sapientinc/sudoku-extreme

---
task_categories:
- question-answering
---
# Hardest Sudoku Puzzle Dataset V2

This dataset contains a mixture of easy and very hard Sudoku puzzles collected from the Sudoku community.

## Dataset Composition

### Sources

- [tdoku benchmarks](https://github.com/t-dillon/tdoku/blob/master/benchmarks/README.md#benchmarked-data-sets)
- [enjoysudoku](http://forum.enjoysudoku.com/the-hardest-sudokus-new-thread-t6539-600.html#p277835)

### Easy Puzzles (1.1M)

- puzzles0_kaggle
- puzzles1_unbiased
- puzzles2_17_clue

### Hard Puzzles (3.1M)

- puzzles3_magictour_top1465
- puzzles4_forum_hardest_1905
- puzzles6_forum_hardest_1106
- ph_2010/01_file1.txt

## Dataset Characteristics

- All puzzles have been exact-deduped and randomly permuted by row, column, box, and digit.
- Each puzzle is guaranteed to have a unique solution.
- Puzzles in the train set are [mathematically inequivalent](http://sudopedia.enjoysudoku.com/Mathematically_equivalent.html) to those in the test set.

## Dataset Structure

- Train set: `train.csv` (3.8M examples)
- Test set: `test.csv` (423k examples)

Puzzles and solutions are flattened in row-major order. Rating is evaluated by number of backtracks needed by [tdoku solver]((https://github.com/t-dillon/tdoku) required to solve the puzzle (higher is harder).

## Usage Guidelines

1. Train models using only the train set.
2. Evaluate models on the test set using exact accuracy (all numbers must be correct).

---
任务类别：
- 问答（Question Answering）
---
# 最难数独（Sudoku）谜题数据集V2

本数据集包含从数独社区收集的简易与极难数独谜题的混合集合。

## 数据集构成

### 来源
- [tdoku基准测试集](https://github.com/t-dillon/tdoku/blob/master/benchmarks/README.md#benchmarked-data-sets)
- [enjoysudoku论坛](http://forum.enjoysudoku.com/the-hardest-sudokus-new-thread-t6539-600.html#p277835)

### 简易谜题（110万条）
- puzzles0_kaggle
- puzzles1_unbiased
- puzzles2_17_clue

### 困难谜题（310万条）
- puzzles3_magictour_top1465
- puzzles4_forum_hardest_1905
- puzzles6_forum_hardest_1106
- ph_2010/01_file1.txt

## 数据集特征
- 所有数独谜题均经过精确去重，并通过行、列、宫及数字的随机置换进行了处理。
- 每个谜题均保证存在唯一解。
- 训练集与测试集中的数独谜题在数学上非等价（参考：http://sudopedia.enjoysudoku.com/Mathematically_equivalent.html）。

## 数据集结构
- 训练集：`train.csv`（380万条样本）
- 测试集：`test.csv`（42.3万条样本）

数独谜题与答案均以行优先顺序扁平化存储。难度评分由[tdoku求解器](https://github.com/t-dillon/tdoku)求解该谜题所需的回溯次数决定，数值越高则难度越大。

## 使用指南
1. 仅使用训练集训练模型。
2. 在测试集上采用精确准确率评估模型性能，要求所有数字均完全正确。