AdversarialDecoder/baseline.py at main · Isinlor/AdversarialDecoder · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
import os
import json
import pickle

import fire
import numpy as np
from scipy import sparse

from sklearn.model_selection import PredefinedSplit, GridSearchCV
from sklearn.linear_model import LogisticRegression
from sklearn.feature_extraction.text import TfidfVectorizer

def _load_split(data_dir, source, split, n=np.inf):
    path = os.path.join(data_dir, f'{source}.{split}.jsonl')
    texts = []
    for i, line in enumerate(open(path)):
        if i >= n:
            break
        texts.append(json.loads(line)['text'])
    return texts

def load_split(data_dir, human_source, computer_source, split, n=np.inf):
    human = _load_split(data_dir, human_source, split, n=n//2)
    computer = _load_split(data_dir, computer_source, split, n=n//2)
    texts = human+computer
    labels = [0]*len(human)+[1]*len(computer)
    return texts, labels

def get_path(dir, source, name):
    return os.path.join(dir, f'{source}.{name}')

def dump_model(model_dir, source, name, model):
    serializer = pickle
    path = get_path(model_dir, source, name)
    with open(path, 'wb') as f:
        serializer.dump(model, f)

def load_model(model_dir, source, name):
    serializer = pickle
    path = get_path(model_dir, source, name)
    with open(path, 'rb') as f:
        return serializer.load(f)

def has_model(model_dir, source, name):
    return os.path.exists(get_path(model_dir, source, name))

# adapted from: https://github.com/openai/gpt-2-output-dataset
def main(
    human_source,
    computer_source,
    data_dir='./data',
    model_dir='./model',
    log_dir='./log',
    n_train=18000,
    n_valid=1000,
    n_test=1000,
    max_features = 2**16,
    n_jobs=None,
    verbose=False
):

    print("Start loading datasets")

    train_texts, train_labels = load_split(data_dir, human_source, computer_source, 'train', n=n_train)
    valid_texts, valid_labels = load_split(data_dir, human_source, computer_source, 'valid', n=n_valid)
    test_texts, test_labels = load_split(data_dir, human_source, computer_source, 'test', n=n_test)

    print("Datasets loaded")

    vect_name = f'tfidf_vect_{max_features}_feat.bin'
    model_name = f'tfidf_model_{max_features}_feat.bin'
    log_name = f'tfidf_model_{max_features}_feat.json'

    if has_model(model_dir, computer_source, vect_name):

        print(f"Load {vect_name} vect")
        vect = load_model(model_dir, computer_source, vect_name)

    else:

        print(f"Initialize {vect_name}")
        vect = TfidfVectorizer(ngram_range=(1, 2), min_df=5, max_features=max_features)
        print(f"Start training vect")
        vect.fit_transform(train_texts)
        print(f"Dump vect")
        dump_model(model_dir, computer_source, vect_name, vect)
        print(f"Vect dumped")

    valid_features = vect.transform(valid_texts)
    test_features = vect.transform(test_texts)


    if has_model(model_dir, computer_source, model_name):

        print(f"Load {model_name} model")
        model = load_model(model_dir, computer_source, model_name)

    else:

        train_features = vect.transform(train_texts)

        model = LogisticRegression(solver='liblinear')

        print("LogisticRegression initialized!")

        params = {'C': [1/64, 1/32, 1/16, 1/8, 1/4, 1/2, 1, 2, 4, 8, 16, 32, 64]}
        split = PredefinedSplit([-1]*n_train+[0]*n_valid)
        search = GridSearchCV(model, params, cv=split, n_jobs=n_jobs, verbose=verbose, refit=False)

        print("Start grid search!")

        search.fit(sparse.vstack([train_features, valid_features]), train_labels+valid_labels)
        model = model.set_params(**search.best_params_)

        print("Fit model!")

        model.fit(train_features, train_labels)
        dump_model(model_dir, computer_source, model_name, model)
        print(f"Model dumped")

    valid_accuracy = model.score(valid_features, valid_labels)*100.
    test_accuracy = model.score(test_features, test_labels)*100.

    data = {
        'source':computer_source,
        'n_train':n_train,
        'valid_accuracy':valid_accuracy,
        'test_accuracy':test_accuracy
    }
    print(data)
    json.dump(data, open(get_path(log_dir, computer_source, log_name), 'w'))

if __name__ == '__main__':
    fire.Fire(main)