Spaces:
Sleeping
Sleeping
File size: 17,622 Bytes
9eba547 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 | import os
import json
import faiss
import numpy as np
from tqdm import tqdm
import argparse
import logging
from typing import Any, Dict, List, Tuple, Union
from openai import OpenAI
from utils.model_loader import ModelLoader
# 配置日志
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger("RagTitleGenerator")
_GENERATOR_CACHE = {}
def _has_required_titles(data: Dict[str, Any]) -> bool:
titles = data.get("titles")
return (
isinstance(titles, dict)
and bool(str(titles.get("main_title", "")).strip())
and bool(str(titles.get("sub_title", "")).strip())
)
def _get_generator(
index_path: str,
data_path: str,
embed_model_path: str,
api_key: str,
base_url: str
) -> "RagTitleGenerator":
key = (index_path, data_path, embed_model_path, api_key, base_url)
generator = _GENERATOR_CACHE.get(key)
if generator is None:
generator = RagTitleGenerator(
index_path=index_path,
data_path=data_path,
embed_model_path=embed_model_path,
api_key=api_key,
base_url=base_url
)
_GENERATOR_CACHE[key] = generator
return generator
class RagTitleGenerator:
"""
RagTitleGenerator is a class that handles both:
1. Building/maintaining the FAISS index from a given dataset.
2. Generating infographic titles and descriptions for a single data item using RAG.
"""
def __init__(
self,
index_path: str = "faiss_infographics.index",
data_path: str = "infographics_data.npy",
embed_model_path="",
api_key: str="",
base_url: str=""
) -> None:
"""
Initialize the RagTitleGenerator. It attempts to load any existing FAISS index
and corresponding training data from disk. If no index file is found, the index
will be None until build_faiss_index is called.
Args:
index_path (str, optional): Path where the FAISS index file is or will be stored.
data_path (str, optional): Path where the training data embeddings are stored.
embed_model_path: Custom embedding model. If None, defaults to SentenceTransformer("all-MiniLM-L6-v2").
"""
self.index_path = index_path
self.data_path = data_path
if embed_model_path:
self.embed_model = ModelLoader.get_model(embed_model_path)
else:
print("fuck")
#else:
# self.embed_model = SentenceTransformer("all-MiniLM-L6-v2")
self.index = None
self.training_data = [] # List of tuples: (input_text, title, description)
print(api_key, base_url)
self.client = OpenAI(
api_key=api_key,
base_url=base_url
# "https://aihubmix.com/v1"
)
# Try loading an existing FAISS index and data
if os.path.exists(self.index_path) and os.path.exists(self.data_path):
print("Load existing FAISS index from disk.")
self.index = faiss.read_index(self.index_path)
with open(self.data_path, "rb") as f:
self.training_data = np.load(f, allow_pickle=True).tolist()
else:
print("No existing FAISS index found; you can build a new one via build_faiss_index().")
def build_faiss_index(self, data_json_path):
"""Build a FAISS index from a JSON dataset of chart data."""
try:
if not os.path.exists(data_json_path):
raise ValueError(f"Provided data_json_path {data_json_path} does not exist.")
# Remove existing index
self.clear_faiss_data()
# Load the entire dataset from JSON
with open(data_json_path, "r", encoding="utf-8") as f:
dataset = json.load(f)
print("dataset", type(dataset))
# Build the FAISS index from scratch
if isinstance(dataset, dict):
for name, details in tqdm(dataset.items(), desc="Building new FAISS index"):
processed_text = self.process_single_data(details)
title = details.get("metadata", {}).get("title", "")
description = details.get("metadata", {}).get("description", "")
self.add_training_data(processed_text, title, description)
else:
raise ValueError("Dataset must be a dictionary")
except Exception as e:
logger.error(f"处理记录时出错: {str(e)}")
raise
def clear_faiss_data(self) -> None:
"""Remove existing FAISS index and training data from disk."""
if os.path.exists(self.index_path):
os.remove(self.index_path)
if os.path.exists(self.data_path):
os.remove(self.data_path)
self.index = None
self.training_data = []
print("Original FAISS has been cleared.")
def process_single_data(
self,
data: Dict[str, Any]
) -> str:
"""
Convert a single data dictionary into a textual representation for retrieval and generation.
This includes metadata, chart_type, datafacts, etc.
Args:
data (Dict[str, Any]): A dictionary containing chart data, metadata, etc.
Returns:
str: A concatenated text representation of the data.
"""
metadata = data.get("metadata", {})
chart_type = data.get("chart_type", [])
datafacts = data.get("datafacts", [])
data_columns = data["data"].get("columns", [])
chart_data = data["data"].get("data", [])
main_insight = metadata.get("main_insight", "")
# Convert chart type to text
chart_type_text = "Chart Type: " + ", ".join(chart_type) + "\n" if chart_type else ""
# Convert datafacts to text
datafacts_text = "Data Facts:\n"
if datafacts:
for fact in datafacts:
annotation = fact.get("annotation", "")
if annotation:
datafacts_text += f"- {annotation}\n"
else:
datafacts_text = ""
# Main insight
main_insight_text = f"Main Insight: {main_insight}\n" if main_insight else ""
# Data columns
column_text = "Columns: "
if data_columns:
column_names = [col.get("name", "") for col in data_columns]
# column_names = [f"{col.get('name', '')} ({col.get('description', '')})" for col in data_columns]
column_text += ", ".join(column_names) + "\n"
else:
column_text = ""
# Data sample (limit to first few rows)
data_text = "Data Sample:\n"
for row in chart_data:
row_text = ", ".join([f"{k}: {v}" for k, v in row.items()])
data_text += f"{row_text}\n"
final_text = (
f"{chart_type_text}\n"
f"{datafacts_text}\n"
f"{main_insight_text}\n"
f"{column_text}\n"
f"{data_text}"
)
return final_text
def add_training_data(
self,
input_text: str,
title: str,
description: str
) -> None:
"""
Add a single training sample into the in-memory list and FAISS index.
After adding, it writes the updated index and data to disk.
Args:
input_text (str): Concatenated text derived from chart data.
title (str): Ground truth or known title from the data's metadata.
description (str): Ground truth or known description from the data's metadata.
"""
self.training_data.append((input_text, title, description))
embedding = self.embed_model.encode([input_text])
if self.index is None:
dim = embedding.shape[1]
self.index = faiss.IndexFlatL2(dim)
self.index.add(np.array(embedding))
faiss.write_index(self.index, self.index_path)
with open(self.data_path, "wb") as f:
np.save(f, np.array(self.training_data, dtype=object))
def retrieve_similar(
self,
new_input: str,
topk: int = 7
) -> List[Tuple[str, str, str]]:
"""
Retrieve the top-k most similar training samples from the FAISS index.
Args:
new_input (str): The new text query to encode.
topk (int, optional): Number of similar samples to retrieve.
Returns:
List[Tuple[str, str, str]]: A list of (input_text, title, description).
"""
if self.index is None or len(self.training_data) == 0:
return []
new_embedding = self.embed_model.encode([new_input])
topk = min(topk, len(self.training_data))
_, I = self.index.search(np.array(new_embedding), k=topk)
retrieved_list = []
for idx in I[0]:
data = self.training_data[idx]
retrieved_list.append((data[0], data[1], data[2]))
return retrieved_list
def generate_title_description(
self,
data: Dict[str, Any],
topk: int = 7
) -> Tuple[str, str]:
"""
Generate a title and subtitle for a single data dictionary using RAG.
Args:
data (Dict[str, Any]): The data dictionary containing metadata, chart_type, datafacts, etc.
topk (int, optional): Number of similar examples to retrieve for prompt augmentation.
Returns:
Tuple[str, str]: (generated_title, generated_description)
"""
processed_text = self.process_single_data(data)
retrieved_examples = self.retrieve_similar(processed_text, topk=topk) if topk > 0 else []
max_title_words = 8
max_description_words = 13
for _, rt, rd in retrieved_examples:
rt = rt or ""
rd = rd or ""
max_title_words = max(max_title_words, len(rt.split()))
max_description_words = max(max_description_words, len(rd.split()))
example_text = ""
if retrieved_examples:
example_text += "Here are some similar examples:\n"
for i, (r_data, rt, rd) in enumerate(retrieved_examples, start=1):
example_text += (
f"\n[Similar Example {i}]\n"
f"Input Data:\n{r_data}\n"
f"Title: {rt if rt else ''}\n"
f"Description: {rd if rd else ''}\n"
)
title_prompt = (
f"{example_text}\n"
"Based on the above examples (if any), please generate a clear and concise TITLE for the following data.\n"
f"{processed_text}\n\n"
"Important instructions:\n"
"1. The title should focus on the most significant feature of the data. "
"You can choose one or more key insights from the Data Facts that best "
"illustrate the issue, or identify the most notable feature yourself.\n"
# "2. The title should focus solely on what the data is about, without analyzing specific "
# "data characteristics, trends, distributions, or comparisons.\n"
f"2. The title should be strictly under {max_title_words} words.\n"
"3. Use exact terminology from the data sources.\n"
"4. Do NOT use these verbs: show, reveal, illustrate, analyze.\n"
"5. ONLY return the title as a string, no extra text."
)
response_title = self.client.chat.completions.create(
model="gpt-4o-mini",
messages=[
{"role": "system", "content": "You are an AI assistant that generates infographic titles."},
{"role": "user", "content": title_prompt}
]
)
generated_title = response_title.choices[0].message.content
generated_title = generated_title.strip() if generated_title else ""
description_prompt = (
f"{example_text}\n"
"Based on the above examples (if any), please generate a precise DESCRIPTION for the following data.\n"
f"{processed_text}\n\n"
"Important instructions:\n"
"1. Do NOT describe statistical properties (e.g., highest/lowest values, changes over time, "
"ratios, percentages). Simply summarize what the dataset reports.\n"
"2. Use one of the following structured templates where applicable (choose the highest-priority one that fits):\n"
" - Share/Percentage of [group] (who [action/characteristic]) (by [region/timeframe] (in [units])).\n"
" - Number/Total/Amount of [entity] (in [region/timeframe]) (, measured in [units]).\n"
" - Top/Leading N [entities] by [indicator] (, in [timeframe/region]).\n"
" - [Indicator] for [group/topic] (in [region/timeframe]) (, measured in [units]).\n"
f"3. The description should be strictly under {max_description_words} words.\n"
"4. Use exact terminology from the data sources.\n"
"5. Do NOT use these verbs: show, reveal, illustrate, analyze.\n"
"6. ONLY return the description as a string, no extra text."
)
response_description = self.client.chat.completions.create(
model="gpt-4o-mini",
messages=[
{"role": "system", "content": "You are an AI assistant that generates infographic descriptions."},
{"role": "user", "content": description_prompt}
]
)
generated_description = response_description.choices[0].message.content
generated_description = generated_description.strip() if generated_description else ""
return generated_title, generated_description
def process(
input: str = None,
output: str = None,
input_data: Dict = None,
index_path: str = "faiss_infographics.index",
data_path: str = "infographics_data.npy",
topk: int = 7,
embed_model_path = "",
api_key: str="",
base_url: str=""
) -> Union[bool, Dict]:
"""
Process function for generating the title and subtitle for a single data object.
Args:
input (str, optional): Path to the input JSON file with a single data object.
output (str, optional): Path to the output JSON file.
input_data (Dict, optional): A single data dictionary (alternative to file input).
index_path (str, optional): Path to the FAISS index file.
data_path (str, optional): Path to the training data embeddings file.
topk (int, optional): Number of similar examples to retrieve.
Returns:
Union[bool, Dict]:
- If output is provided, returns True/False indicating success/failure.
- Otherwise, returns the updated data dictionary with generated titles.
"""
try:
# Load the single data object
if input_data is None:
if input is None:
return False
with open(input, 'r', encoding='utf-8') as f:
data = json.load(f)
else:
data = input_data
if output is None:
output = input
if _has_required_titles(data):
if output:
with open(output, 'w', encoding='utf-8') as f:
json.dump(data, f, ensure_ascii=False, indent=2)
return True
generator = _get_generator(
index_path=index_path,
data_path=data_path,
embed_model_path=embed_model_path,
api_key=api_key,
base_url=base_url
)
main_title, sub_title = generator.generate_title_description(data, topk=topk)
if "titles" not in data:
data["titles"] = {}
data["titles"]["main_title"] = main_title
data["titles"]["sub_title"] = sub_title
if output:
with open(output, 'w', encoding='utf-8') as f:
json.dump(data, f, ensure_ascii=False, indent=2)
return True
return True
except Exception as e:
print(f"Error in title generation: {str(e)}")
return False
def main():
parser = argparse.ArgumentParser(description='Title generator for chart data')
parser.add_argument('--input', type=str, required=True, help='Input JSON file path (single data object).')
parser.add_argument('--output', type=str, required=True, help='Output JSON file path.')
parser.add_argument('--index_path', type=str, default='faiss_infographics.index', help='FAISS index file path.')
parser.add_argument('--data_path', type=str, default='infographics_data.npy', help='Training data path.')
parser.add_argument('--topk', type=int, default=3, help='Number of similar examples to retrieve.')
parser.add_argument('--embed_model_path', type=str, default='', help='Sentence transformer path')
parser.add_argument('--api_key', type=str, default='', help='API key for LLM.')
parser.add_argument('--base_url', type=str, default='', help='Base URL for LLM.')
args = parser.parse_args()
success = process(
input=args.input,
output=args.output,
index_path=args.index_path,
data_path=args.data_path,
topk=args.topk,
embed_model_path=args.embed_model_path,
api_key=args.api_key,
base_url=args.base_url
)
if success:
print("Title generation completed successfully.")
else:
print("Title generation failed.")
if __name__ == "__main__":
main()
|