langid/create_langid_trainset.py

import pandas as pd
import xmltodict
from sklearn.model_selection import train_test_split
import glob
import sys
import os

filelist = glob.glob('sentences/*.txt')

data = pd.DataFrame()

for tsvfile in filelist:
    print(f"Processing {tsvfile}")
    data = pd.read_csv(tsvfile, sep='\t',on_bad_lines='skip',engine='python',encoding='utf8')
    lang=tsvfile.split('/')[1][0:3]
    if len(data.columns)==1:
        data.insert(0,'id','')
    
    data.columns=['id','source']
    data['target']=lang
    
    data['source'] = "lang: "+data['source']
    data['source'] = data['source'].str.replace('\t',' ')   
    data = data.sample(frac=1).reset_index(drop=True)
    
    data = data[['source','target']]

    # Train - test - dev
    train, test = train_test_split(data, test_size=0.2)
    test, dev = train_test_split(test, test_size=0.5)

    # Write the datasets to disk
    train.to_csv('langid_datafiles/'+lang+'_train.tsv', index=False, header=False, sep='\t')
    test.to_csv('langid_datafiles/'+lang+'_test.tsv', index=False, header=False, sep='\t')
    dev.to_csv('langid_datafiles/'+lang+'_dev.tsv', index=False, header=False, sep='\t')


print("Finished")