GithubRepoClassification/main.py at master · mbornstein/GithubRepoClassification · 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
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
import sys

import numpy as np
import pandas as pd
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import VotingClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_val_score
from sklearn.neural_network import MLPClassifier
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import MinMaxScaler, StandardScaler, minmax_scale
from sklearn.svm import SVC
from sklearn.tree import DecisionTreeClassifier

from customClassifier.TwoStepClassifier import TwoStepClassifier
from importer.datasetImporter import DatasetImporter

# We get a deadlock with the VotingClassifier when running multiple threads under mac os
if sys.platform == "darwin":
    N_JOBS = 1
else:
    N_JOBS = -1


def normalize_data(data):
    skip_log = ['avg_entropy', 'up_to_dateness', 'edu_mail_ratio']
    X_log = np.log(data + 1.0)
    X_log[skip_log] = data[skip_log]
    return pd.DataFrame(minmax_scale(X_log), columns=data.columns)


def logarithmitize(X):
    skip_log = ['avg_entropy', 'up_to_dateness', 'edu_mail_ratio']
    X_log = np.log(X + 1.0)
    X_log[skip_log] = X[skip_log]
    return X_log


def train_and_test_multiple(algos, X, y):
    print('Null accuracy', max([len(y[y == element]) for element in np.unique(y)]) / len(y))
    print('Accuracies:')

    skip_log = ['avg_entropy', 'up_to_dateness', 'edu_mail_ratio']
    X_log = np.log(X + 1.0)
    X_log[skip_log] = X[skip_log]

    for algo in algos:
        accuracy = cross_val_score(algo, X_log, y)
        if type(algo) == TwoStepClassifier:
            print(type(algo.model1).__name__, '+', type(algo.model2).__name__ + ':\t', end='')
        elif type(algo) == Pipeline:
            print(type(algo.named_steps['algo']).__name__ + ':\t', end='')
        else:
            print(type(algo).__name__ + ':\t', end='')
        print('%0.3f +/- %0.2f' % (accuracy.mean(), accuracy.std() * 2))


def learn_full(algos, importer):
    train_and_test_multiple(algos, importer.data, importer.target)


def learn_full_unnormalized(algos, importer):
    train_and_test_multiple(algos, importer.data, importer.target)


def learn_step_one(algos, importer):
    X = normalize_data(importer.data)
    y = np.array(['NO-DEV', 'DEV'])[(importer.target == 'DEV') * 1]
    train_and_test_multiple(algos, X, y)


def learn_step_two(algos, importer):
    y = importer.target[importer.target != 'DEV']
    X = normalize_data(importer.data)[importer.target != 'DEV']
    train_and_test_multiple(algos, X, y)


def test():
    print('Enter test mode: testing different learning models...')

    std_logreg = Pipeline([
        ('std', StandardScaler()),
        ('log_reg', LogisticRegression(C=1.0, max_iter=1000, solver='lbfgs', multi_class='ovr', random_state=1337))
    ])

    algorithms = [
        DecisionTreeClassifier(random_state=1337),
        # LogisticRegression(C=1.0, max_iter=1000, solver='lbfgs', multi_class='ovr', random_state=1337),
        # LogisticRegression(C=1.0, max_iter=100, n_jobs=2, random_state=1337),
        std_logreg,
        SVC(C=20.0, random_state=1337),
        RandomForestClassifier(n_estimators=100, random_state=1337),
        MLPClassifier(max_iter=20000, hidden_layer_sizes=(100,), random_state=1337, shuffle=False,
                      learning_rate='adaptive'),
        MLPClassifier(max_iter=20000, hidden_layer_sizes=(50, 20), random_state=1337, shuffle=False,
                      learning_rate='adaptive'),
        # CustomKMeans(KMeans(n_clusters=15, random_state=1337)),
        # CustomKMeans(KMeans(n_clusters=8, random_state=1337)),
        GradientBoostingClassifier(learning_rate=0.15, random_state=1337),
        VotingClassifier([('log', std_logreg),
                          ('svc', SVC(C=20.0, random_state=1337)),
                          ('rf', RandomForestClassifier(n_estimators=100, random_state=1337)),
                          ('mlp',
                           MLPClassifier(max_iter=20000, hidden_layer_sizes=(50, 20), random_state=1337, shuffle=False,
                                         learning_rate='adaptive')),
                          ('mlp2',
                           MLPClassifier(max_iter=20000, hidden_layer_sizes=(100,), random_state=1337, shuffle=False,
                                         learning_rate='adaptive')),
                          ('gb', GradientBoostingClassifier(learning_rate=0.15, random_state=1337)),
                          ], n_jobs=N_JOBS)
    ]

    algorithms = [Pipeline([
        ('range_scale', MinMaxScaler()),
        ('algo', algo)
    ]) for algo in algorithms]

    importer = DatasetImporter('data/testset.csv')

    #print('\nStep 1 learning')
    # learn_step_one(algorithms, importer)
    #print('\nStep 2 learning')
    # learn_step_two(algorithms, importer)
    #print('\nFull learning without normalization')
    # learn_full_unnormalized(algorithms, importer)
    print('\nFull learning')
    learn_full(algorithms, importer)

    # use only combinations of best N if runtime is too high
    # print('\nTwo step classification:')
    # two_step_algos = [TwoStepClassifier(algo_a, algo_b) for algo_a, algo_b in itertools.product(algorithms, copy.deepcopy(algorithms))]
    # learn_full(two_step_algos, importer)

def trainAndPredict(repos):
    #print('Enter train and predict mode. It trains the model and predicts categories of the given repositories')

    std_logreg = Pipeline([
        ('std', StandardScaler()),
        ('log_reg', LogisticRegression(C=1.0, max_iter=1000, solver='lbfgs', multi_class='ovr', random_state=1337))
    ])

    #this is our final model
    vc = VotingClassifier([('log', std_logreg),
                      ('svc', SVC(C=20.0, random_state=1337)),
                      ('rf', RandomForestClassifier(n_estimators=100, random_state=1337)),
                      ('mlp',
                       MLPClassifier(max_iter=20000, hidden_layer_sizes=(50, 20), random_state=1337, shuffle=False,
                                     learning_rate='adaptive')),
                      ('mlp2',
                       MLPClassifier(max_iter=20000, hidden_layer_sizes=(100,), random_state=1337, shuffle=False,
                                     learning_rate='adaptive', warm_start=True)),
                      ('gb', GradientBoostingClassifier(learning_rate=0.15, random_state=1337, warm_start=True)),
                      ], n_jobs=N_JOBS)

    classifier = Pipeline([
        ('range_scale', MinMaxScaler()),
        ('model', vc)
    ])

    #train the classifier
    importer = DatasetImporter('data/enriched_data.csv', complete_set=True)
    classifier.fit(logarithmitize(importer.data), importer.target)

    # predict gives repositories
    repos = [repo.strip() for repo in repos if repo.strip() != '']
    prediction = classifier.predict(logarithmitize(DatasetImporter.get_data(repos)))
    for repo, category in zip(repos, prediction):
        print(repo + ', ' + category)


if __name__ == '__main__':
    if len(sys.argv) < 2:
        test()
    else:
        repos = open(sys.argv[1], 'r').read().split('\n')
        trainAndPredict(repos)