OpenChess

OrgZ wrote: Fri Oct 06, 2023 2:20 pm

cubail wrote: Fri Oct 06, 2023 1:49 pm https://sites.google.com/site/computers ... nn-vs-book
look nnue number 1 book

My point exactly. nnue is a big FACTOR on these book tournaments.

I think that Cubail member (Mr. Cevdet) is meaning such as
It is better to test the opening books via NNUE assistance

Since he published the Duel, where NNUE was as Winner )

Greetings

Meanwhile, just in case of Book Championship as NNUE enabled,
Then it is not recommended to be used a engine, which produce
Very high draw percentages... e.g such as 98%-99% ...Otherwise,
No much way or meaningless to tune weakest or strongest lines, right?)
If not so clear...because almost all lines are ended such via values: 50%

For this reason (in case of NNUE enabled),
I prefer Rasubfish SC..still super strong. plus less drawish!
And based on tests, produces minus plus 10% less draws...

E.g there some other SF based engines even as without NNUE
Their draws are close or even higher than Raubfish SC NNUE

The main difference,
Raubfish SC is about + 150/+200 Elo stronger than these Drawish engines !

Sedat Canbaz wrote: Fri Oct 06, 2023 3:30 pm Meanwhile, just in case of Book Championship as NNUE enabled,
Then it is not recommended to be used a engine, which produce
Very high draw percentages... e.g such as 98%-99% ...Otherwise,
No much way or meaningless to tune weakest or strongest lines, right?)
If not so clear...because almost all lines are ended such via values: 50%

For this reason (in case of NNUE enabled),
I prefer Rasubfish SC..still super strong. plus less drawish!
And based on tests, produces minus plus 10% less draws...

E.g there some other SF based engines even as without NNUE
Their draws are close or even higher than Raubfish SC NNUE

The main difference,
Raubfish SC is about + 150/+200 Elo stronger than these Drawish engines !

that engine gives some exaggerated evaluations on positions . even if its a draw it will read +0.75 or +1.12 or +2.3. i think its this reason that it fights to win in drawen positions.
i have some games of it vs Raid v2.73 it was a war of blood but Raubfisch due to being old was obliterate.

Score of Raid v2.73 vs Raubfisch X48c3_nn_sl-avx2: 46 - 19 - 135 [0.568]
... Raid v2.73 playing White: 41 - 7 - 52 [0.670] 100
... Raid v2.73 playing Black: 5 - 12 - 83 [0.465] 100
... White vs Black: 53 - 12 - 135 [0.603] 200
Elo difference: 47.2 +/- 27.2, LOS: 100.0 %, DrawRatio: 67.5 %
200 of 200 games finished.

Thats what happened. unfortunatly am using the games to add to my book hence i cant produce the pgn. just to keep my book unique from the rest

OrgZ wrote: Fri Oct 06, 2023 4:16 pm

Sedat Canbaz wrote: Fri Oct 06, 2023 3:30 pm Meanwhile, just in case of Book Championship as NNUE enabled,
Then it is not recommended to be used a engine, which produce
Very high draw percentages... e.g such as 98%-99% ...Otherwise,
No much way or meaningless to tune weakest or strongest lines, right?)
If not so clear...because almost all lines are ended such via values: 50%

For this reason (in case of NNUE enabled),
I prefer Rasubfish SC..still super strong. plus less drawish!
And based on tests, produces minus plus 10% less draws...

E.g there some other SF based engines even as without NNUE
Their draws are close or even higher than Raubfish SC NNUE

The main difference,
Raubfish SC is about + 150/+200 Elo stronger than these Drawish engines !

that engine gives some exaggerated evaluations on positions . even if its a draw it will read +0.75 or +1.12 or +2.3. i think its this reason that it fights to win in drawen positions.
i have some games of it vs Raid v2.73 it was a war of blood but Raubfisch due to being old was obliterate.

Score of Raid v2.73 vs Raubfisch X48c3_nn_sl-avx2: 46 - 19 - 135 [0.568]
... Raid v2.73 playing White: 41 - 7 - 52 [0.670] 100
... Raid v2.73 playing Black: 5 - 12 - 83 [0.465] 100
... White vs Black: 53 - 12 - 135 [0.603] 200
Elo difference: 47.2 +/- 27.2, LOS: 100.0 %, DrawRatio: 67.5 %
200 of 200 games finished.

Thats what happened. unfortunatly am using the games to add to my book hence i cant produce the pgn. just to keep my book unique from the rest

Thanks for the test...

However, your test is not clear...rather I'd say a mystery )
Reason, no any info over test cond. ...exc. names and results ))

But however, according to my experience,
It seems you used again weak suite openings, right ?
Because via Strong suite..no way to see these draw numbers!!
Exc. very likely with ultra fast time controls, such as 5 sec / game

Btw, I tested Raubfish SC in recent tours as well
And Raubfish scored just about 5 to 15 Elo less:
https://sites.google.com/site/computers ... s-37-blitz
https://sites.google.com/site/computers ... nnue-cs-37
https://sites.google.com/site/computers ... nnue-cs-36

In same time that means:
SF improvement is almost dead.. since 2 years !!
Exc. we can see some improvements via weaker lines...
But via strong openings... no much Elo gain for sure!
At least all my recent tours indicates this...

By the way, who care much about these weak opening suites?
At least these handicapped games are not so useful for my side...
If not so clear...just do not come with these critical lines in SCCT tours )
I assume these kind of books will be down and down...fighting for last places!


Greetings

TC: 3+3.60

Raid exp OFF

Opening suite: adrian Oliver does tests yes. so i created my test suite from some of the games which he did. the openings correspond to openings which are found in bin books. so their normal strong openings. nothing else was added on the engines.

Mr Sedat why would i add low time control games to my book . all book games that i add are from TC 3+2 nothing less

OrgZ wrote: Fri Oct 06, 2023 6:05 pm TC: 3+3.60

Raid exp OFF

Opening suite: adrian Oliver does tests yes. so i created my test suite from some of the games which he did. the openings correspond to openings which are found in bin books. so their normal strong openings. nothing else was added on the engines.

Interesting...and Congrats!

Because via NNUE 3700+ to see such low draw ratio: 67.5 %
Never seen before...sure I mean if testings via strong openings...
And of course, if we use and run tour with optimal test conditions too )

On other hand,
I wonder...Oliprog tests are usually close to 95-99% draws
That's why ...how did you manage ... ?)

Btw, can you share please your suite ?
And the played 200 games as well...

Later I may try tıo test and use your suite under
SCCT conditions as well....

Greetings

OrgZ wrote: Fri Oct 06, 2023 6:38 pm Mr Sedat why would i add low time control games to my book . all book games that i add are from TC 3+2 nothing less

Good... but actually it all depends...

For examples,
My produced Bullet games can be counted as Blitz on my older machines..
Or as other example, the played Blitz games can be counted as Rapid etc. e.g
2x EPYC 7B12 128 Cores/256 threads is one of fastest computers for chess...

If still not so clear,
You can check Bullet's power and your engine is another Hercules for sure :
https://sites.google.com/site/computers ... gth-test-4
https://sites.google.com/site/computers ... testings-3

Greetings

Sedat Canbaz wrote: Fri Oct 06, 2023 7:03 pm

OrgZ wrote: Fri Oct 06, 2023 6:05 pm TC: 3+3.60

Raid exp OFF

Opening suite: adrian Oliver does tests yes. so i created my test suite from some of the games which he did. the openings correspond to openings which are found in bin books. so their normal strong openings. nothing else was added on the engines.

Interesting...and Congrats!

Because via NNUE 3700+ to see such low draw ratio: 67.5 %
Never seen before...sure I mean if testings via strong openings...
And of course, if we use and run tour with optimal test conditions too )

On other hand,
I wonder...Oliprog tests are usually close to 95-99% draws
That's why ...how did you manage ... ?)

Btw, can you share please your suite ?
And the played 200 games as well...

Later I may try tıo test and use your suite under
SCCT conditions as well....

Greetings

My bad seems i mixed info. Raid vs Raubfisch were using SCCT_Champ_182.pgn
my question is how are these results from your test suite?

OpenX 011023.pgn (my open suite created using Adrian test games)
was use by my personal Engine.
TC: 3+2.60
Score of Tanick X-011023_modern vs Raubfisch X48c3_nn_sl-avx2: 26 - 11 - 163 [0.537]
... Tanick X-011023_modern playing White: 16 - 5 - 79 [0.555] 100
... Tanick X-011023_modern playing Black: 10 - 6 - 84 [0.520] 100
... White vs Black: 22 - 15 - 163 [0.517] 200
Elo difference: 26.1 +/- 20.5, LOS: 99.3 %, DrawRatio: 81.5 %
200 of 200 games finished.

if you want OpenX 011023 i will send it to you

i have this question, i have always wondered about giving engines these SYZYGY, don't you think their the reason for a high rate of draws? i believe it messes with some interesting choices which an engine can take in a certain position. you have all the experience Mr Sedat.

look at these engines using OpenX 011023

Score of DeepSync 1 sl_modern vs Tanick X-011023_modern: 14 - 11 - 75 [0.515]
... DeepSync 1 sl_modern playing White: 8 - 5 - 37 [0.530] 50
... DeepSync 1 sl_modern playing Black: 6 - 6 - 38 [0.500] 50
... White vs Black: 14 - 11 - 75 [0.515] 100
Elo difference: 10.4 +/- 34.1, LOS: 72.6 %, DrawRatio: 75.0 %
100 of 100 games finished.