Wasp 6.00 NN available ...

[Frank Quisinsky]

Hi there,

John Stanback released Wasp 6.00 NN.

Note:
Released on:
https://www.waspchess.com (for the moment a mirror site).
Will be transfair from a German provider to an US provider in the next months.
At the moment is www.waspchess.com more or less the mirror site.

Wasp 6.00 NN will be available on
https://waspchess.stanback.net/index.html
soon.

Small problems with my provider but not important for the community.

---

I have more interest to switch to the much more important information for users:

Elo gain in my testing with 40 in 8 vs. others attackers around +60 (512Mb for hash, 5-pieces, without ponder, 1 core, 4.1Ghz current AMD / Intel hardware).

Vs. all TOP-41 engines Elo gain is around +70.
In blitz chess Elo gain is around +80 Elo.

I like that move average can be hold with around 82 moves (below mate, vs. others attackers).
The attaicking style hold if I compare to the last release v5.50 NN, I think a bit more aggressive.

So, this is the strongest Wasp version ever, all great stats hold or a bit better as Wasp 5.50 NN (last release).

Super work from John Stanback.
John is working many time on this version and I am very happy for his great result.

Elo = 3250 - 3260 with 1 core and middle time controls if Shredder 13 with 1 core = 3125.
This should be around place 17 in the world. Best place ever! Maybe place 18!?

Have fun with Wasp 6.00 NN and thanks to John Stanback.
The biggest Wasp fan is more as happy ...

v6.00 = 3250 Elo = 2.850 Elo on Raspbery Pi on PihPhisto project (not realeased).
So, Wasp 6.00 NN with 1Ghz and the great human style Wasp have is on the level from World Champion Carlson on 1Ghz Raspbery PI.

FANTASTIC!

Thanks John!!

Note again:
Please donwload Wasp from
https://www.waspchess.com

Best
Frank

[Frank Quisinsky]

Ah ...

Here are the results vs. other attackers (+ Seer, not an attacker).
+61 Elo with 40 moves in 8 minutes without resign mode

https://www.amateurschach.de/frank/fcp_ ... ecret.html

At the moment I am testing Fire 8.3 and Velvet 4.1.0 NN
https://www.amateurschach.de/fling/fcp_ ... ecret.html

Best
Frank

If anywhere need the games, please let me know.
My website is closed since a while.

PS: Wasp 5.62 NN dev and Wasp 6.00 NN (release version) is the same!

[Frank Quisinsky]

Note 2:
Stockfish 11 (last version without NN is in my test).
Move average is clearly lesser as from Stockfish 15.
Quantity of short won games (miniatures) from Stockfish 11 to Stockfish 15 is the same with 40 in 8.
I am working on a test-set ... so Stockfish 11 is more interesting for me because Stockfish 11 lost only 1 of 2.000 games vs. the group of attackers below 60 moves without resign mode.

The reason for Stockfish 11 in my tourney!
No engines are inside with a move average higher as 92 below mate !!!

Test tourney are with Contempt = 0!
Seer is a great engine in defense and endgames with a great move average, so Seer is in the group of test-engines (attackers + Seer).

Best
Frank

But the main topic is Wasp 6.00 NN.
I am very happy with this great work John do here.

Example:
In the group of attackers (short lost games ... Wasp is on place 3).

[chessica]

Hi there,

John Stanback released Wasp 6.00 NN.

Note:
Released on:
https://www.waspchess.com (for the moment a mirror site).
Will be transfair from a German provider to an US provider in the next months.
At the moment is www.waspchess.com more or less the mirror site.

Wasp 6.00 NN will be available on
https://waspchess.stanback.net/index.html
soon.


Best
Frank

Hi Frank, are there two versions? Once with NN and once without? How big is the file?

[Scally]

Great stuff,

Cheers lads.


Al.

[Chessqueen]

Note 2:
Stockfish 11 (last version without NN is in my test).
Move average is clearly lesser as from Stockfish 15.
Quantity of short won games (miniatures) from Stockfish 11 to Stockfish 15 is the same with 40 in 8.
I am working on a test-set ... so Stockfish 11 is more interesting for me because Stockfish 11 lost only 1 of 2.000 games vs. the group of attackers below 60 moves without resign mode.

The reason for Stockfish 11 in my tourney!
No engines are inside with a move average higher as 92 below mate !!!

Test tourney are with Contempt = 0!
Seer is a great engine in defense and endgames with a great move average, so Seer is in the group of test-engines (attackers + Seer).

Best
Frank

But the main topic is Wasp 6.00 NN.
I am very happy with this great work John do here.

Example:
In the group of attackers (short lost games ... Wasp is on place 3).

What is the estimated rating of WASP 6.00 NN ?

[Scally]

It’s graded at 2960 on my RPi4, so I’d imagine it’s around 3360 on better hardware …






(click on a thumbnail for a larger image)


Al.

[Frank Quisinsky]

Scally,

can you bring me that ... directly!



Chessqueen,

if Shredder 13 with 1 core = 3125 ... Wasp 6.00 NN should be have 3250 - 3260 with middle time controls (not testing fast blitz).
Have a look in the links I posted with the 40 in 8 tourney. Elo is not my main interest. Much more interesting is that Wasp is great for opening book analyzes and the style of chess is really great, I think a "human-like-style".

Chessica,

NN file is embedded ...
Last version without Neural Network = Wasp 4.50


Short information for Scally,

I think start of December Alex is ready for a PiPhisto release.
At the moment Alex is working in different ideas I had.

Example:
On display you can see how many moves to "50-moves rule".
On some other nice ideas Alex is working.

Best
Frank

[jstanback]

Hi All,

Wasp can now be downloaded from //waspchess.stanback.net as well as from //www.waspchess.com. The NN is embedded in the binaries, there is no separate nn.bin file. I estimate Wasp 6.00 to be about 65 Elo stronger than Wasp 5.50. The strength increase comes from a larger net with more training data with the positions scored with a more recent, stronger net.

John

[Scally]

John, thank you

Frank, yes I’ve spoken to Alex thanks. I have the Elfacun Module which is similar to PiPhisto.

Al