stsb_multi_mt_extend.py

# coding=utf-8

"""STS Benchmark Multilingual dataset"""

import json
import os
import gzip

import datasets

logger = datasets.logging.get_logger(__name__)

_DESCRIPTION = """\
        STS Benchmark Multilingual dataset by deepl and google (only ar)
"""

_LANGUAGES = ["de", "en", "es", "fr", "it", "nl", "pl", "pt", "ru", "zh"]

_NEW_LANGUAGES = ['ar', 'id']


class STSBMultilingual(datasets.GeneratorBasedBuilder):
    """STS Benchmark Multilingual"""

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name=name,
            version=datasets.Version("1.0.0"),
            description=f"The STS Benchmark for {name} language{'' if name == 'en' else ' by deepl'}.",
        )
        for name in _LANGUAGES
    ] + [
        datasets.BuilderConfig(
            name='id',
            version=datasets.Version("1.0.0"),
            description=f"The STS Benchmark for id language by deepl.",
        ),
        datasets.BuilderConfig(
            name='ar',
            version=datasets.Version("1.0.0"),
            description=f"The STS Benchmark for id language by google translate.",
        )
    ]

    DEFAULT_CONFIG_NAME = 'en'

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "sentence1": datasets.Value("string"),
                    "sentence2": datasets.Value("string"),
                    "score": datasets.Value("float32"),
                },
            ),
            supervised_keys=None,
            homepage="https://github.com/PhilipMay/stsb-multi-mt",
        )

    def _split_generators(self, dl_manager):
        if self.config.name == 'ar':
            path_or_ds = dl_manager.download('test_ar_google.jsonl')
        elif self.config.name == 'id':
            path_or_ds = dl_manager.download('test_id_deepl.jsonl')
        else:
            path_or_ds = datasets.load_dataset("stsb_multi_mt", self.config.name, split="test")
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TEST, gen_kwargs={"path_or_ds": path_or_ds},
            ),
        ]

    def _generate_examples(self, path_or_ds):
        """Yields examples."""
        if isinstance(path_or_ds, datasets.Dataset):
            for i, ins in enumerate(path_or_ds):
                yield i, {'sentence1': ins['sentence1'], 'sentence2': ins['sentence2'], 'score': ins['similarity_score']}
        else:
            with open(path_or_ds) as f:
                for i, line in enumerate(f):
                    yield i, json.loads(line)