Geschrieben von: / Posted by: Arturo Ochoa at 12 January 2004 19:14:26:
Terra Book Tournament 2004-01
Dual Xeon HP 2.4Ghz - Ram: 1Gb.
Hash tables: 128M.
Ponder On.
All 3,4 and 5 EGTBs.
Moves per session: 40/20'
Gui: Winboard 4.2.7.
Cycles: 50
Participants:
Engine Author (Country)
------------------------------------------------------------
Terra 3.2 AO2003Book Peter Fendrich (Sweden)
Terra 3.2 DRBook Peter Fendrich (Sweden)
Terra 3.2 MAIN2Book Peter Fendrich (Sweden)
Crosstable:
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Engine Score 1 2 3
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
1. Terra 3.2 MAIN2Book 55,5/100 ################################################## 110=11110=1===0111=10=00==0=10==10==10=0======1=11 1==000111=111=110000=111=11==00000=1==101110001==1
2. Terra 3.2 DRBook 48,5/100 001=00001=0===1000=01=11==1=01==01==01=1======0=00 ################################################## 0===100==1====100010=111=101=01=0010=10=11=0=1=11=
3. Terra 3.2 AO2003Book 46,0/100 0==111000=000=001111=000=00==11111=0==010001110==0 1===011==0====011101=000=010=10=1101=01=00=1=0=00= ##################################################
Results:
ROUND 1
Terra 3.2 DRBook - Terra 3.2 MAIN2Book 22,0:28,0
ROUND 2
Terra 3.2 AO2003Book - Terra 3.2 DRBook 23,5:26,5
ROUND 3
Terra 3.2 MAIN2Book - Terra 3.2 AO2003Book 27,5:22,5
CONCLUSIONS:
-----------
I cannnot establish a definitive conclusion about the use of random books. Again, the test was run with 3 different books. The mine was a random book with some nags at the beginning. Of course, I am not pretty satisfied with the result but the test doesnt show a clear evidence that RANDOM books makes any difference in the strength of the program (Probably, this conclusion can be as hasty as other previous conclusions written in this Forum). Of course, I would have to create a tuned book for Terra but it is not my mission. By the way, the test was a very interesting task. It is evident that tuned books is a time consuming task and I am already dedicated for a specific engine.
Regards, Arturo Ochoa M.
PD: My email was also posted for those winboard-mates who could be interested in the games.
Winboard Forum
Discussions (not only) about Xboard/Winboard, Computer Chess and Chess Programming
Terra Book Tournament 2004-01: Final Results.
2 posts
• Page 1 of 1
Re: Terra Book Tournament 2004-01: Final Results.
by Peter Fendrich » 13 Jan 2004, 00:57
Geschrieben von: / Posted by: Peter Fendrich at 13 January 2004 00:57:17:
Als Antwort auf: / In reply to: Terra Book Tournament 2004-01: Final Results. geschrieben von: / posted by: Arturo Ochoa at 12 January 2004 19:14:26:
Thank you Arturo for doing these tests!
I'm a bit puzzled however. Is AO2003Book the one you sent me just before christmas or is it the big one even before that?
When I study your books they are very solid and the only problem is that Terra often lacks
the ability to handle a solid position and quickly ruins it instantly after leaving the book.
It's like learning an ape to dance Salsa
The main2.book is a book generated from GM games with a few nags in the
beginning. In addition to that it has been adjusted with the learning data
from numerous computer games to make Terra avoiding lines ending with bad
scores like a pawn down. These bad scores are not because of a bad position
but more often because of a complete lack of understanding of the position.
Also the smaller dr.book is adjusted with learned data in the same way.
I believe that adding learning data in the same manner into the AO2003 book
would make it more adapted to Terra and score better.
If it's ok with you I would like to do exactly that and publish it for download.
Yes you're right, a book has to be tuned in order to really work together with
the engine and that's really a hard work. You know all about it of course!
Best Regards,
Peter
Terra
Als Antwort auf: / In reply to: Terra Book Tournament 2004-01: Final Results. geschrieben von: / posted by: Arturo Ochoa at 12 January 2004 19:14:26:
Thank you Arturo for doing these tests!
I'm a bit puzzled however. Is AO2003Book the one you sent me just before christmas or is it the big one even before that?
When I study your books they are very solid and the only problem is that Terra often lacks
the ability to handle a solid position and quickly ruins it instantly after leaving the book.
It's like learning an ape to dance Salsa
The main2.book is a book generated from GM games with a few nags in the
beginning. In addition to that it has been adjusted with the learning data
from numerous computer games to make Terra avoiding lines ending with bad
scores like a pawn down. These bad scores are not because of a bad position
but more often because of a complete lack of understanding of the position.
Also the smaller dr.book is adjusted with learned data in the same way.
I believe that adding learning data in the same manner into the AO2003 book
would make it more adapted to Terra and score better.
If it's ok with you I would like to do exactly that and publish it for download.
Yes you're right, a book has to be tuned in order to really work together with
the engine and that's really a hard work. You know all about it of course!
Best Regards,
Peter
Terra Book Tournament 2004-01
Dual Xeon HP 2.4Ghz - Ram: 1Gb.
Hash tables: 128M.
Ponder On.
All 3,4 and 5 EGTBs.
Moves per session: 40/20'
Gui: Winboard 4.2.7.
Cycles: 50
Participants:
Engine Author (Country)
------------------------------------------------------------
Terra 3.2 AO2003Book Peter Fendrich (Sweden)
Terra 3.2 DRBook Peter Fendrich (Sweden)
Terra 3.2 MAIN2Book Peter Fendrich (Sweden)
Crosstable:
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Engine Score 1 2 3
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
1. Terra 3.2 MAIN2Book 55,5/100 ################################################## 110=11110=1===0111=10=00==0=10==10==10=0======1=11 1==000111=111=110000=111=11==00000=1==101110001==1
2. Terra 3.2 DRBook 48,5/100 001=00001=0===1000=01=11==1=01==01==01=1======0=00 ################################################## 0===100==1====100010=111=101=01=0010=10=11=0=1=11=
3. Terra 3.2 AO2003Book 46,0/100 0==111000=000=001111=000=00==11111=0==010001110==0 1===011==0====011101=000=010=10=1101=01=00=1=0=00= ##################################################
Results:
ROUND 1
Terra 3.2 DRBook - Terra 3.2 MAIN2Book 22,0:28,0
ROUND 2
Terra 3.2 AO2003Book - Terra 3.2 DRBook 23,5:26,5
ROUND 3
Terra 3.2 MAIN2Book - Terra 3.2 AO2003Book 27,5:22,5
CONCLUSIONS:
-----------
I cannnot establish a definitive conclusion about the use of random books. Again, the test was run with 3 different books. The mine was a random book with some nags at the beginning. Of course, I am not pretty satisfied with the result but the test doesnt show a clear evidence that RANDOM books makes any difference in the strength of the program (Probably, this conclusion can be as hasty as other previous conclusions written in this Forum). Of course, I would have to create a tuned book for Terra but it is not my mission. By the way, the test was a very interesting task. It is evident that tuned books is a time consuming task and I am already dedicated for a specific engine.
Regards, Arturo Ochoa M.
PD: My email was also posted for those winboard-mates who could be interested in the games.
Terra
- Peter Fendrich
2 posts
• Page 1 of 1
Return to Archive (Old Parsimony Forum)
Who is online
Users browsing this forum: No registered users and 27 guests