import random
import pandas as pd
from tqdm import tqdm
from sklearn.model_selection import train_test_split

df = pd.read_parquet('gemago2_dataset.parquet')

texts = []
for _, row in tqdm(df.iterrows(), desc="row", leave=False, total=len(df)):
    texts.append(rf"<kor>{row['korean']}</kor>\n\n<{row['language']}>{row['target']}</{row['language']}>")
    texts.append(rf"<{row['language']}>{row['target']}</{row['language']}>\n\n<kor>{row['korean']}</kor>")
del df

train_texts, test_texts = train_test_split(texts, test_size=0.2, random_state=42)
del texts

random.shuffle(test_texts)
with open("test.txt", "w", encoding="UTF-8") as f:
    f.write("\n".join(test_texts))
del test_texts

random.shuffle(train_texts)
with open("train.txt", "w", encoding="UTF-8") as f:
    f.write("\n".join(train_texts))
del train_texts

# gemago2_dataset_final