Neural Net

[Alex_Zet]

Neural Networks,
a series of connected neurons which communicate due to neurotransmission. The interface through which neurons interact with their neighbors consists of axon terminals connected via synapses to dendrites on other neurons. If the sum of the input signals into one neuron surpasses a certain threshold, the neuron sends an action potential at the axon hillock and transmits this electrical signal along the axon.

In 1949, Donald O. Hebb introduced his theory in The Organization of Behavior, stating that learning is about to adapt weight vectors (persistent synaptic plasticity) of the neuron pre-synaptic inputs, whose dot-product activates or controls the post-synaptic output, which is the base of Neural network learning [2].

Chess
Logistic regression as applied in Texel's Tuning Method may be interpreted as supervised learning application of the single-layer perceptron with one neuron. This is also true for reinforcement learning approaches, such as TD-Leaf in KnightCap or Meep's TreeStrap, where the evaluation consists of a weighted linear combination of features. Despite these similarities with the perceptron, these engines are not considered using ANNs - since they use manually selected chess specific feature construction concepts like material, piece square tables, pawn structure, mobility etc..

More sophisticated attempts to replace static evaluation by neural networks and perceptrons feeding in more unaffiliated feature sets like board representation and attack tables etc., where not yet that successful like in other games. Chess evaluation seems not that well suited for neural nets, but there are also aspects of too weak models and feature recognizers as addressed by Gian-Carlo Pascutto with Stoofvlees [41], huge training effort, and weak floating point performance - but there is still hope due to progress in hardware and parallelization using SIMD instructions and GPUs, and deeper and more powerful neural network structures and methods successful in other domains. In December 2017, Google DeepMind published about their generalized AlphaZero algorithm.

Move Ordering
Concerning move ordering - there were interesting NN proposals like the Chessmaps Heuristic by Kieran Greer et al. [42], and the Neural MoveMap Heuristic by Levente Kocsis et al. [43].

Giraffe & Zurichess
In 2015, Matthew Lai trained Giraffe's deep neural network by TD-Leaf [44]. Zurichess by Alexandru Moșoi uses the TensorFlow library for automated tuning - in a two layers neural network, the second layer is responsible for a tapered eval to phase endgame and middlegame scores [45].

DeepChess
In 2016, Omid E. David, Nathan S. Netanyahu, and Lior Wolf introduced DeepChess obtaining a grandmaster-level chess playing performance using a learning method incorporating two deep neural networks, which are trained using a combination of unsupervised pretraining and supervised training. The unsupervised training extracts high level features from a given chess position, and the supervised training learns to compare two chess positions to select the more favorable one. In order to use DeepChess inside a chess program, a novel version of alpha-beta is used that does not require bounds but positions αpos and βpos [46].

Alpha Zero
In December 2017, the Google DeepMind team along with former Giraffe author Matthew Lai reported on their generalized AlphaZero algorithm, combining Deep learning with Monte-Carlo Tree Search. AlphaZero can achieve, tabula rasa, superhuman performance in many challenging domains with some training effort. Starting from random play, and given no domain knowledge except the game rules, AlphaZero achieved a superhuman level of play in the games of chess and Shogi as well as Go, and convincingly defeated a world-champion program in each case [47].

[Alex_Zet]

Chess Fighter Neural Net

http://www.talkchess.com/forum3/view...hp?f=2&t=70206

[Alex_Zet]

Allie


AT
manyoso

https://github.com/manyoso/allie
https://github.com/manyoso
https://www.patreon.com/gonzochess75

[Alex_Zet]

Scorpio chess engine


Daniel Shawul

https://github.com/dshawul/Scorpio
https://github.com/dshawul
https://sites.google.com/site/dshawul/

[Alex_Zet]

Leela Chess Zero




https://github.com/glinscott
https://github.com/glinscott/leela-chess
https://blog.lczero.org/
https://lichess.org/@/LeelaChess

[svoitsl]

Вышел chess22k 1.13, с пофикснным крашем, если бы удалось заметить раньше то может и в Лигу 2 попал бы....

[Alex_Zet]

LCZero

TCEC бонусный матчLCZero ID125 Elo 2400 против Скорпиона 2.79
Лила одержала 4 победы, 3 ничьи и 13 поражений против Скорпиона.
https://lichess.org/study/MFUV2n3I

Where to play Leela online? https://blog.lczero.org/2018/11/wher...la-online.html
What is Leela Chess Zero? https://lichess.org/blog/WtNG7CcAAFM...ela-chess-zero
discord.gg https://discordapp.com/invite/pKujYxD

All the Leela's https://lichess.org/team/all-the-leelas
Bot and Fans https://lichess.org/team/bot-and-fans
Lichess Bots https://lichess.org/team/lichess-bots
The Bottios Fanclub https://lichess.org/team/the-bottios-fanclub
best bots https://lichess.org/team/best-bots
Сhess champions https://lichess.org/team/hess-champions
Lichess Bots Championship https://lichess.org/team/lichess-bots-championship
Titled players https://lichess.org/team/titled-players

[Alex_Zet]

The Chess Engine that Died So AlphaGo Could Live

Google DeepMind's Go engine is being praised for its "intuition." But everyone seems to have forgotten AlphaGo's predecessor, Giraffe.

Mar 14 2016, 4:00pm

https://www.vice.com/en_us/article/d...fe-matthew-lai

[sovaz1997]

Alex_Zet, статья 2016 года. Не взлетел Жираф потому, что не стал #1.

[Alex_Zet]

AylerKupp: <<keypusher> Yes, I think AK was just trying to define the four terms I was asking about (thanks, AK!), not say that Lc0 had prior knowledge or whatever.>
Correct as I mentioned to <alexmagnus> above. I hope that I succeeded to at least some extent.

Along those lines you and others might find this interesting. https://www.chessprogramming.org/Ne... gives a list (likely incomplete, but it's a start) of neural network-based chess engines in alphabetical order. I thought it would be more interesting to sort the list in chronological order and this is what I came up with:

1989 Morph
1994 SAL
1995 NeuroChess
1996 Alexs
1998 Chessterfield
1999 Octavius
2000 ChessMaps
2004 Tempo
2005 Hermann
2005 Scorpio
2006 Blondie25
2007 Stoofvlees
2011 Arminius
2014 Deep Pink
2015 Giraffe
2015 Zurichess
2016 Spawkfish
2017 AlphaZero
2017 Gosu
2018 Leela Chess Zero
(unk) Golch

The dates are approximate; they refer either to the time a paper described that particular engine or the time that the engine first appeared in a tournament or match. Still, I think it's interesting to see where AlphaZero and LeelaC0 fit in the chronology of neural network-based engines.

I think that Giraffe is of special interest. Its developer, Mathew Lai, developed it as part of his Master's thesis and continued to improve it. But he was hired by Google's Deep Mind and decided that it would be a conflict of interest to continue developing it, given his position as part of the AlphaZero development team and his knowledge of trade secret information obtained.

Of particular interest, I think, is Lai's paper "Giraffe: Using Deep Reinforcement Learning to Play Chess" which, unfortunately, I could no longer find a *.pdf version. And I think that the following is also interesting: https://motherboard.vice.com/en_us/...

AlphaZero was not the first application of reinforced learning as it applied to chess (see, for example, http://citeseerx.ist.psu.edu/viewdo..., published in 2008, but it's certainly the best known one. I think that reinforced learning in chess, as opposed to supervised training where the engine's neural network is trained, like LeelaC0, by providing it as inputs a set of games, holds the most promise in uncovering new chess principles because it's not constrained by the biases inherent in games played by human players. But again, what do I know?

http://www.chessgames.com/perl/chess...60016&kpage=17

AylerKupp http://www.chessgames.com/perl/chess...name=AylerKupp

[Alex_Zet]

Stoofvlees

https://www.chessprogramming.org/Stoofvlees

Stoofvlees, DarkQueen

http://forum.computerschach.de/cgi-b...w.pl?tid=10480

[Alex_Zet]

AlphaZero - Stockfish Match (2017 - 2018), London ENG,

Number of games in database: 220
Years covered: 2017 to 2018
Overall record: +62 -11 =147 (61.6%)

Most played openings
E15 Queen's Indian (24 games)
C67 Ruy Lopez (12 games)
A17 English (11 games)
E17 Queen's Indian (10 games)
C11 French (8 games)
C65 Ruy Lopez, Berlin Defense (8 games)
E16 Queen's Indian (6 games)
D43 Queen's Gambit Declined Semi-Slav (6 games)
C50 Giuoco Piano (5 games)
E00 Queen's Pawn Game (4 games)

http://www.chessgames.com/perl/chess...e=1&pid=160016
http://www.chessgames.com/perl/chess...60016&kpage=17

[Alex_Zet]

AlphaZero

chessprogramming -https://www.chessprogramming.org/AlphaZero
Wikipedia -https://en.wikipedia.org/wiki/AlphaZero

[Alex_Zet]

LCZero

v0.21.4

Two small changes in this release.

A fix for crashes that can occur during use of sticky-endgames
Change the false positive value reported when in wdl style resign and display average nodes per move as part of tournament stats in selfplay mode.


https://github.com/LeelaChessZero/lc0/releases

[Alex_Zet]

2019-07-26

End of era



The era of test40 has finished, that training of test40 has been stopped.

Instead we've just started to train test60. What's new there (relative to test40):

Residual tower size is 320x24 (24 blocks, 320 filters).
Policy head is AlphaZero-style (8×8×73 instead of plain 1858 vector).
Value head is WDL.
Instead of using fixed nodes for every move while training, KLD threshold is used instead (think less in obvious cases, think more in complicated cases, but in average it's still about the same as it was).
[disabled after net 60021 due to issues] Illegal moves are masked from policy head, so that neural network doesn't have to learn detecting legal moves itself.
Instead of training value head just on game outcome {-1; 0; 1}, a fraction of tree's Q value after the move is added.
Using different FPU at root vs for the rest of the tree.
As it was written in earlier post, it's not going to be a "clean run". If we discover improvements while test60 is being trained, it's likely that they will be incorporated in the middle of the training process.

Let's see how it goes.

https://blog.lczero.org/

[Alex_Zet]

Deep Chess (2018) - https://arxiv.org/abs/1711.09667

[Alex_Zet]

Winter

Jonathan Rosenthal
rosenthj

As of Winter 0.6.2, the evaluation function relies on a small neural network for more precise evaluations.

https://github.com/rosenthj/Winter

[Alex_Zet]

LCZero

(Do Not Use - incorrectly tagged) v0.21.5-rc1

Remove softmax calculation from backends and apply it after filtering for illegal moves to ensure spurious outputs on illegal moves don't reduce (or entirely remove) the quality of the policy values on the legal moves. This was especially noticeable on fp16 backends for nets trained with legal move masking, but could theoretically be an improvement for any net.

https://github.com/LeelaChessZero/lc0/releases

[Alex_Zet]

LCZero

v0.22.0-rc1

Bunch of small changes that piled up from last major release.

https://github.com/LeelaChessZero/lc0/releases

[Alex_Zet]

LCZero

v0.22.0

lc0-v0.22.0-windows-blas.zip - https://github.com/LeelaChessZero/lc...ndows-blas.zip 8.21 MB
lc0-v0.22.0-windows-cuda.zip lc0-v0.22.0-windows-cuda.zip 251 MB 251 MB
lc0-v0.22.0-windows-opencl.zip lc0-v0.22.0-windows-opencl.zip 4.19 MB 4.19 MB

https://github.com/LeelaChessZero/lc0/releases

[svoitsl]

Элементы сладкой жизни
SugaR-NN_2019-08-13

[byan_mivv]

И как его настраивать... или все по умолчанию?

[ChessMan]

Цитата: Сообщение от byan_mivv

И как его настраивать... или все по умолчанию?

Это обычный A-B движок. Автор пытается сделать что-то типа самообучения, но оно не работает (в раубфише и в ему подобных тоже).
А я не хочу его расстраивать, поэтому молчу

[byan_mivv]

Ну ты садист Надо сказать!

[Maratka]

Цитата: Сообщение от ChessMan

Это обычный A-B движок. Автор пытается сделать что-то типа самообучения

Для a-b самообучение имеет смысл (да и возможность реализации) лишь в дебютах, так что в общем-то ничего нового, по смыслу оно ничуть не отличается от обучения книги на имеющихся партиях: чем больше выигрышей после некого хода, тем больше рейтинг этого хода в базе/книге.

Ограничения при этом опять же ровно те же: представим, что некий a-b движок с реальным рейтингом 2500-3000 играет против SF (да пусть даже и против Fire): очевидно, что практически любой ход, который он делает в итоге закончится поражением, причем будет это не потому, что сам ход в обязательном порядке плохой (на 1.e4 ответить e5 вполне себе разумно), а потому, что последующие ходы приведут к поражению в 95% случаев, что забракует всю последовательность ходов, начиная с первого.

[Alex_Zet]

Цитата: Сообщение от ChessMan

Это обычный A-B движок. Автор пытается сделать что-то типа самообучения, но оно не работает (в раубфише и в ему подобных тоже).
А я не хочу его расстраивать, поэтому молчу

А как же тогда обновление #7-26-2019: SugaR NN vs Lc0 2.1.3 video and downloadable games. на всеми уважаемом сайте https://chess.massimilianogoi.com/
Как с этим быть?

[ChessMan]

Цитата: Сообщение от Alex_Zet

А как же тогда обновление #7-26-2019: SugaR NN vs Lc0 2.1.3 video and downloadable games. на всеми уважаемом сайте https://chess.massimilianogoi.com/
Как с этим быть?

А вот так. Как я и написал - это обычный А-Б движок.

=========
Если я сделаю движок и в его имени добавлю "NN" - он будет само-обучаться? Нет.

[Alex_Zet]

UCI "шахматный движок", который сочетает в себе мощь Lc0 и Stockfish.

Встречайте py-goratschin

«Шахматный движок» UCI, который сочетает в себе мощь Lc0 и Stockfish - или любых двух других движков, которые вам нравятся.

Код сильно заимствует из проекта CombiChess. Большое спасибо Тому Фридриху за его работу!

GoratschinChess - «шахматный движок», который поддерживает шахматный протокол UCI и объединяет два движка (соответственно, «босс» и «советник») в один. Он работает, спрашивая движки, что они думают, лучший ход для данной позиции, а затем применяя некоторую логику, чтобы определить, какое движение на самом деле делать.

Правила, которые он использует, довольно просты:

Если двигатель видит помощника, то немедленно сделайте это движение, ведущее к помощнику.

Если оба двигателя дают один и тот же лучший ход, то сделайте этот ход.

если движки говорят что-то другое, и результат советника лучше, чем у босса, с полем «cp» (см. self.score_margin в коде), то ход советника должен двигаться. Маржа по умолчанию составляет 0,5 сантипавонны.

Иначе, всегда слушайте «двигатель босса».

«Горачин» - это имя двуглавого персонажа из немецкого научно-фантастического сериала «Перри Родан».

https://github.com/feldi/py-goratschin

[Alex_Zet]

Оказывается это не все. Вот ещё: -

Re: Trying out the Leela Hybrid engine (NN with AB)
Post by Jaglo » Thu May 16, 2019 7:29 am

My personal ranking of hybrids:

1) Leelafish
2) Combichess (Python)
3) Triplebrain (Shredder)
4) Chesscombi v1

killerducky, is the Leelafish project evolving?

http://talkchess.com/forum3/viewtopi...9958&start=140

[sovaz1997]

Если этот "движок" показывает результаты лучше, чем Stockfish и Lc0 по-отдельности, то это хорошо. В ином случае это просто бессмысленно. И да, это не движок. Это просто слектор ходов)) Причем он даже не проверяет зевки Лилы, судя по всему.