8-puzzle
收藏资源简介:
该数据集是一个固定的3x3/8-puzzle(八数码/滑动拼图)评估集,专为滑动拼图推理实验而设计。数据集包含45个样本,按最优解长度分为三个难度级别:easy(1-5步)、medium(6-15步)和hard(16-31步),每个级别各15个样本。每个样本包含以下字段:board(平铺的3x3棋盘,用0表示空白方块)、scramble_depth(生成该评测集时采用的确切最优解长度)、bucket(难度标签:easy/medium/hard)、split(源分割标签,固定为eval_3x3)、optimal_moves(从当前棋盘到目标状态的最优移动序列)、optimal_length(最优移动序列的长度)。注意:移动采用方块移动约定,即移动名称描述的是编号方块移至空白处,而非空白移动。该数据集适用于评估AI模型或规划算法在滑动拼图推理任务上的性能。
This dataset is a fixed 3x3/8-puzzle evaluation set designed for sliding puzzle reasoning experiments. It contains 45 samples divided into three difficulty levels based on optimal solution length: easy (1-5 moves), medium (6-15 moves), and hard (16-31 moves), with 15 samples per level. Each sample includes the following fields: board (a flattened 3x3 board with 0 representing the blank tile), scramble_depth (the exact optimal solution length used when generating the evaluation set), bucket (difficulty label: easy/medium/hard), split (source split label, fixed as eval_3x3), optimal_moves (the sequence of optimal moves from the current board to the goal state), and optimal_length (the length of the optimal move sequence). Note: Moves follow the tile-moving convention, where the move name describes the numbered tile moving to the blank space, rather than the blank moving. This dataset is suitable for evaluating the performance of AI models or planning algorithms on sliding puzzle reasoning tasks.
8-puzzle / 3x3 滑动拼图数据集概述
该数据集是一个用于滑动拼图推理实验的固定 3x3(8-puzzle)评估集。
数据集配置
- 子集/配置名称:
eval - 数据分割:
eval - 数据文件格式:Parquet(路径:
data/eval/eval.parquet)
数据字段说明
| 字段名 | 描述 |
|---|---|
board |
展平的 3x3 棋盘,其中 0 表示空白格 |
scramble_depth |
该评估集的精确最优解长度 |
bucket |
难度分桶:easy、medium 或 hard |
split |
源分割标签:eval_3x3 |
optimal_moves |
从当前棋盘到已解决状态的最优移动序列 |
optimal_length |
optimal_moves 的长度 |
难度分桶统计
| 分桶 | 最优解长度范围 | 样本数量 |
|---|---|---|
| easy(简单) | 1-5 | 15 |
| medium(中等) | 6-15 | 15 |
| hard(困难) | 16-31 | 15 |
移动规则说明
移动采用“棋子移动”约定:移动名称描述的是编号棋子移动到空白格的动作,而非空白格的移动方向。





