Looking for a *.pgn tool that can do the following ... can be very interesting!!

[Frank Quisinsky]

Hi there,

some engines evaluate the first eval after opening book moves very high / to high.

I am looking for a *.pgn tool that can read a *.pgn database and give me a list with the average of all calculated first, second, third up to eighth (I think that is enough) evals, sorted by highest average for the first own calculated eval at first.

Example:
Database with 3280 games, 41 engines, all engines plays against the others = 160 games per engine.

Step 1:

1. Wasp X ... 0,23 / 160 games (average of first own eval from 160 games, black and white), 0,34 / 160 games (average of second eval from 200 games, black and white), 0,41 / 200 games (average of third own eval from 200 games, black and white)
2. SlowChess X ... 0,25 / 200 games (average of first own eval from 160 games, black and white), 0,28 / 160 games (average of second eval from 160 games, black and white) ...

and all that to all the other engines in database

Step 2:

In Step 2 the tool sorted all by white games.
Same list but all white games by engines only!

1. Wasp x ... 0,18 / 80 games (average of first own eval from 100 white games in database), 0,20 / 80 (average of second own eval from 100 white games in database) ... and so on up to the first 8 own calculated moves.

Step 3:

In Step 3 the tool sorted all by black games.
Same list but all black games by engines only!

End of the work by the tool I have three tables ...

1. Average of eval from black and white
2. Average of eval from white
3. Average of eval from black

and all for the first 8 own calculated moves!

With the final results that I can see from all engines in a *.pgn file important information for optimize opening books.

Very helpful, and for see much other things!

Tool can be give also such a list for the first 10 or 12 or more averages for own calculated moves! But in this case it's complicated because if a mate in the near (engine lost a game after 20 moves below mate for an example) it's complicated to create this. Tool have to give an information in how many games a mate is in the near. So I think for the first 8 moves it's enough.

Best
Frank

Maybe a programmer have interest to do that if the day is long and the night ...

[Frank Quisinsky]

Hi there,

can be also interesting for engine programmers ...

If the average of eval for the first 8 moves goes higher and higher, engine can play with contempt the next moves or can increase the aggressiveness.
if the average of eval for the first 8 moves will be reduced, engine can go in defensive mode for fighting for a draw.

More interesting is if engines do that during the complete game. Example: If after move 45 the average for the last 8 or 6 moves reduced, engine have to fight for a draw and contempt goes to 0. Or if the average constantly 0,00 ... game must be very fast ended (before 50-moves rule) with 3-fold. Engine get the order ... forced 3-fold.

Only loud thinking!

An engine, around 200 Elo weaker as the opponent can see that during the game by itself and can give the order ... try a draw or if stronger ... attacke with most aggressiveness. Different other ways with average of evals are for sure possible. In analyze line engine can give the info ... I believe my opponent is stronger or ... I believe my opponent is weaker, or ... the opponent have maybe the same level of strength.



And I will be happy if I can see that during the game is still running.

Best
Frank

[Frank Quisinsky]

Or more in detail ...

Engine can give the information ...
I believe my opponent is in that gaming-phase 50, 100, 150, 200 Elo stronger or weaker.



And we have the feeling, that the intelligence from engines are better to see during the game is still running in analyze line. Just an idea in NN-times, maybe again for the trash, maybe it's the right time for it. For years I had spoken with Martin Blume about it but after all I remember ... the time was wrong for such things. Engines aren't strong enough for such things.

In my opinion it's just as important to evaluate the opponent ... with nice comments ... as to evaluate the own moves only !! The only way todo that is with "average of eval".

Furthermore, I think today it's more interesting for engine programmers as for GUI programmers.

Best
Frank

[Rebel]

Suggested output with imaginary data -

Code: Select all

White and Black
Engine        Games    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]  Average 
Wasp X         160    0.20  -0.10  -0.02   0.08   0.13   0.16   0.22   0.26   0.15
Slowchess X    160   -0.12  -0.05   0.03   0.12   0.17   0.23   0.34   0.50   0.22

Something like this?

I have some time free.

[Frank Quisinsky]

Hi Ed,

yes, yes ... so great!

Note:
Same table you added as example:

For all games ... 160
For white games only ... 80 games
For black games only ... 80 games

and all in one *.txt file.
The tables maybe sorted ... engine with lowest eval at first or so on.
You know it better ...

And with the complete name of engine ...
Wasp 6.50 NN ... Wasp X was only an example.

So great if you have time for it.
Thanks for your time and interest on it.

Such things will give us much more information about the available engines.

Here you can find a small database (with 41 engines, my still running FCP-Tourney-2024):
https://www.amateurschach.de/download/_ ... y-2024.zip

You have enough tourneys and games on your hd.
But maybe you will use my database for it.



Thank you again!!

Best
Frank

[Frank Quisinsky]

Hi Ed,

you know ...
We can search the clear draw positions in database (means I can do that in my balanced FEOBOS book). So I can make my book database better.

Different other things we can read or can do with such information!

Enough ... I have a long day.
If you have questions, I can give an answere in the late evening.



Best
Frank

[Frank Quisinsky]

I had forgotten ...

If we use the tool later for opening book optimazition also ...
1 game for black and 1 game for white must be available before the tool is working / running.

So I can use the tool for check positions goes in a opening databases directly with an engine tournament.

All 41 engines have to play a B01 line with black and white and later I am looking ... to drawish or not by the eval of 41 engines.

Now is enough ...

And again, thank you!

Best
Frank

[Frank Quisinsky]

Hi Ed,

if you used my database for experiments ...
My FCP-Tourney-2024 database was a chaos database after switching engines during the first 4 rounds.

I made for minutes an update (v. 5.3.1)
Now ...
Database was sorted by round with GUI
Hiarcs Chess Explorer Pro 1.2.1 with ECO-codes.

The first 3.280 games = round 1- round 4.
Tomorrow in the late evening round 6 is ready for take-off.
So you can use tomorrow the first 4.920 games = 240 games per engines for your experiments.

Again the download file:
https://www.amateurschach.de/download/_ ... y-2024.zip



Sure, the work in programming is not easy!
I have a guilty conscience!

Best
Frank

[Rebel]

Again the download file:
https://www.amateurschach.de/download/_ ... y-2024.zip

Example for playing the black pieces, using the games from Stefan.

Code: Select all

Engine                    Games    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]
Ethereal 14.00 nnue        3000  -0.88  -0.85  -0.80  -0.80  -0.81  -0.85  -0.91  -1.01
KomodoDragon 3.2 avx2      2500  -0.47  -0.40  -0.35  -0.31  -0.26  -0.22  -0.20  -0.16
KomodoDragon 3.2 MCTS      2500  -0.68  -0.62  -0.57  -0.53  -0.58  -0.58  -0.92  -0.84
CSTal 2.0 avx2             2500  -0.86  -0.82  -0.82  -0.82  -0.85  -0.86  -0.90  -0.96
CSTal 2.0 EAS avx2         2000  -0.88  -0.82  -0.82  -0.84  -0.83  -0.84  -0.88  -0.88
Stockfish 16 230630        2500  -0.52  -0.44  -0.38  -0.32  -0.26  -0.19  -0.11  -0.04
Berserk 230818 avx2        3000  -0.40  -0.37  -0.35  -0.33  -0.33  -0.32  -0.33  -0.34

Tried the pgn in your download and noticed the scores are from the white POV, so I have to negate the black scores first to get a reliable overview, tomorrow is another day

[Frank Quisinsky]

Hi Ed,

Engines stronger after opening book moves, with a great move-average from win games, should have a better eval move by move. Fore search opening blunders is important that the eval for the first moves is not to high. A lot of engines give us clearly to high evals and we are thinking often the opening theory is bad. Stockfish stats are really great, also the results from Dragon avx2!!

I am very interested in the results from the other TOP-50 engines.

I can give your tool later:
Database with white / black win, lose, draw games only.
Results will be different and interesting too.

Best
Frank