WBEC Ridderkerk new results.

[Leo Dijksman]

Geschrieben von:/Posted by: Leo Dijksman at 10 June 2004 17:48:00:

Promo B tourney:
Round 2:
Gothmog 1.0b2 ½110 2.5/4
Phalanx XXII ½001 1.5/4
------------------------------------
Gromit 3.82 1½10 2.5/4
King Of Kings 2.56 0½01 1.5/4
------------------------------------
Gothmog 1.0b2 1101 3.0/4
Movei 0.08.232 0010 1.0/4
------------------------------------

Promo C tourney:
Round 2:
Esc 1.16 011½ 2.5/4
Horizon 4.1b11 100½ 1.5/4
------------------------------------
Round 3:
Terra 3.3b11 ½½10 2.0/4
Ufim 5.00 ½½01 2.0/4
------------------------------------
Queen 2.44 1½11 3.5/4
Esc 1.16 0½00 0.5/4
------------------------------------
Frenzee 146 - Horizon 4.1b11 0-1 = 3 games to go!
------------------------------------

Promo D tourney:
Round 3:
Chispa 4.032 01½1 2.5/4
CyberPagno 2.01 10½0 1.5/4
------------------------------------
ZZZZZZ 6.436 0½1½ 2.0/4
Alarm 0.93.1 1½0½ 2.0/4
------------------------------------
GreKo 2.80 ½110 2.5/4
Hagrid 0.7.56 ½001 1.5/4
------------------------------------

Crosstables and pgn on the WBEC homepage.
Enginelist, latest updates/new engines:
10/06/2004:
Updated: Tytan v3.64
Updated: Yace v0.99.87
Best wishes,
Leo.


WBEC Ridderkerk homepage.

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 10 June 2004 18:10:47:
Als Antwort auf:/In reply to: WBEC Ridderkerk new results. geschrieben von:/posted by: Leo Dijksman at 10 June 2004 17:48:00:

Promo B tourney:
Round 2:
Gothmog 1.0b2 ½110 2.5/4
Phalanx XXII ½001 1.5/4
------------------------------------
Gromit 3.82 1½10 2.5/4
King Of Kings 2.56 0½01 1.5/4
------------------------------------
Gothmog 1.0b2 1101 3.0/4
Movei 0.08.232 0010 1.0/4
------------------------------------

Why Gothmog1.0b2 and not 0.4.9 or 0.4.9beta that is the natural number after 0.4.8?
Is this something that tord started from scratch and not modified 0.4.8?
Uri

[Wael Deeb]

Geschrieben von:/Posted by: Wael Deeb at 10 June 2004 18:24:30:
Als Antwort auf:/In reply to: Re: WBEC Ridderkerk new results. geschrieben von:/posted by: Uri Blass at 10 June 2004 18:10:47:

Promo B tourney:
Round 2:
Gothmog 1.0b2 ½110 2.5/4
Phalanx XXII ½001 1.5/4
------------------------------------
Gromit 3.82 1½10 2.5/4
King Of Kings 2.56 0½01 1.5/4
------------------------------------
Gothmog 1.0b2 1101 3.0/4
Movei 0.08.232 0010 1.0/4
------------------------------------

Why Gothmog1.0b2 and not 0.4.9 or 0.4.9beta that is the natural number after 0.4.8?
Is this something that tord started from scratch and not modified 0.4.8?
Uri

Hi Uri,
Because Tord is rewriting his code and cleaning it up,that's why....
Regards,
Dr.WAEL DEEB

[Fabien Letouzey]

Geschrieben von:/Posted by: Fabien Letouzey at 10 June 2004 18:27:33:
Als Antwort auf:/In reply to: Re: WBEC Ridderkerk new results. geschrieben von:/posted by: Uri Blass at 10 June 2004 18:10:47:

Why Gothmog1.0b2 and not 0.4.9 or 0.4.9beta that is the natural number after 0.4.8?
Is this something that tord started from scratch and not modified 0.4.8?
Uri

My own guess is that he wants to offer a last stable version 1.0 before he can focus on his next engine.
Fabien.

[Roger Brown]

Geschrieben von:/Posted by: Roger Brown at 10 June 2004 18:54:29:
Als Antwort auf:/In reply to: WBEC Ridderkerk new results. geschrieben von:/posted by: Leo Dijksman at 10 June 2004 17:48:00:

Let me save Tord the time and trouble:

Gothmog got lucky
The other engines were unfortunate
Opening Book
Endgame knowledge
Gothmog was less unlucky.....
You do get the picture.

Is it possible that Tord actually knows how to write a competitive chess engine?

My goodness, what a thought!



Later.

[Marcus Prewarski]

Geschrieben von:/Posted by: Marcus Prewarski at 10 June 2004 19:03:42:
Als Antwort auf:/In reply to: Re: WBEC Ridderkerk new results. geschrieben von:/posted by: Roger Brown at 10 June 2004 18:54:29:

Let me save Tord the time and trouble:

Gothmog got lucky
The other engines were unfortunate
Opening Book
Endgame knowledge
Gothmog was less unlucky.....
You do get the picture.

Is it possible that Tord actually knows how to write a competitive chess engine?

My goodness, what a thought!



Later.

As ones engine improves it seems to lose just as many games, but to better engines. So it always seems like the engine is weak. Tord knows he has a good engine, but he is obviously not satisfied with this and perhaps this is what fuels his interest. Everyone else should be happy because he continues to improve whatever he is working on and is willing to share his ideas. Just a thought, I haven't spoken to him about it.
-Marcus

[Roger Brown]

Geschrieben von:/Posted by: Roger Brown at 10 June 2004 19:47:08:
Als Antwort auf:/In reply to: Re: WBEC Ridderkerk new results. geschrieben von:/posted by: Marcus Prewarski at 10 June 2004 19:03:42:

As ones engine improves it seems to lose just as many games, but to better engines. So it always seems like the engine is weak. Tord knows he has a good engine, but he is obviously not satisfied with this and perhaps this is what fuels his interest. Everyone else should be happy because he continues to improve whatever he is working on and is willing to share his ideas. Just a thought, I haven't spoken to him about it.
-Marcus

Hello Marcus,
I can assure you that I am thrilled with Gothmog and Tord's boldness in engine design.
My post was a cheeky jab at Tord's well known modesty with respect to his engine's strength, performance etc.
I totally agree with your post incidentally.
Later.

[Tord Romstad]

Geschrieben von:/Posted by: Tord Romstad at 11 June 2004 00:22:48:
Als Antwort auf:/In reply to: Re: WBEC Ridderkerk new results. geschrieben von:/posted by: Uri Blass at 10 June 2004 18:10:47:

Why Gothmog1.0b2 and not 0.4.9 or 0.4.9beta that is the natural number after 0.4.8?
Is this something that tord started from scratch and not modified 0.4.8?

Fabien is right. I want to release a final 1.0 version before I stop working
on Gothmog. Version 1.0 beta 2 isn't really very different from 0.4.8. There
is some new endgame knowledge and a few new pawn structure evaluation terms,
but not much else. I don't think 1.0 beta 2 is measurably stronger.
My new engine has a very long way to go before it is ready for tournament play.
Tord

[Tord Romstad]

Geschrieben von:/Posted by: Tord Romstad at 11 June 2004 00:30:59:
Als Antwort auf:/In reply to: Re: WBEC Ridderkerk new results. geschrieben von:/posted by: Roger Brown at 10 June 2004 18:54:29:

Let me save Tord the time and trouble:

I'll spend a minimum of time and trouble on this myself anyway:
As we all know, in such a short tournament with six engines of comparable
strength, *anything* is possible. There is no need to try to explain the
results.
Tord

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 11 June 2004 01:40:34:
Als Antwort auf:/In reply to: Re: WBEC Ridderkerk new results. geschrieben von:/posted by: Tord Romstad at 11 June 2004 00:22:48:

Why Gothmog1.0b2 and not 0.4.9 or 0.4.9beta that is the natural number after 0.4.8?
Is this something that tord started from scratch and not modified 0.4.8?

Fabien is right. I want to release a final 1.0 version before I stop working
on Gothmog. Version 1.0 beta 2 isn't really very different from 0.4.8. There
is some new endgame knowledge and a few new pawn structure evaluation terms,
but not much else. I don't think 1.0 beta 2 is measurably stronger.
My new engine has a very long way to go before it is ready for tournament play.

I am curious as to what things are different in your new approach.
Are you still using MTD(f), or have you changed to PVS?
Are you changing the underlying board representation?



my ftp site {remove http:// unless you like error messages}

[Tord Romstad]

Geschrieben von:/Posted by: Tord Romstad at 11 June 2004 16:38:45:
Als Antwort auf:/In reply to: Re: WBEC Ridderkerk new results. geschrieben von:/posted by: Dann Corbit at 11 June 2004 01:40:34:

Hi Dann,

I am curious as to what things are different in your new approach.
Are you still using MTD(f), or have you changed to PVS?
Are you changing the underlying board representation?

There are still many things I haven't yet made up my mind about. I will try
to keep everything as fluid as possible in the early phases of development, in
order to facilitate experimenting with many different ideas.
I hope to end up with something similar to Gothmog, but smaller, simpler, faster
and less buggy.
This is one of the many things which are not clear yet. The easiest option would
be to go for MTD(f) again, but there are also good reasons to try PVS. I will
probably write basic code for both algorithms before I go ahead and perfect one
of them.
Not much. I will still use mailbox on a 16-file board, but I consider
adding some light-weight non-rotated bitboards for use as a piece list
replacement and in the eval.
Tord

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 11 June 2004 16:59:26:
Als Antwort auf:/In reply to: Re: WBEC Ridderkerk new results. geschrieben von:/posted by: Tord Romstad at 11 June 2004 16:38:45:

Hi Dann,

I am curious as to what things are different in your new approach.
Are you still using MTD(f), or have you changed to PVS?

There are still many things I haven't yet made up my mind about. I will try
to keep everything as fluid as possible in the early phases of development, in
order to facilitate experimenting with many different ideas.
I hope to end up with something similar to Gothmog, but smaller, simpler, faster
and less buggy.
This is one of the many things which are not clear yet. The easiest option would
be to go for MTD(f) again, but there are also good reasons to try PVS.

for me the easiest option is pvs(it is the only option I know).
I read that you have problems of instability with MTD if you try to use the difference between the evaluation and your bounds for decision about pruning(I do not say alpha beta because as far as I understand you have only gamma).
I have not these problems and there are a lot of ideas to try with PVS.
I think to rewrite my search but the main reason is not MTD and I decided to delay it and I will not do it before WCCC.
There are a lot of ideas to try to improve the search in movei even without rewriting the search and I plan to try some of them before WCCC and also to try to improve my evaluation.
Uri

[Tord Romstad]

Geschrieben von:/Posted by: Tord Romstad at 11 June 2004 17:43:45:
Als Antwort auf:/In reply to: Re: WBEC Ridderkerk new results. geschrieben von:/posted by: Uri Blass at 11 June 2004 16:59:26:

Hi Dann,

I am curious as to what things are different in your new approach.
Are you still using MTD(f), or have you changed to PVS?

There are still many things I haven't yet made up my mind about. I will try
to keep everything as fluid as possible in the early phases of development, in
order to facilitate experimenting with many different ideas.
I hope to end up with something similar to Gothmog, but smaller, simpler, faster
and less buggy.
This is one of the many things which are not clear yet. The easiest option would
be to go for MTD(f) again, but there are also good reasons to try PVS.

for me the easiest option is pvs(it is the only option I know).
I read that you have problems of instability with MTD if you try to use the difference between the evaluation and your bounds for decision about pruning
(I do not say alpha beta because as far as I understand you have only gamma).
I have not these problems and there are a lot of ideas to try with PVS.

Yes, but in cases where this is a problem, there are often other bounds to use.
The upper and lower bounds for the root score are often useful. I also use
a sort of "aspiration window" around the root score from the previous iteration.
At the beginning of every new iteration, I assume that the score will not change
by more than +- 3 pawns compared to the score from the last iteration, and use
this assumption for pruning decisions. If it turns out that this assumption
is likely to be wrong, I clear the hash table and switch off all selectivity
until the current iteration is finished. This can be rather expensive,
but in practise it isn't a big problem because it usually only occurs when
the game is won or lost.
Thanks to Andrew Williams, from whom I learned this technique.
Yes, in MTD you search with only a single bound, which is nothing more than
a guess of the root score.
There are. All algorithms have their unique advantages, problems and tricks.
This is why I it is probably a good idea to study several algorithms in
depth.
Tord