Andscacs,
an UCI compliant chess engine by Daniel José Queraltó. Its development started in September 2013, and the first release was published in February 2014 [1]. Andscacs soon evolved to a top engine, becoming runner-up at the IGWT III losing the final from Chiron with the narrowest margin of 4½ - 5½, but taking revenge one year later at the IGWT IV, this time with 6½ - 5½ to its favour. At the Tenth Annual ACCA World Computer Rapid Chess Championship 2016, Andscacs emphasized its ambitions, and became runner-up without losing a single game, but 12 wins and 3 draws, and thus the only program preventing later winner Komodo from a 100% score.
The about 200 evaluation features were tuned with 750.000 positions to minimize the standard deviation of Stockfish's and Andscacs' static evaluation, which initially added a nice Elo boost to Andscacs [10], without playing too similar.
an UCI compliant chess engine by Daniel José Queraltó. Its development started in September 2013, and the first release was published in February 2014 [1]. Andscacs soon evolved to a top engine, becoming runner-up at the IGWT III losing the final from Chiron with the narrowest margin of 4½ - 5½, but taking revenge one year later at the IGWT IV, this time with 6½ - 5½ to its favour. At the Tenth Annual ACCA World Computer Rapid Chess Championship 2016, Andscacs emphasized its ambitions, and became runner-up without losing a single game, but 12 wins and 3 draws, and thus the only program preventing later winner Komodo from a 100% score.
Table of Contents
Description
Being a bitboard engine, Andscacs determines sliding piece attacks ray-wise, quite similar to the classical approach [3]. Since version 0.70, Andscacs uses magic bitboards to speed the attack calculations [4].Search
Andscacs applies a principal variation search with tranposition table inside an iterative deepening framework with aspiration windows [5]. Beside quiescence, selectivity is due to adaptive null move pruning, late move reductions, check extensions, singular extensions, razoring and futility pruning. Move ordering is further improved by the killer heuristic, IID and static exchange evaluation, the latter also used to prune bad captures in quiescence and futile moves near the horizon. Since version 0.70, a hash move is tried in quiescence search even if it’s a quiet move. Trying only captures and promotions was clearly weaker [6]. Version 0.80 in Apil 2015 introduced a threaded parallel search [7], a Lazy SMP approach [8].Evaluation
The evaluation considers material imbalances, and has piece-square tables for middlegame and endgame, piece values adapted by type of positions (open, closed), different types of piece mobility, king safety and sophisticated pawn structure evaluation with focus on passed pawns, most features smoothed out between game phases by a tapered eval. Since version 0.62u, evaluation scores are cached with a small evaluation hash table [9].Tuning
The about 200 evaluation features were tuned with 750.000 positions to minimize the standard deviation of Stockfish's and Andscacs' static evaluation, which initially added a nice Elo boost to Andscacs [10], without playing too similar.Forum Posts
2014
2015
Andscacs - New version 0.81 - Detailed changes by Daniel José Queraltó, CCC, June 20, 2015
2016
2017
Re: Andscacs - New version 0.921 with source by Jon Dart, CCC, November 09, 2017 » TCEC Season 10
Re: Andscacs - New version 0.921 with source by Ronald de Man, CCC, November 11, 2017
2018
External Links
Chess Engine
Misc
Claudio Constantini, Alejandro Di Costanzo, Hernán Hock, Jorge Retamoza
References
What links here?
Up one level