-
Notifications
You must be signed in to change notification settings - Fork 113
/
generate_qa.py
245 lines (217 loc) · 9.56 KB
/
generate_qa.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
import argparse
import glob
import io
import json
import multiprocessing
import os
import time
from xml.etree import ElementTree
import cairosvg
import requests
from openai import OpenAI
from PIL import Image
from ratelimit import limits, sleep_and_retry
RATE_LIMIT = 7500 # as per your resources
RATE_LIMIT_INTERVAL = 60 # Time interval of 1 minute
OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
client = OpenAI(api_key=OPENAI_API_KEY)
user_agent = os.getenv('USER_AGENT')
session = requests.Session()
session.headers.update({'User-Agent': user_agent})
num_processes = 4 # can go for 32 or 64 based on no. of cpus
processed_queue = multiprocessing.Queue()
def convert_thumb_url(url):
parts = url.split('/')
# Remove the 'thumb' part
if 'thumb' in parts:
parts.remove('thumb')
# Detect if the last segment contains a size indication and remove it
if 'px' in parts[-1]:
parts.pop() # Remove the last part, which is the size-indicated duplicate
original_url = '/'.join(parts)
return original_url
def download_image(session, image_url, filename, save_directory, delay=1):
try:
# Wait for a specified delay between requests to avoid rate limiting
time.sleep(delay)
response = session.get(image_url)
response.raise_for_status()
if not os.path.exists(save_directory):
os.makedirs(save_directory)
image_format = image_url.split('.')[-1].lower().split('?')[0]
file_path = os.path.join(save_directory, f"{filename}.png")
if image_format == 'svg':
# Parse the SVG
svg_root = ElementTree.fromstring(response.content)
# Set the width and height if not set (necessary for rendering)
if 'width' not in svg_root.attrib:
svg_root.set('width', '800')
if 'height' not in svg_root.attrib:
svg_root.set('height', '600')
# Create a white background rectangle
background = ElementTree.Element(
'rect', width='100%', height='100%', x='0', y='0', fill='white')
# Insert the background rectangle at the beginning of the SVG file
svg_root.insert(0, background)
# Get the modified SVG content as a string
modified_svg = ElementTree.tostring(
svg_root, encoding='utf-8', method='xml')
# Convert the modified SVG to PNG
cairosvg.svg2png(bytestring=modified_svg, write_to=file_path)
elif image_format in ['gif', 'jpg']:
# Handle GIF and JPG files
with Image.open(io.BytesIO(response.content)) as image:
# Convert to RGBA to manage any transparency in GIFs
image = image.convert('RGBA')
if image_format == 'gif':
# Create a white canvas for GIFs
canvas = Image.new('RGBA', image.size, 'WHITE')
canvas.paste(image, mask=image)
image = canvas.convert('RGB') # Convert back to RGB
image.save(file_path, 'PNG') # Save as PNG
else:
# Directly save other formats as PNG
with open(file_path, 'wb') as f:
f.write(response.content)
return 1
except Exception as e:
print(f"Failed to download {image_url}: {e}")
return 0
def get_topic_files(parent_directory, field, topics_directory):
topic_files = []
field_file_path = os.path.join(topics_directory, f'{field}.json')
with open(field_file_path, 'r') as file:
field_data = json.load(file)
# print(field_name)
for subfield, topics in field_data.items():
for topic in topics:
topic_files.append(os.path.join(
parent_directory, subfield, f'{topic}.json'))
print(topic_files)
return topic_files
def get_data(topic_files):
all_data = []
for file_path in topic_files:
if os.path.exists(file_path):
with open(file_path, 'r') as file:
data = json.load(file)
print(f'Processing {file_path}')
# Filter sections with less than 50 words in the 'text'
filtered_data = [item for item in data if len(
item['text'].split()) > 50]
all_data.extend(filtered_data)
else:
print(f'Skipping {file_path} as the file does not exist.')
print(len(all_data))
return all_data
# return all_data[:50]
@sleep_and_retry
@limits(calls=RATE_LIMIT, period=RATE_LIMIT_INTERVAL)
def process_data_point(data_point, client, proc_index, data_point_index, image_directory):
processed_data_points = []
curr_data_point_index = data_point_index
for image_index, image_info in enumerate(data_point['images']):
image_url = image_info['url']
image_caption = image_info['caption']
image_id = image_url.split('/')[-1].split('.')[0]
# Download the image
customurl = convert_thumb_url(image_url)
current_id = f"img_{proc_index}_{curr_data_point_index}"
status = download_image(
session, customurl, current_id, image_directory)
if status == 1:
# Generate question and answer using OpenAI
completion = client.chat.completions.create(
model="gpt-3.5-turbo",
messages=[
{
"role": "system",
"content": "Generate a question and answer pair based on the image's caption, its contextual text, and associated metadata. The question should be insightful, engaging, and directly related to the image's content as described by the caption. The answer should be concise and informative, drawing upon the text and contextual details provided."
},
{
"role": "user",
"content": json.dumps(dict(
data_point,
image_url=image_url,
caption=image_caption
))
}
]
)
response = completion.choices[0].message.content
print(response)
processed_data_point = {
"id": current_id,
"image_id": image_id,
"image_url": image_url,
"text": data_point['text'],
"caption": image_caption,
"section": data_point['section'],
"subfield": data_point['subfield'],
"field": data_point['field'],
"topic": data_point['topic'],
"pagelink": data_point['link'],
"openairesponse": response
}
processed_data_points.append(processed_data_point)
curr_data_point_index += 1
return processed_data_points, curr_data_point_index
def worker(data_chunk, proc_index, result_list, image_directory):
client = OpenAI(api_key=OPENAI_API_KEY)
curr_data_point_index = 0
for i, data_point in enumerate(data_chunk):
results, curr_data_point_index = process_data_point(
data_point, client, proc_index, curr_data_point_index, image_directory)
for result in results:
result_list.append(result)
def main(topics_dir, data_dir, image_directory, qa_directory):
os.makedirs(image_directory, exist_ok=True)
os.makedirs(qa_directory, exist_ok=True)
fields = [
os.path.splitext(f)[0]
for f in os.listdir(topics_dir)
if f.endswith('.json')
]
for field in fields:
field_image_directory = f'{image_directory}/{field}_images'
os.makedirs(field_image_directory, exist_ok=True)
id = len(glob.glob(os.path.join(field_image_directory, '*')))
print(f'starting images with id {id}')
topic_files = get_topic_files(data_dir, field, topics_dir)
filtered_data = get_data(topic_files)
processed_data_file_path = f'{qa_directory}/{field}.json'
print(processed_data_file_path)
if os.path.exists(processed_data_file_path):
with open(processed_data_file_path, 'r') as file:
processed_data = json.load(file)
else:
processed_data = []
manager = multiprocessing.Manager()
results = manager.list()
# Split data into chunks for each process
data_chunks = [filtered_data[i::num_processes]
for i in range(num_processes)]
processes = []
for index, chunk in enumerate(data_chunks):
p = multiprocessing.Process(target=worker, args=(
chunk, index, results, field_image_directory))
processes.append(p)
p.start()
for p in processes:
p.join()
processed_data.extend(results)
with open(processed_data_file_path, 'w') as f:
json.dump(list(processed_data), f, indent=4)
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Process some topics.')
parser.add_argument('--topics_dir', type=str,
default='./data/topics', help='Directory of topics')
parser.add_argument('--data_dir', type=str,
default='./data/wikidata/data/', help='Data path')
parser.add_argument('--image_dir', type=str,
default='./data/images', help='Directory to store images')
parser.add_argument('--qa_dir', type=str,
default='./data/qadata', help='Directory to store q&a processed')
args = parser.parse_args()
main(args.topics_dir, args.data_dir,
args.image_dir, args.qa_dir)