Spaces:
Runtime error
Runtime error
File size: 54,433 Bytes
c76bc58 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 |
"""
Vector Chunking and RAG Module
Handles document chunking, vector embeddings, and RAG question-answering
"""
import os
import json
import numpy as np
from typing import Dict, Any, List, Optional, Tuple
from langchain.text_splitter import RecursiveCharacterTextSplitter, CharacterTextSplitter
from langchain.schema import Document
from langchain_community.vectorstores import FAISS, Chroma
from langchain.chains import RetrievalQA, ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
from langchain.prompts import PromptTemplate
import tempfile
import shutil
class VectorChunker:
"""Main class for document chunking and vector operations"""
def __init__(self, embeddings_model, chunk_size: int = 1000, chunk_overlap: int = 200):
self.embeddings = embeddings_model
self.chunk_size = chunk_size
self.chunk_overlap = chunk_overlap
self.setup_text_splitters()
self.vector_stores = {} # Cache for vector stores
def setup_text_splitters(self):
"""Initialize different text splitting strategies"""
# Default recursive splitter
self.recursive_splitter = RecursiveCharacterTextSplitter(
chunk_size=self.chunk_size,
chunk_overlap=self.chunk_overlap,
length_function=len,
separators=["\n\n", "\n", " ", ""]
)
# Character-based splitter
self.character_splitter = CharacterTextSplitter(
chunk_size=self.chunk_size,
chunk_overlap=self.chunk_overlap,
separator="\n\n"
)
# Semantic splitter for better context preservation
self.semantic_splitter = RecursiveCharacterTextSplitter(
chunk_size=800, # Smaller chunks for better semantic coherence
chunk_overlap=150,
length_function=len,
separators=["\n\n", "\n", ". ", " ", ""]
)
def chunk_documents(self, documents: List[Document], strategy: str = "recursive") -> List[Document]:
"""
Chunk documents using specified strategy
Args:
documents (List[Document]): List of documents to chunk
strategy (str): Chunking strategy ("recursive", "character", "semantic")
Returns:
List[Document]: List of chunked documents
"""
try:
# Choose splitter based on strategy
if strategy == "character":
splitter = self.character_splitter
elif strategy == "semantic":
splitter = self.semantic_splitter
else:
splitter = self.recursive_splitter
# Split documents
chunked_docs = []
for doc in documents:
chunks = splitter.split_documents([doc])
# Add chunk metadata
for i, chunk in enumerate(chunks):
chunk.metadata.update({
'chunk_index': i,
'total_chunks': len(chunks),
'chunk_strategy': strategy,
'original_source': doc.metadata.get('source', 'unknown'),
'chunk_size': len(chunk.page_content),
'chunk_word_count': len(chunk.page_content.split())
})
chunked_docs.extend(chunks)
return chunked_docs
except Exception as e:
raise Exception(f"Document chunking failed: {str(e)}")
def create_vector_store(self, documents: List[Document], store_type: str = "faiss",
persist_directory: Optional[str] = None) -> Any:
"""
Create vector store from documents
Args:
documents (List[Document]): Documents to vectorize
store_type (str): Type of vector store ("faiss", "chroma")
persist_directory (str): Optional directory to persist the store
Returns:
Vector store instance
"""
try:
if not documents:
raise ValueError("No documents provided for vector store creation")
if store_type.lower() == "chroma":
if persist_directory:
vector_store = Chroma.from_documents(
documents=documents,
embedding=self.embeddings,
persist_directory=persist_directory
)
vector_store.persist()
else:
vector_store = Chroma.from_documents(
documents=documents,
embedding=self.embeddings
)
else: # Default to FAISS
vector_store = FAISS.from_documents(
documents=documents,
embedding=self.embeddings
)
# Save FAISS index if persist directory provided
if persist_directory:
os.makedirs(persist_directory, exist_ok=True)
vector_store.save_local(persist_directory)
return vector_store
except Exception as e:
raise Exception(f"Vector store creation failed: {str(e)}")
def create_qa_chain(self, documents: List[Document], llm, chain_type: str = "stuff") -> RetrievalQA:
"""
Create a Question-Answering chain from documents
Args:
documents (List[Document]): Documents for the knowledge base
llm: Language model for answering questions
chain_type (str): Type of QA chain ("stuff", "map_reduce", "refine")
Returns:
RetrievalQA: Configured QA chain
"""
try:
# Chunk documents
chunked_docs = self.chunk_documents(documents, strategy="semantic")
# Create vector store
vector_store = self.create_vector_store(chunked_docs, store_type="faiss")
# Create retriever
retriever = vector_store.as_retriever(
search_type="similarity",
search_kwargs={"k": 4} # Retrieve top 4 most relevant chunks
)
# Custom prompt for GEO-focused QA
qa_prompt_template = """Use the following pieces of context to answer the question at the end.
If you don't know the answer, just say that you don't know, don't try to make up an answer.
Focus on providing clear, accurate, and complete answers that would be suitable for AI search engines.
Context:
{context}
Question: {question}
Answer:"""
qa_prompt = PromptTemplate(
template=qa_prompt_template,
input_variables=["context", "question"]
)
# Create QA chain
qa_chain = RetrievalQA.from_chain_type(
llm=llm,
chain_type=chain_type,
retriever=retriever,
return_source_documents=True,
chain_type_kwargs={"prompt": qa_prompt}
)
return qa_chain
except Exception as e:
raise Exception(f"QA chain creation failed: {str(e)}")
def create_conversational_chain(self, documents: List[Document], llm) -> ConversationalRetrievalChain:
"""
Create a conversational retrieval chain with memory
Args:
documents (List[Document]): Documents for the knowledge base
llm: Language model for conversation
Returns:
ConversationalRetrievalChain: Configured conversational chain
"""
try:
# Chunk documents
chunked_docs = self.chunk_documents(documents, strategy="semantic")
# Create vector store
vector_store = self.create_vector_store(chunked_docs, store_type="faiss")
# Create retriever
retriever = vector_store.as_retriever(
search_type="similarity",
search_kwargs={"k": 3}
)
# Create memory
memory = ConversationBufferMemory(
memory_key="chat_history",
return_messages=True,
output_key="answer"
)
# Custom prompt for conversational QA
condense_question_prompt = """Given the following conversation and a follow up question,
rephrase the follow up question to be a standalone question that can be understood without the chat history.
Chat History:
{chat_history}
Follow Up Input: {question}
Standalone question:"""
# Create conversational chain
conv_chain = ConversationalRetrievalChain.from_llm(
llm=llm,
retriever=retriever,
memory=memory,
return_source_documents=True,
condense_question_prompt=PromptTemplate.from_template(condense_question_prompt)
)
return conv_chain
except Exception as e:
raise Exception(f"Conversational chain creation failed: {str(e)}")
def semantic_search(self, query: str, documents: List[Document], top_k: int = 5) -> List[Dict[str, Any]]:
"""
Perform semantic search on documents
Args:
query (str): Search query
documents (List[Document]): Documents to search
top_k (int): Number of top results to return
Returns:
List[Dict]: Search results with scores
"""
try:
# Chunk documents
chunked_docs = self.chunk_documents(documents, strategy="semantic")
# Create vector store
vector_store = self.create_vector_store(chunked_docs, store_type="faiss")
# Perform similarity search with scores
results = vector_store.similarity_search_with_score(query, k=top_k)
# Format results
formatted_results = []
for doc, score in results:
result = {
'content': doc.page_content,
'metadata': doc.metadata,
'similarity_score': float(score),
'relevance_rank': len(formatted_results) + 1
}
formatted_results.append(result)
return formatted_results
except Exception as e:
raise Exception(f"Semantic search failed: {str(e)}")
def analyze_document_similarity(self, documents: List[Document]) -> Dict[str, Any]:
"""
Analyze similarity between documents
Args:
documents (List[Document]): Documents to analyze
Returns:
Dict: Similarity analysis results
"""
try:
if len(documents) < 2:
return {'error': 'Need at least 2 documents for similarity analysis'}
# Chunk documents
chunked_docs = self.chunk_documents(documents, strategy="semantic")
# Create embeddings for each document
doc_embeddings = []
doc_metadata = []
for doc in chunked_docs:
# Get embedding for the document
embedding = self.embeddings.embed_query(doc.page_content)
doc_embeddings.append(embedding)
doc_metadata.append({
'content_preview': doc.page_content[:200] + "...",
'metadata': doc.metadata,
'length': len(doc.page_content)
})
# Calculate pairwise similarities
similarities = []
embeddings_array = np.array(doc_embeddings)
for i in range(len(embeddings_array)):
for j in range(i + 1, len(embeddings_array)):
# Calculate cosine similarity
similarity = np.dot(embeddings_array[i], embeddings_array[j]) / (
np.linalg.norm(embeddings_array[i]) * np.linalg.norm(embeddings_array[j])
)
similarities.append({
'doc_1_index': i,
'doc_2_index': j,
'similarity_score': float(similarity),
'doc_1_preview': doc_metadata[i]['content_preview'],
'doc_2_preview': doc_metadata[j]['content_preview']
})
# Sort by similarity score
similarities.sort(key=lambda x: x['similarity_score'], reverse=True)
# Calculate statistics
similarity_scores = [s['similarity_score'] for s in similarities]
return {
'total_comparisons': len(similarities),
'average_similarity': np.mean(similarity_scores),
'max_similarity': max(similarity_scores),
'min_similarity': min(similarity_scores),
'similarity_distribution': {
'high_similarity': len([s for s in similarity_scores if s > 0.8]),
'medium_similarity': len([s for s in similarity_scores if 0.5 < s <= 0.8]),
'low_similarity': len([s for s in similarity_scores if s <= 0.5])
},
'top_similar_pairs': similarities[:5],
'most_dissimilar_pairs': similarities[-3:]
}
except Exception as e:
return {'error': f"Similarity analysis failed: {str(e)}"}
def extract_key_passages(self, documents: List[Document], queries: List[str],
passages_per_query: int = 3) -> Dict[str, List[Dict[str, Any]]]:
"""
Extract key passages from documents based on multiple queries
Args:
documents (List[Document]): Documents to search
queries (List[str]): List of queries to search for
passages_per_query (int): Number of passages to extract per query
Returns:
Dict: Key passages organized by query
"""
try:
# Chunk documents
chunked_docs = self.chunk_documents(documents, strategy="semantic")
# Create vector store
vector_store = self.create_vector_store(chunked_docs, store_type="faiss")
key_passages = {}
for query in queries:
# Search for relevant passages
results = vector_store.similarity_search_with_score(query, k=passages_per_query)
passages = []
for doc, score in results:
passage = {
'content': doc.page_content,
'relevance_score': float(score),
'metadata': doc.metadata,
'word_count': len(doc.page_content.split()),
'query_match': query
}
passages.append(passage)
key_passages[query] = passages
return key_passages
except Exception as e:
return {'error': f"Key passage extraction failed: {str(e)}"}
def optimize_chunking_strategy(self, documents: List[Document],
test_queries: List[str]) -> Dict[str, Any]:
"""
Test different chunking strategies and recommend the best one
Args:
documents (List[Document]): Documents to test
test_queries (List[str]): Queries to test retrieval performance
Returns:
Dict: Optimization results and recommendations
"""
try:
strategies = ["recursive", "character", "semantic"]
strategy_results = {}
for strategy in strategies:
try:
# Test this strategy
chunked_docs = self.chunk_documents(documents, strategy=strategy)
vector_store = self.create_vector_store(chunked_docs, store_type="faiss")
# Test retrieval performance
retrieval_scores = []
for query in test_queries:
results = vector_store.similarity_search_with_score(query, k=3)
# Calculate average relevance score
if results:
avg_score = sum(score for _, score in results) / len(results)
retrieval_scores.append(float(avg_score))
# Calculate strategy metrics
avg_retrieval_score = np.mean(retrieval_scores) if retrieval_scores else 0
total_chunks = len(chunked_docs)
avg_chunk_size = np.mean([len(doc.page_content) for doc in chunked_docs])
strategy_results[strategy] = {
'average_retrieval_score': avg_retrieval_score,
'total_chunks': total_chunks,
'average_chunk_size': avg_chunk_size,
'retrieval_scores': retrieval_scores,
'chunk_size_distribution': {
'min': min(len(doc.page_content) for doc in chunked_docs),
'max': max(len(doc.page_content) for doc in chunked_docs),
'std': float(np.std([len(doc.page_content) for doc in chunked_docs]))
}
}
except Exception as e:
strategy_results[strategy] = {'error': f"Strategy test failed: {str(e)}"}
# Determine best strategy
valid_strategies = {k: v for k, v in strategy_results.items() if 'error' not in v}
if valid_strategies:
best_strategy = max(valid_strategies.keys(),
key=lambda k: valid_strategies[k]['average_retrieval_score'])
recommendation = {
'recommended_strategy': best_strategy,
'reason': f"Best average retrieval score: {valid_strategies[best_strategy]['average_retrieval_score']:.4f}",
'all_results': strategy_results,
'performance_summary': {
strategy: result.get('average_retrieval_score', 0)
for strategy, result in valid_strategies.items()
}
}
else:
recommendation = {
'recommended_strategy': 'recursive', # Default fallback
'reason': 'All strategies failed, using default',
'all_results': strategy_results
}
return recommendation
except Exception as e:
return {'error': f"Chunking optimization failed: {str(e)}"}
def create_document_summary(self, documents: List[Document], llm,
summary_type: str = "extractive") -> Dict[str, Any]:
"""
Create document summaries using the chunked content
Args:
documents (List[Document]): Documents to summarize
llm: Language model for summarization
summary_type (str): Type of summary ("extractive", "abstractive")
Returns:
Dict: Summary results
"""
try:
# Chunk documents for better processing
chunked_docs = self.chunk_documents(documents, strategy="semantic")
if summary_type == "extractive":
# Extract key sentences/chunks
return self._create_extractive_summary(chunked_docs)
else:
# Generate abstractive summary using LLM
return self._create_abstractive_summary(chunked_docs, llm)
except Exception as e:
return {'error': f"Document summarization failed: {str(e)}"}
def _create_extractive_summary(self, chunked_docs: List[Document]) -> Dict[str, Any]:
"""Create extractive summary by selecting key chunks"""
try:
# Simple extractive approach: select chunks with highest semantic density
chunk_scores = []
for doc in chunked_docs:
content = doc.page_content
# Simple scoring based on content characteristics
word_count = len(content.split())
sentence_count = len([s for s in content.split('.') if s.strip()])
# Score based on information density
density_score = word_count / max(sentence_count, 1)
# Bonus for chunks with questions, definitions, or lists
structure_bonus = 0
if '?' in content:
structure_bonus += 1
if any(word in content.lower() for word in ['define', 'definition', 'means', 'refers to']):
structure_bonus += 2
if content.count('\n•') > 0 or content.count('1.') > 0:
structure_bonus += 1
total_score = density_score + structure_bonus
chunk_scores.append((doc, total_score))
# Sort by score and select top chunks for summary
chunk_scores.sort(key=lambda x: x[1], reverse=True)
top_chunks = chunk_scores[:min(5, len(chunk_scores))]
summary_content = []
for doc, score in top_chunks:
summary_content.append({
'content': doc.page_content,
'score': score,
'metadata': doc.metadata
})
return {
'summary_type': 'extractive',
'key_chunks': summary_content,
'total_chunks_analyzed': len(chunked_docs),
'chunks_selected': len(top_chunks)
}
except Exception as e:
return {'error': f"Extractive summary failed: {str(e)}"}
def _create_abstractive_summary(self, chunked_docs: List[Document], llm) -> Dict[str, Any]:
"""Create abstractive summary using language model"""
try:
# Combine content from top chunks
combined_content = "\n\n".join([doc.page_content for doc in chunked_docs[:10]])
summary_prompt = f"""Please provide a comprehensive summary of the following content.
Focus on the main topics, key insights, and important details that would be valuable for AI search engines.
Content:
{combined_content[:5000]}
Summary:"""
from langchain.prompts import ChatPromptTemplate
prompt_template = ChatPromptTemplate.from_messages([
("system", "You are a professional content summarizer. Create clear, informative summaries."),
("user", summary_prompt)
])
chain = prompt_template | llm
result = chain.invoke({})
summary_text = result.content if hasattr(result, 'content') else str(result)
return {
'summary_type': 'abstractive',
'summary': summary_text,
'source_chunks': len(chunked_docs),
'content_length_processed': len(combined_content)
}
except Exception as e:
return {'error': f"Abstractive summary failed: {str(e)}"}
def save_vector_store(self, vector_store, directory_path: str, store_type: str = "faiss") -> bool:
"""
Save vector store to disk
Args:
vector_store: Vector store instance to save
directory_path (str): Directory to save the store
store_type (str): Type of vector store
Returns:
bool: Success status
"""
try:
os.makedirs(directory_path, exist_ok=True)
if store_type.lower() == "faiss":
vector_store.save_local(directory_path)
elif store_type.lower() == "chroma":
# Chroma stores are typically persisted during creation
pass
return True
except Exception as e:
print(f"Failed to save vector store: {str(e)}")
return False
def load_vector_store(self, directory_path: str, store_type: str = "faiss"):
"""
Load vector store from disk
Args:
directory_path (str): Directory containing the saved store
store_type (str): Type of vector store
Returns:
Vector store instance or None if failed
"""
try:
if not os.path.exists(directory_path):
return None
if store_type.lower() == "faiss":
vector_store = FAISS.load_local(
directory_path,
self.embeddings,
allow_dangerous_deserialization=True
)
return vector_store
elif store_type.lower() == "chroma":
vector_store = Chroma(
persist_directory=directory_path,
embedding_function=self.embeddings
)
return vector_store
return None
except Exception as e:
print(f"Failed to load vector store: {str(e)}")
return None
def get_chunking_stats(self, documents: List[Document], strategy: str = "recursive") -> Dict[str, Any]:
"""
Get detailed statistics about document chunking
Args:
documents (List[Document]): Documents to analyze
strategy (str): Chunking strategy to use
Returns:
Dict: Detailed chunking statistics
"""
try:
# Chunk documents
chunked_docs = self.chunk_documents(documents, strategy=strategy)
# Calculate statistics
chunk_sizes = [len(doc.page_content) for doc in chunked_docs]
word_counts = [len(doc.page_content.split()) for doc in chunked_docs]
stats = {
'strategy_used': strategy,
'original_documents': len(documents),
'total_chunks': len(chunked_docs),
'chunk_size_stats': {
'min': min(chunk_sizes) if chunk_sizes else 0,
'max': max(chunk_sizes) if chunk_sizes else 0,
'mean': np.mean(chunk_sizes) if chunk_sizes else 0,
'median': np.median(chunk_sizes) if chunk_sizes else 0,
'std': np.std(chunk_sizes) if chunk_sizes else 0
},
'word_count_stats': {
'min': min(word_counts) if word_counts else 0,
'max': max(word_counts) if word_counts else 0,
'mean': np.mean(word_counts) if word_counts else 0,
'median': np.median(word_counts) if word_counts else 0,
'std': np.std(word_counts) if word_counts else 0
},
'chunk_distribution': {
'very_small': len([s for s in chunk_sizes if s < 200]),
'small': len([s for s in chunk_sizes if 200 <= s < 500]),
'medium': len([s for s in chunk_sizes if 500 <= s < 1000]),
'large': len([s for s in chunk_sizes if 1000 <= s < 2000]),
'very_large': len([s for s in chunk_sizes if s >= 2000])
},
'overlap_efficiency': self._calculate_overlap_efficiency(chunked_docs),
'content_coverage': self._calculate_content_coverage(documents, chunked_docs)
}
return stats
except Exception as e:
return {'error': f"Chunking statistics failed: {str(e)}"}
def _calculate_overlap_efficiency(self, chunked_docs: List[Document]) -> float:
"""Calculate efficiency of chunk overlaps"""
try:
if len(chunked_docs) < 2:
return 1.0
total_content_length = sum(len(doc.page_content) for doc in chunked_docs)
unique_content = set()
# Rough estimate of content uniqueness
for doc in chunked_docs:
words = doc.page_content.split()
for i in range(0, len(words), 10): # Sample every 10th word
unique_content.add(' '.join(words[i:i+10]))
# Efficiency as ratio of unique content to total content
efficiency = len(unique_content) * 10 / total_content_length if total_content_length > 0 else 0
return min(efficiency, 1.0)
except Exception:
return 0.5 # Default neutral efficiency
def _calculate_content_coverage(self, original_docs: List[Document],
chunked_docs: List[Document]) -> float:
"""Calculate how well chunks cover original content"""
try:
original_content = ' '.join([doc.page_content for doc in original_docs])
chunked_content = ' '.join([doc.page_content for doc in chunked_docs])
# Simple coverage metric based on length
coverage = len(chunked_content) / len(original_content) if original_content else 0
return min(coverage, 1.0)
except Exception:
return 0.0
class ChunkingOptimizer:
"""Helper class for optimizing chunking parameters"""
def __init__(self, embeddings_model):
self.embeddings = embeddings_model
def optimize_chunk_size(self, documents: List[Document], test_queries: List[str],
size_range: Tuple[int, int] = (200, 2000),
step_size: int = 200) -> Dict[str, Any]:
"""
Find optimal chunk size for given documents and queries
Args:
documents (List[Document]): Documents to test
test_queries (List[str]): Queries for testing retrieval
size_range (Tuple[int, int]): Range of chunk sizes to test
step_size (int): Step size for testing
Returns:
Dict: Optimization results with recommended chunk size
"""
try:
results = {}
min_size, max_size = size_range
for chunk_size in range(min_size, max_size + 1, step_size):
# Test this chunk size
chunker = VectorChunker(self.embeddings, chunk_size=chunk_size)
try:
chunked_docs = chunker.chunk_documents(documents)
vector_store = chunker.create_vector_store(chunked_docs)
# Test retrieval performance
retrieval_scores = []
for query in test_queries:
search_results = vector_store.similarity_search_with_score(query, k=3)
if search_results:
avg_score = sum(score for _, score in search_results) / len(search_results)
retrieval_scores.append(float(avg_score))
avg_performance = np.mean(retrieval_scores) if retrieval_scores else 0
results[chunk_size] = {
'average_retrieval_score': avg_performance,
'total_chunks': len(chunked_docs),
'retrieval_scores': retrieval_scores
}
except Exception as e:
results[chunk_size] = {'error': str(e)}
# Find optimal chunk size
valid_results = {k: v for k, v in results.items() if 'error' not in v}
if valid_results:
optimal_size = max(valid_results.keys(),
key=lambda k: valid_results[k]['average_retrieval_score'])
return {
'optimal_chunk_size': optimal_size,
'optimal_performance': valid_results[optimal_size]['average_retrieval_score'],
'all_results': results,
'performance_trend': self._analyze_performance_trend(valid_results),
'recommendation': f"Use chunk size {optimal_size} for best retrieval performance"
}
else:
return {
'error': 'No valid chunk sizes could be tested',
'all_results': results
}
except Exception as e:
return {'error': f"Chunk size optimization failed: {str(e)}"}
def _analyze_performance_trend(self, results: Dict[int, Dict[str, Any]]) -> Dict[str, Any]:
"""Analyze performance trend across different chunk sizes"""
try:
sizes = sorted(results.keys())
performances = [results[size]['average_retrieval_score'] for size in sizes]
# Find trend direction
if len(performances) >= 2:
trend_direction = "increasing" if performances[-1] > performances[0] else "decreasing"
peak_performance = max(performances)
peak_size = sizes[performances.index(peak_performance)]
return {
'trend_direction': trend_direction,
'peak_performance': peak_performance,
'peak_size': peak_size,
'performance_range': max(performances) - min(performances),
'stable_performance': max(performances) - min(performances) < 0.1
}
else:
return {'error': 'Insufficient data for trend analysis'}
except Exception:
return {'error': 'Trend analysis failed'}
class RAGPipeline:
"""Complete RAG pipeline for document question-answering"""
def __init__(self, embeddings_model, llm):
self.embeddings = embeddings_model
self.llm = llm
self.chunker = VectorChunker(embeddings_model)
self.vector_stores = {}
self.qa_chains = {}
def create_pipeline(self, documents: List[Document], pipeline_id: str,
chunking_strategy: str = "semantic") -> Dict[str, Any]:
"""
Create a complete RAG pipeline for documents
Args:
documents (List[Document]): Documents to process
pipeline_id (str): Unique identifier for this pipeline
chunking_strategy (str): Strategy for document chunking
Returns:
Dict: Pipeline creation results
"""
try:
# Step 1: Chunk documents
chunked_docs = self.chunker.chunk_documents(documents, strategy=chunking_strategy)
# Step 2: Create vector store
vector_store = self.chunker.create_vector_store(chunked_docs, store_type="faiss")
# Step 3: Create QA chain
qa_chain = self.chunker.create_qa_chain(documents, self.llm)
# Store pipeline components
self.vector_stores[pipeline_id] = vector_store
self.qa_chains[pipeline_id] = qa_chain
# Pipeline statistics
stats = {
'pipeline_id': pipeline_id,
'documents_processed': len(documents),
'chunks_created': len(chunked_docs),
'chunking_strategy': chunking_strategy,
'vector_store_type': 'faiss',
'embedding_model': str(self.embeddings),
'created_at': self._get_timestamp()
}
return {
'success': True,
'pipeline_stats': stats,
'chunking_info': self.chunker.get_chunking_stats(documents, chunking_strategy)
}
except Exception as e:
return {'error': f"Pipeline creation failed: {str(e)}"}
def query_pipeline(self, pipeline_id: str, query: str,
return_sources: bool = True) -> Dict[str, Any]:
"""
Query a created RAG pipeline
Args:
pipeline_id (str): ID of the pipeline to query
query (str): Question to ask
return_sources (bool): Whether to return source documents
Returns:
Dict: Query results with answer and sources
"""
try:
if pipeline_id not in self.qa_chains:
return {'error': f"Pipeline '{pipeline_id}' not found"}
qa_chain = self.qa_chains[pipeline_id]
# Execute query
result = qa_chain({"query": query})
# Format response
response = {
'query': query,
'answer': result.get('result', 'No answer generated'),
'pipeline_id': pipeline_id,
'query_timestamp': self._get_timestamp()
}
# Add source documents if requested
if return_sources and 'source_documents' in result:
sources = []
for i, doc in enumerate(result['source_documents']):
source = {
'source_index': i,
'content': doc.page_content,
'metadata': doc.metadata,
'relevance_rank': i + 1
}
sources.append(source)
response['sources'] = sources
response['num_sources'] = len(sources)
return response
except Exception as e:
return {'error': f"Pipeline query failed: {str(e)}"}
def batch_query_pipeline(self, pipeline_id: str, queries: List[str]) -> List[Dict[str, Any]]:
"""
Execute multiple queries on a pipeline
Args:
pipeline_id (str): ID of the pipeline to query
queries (List[str]): List of questions to ask
Returns:
List[Dict]: List of query results
"""
results = []
for i, query in enumerate(queries):
try:
result = self.query_pipeline(pipeline_id, query, return_sources=False)
result['batch_index'] = i
results.append(result)
except Exception as e:
results.append({
'batch_index': i,
'query': query,
'error': f"Batch query failed: {str(e)}"
})
return results
def evaluate_pipeline(self, pipeline_id: str, test_queries: List[str],
expected_answers: List[str] = None) -> Dict[str, Any]:
"""
Evaluate pipeline performance on test queries
Args:
pipeline_id (str): ID of the pipeline to evaluate
test_queries (List[str]): Test questions
expected_answers (List[str]): Optional expected answers for comparison
Returns:
Dict: Evaluation results
"""
try:
if pipeline_id not in self.qa_chains:
return {'error': f"Pipeline '{pipeline_id}' not found"}
evaluation_results = []
response_times = []
for i, query in enumerate(test_queries):
import time
start_time = time.time()
# Execute query
result = self.query_pipeline(pipeline_id, query, return_sources=True)
end_time = time.time()
response_time = end_time - start_time
response_times.append(response_time)
# Evaluate result
eval_result = {
'query_index': i,
'query': query,
'answer_generated': not result.get('error'),
'response_time': response_time,
'answer_length': len(result.get('answer', '')),
'sources_returned': result.get('num_sources', 0)
}
# If expected answer provided, calculate similarity
if expected_answers and i < len(expected_answers):
expected = expected_answers[i]
generated = result.get('answer', '')
# Simple similarity metric
similarity = self._calculate_answer_similarity(expected, generated)
eval_result['answer_similarity'] = similarity
eval_result['expected_answer'] = expected
evaluation_results.append(eval_result)
# Calculate aggregate metrics
successful_queries = len([r for r in evaluation_results if r['answer_generated']])
avg_response_time = np.mean(response_times) if response_times else 0
if expected_answers:
similarities = [r.get('answer_similarity', 0) for r in evaluation_results
if 'answer_similarity' in r]
avg_similarity = np.mean(similarities) if similarities else 0
else:
avg_similarity = None
return {
'pipeline_id': pipeline_id,
'total_queries': len(test_queries),
'successful_queries': successful_queries,
'success_rate': successful_queries / len(test_queries) if test_queries else 0,
'average_response_time': avg_response_time,
'average_answer_similarity': avg_similarity,
'detailed_results': evaluation_results,
'evaluation_timestamp': self._get_timestamp()
}
except Exception as e:
return {'error': f"Pipeline evaluation failed: {str(e)}"}
def _calculate_answer_similarity(self, expected: str, generated: str) -> float:
"""Calculate similarity between expected and generated answers"""
try:
# Simple word overlap similarity
expected_words = set(expected.lower().split())
generated_words = set(generated.lower().split())
if not expected_words and not generated_words:
return 1.0
intersection = expected_words.intersection(generated_words)
union = expected_words.union(generated_words)
return len(intersection) / len(union) if union else 0.0
except Exception:
return 0.0
def get_pipeline_info(self, pipeline_id: str) -> Dict[str, Any]:
"""Get information about a specific pipeline"""
try:
if pipeline_id not in self.qa_chains:
return {'error': f"Pipeline '{pipeline_id}' not found"}
# Get vector store info
vector_store = self.vector_stores.get(pipeline_id)
if vector_store:
try:
# Try to get vector store statistics
total_vectors = vector_store.index.ntotal if hasattr(vector_store, 'index') else 'unknown'
except:
total_vectors = 'unknown'
else:
total_vectors = 'unknown'
return {
'pipeline_id': pipeline_id,
'has_qa_chain': pipeline_id in self.qa_chains,
'has_vector_store': pipeline_id in self.vector_stores,
'total_vectors': total_vectors,
'embedding_model': str(self.embeddings),
'llm_model': str(self.llm)
}
except Exception as e:
return {'error': f"Failed to get pipeline info: {str(e)}"}
def list_pipelines(self) -> Dict[str, Any]:
"""List all created pipelines"""
return {
'total_pipelines': len(self.qa_chains),
'pipeline_ids': list(self.qa_chains.keys()),
'vector_stores': list(self.vector_stores.keys())
}
def delete_pipeline(self, pipeline_id: str) -> Dict[str, Any]:
"""Delete a pipeline and free resources"""
try:
deleted_components = []
if pipeline_id in self.qa_chains:
del self.qa_chains[pipeline_id]
deleted_components.append('qa_chain')
if pipeline_id in self.vector_stores:
del self.vector_stores[pipeline_id]
deleted_components.append('vector_store')
if deleted_components:
return {
'success': True,
'pipeline_id': pipeline_id,
'deleted_components': deleted_components
}
else:
return {'error': f"Pipeline '{pipeline_id}' not found"}
except Exception as e:
return {'error': f"Pipeline deletion failed: {str(e)}"}
def export_pipeline_config(self, pipeline_id: str) -> Dict[str, Any]:
"""Export pipeline configuration for recreation"""
try:
if pipeline_id not in self.qa_chains:
return {'error': f"Pipeline '{pipeline_id}' not found"}
config = {
'pipeline_id': pipeline_id,
'embedding_model_name': getattr(self.embeddings, 'model_name', 'unknown'),
'llm_model_name': getattr(self.llm, 'model_name', 'unknown'),
'chunker_config': {
'chunk_size': self.chunker.chunk_size,
'chunk_overlap': self.chunker.chunk_overlap
},
'export_timestamp': self._get_timestamp(),
'vector_store_type': 'faiss'
}
return config
except Exception as e:
return {'error': f"Pipeline export failed: {str(e)}"}
def _get_timestamp(self) -> str:
"""Get current timestamp"""
from datetime import datetime
return datetime.now().strftime('%Y-%m-%d %H:%M:%S')
# Utility functions for the module
def optimize_rag_pipeline(documents: List[Document], embeddings_model, llm,
test_queries: List[str]) -> Dict[str, Any]:
"""
Optimize RAG pipeline configuration for given documents and queries
Args:
documents (List[Document]): Documents to optimize for
embeddings_model: Embedding model to use
llm: Language model to use
test_queries (List[str]): Test queries for optimization
Returns:
Dict: Optimization recommendations
"""
try:
# Test different chunking strategies
chunker = VectorChunker(embeddings_model)
chunking_results = chunker.optimize_chunking_strategy(documents, test_queries)
# Test different chunk sizes
optimizer = ChunkingOptimizer(embeddings_model)
size_results = optimizer.optimize_chunk_size(documents, test_queries)
# Create optimized pipeline
best_strategy = chunking_results.get('recommended_strategy', 'semantic')
best_size = size_results.get('optimal_chunk_size', 1000)
# Create optimized chunker
optimized_chunker = VectorChunker(
embeddings_model,
chunk_size=best_size,
chunk_overlap=best_size // 5 # 20% overlap
)
# Test the optimized configuration
pipeline = RAGPipeline(embeddings_model, llm)
pipeline.chunker = optimized_chunker
test_pipeline_id = "optimization_test"
creation_result = pipeline.create_pipeline(documents, test_pipeline_id, best_strategy)
if not creation_result.get('error'):
evaluation_result = pipeline.evaluate_pipeline(test_pipeline_id, test_queries)
pipeline.delete_pipeline(test_pipeline_id) # Clean up
else:
evaluation_result = {'error': 'Could not evaluate optimized pipeline'}
return {
'optimization_complete': True,
'recommended_config': {
'chunking_strategy': best_strategy,
'chunk_size': best_size,
'chunk_overlap': best_size // 5
},
'chunking_optimization': chunking_results,
'size_optimization': size_results,
'performance_evaluation': evaluation_result,
'recommendations': [
f"Use {best_strategy} chunking strategy",
f"Set chunk size to {best_size} characters",
f"Use {best_size // 5} character overlap",
"Monitor and adjust based on query performance"
]
}
except Exception as e:
return {'error': f"RAG optimization failed: {str(e)}"}
def create_demo_rag_system(sample_documents: List[Document], embeddings_model, llm) -> Dict[str, Any]:
"""
Create a demonstration RAG system with sample documents
Args:
sample_documents (List[Document]): Sample documents for demo
embeddings_model: Embedding model
llm: Language model
Returns:
Dict: Demo system information and sample interactions
"""
try:
# Create RAG pipeline
pipeline = RAGPipeline(embeddings_model, llm)
demo_id = "demo_system"
# Create the pipeline
creation_result = pipeline.create_pipeline(sample_documents, demo_id, "semantic")
if creation_result.get('error'):
return {'error': f"Demo system creation failed: {creation_result['error']}"}
# Sample queries for demonstration
demo_queries = [
"What is the main topic of these documents?",
"Can you summarize the key points?",
"What are the most important concepts mentioned?"
]
# Execute demo queries
demo_results = []
for query in demo_queries:
result = pipeline.query_pipeline(demo_id, query, return_sources=True)
demo_results.append(result)
# Get system statistics
pipeline_info = pipeline.get_pipeline_info(demo_id)
return {
'demo_system_created': True,
'pipeline_id': demo_id,
'creation_stats': creation_result,
'pipeline_info': pipeline_info,
'demo_queries': demo_queries,
'demo_results': demo_results,
'usage_instructions': [
f"Use pipeline.query_pipeline('{demo_id}', 'your question') to ask questions",
"The system will return answers with source document references",
"Sources show which parts of the documents were used for the answer"
]
}
except Exception as e:
return {'error': f"Demo system creation failed: {str(e)}"}
# Export the main classes for use in other modules
__all__ = [
'VectorChunker',
'ChunkingOptimizer',
'RAGPipeline',
'optimize_rag_pipeline',
'create_demo_rag_system'
] |