Spaces:
Sleeping
Sleeping
Update app/policy_vector_db.py
Browse files- app/policy_vector_db.py +17 -16
app/policy_vector_db.py
CHANGED
|
@@ -40,39 +40,40 @@ class PolicyVectorDB:
|
|
| 40 |
try:
|
| 41 |
existing_ids = set(collection.get(include=[])['ids'])
|
| 42 |
except Exception as e:
|
| 43 |
-
logger.warning(f"Could not retrieve existing IDs from ChromaDB: {e}")
|
| 44 |
|
| 45 |
new_chunks = [chunk for chunk in chunks if chunk.get('id') and chunk['id'] not in existing_ids]
|
|
|
|
| 46 |
if not new_chunks:
|
| 47 |
logger.info("No new chunks to add.")
|
| 48 |
return
|
| 49 |
|
|
|
|
| 50 |
batch_size = 128
|
| 51 |
for i in range(0, len(new_chunks), batch_size):
|
| 52 |
batch = new_chunks[i:i + batch_size]
|
| 53 |
texts = [chunk['text'] for chunk in batch]
|
| 54 |
ids = [chunk['id'] for chunk in batch]
|
| 55 |
-
metadatas = [self._flatten_metadata(chunk.get('metadata'
|
| 56 |
embeddings = self.embedding_model.encode(texts, show_progress_bar=False).tolist()
|
| 57 |
collection.add(ids=ids, embeddings=embeddings, documents=texts, metadatas=metadatas)
|
| 58 |
-
logger.info(f"Added batch {i
|
| 59 |
logger.info(f"Finished adding {len(new_chunks)} chunks.")
|
| 60 |
|
| 61 |
def search(self, query_text: str, top_k: int = None) -> List[Dict]:
|
| 62 |
collection = self._get_collection()
|
| 63 |
query_embedding = self.embedding_model.encode([query_text]).tolist()
|
| 64 |
-
top_k = top_k
|
| 65 |
-
|
| 66 |
results = collection.query(
|
| 67 |
query_embeddings=query_embedding,
|
| 68 |
n_results=top_k,
|
| 69 |
include=["documents", "metadatas", "distances"]
|
| 70 |
)
|
| 71 |
-
|
| 72 |
search_results = []
|
| 73 |
-
if results and results['documents'][0]:
|
| 74 |
for i, doc in enumerate(results['documents'][0]):
|
| 75 |
-
relevance_score = 1 - results['distances'][0][i]
|
| 76 |
search_results.append({
|
| 77 |
'text': doc,
|
| 78 |
'metadata': results['metadatas'][0][i],
|
|
@@ -83,24 +84,24 @@ class PolicyVectorDB:
|
|
| 83 |
def ensure_db_populated(db_instance: PolicyVectorDB, chunks_file_path: str):
|
| 84 |
try:
|
| 85 |
if db_instance._get_collection().count() == 0:
|
| 86 |
-
logger.info("Vector database is empty. Attempting to populate
|
| 87 |
if not os.path.exists(chunks_file_path):
|
| 88 |
-
logger.error(f"Chunks file not found at {chunks_file_path}")
|
| 89 |
return False
|
| 90 |
-
|
| 91 |
with open(chunks_file_path, 'r', encoding='utf-8') as f:
|
| 92 |
chunks_to_add = json.load(f)
|
| 93 |
-
|
| 94 |
if not chunks_to_add:
|
| 95 |
-
logger.warning("Chunks file is empty.")
|
| 96 |
return False
|
| 97 |
|
| 98 |
db_instance.add_chunks(chunks_to_add)
|
| 99 |
-
logger.info("
|
| 100 |
return True
|
| 101 |
else:
|
| 102 |
-
logger.info("
|
| 103 |
return True
|
| 104 |
except Exception as e:
|
| 105 |
logger.error(f"DB Population Error: {e}", exc_info=True)
|
| 106 |
-
return False
|
|
|
|
| 40 |
try:
|
| 41 |
existing_ids = set(collection.get(include=[])['ids'])
|
| 42 |
except Exception as e:
|
| 43 |
+
logger.warning(f"Could not retrieve existing IDs from ChromaDB: {e}. Assuming no existing IDs for now.")
|
| 44 |
|
| 45 |
new_chunks = [chunk for chunk in chunks if chunk.get('id') and chunk['id'] not in existing_ids]
|
| 46 |
+
|
| 47 |
if not new_chunks:
|
| 48 |
logger.info("No new chunks to add.")
|
| 49 |
return
|
| 50 |
|
| 51 |
+
logger.info(f"Adding {len(new_chunks)} new chunks to the vector database...")
|
| 52 |
batch_size = 128
|
| 53 |
for i in range(0, len(new_chunks), batch_size):
|
| 54 |
batch = new_chunks[i:i + batch_size]
|
| 55 |
texts = [chunk['text'] for chunk in batch]
|
| 56 |
ids = [chunk['id'] for chunk in batch]
|
| 57 |
+
metadatas = [self._flatten_metadata(chunk['metadata']) if chunk.get('metadata') else {} for chunk in batch]
|
| 58 |
embeddings = self.embedding_model.encode(texts, show_progress_bar=False).tolist()
|
| 59 |
collection.add(ids=ids, embeddings=embeddings, documents=texts, metadatas=metadatas)
|
| 60 |
+
logger.info(f"Added batch {i//batch_size + 1}/{(len(new_chunks) + batch_size - 1) // batch_size}")
|
| 61 |
logger.info(f"Finished adding {len(new_chunks)} chunks.")
|
| 62 |
|
| 63 |
def search(self, query_text: str, top_k: int = None) -> List[Dict]:
|
| 64 |
collection = self._get_collection()
|
| 65 |
query_embedding = self.embedding_model.encode([query_text]).tolist()
|
| 66 |
+
top_k = top_k if top_k else self.top_k_default
|
|
|
|
| 67 |
results = collection.query(
|
| 68 |
query_embeddings=query_embedding,
|
| 69 |
n_results=top_k,
|
| 70 |
include=["documents", "metadatas", "distances"]
|
| 71 |
)
|
| 72 |
+
|
| 73 |
search_results = []
|
| 74 |
+
if results and results['documents'] and results['documents'][0]:
|
| 75 |
for i, doc in enumerate(results['documents'][0]):
|
| 76 |
+
relevance_score = 1 - results['distances'][0][i]
|
| 77 |
search_results.append({
|
| 78 |
'text': doc,
|
| 79 |
'metadata': results['metadatas'][0][i],
|
|
|
|
| 84 |
def ensure_db_populated(db_instance: PolicyVectorDB, chunks_file_path: str):
|
| 85 |
try:
|
| 86 |
if db_instance._get_collection().count() == 0:
|
| 87 |
+
logger.info("Vector database is empty. Attempting to populate from chunks file.")
|
| 88 |
if not os.path.exists(chunks_file_path):
|
| 89 |
+
logger.error(f"Chunks file not found at {chunks_file_path}. Cannot populate DB.")
|
| 90 |
return False
|
| 91 |
+
|
| 92 |
with open(chunks_file_path, 'r', encoding='utf-8') as f:
|
| 93 |
chunks_to_add = json.load(f)
|
| 94 |
+
|
| 95 |
if not chunks_to_add:
|
| 96 |
+
logger.warning(f"Chunks file at {chunks_file_path} is empty. No data to add to DB.")
|
| 97 |
return False
|
| 98 |
|
| 99 |
db_instance.add_chunks(chunks_to_add)
|
| 100 |
+
logger.info("Vector database population attempt complete.")
|
| 101 |
return True
|
| 102 |
else:
|
| 103 |
+
logger.info("Vector database already contains data. Skipping population.")
|
| 104 |
return True
|
| 105 |
except Exception as e:
|
| 106 |
logger.error(f"DB Population Error: {e}", exc_info=True)
|
| 107 |
+
return False
|