Nunn Top 2 finished

[Heinz van Kempen]

Geschrieben von:/Posted by: Heinz van Kempen at 18 June 2004 15:42:20:

Hi ,
after 7600 games Nunn Top 2 is finished.
Here are the final standings after all twenty positions:
Conditions:
Athlon 2600+
64 MB Hash
5 men EGTB
all Nunn2 positions
Time control: 4 min. + 2 sec.


1 Aristarch 4.50              515.5 / 760 
2 Crafty 19.12 SE Capa        495.0 / 760 
3 Thinker 4.6b                482.5 / 760 
4 Crafty 19.13 DC             472.0 / 760 
5 Patriot 0.172-Light         469.0 / 760 
6 Little Goliath 2000 v3.9    460.5 / 760 
7 Slow Chess 2.93             397.5 / 760 
8 Gothmog 0.4.8f              383.0 / 760 
9 King of Kings 2.56          383.0 / 760 
10 LambChop 10.99             383.0 / 760 
11 Jonny 2.63b                367.0 / 760 
12 Movei 00_8_198             346.5 / 760 
13 PostModernist 1010a        342.0 / 760 
14 Fruit X 05/09              336.5 / 760 
15 Amy 0.8.7b                 327.5 / 760 
16 Naum 1.0                   321.0 / 760 
17 Terra 3.3B11               300.5 / 760 
18 The Baron 1.3.1b2          300.5 / 760 
19 Frenzee 157                294.5 / 760 
20 Booot 3.3                  223.0 / 760 

So Aristarch won not unexpectedly. For Crafty the Capablanca settings from Mike Byrne´s SE are pretty strong. I had also good results with Fischer and Morphy settings before. Others I did not test.
Little Goliath 2000 v.3.9 stronger than Nemesis version what is no real news. Naum proved to be a strong newcomer and Booot did well in this strong field.
Fruit much improved already with this beta, but this was topped by far in the meantime by Fruit 1.5. See previous postings.
Most improved other versions:
PostModernist 1008    2501  
PostModernist 1010a   2559  +58
Terra 3.3B5           2476
Terra 3.3B11          2525  +49
LambChop 10.88        2560    
LambChop 10.99        2598  +38
Jonny 2.62            2553
Jonny 2.63b           2586  +33
SlowChess 2.89b       2584  
SlowChess 2.93        2614  +30
King of Kings 2.52    2573
King of Kings 2.56    2595  +22
Kaissa 1.7            2653
Patriot 0,172light    2673  +20
Thinker 4.5e          2673
Thinker 4.6b          2689  +16
Movei 00_8_178        2551
Movei 00_8_198        2567  +16
Aristarch, Amy and Booot also improved by a few points.


Best Regards
Heinz
http://www.husvankempen.de/nunn/

[Peter Fendrich]

Geschrieben von:/Posted by: Peter Fendrich at 18 June 2004 15:51:34:
Als Antwort auf:/In reply to: Nunn Top 2 finished geschrieben von:/posted by: Heinz van Kempen at 18 June 2004 15:42:20:

Hi,
thank you for this tournament!
I was astonsihed by tha fact that Terra's results jumped up and down that much for different Nunn positions. I haven't studied the other programs to see if this is normal though. This version of Terra is very speculative in its play (not so much in tactics as positionally). When time comes, I have to study the games in order to reduce the worst stupidness of this approach.
/Peter

Hi ,
after 7600 games Nunn Top 2 is finished.
Here are the final standings after all twenty positions:
Conditions:
Athlon 2600+
64 MB Hash
5 men EGTB
all Nunn2 positions
Time control: 4 min. + 2 sec.



Best Regards
Heinz
http://www.husvankempen.de/nunn/

>1 Aristarch 4.50              515.5 / 760 
>2 Crafty 19.12 SE Capa        495.0 / 760 
>3 Thinker 4.6b                482.5 / 760 
>4 Crafty 19.13 DC             472.0 / 760 
>5 Patriot 0.172-Light         469.0 / 760 
>6 Little Goliath 2000 v3.9    460.5 / 760 
>7 Slow Chess 2.93             397.5 / 760 
>8 Gothmog 0.4.8f              383.0 / 760 
>9 King of Kings 2.56          383.0 / 760 
>10 LambChop 10.99             383.0 / 760 
>11 Jonny 2.63b                367.0 / 760 
>12 Movei 00_8_198             346.5 / 760 
>13 PostModernist 1010a        342.0 / 760 
>14 Fruit X 05/09              336.5 / 760 
>15 Amy 0.8.7b                 327.5 / 760 
>16 Naum 1.0                   321.0 / 760 
>17 Terra 3.3B11               300.5 / 760 
>18 The Baron 1.3.1b2          300.5 / 760 
>19 Frenzee 157                294.5 / 760 
>20 Booot 3.3                  223.0 / 760 
>
>So Aristarch won not unexpectedly. For Crafty the Capablanca settings from Mike Byrne´s SE are pretty strong. I had also good results with Fischer and Morphy settings before. Others I did not test.
>Little Goliath 2000 v.3.9 stronger than Nemesis version what is no real news. Naum proved to be a strong newcomer and Booot did well in this strong field.
>Fruit much improved already with this beta, but this was topped by far in the meantime by Fruit 1.5. See previous postings.
>Most improved other versions:
>PostModernist 1008    2501  
>PostModernist 1010a   2559  +58
>Terra 3.3B5           2476
>Terra 3.3B11          2525  +49
>LambChop 10.88        2560    
>LambChop 10.99        2598  +38
>Jonny 2.62            2553
>Jonny 2.63b           2586  +33
>SlowChess 2.89b       2584  
>SlowChess 2.93        2614  +30
>King of Kings 2.52    2573
>King of Kings 2.56    2595  +22
>Kaissa 1.7            2653
>Patriot 0,172light    2673  +20
>Thinker 4.5e          2673
>Thinker 4.6b          2689  +16
>Movei 00_8_178        2551
>Movei 00_8_198        2567  +16
>Aristarch, Amy and Booot also improved by a few points.
>



Terra

[Heinz van Kempen]

Geschrieben von:/Posted by: Heinz van Kempen at 18 June 2004 16:14:47:
Als Antwort auf:/In reply to: Re: Nunn Top 2 finished geschrieben von:/posted by: Peter Fendrich at 18 June 2004 15:51:34:

Hi,
thank you for this tournament!
I was astonsihed by tha fact that Terra's results jumped up and down that much for different Nunn positions. I haven't studied the other programs to see if this is normal though.
/Peter

Hello Peter,
yes, this is typical for such different positions for all engines. Not that much that Aristarch could finish last with any position, but for most engines in this group positions will vary a lot, sometimes top, sometimes catastrophic. Anyway all this positions are well known in tournament practice.

This version of Terra is very speculative in its play (not so much in tactics as positionally). When time comes, I have to study the games in order to reduce the worst stupidness of this approach.
Best Regards
Heinz

[Tom Likens]

Geschrieben von:/Posted by: Tom Likens at 18 June 2004 17:45:12:
Als Antwort auf:/In reply to: Re: Nunn Top 2 finished geschrieben von:/posted by: Heinz van Kempen at 18 June 2004 16:14:47:

Hi,
thank you for this tournament!
I was astonsihed by tha fact that Terra's results jumped up and down that much for different Nunn positions. I haven't studied the other programs to see if this is normal though.
/Peter

Hello Peter,
yes, this is typical for such different positions for all engines. Not that much that Aristarch could finish last with any position, but for most engines in this group positions will vary a lot, sometimes top, sometimes catastrophic. Anyway all this positions are well known in tournament practice.

This version of Terra is very speculative in its play (not so much in tactics as positionally). When time comes, I have to study the games in order to reduce the worst stupidness of this approach.
Best Regards
Heinz

Hello Heinz,
One thing I've always wondered about, since many programmers (myself
included) use the Nunn positions to test program improvements, are we
inadvertently making our engines myopic (i.e. blind). By tuning against
a limited set of starting positions (and 20 *is* limited) it seems we are
falling into the same trap as using WAC to test the tactical strength of
an engine.
Of course, this has no bearing on your tournaments, which I enjoy immensely,
but I am curious what others think.
regards,
--tom

[Heinz van Kempen]

Geschrieben von:/Posted by: Heinz van Kempen at 18 June 2004 17:56:40:
Als Antwort auf:/In reply to: Re: Nunn Top 2 finished geschrieben von:/posted by: Tom Likens at 18 June 2004 17:45:12:

Hello Heinz,
One thing I've always wondered about, since many programmers (myself
included) use the Nunn positions to test program improvements, are we
inadvertently making our engines myopic (i.e. blind). By tuning against
a limited set of starting positions (and 20 *is* limited) it seems we are
falling into the same trap as using WAC to test the tactical strength of
an engine.
Of course, this has no bearing on your tournaments, which I enjoy immensely,
but I am curious what others think.
regards,
--tom

Hello Tom,
yes, I was often suspicious concerning this possible tuning. Up to now I could not detect that engines are tuned for the Nunn positions, but this will not signify that they are not. As soon as this tuning becomes obvious it will be time to do something else (tests with own books etc.).
I would also like to have other opinions here, as this is important.
Best Regards
Heinz

[Peter Fendrich]

Geschrieben von:/Posted by: Peter Fendrich at 18 June 2004 19:01:11:
Als Antwort auf:/In reply to: Re: Nunn Top 2 finished geschrieben von:/posted by: Tom Likens at 18 June 2004 17:45:12:

Hi,
thank you for this tournament!
I was astonsihed by tha fact that Terra's results jumped up and down that much for different Nunn positions. I haven't studied the other programs to see if this is normal though.
/Peter

Hello Peter,
yes, this is typical for such different positions for all engines. Not that much that Aristarch could finish last with any position, but for most engines in this group positions will vary a lot, sometimes top, sometimes catastrophic. Anyway all this positions are well known in tournament practice.

This version of Terra is very speculative in its play (not so much in tactics as positionally). When time comes, I have to study the games in order to reduce the worst stupidness of this approach.
Best Regards
Heinz

Hello Heinz,
One thing I've always wondered about, since many programmers (myself
included) use the Nunn positions to test program improvements, are we
inadvertently making our engines myopic (i.e. blind). By tuning against
a limited set of starting positions (and 20 *is* limited) it seems we are
falling into the same trap as using WAC to test the tactical strength of
an engine.
Of course, this has no bearing on your tournaments, which I enjoy immensely,
but I am curious what others think.
regards,
--tom

I never used the Nunn positions for testing but I'm sure that I could make my program play specialized for the Nunn set and earn a few points. If the program can be better in all or most of the Nunn set it will probably be better allround as well. It's not like adjusting it to the WAC set where the solution is known.
/Peter



Terra

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 18 June 2004 20:46:58:
Als Antwort auf:/In reply to: Re: Nunn Top 2 finished geschrieben von:/posted by: Peter Fendrich at 18 June 2004 19:01:11:

Hi,
thank you for this tournament!
I was astonsihed by tha fact that Terra's results jumped up and down that much for different Nunn positions. I haven't studied the other programs to see if this is normal though.
/Peter

Hello Peter,
yes, this is typical for such different positions for all engines. Not that much that Aristarch could finish last with any position, but for most engines in this group positions will vary a lot, sometimes top, sometimes catastrophic. Anyway all this positions are well known in tournament practice.

This version of Terra is very speculative in its play (not so much in tactics as positionally). When time comes, I have to study the games in order to reduce the worst stupidness of this approach.
Best Regards
Heinz

Hello Heinz,
One thing I've always wondered about, since many programmers (myself
included) use the Nunn positions to test program improvements, are we
inadvertently making our engines myopic (i.e. blind). By tuning against
a limited set of starting positions (and 20 *is* limited) it seems we are
falling into the same trap as using WAC to test the tactical strength of
an engine.
Of course, this has no bearing on your tournaments, which I enjoy immensely,
but I am curious what others think.
regards,
--tom

I never used the Nunn positions for testing but I'm sure that I could make my program play specialized for the Nunn set and earn a few points. If the program can be better in all or most of the Nunn set it will probably be better allround as well. It's not like adjusting it to the WAC set where the solution is known.
/Peter

If the target is to tune the program for the nunn position then it is possible
to find personality that will score better for every position and to tell
the program to use personality i in position i.
Note that I never tuned movei for the nunn positions in that way and I believe
no programmer did it but it is possible to tune programs for the nunn positions in that way so the program earn 50 elo in the nunn position and earns nothing in other positions that it does not know which personality to choose.
I have book based of positions and moves and one of my ideas that is not implemented is that the book can include not only positions and moves but also personality that the program will use in case of specific opening choice.
I wonder if part of the commercial programs use that idea for the ssdf games.
That idea may do the job of the person that write the opening book more important because he can write not only moves into the book.
Uri

[Tom Likens]

Geschrieben von:/Posted by: Tom Likens at 18 June 2004 21:17:00:
Als Antwort auf:/In reply to: Re: Nunn Top 2 finished geschrieben von:/posted by: Uri Blass at 18 June 2004 20:46:58:

Hi,
thank you for this tournament!
I was astonsihed by tha fact that Terra's results jumped up and down that much for different Nunn positions. I haven't studied the other programs to see if this is normal though.
/Peter

Hello Peter,
yes, this is typical for such different positions for all engines. Not that much that Aristarch could finish last with any position, but for most engines in this group positions will vary a lot, sometimes top, sometimes catastrophic. Anyway all this positions are well known in tournament practice.

This version of Terra is very speculative in its play (not so much in tactics as positionally). When time comes, I have to study the games in order to reduce the worst stupidness of this approach.
Best Regards
Heinz

Hello Heinz,
One thing I've always wondered about, since many programmers (myself
included) use the Nunn positions to test program improvements, are we
inadvertently making our engines myopic (i.e. blind). By tuning against
a limited set of starting positions (and 20 *is* limited) it seems we are
falling into the same trap as using WAC to test the tactical strength of
an engine.
Of course, this has no bearing on your tournaments, which I enjoy immensely,
but I am curious what others think.
regards,
--tom

I never used the Nunn positions for testing but I'm sure that I could make my program play specialized for the Nunn set and earn a few points. If the program can be better in all or most of the Nunn set it will probably be better allround as well. It's not like adjusting it to the WAC set where the solution is known.
/Peter

If the target is to tune the program for the nunn position then it is possible
to find personality that will score better for every position and to tell
the program to use personality i in position i.
Note that I never tuned movei for the nunn positions in that way and I believe
no programmer did it but it is possible to tune programs for the nunn positions in that way so the program earn 50 elo in the nunn position and earns nothing in other positions that it does not know which personality to choose.
I have book based of positions and moves and one of my ideas that is not implemented is that the book can include not only positions and moves but also personality that the program will use in case of specific opening choice.
I wonder if part of the commercial programs use that idea for the ssdf games.
That idea may do the job of the person that write the opening book more important because he can write not only moves into the book.
Uri

Interestingly, this may be a legitimate way to program an engine. Not to push up your SSDF
or any other rating, but to change the program's style based on the board position. To make
it more general, rather than simply recognizing one of the Nunn positions (or any other *specific*
position for that matter) you could program the engine to enable certain personalities based
on specific themes or general board considerations.
For example, if the position was open and the engine deemed it had an advantage it could
enable its "Tal" personality. Or if it anticipated a combination was in the air it could fire up
its "Alekhine" or "Kasparov" persona. For quiet positions, it could "become" Karpov or
Petrosian. Obviously, these would just be shifts in the weights certain features (such as king
saftey) would receive, but it could make things interesting. I also think this would make the
playing style more attractive.
--tom