generate_dataset.py

from datasets import load_dataset
from underthesea import word_tokenize
from underthesea.pipeline.word_tokenize.regex_tokenize import tokenize
from os.path import dirname, join
from underthesea.utils import logger

dataset = load_dataset("undertheseanlp/UTS_Text_v1")

sentences = dataset["train"]["text"]
pwd = dirname(__file__)
train_file = join(pwd, "data/train.txt")
dev_file = join(pwd, "data/dev.txt")
test_file = join(pwd, "data/test.txt")
# write empty files
for file in [train_file, dev_file, test_file]:
    with open(file, "w") as f:
        f.write("")

f = open(train_file, "a")
content = ""
for j, s in enumerate(sentences):
    if j == 8000:
        f.close()
        f = open(dev_file, "a")
    if j == 9000:
        f.close()
        f = open(test_file, "a")
    words = word_tokenize(s)
    for word in words:
        tokens = tokenize(word)
        for i, token in enumerate(tokens):
            if i == 0:
                tag = "B-W"
            else:
                tag = "I-W"
            content += token + "\t" + tag + "\n"
    content += "\n"
    if j % 1000 == 999:
        f.write(content)
        content = ""
        logger.info(j)

f.close()