Prefix Sums, Mazes, Chess Puzzles

本研究介绍了三个数据集：Prefix Sums、Mazes和Chess Puzzles，旨在促进从简单训练数据到复杂测试例子的系统泛化研究。Prefix Sums数据集包含52种不同长度的二进制字符串及其累积和模2的结果，共520,000条数据。Mazes数据集包含迷宫图像及其解决方案，而Chess Puzzles数据集则提供国际象棋中局面的最佳下一步。这些数据集通过Python包提供，便于训练和测试，适用于机器学习和人工智能领域，特别是解决模型泛化能力的问题。

This study introduces three datasets: Prefix Sums, Mazes, and Chess Puzzles, aimed at advancing research on systematic generalization from simple training data to complex test examples. The Prefix Sums dataset comprises binary strings across 52 distinct lengths and their cumulative sums modulo 2, with a total of 520,000 data instances. The Mazes dataset contains maze images and their corresponding solutions, while the Chess Puzzles dataset provides optimal next moves for chess positions. All these datasets are made available via a Python package to facilitate model training and evaluation, and are suitable for the machine learning and artificial intelligence fields, particularly for addressing challenges related to model generalization ability.