MATH-lighteval.py

"""Mathematics Aptitude Test of Heuristics (MATH) dataset, lighteval format with correct builder configs."""

import json
import os


from datasets import load_dataset, Dataset, DatasetDict, GeneratorBasedBuilder, BuilderConfig, DatasetInfo, Value, Features, Split, SplitGenerator, Version


_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}
}
"""


_DESCRIPTION = """\
The Mathematics Aptitude Test of Heuristics (MATH) dataset consists of problems
from mathematics competitions, including the AMC 10, AMC 12, AIME, and more.
Each problem in MATH has a full step-by-step solution, which can be used to teach
models to generate answer derivations and explanations. This version of the dataset
includes appropriate builder configs s.t. it can be used as a drop-in replacement
for the now missing lighteval/MATH dataset.
"""


_HOMEPAGE = "https://github.com/hendrycks/math"


_LICENSE = "https://github.com/hendrycks/math/blob/main/LICENSE"


# Original data URL: "https://people.eecs.berkeley.edu/~hendrycks/MATH.tar"
_URL = "data/MATH.zip"


class FilteredTypeConfig(BuilderConfig):
    def __init__(self, type_value, type_name, **kwargs):
        super().__init__(**kwargs)
        self.type_value = type_value
        self.type_name = type_name

class FilteredTypeDatasetBuilder(GeneratorBasedBuilder):
    """Mathematics Aptitude Test of Heuristics (MATH) dataset."""

    VERSION = Version("1.0.0")

    BUILDER_CONFIGS = [FilteredTypeConfig(
            name="default",
            version="1.0.0",
            description=f"default builder config",
            type_name="default", # for builder config
            type_value="default", # in original dataset
        )] + [
        FilteredTypeConfig(
            name=type_name,
            version="1.0.0",
            description=f"Dataset filtered by type: {type_value}",
            type_name=type_name, # for builder config
            type_value=type_value, # in original dataset
        )
        for type_name, type_value in [("algebra", "Algebra"), ("counting_and_probability", "Counting & Probability"), ("geometry", "Geometry"), ("intermediate_algebra", "Intermediate Algebra"), ("number_theory", "Number Theory"), ("prealgebra", "Prealgebra"), ("precalculus", "Precalculus")]
    ]

    def _info(self):
        return DatasetInfo(
            description=_DESCRIPTION,
            features=Features({
                "problem": Value("string"),
                "level": Value("string"),
                "solution": Value("string"),
                "type": Value("string"),
            }),
            supervised_keys=None,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        download_dir = dl_manager.download_and_extract(_URL)
        return [
            SplitGenerator(
                name=Split.TRAIN,
                gen_kwargs={"data_dir": dl_manager.iter_files(os.path.join(download_dir, "MATH", "train"))},
            ),
            SplitGenerator(
                name=Split.TEST,
                gen_kwargs={"data_dir": dl_manager.iter_files(os.path.join(download_dir, "MATH", "test"))},
            ),
        ]

    def _generate_examples(self, data_dir):
        type_value = self.config.type_value  # Access the type value for the current config

        """Yields examples as (key, example) tuples. Filters by type if appropriate builder config is given."""
        for id_, filepath in enumerate(data_dir):
            with open(filepath, "rb") as fin:
                example = json.load(fin)
                if type_value == "default" or example["type"] == type_value:
                    yield id_, example