In 1980 my wife said to me : search yourself a hobby. I did, bought the TRS-80 and from one thing came another.
Rebel 15
1. New and stronger neural network made from
1.87 billion positions.
2. Estimated strength increase over Rebel 14.2 emulating
CCRL 40/2 testing in the range of 80-130 elo.
3. There are 2 versions of Rebel 15
. Rebel-15-avx2 for modern computers
. Rebel-15-sse for somewhat older computers.
Use Rebel-15-avx2 wherever possible since it
runs considerable faster.
Credits
- Fabien Letouzey, Thomas Gaksch and Jerry Donald Watson for Toga.
- Chris Whittington for introducing me into NNUE, providing me with the necessary tools and writing the NNUE interference code.
- Thorsten Czub for testing Rebel 15.
- Rebel 15 is released under the GPLv3 licence.
___________________________________________________________________________________________________
REBEL 15x2
Double sized neural network
x2 stands the multiplication of the neural net with a factor of 2 from 22Mb to 44Mb. Rebel 15x2 is not meant as a new elo bomb but as a final touch of Rebel 15 and investment for a future Rebel.
Because of the double sized network (based on 2.75 billion positions) Rebel has much more chess knowledge available which can be easily demonstrated by playing a match on depth instead of regular time control. The match ended in favor for the x2 version with a 58% score indicating a 56 elo increase.
However there is an important drawback, because of the bigger neural network the NPS has dropped significantly, depending on your PC with 25% - 35% which has to be compensated by the better chess knowledge hence the elo gain is not much. On the other hand it makes Rebel 15x2 much more suitable as an analysis engine and an excellent starting point for creating an even better neural network.
Rebel 15x2
______________________________________________________________________________________
REBEL 15.1
New network architecture
1. New network architecture, version 10.
2. Increased data set from 2.75 billion positions to 3.6 billion positions.
3. Search improvements.
4. Estimated elo gain based on GRL testing ~56 elo.
5. Rebel 15.1 development diary.
Rebel 15.1
______________________________________________________________________________________
Match results against various popular opening books
BOOK | SIZE | GAMES | SCORE |
Performance.bin | 1.4 Mb | 800 | 65.1% |
ProDeo.bin | 4 Mb | 800 | 65.5% |
Komodo.bin | 9 Mb | 800 | 62.3% |
Cerebellum.bin | 173 Mb | 800 | 55.7% |
Echidna.bin | 308 Mb | 800 | 57.8% |
Opponents | CCRL | CEGT |
Berserk 6 | 3429 | 3347 |
Ethereal 12.75 | 3396 | 3322 |
Rubichess 2.2 | 3445 | 3376 |
Houdini 6.03 | 3440 | 3357 |
Rebel Engine | version 15.1a |
Rebel book size | 243 Kb |
Time control | 40/40 |
The Noomen book comes with the latest neural network (as Rebel 15.1a) estimated 15-20 elo stronger, pre-installed and ready to use in Arena 3.0
Get the lastest ARENA inferface. ARENA is copyrighted © by Martin Blume and its distribution is limited to the official ARENA website only.
It's adviced to install Rebel-15.1a into the root of the C-drive else modify the polyglot.ini file in the engine folder accordingly.
PGN 4000 games
Rebel with Noomen Book
Click Me
Page views since the release of Rebel 14
Last modification August 20, 2022