SedatChess

[Sedat Canbaz]

Thank you very much, Sedat, for your incredible work.

However, without detracting from the work Dorsz has done, this contradicts the metrics on the following page.

Something doesn't add up.

No offense to anyone: https://nextchessmove.com/dev-builds

Best regards!

Dear Anton,

You are welcome and nice to read your valuable comments!

On other hand,
If you are meaning regarding the test via UHO opening suite,
You've right such as SF dev improvements did not work much!

But in the Official tour (via Unique opening suite),
SF 020425 dev. improvement seems to be better...!
Otherwise I hardly doubt to be ranked in 2nd place!

By the way,
if you're going to ask my view: why SF performed worse via UHO?
It's hard to know the true but often SF team's pre Elo diagrams,
- Does not reflect much to reality...at least under SCCT's cond. !

And just a few more words about SCCT conditions,
Hyper-Threading disabled, Large-Pages disabled,
Almost daily speed benchmarks to see stability, plus
Hash-table tests done, e.g to see which is best etc.

In other words,
Instead of SF' Team Elo diagrams, I'd prefer to rely mainly on
SCCT Hardware, Tournament Conditions, Improvements etc. !

Greetings )

[Sedat Canbaz]

As FINAL Conclusion:

1st of all, it is ALWAYS your CHOICE and
You can appreciate any standıng which you like more !)

But anyhow, according to my experience:
Definitely the standings via Unique suites represent better, more
Useful data (than UHO suite) to our actual chess reality... and why?
Because no so HUGE Elo differences between these Top engines!
Note: Sure here I refer about the Elo diff. for the test via UHO suite..
Moreover, Chess is a game for two sides..fl not clear: not mainly for 'Whites'!
As clear proof, please check UHO's win. percentages...almost all by WHITES!

What I can say more,
-Via Unique openings: better quality, more correct, safer!
-Via UHO openings: more FUN, more exciting, but a mystery
Eng strength, especially if the number of games is small !
For this, it's required many thousands of games (per player)
Otherwise, just a waste of time...because 'time' is money!

That's all for now and thanks for your interest... )


Best,
Sedat

[Anton101]

This book is generally used for fish testing: https://pixeldrain.com/u/2kBgYmHS

Here you can see where it comes from: https://github.com/official-stockfish/b ... a7f185c4b9

Here is an example of how the test is prepared, book depth, TC, hash, threads, etc: https://tests.stockfishchess.org/tests/ ... 2d74b1668a

Without a doubt, I understand that as a TD you have complete authority to decide what to do and what not to do in each test, however, one thing these people have taught us over time is to have created an almost invincible engine.

I don't have the computing power that you have or the tournament experience that you have, but I invite you to do this type of test at some point.

The final idea is to compare the results of the SF developers with your work and know the results obtained.

I can do a tour in fastchess, using cmd, but I don't have some engines.

Plus it will take me many hours to reach a thousand games per engine.

It's just an invitation, but in the end, you'll decide.

Best regards Mr. Sedat.

[Sedat Canbaz]

@ to Mr. Anton CES

Many thanks for your input...

As far as I understand,
Do you invite Me to be exactly 'in shoes' as SF team, right ?)

If so... I am afraid to say that
I can not adapt 100% SCCT exactly to SF team's testing conditions,
Since my tournament hardware is not exactly same as SF team..

Moreover, I always use Core/s (Hyper-Threading disabled) per engine!
The best testing system way, as most stable for my engine testings !

Where I assume SF team is using Thread/s (Hyper-Threading enabled)..

Besides, I noticed that UHO_Lichess_4852_v1 is EPD file, In short:
I am not a BIG fun of using Epd openings (in case of eng testings)
Due to produced games are not useful at all for creating books etc!

Anyhow, just out of curiosity...very soon
I will give a try via UHO_Lichess_4852_v1.epd as well!

And let's see what will be the 3rd tour's standings ?!

And please stay tuned..soon as possible new updates...


Greetings )

[Sedat Canbaz]

UPDATE

Due to respect/wish to Mr. Anton CES's interesting idea,
The tour via 'UHO_Lichess_4852_v1' openings started !

And here are the first unofficial results, standings after 250+ games:

And as usually,
Kookaburra is 1st, Stockfish dev 2nd, where SF PRO2 190325 is 3rd!

But what is unusual ?
Omega MPV is last ? any ideas over these issues?
Btw, I know very well too that 250 games (per player) is small data but...

In other words,
You will never notice via well-tuned openings such mystery !

BUT ANYHOW, NEVER MIND,
AS MOST IMPORTANT FUN (at least for this tour)

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 Kookaburra 3.10                34      30     253   54.9%   49.8% 
   2 Stockfish 020425               29      32     252   54.2%   45.6% 
   3 SF PRO2 190325                 21      31     252   53.0%   48.8% 
   4 Corchess 210325                14      31     251   52.0%   46.6% 
   5 SF-X 130425                    13      32     250   51.8%   44.4% 
   6 KillfishPB 110425              11      32     250   51.6%   44.8% 
   7 Artemis 12TR                    1      30     251   50.2%   51.0% 
   8 HypnoS++ 1.02                 -61      30     252   41.3%   50.8% 
   9 Omega MPV                     -63      31     251   41.0%   48.6% 

1131 of 4680 games finished.

[Sedat Canbaz]

UPDATE 2

And here are latest standings (after 550+ games per player)

Kookaburra leads as before, where Omega MPV is still last !!
Very strange really....normally HypnoS was as last in older tours..
It seems Omega MPV did not like/suit UHO_Lichess openings !?

On other hand,
The current produced draw ratio is incredible low...nice..but so far,
Determining the strength of overall chess engines is just a mystery!
Not so correct via 'UHO_Lichess openings' at least so far ...

And moreover,
I see these testings quite useful...otherwise without running them..
We would not know about all over...right ?

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 Kookaburra 3.10                34      21     553   54.9%   48.6% 
   2 Stockfish 020425               28      20     553   54.1%   50.3% 
   3 SF-X 130425                    17      21     553   52.4%   47.7% 
   4 Corchess 210325                16      21     552   52.4%   47.8% 
   5 SF PRO2 190325                 15      21     555   52.2%   48.8% 
   6 KillfishPB 110425              14      21     551   52.0%   46.6% 
   7 Artemis 12TR                   -5      21     553   49.3%   49.0% 
   8 HypnoS++ 1.02                 -56      20     553   42.0%   49.9% 
   9 Omega MPV                     -64      21     553   40.9%   49.5% 

2488 of 4680 games finished.

[Sedat Canbaz]

UPDATE 3

This post is going about the fake Troll members,
Like Deeds...who often loves to blame my tours )

One of SOURCES:
viewtopic.php?f=38&p=40059#p40059

What do you wait? SCCT ?
Why don't run and organize a useful/great rating under
Optimal cond. such as MP, Blitz, Rapid.. Ponder ON etc. ?!

Just in case, if via weak openings and eng 3800+ Elo,
Do not forget to run min. 10.000 games (per player) ,
And vs various participants as well... and as most
Important: it will be good, if NPS values to be shared..

Otherwise, sorry to say that but NO NPS = NO TRUST
Because for gaining TRUST..it's required many years, but
Just when you lose the trust.. it's required centuries!

Hope helps )

[Sedat Canbaz]

UPDATE 4

The tour via UHO_Lichess_4852_v1 is coming to its end
And about 120+ games (per player) are left..

As before, Kookaburra is 1st, Corchess 2nd, Stockfish 3rd!
Omega MPV is still last...

Btw, what I said before to all ?
https://sites.google.com/site/computers ... t-nn-cs-56
-Unpredictable (unbalanced) performances by many participants!
Reason? due to I use to play weak openings...no other explanation
Of that..sure this applies for today... not sure for tomorrow!

And here are the latest played results:

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 Kookaburra 3.10                32      16     919   54.6%   49.4% 
   2 Corchess 210325                24      16     919   53.4%   48.5% 
   3 Stockfish 020425               20      16     918   52.8%   49.9% 
   4 SF-X 130425                    19      16     917   52.7%   47.3% 
   5 SF PRO2 190325                 14      16     917   52.1%   49.8% 
   6 KillfishPB 110425               6      16     915   50.9%   47.5% 
   7 Artemis 12TR                    2      16     916   50.3%   50.3% 
   8 HypnoS++ 1.02                 -56      16     918   42.0%   50.2% 
   9 Omega MPV                     -62      16     917   41.2%   49.7% 

4128 of 4680 games finished.

[Anton101]

Great tour, many thanks Mr. Sedat.

[Sedat Canbaz]

Great tour, many thanks Mr. Sedat.

My pleasure Mr. Anton )