-
Notifications
You must be signed in to change notification settings - Fork 3
/
util.py
400 lines (372 loc) · 10.7 KB
/
util.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
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
# coding:utf-8
import re, os, sys, time, codecs
import collections, copy, cPickle, operator
import math, string
# This file provides several utility functions for further use.
# file path for converting the original train/test file
# to format of CRF++
train_file = "data/MSRA/train.txt"
output_train_file = "data/MSRA/train.out"
test_file = "data/MSRA/testright.txt"
output_test_file = "data/MSRA/testright.out"
# input file for further analysis and ground-truth file.
test_output_file = "data/MSRA/testoutput_char_e_85.out"
testright_file = "dict/testright.txt"
# resource files: surname, place and org.
surname_file = "dict/surname.txt"
surname_save = "dict/surname.save"
place_file = "dict/place_surf.txt"
place_save = "dict/place_surf.save"
org_file = "dict/org_surf.txt"
org_save = "dict/org_surf.save"
surname_list = {"single": [], "double": []}
place_list = []
org_list = []
# predefined list...
num_list = string.digits + u"一二三四五六七八九十两零几多"
punct_list = string.punctuation + "be" + u"。”’、—"
sw_list = u"些仅不个乎也了仍们你我他她但借假再几别既即却又另只的"\
u"叫各吓吗嘛否吧吱呀呃哦呗呢呵呜呸咋咦咧咱咳哇哈哎哒哟哦噢哪哼唉啥啦啪"\
u"喂喏喔喽嗡嗬嗯嗳嘎嘘嘻嘿里是还到"\
u"它就很得怎么打把某死每没而虽被谁说贼这俄"
org_sw = u"些仅不个乎也了仍们你我他她但再几别既即却又另只的"\
u"叫各吓吗嘛否吧吱呀呃哦呗呢呵呜呸咋咦咧咱咳哇哈哎哒哟哦噢哪哼唉啥啦啪"\
u"喂喏喔喽嗡嗬嗯嗳嘎嘘嘻嘿里是还到"\
u"它很得怎么把某每没而虽被谁说贼这"
org_sw += punct_list
# note that \u1111, 1112, 1113 are PER, LOC, ORG, respectively.
name_sw = sw_list + u"说摄地会使委市" + u"\u1111\u1112\u1113"
# convert full-mode into half-mode
def strQ2B(ustring):
"""全角转半角"""
rstring = ""
for uchar in ustring:
inside_code=ord(uchar)
if inside_code == 12288:
inside_code = 32
# # ,?!:;() are not converted
# elif inside_code in [65292,65311,65281,65307,65306,65288,65289]:
# inside_code = inside_code
elif (inside_code >= 65281 and inside_code <= 65374):
inside_code -= 65248
rstring += unichr(inside_code)
return rstring
# currently BIEO repr.
# e.g.: Apple/ns -> A: b_ns, p: i_ns, e: e_ns
# input: i: idx of char, l: length of word, t: type of word.
def get_ne_type(i, l, t):
if t == "o":
return "o"
if i == 0:
return "b_"+t
elif i == l-1:
return "e_"+t
else:
return "i_"+sList[1]
def compile_surname(sent, feat_list):
global surname_list
idx = len(feat_list[0])
for f in feat_list:
f.append("0")
i = 0
while True:
if i == len(sent):
break;
if sent[i] in surname_list["single"]:
feat_list[i][idx] = "1"
if sent[i:i+1] in surname_list["double"]:
feat_list[i][idx] = "2"
feat_list[i+1][idx] = "2"
i += 1
i += 1
def compile_place(sent, feat_list):
global place_list
idx = len(feat_list[0])
for f in feat_list:
f.append("0")
i = 0
while True:
if i == len(sent):
break;
if sent[i] in place_list:
feat_list[i][idx] = "1"
elif sent[i:i+2] in place_list:
feat_list[i][idx] = "1"
feat_list[i+1][idx] = "1"
i += 1
i += 1
def compile_org(sent, feat_list):
global org_list
idx = len(feat_list[0])
for f in feat_list:
f.append("0")
i = 0
while True:
if i == len(sent):
break;
if sent[i] in org_list:
feat_list[i][idx] = "1"
elif sent[i:i+2] in org_list:
feat_list[i][idx] = "1"
feat_list[i+1][idx] = "1"
i += 1
elif sent[i:i+3] in org_list:
feat_list[i][idx] = "1"
feat_list[i+1][idx] = "1"
feat_list[i+2][idx] = "1"
i += 2
i += 1
# input: original sentence.
# output: a list containing the features extracted. (like is_surname, is_place)
def compile_features(sent):
feat_list = []
for i in range(len(sent)):
feat_list.append([])
compile_surname(sent, feat_list)
compile_place(sent, feat_list)
compile_org(sent, feat_list)
return feat_list
def read_output(fname, rname):
f = open(fname, "r")
f2 = open(rname, "r")
true_y = []
pred_y = []
while(True):
true_list = {}
pred_list = {}
test = f2.readline().decode("utf8", "ignore")
seg2 = test.split()
if not seg2:
break
sent = ""
for s in seg2:
sent += s.split("/")[0]
# get true NEs.
cur_idx = 0
for i in range(len(seg2)):
s = seg2[i]
if s[-1] == "o":
continue
else:
ne = s.split("/")[0]
idx = sent.find(ne, cur_idx)
cur_idx = idx+len(ne)
ne = ne + "_" + str(idx)
true_list[ne] = s.split("/")[-1]
# integrate predicted output.
ne_type = ""
ne = ""
cur_idx = 0
is_end = False
while True:
tmp_ne = ""
tmp_ne_type = ""
line = f.readline().strip().decode("utf8")
l = line.split()
if not line:
# print "break"
break
tag = l[-1]
if tag == "o":
if ne:
is_end = True
elif ne_type and tag[2:] != ne_type:
is_end = True
tmp_ne = l[0]
tmp_ne_type = tag.split('_')[-1]
elif tag[0] == "b":
if ne:
is_end = True
tmp_ne = l[0]
tmp_ne_type = tag.split('_')[-1]
else:
ne = l[0]
ne_type = tag.split('_')[-1]
else:
ne += l[0]
if is_end and ne:
idx = sent.find(ne, cur_idx)
cur_idx = idx + len(ne)
ne = ne + "_" + str(idx)
# emit NE!
pred_list[ne] = ne_type
is_end = False
ne = ""
ne_type = ""
if tmp_ne:
ne = tmp_ne
ne_type = tmp_ne_type
total_ne = set(true_list.items()) | set(pred_list.items())
if total_ne == set(true_list.items()):
continue
print test
for item in set(pred_list.items()) - set(true_list.items()):
# for item in total_ne:
print item[0], item[1]
# if item[0] in true_list:
# true_y.append(item[1])
# else:
# true_y.append(u"o")
# if item[0] in pred_list:
# pred_y.append(item[1])
# else:
# pred_y.append(u"o")
print "\n\n"
true_list = {}
pred_list = {}
# print "\n"
f.close()
return true_y, pred_y
# given texts with format: xxx/nr xx...
# output as format: x 0 b_nr
# which is the format for CRF.
# isTest: provide a different precessing method for test file (no ground truth)
# extract_ne: a function var, we use it to extract pre-defined ne from rules
# , and use it as a refernce feature in CRF++. Example: see rule.py
def read_write(fname, oname, isTest=False, extract_ne=None):
f = open(fname, "r")
out = open(oname, "w")
cnt = 0
for line in f.readlines():
print cnt
cnt += 1
ne_list = []
sent = ""
if line.strip() == "":
continue
line = strQ2B(line.strip().decode("utf8"))
if not isTest:
for seg in line.split():
if seg[0] == u"\ufeff":
seg = seg[1:]
sList = seg.split("/")
i = 0
sent += strQ2B(sList[0])
l = len(sList[0])
for i in range(l):
ne_type = get_ne_type(i, l, sList[1])
ne_list.append(ne_type)
feat_list = compile_features(sent)
if extract_ne:
pred_list = extract_ne(sent)
for i in range(len(ne_list)):
tmp_list = [sent[i]] + feat_list[i]
if extract_ne:
tmp_list.append(pred_list[i])
tmp_list.append(ne_list[i])
info = "\t".join(tmp_list)
out.write(info.encode("utf8") + "\n")
else:
sent = line
feat_list = compile_features(sent)
if extract_ne:
pred_list = extract_ne(sent)
for i in range(len(sent)):
tmp_list = [sent[i]] + feat_list[i]
if extract_ne:
tmp_list.append(pred_list[i])
info = "\t".join(tmp_list)
out.write(info.encode("utf8") + "\n")
out.write("\n")
f.close()
out.close()
# utility function to load NEs from dicts.
def load_ne_from_file(fname, sname, tag):
if os.path.exists(sname):
temp_trie = cPickle.load(open(sname, "r"))
else:
temp_ne = get_ne(fname, tag)
temp_trie = dawg.IntDAWG(zip(temp_ne.keys(), temp_ne.values()))
cPickle.dump(temp_trie, open(sname, "w"))
return temp_trie
def load_suf_from_file(fname, sname):
temp_list = []
if os.path.exists(sname):
temp_list = cPickle.load(open(sname, "r"))
else:
f = open(fname, "r")
for line in f.readlines():
seg = line.strip().decode("utf8").split()
if not seg:
continue
temp_list.append(seg[0])
f.close()
cPickle.dump(temp_list, open(sname, "w"))
return temp_list
# whether a character is valid to consist a possible NE.
def is_valid(c, d=[]):
flag = (c >= u"\u4e00" and c <= u"\u9fa5")
flag = flag and (c not in org_sw+num_list)
if d != []:
flag = flag and (c in d)
return flag
# calculate PMI (pointwise mutual info) to extract top-k words.
def get_pmi(fname, dname, k):
f = open(fname, "r")
sur = {}
pre = {}
with open(dname, "r") as f1:
for l in f1.readlines():
l = l.strip().decode("utf8").split()
if not l:
continue
l = l[0]
pre[l[0]] = pre.get(l[0], 0) + 1
sur[l[-1]] = sur.get(l[-1], 0) + 1
sur = sorted(i for i in sur if i >= 5)
pre = sorted(i for i in pre if i >= 5)
bi = {}
uni = {}
pmi = {}
for line in f.xreadlines():
# preprocess
line = strQ2B(line.decode("utf8").strip())
for i in range(len(line[:])):
if is_valid(line[i]):
uni[line[i]] = uni.get(line[i], 0) + 1
if i < len(line)-1 and is_valid(line[i]) and is_valid(line[i+1]):
bi[line[i:i+2]] = bi.get(line[i:i+2], 0) + 1
f.close()
temp = sum(bi.values())
for b in bi:
bi[b] /= float(temp)
temp = sum(uni.values())
for u in uni:
uni[u] /= float(temp)
print len(bi)
for b in bi:
if uni[b[0]] == 0 or uni[b[1]] == 0 or (b[0] not in sur and b[1] not in pre):
continue
pmi[b] = math.log( bi[b]**k / (uni[b[0]]*uni[b[1]]) )
pmi = sorted(pmi.items(), key=operator.itemgetter(1), reverse=True)[:5000]
o = open("pmi_res2.out", "w")
for p, v in pmi:
o.write(p.encode("utf8") + "\t" + str(v) + "\n")
o.close()
def init():
global surname_list, place_list, org_list
if os.path.exists(surname_save):
surname_list = cPickle.load(open(surname_save, "r"))
else:
f = open(surname_file, "r")
for line in f.readlines():
seg = line.strip().decode("utf8").split()
if not seg:
continue
surname = seg[0]
if len(surname) == 1:
surname_list["single"].append(surname)
else:
surname_list["double"].append(surname)
f.close()
cPickle.dump(surname_list, open(surname_save, "w"))
place_list = load_suf_from_file(place_file, place_save)
org_list = load_suf_from_file(org_file, org_save)
init()
def main():
# read_write(train_file, output_train_file)
# read_write(test_file, output_test_file)
# true_y, pred_y = read_output(test_output_file, testright_file)
get_pmi("./data/pmi_text.txt", "./dict/all.txt", 10)
if __name__ == '__main__':
main()