File size: 21,995 Bytes
ac0f906 0e5b8f8 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 0e5b8f8 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 e83f5e9 ac0f906 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 |
import os
from pinecone import Pinecone
from dotenv import load_dotenv
import logging
from typing import Optional, List, Dict, Any, Union, Tuple
import time
from langchain_google_genai import GoogleGenerativeAIEmbeddings
import google.generativeai as genai
from langchain_core.retrievers import BaseRetriever
from langchain.callbacks.manager import Callbacks
from langchain_core.documents import Document
from langchain_core.pydantic_v1 import Field
# Configure logging
logger = logging.getLogger(__name__)
# Load environment variables
load_dotenv()
# Pinecone API key and index name
PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")
PINECONE_INDEX_NAME = os.getenv("PINECONE_INDEX_NAME")
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
# Pinecone retrieval configuration
DEFAULT_LIMIT_K = int(os.getenv("PINECONE_DEFAULT_LIMIT_K", "10"))
DEFAULT_TOP_K = int(os.getenv("PINECONE_DEFAULT_TOP_K", "6"))
DEFAULT_SIMILARITY_METRIC = os.getenv("PINECONE_DEFAULT_SIMILARITY_METRIC", "cosine")
DEFAULT_SIMILARITY_THRESHOLD = float(os.getenv("PINECONE_DEFAULT_SIMILARITY_THRESHOLD", "0.75"))
ALLOWED_METRICS = os.getenv("PINECONE_ALLOWED_METRICS", "cosine,dotproduct,euclidean").split(",")
# Export constants for importing elsewhere
__all__ = [
'get_pinecone_index',
'check_db_connection',
'search_vectors',
'upsert_vectors',
'delete_vectors',
'fetch_metadata',
'get_chain',
'DEFAULT_TOP_K',
'DEFAULT_LIMIT_K',
'DEFAULT_SIMILARITY_METRIC',
'DEFAULT_SIMILARITY_THRESHOLD',
'ALLOWED_METRICS',
'ThresholdRetriever'
]
# Configure Google API
if GOOGLE_API_KEY:
genai.configure(api_key=GOOGLE_API_KEY)
# Initialize global variables to store instances of Pinecone and index
pc = None
index = None
_retriever_instance = None
# Check environment variables
if not PINECONE_API_KEY:
logger.error("PINECONE_API_KEY is not set in environment variables")
if not PINECONE_INDEX_NAME:
logger.error("PINECONE_INDEX_NAME is not set in environment variables")
# Initialize Pinecone
def init_pinecone():
"""Initialize pinecone connection using new API"""
global pc, index
try:
# Only initialize if not already initialized
if pc is None:
logger.info(f"Initializing Pinecone connection to index {PINECONE_INDEX_NAME}...")
# Check if API key and index name are set
if not PINECONE_API_KEY:
logger.error("PINECONE_API_KEY is not set in environment variables")
return None
if not PINECONE_INDEX_NAME:
logger.error("PINECONE_INDEX_NAME is not set in environment variables")
return None
# Initialize Pinecone client using the new API
pc = Pinecone(api_key=PINECONE_API_KEY)
try:
# Check if index exists
index_list = pc.list_indexes()
if not hasattr(index_list, 'names') or PINECONE_INDEX_NAME not in index_list.names():
logger.error(f"Index {PINECONE_INDEX_NAME} does not exist in Pinecone")
return None
# Get existing index
index = pc.Index(PINECONE_INDEX_NAME)
logger.info(f"Pinecone connection established to index {PINECONE_INDEX_NAME}")
except Exception as connection_error:
logger.error(f"Error connecting to Pinecone index: {connection_error}")
return None
return index
except ImportError as e:
logger.error(f"Required package for Pinecone is missing: {e}")
return None
except Exception as e:
logger.error(f"Unexpected error initializing Pinecone: {e}")
return None
# Get Pinecone index singleton
def get_pinecone_index():
"""Get Pinecone index"""
global index
if index is None:
index = init_pinecone()
return index
# Check Pinecone connection
def check_db_connection():
"""Check Pinecone connection"""
try:
pinecone_index = get_pinecone_index()
if pinecone_index is None:
return False
# Check index information to confirm connection is working
stats = pinecone_index.describe_index_stats()
# Get total vector count from the new result structure
total_vectors = stats.get('total_vector_count', 0)
if hasattr(stats, 'namespaces'):
# If there are namespaces, calculate total vector count from namespaces
total_vectors = sum(ns.get('vector_count', 0) for ns in stats.namespaces.values())
logger.info(f"Pinecone connection is working. Total vectors: {total_vectors}")
return True
except Exception as e:
logger.error(f"Error in Pinecone connection: {e}")
return False
# Convert similarity score based on the metric
def convert_score(score: float, metric: str) -> float:
"""
Convert similarity score to a 0-1 scale based on the metric used.
For metrics like euclidean distance where lower is better, we invert the score.
Args:
score: The raw similarity score
metric: The similarity metric used
Returns:
A normalized score between 0-1 where higher means more similar
"""
if metric.lower() in ["euclidean", "l2"]:
# For distance metrics (lower is better), we inverse and normalize
# Assuming max reasonable distance is 2.0 for normalized vectors
return max(0, 1 - (score / 2.0))
else:
# For cosine and dot product (higher is better), return as is
return score
# Filter results based on similarity threshold
def filter_by_threshold(results, threshold: float, metric: str) -> List[Dict]:
"""
Filter query results based on similarity threshold.
Args:
results: The query results from Pinecone
threshold: The similarity threshold (0-1)
metric: The similarity metric used
Returns:
Filtered list of matches
"""
filtered_matches = []
if not hasattr(results, 'matches'):
return filtered_matches
for match in results.matches:
# Get the score
score = getattr(match, 'score', 0)
# Convert score based on metric
normalized_score = convert_score(score, metric)
# Filter based on threshold
if normalized_score >= threshold:
# Add normalized score as an additional attribute
match.normalized_score = normalized_score
filtered_matches.append(match)
return filtered_matches
# Search vectors in Pinecone with advanced options
async def search_vectors(
query_vector,
top_k: int = DEFAULT_TOP_K,
limit_k: int = DEFAULT_LIMIT_K,
similarity_metric: str = DEFAULT_SIMILARITY_METRIC,
similarity_threshold: float = DEFAULT_SIMILARITY_THRESHOLD,
namespace: str = "Default",
filter: Optional[Dict] = None
) -> Dict:
"""
Search for most similar vectors in Pinecone with advanced filtering options.
Args:
query_vector: The query vector
top_k: Number of results to return (after threshold filtering)
limit_k: Maximum number of results to retrieve from Pinecone
similarity_metric: Similarity metric to use (cosine, dotproduct, euclidean)
similarity_threshold: Threshold for similarity (0-1)
namespace: Namespace to search in
filter: Filter query
Returns:
Search results with matches filtered by threshold
"""
try:
# Validate parameters
if similarity_metric not in ALLOWED_METRICS:
logger.warning(f"Invalid similarity metric: {similarity_metric}. Using default: {DEFAULT_SIMILARITY_METRIC}")
similarity_metric = DEFAULT_SIMILARITY_METRIC
if limit_k < top_k:
logger.warning(f"limit_k ({limit_k}) must be greater than or equal to top_k ({top_k}). Setting limit_k to {top_k}")
limit_k = top_k
# Perform search directly without cache
pinecone_index = get_pinecone_index()
if pinecone_index is None:
logger.error("Failed to get Pinecone index for search")
return None
# Query Pinecone with the provided metric and higher limit_k to allow for threshold filtering
results = pinecone_index.query(
vector=query_vector,
top_k=limit_k, # Retrieve more results than needed to allow for threshold filtering
namespace=namespace,
filter=filter,
include_metadata=True,
include_values=False, # No need to return vector values to save bandwidth
metric=similarity_metric # Specify similarity metric
)
# Filter results by threshold
filtered_matches = filter_by_threshold(results, similarity_threshold, similarity_metric)
# Limit to top_k after filtering
filtered_matches = filtered_matches[:top_k]
# Create a new results object with filtered matches
results.matches = filtered_matches
# Log search result metrics
match_count = len(filtered_matches)
logger.info(f"Pinecone search returned {match_count} matches after threshold filtering (metric: {similarity_metric}, threshold: {similarity_threshold}, namespace: {namespace})")
return results
except Exception as e:
logger.error(f"Error searching vectors: {e}")
return None
# Upsert vectors to Pinecone
async def upsert_vectors(vectors, namespace="Default"):
"""Upsert vectors to Pinecone index"""
try:
pinecone_index = get_pinecone_index()
if pinecone_index is None:
logger.error("Failed to get Pinecone index for upsert")
return None
response = pinecone_index.upsert(
vectors=vectors,
namespace=namespace
)
# Log upsert metrics
upserted_count = response.get('upserted_count', 0)
logger.info(f"Upserted {upserted_count} vectors to Pinecone")
return response
except Exception as e:
logger.error(f"Error upserting vectors: {e}")
return None
# Delete vectors from Pinecone
async def delete_vectors(ids, namespace="Default"):
"""Delete vectors from Pinecone index"""
try:
pinecone_index = get_pinecone_index()
if pinecone_index is None:
logger.error("Failed to get Pinecone index for delete")
return False
response = pinecone_index.delete(
ids=ids,
namespace=namespace
)
logger.info(f"Deleted vectors with IDs {ids} from Pinecone")
return True
except Exception as e:
logger.error(f"Error deleting vectors: {e}")
return False
# Fetch vector metadata from Pinecone
async def fetch_metadata(ids, namespace="Default"):
"""Fetch metadata for specific vector IDs"""
try:
pinecone_index = get_pinecone_index()
if pinecone_index is None:
logger.error("Failed to get Pinecone index for fetch")
return None
response = pinecone_index.fetch(
ids=ids,
namespace=namespace
)
return response
except Exception as e:
logger.error(f"Error fetching vector metadata: {e}")
return None
# Create a custom retriever class for Langchain integration
class ThresholdRetriever(BaseRetriever):
"""
Custom retriever that supports threshold-based filtering and multiple similarity metrics.
This integrates with the Langchain ecosystem while using our advanced retrieval logic.
"""
vectorstore: Any = Field(description="Vector store to use for retrieval")
embeddings: Any = Field(description="Embeddings model to use for retrieval")
search_kwargs: Dict[str, Any] = Field(default_factory=dict, description="Search kwargs for the vectorstore")
top_k: int = Field(default=DEFAULT_TOP_K, description="Number of results to return after filtering")
limit_k: int = Field(default=DEFAULT_LIMIT_K, description="Maximum number of results to retrieve from Pinecone")
similarity_metric: str = Field(default=DEFAULT_SIMILARITY_METRIC, description="Similarity metric to use")
similarity_threshold: float = Field(default=DEFAULT_SIMILARITY_THRESHOLD, description="Threshold for similarity")
namespace: str = "Default"
class Config:
"""Configuration for this pydantic object."""
arbitrary_types_allowed = True
async def search_vectors_sync(
self, query_vector,
top_k: int = DEFAULT_TOP_K,
limit_k: int = DEFAULT_LIMIT_K,
similarity_metric: str = DEFAULT_SIMILARITY_METRIC,
similarity_threshold: float = DEFAULT_SIMILARITY_THRESHOLD,
namespace: str = "Default",
filter: Optional[Dict] = None
) -> Dict:
"""Synchronous wrapper for search_vectors"""
import asyncio
try:
# Get current event loop or create a new one
try:
loop = asyncio.get_event_loop()
except RuntimeError:
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
# Use event loop to run async function
if loop.is_running():
# If we're in an event loop, use asyncio.create_task
task = asyncio.create_task(search_vectors(
query_vector=query_vector,
top_k=top_k,
limit_k=limit_k,
similarity_metric=similarity_metric,
similarity_threshold=similarity_threshold,
namespace=namespace,
filter=filter
))
return await task
else:
# If not in an event loop, just await directly
return await search_vectors(
query_vector=query_vector,
top_k=top_k,
limit_k=limit_k,
similarity_metric=similarity_metric,
similarity_threshold=similarity_threshold,
namespace=namespace,
filter=filter
)
except Exception as e:
logger.error(f"Error in search_vectors_sync: {e}")
return None
def _get_relevant_documents(
self, query: str, *, run_manager: Callbacks = None
) -> List[Document]:
"""
Get documents relevant to the query using threshold-based retrieval.
Args:
query: The query string
run_manager: The callbacks manager
Returns:
List of relevant documents
"""
# Generate embedding for query using the embeddings model
try:
# Use the embeddings model we stored in the class
embedding = self.embeddings.embed_query(query)
except Exception as e:
logger.error(f"Error generating embedding: {e}")
# Fallback to creating a new embedding model if needed
embedding_model = GoogleGenerativeAIEmbeddings(model="models/embedding-001")
embedding = embedding_model.embed_query(query)
# Perform search with advanced options - avoid asyncio.run()
import asyncio
# Get or create event loop
try:
loop = asyncio.get_event_loop()
except RuntimeError:
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
# Run asynchronous search in a safe way
if loop.is_running():
# We're inside an existing event loop (like in FastAPI)
# Use a different approach - convert it to a synchronous call
from concurrent.futures import ThreadPoolExecutor
import functools
# Define a wrapper function to run in a thread
def run_async_in_thread():
# Create a new event loop for this thread
thread_loop = asyncio.new_event_loop()
asyncio.set_event_loop(thread_loop)
# Run the coroutine and return the result
return thread_loop.run_until_complete(search_vectors(
query_vector=embedding,
top_k=self.top_k,
limit_k=self.limit_k,
similarity_metric=self.similarity_metric,
similarity_threshold=self.similarity_threshold,
namespace=self.namespace,
# filter=self.search_kwargs.get("filter", None)
))
# Run the async function in a thread
with ThreadPoolExecutor() as executor:
search_result = executor.submit(run_async_in_thread).result()
else:
# No event loop running, we can use run_until_complete
search_result = loop.run_until_complete(search_vectors(
query_vector=embedding,
top_k=self.top_k,
limit_k=self.limit_k,
similarity_metric=self.similarity_metric,
similarity_threshold=self.similarity_threshold,
namespace=self.namespace,
# filter=self.search_kwargs.get("filter", None)
))
# Convert to documents
documents = []
if search_result and hasattr(search_result, 'matches'):
for match in search_result.matches:
# Extract metadata
metadata = {}
if hasattr(match, 'metadata'):
metadata = match.metadata
# Add score to metadata
score = getattr(match, 'score', 0)
normalized_score = getattr(match, 'normalized_score', score)
metadata['score'] = score
metadata['normalized_score'] = normalized_score
# Extract text
text = metadata.get('text', '')
if 'text' in metadata:
del metadata['text'] # Remove from metadata since it's the content
# Create Document
doc = Document(
page_content=text,
metadata=metadata
)
documents.append(doc)
return documents
# Get the retrieval chain with Pinecone vector store
def get_chain(
index_name=PINECONE_INDEX_NAME,
namespace="Default",
top_k=DEFAULT_TOP_K,
limit_k=DEFAULT_LIMIT_K,
similarity_metric=DEFAULT_SIMILARITY_METRIC,
similarity_threshold=DEFAULT_SIMILARITY_THRESHOLD
):
"""
Get the retrieval chain with Pinecone vector store using threshold-based retrieval.
Args:
index_name: Pinecone index name
namespace: Pinecone namespace
top_k: Number of results to return after filtering
limit_k: Maximum number of results to retrieve from Pinecone
similarity_metric: Similarity metric to use (cosine, dotproduct, euclidean)
similarity_threshold: Threshold for similarity (0-1)
Returns:
ThresholdRetriever instance
"""
global _retriever_instance
try:
# If already initialized with same parameters, return cached instance
if _retriever_instance is not None:
return _retriever_instance
start_time = time.time()
logger.info("Initializing new retriever chain with threshold-based filtering")
# Initialize embeddings model
embeddings = GoogleGenerativeAIEmbeddings(model="models/embedding-001")
# Get index
pinecone_index = get_pinecone_index()
if not pinecone_index:
logger.error("Failed to get Pinecone index for retriever chain")
return None
# Get statistics for logging
try:
stats = pinecone_index.describe_index_stats()
total_vectors = stats.get('total_vector_count', 0)
logger.info(f"Pinecone index stats - Total vectors: {total_vectors}")
except Exception as e:
logger.error(f"Error getting index stats: {e}")
# Use Pinecone from langchain_community.vectorstores
from langchain_community.vectorstores import Pinecone as LangchainPinecone
logger.info(f"Creating Pinecone vectorstore with index: {index_name}, namespace: {namespace}")
vectorstore = LangchainPinecone.from_existing_index(
embedding=embeddings,
index_name=index_name,
namespace=namespace,
text_key="text"
)
# Create threshold-based retriever
logger.info(f"Creating ThresholdRetriever with top_k={top_k}, limit_k={limit_k}, " +
f"metric={similarity_metric}, threshold={similarity_threshold}")
# Create ThresholdRetriever with both vectorstore and embeddings
_retriever_instance = ThresholdRetriever(
vectorstore=vectorstore,
embeddings=embeddings, # Pass embeddings separately
top_k=top_k,
limit_k=limit_k,
similarity_metric=similarity_metric,
similarity_threshold=similarity_threshold
)
logger.info(f"Pinecone retriever initialized in {time.time() - start_time:.2f} seconds")
return _retriever_instance
except Exception as e:
logger.error(f"Error creating retrieval chain: {e}")
return None |