Difference between revisions of "Bonanza"

From Chessprogramming wiki
Jump to: navigation, search
Line 16: Line 16:
 
a [[Supervised Learning|supervised]] [[Automated Tuning|tuning method]] based on [[Automated Tuning#MoveAdaption|move adaptation]],  
 
a [[Supervised Learning|supervised]] [[Automated Tuning|tuning method]] based on [[Automated Tuning#MoveAdaption|move adaptation]],  
 
dubbed the '''Bonanza Method''' which evolved to [[Minimax Tree Optimization]] (MMTO).  
 
dubbed the '''Bonanza Method''' which evolved to [[Minimax Tree Optimization]] (MMTO).  
Bonanza utilizes and tuned [[Piece-Square Tables|piece-square tables]] indexed by king location and further two-piece locations and side to move (turn), dubbed '''KPP''' or '''KPPT''', which was used in many other Shogi programs <ref>[https://groups.google.com/d/msg/shogi-l/c4-dY44P8Mw/M3z-RtFR-tsJ The 25th World Computer Shogi Championships] by [[Reijer Grimbergen]] on behalf of [[Takenobu Takizawa]], [[Computer Chess Forums|SHOGI-L]], February 11, 2015</ref>, and has influenced the design of [[NNUE]] <ref>[http://yaneuraou.yaneu.com/2020/05/03/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92%E3%82%A8%E3%83%B3%E3%82%B8%E3%83%8B%E3%82%A2%E3%81%AE%E3%81%9F%E3%82%81%E3%81%AE%E5%B0%86%E6%A3%8Bai%E9%96%8B%E7%99%BA%E5%85%A5%E9%96%80%E3%81%9D%E3%81%AE1/ 機械学習エンジニアのための将棋AI開発入門その1 Introduction to Shogi AI development for machine learning engineers Part 1], May 03, 2020 (Japanese)</ref>.  
+
Bonanza utilizes and tuned [[Piece-Square Tables|piece-square tables]] indexed by king location and further two-piece locations and side to move (turn), dubbed '''KPP''', '''KKP''' or '''KPPT''', which was used in many other Shogi programs <ref>[https://groups.google.com/d/msg/shogi-l/c4-dY44P8Mw/M3z-RtFR-tsJ The 25th World Computer Shogi Championships] by [[Reijer Grimbergen]] on behalf of [[Takenobu Takizawa]], [[Computer Chess Forums|SHOGI-L]], February 11, 2015</ref>, and has influenced the design of [[NNUE]] <ref>[http://yaneuraou.yaneu.com/2020/05/03/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92%E3%82%A8%E3%83%B3%E3%82%B8%E3%83%8B%E3%82%A2%E3%81%AE%E3%81%9F%E3%82%81%E3%81%AE%E5%B0%86%E6%A3%8Bai%E9%96%8B%E7%99%BA%E5%85%A5%E9%96%80%E3%81%9D%E3%81%AE1/ 機械学習エンジニアのための将棋AI開発入門その1 Introduction to Shogi AI development for machine learning engineers Part 1], May 03, 2020 (Japanese)</ref>.  
  
 
Beside a [[Parallel Search|parallel tree search]], Bonanza is able to apply parallelization by the so called '''Consensus''' method  
 
Beside a [[Parallel Search|parallel tree search]], Bonanza is able to apply parallelization by the so called '''Consensus''' method  

Revision as of 21:07, 4 October 2020

Home * Engines * Bonanza

Bonanza team performs distributed tree search [1]

Bonanza, (Bonanza Feliz)
an XBoard compliant Category:Open Source Shogi engine developed by primary author Kunihito Hoki, started in 2004, at times supported by Takuya Obata, Takuya Sugiyama, and Takeshi Ito [2]. Bonanza is top contender at Computer Olympiads and World Computer Shogi Championships and so far two times champion, winning the WCSC16 in 2006, and the WCSC23 in 2013 [3].

Description

Bonanza is written in C and utilizes 9x9 Bitboards in form of three 32-bit unsigned integers. It is a fractional ply alpha-beta engine performing a principal variation search with transposition table, null move pruning, recursive iterative-deepening for PV-nodes, various extensions, reductions, and futility pruning. Bonanza pioneered in large-scale machine learning of static evaluation functions, a supervised tuning method based on move adaptation, dubbed the Bonanza Method which evolved to Minimax Tree Optimization (MMTO). Bonanza utilizes and tuned piece-square tables indexed by king location and further two-piece locations and side to move (turn), dubbed KPP, KKP or KPPT, which was used in many other Shogi programs [4], and has influenced the design of NNUE [5].

Beside a parallel tree search, Bonanza is able to apply parallelization by the so called Consensus method [6], a kind of triple-brain approach where multiple, slightly modified instances of the same engine vote for the best move [7] [8].

Publications

Forum Posts

External Links

Shogi Engine

Misc

References

Up one Level