from __future__ import annotations

from mteb.abstasks import AbsTaskClassification
from mteb.abstasks.TaskMetadata import TaskMetadata


class NepaliNewsClassification(AbsTaskClassification):
    metadata = TaskMetadata(
        name="NepaliNewsClassification",
        description="A Nepali dataset for 7500 news articles ",
        reference="https://github.com/goru001/nlp-for-nepali",
        dataset={
            "path": "bpHigh/iNLTK_Nepali_News_Dataset",
            "revision": "79125f20d858a08f71ec4923169a6545221725c4",
        },
        type="Classification",
        category="s2s",
        date=("2019-01-01", "2020-01-01"),
        eval_splits=["train"],
        eval_langs=["nep-Deva"],
        main_score="accuracy",
        form=["written"],
        domains=["News"],
        task_subtypes=["Topic classification"],
        license="CC BY-SA 4.0",
        socioeconomic_status="mixed",
        annotations_creators="derived",
        dialect=[],
        text_creation="found",
        bibtex_citation="""
        @inproceedings{arora-2020-inltk,
    title = "i{NLTK}: Natural Language Toolkit for Indic Languages",
    author = "Arora, Gaurav",
    editor = "Park, Eunjeong L.  and
      Hagiwara, Masato  and
      Milajevs, Dmitrijs  and
      Liu, Nelson F.  and
      Chauhan, Geeticka  and
      Tan, Liling",
    booktitle = "Proceedings of Second Workshop for NLP Open Source Software (NLP-OSS)",
    month = nov,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2020.nlposs-1.10",
    doi = "10.18653/v1/2020.nlposs-1.10",
    pages = "66--71",
    abstract = "We present iNLTK, an open-source NLP library consisting of pre-trained language models and out-of-the-box support for Data Augmentation, Textual Similarity, Sentence Embeddings, Word Embeddings, Tokenization and Text Generation in 13 Indic Languages. By using pre-trained models from iNLTK for text classification on publicly available datasets, we significantly outperform previously reported results. On these datasets, we also show that by using pre-trained models and data augmentation from iNLTK, we can achieve more than 95{\%} of the previous best performance by using less than 10{\%} of the training data. iNLTK is already being widely used by the community and has 40,000+ downloads, 600+ stars and 100+ forks on GitHub.",
    }
    """,
        n_samples={"train": 5975, "test": 1495},
        avg_character_length={"train": 196.61, "test": 196.017},
    )

    def dataset_transform(self):
        self.dataset = self.dataset.rename_column("paras", "text")
        self.dataset = self.stratified_subsampling(
            self.dataset, seed=self.seed, splits=["train"]
        )