Upload preprocess.py
Browse files- preprocess.py +356 -0
preprocess.py
ADDED
@@ -0,0 +1,356 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from zipfile import ZipFile, ZIP_DEFLATED
|
2 |
+
import json
|
3 |
+
import os
|
4 |
+
from pprint import pprint
|
5 |
+
from copy import deepcopy
|
6 |
+
from collections import Counter
|
7 |
+
from tqdm import tqdm
|
8 |
+
from shutil import rmtree
|
9 |
+
import re
|
10 |
+
|
11 |
+
digit2word = {
|
12 |
+
'0': 'zero', '1': 'one', '2': 'two', '3': 'three', '4': 'four', '5': 'five',
|
13 |
+
'6': 'six', '7': 'seven', '8': 'eight', '9': 'nine', '10': 'ten'
|
14 |
+
}
|
15 |
+
|
16 |
+
match = {
|
17 |
+
'0': 0,
|
18 |
+
'1': 0,
|
19 |
+
'>1': 0,
|
20 |
+
}
|
21 |
+
|
22 |
+
|
23 |
+
def pharse_in_sen(phrase, sen):
|
24 |
+
'''
|
25 |
+
match value in the sentence
|
26 |
+
:param phrase: str
|
27 |
+
:param sen: str
|
28 |
+
:return: start, end if matched, else None, None
|
29 |
+
'''
|
30 |
+
assert isinstance(phrase, str)
|
31 |
+
pw = '(^|[\s,\.:\?!-])(?P<v>{})([\s,\.:\?!-]|$)'
|
32 |
+
pn = '(^|[\s\?!-]|\D[,\.:])(?P<v>{})($|[\s\?!-]|[,\.:]\D|[,\.:]$)'
|
33 |
+
if phrase.isdigit():
|
34 |
+
pattern = pn
|
35 |
+
else:
|
36 |
+
pattern = pw
|
37 |
+
p = re.compile(pattern.format(re.escape(phrase)), re.I)
|
38 |
+
m = re.search(p, sen)
|
39 |
+
if m:
|
40 |
+
num = len(re.findall(p, sen))
|
41 |
+
if num > 1:
|
42 |
+
match['>1'] += 1
|
43 |
+
else:
|
44 |
+
match['1'] += 1
|
45 |
+
return m.span('v'), num
|
46 |
+
if phrase.isdigit() and phrase in digit2word:
|
47 |
+
phrase = digit2word[phrase]
|
48 |
+
p = re.compile(pw.format(re.escape(phrase)), re.I)
|
49 |
+
m = re.search(p, sen)
|
50 |
+
if m:
|
51 |
+
num = len(re.findall(p, sen))
|
52 |
+
if num > 1:
|
53 |
+
match['>1'] += 1
|
54 |
+
else:
|
55 |
+
match['1'] += 1
|
56 |
+
return m.span('v'), num
|
57 |
+
match['0'] += 1
|
58 |
+
return (None, None), 0
|
59 |
+
|
60 |
+
|
61 |
+
def sys_intent():
|
62 |
+
"""from original data README.md"""
|
63 |
+
return {
|
64 |
+
"inform": {"description": "Inform the value for a slot to the user."},
|
65 |
+
"request": {"description": "Request the value of a slot from the user."},
|
66 |
+
"confirm": {"description": "Confirm the value of a slot before making a transactional service call."},
|
67 |
+
"offer": {"description": "Offer a certain value for a slot to the user."},
|
68 |
+
"notify_success": {"description": "Inform the user that their request was successful."},
|
69 |
+
"notify_failure": {"description": "Inform the user that their request failed."},
|
70 |
+
"inform_count": {"description": "Inform the number of items found that satisfy the user's request."},
|
71 |
+
"offer_intent": {"description": "Offer a new intent to the user."},
|
72 |
+
"req_more": {"description": "Asking the user if they need anything else."},
|
73 |
+
"goodbye": {"description": "End the dialogue."},
|
74 |
+
}
|
75 |
+
|
76 |
+
|
77 |
+
def usr_intent():
|
78 |
+
"""from original data README.md"""
|
79 |
+
return {
|
80 |
+
"inform_intent": {"description": "Express the desire to perform a certain task to the system."},
|
81 |
+
"negate_intent": {"description": "Negate the intent which has been offered by the system."},
|
82 |
+
"affirm_intent": {"description": "Agree to the intent which has been offered by the system."},
|
83 |
+
"inform": {"description": "Inform the value of a slot to the system."},
|
84 |
+
"request": {"description": "Request the value of a slot from the system."},
|
85 |
+
"affirm": {"description": "Agree to the system's proposition. "},
|
86 |
+
"negate": {"description": "Deny the system's proposal."},
|
87 |
+
"select": {"description": "Select a result being offered by the system."},
|
88 |
+
"request_alts": {"description": "Ask for more results besides the ones offered by the system."},
|
89 |
+
"thank_you": {"description": "Thank the system."},
|
90 |
+
"goodbye": {"description": "End the dialogue."},
|
91 |
+
}
|
92 |
+
|
93 |
+
|
94 |
+
def get_intent():
|
95 |
+
"""merge sys & usr intent"""
|
96 |
+
return {
|
97 |
+
"inform": {"description": "Inform the value for a slot."},
|
98 |
+
"request": {"description": "Request the value of a slot."},
|
99 |
+
"confirm": {"description": "Confirm the value of a slot before making a transactional service call."},
|
100 |
+
"offer": {"description": "Offer a certain value for a slot to the user."},
|
101 |
+
"notify_success": {"description": "Inform the user that their request was successful."},
|
102 |
+
"notify_failure": {"description": "Inform the user that their request failed."},
|
103 |
+
"inform_count": {"description": "Inform the number of items found that satisfy the user's request."},
|
104 |
+
"offer_intent": {"description": "Offer a new intent to the user."},
|
105 |
+
"req_more": {"description": "Asking the user if they need anything else."},
|
106 |
+
"goodbye": {"description": "End the dialogue."},
|
107 |
+
"inform_intent": {"description": "Express the desire to perform a certain task to the system."},
|
108 |
+
"negate_intent": {"description": "Negate the intent which has been offered by the system."},
|
109 |
+
"affirm_intent": {"description": "Agree to the intent which has been offered by the system."},
|
110 |
+
"affirm": {"description": "Agree to the system's proposition. "},
|
111 |
+
"negate": {"description": "Deny the system's proposal."},
|
112 |
+
"select": {"description": "Select a result being offered by the system."},
|
113 |
+
"request_alts": {"description": "Ask for more results besides the ones offered by the system."},
|
114 |
+
"thank_you": {"description": "Thank the system."},
|
115 |
+
}
|
116 |
+
|
117 |
+
|
118 |
+
def preprocess():
|
119 |
+
original_data_dir = 'dstc8-schema-guided-dialogue-master'
|
120 |
+
new_data_dir = 'data'
|
121 |
+
|
122 |
+
if not os.path.exists(original_data_dir):
|
123 |
+
original_data_zip = 'dstc8-schema-guided-dialogue-master.zip'
|
124 |
+
if not os.path.exists(original_data_zip):
|
125 |
+
raise FileNotFoundError(f'cannot find original data {original_data_zip} in sgd/, should manually download dstc8-schema-guided-dialogue-master.zip from https://github.com/google-research-datasets/dstc8-schema-guided-dialogue')
|
126 |
+
else:
|
127 |
+
archive = ZipFile(original_data_zip)
|
128 |
+
archive.extractall()
|
129 |
+
|
130 |
+
os.makedirs(new_data_dir, exist_ok=True)
|
131 |
+
|
132 |
+
dialogues = []
|
133 |
+
ontology = {'domains': {},
|
134 |
+
'intents': get_intent(),
|
135 |
+
'state': {},
|
136 |
+
'dialogue_acts': {
|
137 |
+
"categorical": {},
|
138 |
+
"non-categorical": {},
|
139 |
+
"binary": {}
|
140 |
+
}}
|
141 |
+
splits = ['train', 'validation', 'test']
|
142 |
+
dataset_name = 'sgd'
|
143 |
+
for data_split in splits:
|
144 |
+
data_dir = os.path.join(original_data_dir, data_split if data_split != 'validation' else 'dev')
|
145 |
+
# schema => ontology
|
146 |
+
with open(os.path.join(data_dir, 'schema.json')) as f:
|
147 |
+
data = json.load(f)
|
148 |
+
for schema in data:
|
149 |
+
domain = schema['service_name']
|
150 |
+
ontology['domains'].setdefault(domain, {})
|
151 |
+
ontology['domains'][domain]['description'] = schema['description']
|
152 |
+
ontology['domains'][domain].setdefault('slots', {})
|
153 |
+
ontology['domains'][domain]['active_intents'] = schema['intents']
|
154 |
+
ontology['state'].setdefault(domain, {})
|
155 |
+
for slot in schema['slots']:
|
156 |
+
ontology['domains'][domain]['slots'][slot['name']] = {
|
157 |
+
"description": slot['description'],
|
158 |
+
"is_categorical": slot['is_categorical'],
|
159 |
+
"possible_values": slot['possible_values']
|
160 |
+
}
|
161 |
+
ontology['state'][domain][slot['name']] = ''
|
162 |
+
# add 'count' slot
|
163 |
+
ontology['domains'][domain]['slots']['count'] = {
|
164 |
+
"description": "the number of items found that satisfy the user's request.",
|
165 |
+
"is_categorical": False,
|
166 |
+
"possible_values": []
|
167 |
+
}
|
168 |
+
|
169 |
+
# dialog
|
170 |
+
cnt = 0
|
171 |
+
for root, dirs, files in os.walk(data_dir):
|
172 |
+
fs = sorted([x for x in files if 'dialogues' in x])
|
173 |
+
for f in tqdm(fs, desc='processing schema-guided-{}'.format(data_split)):
|
174 |
+
data = json.load(open(os.path.join(data_dir, f)))
|
175 |
+
for d in data:
|
176 |
+
dialogue = {
|
177 |
+
"dataset": dataset_name,
|
178 |
+
"data_split": data_split,
|
179 |
+
"dialogue_id": f'{dataset_name}-{data_split}-{cnt}',
|
180 |
+
"original_id": d['dialogue_id'],
|
181 |
+
"domains": d['services'],
|
182 |
+
"goal": { # no goal
|
183 |
+
'description': '',
|
184 |
+
'inform': {},
|
185 |
+
'request': {}
|
186 |
+
},
|
187 |
+
"turns": []
|
188 |
+
}
|
189 |
+
cnt += 1
|
190 |
+
init_state = {}
|
191 |
+
for domain in dialogue['domains']:
|
192 |
+
init_state.setdefault(domain, deepcopy(ontology['state'][domain]))
|
193 |
+
|
194 |
+
for utt_idx, t in enumerate(d['turns']):
|
195 |
+
speaker = t['speaker'].lower()
|
196 |
+
turn = {
|
197 |
+
'speaker': speaker,
|
198 |
+
'utterance': t['utterance'],
|
199 |
+
'utt_idx': utt_idx,
|
200 |
+
'dialogue_acts': {
|
201 |
+
'binary': [],
|
202 |
+
'categorical': [],
|
203 |
+
'non-categorical': [],
|
204 |
+
},
|
205 |
+
}
|
206 |
+
for frame in t['frames']:
|
207 |
+
domain = frame['service']
|
208 |
+
for action in frame['actions']:
|
209 |
+
intent = action['act'].lower() # lowercase intent
|
210 |
+
assert intent in ontology['intents'], [intent]
|
211 |
+
slot = action['slot']
|
212 |
+
value_list = action['values']
|
213 |
+
if action['act'] in ['REQ_MORE', 'AFFIRM', 'NEGATE', 'THANK_YOU', 'GOODBYE']:
|
214 |
+
# Slot and values are always empty
|
215 |
+
assert slot == "" and len(value_list) == 0
|
216 |
+
turn['dialogue_acts']['binary'].append({
|
217 |
+
"intent": intent,
|
218 |
+
"domain": '',
|
219 |
+
"slot": ''
|
220 |
+
})
|
221 |
+
elif action['act'] in ['NOTIFY_SUCCESS', 'NOTIFY_FAILURE', 'REQUEST_ALTS', 'AFFIRM_INTENT', 'NEGATE_INTENT']:
|
222 |
+
# Slot and values are always empty
|
223 |
+
assert slot == "" and len(value_list) == 0
|
224 |
+
turn['dialogue_acts']['binary'].append({
|
225 |
+
"intent": intent,
|
226 |
+
"domain": domain,
|
227 |
+
"slot": ''
|
228 |
+
})
|
229 |
+
elif action['act'] in ['OFFER_INTENT', 'INFORM_INTENT']:
|
230 |
+
# slot containing the intent being offered.
|
231 |
+
assert slot == 'intent' and len(value_list) == 1
|
232 |
+
turn['dialogue_acts']['binary'].append({
|
233 |
+
"intent": intent,
|
234 |
+
"domain": domain,
|
235 |
+
"slot": value_list[0]
|
236 |
+
})
|
237 |
+
elif action['act'] in ['REQUEST'] and len(value_list) == 0:
|
238 |
+
# always contains a slot, but values are optional.
|
239 |
+
assert slot in ontology['domains'][domain]['slots'], f'{domain}-{slot}'
|
240 |
+
turn['dialogue_acts']['binary'].append({
|
241 |
+
"intent": intent,
|
242 |
+
"domain": domain,
|
243 |
+
"slot": slot
|
244 |
+
})
|
245 |
+
elif action['act'] in ['SELECT'] and len(value_list) == 0:
|
246 |
+
# (slot=='' and len(value_list) == 0) or (slot!='' and len(value_list) > 0)
|
247 |
+
assert slot == '', f'{domain}-{slot}-{action}'
|
248 |
+
turn['dialogue_acts']['binary'].append({
|
249 |
+
"intent": intent,
|
250 |
+
"domain": domain,
|
251 |
+
"slot": slot
|
252 |
+
})
|
253 |
+
elif action['act'] in ['INFORM_COUNT']:
|
254 |
+
# always has "count" as the slot, and a single element in values for the number of results obtained by the system.
|
255 |
+
assert slot == 'count' and len(value_list) == 1
|
256 |
+
value = value_list[0]
|
257 |
+
|
258 |
+
turn['dialogue_acts']['non-categorical'].append({
|
259 |
+
"intent": intent,
|
260 |
+
"domain": domain,
|
261 |
+
"slot": slot,
|
262 |
+
"value": value,
|
263 |
+
})
|
264 |
+
|
265 |
+
# find char span
|
266 |
+
(start, end), num = pharse_in_sen(value, t['utterance'])
|
267 |
+
assert num > 0, f'{value}-{t["utterance"]}' # {1:20086, 2:341, 3:19}
|
268 |
+
assert value.lower() == t['utterance'][start:end].lower() \
|
269 |
+
or digit2word[value].lower() == t['utterance'][start:end].lower()
|
270 |
+
# first match is always the choice
|
271 |
+
turn['dialogue_acts']['non-categorical'][-1].update({
|
272 |
+
"value": t['utterance'][start:end], "start": start, "end": end
|
273 |
+
})
|
274 |
+
else:
|
275 |
+
# have slot & value
|
276 |
+
if ontology['domains'][domain]['slots'][slot]['is_categorical']:
|
277 |
+
possible_values = [value.lower() for value in ontology['domains'][domain]['slots'][slot]['possible_values']]
|
278 |
+
for value in value_list:
|
279 |
+
if value.lower() not in possible_values and value != 'dontcare':
|
280 |
+
ontology['domains'][domain]['slots'][slot]['possible_values'].append(value)
|
281 |
+
print(f'add value to ontology\t{domain}-{slot}-{value}', possible_values)
|
282 |
+
turn['dialogue_acts']['categorical'].append({
|
283 |
+
"intent": intent,
|
284 |
+
"domain": domain,
|
285 |
+
"slot": slot,
|
286 |
+
"value": value,
|
287 |
+
})
|
288 |
+
else:
|
289 |
+
# span info in frame['slots']
|
290 |
+
for value in value_list:
|
291 |
+
for slot_info in frame['slots']:
|
292 |
+
start = slot_info['start']
|
293 |
+
end = slot_info['exclusive_end']
|
294 |
+
if slot_info['slot'] == slot and t['utterance'][start:end].lower() == value.lower():
|
295 |
+
assert t['utterance'][start:end] == value, f'{action}-{slot_info}-{t["utterance"][start:end]}'
|
296 |
+
turn['dialogue_acts']['non-categorical'].append({
|
297 |
+
"intent": intent,
|
298 |
+
"domain": domain,
|
299 |
+
"slot": slot,
|
300 |
+
"value": value,
|
301 |
+
"start": start,
|
302 |
+
"end": end
|
303 |
+
})
|
304 |
+
break
|
305 |
+
else:
|
306 |
+
assert value == 'dontcare', f'{action}-{slot_info}'
|
307 |
+
|
308 |
+
if speaker == 'user':
|
309 |
+
state = deepcopy(init_state)
|
310 |
+
active_intent = {}
|
311 |
+
requested_slots = {}
|
312 |
+
for frame in t['frames']:
|
313 |
+
domain = frame['service']
|
314 |
+
active_intent[domain] = frame['state']['active_intent']
|
315 |
+
requested_slots[domain] = frame['state']['requested_slots']
|
316 |
+
for slot, value_list in frame['state']['slot_values'].items():
|
317 |
+
state[domain][slot] = value_list[0]
|
318 |
+
for value in value_list[1:]:
|
319 |
+
state[domain][slot] += '|' + value
|
320 |
+
turn['state'] = state
|
321 |
+
turn['active_intent'] = active_intent
|
322 |
+
turn['requested_slots'] = requested_slots
|
323 |
+
else:
|
324 |
+
# service_call and service_results
|
325 |
+
turn['service_call'] = {}
|
326 |
+
turn['db_results'] = {}
|
327 |
+
for frame in t['frames']:
|
328 |
+
if 'service_call' not in frame:
|
329 |
+
continue
|
330 |
+
domain = frame['service']
|
331 |
+
turn['service_call'][domain] = frame['service_call']
|
332 |
+
turn['db_results'][domain] = frame['service_results']
|
333 |
+
|
334 |
+
# add to dialogue_acts dictionary in the ontology
|
335 |
+
for da_type in turn['dialogue_acts']:
|
336 |
+
das = turn['dialogue_acts'][da_type]
|
337 |
+
for da in das:
|
338 |
+
ontology["dialogue_acts"][da_type].setdefault((da['intent'], da['domain'], da['slot']), {})
|
339 |
+
ontology["dialogue_acts"][da_type][(da['intent'], da['domain'], da['slot'])][speaker] = True
|
340 |
+
dialogue['turns'].append(turn)
|
341 |
+
dialogues.append(dialogue)
|
342 |
+
|
343 |
+
for da_type in ontology['dialogue_acts']:
|
344 |
+
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()])
|
345 |
+
json.dump(dialogues[:10], open(f'dummy_data.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
346 |
+
json.dump(ontology, open(f'{new_data_dir}/ontology.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
347 |
+
json.dump(dialogues, open(f'{new_data_dir}/dialogues.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
348 |
+
with ZipFile('data.zip', 'w', ZIP_DEFLATED) as zf:
|
349 |
+
for filename in os.listdir(new_data_dir):
|
350 |
+
zf.write(f'{new_data_dir}/{filename}')
|
351 |
+
rmtree(original_data_dir)
|
352 |
+
rmtree(new_data_dir)
|
353 |
+
return dialogues, ontology
|
354 |
+
|
355 |
+
if __name__ == '__main__':
|
356 |
+
preprocess()
|