Upload preprocess.py
Browse files- preprocess.py +97 -0
preprocess.py
ADDED
|
@@ -0,0 +1,97 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import json
|
| 2 |
+
import os
|
| 3 |
+
from zipfile import ZipFile, ZIP_DEFLATED
|
| 4 |
+
import random
|
| 5 |
+
import json_lines
|
| 6 |
+
from collections import Counter
|
| 7 |
+
from shutil import rmtree
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
def preprocess():
|
| 11 |
+
random.seed(42)
|
| 12 |
+
|
| 13 |
+
ontology = {
|
| 14 |
+
'domains': {},
|
| 15 |
+
'intents': {},
|
| 16 |
+
'state': {},
|
| 17 |
+
"dialogue_acts": {
|
| 18 |
+
"categorical": {},
|
| 19 |
+
"non-categorical": {},
|
| 20 |
+
"binary": {}
|
| 21 |
+
}
|
| 22 |
+
}
|
| 23 |
+
|
| 24 |
+
dataset = 'metalwoz'
|
| 25 |
+
splits = ['train', 'validation', 'test']
|
| 26 |
+
dialogues_by_split = {split: [] for split in splits}
|
| 27 |
+
ZipFile('metalwoz-test-v1.zip').extract('dstc8_metalwoz_heldout.zip')
|
| 28 |
+
cnt = Counter()
|
| 29 |
+
for filename in ['metalwoz-v1.zip', 'dstc8_metalwoz_heldout.zip']:
|
| 30 |
+
with ZipFile(filename) as zipfile:
|
| 31 |
+
task_id2description = {x['task_id']: x for x in json_lines.reader(zipfile.open('tasks.txt'))}
|
| 32 |
+
for path in zipfile.namelist():
|
| 33 |
+
if path.startswith('dialogues'):
|
| 34 |
+
if filename == 'metalwoz-v1.zip':
|
| 35 |
+
split = random.choice(['train']*9+['validation'])
|
| 36 |
+
else:
|
| 37 |
+
split = 'test'
|
| 38 |
+
if split == 'validation':
|
| 39 |
+
print(path, split)
|
| 40 |
+
for ori_dialog in json_lines.reader(zipfile.open(path)):
|
| 41 |
+
dialogue_id = f'{dataset}-{split}-{len(dialogues_by_split[split])}'
|
| 42 |
+
domain = ori_dialog['domain']
|
| 43 |
+
|
| 44 |
+
task_des = task_id2description[ori_dialog['task_id']]
|
| 45 |
+
|
| 46 |
+
goal = {
|
| 47 |
+
'description': "user role: {}. user prompt: {}. system role: {}. system prompt: {}.".format(
|
| 48 |
+
task_des['user_role'], task_des['user_prompt'], task_des['bot_role'], task_des['bot_prompt']),
|
| 49 |
+
'inform': {},
|
| 50 |
+
'request': {}
|
| 51 |
+
}
|
| 52 |
+
|
| 53 |
+
dialogue = {
|
| 54 |
+
'dataset': dataset,
|
| 55 |
+
'data_split': split,
|
| 56 |
+
'dialogue_id': dialogue_id,
|
| 57 |
+
'original_id': ori_dialog['id'],
|
| 58 |
+
'domains': [domain], # will be updated by dialog_acts and state
|
| 59 |
+
'goal': goal,
|
| 60 |
+
'turns': []
|
| 61 |
+
}
|
| 62 |
+
|
| 63 |
+
ontology['domains'][domain] = {
|
| 64 |
+
'description': task_des['bot_role'],
|
| 65 |
+
'slots': {}
|
| 66 |
+
}
|
| 67 |
+
cnt[ori_dialog['turns'][0]] += 1
|
| 68 |
+
# assert ori_dialog['turns'][0] == "how may I help you?", print(ori_dialog['turns'])
|
| 69 |
+
for utt_idx, utt in enumerate(ori_dialog['turns'][1:]):
|
| 70 |
+
speaker = 'user' if utt_idx % 2 == 0 else 'system'
|
| 71 |
+
turn = {
|
| 72 |
+
'speaker': speaker,
|
| 73 |
+
'utterance': utt,
|
| 74 |
+
'utt_idx': utt_idx,
|
| 75 |
+
}
|
| 76 |
+
dialogue['turns'].append(turn)
|
| 77 |
+
|
| 78 |
+
dialogues_by_split[split].append(dialogue)
|
| 79 |
+
|
| 80 |
+
os.remove('dstc8_metalwoz_heldout.zip')
|
| 81 |
+
new_data_dir = 'data'
|
| 82 |
+
os.makedirs(new_data_dir, exist_ok=True)
|
| 83 |
+
dialogues = []
|
| 84 |
+
for split in splits:
|
| 85 |
+
dialogues += dialogues_by_split[split]
|
| 86 |
+
json.dump(dialogues[:10], open(f'dummy_data.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 87 |
+
json.dump(ontology, open(f'{new_data_dir}/ontology.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 88 |
+
json.dump(dialogues, open(f'{new_data_dir}/dialogues.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 89 |
+
with ZipFile('data.zip', 'w', ZIP_DEFLATED) as zf:
|
| 90 |
+
for filename in os.listdir(new_data_dir):
|
| 91 |
+
zf.write(f'{new_data_dir}/{filename}')
|
| 92 |
+
rmtree(new_data_dir)
|
| 93 |
+
return dialogues, ontology
|
| 94 |
+
|
| 95 |
+
|
| 96 |
+
if __name__ == '__main__':
|
| 97 |
+
preprocess()
|