Hendrycks MATH Dataset

Dataset Description

The MATH dataset is a collection of mathematics competition problems designed to evaluate mathematical reasoning and problem-solving capabilities in computational systems. Containing 12,500 high school competition-level mathematics problems, this dataset is notable for including detailed step-by-step solutions alongside each problem.

Dataset Summary

The dataset consists of mathematics problems spanning multiple difficulty levels (1-5) and various mathematical subjects including:

• Prealgebra
• Algebra
• Number Theory
• Counting and Probability
• Geometry
• Intermediate Algebra
• Precalculus

Each problem comes with:

• A complete problem statement
• A step-by-step solution
• A final answer
• Difficulty rating
• Subject classification

Data Split

The dataset is divided into:

• Training set: 12,000
• Test set: 500 problems

Dataset Creation

Citation

@article{hendrycksmath2021,
    title={Measuring Mathematical Problem Solving With the MATH Dataset},
    author={Dan Hendrycks
    and Collin Burns
    and Saurav Kadavath
    and Akul Arora
    and Steven Basart
    and Eric Tang
    and Dawn Song
    and Jacob Steinhardt},
    journal={arXiv preprint arXiv:2103.03874},
    year={2021}
}

Source Data

The problems originate from high school mathematics competitions, including competitions like the AMC 10, AMC 12, and AIME. These represent carefully curated, high-quality mathematical problems that test conceptual understanding and problem-solving abilities rather than just computational skills.

Annotations

Each problem includes:

• Complete problem text in LaTeX format
• Detailed solution steps
• Final answer in a standardized format
• Subject category
• Difficulty level (1-5)

Papers and References

For detailed information about the dataset and its evaluation, refer to "Measuring Mathematical Problem Solving With the MATH Dataset" presented at NeurIPS 2021.

https://arxiv.org/pdf/2103.03874