#STOCKFISH CHESS MCTS PROFESSIONAL#

Furthermore, crazyhouse is commonly played in short time controls, and is increasing in popularity particularly for the online community. Moreover, the ability to drop pieces to set your opponent in check often enables long forced checkmating sequences. The chance of drawing and the average game length are significantly reduced because games almost always end in checkmate, and the element of the chess endgame is missing. The fact that pieces never fully leave the game makes crazyhouse a highly tactical game with a considerably larger branching factor than conventional chess. The element of dropping captured pieces is similar to the game of shogi, with the difference that in crazyhouse pawns can also be dropped to deliver immediate checkmate. In addition, instead of playing a classical move, the player has the option to drop any pocket piece onto an empty square of the board, with the exception that pawns cannot be dropped on the first or eighth rank. Crazyhouse incorporates all classical chess rules including castling, en passant capture, and draw by three-fold repetition. The captured piece switches its color, and is henceforth held in the capturing so-called pocket of the respective play. Crazyhouse, also known as drop chess, is the single-player version of the game bughouse and introduces the ability to re-introduce pieces that have been captured. We present the neural network based engine CrazyAra which learned to play the chess variant crazyhouse solely in a supervised fashion. In this article, we extend this family of games. The project AlphaZero ( Silver et al., 2017a) with its predecessors AlphaGoZero ( Silver et al., 2017b) and AlphaGo ( Silver et al., 2016) marks a milestone in the field of artificial intelligence, demonstrating that the board games Go, chess, and shogi can be learned from zero human knowledge. In 10 long-time control matches playing Stockfish 10, CrazyAraFish wins three games and draws one out of 10 matches. Finally, for CrazyAraFish we continue training our model on generated engine games. Furthermore, we test the playing strength of CrazyAra on CPU against all participants of the second Crazyhouse Computer Championships 2017, winning against twelve of the thirteen participants. Most notably, CrazyAra achieved a four to one win over 2017 crazyhouse world champion Justin Tan (aka LM Jann Lee) who is more than 400 Elo higher rated compared to the average player in our training set.

#STOCKFISH CHESS MCTS PROFESSIONAL#

During development, versions of CrazyAra played professional human players. After training on 569537 human games for 1.5 days we achieve a move prediction accuracy of 60.4%. These improvements include modifications in the neural network design and training configuration, the introduction of a data normalization step and a more sample efficient Monte-Carlo tree search which has a lower chance to blunder. Therefore, we focus on improving efficiency in multiple aspects while relying on low computational resources.

Crazyhouse is a game with a higher branching factor than chess and there is only limited data of lower quality available compared to AlphaGo. With this paper, we present CrazyAra which is a neural network based engine solely trained in supervised manner for the chess variant crazyhouse. Although starting from zero knowledge has been shown to yield impressive results, it is associated with high computationally costs especially for complex games.

3Department of Computer Science, JKU Linz, Linz, AustriaДeep neural networks have been successfully applied in learning the board games Go, chess, and shogi without prior knowledge by making use of reinforcement learning.

2Centre for Cognitive Science, TU Darmstadt, Darmstadt, Germany.

1Department of Computer Science, TU Darmstadt, Darmstadt, Germany.

Johannes Czech 1 *, Moritz Willig 1, Alena Beyer 1, Kristian Kersting 1,2 and Johannes FГјrnkranz 3

Stockfish chess mcts