process.py

import json
import os
import gzip
import requests

import pandas as pd

urls = {
    'dev1': 'https://home.ttic.edu/~kgimpel/comsense_resources/dev1.txt.gz',
    'dev2': 'https://home.ttic.edu/~kgimpel/comsense_resources/dev2.txt.gz',
    'test': 'https://home.ttic.edu/~kgimpel/comsense_resources/test.txt.gz',
    'train': "https://home.ttic.edu/~kgimpel/comsense_resources/train600k.txt.gz"
}
os.makedirs("dataset", exist_ok=True)


def wget(url, cache_dir: str = './cache'):
    """ wget and uncompress data_iterator """
    os.makedirs(cache_dir, exist_ok=True)
    filename = os.path.basename(url)
    path = f'{cache_dir}/{filename}'
    if os.path.exists(path):
        return path.replace('.gz', '')
    with open(path, "wb") as f_:
        r = requests.get(url)
        f_.write(r.content)
    with gzip.open(path, 'rb') as f_:
        with open(path.replace('.gz', ''), 'wb') as f_write:
            f_write.write(f_.read())
    os.remove(path)
    return path.replace('.gz', '')


def read_file(file_name):
    with open(file_name) as f_reader:
        df = pd.DataFrame([i.split('\t') for i in f_reader.read().split('\n') if len(i) > 0], columns=['relation', 'head', 'tail', 'flag'])
    df = df[[not i.startswith("Not") for i in df.relation]]
    df_positive = df[df['flag'] != '0']
    df_negative = df[df['flag'] == '0']    
    df_positive.pop('flag')
    df_negative.pop('flag')
    return df_positive, df_negative


if __name__ == '__main__':
    test_p, test_n = read_file(wget(urls['test']))
    dev1_p, dev1_n = read_file(wget(urls['dev1']))
    dev2_p, dev2_n = read_file(wget(urls['dev2']))
    train_p, _ = read_file(wget(urls['train']))

    with open(f'dataset/test.jsonl', 'w') as f:
        for relation, df_p in test_p.groupby('relation'):
            if len(df_p) < 2:
                continue
            df_n = test_n[test_n['relation'] == relation]
            f.write(json.dumps({
                'relation_type': relation,
                'positives': df_p[['head', 'tail']].to_numpy().tolist(),
                'negatives': df_n[['head', 'tail']].to_numpy().tolist()
            }) + '\n')

    with open(f'dataset/train.jsonl', 'w') as f:
        for relation, df_p in train_p.groupby('relation'):
            if len(df_p) < 2:
                continue
            f.write(json.dumps({
                'relation_type': relation,
                'positives': df_p[['head', 'tail']].to_numpy().tolist(),
                'negatives': []
            }) + '\n')

    with open(f'dataset/valid.jsonl', 'w') as f:
        for relation, df_p in dev1_p.groupby('relation'):
            if len(df_p) < 2:
                continue
            df_n = dev1_n[dev1_n['relation'] == relation]
            f.write(json.dumps({
                'relation_type': relation,
                'positives': df_p[['head', 'tail']].to_numpy().tolist(),
                'negatives': df_n[['head', 'tail']].to_numpy().tolist()
            }) + '\n')
        for relation, df_p in dev2_p.groupby('relation'):
            if len(df_p) < 2:
                continue
            df_n = dev2_n[dev2_n['relation'] == relation]
            f.write(json.dumps({
                'relation_type': relation,
                'positives': df_p[['head', 'tail']].to_numpy().tolist(),
                'negatives': df_n[['head', 'tail']].to_numpy().tolist()
            }) + '\n')