Seer 2.50 NN vs. 46 engines, gauntlet 80 min. per game!

[Frank Quisinsky]

Hi there,

I would like to see Seer 2.5.0 NN:
A "dramatic" current situation in Trier, Germany!
You know, not for my wife but for me!



I stopped my still running "run-4" for 2.5 days and started the good and old gauntlet idea again I do in FCP ratinglist / SWCR ratinglist times.

With FCP Tourney-KI rules (40/20, 80 minutes games on 4.4Ghz).
Seer 2.5.0 have to play 1.380 games vs. 46 engines.
This need 2.5 days, started for around 20 minutes (22:20 German time).

Two live links only:

1. *.html tournament table:
https://www.amateurschach.de/fling/Seer_250-nn.html

2. *.pgn file:
https://www.amateurschach.de/fling/Seer_250-nn.pgn

Good luck to the programmer of Seer!

Ah, I added my current ratings after 82.050 games!
Can be checked here:

*.jpg: https://www.amateurschach.de/t-ki/common/ki-ratings.jpg
*.txt: https://www.amateurschach.de/t-ki/ratin ... atings.txt

Best
Frank

PS: Later this evening the information will be available on my site!

[Frank Quisinsky]

Hi there,

Seer 2.5.0 NN by Conner McMonigle
https://github.com/connormcmonigle/seer-nnue/

FCP Tourney-KI, run-g1 ... gauntlet-results
https://www.amateurschach.de/fling/seer_250-nn.html

FCP Tourney-Ki, games (*.pgn or *.cbh)
https://www.amateurschach.de/main/_download-ki.htm

FCP Tourney-Ki, miniatures for g-runs
https://www.amateurschach.de/main/_miniatures_g-run.htm

Seer 2.5.0 NN, final results:
+62,08 Elo gain to v2.4.0 NN

Current ratings v.4.14g1
https://www.amateurschach.de/t-ki/ratin ... atings.txt
https://www.amateurschach.de/t-ki/common/ki-ratings.jpg

April 26th, 2022 - Apr. 29th, 2022
run-g1: Seer 2.5.0 NN

Code: Select all

Miniatures, mate below 60 moves (won)  =  30 games = 2,173%
Miniatures, mate below 60 moves (lost) =  06 games = 0,434%
Fast draws below 40 moves              =  68 games = 4,927%

CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
 all games         87     1380
 white won         88      356
 black won         90      185
white draw         89      315
black draw         85      433
white loss         86       19
black loss         81       72
  all wins         89      541     1380      39.2%
 all draws         86      748     1380      54.2%
 all loses         82       91     1380       6.6%

What I like:
Seer played more aggressively than in version 2.4.0!
Very nice games in fighting end or better the late mid-games ... that is really the strength of Seer.
Great improvements!

What I dislike:
- fast draws below 40 moves ... clearly to high!
- Seer is thinking for mate in x (example often for mate in 1 move over a minute). Without resign mode I lost really a lot of time and ressources if engines are not able to play moves before mate directly.

Fact is:
Seer is absolutely a very nice TOP-10 CPU-Engine and I like the work a lot.

Thanks again for Seer 2.5.0 NN.
Have a nice day!

Best
Frank

PS: If I start a gauntlet, I am thinking about such a profile in my messages.
The idea is not yet well implemented and will certainly be improved.

[Frank Quisinsky]

Hi there,

Unfortunately, I produced a bit of a chaos. During my still running "run-4", 2 engines are replaced.
I need 2 days to correct all data. Then I start run-g2 = Dragon 3 NN (Komodo).

It makes no sense to test Stockfish 120422 dev when a few days later Stockfish 15 was released. Furthermore I have to replace Bit-Genie 9.26 dev, because the version is much weaker than Bit-Genie 9.19 dev.

Sunday in the evening I will start my second CPU-Gauntlet, called run-g2.
After half a year of tournaments, I need to see some Gauntlet's again.

I updated my organization page:
https://www.amateurschach.de/main/_fcp-tourney-ki.htm

Best and sorry for my little chaos on weekend!
Frank

[connor_mcmonigle]

Hi there,

Seer 2.5.0 NN by Conner McMonigle
https://github.com/connormcmonigle/seer-nnue/

FCP Tourney-KI, run-g1 ... gauntlet-results
https://www.amateurschach.de/fling/seer_250-nn.html

FCP Tourney-Ki, games (*.pgn or *.cbh)
https://www.amateurschach.de/main/_download-ki.htm

FCP Tourney-Ki, miniatures for g-runs
https://www.amateurschach.de/main/_miniatures_g-run.htm

Seer 2.5.0 NN, final results:
+62,08 Elo gain to v2.4.0 NN

Current ratings v.4.14g1
https://www.amateurschach.de/t-ki/ratin ... atings.txt
https://www.amateurschach.de/t-ki/common/ki-ratings.jpg

April 26th, 2022 - Apr. 29th, 2022
run-g1: Seer 2.5.0 NN

Code: Select all

Miniatures, mate below 60 moves (won)  =  30 games = 2,173%
Miniatures, mate below 60 moves (lost) =  06 games = 0,434%
Fast draws below 40 moves              =  68 games = 4,927%

CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
 all games         87     1380
 white won         88      356
 black won         90      185
white draw         89      315
black draw         85      433
white loss         86       19
black loss         81       72
  all wins         89      541     1380      39.2%
 all draws         86      748     1380      54.2%
 all loses         82       91     1380       6.6%

What I like:
Seer played more aggressively than in version 2.4.0!
Very nice games in fighting end or better the late mid-games ... that is really the strength of Seer.
Great improvements!

What I dislike:
- fast draws below 40 moves ... clearly to high!
- Seer is thinking for mate in x (example often for mate in 1 move over a minute). Without resign mode I lost really a lot of time and ressources if engines are not able to play moves before mate directly.

Fact is:
Seer is absolutely a very nice TOP-10 CPU-Engine and I like the work a lot.

Thanks again for Seer 2.5.0 NN.
Have a nice day!

Best
Frank

PS: If I start a gauntlet, I am thinking about such a profile in my messages.
The idea is not yet well implemented and will certainly be improved.

Thanks for testing.

[Frank Quisinsky]

Hi Conner,

and Winter is in the group of opponents.
I am thinking on your hint, other thread for some weeks.
It seems you like Winter, indeed a very interesting engine!

Furthermore, we discussed about NN by Stockfish engines:
After the dicussion I added a new rule:
Max. 4 engines in one gauntlet = derived from Stockfish or NN from Stockfish or NN from Rebel = not more as 10% of opponents!

For me are the NNSf (NN file based on Stockfish) important because in my opinion the style of engines is not changed with NN file based on Stockfish. So it's interesting for testing and looking for me.

But I saw yesterday that I forgot in Seer 2.5.0 NN gauntlet Hallogen as opponent!



I will make during this week a small correction (add Halogon in the group of participant).
After the still running Wasp 5.53 NN dev gauntlet.

Go Seer go ... on the way to a TOP-5 engine!



Best
Frank

[Frank Quisinsky]

Hi there,

the two missed matches played on FQ-2 (i7 11th, 3.2Ghz) with 40 moves in 32 minutes.
Need it for future tourneys, wrote that before!

https://www.amateurschach.de/fling/seer ... issed.html

Code: Select all

  09. Seer 2.5.0 NN               :  3346.48   1470    66.6   582   795    93   979.5   54.1  12.19  3211.30  12.00   49.0
  --. Seer 2.4.0 NN               :  3283.37   2505    61.1   830  1400   275  1530.0   55.9   8.86  3194.23  11.57   88.9

At the moment after 1.470 games = +63,11

Code: Select all

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

run-g1
April 26th, 2022 - Mai 04th, 2022
Seer 2.5.0 NN, USA
1.470 games
https://github.com/connormcmonigle/seer-nnue/

Miniatures, mate below 60 moves (won)  =  32 games = 2,176%
Miniatures, mate below 60 moves (lost) =   7 games = 0,476%
Fast draws below 40 moves              =  70 games = 4,761%

CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
 all games         87     1470
 white won         89      386
 black won         91      196
white draw         88      330
black draw         85      465
white loss         86       19
black loss         81       74
  all wins         90      582     1470      39.6%
 all draws         86      795     1470      54.1%
 all loses         82       93     1470       6.3%
 
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Best
Frank