Spaces:
Sleeping
Sleeping
File size: 9,179 Bytes
939a9f4 |
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 |
"""
Vector Store Module
===================
Purpose: Store embeddings and retrieve similar ones
This module uses Chroma for persistent, efficient vector storage.
Chroma is free, local, and production-ready.
Key Concepts:
• Vector storage: Persistent storage mapping chunk_id → embedding
• Metadata: Source info, text preview, etc.
• Retrieval: Find top-k most similar vectors using cosine similarity
• Persistence: Data survives application restarts
"""
from typing import List, Dict, Any
from dataclasses import dataclass, field
import logging
import chromadb
import os
logger = logging.getLogger(__name__)
@dataclass
class RetrievalResult:
"""A single retrieved chunk with metadata."""
chunk_id: str
text: str
similarity: float
metadata: Dict[str, Any] = field(default_factory=dict)
class ChromaVectorStore:
"""
Vector store using Chroma (persistent, free, production-ready).
Chroma is a modern vector database that:
• Stores embeddings persistently on disk
• Provides similarity search
• Is completely free and open source
• Works locally (no API calls)
This is the recommended implementation for production RAG systems.
"""
def __init__(self, persist_directory: str = ".chromadb", collection_name: str = "rag"):
"""
Initialize Chroma vector store.
Args:
persist_directory: Where to store vectors on disk
collection_name: Name of the collection (namespace)
Example:
>>> store = ChromaVectorStore(persist_directory="./data/vectors")
"""
self.persist_directory = persist_directory
self.collection_name = collection_name
# Ensure persist directory exists
os.makedirs(persist_directory, exist_ok=True)
try:
# Create persistent client
self.client = chromadb.PersistentClient(path=persist_directory)
# Get or create collection
self.collection = self.client.get_or_create_collection(
name=collection_name,
metadata={"hnsw:space": "cosine"} # Use cosine similarity
)
logger.info(
f"✓ Initialized Chroma vector store at {persist_directory} "
f"(collection: {collection_name})"
)
except Exception as e:
logger.error(f"Failed to initialize Chroma: {e}")
raise
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
try:
self.client.persist()
self.client.shutdown()
except Exception:
pass
def add(
self,
chunk_id: str,
text: str,
embedding: List[float],
metadata: Dict[str, Any] = None
) -> None:
"""
Add a chunk with its embedding to the store.
Args:
chunk_id: Unique identifier for chunk
text: Original text content
embedding: Vector representation (list of floats)
metadata: Optional metadata (source, page number, etc.)
Example:
>>> store.add(
... "doc1_chunk_0",
... "Machine learning is AI",
... [0.1, 0.2, ..., 0.384],
... metadata={"doc_id": "doc1", "page": 1}
... )
"""
try:
self.collection.add(
ids=[chunk_id],
documents=[text],
embeddings=[embedding],
metadatas=[metadata or {}]
)
logger.debug(f"Added chunk {chunk_id} ({len(text)} chars)")
except Exception as e:
logger.error(f"Failed to add chunk {chunk_id}: {e}")
raise
def retrieve(
self,
query_embedding: List[float],
top_k: int = 5
) -> List[RetrievalResult]:
"""
Find most similar chunks to query.
Args:
query_embedding: Query vector
top_k: Number of results to return
Returns:
List of RetrievalResult objects, sorted by similarity (highest first)
Example:
>>> results = store.retrieve(query_embedding, top_k=3)
>>> for r in results:
... print(f"{r.similarity:.3f} | {r.text[:60]}")
"""
try:
if self.collection.count() == 0:
logger.warning("Vector store is empty")
return []
# Query Chroma
results = self.collection.query(
query_embeddings=[query_embedding],
n_results=top_k
)
if not results["ids"] or not results["ids"][0]:
logger.debug("No results found for query")
return []
# Convert to RetrievalResult objects
retrieval_results = []
for i, chunk_id in enumerate(results["ids"][0]):
# Chroma returns distances, convert to similarity (1 - distance for cosine)
# Note: Chroma with cosine metric returns distances
distance = results["distances"][0][i]
similarity = 1 - distance # Convert distance to similarity
result = RetrievalResult(
chunk_id=chunk_id,
text=results["documents"][0][i],
similarity=similarity,
metadata=results["metadatas"][0][i]
)
retrieval_results.append(result)
logger.debug(f"Retrieved {len(retrieval_results)} chunks")
return retrieval_results
except Exception as e:
logger.error(f"Retrieval failed: {e}")
raise
def size(self) -> int:
"""Return number of chunks in store."""
try:
count = self.collection.count()
return count
except Exception as e:
logger.error(f"Failed to get store size: {e}")
return 0
def delete(self, chunk_id: str) -> bool:
"""
Delete a chunk from the store.
Args:
chunk_id: ID of chunk to delete
Returns:
True if deleted, False if not found
"""
try:
self.collection.delete(ids=[chunk_id])
logger.debug(f"Deleted chunk {chunk_id}")
return True
except Exception as e:
logger.error(f"Failed to delete chunk {chunk_id}: {e}")
return False
def clear(self) -> None:
"""Clear all vectors from store."""
try:
# Get all IDs and delete them
all_data = self.collection.get()
if all_data["ids"]:
self.collection.delete(ids=all_data["ids"])
logger.info("Cleared vector store")
except Exception as e:
logger.error(f"Failed to clear store: {e}")
raise
# ============ TESTS ============
import tempfile
import shutil
import time
def test_chroma_vector_store():
temp_dir = tempfile.mkdtemp()
store = ChromaVectorStore(persist_directory=temp_dir)
try:
# Add chunks
vec1 = [1.0, 0.0, 0.0]
vec2 = [0.9, 0.1, 0.0]
vec3 = [0.0, 1.0, 0.0]
store.add("chunk1", "Machine learning", vec1, metadata={"source": "test"})
store.add("chunk2", "Deep learning networks", vec2, metadata={"source": "test"})
store.add("chunk3", "Cooking recipes", vec3, metadata={"source": "test"})
# Retrieve
results = store.retrieve(vec1, top_k=2)
assert len(results) == 2
assert results[0].chunk_id == "chunk1"
print("✓ Chroma test passed!")
finally:
# Cleanup Chroma resources
try:
if hasattr(store, "client"):
store.client.close()
del store.client
del store.collection
except Exception as e:
logger.warning(f"Error closing Chroma client: {e}")
# Give Windows time to release file handles
time.sleep(1.0)
# Retry logic for Windows file deletion
retry_count = 0
max_retries = 5
while retry_count < max_retries:
try:
shutil.rmtree(temp_dir)
break
except PermissionError:
retry_count += 1
if retry_count < max_retries:
time.sleep(0.5)
else:
logger.warning(f"Could not delete temp directory {temp_dir}, skipping")
break
if __name__ == "__main__":
logging.basicConfig(level=logging.INFO)
# Test Chroma
try:
test_chroma_vector_store()
except ImportError:
print("Chroma not installed, skipping test")
# Test SimpleVectorStore
test_simple_vector_store() |