Upload preprocess.py
Browse files- preprocess.py +156 -0
preprocess.py
ADDED
|
@@ -0,0 +1,156 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from zipfile import ZipFile, ZIP_DEFLATED
|
| 2 |
+
from shutil import rmtree
|
| 3 |
+
import json
|
| 4 |
+
import os
|
| 5 |
+
from tqdm import tqdm
|
| 6 |
+
from collections import Counter
|
| 7 |
+
from pprint import pprint
|
| 8 |
+
from nltk.tokenize import sent_tokenize, word_tokenize
|
| 9 |
+
from nltk.tokenize.treebank import TreebankWordDetokenizer
|
| 10 |
+
import re
|
| 11 |
+
|
| 12 |
+
topic_map = {
|
| 13 |
+
1: "Ordinary Life",
|
| 14 |
+
2: "School Life",
|
| 15 |
+
3: "Culture & Education",
|
| 16 |
+
4: "Attitude & Emotion",
|
| 17 |
+
5: "Relationship",
|
| 18 |
+
6: "Tourism",
|
| 19 |
+
7: "Health",
|
| 20 |
+
8: "Work",
|
| 21 |
+
9: "Politics",
|
| 22 |
+
10: "Finance"
|
| 23 |
+
}
|
| 24 |
+
|
| 25 |
+
act_map = {
|
| 26 |
+
1: "inform",
|
| 27 |
+
2: "question",
|
| 28 |
+
3: "directive",
|
| 29 |
+
4: "commissive"
|
| 30 |
+
}
|
| 31 |
+
|
| 32 |
+
emotion_map = {
|
| 33 |
+
0: "no emotion",
|
| 34 |
+
1: "anger",
|
| 35 |
+
2: "disgust",
|
| 36 |
+
3: "fear",
|
| 37 |
+
4: "happiness",
|
| 38 |
+
5: "sadness",
|
| 39 |
+
6: "surprise"
|
| 40 |
+
}
|
| 41 |
+
|
| 42 |
+
def preprocess():
|
| 43 |
+
original_data_dir = 'ijcnlp_dailydialog'
|
| 44 |
+
new_data_dir = 'data'
|
| 45 |
+
|
| 46 |
+
if not os.path.exists(original_data_dir):
|
| 47 |
+
original_data_zip = 'ijcnlp_dailydialog.zip'
|
| 48 |
+
if not os.path.exists(original_data_zip):
|
| 49 |
+
raise FileNotFoundError(f'cannot find original data {original_data_zip} in dailydialog/, should manually download ijcnlp_dailydialog.zip from http://yanran.li/files/ijcnlp_dailydialog.zip')
|
| 50 |
+
else:
|
| 51 |
+
archive = ZipFile(original_data_zip)
|
| 52 |
+
archive.extractall()
|
| 53 |
+
|
| 54 |
+
os.makedirs(new_data_dir, exist_ok=True)
|
| 55 |
+
|
| 56 |
+
dataset = 'dailydialog'
|
| 57 |
+
splits = ['train', 'validation', 'test']
|
| 58 |
+
dialogues_by_split = {split:[] for split in splits}
|
| 59 |
+
dial2topics = {}
|
| 60 |
+
with open(os.path.join(original_data_dir, 'dialogues_text.txt')) as dialog_file, \
|
| 61 |
+
open(os.path.join(original_data_dir, 'dialogues_topic.txt')) as topic_file:
|
| 62 |
+
for dialog, topic in zip(dialog_file, topic_file):
|
| 63 |
+
topic = int(topic.strip())
|
| 64 |
+
dialog = dialog.replace(' __eou__ ', ' ')
|
| 65 |
+
if dialog in dial2topics:
|
| 66 |
+
dial2topics[dialog].append(topic)
|
| 67 |
+
else:
|
| 68 |
+
dial2topics[dialog] = [topic]
|
| 69 |
+
|
| 70 |
+
global topic_map, act_map, emotion_map
|
| 71 |
+
|
| 72 |
+
ontology = {'domains': {x:{'description': '', 'slots': {}} for x in topic_map.values()},
|
| 73 |
+
'intents': {x:{'description': ''} for x in act_map.values()},
|
| 74 |
+
'state': {},
|
| 75 |
+
'dialogue_acts': {
|
| 76 |
+
"categorical": [],
|
| 77 |
+
"non-categorical": [],
|
| 78 |
+
"binary": {}
|
| 79 |
+
}}
|
| 80 |
+
|
| 81 |
+
detokenizer = TreebankWordDetokenizer()
|
| 82 |
+
|
| 83 |
+
for data_split in splits:
|
| 84 |
+
archive = ZipFile(os.path.join(original_data_dir, f'{data_split}.zip'))
|
| 85 |
+
with archive.open(f'{data_split}/dialogues_{data_split}.txt') as dialog_file, \
|
| 86 |
+
archive.open(f'{data_split}/dialogues_act_{data_split}.txt') as act_file, \
|
| 87 |
+
archive.open(f'{data_split}/dialogues_emotion_{data_split}.txt') as emotion_file:
|
| 88 |
+
for dialog_line, act_line, emotion_line in tqdm(zip(dialog_file, act_file, emotion_file)):
|
| 89 |
+
if not dialog_line.strip():
|
| 90 |
+
break
|
| 91 |
+
utts = dialog_line.decode().split("__eou__")[:-1]
|
| 92 |
+
acts = act_line.decode().split(" ")[:-1]
|
| 93 |
+
emotions = emotion_line.decode().split(" ")[:-1]
|
| 94 |
+
assert (len(utts) == len(acts) == len(emotions)), "Different turns btw dialogue & emotion & action"
|
| 95 |
+
|
| 96 |
+
topics = dial2topics[dialog_line.decode().replace(' __eou__ ', ' ')]
|
| 97 |
+
topic = Counter(topics).most_common(1)[0][0]
|
| 98 |
+
domain = topic_map[topic]
|
| 99 |
+
|
| 100 |
+
dialogue_id = f'{dataset}-{data_split}-{len(dialogues_by_split[data_split])}'
|
| 101 |
+
dialogue = {
|
| 102 |
+
'dataset': dataset,
|
| 103 |
+
'data_split': data_split,
|
| 104 |
+
'dialogue_id': dialogue_id,
|
| 105 |
+
'original_id': f'{data_split}-{len(dialogues_by_split[data_split])}',
|
| 106 |
+
'domains': [domain],
|
| 107 |
+
'turns': []
|
| 108 |
+
}
|
| 109 |
+
|
| 110 |
+
for utt, act, emotion in zip(utts, acts, emotions):
|
| 111 |
+
speaker = 'user' if len(dialogue['turns']) % 2 == 0 else 'system'
|
| 112 |
+
intent = act_map[int(act)]
|
| 113 |
+
emotion = emotion_map[int(emotion)]
|
| 114 |
+
# re-tokenize
|
| 115 |
+
utt = ' '.join([detokenizer.detokenize(word_tokenize(s)) for s in sent_tokenize(utt)])
|
| 116 |
+
# replace with common apostrophe
|
| 117 |
+
utt = utt.replace(' ’ ', "'")
|
| 118 |
+
# add space after full-stop
|
| 119 |
+
utt = re.sub('\.(?!com)(\w)', lambda x: '. '+x.group(1), utt)
|
| 120 |
+
|
| 121 |
+
dialogue['turns'].append({
|
| 122 |
+
'speaker': speaker,
|
| 123 |
+
'utterance': utt.strip(),
|
| 124 |
+
'utt_idx': len(dialogue['turns']),
|
| 125 |
+
'dialogue_acts': {
|
| 126 |
+
'binary': [{
|
| 127 |
+
'intent': intent,
|
| 128 |
+
'domain': '',
|
| 129 |
+
'slot': ''
|
| 130 |
+
}],
|
| 131 |
+
'categorical': [],
|
| 132 |
+
'non-categorical': [],
|
| 133 |
+
},
|
| 134 |
+
'emotion': emotion,
|
| 135 |
+
})
|
| 136 |
+
|
| 137 |
+
ontology["dialogue_acts"]['binary'].setdefault((intent, '', ''), {})
|
| 138 |
+
ontology["dialogue_acts"]['binary'][(intent, '', '')][speaker] = True
|
| 139 |
+
|
| 140 |
+
dialogues_by_split[data_split].append(dialogue)
|
| 141 |
+
|
| 142 |
+
ontology["dialogue_acts"]['binary'] = 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"]['binary'].items()])
|
| 143 |
+
dialogues = dialogues_by_split['train']+dialogues_by_split['validation']+dialogues_by_split['test']
|
| 144 |
+
json.dump(dialogues[:10], open(f'dummy_data.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 145 |
+
json.dump(ontology, open(f'{new_data_dir}/ontology.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 146 |
+
json.dump(dialogues, open(f'{new_data_dir}/dialogues.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 147 |
+
with ZipFile('data.zip', 'w', ZIP_DEFLATED) as zf:
|
| 148 |
+
for filename in os.listdir(new_data_dir):
|
| 149 |
+
zf.write(f'{new_data_dir}/{filename}')
|
| 150 |
+
rmtree(original_data_dir)
|
| 151 |
+
rmtree(new_data_dir)
|
| 152 |
+
return dialogues, ontology
|
| 153 |
+
|
| 154 |
+
|
| 155 |
+
if __name__ == '__main__':
|
| 156 |
+
preprocess()
|