testbed/PyThaiNLP__pythainlp/pythainlp/generate/thai2fit.py

# -*- coding: utf-8 -*-
# SPDX-FileCopyrightText: 2016-2025 PyThaiNLP Project
# SPDX-FileType: SOURCE
# SPDX-License-Identifier: Apache-2.0
"""
Thai2fit: Thai Wikipeida Language Model for Text Generation

Codes are from
https://github.com/PyThaiNLP/tutorials/blob/master/source/notebooks/text_generation.ipynb
"""

__all__ = ["gen_sentence"]

import pickle
import random
from typing import List, Union

# fastai
import fastai
import pandas as pd
from fastai.text import *

# pythainlp
from pythainlp.ulmfit import (
    THWIKI_LSTM,
    ThaiTokenizer,
    post_rules_th,
    pre_rules_th,
)

# get dummy data
imdb = untar_data(URLs.IMDB_SAMPLE)
dummy_df = pd.read_csv(imdb / "texts.csv")

# get vocab
thwiki = THWIKI_LSTM

thwiki_itos = pickle.load(open(thwiki["itos_fname"], "rb"))
thwiki_vocab = fastai.text.transform.Vocab(thwiki_itos)

# dummy databunch
tt = Tokenizer(
    tok_func=ThaiTokenizer,
    lang="th",
    pre_rules=pre_rules_th,
    post_rules=post_rules_th,
)
processor = [
    TokenizeProcessor(tokenizer=tt, chunksize=10000, mark_fields=False),
    NumericalizeProcessor(vocab=thwiki_vocab, max_vocab=60000, min_freq=3),
]
data_lm = (
    TextList.from_df(dummy_df, imdb, cols=["text"], processor=processor)
    .split_by_rand_pct(0.2)
    .label_for_lm()
    .databunch(bs=64)
)


data_lm.sanity_check()

config = {
    "emb_sz": 400,
    "n_hid": 1550,
    "n_layers": 4,
    "pad_token": 1,
    "qrnn": False,
    "tie_weights": True,
    "out_bias": True,
    "output_p": 0.25,
    "hidden_p": 0.1,
    "input_p": 0.2,
    "embed_p": 0.02,
    "weight_p": 0.15,
}
trn_args = {"drop_mult": 0.9, "clip": 0.12, "alpha": 2, "beta": 1}

learn = language_model_learner(
    data_lm, AWD_LSTM, config=config, pretrained=False, **trn_args
)

# load pretrained models
learn.load_pretrained(**thwiki)


def gen_sentence(
    start_seq: str = "",
    N: int = 4,
    prob: float = 0.001,
    output_str: bool = True,
) -> Union[List[str], str]:
    """
    Text generator using Thai2fit

    :param str start_seq: word to begin sentence with
    :param int N: number of words
    :param bool output_str: output as string
    :param bool duplicate: allow duplicate words in sentence

    :return: list words or str words
    :rtype: List[str], str

    :Example:
    ::

      from pythainlp.generate.thai2fit import gen_sentence

      gen_sentence()
      # output: 'แคทรียา อิงลิช  (นักแสดง'

      gen_sentence("แมว")
      # output: 'แมว คุณหลวง '
    """
    if not start_seq:
        start_seq = random.choice(list(thwiki_itos))
    list_word = learn.predict(
        start_seq, N, temperature=0.8, min_p=prob, sep="-*-"
    ).split("-*-")

    if output_str:
        return "".join(list_word)

    return list_word