IvanHoe 0B.01.08

Discussion about chess-playing software (engines, hosts, opening books, platforms, etc...)
Damir Desevac
Posts: 57
Joined: Thu Jun 10, 2010 2:58 pm
Real Name: Damir Desevac

Re: IvanHoe 0B.01.08

Post by Damir Desevac » Sun Jan 09, 2011 9:38 pm

:) I have managed to measure the speed increase between Ivanhoe and Houdini. Houdini has a speed advantage of almost 2000 kN/s. If you can make Ivanhoe as fast as Houdini + with its new time management, I am sure it would be able to surpass Houdini. ;)

User avatar
robbolito
Posts: 601
Joined: Thu Jun 10, 2010 3:48 am

Re: IvanHoe 0B.01.08

Post by robbolito » Sun Jan 09, 2011 9:38 pm

Intel(R) Core(TM) i7 CPU 870 @ 2.93GHz 8.0 GB Memory
Windows 7 Home Premium Edition (Build 7600) 64 bit
Fritz Benchmark:
Speed: 13.09
KNS: 6280
GUI: Rybka 3
Book: Megabook 2011-10 moves
Hash:256
TB and GTB: ON
Ponder: OFF
Blitz: 5.0

Code: Select all

1   Houdini 1.5 x64       3190  +7/-5/=8 55.00%   11.0/20
2   IvanHoe 0B.01.09 x64  3190  +5/-7/=8 45.00%    9.0/20

User avatar
kingliveson
Posts: 1388
Joined: Thu Jun 10, 2010 1:22 am
Real Name: Franklin Titus
Location: 28°32'1"N 81°22'33"W

Re: IvanHoe 0B.01.08

Post by kingliveson » Sun Jan 09, 2011 10:59 pm

Damir Desevac wrote::) I have managed to measure the speed increase between Ivanhoe and Houdini. Houdini has a speed advantage of almost 2000 kN/s. If you can make Ivanhoe as fast as Houdini + with its new time management, I am sure it would be able to surpass Houdini. ;)
Not quite. There are several reasons Houdini is much stronger; better eval, search (pruning), and SMP implementation -- which also contribute to the speed difference. All I did was changed 2 lines of code and that won't be enough...

At very fast time control, I expect this build to be trashed like Rybka and other IvanHoe builds, so am really only looking at results for 10+0 and greater.
PAWN : Knight >> Bishop >> Rook >>Queen

User avatar
Robert Houdart
Posts: 180
Joined: Thu Jun 10, 2010 4:55 pm
Contact:

Re: IvanHoe 0B.01.08

Post by Robert Houdart » Sun Jan 09, 2011 11:08 pm

Damir Desevac wrote::) I have managed to measure the speed increase between Ivanhoe and Houdini. Houdini has a speed advantage of almost 2000 kN/s. If you can make Ivanhoe as fast as Houdini + with its new time management, I am sure it would be able to surpass Houdini. ;)
It's not very useful to compare node speeds between different engines, each engine has its own metric of "node speed".

User avatar
robbolito
Posts: 601
Joined: Thu Jun 10, 2010 3:48 am

Re: IvanHoe 0B.01.08

Post by robbolito » Mon Jan 10, 2011 10:58 am

Intel(R) Core(TM) i7 CPU 870 @ 2.93GHz 8.0 GB Memory
Windows 7 Home Premium Edition (Build 7600) 64 bit
Fritz Benchmark:
Speed: 13.09
KNS: 6280
GUI: Rybka 3
Book: Powerbook 2011-10 moves
Hash:256
RB and GTB: ON
Ponder: OFF
Blitz: 15.0


Code: Select all

1   Houdini 1.5 x64       3190  +10/-3/=17 61.67%   18.5/30
2   IvanHoe 0B.01.09 x64  3190  +3/-10/=17 38.33%   11.5/30


User avatar
kingliveson
Posts: 1388
Joined: Thu Jun 10, 2010 1:22 am
Real Name: Franklin Titus
Location: 28°32'1"N 81°22'33"W

Re: IvanHoe 0B.01.08

Post by kingliveson » Mon Jan 10, 2011 1:29 pm

robbolito wrote:Intel(R) Core(TM) i7 CPU 870 @ 2.93GHz 8.0 GB Memory
Windows 7 Home Premium Edition (Build 7600) 64 bit
Fritz Benchmark:
Speed: 13.09
KNS: 6280
GUI: Rybka 3
Book: Powerbook 2011-10 moves
Hash:256
RB and GTB: ON
Ponder: OFF
Blitz: 15.0


Code: Select all

1   Houdini 1.5 x64       3190  +10/-3/=17 61.67%   18.5/30
2   IvanHoe 0B.01.09 x64  3190  +3/-10/=17 38.33%   11.5/30

Thanks for these -- I am getting similar results at 15+10 with current standing, +10/=10/-3. After the first few games, Houdini's chess engine pattern learning (CEPL) feature engaged full throttle.
PAWN : Knight >> Bishop >> Rook >>Queen

ernest
Posts: 247
Joined: Thu Sep 02, 2010 10:33 am

Re: IvanHoe 0B.01.08

Post by ernest » Mon Jan 10, 2011 6:29 pm

kingliveson wrote:Houdini's chess engine pattern learning (CEPL) feature
Hi,
Never heard about that... :shock:
Do you have more details, or links?

User avatar
kingliveson
Posts: 1388
Joined: Thu Jun 10, 2010 1:22 am
Real Name: Franklin Titus
Location: 28°32'1"N 81°22'33"W

Re: IvanHoe 0B.01.08

Post by kingliveson » Fri Jan 14, 2011 12:06 am

ernest wrote:
kingliveson wrote:Houdini's chess engine pattern learning (CEPL) feature
Hi,
Never heard about that... :shock:
Do you have more details, or links?

You'll have to ask Robert, though am not sure he's willing to share... :)
PAWN : Knight >> Bishop >> Rook >>Queen

ernest
Posts: 247
Joined: Thu Sep 02, 2010 10:33 am

Re: IvanHoe 0B.01.08

Post by ernest » Fri Jan 14, 2011 12:51 am

kingliveson wrote:You'll have to ask Robert
OK, I'll phrase that differently:
where did Robert mention that? (or was it privately with you? ;) )

UncombedCoconut
Posts: 44
Joined: Thu Jun 10, 2010 1:43 am
Real Name: Justin Blanchard
Location: United States

Re: IvanHoe 0B.01.08

Post by UncombedCoconut » Fri Jan 14, 2011 10:04 am

Perhaps related to this joke?
Laskos wrote:
overtond wrote:Hi Robert,

Without going in to technical feedback {that I for one would not understand} have you written the code so that H1.5 makes use of previous analysis? I ask as I have read in many places that when H1.5 is used to play against other engines it is very strong in terms of results obtained, however, when a position is offered to H1.5 such that no previous information was available, the difference between H1.5 and other top engines {for example, R4} is less significant.

Is this something you are aware of?

ATB
David
Yes, Houdini 1.5 has pattern recognition and learning from previous versions games against all the engines. It uses adaptive Monte Carlo with multidimensional parameter adapter specific to each engine it's playing. If you give Houdini 1.5 a new position to analyze, its level is much lower than Rybka 4, which is known to have the best analysis of the new positions. Sorry, new positions are not the realm of non-Rybka 4 engines.

Kai

Post Reply