Upload preprocess.py
Browse files- preprocess.py +232 -0
preprocess.py
ADDED
|
@@ -0,0 +1,232 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import copy
|
| 2 |
+
import json
|
| 3 |
+
import os
|
| 4 |
+
from zipfile import ZipFile, ZIP_DEFLATED
|
| 5 |
+
from shutil import rmtree
|
| 6 |
+
|
| 7 |
+
ontology = {
|
| 8 |
+
'domains': {
|
| 9 |
+
'restaurant': {
|
| 10 |
+
'description': 'search for a restaurant to dine',
|
| 11 |
+
'slots': {
|
| 12 |
+
'food': {
|
| 13 |
+
'description': 'food type of the restaurant',
|
| 14 |
+
'is_categorical': False,
|
| 15 |
+
'possible_values': []
|
| 16 |
+
},
|
| 17 |
+
'area': {
|
| 18 |
+
'description': 'area of the restaurant',
|
| 19 |
+
'is_categorical': True,
|
| 20 |
+
'possible_values': ["east", "west", "centre", "north", "south"]
|
| 21 |
+
},
|
| 22 |
+
'postcode': {
|
| 23 |
+
'description': 'postal code of the restaurant',
|
| 24 |
+
'is_categorical': False,
|
| 25 |
+
'possible_values': []
|
| 26 |
+
},
|
| 27 |
+
'phone': {
|
| 28 |
+
'description': 'phone number of the restaurant',
|
| 29 |
+
'is_categorical': False,
|
| 30 |
+
'possible_values': []
|
| 31 |
+
},
|
| 32 |
+
'address': {
|
| 33 |
+
'description': 'address of the restaurant',
|
| 34 |
+
'is_categorical': False,
|
| 35 |
+
'possible_values': []
|
| 36 |
+
},
|
| 37 |
+
'price range': {
|
| 38 |
+
'description': 'price range of the restaurant',
|
| 39 |
+
'is_categorical': True,
|
| 40 |
+
'possible_values': ["expensive", "moderate", "cheap"]
|
| 41 |
+
},
|
| 42 |
+
'name': {
|
| 43 |
+
'description': 'name of the restaurant',
|
| 44 |
+
'is_categorical': False,
|
| 45 |
+
'possible_values': []
|
| 46 |
+
}
|
| 47 |
+
}
|
| 48 |
+
}
|
| 49 |
+
},
|
| 50 |
+
'intents': {
|
| 51 |
+
'inform': {
|
| 52 |
+
'description': 'system informs user the value of a slot'
|
| 53 |
+
},
|
| 54 |
+
'request': {
|
| 55 |
+
'description': 'system asks the user to provide value of a slot'
|
| 56 |
+
}
|
| 57 |
+
},
|
| 58 |
+
'state': {
|
| 59 |
+
'restaurant': {
|
| 60 |
+
'food': '',
|
| 61 |
+
'area': '',
|
| 62 |
+
'postcode': '',
|
| 63 |
+
'phone': '',
|
| 64 |
+
'address': '',
|
| 65 |
+
'price range': '',
|
| 66 |
+
'name': ''
|
| 67 |
+
}
|
| 68 |
+
},
|
| 69 |
+
"dialogue_acts": {
|
| 70 |
+
"categorical": {},
|
| 71 |
+
"non-categorical": {},
|
| 72 |
+
"binary": {}
|
| 73 |
+
}
|
| 74 |
+
}
|
| 75 |
+
|
| 76 |
+
|
| 77 |
+
def convert_da(da, utt):
|
| 78 |
+
global ontology
|
| 79 |
+
|
| 80 |
+
converted = {
|
| 81 |
+
'binary': [],
|
| 82 |
+
'categorical': [],
|
| 83 |
+
'non-categorical': []
|
| 84 |
+
}
|
| 85 |
+
|
| 86 |
+
for s, v in da:
|
| 87 |
+
if s == 'request':
|
| 88 |
+
converted['binary'].append({
|
| 89 |
+
'intent': 'request',
|
| 90 |
+
'domain': 'restaurant',
|
| 91 |
+
'slot': v,
|
| 92 |
+
})
|
| 93 |
+
|
| 94 |
+
else:
|
| 95 |
+
slot_type = 'categorical' if ontology['domains']['restaurant']['slots'][s]['is_categorical'] else 'non-categorical'
|
| 96 |
+
|
| 97 |
+
v = v.strip()
|
| 98 |
+
if v != 'dontcare' and ontology['domains']['restaurant']['slots'][s]['is_categorical']:
|
| 99 |
+
if v == 'center':
|
| 100 |
+
v = 'centre'
|
| 101 |
+
elif v == 'east side':
|
| 102 |
+
v = 'east'
|
| 103 |
+
assert v in ontology['domains']['restaurant']['slots'][s]['possible_values'], print([s,v, utt])
|
| 104 |
+
|
| 105 |
+
converted[slot_type].append({
|
| 106 |
+
'intent': 'inform',
|
| 107 |
+
'domain': 'restaurant',
|
| 108 |
+
'slot': s,
|
| 109 |
+
'value': v
|
| 110 |
+
})
|
| 111 |
+
|
| 112 |
+
if slot_type == 'non-categorical' and v != 'dontcare':
|
| 113 |
+
|
| 114 |
+
start = utt.lower().find(v)
|
| 115 |
+
|
| 116 |
+
if start != -1:
|
| 117 |
+
end = start + len(v)
|
| 118 |
+
converted[slot_type][-1]['start'] = start
|
| 119 |
+
converted[slot_type][-1]['end'] = end
|
| 120 |
+
converted[slot_type][-1]['value'] = utt[start:end]
|
| 121 |
+
|
| 122 |
+
return converted
|
| 123 |
+
|
| 124 |
+
|
| 125 |
+
def preprocess():
|
| 126 |
+
original_data_dir = 'woz'
|
| 127 |
+
new_data_dir = 'data'
|
| 128 |
+
os.makedirs(new_data_dir, exist_ok=True)
|
| 129 |
+
|
| 130 |
+
dataset = 'woz'
|
| 131 |
+
splits = ['train', 'validation', 'test']
|
| 132 |
+
domain = 'restaurant'
|
| 133 |
+
dialogues_by_split = {split: [] for split in splits}
|
| 134 |
+
global ontology
|
| 135 |
+
|
| 136 |
+
for split in splits:
|
| 137 |
+
if split != 'validation':
|
| 138 |
+
filename = os.path.join(original_data_dir, f'woz_{split}_en.json')
|
| 139 |
+
else:
|
| 140 |
+
filename = os.path.join(original_data_dir, 'woz_validate_en.json')
|
| 141 |
+
if not os.path.exists(filename):
|
| 142 |
+
raise FileNotFoundError(
|
| 143 |
+
f'cannot find {filename}, should manually download from https://github.com/nmrksic/neural-belief-tracker/tree/master/data/woz')
|
| 144 |
+
|
| 145 |
+
data = json.load(open(filename))
|
| 146 |
+
|
| 147 |
+
for item in data:
|
| 148 |
+
dialogue = {
|
| 149 |
+
'dataset': dataset,
|
| 150 |
+
'data_split': split,
|
| 151 |
+
'dialogue_id': f'{dataset}-{split}-{len(dialogues_by_split[split])}',
|
| 152 |
+
'original_id': item['dialogue_idx'],
|
| 153 |
+
'domains': [domain],
|
| 154 |
+
'turns': []
|
| 155 |
+
}
|
| 156 |
+
|
| 157 |
+
turns = item['dialogue']
|
| 158 |
+
n_turn = len(turns)
|
| 159 |
+
|
| 160 |
+
for i in range(n_turn):
|
| 161 |
+
sys_utt = turns[i]['system_transcript'].strip()
|
| 162 |
+
usr_utt = turns[i]['transcript'].strip()
|
| 163 |
+
usr_da = turns[i]['turn_label']
|
| 164 |
+
|
| 165 |
+
for s, v in usr_da:
|
| 166 |
+
if s == 'request':
|
| 167 |
+
assert v in ontology['domains']['restaurant']['slots']
|
| 168 |
+
else:
|
| 169 |
+
assert s in ontology['domains']['restaurant']['slots']
|
| 170 |
+
|
| 171 |
+
if i != 0:
|
| 172 |
+
dialogue['turns'].append({
|
| 173 |
+
'utt_idx': len(dialogue['turns']),
|
| 174 |
+
'speaker': 'system',
|
| 175 |
+
'utterance': sys_utt,
|
| 176 |
+
})
|
| 177 |
+
|
| 178 |
+
cur_state = copy.deepcopy(ontology['state'])
|
| 179 |
+
for act_slots in turns[i]['belief_state']:
|
| 180 |
+
act, slots = act_slots['act'], act_slots['slots']
|
| 181 |
+
if act == 'inform':
|
| 182 |
+
for s, v in slots:
|
| 183 |
+
v = v.strip()
|
| 184 |
+
if v != 'dontcare' and ontology['domains']['restaurant']['slots'][s]['is_categorical']:
|
| 185 |
+
if v not in ontology['domains']['restaurant']['slots'][s]['possible_values']:
|
| 186 |
+
if v == 'center':
|
| 187 |
+
v = 'centre'
|
| 188 |
+
elif v == 'east side':
|
| 189 |
+
v = 'east'
|
| 190 |
+
assert v in ontology['domains']['restaurant']['slots'][s]['possible_values']
|
| 191 |
+
|
| 192 |
+
cur_state[domain][s] = v
|
| 193 |
+
|
| 194 |
+
cur_usr_da = convert_da(usr_da, usr_utt)
|
| 195 |
+
|
| 196 |
+
# add to dialogue_acts dictionary in the ontology
|
| 197 |
+
for da_type in cur_usr_da:
|
| 198 |
+
das = cur_usr_da[da_type]
|
| 199 |
+
for da in das:
|
| 200 |
+
ontology["dialogue_acts"][da_type].setdefault((da['intent'], da['domain'], da['slot']), {})
|
| 201 |
+
ontology["dialogue_acts"][da_type][(da['intent'], da['domain'], da['slot'])]['user'] = True
|
| 202 |
+
|
| 203 |
+
dialogue['turns'].append({
|
| 204 |
+
'utt_idx': len(dialogue['turns']),
|
| 205 |
+
'speaker': 'user',
|
| 206 |
+
'utterance': usr_utt,
|
| 207 |
+
'state': cur_state,
|
| 208 |
+
'dialogue_acts': cur_usr_da,
|
| 209 |
+
})
|
| 210 |
+
|
| 211 |
+
dialogues_by_split[split].append(dialogue)
|
| 212 |
+
|
| 213 |
+
dialogues = []
|
| 214 |
+
for split in splits:
|
| 215 |
+
dialogues += dialogues_by_split[split]
|
| 216 |
+
for da_type in ontology['dialogue_acts']:
|
| 217 |
+
ontology["dialogue_acts"][da_type] = sorted([str(
|
| 218 |
+
{'user': speakers.get('user', False), 'system': speakers.get('system', False), 'intent': da[0],
|
| 219 |
+
'domain': da[1], 'slot': da[2]}) for da, speakers in ontology["dialogue_acts"][da_type].items()])
|
| 220 |
+
json.dump(dialogues[:10], open(f'dummy_data.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 221 |
+
json.dump(ontology, open(f'{new_data_dir}/ontology.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 222 |
+
json.dump(dialogues, open(f'{new_data_dir}/dialogues.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 223 |
+
with ZipFile('data.zip', 'w', ZIP_DEFLATED) as zf:
|
| 224 |
+
for filename in os.listdir(new_data_dir):
|
| 225 |
+
zf.write(f'{new_data_dir}/{filename}')
|
| 226 |
+
rmtree(original_data_dir)
|
| 227 |
+
rmtree(new_data_dir)
|
| 228 |
+
return dialogues, ontology
|
| 229 |
+
|
| 230 |
+
|
| 231 |
+
if __name__ == '__main__':
|
| 232 |
+
preprocess()
|