from datasets import load_dataset
import numpy as np
import unicodedata

dataset = load_dataset("Fizzarolli/wattpad2", "default", split="train")

from lingua import LanguageDetectorBuilder

detector = LanguageDetectorBuilder.from_all_languages().with_preloaded_language_models().build()

def fix_language_column(example):
    res = detector.compute_language_confidence_values(unicodedata.normalize('NFKD', example["description"]))
    example["language"] = res[0].language.iso_code_639_3.name
    confidences = list(map(lambda x: x.value, res))
    example["language_confidence"] = np.max(confidences)
    return example

dataset = dataset.map(fix_language_column)

dataset.push_to_hub("Fizzarolli/wattpad2", "default")