-
Notifications
You must be signed in to change notification settings - Fork 8
/
phrase_segment.py
226 lines (189 loc) · 7.2 KB
/
phrase_segment.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
"""
This code is updated version of "http://chasen.org/~daiti-m/diary/misc/phraser.py" by Ye Kyaw Thu, LST, NECTEC, Thailand.
Last Updated: 10 Sept 2021
### Experiment Note by Ye
parameters for NPMI, -l {1..2} -t 0.1 -f {1..3} (ဆိုရင် phrase level ကောင်းကောင်း အလုပ်လုပ်ပေးတယ်)
### References
Experiment Note by Assoc. Prof. Daichi Mochihashi: http://chasen.org/~daiti-m/diary/
Statistically recognize long phrases with Normalized PMI: http://chasen.org/~daiti-m/diary/misc/phraser.py
https://courses.engr.illinois.edu/cs440/fa2018/lectures/lect36.html
https://courses.engr.illinois.edu/cs447/fa2018/Slides/Lecture17HO.pdf
https://en.wikipedia.org/wiki/Pointwise_mutual_information
https://stackoverflow.com/questions/6589814/what-is-the-difference-between-dict-and-collections-defaultdict
https://stackoverflow.com/questions/4406501/change-the-name-of-a-key-in-dictionary
https://stackoverflow.com/questions/47606995/python3-change-dictionary-key-from-string-to-tuple-of-strings
"""
import os
import sys
import tempfile
import numpy as np
from collections import defaultdict
from pylab import *
import pickle
import argparse
def connect (words, bond):
N = len (words)
n = 0
sentence = []
while (n < N):
flag = bond[n]
if flag == 0:
sentence.append (words[n])
n += 1
else:
sentence.append (words[n] + '_' + words[n+1])
n += 2
return sentence
def collocate (words, phrases):
N = len(words)
bond = []
for n in range(N-1):
(v,w) = (words[n],words[n+1])
if (v,w) in phrases:
bond.append (phrases[(v,w)]) # NPMI > 0
else:
bond.append (0)
bond.append (0)
# collocate max-first
while True:
s = max (bond)
n = bond.index (s)
if s == 0:
break
# connect maximum
bond[n] = -1
if n > 0:
bond[n-1] = 0
if n < N-2:
bond[n+1] = 0
# join words
return connect (words, bond)
def parse_write (file, phrases, output):
with open (file, 'r') as fh:
with open (output, 'w') as oh:
for line in fh:
words = line.rstrip('\n').split()
if len(words) > 0:
sentence = collocate (words, phrases)
#print("sentence:", sentence)
oh.write (' '.join (sentence) + '\n')
return output
# for Myanmar language, default threshold=0.2 and default minfreq=5 is best with "corpus2" training corpus
# for Japanese language, threshold=0.1 and minfreq=1 set by researcher Daichi Mochihashi.
def compute_phrase (unigram, bigram, threshold, minfreq):
N = sum (list(unigram.values()))
phrases = {}
#print(bigram.items())
for bi,freq in bigram.items():
if freq >= minfreq:
v = bi[0]; w = bi[1]
# print("N:", N, "v:", v, "w:", w, "unigram[v]:", unigram[v], "unigram[w]:", unigram[w])
npmi = (log(N) + log(freq) - log(unigram[v]) - log(unigram[w])) \
/ (log(N) - log(freq))
if npmi > threshold:
phrases[bi] = npmi
#print("phrases: ", phrases)
return phrases
def count_bigram (file, bigram_dict_txt, bigram_dict_bin):
fileBI_txt = open(bigram_dict_txt, "w")
bigram = defaultdict (int)
with open (file, 'r') as fh:
for line in fh:
words = line.rstrip('\n').split()
if len(words) > 0:
pword = words[0]
for word in words[1:]:
bigram[(pword,word)] += 1
pword = word
for key, value in bigram.items():
fileBI_txt.write (str(key)+'\t'+str(value)+'\n')
fileBI_txt.close()
#new_bigram = {} # got error
new_bigram = defaultdict(int)
# removed "_" for phrase option or phrase segmentation
for key, value in bigram.items():
keyi1 = key[0].replace('_', '')
keyi2 = key[1].replace('_', '')
new_bigram[(keyi1,keyi2)] = value
# write binary dictionary
fileBI_bin = open(bigram_dict_bin, "wb")
pickle.dump(new_bigram, fileBI_bin)
fileBI_bin.close()
new_bigram.clear()
return bigram
def count_unigram (file, unigram_dict_txt, unigram_dict_bin):
fileUNI_txt = open(unigram_dict_txt, "w")
unigram = defaultdict (int)
with open (file, 'r') as fh:
for line in fh:
words = line.rstrip('\n').split()
for word in words:
unigram[word] += 1
for key, value in unigram.items():
fileUNI_txt.write (str(key)+'\t'+str(value)+'\n')
fileUNI_txt.close()
#new_unigram = {}, got error
new_unigram = defaultdict(int)
# removed "_" for phrase option or phrase segmentation
new_unigram = { key.replace('_', ''): value for key, value in unigram.items() }
# write binary dictionary
fileUNI_bin = open(unigram_dict_bin, "wb")
pickle.dump(new_unigram, fileUNI_bin)
fileUNI_bin.close()
#new_unigram.clear()
return unigram
def eprint (s,clear=True):
if clear:
sys.stderr.write ('\x1b[K')
sys.stderr.write (s + "\n")
sys.stderr.flush ()
def train_phrase(iters, threshold, minfreq, uni_dict_txt, bi_dict_txt, uni_dict_bin, bi_dict_bin, file_input, file_output):
eprint ('computing phrases: threshold = %g minfreq = %d' % (threshold,minfreq))
filein = file_input
for iter in range(1,iters+1):
eprint ('pass [%d/%d]..' % (iter,iters))
# count n-grams
eprint ('- computing phrases..')
unigram = count_unigram (filein, uni_dict_txt, uni_dict_bin)
bigram = count_bigram (filein, bi_dict_txt, bi_dict_bin)
phrases = compute_phrase (unigram, bigram, threshold, minfreq)
# save intermediate file
if iter == iters:
fileout = file_output
else:
fileout = tempfile.mktemp()
eprint ('- writing output..')
parse_write (filein, phrases, fileout)
if (filein != file_input):
os.remove (filein)
filein = fileout
eprint ('done.')
def read_dict (fileDICT):
try:
with open(fileDICT, 'rb') as input_file:
dictionary = pickle.load(input_file)
input_file.close()
except FileNotFoundError:
print('Dictionary file', fileDICT, ' not found!')
return dictionary
def phrase_segmentation(threshold, minfreq, uni_dict_bin, bi_dict_bin, input, output):
eprint ('computing phrases: threshold = %g minfreq = %d' % (threshold,minfreq))
filein = input
eprint ('phrase segmentation...')
eprint ('- read unigram dictionary')
unigram = read_dict(uni_dict_bin)
#print(unigram)
eprint ('- read bigram dictionary')
bigram = read_dict(bi_dict_bin)
#print(bigram)
eprint ('- computing phrases..')
phrases = compute_phrase (unigram, bigram, threshold, minfreq)
#print(phrases)
# save intermediate file
fileout = output
print ('- writing output..., filename: ', output)
parse_write (filein, phrases, fileout)
#if (filein != input):
# os.remove (filein)
filein = output
eprint ('done.')