msvamp.py

import csv
import datasets

_DESCRIPTION = """\
Multilingual Simple Variations on Arithmetic Math word Problems (MSVAMP).

The same 1000 problems from [MSVAMP](https://arxiv.org/pdf/2310.20246) are each translated via human annotators in Italian.

You can find the input and targets in Italian and English as `.csv` files.
"""

_HOMEPAGE = "https://github.com/lranaldii/italian_arithmetic_reasoning"

_LICENSE = "CC BY SA 4.0"


_BASE_URL = "msvamp_{lang}.tsv"

_LANG = ["en", "it"]



class MSVAMP(datasets.GeneratorBasedBuilder):
    """MSVAMP"""

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name=lang,
            description=f"MSVAMP {lang} set",
            version=datasets.Version("1.0.0"),
        )
        for lang in _LANG
    ]

    def _info(self):
        features = datasets.Features(
            {
                "question": datasets.Value("string"),
                "answer_number": datasets.Value("int32"),
            }
        )
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        name = self.config.name

        filepaths = dl_manager.download_and_extract(
            {
                datasets.Split.TEST: _BASE_URL.format(lang=name),
            }
        )

        return [
            datasets.SplitGenerator(
                name=split,
                gen_kwargs={"filepath": path},
            )
            for split, path in filepaths.items()
        ]

    def _generate_examples(self, filepath):
        with open(filepath, encoding="utf-8") as csv_file:
            csv_reader = csv.reader(
                csv_file,
                delimiter=",",
            )
            for key, row in enumerate(csv_reader):
                yield key, {
                    "question": row[0],
                    "answer_number": int(row[1]),  
                }