testbed/embeddings-benchmark__mteb/mteb/tasks/Classification/multilingual/MasakhaNEWSClassification.py

from __future__ import annotations

from mteb.abstasks.TaskMetadata import TaskMetadata

from ....abstasks import AbsTaskClassification, MultilingualTask

_LANGUAGES = {
    "amh": ["amh-Ethi"],
    "eng": ["eng-Latn"],
    "fra": ["fra-Latn"],
    "hau": ["hau-Latn"],
    "ibo": ["ibo-Latn"],
    "lin": ["lin-Latn"],
    "lug": ["lug-Latn"],
    "orm": ["orm-Ethi"],
    "pcm": ["pcm-Latn"],
    "run": ["run-Latn"],
    "sna": ["sna-Latn"],
    "som": ["som-Latn"],
    "swa": ["swa-Latn"],
    "tir": ["tir-Ethi"],
    "xho": ["xho-Latn"],
    "yor": ["yor-Latn"],
}


class MasakhaNEWSClassification(AbsTaskClassification, MultilingualTask):
    metadata = TaskMetadata(
        name="MasakhaNEWSClassification",
        dataset={
            "path": "mteb/masakhanews",
            "revision": "18193f187b92da67168c655c9973a165ed9593dd",
        },
        description="MasakhaNEWS is the largest publicly available dataset for news topic classification in 16 languages widely spoken in Africa. The train/validation/test sets are available for all the 16 languages.",
        reference="https://arxiv.org/abs/2304.09972",
        category="s2s",
        type="Classification",
        eval_splits=["test"],
        eval_langs=_LANGUAGES,
        main_score="accuracy",
        date=None,
        form=None,
        domains=None,
        task_subtypes=None,
        license=None,
        socioeconomic_status=None,
        annotations_creators=None,
        dialect=None,
        text_creation=None,
        bibtex_citation=None,
        n_samples={"test": 422},
        avg_character_length={"test": 5116.6},
    )

    def dataset_transform(self):
        for lang in self.dataset.keys():
            self.dataset[lang] = self.dataset[lang].rename_columns(
                {"category": "label"}
            )