Winboard Forum

Hi all ,

with new activities like AEGT and CEGT I have finally decided to declare Nunn Blitz tournaments as concluded. Thanks to all authors who supported me with participation, betas and advice.

For game collectors the database with 79 000 Nunn 2 Suite games is available for download (unstripped from all comments still more than 23 MB).

http://www.husvankempen.de/nunn/downloads.htm

The games were mostly played in 2004 on Athlon?s 2600+ and 4+2. After all what I checked to games with longer time control like AEGT, results are suprisingly similar, although the quality even on fast hardware might be inferior. I know this will not stop testers to claim that Blitz games are nonsense and at the same time offer games on old Pentium instead wanting to tell us that this are games with long time controls with high quality.

Robert Allgeuer made great statistical analyses for this database that is also available as Excel sheet. Thanks, Robert and good luck with YABRL, that in my opinion is underestimated.

http://www.husvankempen.de/nunn/downloa ... alyses.xls

Best Regards
Heinz

Thanks Heinz.

My experience is also that results from Blitz games are surprisingly representative, as long as losses on time do not dominate the results (which is definitely the case when choosing X+2 type of time controls). With a sufficient number of games there is also never a "shot of luck" by an engine, even when some people claim - without any foundation - that Blitz is nothing but gambling....

Generally I believe that it is more important to have a sufficient number of games for a rating list and thus small error bars, rather than this or that time control.

An approach I find quite useful is to compare individual match results of engines in Blitz rating lists with those at longer time contols posted in this or other forums. This allows to get a feeling whether a specific engine is balanced, gets weaker or stronger at longer time controls. E.g. I know that ratings of Delfi or KnightDreamer will somewhat go down at longer time controls compared to its position in YABRL, while e.g. Aristarch, Gothmog and Gandalf would go up. However, we are talking some 30-40 points absolute max and only in extreme, rare cases. For the vast majority of engines Blitz ratings are probably quite identical to the ones at longer time controls.

True, I have also been considering for some time now already to stop YABRL or to continue in private due to the apparent lack of interest in Blitz results.

Good luck and success with AEGT, I am sure it will be something like a "reference".

Robert

Hi Heinz and Robert,

First of all, thanks to Heinz for Nunn for playing the Nunn blitz tournaments. They were great fun to follow, and I will miss them in the future. But of course, AEGT is not a bad compensation.

I would like to second Heinz' statement that YABRL is underestimated. Your list is definitely among the most interesting out there, and deserves to get more attention.

Robert, I agree with most of what you write. One small comment below:

Robert Allgeuer wrote:E.g. I know that ratings of Delfi or KnightDreamer will somewhat go down at longer time controls compared to its position in YABRL, while e.g. Aristarch, Gothmog and Gandalf would go up.

I don't know about Delfi, KnightDreamer, Aristarch and Gandalf, but in the case of Gothmog I think you are wrong. Your rating for Gothmog looks remarkably accurate, and I am pretty sure it would remain almost unchanged at longer time controls.

Tord

I don't know about Delfi, KnightDreamer, Aristarch and Gandalf, but in the case of Gothmog I think you are wrong. Your rating for Gothmog looks remarkably accurate, and I am pretty sure it would remain almost unchanged at longer time controls.

That is based on tests I have done (with 0.4.8 at the time, if I remember right) where I compared results between 5+2 and 15+6 (time control exactly 3 times as long, and ~1+ ply more) on identical hardware (Athlon 2400+) and Gothmog was amongst the ones (if not the one) who gained the most.

My guess is that Gothmog has a very good branching factor and this should help at longer time controls.

I intend to do the same sort of stuff during my next absence (i.e. Easter).

Robert

Heinz van Kempen wrote:Hi all ,

with new activities like AEGT and CEGT I have finally decided to declare Nunn Blitz tournaments as concluded.

Heinz

I think this is a shame, actually.

I don't believe as that the blitz results were so bad. I run tests for GLC on a Duron 800MHz with 40/5 time control. If it was lightning... well. And it had the great benefit that a program got a rating very quickly.

Never-the-less, as Tord said, AEGT and CEGT are great compensation.

Keep up the good work.

Cheers, Tim.

Hi Tim ,

thanks for pointing out what is important for you.

For AEGT there is also 40/4 repeated additionally for the best engines and for fast statistics. We ran 800-1200 games mainly for King Class engines.

I also asked in the forum if we should do the same for Queen Class etc., but the usual response was silence. So nothing more was done in this respect.

AEGT will be a tournament especially for the authors and only interested will play and have many games. But of course without feedback, testers will lose motivation and enthusiasm and it will finally stop. That is all I can say at this point.

Let us hope that at least a few people will understand this and that there will be more feedback for all tournaments (not only AEGT) in the future. I saw already that some stopped to test amateur engines, to have at least feedback from those readers who are only interested in the very best.

Best Regards
Heinz