math_qa

Stay organized with collections Save and categorize content based on your preferences.

  • Description:

A large-scale dataset of math word problems and an interpretable neural math problem solver that learns to map problems to operation programs.

Split Examples
'test' 2,985
'train' 29,837
'validation' 4,475
  • Feature structure:
FeaturesDict({
    'Problem': Text(shape=(), dtype=tf.string),
    'Rationale': Text(shape=(), dtype=tf.string),
    'annotated_formula': Text(shape=(), dtype=tf.string),
    'category': Text(shape=(), dtype=tf.string),
    'correct': Text(shape=(), dtype=tf.string),
    'correct_option': Text(shape=(), dtype=tf.string),
    'linear_formula': Text(shape=(), dtype=tf.string),
    'options': Text(shape=(), dtype=tf.string),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
Problem Text tf.string
Rationale Text tf.string
annotated_formula Text tf.string
category Text tf.string
correct Text tf.string
correct_option Text tf.string
linear_formula Text tf.string
options Text tf.string
  • Citation:
@misc{amini2019mathqa,
      title={MathQA: Towards Interpretable Math Word Problem Solving with Operation-Based Formalisms},
      author={Aida Amini and Saadia Gabriel and Peter Lin and Rik Koncel-Kedziorski and Yejin Choi and Hannaneh Hajishirzi},
      year={2019},
      eprint={1905.13319},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}