-
Notifications
You must be signed in to change notification settings - Fork 5
/
clone_detection.py
173 lines (150 loc) · 5.46 KB
/
clone_detection.py
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
from collections import Counter
import json
import random
import re
import time
from pygments.lexers.jvm import JavaLexer
from pygments.token import Comment
from bleu_ignoring import corpus_bleu, SmoothingFunction
from CodeBLEU.code_bleu import code_bleu
from nltk.util import ngrams
import numpy as np
def print_results(tl, bl, cl):
print('BLEU:')
TP = ((bl == 1) & (tl == 1)).sum()
FP = ((bl == 1) & (tl == 0)).sum()
FN = ((bl == 0) & (tl == 1)).sum()
TN = ((bl == 0) & (tl == 0)).sum()
print(f' TP: {TP}, FP: {FP}, TN: {TN}, FN: {FN}')
print(f' Accuracy: {(TP + TN)/(TP+TN+FP+FN)}')
print(f' Precision: {TP/(TP+FP)}')
print(f' Recall: {TP/(TP+FN)}')
print(f' F1: {(2*TP)/(2*TP + FP + FN)}')
print()
print('CrystalBLEU:')
TP = ((cl == 1) & (tl == 1)).sum()
FP = ((cl == 1) & (tl == 0)).sum()
FN = ((cl == 0) & (tl == 1)).sum()
TN = ((cl == 0) & (tl == 0)).sum()
print(f' TP: {TP}, FP: {FP}, TN: {TN}, FN: {FN}')
print(f' Accuracy: {(TP + TN)/(TP+TN+FP+FN)}')
print(f' Precision: {TP/(TP+FP)}')
print(f' Recall: {TP/(TP+FN)}')
print(f' F1: {(2*TP)/(2*TP + FP + FN)}')
sm_func = SmoothingFunction(epsilon=0.0001).method1
lexer = JavaLexer()
code = {}
with open('clone_detection/data.jsonl') as f:
tmp = list(map(lambda x: json.loads(x), f.read().split('\n')[:-1]))
# tmp = tmp[:3000]
start_time = time.process_time()
all_ngrams = []
total = 0
for j in tmp:
try:
tok = [i[1] for i in lexer.get_tokens(j['func']) if not (re.fullmatch('\s+', i[1]) or (i[0] in Comment))]
total += len(tok)
code[j['idx']] = tok
if True:#random.random() < 0.3:
for i in range(1, 5):
all_ngrams.extend(ngrams(tok, i))
except:
break
freq = Counter(all_ngrams)
print('Preprocessing time:', time.process_time() - start_time)
print('Number of preprocessed tokens:', total)
# print(len(code.items()))
most_common_dict = dict(freq.most_common(500))
with open('clone_detection/train.txt') as f:
tmp = f.read().split('\n')
# tmp = tmp[::1000]
blues = [[], []]
crystals = [[], []]
for j in tmp:
x = re.split('\s+', j)
if len(x) == 3:
c1, c2, label = x
# if (c1 not in code) or (c2 not in code):
# continue
code1 = code[c1]
code2 = code[c2]
bleuscore = corpus_bleu([[code1]], [code2], smoothing_function=sm_func)
crystalbleuscore = corpus_bleu([[code1]], [code2], smoothing_function=sm_func, ignoring=most_common_dict)
blues[int(label)].append(bleuscore)
crystals[int(label)].append(crystalbleuscore)
f_b = np.mean(blues[0])
t_b = np.mean(blues[1])
f_c = np.mean(crystals[0])
t_c = np.mean(crystals[1])
th_b = (f_b + t_b)/2
th_c = (f_c + t_c)/2
# print(f_b, t_b, th_b)
# print(f_c, t_c, th_c)
with open('clone_detection/test.txt') as f:
tmp = f.read().split('\n')
true_label = []
bleu_label = []
crystal_label = []
intra_h = []
intra_r = []
inter_h = []
inter_r = []
bs = [[], []]
cs = [[], []]
for j in tmp[:30000]:
x = re.split('\s+', j)
if len(x) == 3:
c1, c2, label = x
# if (c1 not in code) or (c2 not in code):
# continue
code1 = code[c1]
code2 = code[c2]
if int(label) == 0:
inter_h.append(code2)
inter_r.append([code1])
else:
intra_h.append(code2)
intra_r.append([code1])
bleuscore = corpus_bleu([[code1]], [code2], smoothing_function=sm_func)
bs[int(label)].append(bleuscore)
crystalbleuscore = corpus_bleu([[code1]], [code2], smoothing_function=sm_func, ignoring=most_common_dict)
cs[int(label)].append(crystalbleuscore)
l_b = 1 if bleuscore > th_b else 0
l_c = 1 if crystalbleuscore > th_c else 0
bleu_label.append(l_b)
crystal_label.append(l_c)
true_label.append(int(label))
with open('clone_sim_scores.npy', 'wb') as f:
np.save(f, np.array(bs[0]))
np.save(f, np.array(bs[1]))
np.save(f, np.array(cs[0]))
np.save(f, np.array(cs[1]))
print_results(np.array(true_label), np.array(bleu_label), np.array(crystal_label))
print('# inter:', len(inter_h), '# intra:', len(intra_h))
print('BLEU inter')
start_time = time.process_time()
bleu_inter = corpus_bleu(inter_r, inter_h, smoothing_function=sm_func)
print('Calculation time:', time.process_time() - start_time)
print('BLEU intra')
start_time = time.process_time()
bleu_intra = corpus_bleu(intra_r, intra_h, smoothing_function=sm_func)
print('Calculation time:', time.process_time() - start_time)
print('CrystalBLEU inter')
start_time = time.process_time()
crystal_inter = corpus_bleu(inter_r, inter_h, smoothing_function=sm_func, ignoring=most_common_dict)
print('Calculation time:', time.process_time() - start_time)
print('CrystalBLEU intra')
start_time = time.process_time()
crystal_intra = corpus_bleu(intra_r, intra_h, smoothing_function=sm_func, ignoring=most_common_dict)
print('Calculation time:', time.process_time() - start_time)
print('CodeBLEU inter')
start_time = time.process_time()
code_inter = code_bleu(inter_r, inter_h)
print('Calculation time:', time.process_time() - start_time)
print('CodeBLEU intra')
start_time = time.process_time()
code_intra = code_bleu(intra_r, intra_h)
print('Calculation time:', time.process_time() - start_time)
print(f'BLEU distinguishability = {bleu_intra/bleu_inter}')
print(f'CrystalBLEU distinguishability = {crystal_intra/crystal_inter}')
print(f'CodeBLEU distinguishability = {code_intra/code_inter}')