中国象棋训练数据

数据为sqlite3数据库格式，每行是一个PGN格式字符串，代表一个棋局，其中"start_fen"是起始局面，"skip_step"表示棋局开始的随机步数，即最开始的几步是随机行棋，这几步的分值为0，"result"是棋局最终结果，1表示红胜，0表示和棋，-1表示黑胜，2表示棋局中存在随机招法从而结果无参考价值，"mvs"是招法列表，招法包括"mv"和"score"，mv是招法(2字节无符号整数，低字节为起始位置，高字节为目标位置)，score是走完这个招法后的局面得分。 数据总量大约有2000万个棋局。 棋盘定义如下，黑上红下 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3A, 0x3B, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 0x4A, 0x4B, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5A, 0x5B, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x6A, 0x6B, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7A, 0x7B, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, 0x8A, 0x8B, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, 0x99, 0x9A, 0x9B, 0xA3, 0xA4, 0xA5, 0xA6, 0xA7, 0xA8, 0xA9, 0xAA, 0xAB, 0xB3, 0xB4, 0xB5, 0xB6, 0xB7, 0xB8, 0xB9, 0xBA, 0xBB, 0xC3, 0xC4, 0xC5, 0xC6, 0xC7, 0xC8, 0xC9, 0xCA, 0xCBThis dataset is stored in SQLite3 database format, with each line containing a PGN-formatted string representing a chess game. It includes the following fields: "start_fen" refers to the initial starting position of the game; "skip_step" indicates the number of random initial moves, where the first several moves are played randomly and their scores are set to 0; "result" is the final outcome of the game, where 1 denotes a win for the red side, 0 denotes a draw, -1 denotes a win for the black side, and 2 indicates that the game contains random moves thus its result is not referenceable; "mvs" is a move list, with each entry containing "mv" and "score", where "mv" is the specific chess move and "score" is the position evaluation score after executing this move. The total number of chess games in this dataset is approximately 20 million.

搜集汇总

背景与挑战

背景概述

该数据集是一个大规模的中国象棋训练数据集合，包含约2000万个棋局，以sqlite3数据库格式存储，每个棋局以PGN字符串表示，并详细记录了起始局面、随机步数、最终结果和招法得分。数据适用于象棋AI模型训练，支持局面分析和招法评估，具有Apache License 2.0开源许可，便于研究和开发使用。