CrazyAra, ClassicAra, MultiAra 0.9.5 release

[IQ_QI]

Hello chess engine enthusiasts,
we are happy to announce that the first version of MultiAra is now available.
MultiAra is a version of the Ara project which supports all chess variants available on lichess.org (antichess, atomic, chess960, crazyhouse, king-of-the-hill, horde, racing kings and three-check).

The neural network models have been initialized by lichess.org game data from August 2013 until July 2020.
All variant (except chess960) have been improved using reinforcement learning.
Details can be found in the master thesis by Maximilian Alexander Gehrke which among others covers an ablation study between training from zero knowledge and supervised initialization, as well as a preliminary strength comparison against Fairy-Stockfish:
Assessing Popular Chess Variants Using Deep Reinforcement Learning (https://ml-research.github.io/papers/ge ... essing.pdf).

ClassicAra
The model for ClassicAra uses a new input representation (https://github.com/QueensGambit/CrazyAra/pull/134) and a new WDLP value output head (https://github.com/QueensGambit/CrazyAra/pull/123).
The old model and input representation for chess is still supported as the input representation version is now inferred by the model name.

More information and binaries for Linux, Windows, Mac for GPU and CPU can be found in the official release notes:

• https://github.com/QueensGambit/CrazyAr ... /tag/0.9.5

[dkappe]

Thanks for this. Do you have a sense of playing strength versus the previous release?

[IQ_QI]

Thanks for this. Do you have a sense of playing strength versus the previous release?

Yes, the new input representation and WDLP head resulted in a 44.83 +/- 20.54 Elo increase.
The neural network was trained on the same data set using the same training configuration.
Most of the strength increase is attributed to the new input representation.

ClassicAra

Code: Select all

tc=0/1:0+0.25
-openings file="UHO_V3_6mvs_+090_+099.pgn"
ClassicAra-0.9.5-rise_3.3-input_3.0-wldp vs ClassicAra-0.9.4-rise_3.3-input_1.0:
202 - 125 - 273 (WLD)
Elo difference: 44.83 +/-  20.54

The opening suite UHO_V3_6mvs_+090_+099.pgn is available on https://www.sp-cc.de/anti-draw-openings.htm.

CrazyAra

The crazyhouse version, CrazyAra, was only affected by a small speed increase and dynamically loading the neural network input shape.

Code: Select all

tc=0/1:0+0.25
-openings file="crazyhouse_mix_cp_130.epd"
CrazyAra-0.9.5 vs CrazyAra-0.9.4
119 - 112 - 369 (WLD)
Elo difference: 4.05 +/-  17.24

[dkappe]

I’ve used lichess data for training Bad Gyal. One of the challenges I faced is that agreed draws and resignations meant I didn’t have enough endgame and imbalanced positions. To that end, I played all the kingbase 2300+ FIDE games out to terminal with sf10 100k nodes. Made a huge difference in the training.

The data can be found here: https://github.com/dkappe/leela-chess-w ... -Gyal-Data

[AdminX]

Okay, I started a LTC game with new version of Slowchess 2.7

[d]rnbq1rk1/ppp1bppp/4pn2/8/2pP4/1PN1PN2/P4PPP/R1BQKB1R w KQ - 0 1

Both engines are thinking now.

[AdminX]

18 moves in ...

[d]2rr3k/4qpp1/p3pb2/1pn5/8/1P3P2/PBB2P1P/RQ1R2K1 w - - 0 1

[carldaman]

Hello chess engine enthusiasts,
we are happy to announce that the first version of MultiAra is now available.
MultiAra is a version of the Ara project which supports all chess variants available on lichess.org (antichess, atomic, chess960, crazyhouse, king-of-the-hill, horde, racing kings and three-check).

The neural network models have been initialized by lichess.org game data from August 2013 until July 2020.
All variant (except chess960) have been improved using reinforcement learning.
Details can be found in the master thesis by Maximilian Alexander Gehrke which among others covers an ablation study between training from zero knowledge and supervised initialization, as well as a preliminary strength comparison against Fairy-Stockfish:
Assessing Popular Chess Variants Using Deep Reinforcement Learning (https://ml-research.github.io/papers/ge ... essing.pdf).

ClassicAra
The model for ClassicAra uses a new input representation (https://github.com/QueensGambit/CrazyAra/pull/134) and a new WDLP value output head (https://github.com/QueensGambit/CrazyAra/pull/123).
The old model and input representation for chess is still supported as the input representation version is now inferred by the model name.

More information and binaries for Linux, Windows, Mac for GPU and CPU can be found in the official release notes:

• https://github.com/QueensGambit/CrazyAr ... /tag/0.9.5

Thanks for the new release.

The previous versions work well on my i7, but the new one just gives a nondescript error and will not run.

[AdminX]

Hello chess engine enthusiasts,
we are happy to announce that the first version of MultiAra is now available.
MultiAra is a version of the Ara project which supports all chess variants available on lichess.org (antichess, atomic, chess960, crazyhouse, king-of-the-hill, horde, racing kings and three-check).

The neural network models have been initialized by lichess.org game data from August 2013 until July 2020.
All variant (except chess960) have been improved using reinforcement learning.
Details can be found in the master thesis by Maximilian Alexander Gehrke which among others covers an ablation study between training from zero knowledge and supervised initialization, as well as a preliminary strength comparison against Fairy-Stockfish:
Assessing Popular Chess Variants Using Deep Reinforcement Learning (https://ml-research.github.io/papers/ge ... essing.pdf).

ClassicAra
The model for ClassicAra uses a new input representation (https://github.com/QueensGambit/CrazyAra/pull/134) and a new WDLP value output head (https://github.com/QueensGambit/CrazyAra/pull/123).
The old model and input representation for chess is still supported as the input representation version is now inferred by the model name.

More information and binaries for Linux, Windows, Mac for GPU and CPU can be found in the official release notes:

• https://github.com/QueensGambit/CrazyAr ... /tag/0.9.5

Thanks for the new release.

The previous versions work well on my i7, but the new one just gives a nondescript error and will not run.

Did you remember to to 1st run it from command line and type isready. The 1st run will take a while, but after that it should be fine. You only have to do this once.

[carldaman]

It opens up when run from the command line, but immediately crashes after isready.
Did not have to do any of this with the previous versions.

[Dokterchen]

It opens up when run from the command line, but immediately crashes after isready.
Did not have to do any of this with the previous versions.

That's how it was for me too, because I had overlooked it:

"The binary packages include the required inference libraries for each platform.
However, the models should be downloaded separately and unzipped.

CrazyAra-rl-model-os-96.zip
ClassicAra-sl-model-wdlp-rise3.3-input3.0.zip
MultiAra-rl-models.zip (improved MultiAra models using reinforcement learning (rl) )
MultiAra-sl-models.zip (initial MultiAra models using supervised learning)

Next, move the model files into the model/<engine-name>/<variant> folder."