Rodent NN

[Rebel]

Rodent from Pawel Koziol is one of those engines with an excellent HCE evaluation.

Rodent IV 0.22 is rated 2988 on CCRL and 2998 on GRL.

We decided to move Rodent to NNUE evaluation, created a dataset of 800 million analyzed positions with Rodent 4.022 and moved the 800 million positions into the learner, tested the new created net with Rebel and Rodent NN gained about 200 elo over Rodent 4.022

In reality the Rodent net is actually 300 elo stronger when Pawel will implement the net into his own (Rodent) source code.

http://rebel13.nl/windows/rodent-nn.html

[Frank Quisinsky]

https://www.amateurschach.de/t-ki/common/ki-ratings.jpg

Inside!
I wish Pawel good luck ... the playing style of the version I tested is fantastic!

[pohl4711]

In reality the Rodent net is actually 300 elo stronger when Pawel will implement the net into his own (Rodent) source code.

The question is: Will that happen or not? Means: Should we test this Toga 4/Rodent nnue Engine/Net-hybrid or should we wait for a "complete" Rodent nnue engine-release of Pawel?

[Rebel]

In reality the Rodent net is actually 300 elo stronger when Pawel will implement the net into his own (Rodent) source code.

The question is: Will that happen or not? Means: Should we test this Toga 4/Rodent nnue Engine/Net-hybrid or should we wait for a "complete" Rodent nnue engine-release of Pawel?

The idea from the start was that Pawel will implement the net in his own source code, thus a complete Rodent nnue as you have called it.

[Rebel]

https://www.amateurschach.de/t-ki/common/ki-ratings.jpg

Inside!
I wish Pawel good luck ... the playing style of the version I tested is fantastic!

It is indeed.

[Frank Quisinsky]

Hi Ed,

BTW:
I am looking in short-game stats from Rebel 14.2 (still running run-3 after 20 of 30 rounds or after 800 from 1.200 games).

Rebel 14.2 NN ... 16x won, 14x lost.

Wasp isn't on place 1 I am await before ...
I forgot Andscacs 0.1 NNSf ...

1. Andscacs 0.1 NNSf ... 76x won, 0x lost
2. Booot 7.0 NN dev ... 46x won, 0x lost
3. Fire 8.2 ... 44x won, 3x lost
4. Igel 3.010 NN ... 43x won, 0x lost
5. Tucano 10.00 NN ... 40x won, 3x lost
6. Wasp 5.30 NN dev ... 40x won, 4x lost
7. Arasan 23.2 NN ... 39x won, 0x lost

Maybe place 2 is possible (for the moment place 6) ... to many others are very strong in "run-3".
4x lost is a lot for Wasp ... the start of tourney with "short-games-statistic" was bad.
Very sure it will go higher.

Do you changed anything what can important here?
My feeling is that Rebel 14.1 NN was stronger in attacking chess!?

Elo gain is great ... for the moment +35
TogaIV 1.1 NNRe will be start soon (run-4).
Wilhelm created a logo for TogaIV.

Best
Frank