvanHoe?


kingliveson wrote:It has been determined that using specific time management configurations, Rybka 4 performs better at repeating time controls. IvanHoe also has time management configuration that counters Rybka's. I have been testing some settings simultaneously on 2 different hardware, and would say the data is inconclusive right now.

Code: Select all

Ivanhoe v Rybka R002, 4'/40+4'/40+4'/40  2010 

1   Deep Rybka 4 x64   +14/=28/-8 56.00%   28.0/50
2   IvanHoe 9.57b x64  +8/=28/-14 44.00%   22.0/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 256 MB Hash
† Ponder Off
† Athlon 64 X2 6000+ @ 3100 MHz
† 2 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 20 (default 15)
TimeOrdinaryFactor 60 (default 75)
TimeAbsolutePercent 20 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)

Pawn Hash 16 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off


On System 2

Code: Select all

Rybka vs IvanHoe , 4'/40+4'/40+4'/40  

1   IvanHoe 9.57b x64       +12/=28/-10 52.00%   26.0/50
2   Deep Rybka 4 SSE42 x64  +10/=28/-12 48.00%   24.0/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 512 MB Hash
† Ponder Off
† Phenom II 940
† 4 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 20 (default 15)
TimeOrdinaryFactor 60 (default 75)
TimeAbsolutePercent 20 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)
Pawn Hash 32 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off

Game data available: http://chess.cygnitec.com/pgn/

Re: Time management: Rybka vs IvanHoe

Posted: Tue Jun 29, 2010 7:32 pm
by kingliveson
Robert Houdart wrote:The standard deviation (sigma) of a 50 game match is about 6% to 7%.
All your results fit within the +-1 sigma range (44%-56%) and are indistinguishable from a random sample of matches between two equally strong engines.

Robert
Hey Robert, you are right this data does not show much due to lack of games. I will try to find one setting and then run more games. I am considering:

TimeEasyFactor 20 (default 15) or 50
TimeOrdinaryFactor 237 (default 75) or 115
TimeAbsolutePercent 80 (default 25) or 95

benstoker wrote:What in the world is the 'BufferTime' setting in IvanHoe?
Prevents time losses.

edit: 50.115.95 hasn't been tested. But using a formula, 20.100.30 looks promising. 20.237.80 is currently being tested on the slower hardware against Rybka 4. After 40 games, standing is +11/=24/-5 in favor of IvanHoe. The result is similar to one above using 20.240.80.

Re: Time management: Rybka vs IvanHoe

Posted: Wed Jun 30, 2010 2:02 pm
by kingliveson

Code: Select all

Rybka vs IvanHoe , 4'/40+4'/40+4'/40 

1   Deep Rybka 4 SSE42 x64  +9/=33/-8 51.00%   25.5/50
2   IvanHoe 9.57b x64       +8/=33/-9 49.00%   24.5/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 512 MB Hash
† Ponder Off
† Phenom II 940
† 4 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 20 (default 15)
TimeOrdinaryFactor 100 (default 75)
TimeAbsolutePercent 30 (default 25)

Pawn Hash 32 MB (1/16 or 1/8 Hash Recommended, default 4)
BufferTime 2000 (1000 = 1 second, default 0)
AllowInstantMoveFromHash Off


Game data available: http://chess.cygnitec.com/pgn/

Re: Time management: Rybka vs IvanHoe

Posted: Wed Jun 30, 2010 2:42 pm
by kingliveson

Code: Select all

Ivanhoe v Rybka R003, 4'/40+4'/40+4'/40  2010

1  IvanHoe 9.57b x64  +19/=64/-17 51.00%   51.0/100
2  Deep Rybka 4 x64   +17/=64/-19 49.00%   49.0/100
† 50 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 256 MB Hash
† Ponder Off
† Athlon 64 X2 6000+ @ 3100 MHz
† 2 Cores
† 40/4 Repeating Time Control

Rybka
TC 3.100.150 (Recommended, default 3.80.100)

IvanHoe
TimeEasyFactor 20 (default 15)
TimeOrdinaryFactor 237 (default 75)
TimeAbsolutePercent 80 (default 25)
BufferTime 2000 (1000 = 1 second, default 0)

Pawn Hash 16 MB (1/16 or 1/8 Hash Recommended, default 4)
AllowInstantMoveFromHash Off


Game data available: http://chess.cygnitec.com/pgn/

Re: Time management: Rybka vs IvanHoe

Posted: Thu Jul 01, 2010 1:28 pm
by kingliveson

Code: Select all

Rybka vs IvanHoe , Blitz:4'  2010                          

1   IvanHoe 9.57b x64       +15/=29/-6 59.00%   29.5/50
2   Deep Rybka 4 SSE42 x64  +6/=29/-15 41.00%   20.5/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 256 MB Hash
† Ponder Off
† Phenom II 940
† 4 Cores
† 4+0 Time Control

Rybka
TC 2.72.115? (Recommended, default 3.80.100)*
*Can't remember if Rybka's time management used recommended or default.

IvanHoe
TimeEasyFactor 12 (default 15)
TimeOrdinaryFactor 51 (default 75)
TimeAbsolutePercent 20 (default 25)

Pawn Hash 16 MB (1/16 or 1/8 Hash Recommended, default 4)
BufferTime 2000 (1000 = 1 second, default 0)
AllowInstantMoveFromHash Off


On System 2

Code: Select all

Ivanhoe v Rybka R004, Blitz:3'  2010                      

1   IvanHoe 9.57b x64  +19/=28/-3 66.00%   33.0/50
2   Deep Rybka 4 x64   +3/=28/-19 34.00%   17.0/50
† 25 Positions (test suite)
† 3-4-5 Nalimov/Robbobase
† 128 MB Hash
† Ponder Off
† Athlon 64 X2 6000+ @ 3100 MHz
† 2 Cores
† 3+0 Time Control

Rybka
TC 3.100.150 (default 3.80.100)*
*Forget to change to 2.72.115 which is recommend for sudden death matches.

IvanHoe
TimeEasyFactor 12 (default 15)
TimeOrdinaryFactor 51 (default 75)
TimeAbsolutePercent 20 (default 25)
Pawn Hash 16 MB (1/16 or 1/8 Hash Recommended, default 4)
BufferTime 2000 (1000 = 1 second, default 0)
AllowInstantMoveFromHash Off


Game data available: http://chess.cygnitec.com/pgn/

More sudden death matches are now being tested. Settings were verified. :)