Spaces:
Building
Building
File size: 15,078 Bytes
872c2a9 3f61806 b9ccd0b 3f61806 7361b6f 3f61806 b9ccd0b 7361b6f b9ccd0b 3f61806 b9ccd0b 3f61806 872c2a9 3f61806 872c2a9 3f61806 872c2a9 b9ccd0b 7361b6f 3f61806 872c2a9 7361b6f 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 3f61806 7361b6f 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 7361b6f b9ccd0b 872c2a9 b9ccd0b 872c2a9 7361b6f b9ccd0b 872c2a9 7361b6f b9ccd0b 7361b6f 872c2a9 b9ccd0b 872c2a9 7361b6f 872c2a9 b9ccd0b 872c2a9 7361b6f 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 7361b6f b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 b9ccd0b 872c2a9 3f61806 872c2a9 3f61806 872c2a9 3f61806 872c2a9 3f61806 872c2a9 3f61806 872c2a9 b9ccd0b 872c2a9 7361b6f 872c2a9 3f61806 872c2a9 b9ccd0b 3f61806 7361b6f |
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 |
# /home/bk_anupam/code/LLM_agents/RAG_BOT/vector_store.py
from collections import defaultdict
import os
import sys
import datetime
# shutil and re are no longer directly used by this class after refactoring
from langchain_chroma import Chroma
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain.schema import HumanMessage, AIMessage
from langchain.prompts import PromptTemplate
from langchain.schema.runnable import RunnablePassthrough
# Add the project root to the Python path
project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), '..'))
sys.path.insert(0, project_root)
from RAG_BOT.logger import logger
from RAG_BOT.config import Config
from RAG_BOT.document_processor import DocumentProcessor
class VectorStore:
def __init__(self, persist_directory=None):
self.config = Config()
self.persist_directory = persist_directory or self.config.VECTOR_STORE_PATH
# Initialize the embedding model once.
self.embeddings = HuggingFaceEmbeddings(model_name=self.config.EMBEDDING_MODEL_NAME)
logger.info("Embedding model initialized successfully.")
# Create or load the Chroma vector database.
# Added more robust error handling and directory creation
os.makedirs(self.persist_directory, exist_ok=True) # Ensure directory exists before checking content
if os.path.exists(self.persist_directory) and os.listdir(self.persist_directory):
try:
self.vectordb = Chroma(persist_directory=self.persist_directory, embedding_function=self.embeddings)
logger.info(f"Existing vector store loaded successfully from: {self.persist_directory}")
except Exception as e:
logger.error(f"Error loading existing vector store from {self.persist_directory}: {e}", exc_info=True)
# Consider if creating a new one is the right fallback or if it should raise
logger.warning(f"Attempting to create a new vector store at {self.persist_directory} due to loading error.")
try:
# Attempt to create anew if loading failed (might indicate corruption)
self.vectordb = Chroma(persist_directory=self.persist_directory, embedding_function=self.embeddings)
logger.info("New vector store created after load failure.")
except Exception as inner_e:
logger.critical(f"Failed to create new vector store after load failure: {inner_e}", exc_info=True)
raise inner_e # Re-raise critical failure
else:
try:
self.vectordb = Chroma(persist_directory=self.persist_directory, embedding_function=self.embeddings)
logger.info(f"New vector store created successfully at: {self.persist_directory}")
except Exception as e:
logger.critical(f"Failed to create new vector store at {self.persist_directory}: {e}", exc_info=True)
raise e # Re-raise critical failure
# Initialize document processors
self.document_processor = DocumentProcessor()
def get_vectordb(self):
return self.vectordb
def add_documents(self, texts):
if not texts:
logger.warning("Attempted to add an empty list of documents. Skipping.")
return
source = texts[0].metadata.get('source', 'N/A') # Get source from first doc for logging
try:
self.vectordb.add_documents(documents=texts)
logger.info(f"Vector store updated with {len(texts)} document chunks from source: {os.path.basename(source)}") # Log only basename
except Exception as e:
logger.error(f"Failed to add documents from source {os.path.basename(source)} to ChromaDB: {e}", exc_info=True)
def document_exists(self, date_str: str, language: str) -> bool:
"""
Checks if any document with the given date string and language already exists in the index.
"""
if not date_str:
# If date extraction failed, we assume it doesn't exist to allow indexing.
# The alternative is to skip indexing files without dates.
logger.warning("Cannot check for existing document without a date string. Assuming it does not exist.")
return False
if not language:
# Similarly, if language is missing, we cannot perform the combined check.
logger.warning("Cannot check for existing document without language metadata. Assuming it does not exist.")
return False
logger.debug(f"Checking for existing documents with date: {date_str} and language: {language}")
try:
# Check if vectordb is initialized
if not hasattr(self, 'vectordb') or self.vectordb is None:
logger.error("VectorDB not initialized. Cannot check for existing documents.")
return False # Cannot check, assume not found
existing_docs = self.vectordb.get(
# Use $and operator for multiple metadata filters
where={
"$and": [
{"date": date_str},
{"language": language}
] },
limit=1, # We only need to know if at least one exists
include=[] # We don't need metadata or documents, just the count implicitly
)
# Check if the 'ids' list is not empty
if existing_docs and existing_docs.get('ids'):
logger.debug(f"Document with date {date_str} and language {language} found in the index.")
return True
else:
logger.debug(f"No document found with date {date_str} and language {language}.")
return False
except Exception as e:
logger.error(f"Error checking ChromaDB for existing date {date_str} and language {language}: {e}. Assuming document does not exist.", exc_info=True)
# Decide how to handle errors - returning False assumes it doesn't exist, allowing indexing to proceed.
return False
def index_document(self, documents, semantic_chunk=False):
"""
Chunks and indexes a list of documents, checking first if a document with the same date metadata already exists.
Returns True if indexing occurred, False if skipped or failed.
This is a private helper method.
"""
if not documents:
logger.warning("Attempted to index an empty list of documents. Skipping.")
return False
# Extract date from the first document's metadata (assuming consistency for a single document/murli)
doc_metadata = documents[0].metadata
extracted_date = doc_metadata.get('date') # Get the date string
extracted_language = doc_metadata.get('language') # Get the language string
source_file = doc_metadata.get('source', 'N/A') # Get source for logging
# 1. Check if document already exists using the helper method
if self.document_exists(extracted_date, extracted_language):
logger.info(f"Document with date {extracted_date} and language {extracted_language} (source: {os.path.basename(source_file)}) already indexed. Skipping.")
return False # Indicate skip
# 2. If not existing, proceed with chunking and adding
logger.info(f"Proceeding with chunking and indexing for {os.path.basename(source_file)}.")
try:
if semantic_chunk:
texts = self.document_processor.semantic_chunking(
documents,
chunk_size=self.config.CHUNK_SIZE,
chunk_overlap=self.config.CHUNK_OVERLAP,
model_name=self.config.EMBEDDING_MODEL_NAME
)
else:
texts = self.document_processor.split_text(
documents,
chunk_size=self.config.CHUNK_SIZE,
chunk_overlap=self.config.CHUNK_OVERLAP
)
if not texts:
logger.warning(f"No text chunks generated after processing {os.path.basename(source_file)}. Nothing to index.")
return False # Indicate skip/failure
self.add_documents(texts)
logger.info(f"Successfully indexed {len(texts)} chunks from {os.path.basename(source_file)}.")
return True # Indicate successful indexing
except Exception as e:
logger.error(f"Error during chunking or adding documents for {os.path.basename(source_file)}: {e}", exc_info=True)
return False # Indicate failure
# _move_indexed_file method has been removed as its logic is now in FileManager.
# index_directory method has been removed as its logic is now in DocumentIndexer.
def log_all_indexed_metadata(self):
"""
Retrieves and logs the date, is_avyakt, and language metadata for ALL indexed documents.
Groups and counts by (date, is_avyakt, language).
"""
if not hasattr(self, 'vectordb') or self.vectordb is None:
logger.error("VectorDB instance not available for metadata retrieval.")
return
try:
logger.info("Attempting to retrieve metadata for ALL indexed documents...")
all_data = self.vectordb.get(include=['metadatas'])
if all_data and all_data.get('ids'):
all_metadatas = all_data.get('metadatas', [])
total_docs = len(all_data['ids'])
logger.info(f"Retrieved metadata for {total_docs} documents.")
if not all_metadatas:
logger.warning("Retrieved document IDs but no corresponding metadata.")
return
# Structure: {(date, is_avyakt, language): count}
metadata_summary = defaultdict(int)
missing_metadata_count = 0
for metadata in all_metadatas:
date = metadata.get('date', 'N/A')
is_avyakt = metadata.get('is_avyakt', 'N/A')
language = metadata.get('language', 'N/A')
if date == 'N/A' and is_avyakt == 'N/A' and language == 'N/A':
missing_metadata_count += 1
else:
metadata_summary[(date, is_avyakt, language)] += 1
logger.info("--- Logging Date, Avyakt Status, and Language for All Indexed Documents ---")
if metadata_summary:
# Sort by date, then is_avyakt (converted to string for comparison), then language for consistent logging
# Use a custom key to handle potential non-boolean types for is_avyakt during sorting
for (date, is_avyakt, language), count in sorted(metadata_summary.items(), key=lambda item: (item[0][0], str(item[0][1]), item[0][2])):
avyakt_str = "Avyakt" if is_avyakt is True else "Sakar/Other" if is_avyakt is False else "Unknown Status"
logger.info(f"Date: {date} - Type: {avyakt_str}, Language: {language}, Count: {count}")
else:
logger.info("No documents with date, is_avyakt, or language metadata found.")
if missing_metadata_count > 0:
logger.warning(f"Found {missing_metadata_count} documents missing date, is_avyakt, and language metadata.")
logger.info("--- Finished Logging All Indexed Metadata ---")
else:
logger.info("ChromaDB index appears to be empty. No metadata to retrieve.")
except Exception as e:
logger.error(f"Error retrieving all metadata from ChromaDB: {e}", exc_info=True)
def query_index(self, query, chain_type="stuff", k=25, model_name="gemini-2.0-flash", date_filter=None):
# Ensure vectordb is initialized
if not hasattr(self, 'vectordb') or self.vectordb is None:
logger.error("VectorDB not initialized. Cannot perform query.")
return "Error: Vector Store is not available."
llm = ChatGoogleGenerativeAI(model=model_name, temperature=0.3)
search_kwargs = {"k": k}
if date_filter:
try:
filter_date = datetime.datetime.strptime(date_filter, '%Y-%m-%d')
formatted_date = filter_date.strftime('%Y-%m-%d')
logger.info(f"Applying date filter: {formatted_date}")
# Use Chroma's metadata filtering syntax
search_kwargs["filter"] = {"date": formatted_date}
except ValueError:
logger.error(f"Invalid date format provided: {date_filter}. Should be YYYY-MM-DD.")
# Return an error or raise? Returning error message for now.
return "Error: Invalid date format for filter. Please use YYYY-MM-DD."
try:
retriever = self.vectordb.as_retriever(
search_type="similarity",
search_kwargs=search_kwargs
)
retrieved_docs = retriever.invoke(query)
context = "\n\n".join([doc.page_content for doc in retrieved_docs])
logger.info(f"Retrieved {len(retrieved_docs)} documents for query: '{query[:50]}...'") # Log snippet
logger.debug(f"Context for LLM: {context}") # Log snippet
custom_prompt = PromptTemplate(
input_variables=["context", "question"],
template=(
self.config.get_system_prompt(language_code="en") + # Call on instance and provide language_code
"\n\nContext:\n{context}\n\nQuestion: {question}" # Added newlines for clarity
),
)
chain = custom_prompt | llm | RunnablePassthrough() # Removed RunnablePassthrough, not needed here
response = chain.invoke({"context": context, "question": query})
if isinstance(response, AIMessage):
return response.content
# Handle potential string or other types returned by the chain
elif isinstance(response, str):
return response
elif hasattr(response, 'content'): # Check for Langchain Core message types
return response.content
else:
logger.warning(f"Unexpected response type from LLM chain: {type(response)}")
return str(response) # Fallback to string representation
except Exception as e:
logger.error(f"Error during query execution: {e}", exc_info=True)
return "Sorry, an error occurred while processing your query."
# Standalone script functions (test_query_index, index_data) and
# if __name__ == "__main__": block have been moved to RAG_BOT/vector_store_cli.py
|