Experiment with move-average ... the reason why!

[Frank Quisinsky]

Hi there,

normaly a TOP-41 tourney have a move-average from 87-90 moves without resign.
The current running tourney (last game is running) have a move average from 95,4 moves.
This must have a reason!

If I put the tourney games from all the "long-movers" in a new database I got this one:

Code: Select all

     Name                        Games     Win     Draw     Lose       Pts         S-B         %

Berserk 12 NN                 :   195  :   82+  :  110=  :    3-  :   137.0  :  12531.50  :  70.26%
Stockfish 16 NN               :   195  :   77+  :  118=  :    0-  :   136.0  :  12568.25  :  69.74%
rofChade 3.1 NN               :   195  :   44+  :  147=  :    4-  :   117.5  :  10818.75  :  60.26%
Obsidian 8.0 NN               :   195  :   47+  :  135=  :   13-  :   114.5  :  10455.75  :  58.72%
Koivisto 9.2 NN dev (hotfix)  :   195  :   48+  :  133=  :   14-  :   114.5  :  10403.25  :  58.72%
Viridithas 11.0.0 NN          :   195  :   40+  :  138=  :   17-  :   109.0  :   9863.50  :  55.9%
Booot 7.2 NN                  :   195  :   16+  :  153=  :   26-  :    92.5  :   8635.25  :  47.44%
Stormphrax 3.0.0 NN           :   195  :   19+  :  137=  :   39-  :    87.5  :   8020.75  :  44.87%
Black Marlin 8.0 NN           :   195  :   16+  :  141=  :   38-  :    86.5  :   7991.25  :  44.36%
Superultra 1.0 NN             :   195  :   16+  :  141=  :   38-  :    86.5  :   7978.75  :  44.36%
Akimbo 0.7.0 NN               :   195  :    7+  :  139=  :   49-  :    76.5  :   7197.25  :  39.23%
Smallbrain 7.0 NN             :   195  :    8+  :  125=  :   62-  :    70.5  :   6566.50  :  36.15%
Halogen 11.4 NN               :   195  :    5+  :  131=  :   59-  :    70.5  :   6565.00  :  36.15%
Willow 3.1 NN                 :   195  :    4+  :  124=  :   67-  :    66.0  :   6080.75  :  33.85%

First view ... congratulation the the programmer of Berserk!

And now this one ...
Such a tourney, inside are all the "Long-Movers" only, produced ...
a move-average from 105 moves!

That is NoGo!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
17-18 moves over standard!
To much of very boring endgames are in the database, fighting for nothing.

I heard that different programmers have no understandings for my messages to the move-average but the reason is to search in the own programming and not in "Frank Quisinsky" is completly crazy. OK, I am completly crazy but honestly ... thats boring chess, nobody like to see. 70% of engines in TOP-50 not have such problems.

Bad bishop endgames is no excuse if others have no problems with it.

I wish me that this group of programmers playing here test-games without resign-mode.
So its more easy to see what for problems the engines produced.
Looking in my tourney games is an other way!

Thinking on it energy is expensive!
To reduce the move-average is not the job from GUI programmers, that is a matter for the boss of an engine!

The programmers of Berserk, Koivisto and Ethereal know what is to do.
If I am looking in the move average results from Stefan Pohl and compare Torch with Berserk, Koivisto and Ethereal.

Best
Frank

Code: Select all

Games   = 1365   ( no result = 0,  FEN tags = 0 )
Players = 14   ( clusters = 1 )
Date Range: 2023.09.24 - 2023.12.23

Games with:  WhiteElo = 0   BlackElo = 0   BothElos = 0

White Wins = 322 ( 23.59 % )
Draws      = 936 ( 68.57 % )
Black Wins = 107 ( 7.84 % )
White Pct = 57.88 %
Black Pct = 42.12 %

ECO:  Total = 1365  A: 302  B: 324  C: 282  D: 215  E: 242
PlyCount:  Total = 1365  Range: 39-600  Average = 210.17  StdDev = 104.08

finished: be sure to rename/copy outSummary

[Frank Quisinsky]

Ah, I made the statistic with Chessbase ...
End of the game is 300 moves with Chessbase.

Many of the games from the 14 engines goes clearly longer as 300 moves.

I will not added the same stat again, but the move average isn't 105 moves ... the move average is in the near of 107 moves for this group of 14 engines.

107 moves

[Frank Quisinsky]

Have nothing against the programs or the programmers, like rofChade a lot, like Stockfish a lot ... OK, I Iike all what can play chess for that reason ... I am a computer chess freak since 44 years.

I am sure all the programmers do her best and they give us the programs for free. I know all this and will give my sorry if I am to aggressive with this topic. But the move-average goes in wrong direction.

Sorry, I must stop the tourney and try out what I can do for reduces the move-average.
Not all of my conditions are perfect for a tourney with such strong engines.

In the past I sent often Mark a mail with a link to the games for TWIC (if I produced a tourney with many games and an higher time-control). Do that since more as 20 years. I can't do this for my last tourney. Chess players are't happy if they find all the long and boring games in the database.

Tomorrow I start a new tourney without the 14 programs I have written before.
I hope that in my future tourneys I can use the 14 programs again with a better move-average.
A tourney without Berserk, without rofChade ... and all the others ... is hard for me.

[Modern Times]

To reduce the move-average is not the job from GUI programmers, that is a matter for the boss of an engine!

The programmers of Berserk, Koivisto and Ethereal know what is to do.

I'd be very surprised if any engine author has a specific objective to reduce game length in developing their engine. A reduced game length if it happens is probably a by-product of other objectives that they may have.

I like the purity of having no adjudication at all and letting games play to their natural conclusion, but long games are a consequence of that, and even humans resign lost games or agree draws. Have you considered using tablebase adjudication, or does that go against your principles ?

[Frank Quisinsky]

Hi,

never I saw that in any tournament before. And I have alone 1.7 million of games produced if I am looking in the own databases. I am looking all the time of move-average because that point is important for different important stats I like to to a very long time. Thinking in the past on Demolito. A great program but in every stats I can see the move-average goes to 100. Very rarely in the past that a program have a move average from 100.

Load the games from each of the engines and looking what the engines like to do in the endgames. Looking in the file pngs-stats. Here you can see such things directly.

Example Stockfish:

Code: Select all

32. Stockfish 16 NN                (236+,   403=,   1-)  68.4%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games        100      640
           white won         81      188
           black won         85       48
          white draw        107      132
          black draw        113      271
          white loss          0        0
          black loss         93        1
            all wins         82      236      640      36.9%
           all draws        111      403      640      63.0%
           all loses         93        1      640       0.2%

Example rofChade

Code: Select all

41. rofChade 3.1 NN                (141+,   472=,   27-)  58.9%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games        103      640
           white won         88      115
           black won         97       26
          white draw        105      201
          black draw        110      271
          white loss         75        4
          black loss         79       23
            all wins         90      141      640      22.0%
           all draws        108      472      640      73.8%
           all loses         78       27      640       4.2%

Example Berserk

Code: Select all

 5. Berserk 12 NN                  (205+,   426=,   9-)  65.3%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games        105      640
           white won         85      164
           black won         82       41
          white draw        126      155
          black draw        107      271
          white loss         74        1
          black loss        111        8
            all wins         85      205      640      32.0%
           all draws        114      426      640      66.6%
           all loses        107        9      640       1.4%

Let us compare this with absolutely normal stats ... engines produced since 20 years!

Code: Select all

25. Pawn 2.0 NN                    (30+,   406=,   204-)  36.4%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         92      640
           white won         84       20
           black won         80       10
          white draw         92      247
          black draw         98      159
          white loss         86       53
          black loss         88      151
            all wins         83       30      640       4.7%
           all draws         95      406      640      63.4%
           all loses         87      204      640      31.9%

Code: Select all

12. Chess.cpp 4.0 NN               (30+,   376=,   234-)  34.1%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         84      640
           white won         84       18
           black won        106       12
          white draw         88      230
          black draw         83      146
          white loss         80       72
          black loss         77      162
            all wins         93       30      640       4.7%
           all draws         86      376      640      58.8%
           all loses         78      234      640      36.6%

Code: Select all

23. Nemorino 6.11 NN dev           (27+,   415=,   198-)  36.6%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         92      640
           white won         86       20
           black won         83        7
          white draw         98      242
          black draw         94      173
          white loss         88       58
          black loss         82      140
            all wins         85       27      640       4.2%
           all draws         97      415      640      64.8%
           all loses         84      198      640      30.9%

And suddenly directly 14 programs offer such bad stats and so many long games.
If a chess player like to replay the long games he get headache.
That have nothing to do with chess.
Must be an other game!

To your question:
In all the years computerchess ... never I used resign=on.
No interest to produce "half games" or "unready" games.
Can nothing do with it.

107 moves ... end of 2024 maybe 30 programs, what is wrong in chess programmings?
A swindle mode ... do nothing the opponent create an error.
Hidden contempt = 24 parameters or what.

What is the genial idea programmers have for such cramp games?
I am very interesting to hear that!

Looking in the games ... before you should search a tablet against headaches!

[Ciekce]

I wish me that this group of programmers playing here test-games without resign-mode.
[...]
Thinking on it energy is expensive!

So you would have us waste ridiculous amounts of electricity making the millions of games that happen in testing longer, so that you can save the odd bit of electricity in tournaments?

How do you expect us to test for move average with a patch, anyway?

Literally just use adjudication in your tournaments if you are this obsessed with shorter games. It is not our job to fix a problem of your own making.

[Frank Quisinsky]

Hi Conor,

thinking about Winboard times, contacts with Tim and others working on GUIs. Thinking about the time I worked with Martin Blume on Arena and all the contacts we had with chess programmers (hundrets of such contacts). On the comments of Bob Hyatt as UCI is being developed (Bob likes the WB protocol and dislikes UCI for various reasons) and on all the discussions around the topic: Adjudication!

Related ... GUI can use Nalimov if engine does not support it itself. So the opinions of so many programmers are very clear. No endgame bases for engine if engine does not support them.

GUI = slave
Engine = matter for the boss, the job of engine programmers!

Not sure, maybe 1,000 or 2,000 or more chess programs are available today. I start with the first 4 available Winboard engines my first engine tournaments. So many programs are tested all the time ... maybe 500 or more? But such games I have never seen and never written by a programmer ... its not our job.

Conor, I often can't keep up with the speed at which younger people implement things and see things more easily. And most of it is new and honestly very, very strong. Much better as the olders are thinking. I saw that ... but again ... the stats from the 14 engines I added before are very clear and fact. And such statistics I have never seen before in all the years of computer chess.

If I were on the other side and younger ... very sure ... I would think that Quisinsky is crazy. Surely you have it in your head!
I am not crazy, but a tournament between 14 engines with a move average of 107 moves without resign ... that is crazy!

Best wishes and thanks again for your reply!
After all ... I can't give a comment like yes you are right in this case.
100x I like more to write ... yes you are right, but I have a completly other opinion!
Nothing to do with your person, because it seems that you are not alone with your opinion!

Frank and ... have a nice day!

[RubiChess]

Instead of repeating your high move average rant again and again, I would like you to explain how (the fuck) an engine like SF should avoid long draws in examples I gave you here viewtopic.php?f=2&t=82814&start=10#p954345 when you deny draw adjudication and so don't allow the engines offering draw by evaluating 0.00 for 300 moves in a row?

[Frank Quisinsky]

Written by a programmer with a move average currently in TOP-15 of TOP-50 engines!
Give the others a hint what you are doing as to play good friend for fellow programmers.
Time can be better used ...

I heard yesterday (I am not on discord) that some did not understand my comments about move-average. For this reason I build a simple example and try again to explain the opinion.

So, Stockfish is open source ...
If you want to help others, do it on the Stockfish site.

Move average RubiChess:

Code: Select all

28. RubiChess 20230918 NN          (161+,   472=,   7-)  62.0%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         91      640
           white won         85      125
           black won         91       36
          white draw         95      195
          black draw         91      277
          white loss          0        0
          black loss        103        7
            all wins         86      161      640      25.2%
           all draws         93      472      640      73.8%
           all loses        103        7      640       1.1%

Again and again such comments by yourself.
Do you need friends?

Best
Frank

[RubiChess]

Written by a programmer with a move average currently in TOP-15 of TOP-50 engines!
Give the others a hint what you are doing as to play good friend for fellow programmers.
Time can be better used ...

I'm not in the position giving SF developers hints about how they should do their job, cause lots of them are much smarter than me.

So, Stockfish is open source ...
If you want to help others, do it on the Stockfish site.

So you refuse to give any answers about how to avoid the draw in the linked examples. Again!

Move average RubiChess:
...

I already gave you the most reasonable answer why Rubi stats are "better" than stats of SF and Berserk:
Rubi is dumb enough to play into drawn positions that are known to be draw by most opponents so that they can agree to play into a 3-fold soon.
Look again at the examples I gave you. So playing as dumb and simple as most other engines is a sign of quality in your small world.

Again and again such comments by yourself.

I will continue as long as you ignore any arguments that don't fit into your narrow-minded attitude.

Do you need friends?

No, thanks.