ethanjtang/PAWN-piece-value-datasets

PAWN（Piece Value Analysis with Neural Networks）数据集是一个用于棋子价值分析的象棋数据集，专注于通过神经网络进行研究。数据集定义了棋子价值为使用Stockfish引擎（版本17，深度20，每评估超时300秒）计算的原始棋盘位置与移除特定棋子后位置之间的评估差异（以百分之一兵为单位）。数据集包含两个主要子集：MC-Large子集源自6,925盘Magnus Carlsen（国际象棋世界冠军）的对局，包含11,673,269个棋子价值条目，覆盖549,410个独特棋盘位置；TF子集源自7,656盘GM级别古典对局（双方玩家FIDE古典Elo评分均≥2500，对局时间为2023年），包含12,263,049个棋子价值条目，覆盖533,540个独特位置。数据列包括游戏ID、FEN字符串、移动编号、走子方、ECO代码/开局名称、双方棋子材质字符串、棋子类型（大写表示白方，小写表示黑方）、行列索引（0-7）、原始评估值、移除棋子后的评估值以及计算得到的棋子价值。该数据集旨在支持象棋引擎分析、棋子价值建模和机器学习应用。

PAWN (Piece Value Analysis with Neural Networks) is a chess dataset for piece value analysis, focusing on research with neural networks. The dataset defines piece value as the difference in Stockfish evaluation (using Stockfish 17 at depth=20 with a timeout=300s per evaluation) between the original position and the position with that piece removed, measured in centipawns. It consists of two main subsets: the MC-Large subset derived from 6,925 Magnus Carlsen games, containing 11,673,269 piece value entries across 549,410 unique positions, and the TF subset derived from 7,656 GM-level Classical games (both players ≥2500 FIDE Classical Elo, played in 2023), containing 12,263,049 piece value entries across 533,540 unique positions. Columns include game_id, fen, move_number, side_to_move, eco_code/opening, white_material, black_material, piece_type, rank, file, original_eval, eval_without_piece, and piece_value. The dataset is designed to support chess engine analysis, piece value modeling, and machine learning applications.