Jiann commited on
Commit
84cb2b1
·
verified ·
1 Parent(s): c3bf24f

Upload 5 files

Browse files
conceptnet_antonym.txt ADDED
The diff for this file is too large to render. See raw diff
 
conceptnet_entity.csv ADDED
The diff for this file is too large to render. See raw diff
 
gen_train_data.py ADDED
@@ -0,0 +1,344 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import numpy as np
2
+ import random
3
+ import re
4
+ import copy
5
+ from nltk.corpus import stopwords
6
+ import nltk
7
+ pos_tag = nltk.pos_tag
8
+ from nltk.stem import WordNetLemmatizer
9
+ lemma = WordNetLemmatizer().lemmatize
10
+ import sys
11
+
12
+ function_word = [".", ",", "!", "?", "male", "female", "neutral"]
13
+ def get_avail_phrases():
14
+ sw = set(stopwords.words('english'))
15
+ avail_phrases = set()
16
+ fin = open("./conceptnet_entity.csv", 'r')
17
+ for i, line in enumerate(fin):
18
+ avail_phrases.add(' '.join(line.strip().split("|||")[:-1]))
19
+ avail_phrases = avail_phrases - sw
20
+ fin.close()
21
+
22
+ fin = open("./negation.txt", 'r')
23
+ negation_word = []
24
+ for i, line in enumerate(fin):
25
+ word = ' '.join(line.strip().split()[1:])
26
+ negation_word.append(word)
27
+ avail_phrases.add(word)
28
+ fin.close()
29
+
30
+ for w in function_word:
31
+ avail_phrases.add(w)
32
+
33
+ with open("avail_phrases.txt", "w") as fout:
34
+ for w in avail_phrases:
35
+ fout.write(w+"\n")
36
+ return avail_phrases, negation_word
37
+
38
+ avail_phrases, negation_word = get_avail_phrases()
39
+
40
+ def output(st, fout):
41
+ if "w" in data_dir:
42
+ fout.write(" ".join(st)+"\n")
43
+ else:
44
+ for sen in st:
45
+ fout.write(sen+"\n")
46
+ fout.write("-"*5+"\n")
47
+
48
+ def repeat_sentence(st):
49
+ # repeat one sentence and delete the original sentence
50
+ idx = np.random.choice(np.arange(len(st))[1:], 1 + int(len(st)/2), replace=False).tolist()
51
+ s = min(idx)
52
+ tmp_st = copy.deepcopy(st)
53
+ for l in idx:
54
+ tmp_st[l] = copy.deepcopy(tmp_st[s])
55
+ return tmp_st
56
+
57
+ def repeat_ngram(st):
58
+ # repeat ngram in one sentence 1~4
59
+ def repeat_sen_gram(st):
60
+ flag = True
61
+ for _ in range(10):
62
+ try:
63
+ idx = np.random.choice(np.arange(len(st))[1:])
64
+ gram_num = np.random.choice(np.arange(5)[1:])
65
+ split_sen = st[idx].strip().split()
66
+ pointer_st = np.random.choice(np.arange(len(split_sen)))
67
+ pointer_ed = pointer_st + gram_num
68
+ if pointer_ed > len(split_sen):
69
+ pointer_ed = pointer_st
70
+ pointer_st = pointer_ed - gram_num
71
+ if pointer_st < 0:
72
+ continue
73
+ else:
74
+ flag = False
75
+ break
76
+ except:
77
+ continue
78
+ if flag:
79
+ return copy.deepcopy(st)
80
+ sen1, sen2, sen3 = " ".join(split_sen[:pointer_st]), " ".join(split_sen[pointer_st:pointer_ed]), " ".join(split_sen[pointer_ed:])
81
+ tmp_st = copy.deepcopy(st)
82
+ tmp_st[idx] = " ".join([sen1, sen2, sen2, sen3]).strip()
83
+ return tmp_st
84
+ for i in range(int(len(st)/2)):
85
+ st = repeat_sen_gram(st)
86
+ return st
87
+
88
+ def replace_sentence(st):
89
+ flag = True
90
+ for _ in range(10):
91
+ try:
92
+ tmp_st = copy.deepcopy(st)
93
+ idxs = np.random.choice(np.arange(len(st))[1:], np.random.choice(np.arange(1, len(st))), replace=False)
94
+ replace_st_id = np.random.choice(np.arange(len(story)))
95
+ for idx in idxs:
96
+ tmp_st[idx] = np.random.choice(story[replace_st_id])
97
+ flag = False
98
+ break
99
+ except:
100
+ continue
101
+ if flag:
102
+ return copy.deepcopy(st)
103
+ return tmp_st
104
+
105
+ def change_neg_helper(sen):
106
+ def pro(s):
107
+ final_sen = " ".join(s)
108
+ return final_sen
109
+ sen = sen.strip().split()
110
+ for i, n in enumerate(sen):
111
+ if n in negation_word:
112
+ del sen[i]
113
+ return pro(sen)
114
+ neg_list = ["not", "n't"]
115
+ for i, n in enumerate(sen):
116
+ if n in ["would", "will", "can", "could", "may", "might", "shall", "should", "do", "does", "did", "am", "is", "are", "was", "were", "be", "been"]:
117
+ sen.insert(i+1, np.random.choice(neg_list))
118
+ return pro(sen)
119
+ pos_sen = pos_tag(sen)
120
+ for i, n in enumerate(pos_sen):
121
+ if n[1] == "VB":
122
+ sen.insert(i, "do " + np.random.choice(neg_list))
123
+ return pro(sen)
124
+ elif n[1] == "VBD":
125
+ sen[i] = lemma(sen[i], "v")
126
+ sen.insert(i, "did " + np.random.choice(neg_list))
127
+ return pro(sen)
128
+ elif n[1] == "VBG":
129
+ sen.insert(i, np.random.choice(neg_list))
130
+ return pro(sen)
131
+ elif n[1] == "VBN":
132
+ sen.insert(i, np.random.choice(neg_list))
133
+ return pro(sen)
134
+ elif n[1] == "VBP":
135
+ sen.insert(i, "do " + np.random.choice(neg_list))
136
+ return pro(sen)
137
+ elif n[1] == "VBZ":
138
+ sen[i] = lemma(sen[i], "v")
139
+ sen.insert(i, "does " + np.random.choice(neg_list))
140
+ return pro(sen)
141
+ print("VERB ERROR")
142
+ return None
143
+
144
+ anotomy_word = {}
145
+ all_num, anotomy_num = 0, 0
146
+ with open("./conceptnet_antonym.txt", "r") as fin:
147
+ for line in fin:
148
+ tmp = line.strip().split("|||")
149
+ if len(tmp) == 3:
150
+ h, t = tmp[0], tmp[2].split()
151
+ if h in anotomy_word:
152
+ anotomy_word[h] += t
153
+ else:
154
+ anotomy_word[h] = t[:]
155
+
156
+ def change_neg_sentence(st):
157
+ flag = True
158
+ for _ in range(10):
159
+ try:
160
+ tmp_st = copy.deepcopy(st)
161
+ idxs = np.random.choice(np.arange(len(st))[1:], np.random.choice(np.arange(1, len(st))), replace=False)
162
+ for idx in idxs:
163
+ tmp_st_idx = change_neg_helper(st[idx])
164
+ if tmp_st_idx is not None:
165
+ tmp_st[idx] = tmp_st_idx
166
+ flag = False
167
+ if flag == False:
168
+ break
169
+ except:
170
+ continue
171
+ if flag:
172
+ return copy.deepcopy(st)
173
+ return tmp_st
174
+
175
+ def replace_word(st):
176
+ global all_num, anotomy_num
177
+ def replace_one_word(st):
178
+ anotomy = False
179
+ flag = True
180
+ for _ in range(100):
181
+ tmp_st = copy.deepcopy(st)
182
+ idx = np.random.choice(np.arange(len(st))[1:])
183
+ split_sen = tmp_st[idx].split()
184
+ pos_split_sen = pos_tag(split_sen)
185
+ avail_w_id = []
186
+ for w_id, w in enumerate(split_sen):
187
+ if (w in avail_phrases and w not in function_word and "[" not in w):
188
+ avail_w_id.append(w_id)
189
+ if len(avail_w_id) == 0: continue
190
+ word_id = np.random.choice(avail_w_id)
191
+ if pos_split_sen[word_id][1] not in pos_vocab_entity: continue
192
+ lemma_word = lemma(pos_split_sen[word_id][0], 'v' if pos_split_sen[word_id][1][0] == 'V' else 'n')
193
+ if lemma_word in anotomy_word:
194
+ replace_word = np.random.choice(anotomy_word[lemma_word])
195
+ anotomy = True
196
+ else:
197
+ word_freq = pos_vocab_entity[pos_split_sen[word_id][1]]
198
+ replace_word = ""
199
+ flag_in = True
200
+ for _ in range(10):
201
+ replace_word = np.random.choice(word_freq["word"], p=word_freq["freq"]/np.sum(word_freq["freq"]))
202
+ if len(word_freq["word"]) == 1 or replace_word != pos_split_sen[word_id][0]:
203
+ flag_in = False
204
+ break
205
+ if flag_in:
206
+ replace_word = pos_split_sen[word_id][0]
207
+ anotomy = False
208
+ tmp_split_sen = copy.deepcopy(split_sen)
209
+ split_sen[word_id] = replace_word
210
+ tmp_st[idx] = " ".join(split_sen)
211
+ flag = False
212
+ break
213
+ if flag:
214
+ return copy.deepcopy(st), False
215
+ return tmp_st, anotomy
216
+ num = 0
217
+ for idx in np.arange(len(st))[1:]:
218
+ for word in st[idx].split():
219
+ if word in avail_phrases:
220
+ num += 1
221
+ try:
222
+ final_num = np.random.choice(np.arange(1, int(num*0.15+1)))
223
+ except:
224
+ final_num = 1
225
+ for _ in range(final_num):
226
+ st, anotomy = replace_one_word(st)
227
+ all_num += 1
228
+ if anotomy: anotomy_num += 1
229
+ return st
230
+
231
+ def shuffle_sentence(st, n_sentence):
232
+ def exchange(l, ids, target_ids):
233
+ tmp_l = copy.deepcopy(l)
234
+ for o_id, t_id in zip(ids, target_ids):
235
+ tmp_l[o_id] = copy.deepcopy(l[t_id])
236
+ return tmp_l
237
+ # exchange n sentences
238
+ flag = True
239
+ for _ in range(10):
240
+ sen_ids = np.random.choice(np.arange(len(st))[1:], n_sentence, replace=False)
241
+ target_ids = np.random.permutation(sen_ids)
242
+ tmp_st = exchange(st, sen_ids, target_ids)
243
+ if st != tmp_st:
244
+ flag = False
245
+ break
246
+ if flag:
247
+ return copy.deepcopy(st)
248
+ return tmp_st
249
+ def get_pos_vocab(dir):
250
+ pos_vocab_entity = {}
251
+ with open("%s/entity_vocab.txt"%dir, "r") as fin:
252
+ for line in fin:
253
+ tmp = line.strip().split("|||")
254
+ word = tmp[0].split()[0]
255
+ pos = tmp[1:]
256
+ for p in pos:
257
+ pp = p.split()
258
+ if pp[0] in pos_vocab_entity:
259
+ pos_vocab_entity[pp[0]]["word"].append(word)
260
+ pos_vocab_entity[pp[0]]["freq"].append(float(pp[1]))
261
+ else:
262
+ pos_vocab_entity[pp[0]] = {"word":[word], "freq":[float(pp[1])]}
263
+ return pos_vocab_entity
264
+ # ========================================================================================
265
+
266
+ name_list = ["test", "dev", "train"]
267
+ data_dir = "./%s/ini_data"%("WritingPrompts" if "w" in sys.argv[1] else "ROCStories")
268
+ output_dir = "%s/train_data"%("WritingPrompts" if "w" in sys.argv[1] else "ROCStories")
269
+
270
+ # type_dict = {"repeat":0.6, "replace":0.15, "shuffle":0.15, "neg":0.1}
271
+ type_dict = {"repeat":0.1, "replace":0.3, "shuffle":0.4, "neg":0.2}
272
+ type_list = list(type_dict.keys())
273
+ type_prob_list = []
274
+ for t in type_list:
275
+ type_prob_list.append(type_dict[t])
276
+
277
+ time_list = [1,2,3,4]
278
+ # time_prob_list = [0.2,0.4,0.3,0.1]
279
+ time_prob_list = [0.5,0.2,0.2,0.1]
280
+
281
+ pos_vocab_entity = get_pos_vocab(data_dir)
282
+ for name in name_list:
283
+ if "w" in data_dir.lower():
284
+ with open("%s/%s.wp_source"%(data_dir, name), "r") as fin1:
285
+ with open("%s/%s.wp_target"%(data_dir, name), "r") as fin2:
286
+ story, tmp = [], []
287
+ for k, line in enumerate(fin2):
288
+ src = fin1.readline().strip()
289
+ if src[-1].isalpha():
290
+ src = src + " ."
291
+ tmp.append(src)
292
+ for sen in line.strip().split(".")[:-1]:
293
+ if sen.strip() != "":
294
+ tmp.append(sen.strip()+" .")
295
+ if len(tmp) >= 4:
296
+ story.append(tmp)
297
+ tmp = []
298
+ else:
299
+ with open("%s/%s.txt"%(data_dir, name), "r") as fin:
300
+ story, tmp = [], []
301
+ for k, line in enumerate(fin):
302
+ i = k + 1
303
+ if i % 6 == 0:
304
+ story.append(tmp)
305
+ tmp = []
306
+ else:
307
+ sen = line.strip()
308
+ tmp.append(sen+" ." if sen[-1].isalpha() else sen)
309
+
310
+ with open("%s/%s_human.txt"%(output_dir, name), "w") as fout:
311
+ for st_id, st in enumerate(story):
312
+ output(st, fout)
313
+
314
+ prefix = "%s/%s_negative"%(output_dir, name)
315
+ with open("%s.txt"%(prefix), "w") as fout:
316
+ for st_id, st in enumerate(story):
317
+ chaotic_list = np.random.choice(type_list,
318
+ np.random.choice(time_list, p=time_prob_list), replace=False, p=type_prob_list/np.sum(type_prob_list)).tolist()
319
+ print(chaotic_list)
320
+ for c in chaotic_list:
321
+ if c == "repeat":
322
+ if random.random() < 0.7:
323
+ st = repeat_sentence(st)
324
+ else:
325
+ st = repeat_ngram(st)
326
+ if c == "replace":
327
+ if random.random() < 0.5:
328
+ # replace one sentence
329
+ st = replace_sentence(st)
330
+ else:
331
+ # replace one word
332
+ st = replace_word(st)
333
+ if c == "shuffle":
334
+ n_sentence = int(np.random.choice(np.arange(1,len(st)-1)+1))
335
+ st = shuffle_sentence(st, n_sentence)
336
+ if c == "neg":
337
+ st = change_neg_sentence(st)
338
+ output(st, fout)
339
+
340
+
341
+
342
+ print("Anotomy:", anotomy_num)
343
+ print("All:", all_num)
344
+
get_vocab.py ADDED
@@ -0,0 +1,56 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from nltk.stem import WordNetLemmatizer
2
+ lemma = WordNetLemmatizer().lemmatize
3
+ import nltk
4
+ pos_tag = nltk.pos_tag
5
+ from nltk.corpus import stopwords
6
+ import sys
7
+
8
+ mode = sys.argv[1]
9
+ file_dir = "./WritingPrompts/ini_data/" if "w" in mode else "./ROCStories/ini_data/"
10
+ file_name = "train.wp_target" if "w" in mode else "train.txt"
11
+
12
+ def get_avail_phrases():
13
+ sw = set(stopwords.words('english'))
14
+ avail_phrases = set()
15
+ fin = open("./conceptnet_entity.csv", 'r')
16
+ for i, line in enumerate(fin):
17
+ avail_phrases.add(' '.join(line.strip().split("|||")[:-1]))
18
+ avail_phrases = avail_phrases - sw
19
+ fin.close()
20
+
21
+ fin = open("./negation.txt", 'r')
22
+ for i, line in enumerate(fin):
23
+ avail_phrases.add(' '.join(line.strip().split()[1:]))
24
+ fin.close()
25
+
26
+ for w in [".", ",", "!", "?", "male", "female", "neutral"]:
27
+ avail_phrases.add(w)
28
+
29
+ return avail_phrases
30
+
31
+ avail_phrases = get_avail_phrases()
32
+
33
+ vocab = {}
34
+ with open("%s/%s"%(file_dir, file_name), "r") as fin1:
35
+ for kkk, line in enumerate(fin1):
36
+ if kkk % 1000 == 0:
37
+ print(kkk)
38
+ tmp = line.strip().split()
39
+ pos = pos_tag(tmp)
40
+ for word_pos in pos:
41
+ if lemma(word_pos[0], 'v' if word_pos[1][0] == 'V' else 'n') not in avail_phrases:
42
+ continue
43
+ if word_pos[0] in vocab:
44
+ vocab[word_pos[0]]["number"] += 1
45
+ if word_pos[1] in vocab[word_pos[0]]:
46
+ vocab[word_pos[0]][word_pos[1]] += 1
47
+ else:
48
+ vocab[word_pos[0]][word_pos[1]] = 1
49
+ else:
50
+ vocab[word_pos[0]] = {word_pos[1]:1, "number":1}
51
+ vocab_list = sorted(vocab, key=lambda x: vocab[x]["number"], reverse=True)
52
+ with open("%s/entity_vocab.txt"%file_dir, "w") as fout:
53
+ for v in vocab_list:
54
+ pos_list = sorted(vocab[v], key=vocab[v].get, reverse=True)
55
+ pos_list.remove("number")
56
+ fout.write("%s %d|||"%(v, vocab[v]["number"]) + "|||".join(["%s %d"%(p, vocab[v][p]) for p in pos_list]) + "\n")
negation.txt ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0 no
2
+ 0 not
3
+ 0 n't
4
+ 0 none
5
+ 0 never
6
+ 0 nobody
7
+ 0 nothing
8
+ 0 nowhere
9
+ 0 neither
10
+ 0 nobody
11
+ 0 hardly
12
+ 0 scarcely
13
+ 0 barely
14
+ 0 seldom
15
+ 0 cannot
16
+ 0 can 't
17
+ 0 may not
18
+ 0 would n't
19
+ 0 would not
20
+ 0 should n't
21
+ 0 should not
22
+ 0 do n't
23
+ 0 do not
24
+ 0 does 't
25
+ 0 dose not
26
+ 0 did n't
27
+ 0 did not
28
+ 0 is n't
29
+ 0 is not
30
+ 0 are n't
31
+ 0 are not
32
+ 0 was n't
33
+ 0 was not