-
Notifications
You must be signed in to change notification settings - Fork 0
/
identifier.py
328 lines (254 loc) · 8.51 KB
/
identifier.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
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
#read from file
import io
import os
import random
import collections
import sys
def print_acc(accuracies):
overall_accuracy = accuracies[0]
lang_accuracies = accuracies[1]
print('Overall Accuracy: ' + str(overall_accuracy))
print()
print('Accuracy per Language: ')
for key, value in lang_accuracies.items():
print(str(key) + ': ' + str(value))
def print_metrics(metrics):
micro = metrics['micro']
macro = metrics['macro']
print('Micro averaged Precision: ' + str(micro[0]))
print('Micro averaged Rank: ' + str(micro[1]))
print('Micro averaged F-measure: ' + str(micro[2]))
print()
print('Macro averaged Precision: ' + str(macro[0]))
print('Macro averaged Rank: ' + str(macro[1]))
print('Macro averaged F-measure: ' + str(macro[2]))
def accuracy(ts):
inverted_test_set = {}
accuracies = {}
overall_accuracy = 0
test_sentence_count = len(test_set)
for lang in languages:
accuracies[lang] = 0
for sentence,lang in test_set:
if lang not in inverted_test_set:
inverted_test_set[lang] = [sentence]
else:
inverted_test_set[lang].append(sentence)
for correct_lang, guessed_lang in ts:
if correct_lang == guessed_lang: #highest probability = predicted language
accuracies[correct_lang] += 1.0
overall_accuracy += 1.0
overall_accuracy/=test_sentence_count
overall_accuracy*=100
for key in accuracies:
accuracies[key]=(accuracies[key]/len(inverted_test_set[key]))*100
return (overall_accuracy, accuracies)
def metrics(orig_guessed_tuples):
matrices = {}
language_count = len(languages)
pisum = 0
rosum = 0
Fsum = 0
for l in languages:
for orig, guessed in orig_guessed_tuples:
TP, FP, FN = 0,0,0
if orig==l and guessed==l:
TP = 1
elif orig==l and guessed!=l:
FN = 1
elif orig!=l and guessed==l:
FP = 1
if l in matrices:
matrices[l] = (matrices[l][0] + TP, matrices[l][1] + FP, matrices[l][2] + FN)
else:
matrices[l] = (TP, FP, FN)
tpsum = 0
fpsum = 0
fnsum = 0
pisum = 0
rosum = 0
Fsum = 0
for key in matrices:
TP, FP, FN = matrices[key]
pii, roi = 0, 0
if (TP + FP) !=0:
pii = 1.0 * TP / (TP + FP)
if (TP + FN) !=0:
roi = 1.0 * TP / (TP + FN)
pisum += pii
rosum += roi
if not (pii==0 and roi==0):
Fsum += (2.0 * pii * roi) / (pii + roi)
tpsum+=TP
fpsum+=FP
fnsum+=FN
pi = 1.0 * tpsum / (tpsum + fpsum)
ro = 1.0 * tpsum / (tpsum + fnsum)
F = (2.0 * pi * ro) / (pi + ro)
pisum /= language_count
rosum /= language_count
Fsum /= language_count
return {'micro': (pi, ro, F), 'macro': (pisum, rosum, Fsum)}
def main():
sentences = [] #list of tuples of (sentence, language) -> we can shuffle list, not map
languages = set() #languages
count_map = {}
f = io.open('corpus.txt', 'r', encoding='utf-16') #read from file in utf-16 format
corpus_utf16 = f.readlines() #read line by line
i=0
corpus = []
for line in corpus_utf16:
#get the last token of the sentence as the language, the others as the sentence
sentence, lang = line.rsplit(None,1)
sentence, lang = sentence.strip().replace(' ', ''), lang.strip()
#in our analysis, no space character is included
sentences.append((sentence, lang))
languages.add(lang.strip())
#randomly split data to 90% training and 10% testing
sentence_count = len(sentences)
random.shuffle(sentences)
training_set = sentences[:int(sentence_count/10 * 9)]
test_set = sentences[int(sentence_count * 9/10):]
training_dict = {}
vocab_sizes = {}
#make sentences of a language into one big string
for value,key in training_set:
if key not in training_dict:
training_dict[key] = value
else:
training_dict[key] += value
#vocab size of language l
for l in training_dict:
vocab_sizes[l] = len(set(training_dict[l]))
if sys.argv[1] == 'naive_bayes' :
letter_probabilities = {}
for key in training_dict:
train_sentence = training_dict[key]
total_chars = len(training_dict[key]) #total number of characters in training set
chars = []
unique_chars = vocab_sizes[key]
for ch in train_sentence:
if ch not in chars:
#laplace smoothing
if key not in letter_probabilities:
letter_probabilities[key] = {ch: ((train_sentence.count(ch)+1.0)/(unique_chars + total_chars))}
else:
letter_probabilities[key][ch] = ((train_sentence.count(ch)+1.0)/(unique_chars + total_chars))
chars.append(ch)
#print vocab_sizes[key], len(letter_probabilities[key])
result = {}
language_count = len(languages)
for test_tuple in test_set:
test_sentence, _ = test_tuple
sum_probabilities = 0
lang_probs = {}#probability that sentence is that language
for l in languages:
total_chars = len(training_dict[l]) #total number of characters in training set
unique_chars = vocab_sizes[l]
probs = letter_probabilities[l] #letter probabilities of that language
for ci in test_sentence:
if ci in probs:
sum_probabilities += probs[ci] #sum(1..n) P(ci|l)
else:
sum_probabilities += (1.0/(total_chars+unique_chars+1))
sum_probabilities*=1.0/language_count #P(l)
lang_probs[l] = sum_probabilities
result[test_tuple] = sorted(lang_probs.items(), key=lambda x:x[1], reverse=True)
ts = []
for test_tuple in result:
correct_lang = test_tuple[1]
guessed_lang = result[test_tuple][0][0]
ts.append((correct_lang, guessed_lang))
print_acc(accuracy(ts))
print()
print_metrics(metrics(ts))
elif sys.argv[1] == 'unigram_svm' or sys.argv[1] == 'super_svm':
if len(sys.argv) < 3:
print ('Missing argument!')
else:
#SVM
vocab = {}
langs = {}
i=1
for key in languages:
langs[key] = i
i+=1
i=1
for sentence, lang in sentences:
for l in sentence:
if l not in vocab.values():
vocab[i] = l
i+=1
i=1
vocab = {v: k for k, v in vocab.items()}
svm_sentence_dict = {}
if sys.argv[1] == 'unigram_svm':
for sentence, lang in sentences:
st = str(langs[lang])
ch_counts = {}
for ch in sentence:
if vocab[ch] not in ch_counts:
ch_counts[vocab[ch]] = sentence.count(ch)
ch_counts = collections.OrderedDict(sorted(ch_counts.items()))
for key in ch_counts:
st+=' ' + str(key) + ':' + str(1)
svm_sentence_dict[sentence] = st #for that sentence, lang and f:v info
elif sys.argv[1] == 'super_svm':
bigram = {}
for sentence, lang in sentences:
for (f, s) in zip(sentence[0::2], sentence[1::2]):
if str(f+s) not in bigram.values():
bigram[i] = str(f+s)
i+=1
bigram = {v: k for k, v in bigram.items()}
for sentence, lang in sentences:
st = str(langs[lang])
ch_counts = {}
bigram_counts = {}
for ch in sentence:
if vocab[ch] not in ch_counts:
ch_counts[vocab[ch]] = sentence.count(ch)
for (f, s) in zip(sentence[0::2], sentence[1::2]):
if bigram[str(f+s)] not in bigram_counts:
bigram_counts[bigram[str(f+s)]] = sentence.count(str(f+s))
ch_counts = collections.OrderedDict(sorted(ch_counts.items()))
bigram_counts = collections.OrderedDict(sorted(bigram_counts.items()))
for key in ch_counts:
st+=' ' + str(key) + ':' + str(1)
for key in ch_counts:
st+=' ' + str(key + len(vocab)) + ':' + str(ch_counts[key])
for key in bigram_counts:
st+=' ' + str(key + 2*len(vocab)) + ':' + str(bigram_counts[key])
# st+=' ' + str(len(bigram) + 2*len(vocab) + 1) + ':' + str(sum(1 for c in sentence if c.isupper()))
svm_sentence_dict[sentence] = st #for that sentence, lang and f:v info
else:
print('Wrong input!')
train_strings = []
test_strings = []
f_train = open("svm_train.txt","w")
f_test = open("svm_test.txt","w")
for sentence, lang in training_set:
f_train.write(svm_sentence_dict[sentence] + '\n')
for sentence, lang in test_set:
f_test.write(svm_sentence_dict[sentence] + '\n')
f_train.close()
f_test.close()
path = str(sys.argv[2])
os.system(path + '/svm_multiclass_learn -c 1.0 svm_train.txt model.txt')
os.system(path + '/svm_multiclass_classify svm_test.txt model.txt output.txt')
out = open("output.txt","r")
test = open("svm_test.txt", "r")
guessed_probs = out.readlines()
orig_sent = test.readlines()
orig_guessed_tuples = []
langs = {v: k for k, v in langs.items()}
for s in guessed_probs:
guessed = langs[int(s.split(' ')[0])]
orig = langs[int(orig_sent[guessed_probs.index(s)].split(' ')[0])]
orig_guessed_tuples.append((orig, guessed))
print()
print_acc(accuracy(orig_guessed_tuples))
print()
print_metrics(metrics(orig_guessed_tuples))
if __name__ == "__main__":
main()