booot 7.3 release

[booot]

Good day, community!

Here is the new Booot release. Lots of experiments, lots of net-tests and finally one small step forward.
This time the net is really big but i can not say 'really smart' . But anyway new version seems about 40-50 elo points stronger on my tests.

Thanks to Andrew Grant for bug-report.
Thanks to Volodymyr Shcherbyna for help with EGTB support for Booot.
Thanks to all of you who find my engine interesting!

Best regards,
Alex Morozov
Kiev 2024

https://github.com/booot76/Booot-chess- ... es/tag/7.3

[Frank Quisinsky]

Hi Alex,

do you know such stats from your engine:

90-minutes games ... with Booot 7.2 NN

Code: Select all

 8. Booot 7.2 NN                   (70+,   479=,   91-)  48.4%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games        107      640
           white won         85       56
           black won         86       14
          white draw        118      250
          black draw        106      229
          white loss         95       14
          black loss         91       77
            all wins         85       70      640      10.9%
           all draws        113      479      640      74.8%

250-minutes games ... with Booot 6.4

Code: Select all

 6. Booot 6.4                      (19+,   110=,   86-)  34.4%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         84      215
           white won         72       12
           black won         97        7
          white draw         79       65
          black draw         80       45
          white loss         94       30
          black loss         90       56
            all wins         81       19      215       8.8%
           all draws         79      110      215      51.2%
           all loses         92       86      215      40.0%

90-minutes games ... with Booot 6.4

Code: Select all

 3. Booot 6.4                      (733+,   1106=,   161-)  64.3%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         78     2000
           white won         74      440
           black won         80      293
          white draw         74      510
          black draw         79      596
          white loss         96       50
          black loss         91      111
            all wins         77      733     2000      36.6%
           all draws         77     1106     2000      55.3%
           all loses         93      161     2000       8.1%

A complete other playing-style.
No other engine I know produced such differents with move-average stats.

So, what is wrong?
From first place in move-average stats (TOP-50 engines around the year 2020) to the last place (TOP-50 engines around the year 2023).

Best
Frank

PS: THANKS for your new version.

I wrote often ... Booot 6.4 ... the superstar from the engines because incredible human-like-play and strengths your versions 6.4 / 6.5 produced. But the style to your neural-network versions is complete changed ... in my opinion to 100%. Your older versions are one of my TOP-3 favorits.

Average of draws ... around 30- over 40 moves higher!!!

Note: All stats based on games with resign = off.

[Frank Quisinsky]

Hi Alex,

you wrote:
Thanks to Volodymyr Shcherbyna for help with EGTB support for Booot.

If I install your version 7.3 I can't set a directory to EGTB support.

Best
Frank

[booot]

Booot uses syzygy EGTB only for FEN-generating during self-playing now. Booot is Pascal engine, so only way to do it - via external DLL and only for WIN64. I will try to improve the probing code and release in next version.

I also like my old HCE engines. NN-evaluation makes engines too boring. But... ELO is the king.

[Frank Quisinsky]

Hi Alex,

back to my first posting ...

I hope you will not misunderstand my first posting.
You know that Booot has been one of my three favourites for so many years (Wasp, Slow and Booot).
I like your work very much, but I can't understand what your engine v.7.2 is playing.

A good example is the bad bishop endgames.
Example pawn on h- or a-line. with a bishop of wrong colour.
Your engine gave advantage and avoided draw ... 50 move rules.

In other absolutely clear endgames your engine avoids draw after 50 moves rules more times.
A pawn move before.

Do you add a secret swindle mode in Booot or a secret contempt or any of that crap?

Notice:
Your engine has no big problems with all this stuff in older versions 6.4 / 6.5.

And the end result of all this ... games do not find an end.
I am talking about "chewing gum draws" when you let the engine play with resign = off.

The attacking style your engine has is lost to more than 50%.
If I compare 6.4 / 6.5 with your latest 7 versions with neural network.

All the nice moves, can open complicated positions (often in the middle of the board) are no longer playable in many games according to statistics.

Alex, I think you must be a very strong chess player.
Because in detail your versions 6.4 / 6.5 are really wonderful.
Comes from a chess god.

Go your own way without all this help, maybe this will make your program weaker.
This is really important, because you develop one of the most interesting chess programs of all time.
The style your engine has produced should not disappear.

Download the database from my 40/20 and 66+6 games here:
v1.05.1 around 44Mb
https://www.amateurschach.de/fling/index.html

And study the games for the 40 in 20 tournament.
You can find much more what isn't very nice and again ... your older versions are wonderful without such problems!

Friendly
Frank

[Modern Times]

I also like my old HCE engines. NN-evaluation makes engines too boring. But... ELO is the king.

That is true, Elo at the end of the day is what counts. It is the most important measure of success.

[booot]

"Example pawn on h- or a-line. with a bishop of wrong colour.
Your engine gave advantage and avoided draw ... 50 move rules."

It is obvious : engine does not understand what 'draw' means. It tries to maximize the evaluation. 50 moves rule is 0. Avoiding it - some small plus. Its still bigger then 0 .

[booot]

"All the nice moves, can open complicated positions (often in the middle of the board) are no longer playable in many games according to statistics."

200+ elo stronger engine does not play weak moves. Even those moves are attractive. Stronger engine - boring games.

[Frank Quisinsky]

Hi Alex,

often I think the same but we have so many examples our opinion about it is wrong:
1. Uralochka, 2. Revenge, 3. Igel

All the engines keep the strong style and keep the strong stats!
And all are very strong in playing-strength.

End of the day ... you are the programmer ... it must have to do with the training method (think so).
Can't believe that all of that what is really great in Booot is futsch (oh, German word) ... is lost.

Furthermore, playing strength isn't all and boring ...
You know that humans have not more as 2900 Elo.
To keep the style is in my opinion much more important as playing strength.

Or you develops a programs, we have more times available with other names.
Means if a program lost the own face.

And your engine is again ... one of the most interesting chess programs "all times".
There is no doubt about it ... to 100% sure!

Please thinking about it ...
Wrote that friendly because I had in the past so many fun with Booot. So many nice hours in looking what your engine can do.

Best
Frank

[Frank Quisinsky]

Alex,

a bit motivation for a great programmer.
I hope all is OK with you and your family
in the bad times around your country, or better in the bad times we life.

Take care of yourself and your family.

Best
Frank