zhuqi commited on
Commit
57a3bf1
1 Parent(s): 6a7ab4a

Upload preprocess.py

Browse files
Files changed (1) hide show
  1. preprocess.py +276 -0
preprocess.py ADDED
@@ -0,0 +1,276 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import zipfile
2
+ import json
3
+ import os
4
+ import copy
5
+ from shutil import copy2, rmtree
6
+ from zipfile import ZipFile, ZIP_DEFLATED
7
+
8
+ ontology = {
9
+ 'domains': {
10
+ 'restaurant': {
11
+ 'description': 'find a restaurant to eat',
12
+ 'slots': {
13
+ 'area': {
14
+ 'description': 'area where the restaurant is located',
15
+ 'is_categorical': True,
16
+ 'possible_values': ["centre","north","west","south","east"]
17
+ },
18
+ 'price range': {
19
+ 'description': 'price range of the restaurant',
20
+ 'is_categorical': True,
21
+ 'possible_values': ["cheap","moderate","expensive"]
22
+ },
23
+ 'food': {
24
+ 'description': 'the cuisine of the restaurant',
25
+ 'is_categorical': False,
26
+ 'possible_values': ["afghan","african","afternoon tea","asian oriental","australasian","australian","austrian","barbeque","basque","belgian","bistro","brazilian","british","canapes","cantonese","caribbean","catalan","chinese","christmas","corsica","creative","crossover","cuban","danish","eastern european","english","eritrean","european","french","fusion","gastropub","german","greek","halal","hungarian","indian","indonesian","international","irish","italian","jamaican","japanese","korean","kosher","latin american","lebanese","light bites","malaysian","mediterranean","mexican","middle eastern","modern american","modern eclectic","modern european","modern global","molecular gastronomy","moroccan","new zealand","north african","north american","north indian","northern european","panasian","persian","polish","polynesian","portuguese","romanian","russian","scandinavian","scottish","seafood","singaporean","south african","south indian","spanish","sri lankan","steakhouse","swedish","swiss","thai","the americas","traditional","turkish","tuscan","unusual","vegetarian","venetian","vietnamese","welsh","world"]
27
+ },
28
+ 'name': {
29
+ 'description': 'name of the restaurant',
30
+ 'is_categorical': False,
31
+ 'possible_values': []
32
+ },
33
+ 'phone': {
34
+ 'description': 'phone number of the restaurant',
35
+ 'is_categorical': False,
36
+ 'possible_values': []
37
+ },
38
+ 'address': {
39
+ 'description': 'exact location of the restaurant',
40
+ 'is_categorical': False,
41
+ 'possible_values': []
42
+ },
43
+ 'postcode': {
44
+ 'description': 'postcode of the restaurant',
45
+ 'is_categorical': False,
46
+ 'possible_values': []
47
+ }
48
+ }
49
+ }
50
+ },
51
+ 'intents': {
52
+ 'inform': {
53
+ 'description': 'inform the value of a slot'
54
+ },
55
+ 'request': {
56
+ 'description': 'ask for the value of a slot'
57
+ },
58
+ 'nooffer': {
59
+ 'description': 'inform the user that there is no result satisfies user requirements'
60
+ }
61
+ },
62
+ 'state': {
63
+ 'restaurant': {
64
+ 'price range': '',
65
+ 'area': '',
66
+ 'food': ''
67
+ }
68
+ },
69
+ 'dialogue_acts': {
70
+ "categorical": {},
71
+ "non-categorical": {},
72
+ "binary": {}
73
+ }
74
+ }
75
+
76
+
77
+ def convert_da(utt, da):
78
+ global ontology
79
+ converted_da = {
80
+ 'binary': [],
81
+ 'categorical': [],
82
+ 'non-categorical': []
83
+ }
84
+
85
+ for intent, svs in da.items():
86
+ assert intent in ontology['intents']
87
+ if intent == 'nooffer':
88
+ assert svs == [['none', 'none']]
89
+ converted_da['binary'].append({
90
+ 'intent': intent,
91
+ 'domain': 'restaurant',
92
+ 'slot': '',
93
+ })
94
+ continue
95
+
96
+ for s, v in svs:
97
+ if 'care' in v:
98
+ assert v == 'dontcare', print(v)
99
+ assert s == s.lower()
100
+ if s == 'pricerange':
101
+ s = 'price range'
102
+ v = v
103
+ if intent == 'request':
104
+ assert v == '?'
105
+ converted_da['binary'].append({
106
+ 'intent': intent,
107
+ 'domain': 'restaurant',
108
+ 'slot': s
109
+ })
110
+ continue
111
+
112
+ if s in ['price range', 'area']:
113
+ assert v.lower() in ontology['domains']['restaurant']['slots'][s]['possible_values'] + ['dontcare'], print(s, v)
114
+ converted_da['categorical'].append({
115
+ 'intent': intent,
116
+ 'domain': 'restaurant',
117
+ 'slot': s,
118
+ 'value': v
119
+ })
120
+
121
+ else:
122
+ # non-categorical
123
+ start_ch = utt.lower().find(v.lower())
124
+
125
+ if start_ch == -1:
126
+ if not v == 'dontcare':
127
+ print('non-categorical slot value not found')
128
+ print('value: {}'.format(v))
129
+ print('sentence: {}'.format(utt))
130
+ print()
131
+
132
+ converted_da['non-categorical'].append({
133
+ 'intent': intent,
134
+ 'domain': 'restaurant',
135
+ 'slot': s,
136
+ 'value': v,
137
+ })
138
+ else:
139
+ converted_da['non-categorical'].append({
140
+ 'intent': intent,
141
+ 'domain': 'restaurant',
142
+ 'slot': s,
143
+ 'value': utt[start_ch: start_ch + len(v)],
144
+ 'start': start_ch,
145
+ 'end': start_ch + len(v)
146
+ })
147
+ assert utt[start_ch: start_ch + len(v)].lower() == v.lower()
148
+
149
+ return converted_da
150
+
151
+
152
+ def convert_state(slu):
153
+ global ontology
154
+ ret_state = copy.deepcopy(ontology['state'])
155
+ for da in slu:
156
+ if da['act'] != 'inform':
157
+ continue
158
+
159
+ for s, v in da['slots']:
160
+ s = s if s != 'pricerange' else 'price range'
161
+ if s not in ret_state['restaurant']:
162
+ print('slot not in state')
163
+ print(da)
164
+ print()
165
+ continue
166
+ ret_state['restaurant'][s] = v
167
+
168
+ return ret_state
169
+
170
+
171
+ def preprocess():
172
+ # use convlab-2 version camrest which already has dialog act annotation
173
+ original_data_dir = '../../camrest/'
174
+ new_data_dir = 'data'
175
+
176
+ os.makedirs(new_data_dir, exist_ok=True)
177
+
178
+ copy2(f'{original_data_dir}/db/CamRestDB.json', new_data_dir)
179
+
180
+ dataset = 'camrest'
181
+ domain = 'restaurant'
182
+ splits = ['train', 'validation', 'test']
183
+ dialogues_by_split = {split:[] for split in splits}
184
+
185
+ for split in ['train', 'val', 'test']:
186
+ data = json.load(zipfile.ZipFile(os.path.join(original_data_dir, f'{split}.json.zip'), 'r').open(f'{split}.json'))
187
+ if split == 'val':
188
+ split = 'validation'
189
+
190
+ cur_domains = [domain]
191
+
192
+ for ori_dialog in data:
193
+ dialogue_id = f'{dataset}-{split}-{len(dialogues_by_split[split])}'
194
+
195
+ goal = {
196
+ 'description': ori_dialog['goal']['text'],
197
+ 'inform': {'restaurant': {}},
198
+ 'request': {'restaurant': {}}
199
+ }
200
+ for slot, value in ori_dialog['goal']['info'].items():
201
+ if slot == 'pricerange':
202
+ slot = 'price range'
203
+ goal['inform'][domain][slot] = value
204
+ for slot in ori_dialog['goal']['reqt']:
205
+ if slot == 'pricerange':
206
+ slot = 'price range'
207
+ goal['request'][domain][slot] = ''
208
+
209
+ dialogue = {
210
+ 'dataset': dataset,
211
+ 'data_split': split,
212
+ 'dialogue_id': dialogue_id,
213
+ 'original_id': ori_dialog['dialogue_id'],
214
+ 'domains': cur_domains,
215
+ 'goal': goal,
216
+ 'finished': ori_dialog['finished'],
217
+ 'turns': []
218
+ }
219
+
220
+ for turn in ori_dialog['dial']:
221
+ usr_text = turn['usr']['transcript']
222
+ usr_da = turn['usr']['dialog_act']
223
+
224
+ sys_text = turn['sys']['sent']
225
+ sys_da = turn['sys']['dialog_act']
226
+
227
+ cur_state = convert_state(turn['usr']['slu'])
228
+ cur_user_da = convert_da(usr_text, usr_da)
229
+
230
+ usr_turn = {
231
+ 'speaker': 'user',
232
+ 'utterance': usr_text,
233
+ 'utt_idx': len(dialogue['turns']),
234
+ 'dialogue_acts': cur_user_da,
235
+ 'state': cur_state,
236
+ }
237
+
238
+ sys_turn = {
239
+ 'speaker': 'system',
240
+ 'utterance': sys_text,
241
+ 'utt_idx': len(dialogue['turns'])+1,
242
+ 'dialogue_acts': convert_da(sys_text, sys_da),
243
+ }
244
+
245
+ dialogue['turns'].append(usr_turn)
246
+ dialogue['turns'].append(sys_turn)
247
+
248
+ for turn in dialogue['turns']:
249
+ speaker = turn['speaker']
250
+ dialogue_acts = turn['dialogue_acts']
251
+
252
+ # add to dialogue_acts dictionary in the ontology
253
+ for da_type in dialogue_acts:
254
+ das = dialogue_acts[da_type]
255
+ for da in das:
256
+ ontology["dialogue_acts"][da_type].setdefault((da['intent'], da['domain'], da['slot']), {})
257
+ ontology["dialogue_acts"][da_type][(da['intent'], da['domain'], da['slot'])][speaker] = True
258
+ dialogues_by_split[split].append(dialogue)
259
+
260
+ dialogues = []
261
+ for split in splits:
262
+ dialogues += dialogues_by_split[split]
263
+ for da_type in ontology['dialogue_acts']:
264
+ ontology["dialogue_acts"][da_type] = sorted([str({'user': speakers.get('user', False), 'system': speakers.get('system', False), 'intent':da[0],'domain':da[1], 'slot':da[2]}) for da, speakers in ontology["dialogue_acts"][da_type].items()])
265
+ json.dump(dialogues[:10], open(f'dummy_data.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
266
+ json.dump(ontology, open(f'{new_data_dir}/ontology.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
267
+ json.dump(dialogues, open(f'{new_data_dir}/dialogues.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
268
+ with ZipFile('data.zip', 'w', ZIP_DEFLATED) as zf:
269
+ for filename in os.listdir(new_data_dir):
270
+ zf.write(f'{new_data_dir}/{filename}')
271
+ rmtree(new_data_dir)
272
+ return dialogues, ontology
273
+
274
+
275
+ if __name__ == '__main__':
276
+ preprocess()