Update rag_utils.py
Browse files- rag_utils.py +170 -123
rag_utils.py
CHANGED
@@ -1,154 +1,201 @@
|
|
|
|
1 |
from langchain.embeddings import HuggingFaceEmbeddings
|
2 |
from langchain.vectorstores import FAISS
|
3 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
4 |
-
from langchain.
|
5 |
-
from langchain.prompts import PromptTemplate
|
6 |
-
from langchain.llms import HuggingFaceHub
|
7 |
-
import os
|
8 |
import logging
|
|
|
|
|
|
|
|
|
9 |
|
10 |
-
#
|
11 |
-
logging.basicConfig(level=logging.INFO
|
|
|
12 |
|
13 |
class RAGSystem:
|
14 |
def __init__(self):
|
|
|
15 |
try:
|
16 |
-
# Initialize embeddings
|
17 |
self.embeddings = HuggingFaceEmbeddings(
|
18 |
-
model_name="sentence-transformers/all-mpnet-base-v2"
|
|
|
19 |
)
|
20 |
self.vector_store = None
|
21 |
self.text_splitter = RecursiveCharacterTextSplitter(
|
22 |
chunk_size=500,
|
23 |
-
chunk_overlap=50
|
|
|
24 |
)
|
25 |
-
|
26 |
-
self.llm = HuggingFaceHub(
|
27 |
-
repo_id="google/flan-t5-large",
|
28 |
-
task="text-generation",
|
29 |
-
model_kwargs={"temperature": 0.7, "max_length": 512}
|
30 |
-
)
|
31 |
-
logging.info("RAG system initialized successfully.")
|
32 |
except Exception as e:
|
33 |
-
|
34 |
-
raise
|
35 |
|
36 |
-
def
|
37 |
-
"""
|
|
|
38 |
try:
|
39 |
-
documents = []
|
40 |
-
# Validate knowledge base
|
41 |
-
self._validate_knowledge_base(knowledge_base)
|
42 |
-
|
43 |
-
# Generate insights and case studies
|
44 |
-
expert_insights = self._generate_expert_insights(knowledge_base)
|
45 |
-
case_studies = self._generate_case_studies()
|
46 |
-
|
47 |
for damage_type, cases in knowledge_base.items():
|
48 |
-
for
|
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 |
-
self.
|
81 |
-
|
82 |
-
|
83 |
-
retriever=self.vector_store.as_retriever(),
|
84 |
-
chain_type_kwargs={
|
85 |
-
"prompt": self._get_qa_prompt()
|
86 |
-
}
|
87 |
)
|
88 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
except Exception as e:
|
90 |
-
|
91 |
-
raise
|
92 |
-
|
93 |
-
def
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
logging.error(f"Missing required field '{key}' in {damage_type}, case {idx + 1}")
|
101 |
-
raise ValueError(f"Missing required field '{key}' in {damage_type}, case {idx + 1}")
|
102 |
-
logging.info("Knowledge base validation passed.")
|
103 |
-
|
104 |
-
def _get_qa_prompt(self):
|
105 |
-
"""Create a custom prompt template for the QA chain"""
|
106 |
-
template = """
|
107 |
-
Context: {context}
|
108 |
-
|
109 |
-
Question: {question}
|
110 |
-
|
111 |
-
Provide a detailed analysis considering:
|
112 |
-
1. Technical aspects
|
113 |
-
2. Safety implications
|
114 |
-
3. Cost-benefit analysis
|
115 |
-
4. Long-term considerations
|
116 |
-
5. Best practices and recommendations
|
117 |
-
|
118 |
-
Answer:
|
119 |
-
"""
|
120 |
-
return PromptTemplate(
|
121 |
-
template=template,
|
122 |
-
input_variables=["context", "question"]
|
123 |
-
)
|
124 |
-
|
125 |
-
def get_enhanced_analysis(self, damage_type, confidence, custom_query=None):
|
126 |
-
"""Get enhanced analysis with dynamic content generation"""
|
127 |
try:
|
128 |
if not self.vector_store:
|
129 |
-
raise ValueError("Vector store
|
130 |
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
Include technical assessment, safety implications, and expert recommendations.
|
135 |
-
"""
|
136 |
else:
|
137 |
-
|
138 |
-
|
|
|
|
|
|
|
139 |
# Get relevant documents
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
|
|
|
|
|
|
|
|
|
|
150 |
}
|
151 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
152 |
except Exception as e:
|
153 |
-
|
154 |
-
return
|
|
|
1 |
+
# rag_utils.py
|
2 |
from langchain.embeddings import HuggingFaceEmbeddings
|
3 |
from langchain.vectorstores import FAISS
|
4 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
5 |
+
from langchain.docstore.document import Document
|
|
|
|
|
|
|
6 |
import logging
|
7 |
+
from typing import List, Dict, Any
|
8 |
+
import numpy as np
|
9 |
+
from tqdm import tqdm
|
10 |
+
import streamlit as st
|
11 |
|
12 |
+
# Set up logging
|
13 |
+
logging.basicConfig(level=logging.INFO)
|
14 |
+
logger = logging.getLogger(__name__)
|
15 |
|
16 |
class RAGSystem:
|
17 |
def __init__(self):
|
18 |
+
"""Initialize RAG system with custom embeddings and configurations"""
|
19 |
try:
|
|
|
20 |
self.embeddings = HuggingFaceEmbeddings(
|
21 |
+
model_name="sentence-transformers/all-mpnet-base-v2",
|
22 |
+
model_kwargs={'device': 'cuda' if st.cuda.is_available() else 'cpu'}
|
23 |
)
|
24 |
self.vector_store = None
|
25 |
self.text_splitter = RecursiveCharacterTextSplitter(
|
26 |
chunk_size=500,
|
27 |
+
chunk_overlap=50,
|
28 |
+
separators=["\n\n", "\n", ". ", ", ", " ", ""]
|
29 |
)
|
30 |
+
logger.info("RAG system initialized successfully")
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
except Exception as e:
|
32 |
+
logger.error(f"Error initializing RAG system: {str(e)}")
|
33 |
+
raise
|
34 |
|
35 |
+
def _create_documents(self, knowledge_base: Dict) -> List[Document]:
|
36 |
+
"""Create documents from knowledge base with structured format"""
|
37 |
+
documents = []
|
38 |
try:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
for damage_type, cases in knowledge_base.items():
|
40 |
+
for case in cases:
|
41 |
+
# Create a detailed document for each case
|
42 |
+
technical_info = f"""
|
43 |
+
Technical Analysis for {damage_type}:
|
44 |
+
Severity Level: {case['severity']}
|
45 |
+
Detailed Description: {case['description']}
|
46 |
+
Primary Location: {case['location']}
|
47 |
+
Required Expertise: {case['required_expertise']}
|
48 |
+
"""
|
49 |
+
|
50 |
+
repair_info = f"""
|
51 |
+
Repair and Maintenance Information:
|
52 |
+
Repair Methods: {' -> '.join(case['repair_method'])}
|
53 |
+
Estimated Cost Range: {case['estimated_cost']}
|
54 |
+
Expected Timeframe: {case['timeframe']}
|
55 |
+
"""
|
56 |
+
|
57 |
+
safety_info = f"""
|
58 |
+
Safety and Prevention Guidelines:
|
59 |
+
Immediate Actions Required: {case['immediate_action']}
|
60 |
+
Preventive Measures: {case['prevention']}
|
61 |
+
Critical Considerations: Special attention needed for {damage_type} in {case['location']}
|
62 |
+
"""
|
63 |
+
|
64 |
+
# Combine all information
|
65 |
+
doc_text = f"{technical_info}\n{repair_info}\n{safety_info}"
|
66 |
+
|
67 |
+
# Create metadata for better retrieval
|
68 |
+
metadata = {
|
69 |
+
'damage_type': damage_type,
|
70 |
+
'severity': case['severity'],
|
71 |
+
'location': case['location'],
|
72 |
+
'document_type': 'construction_damage_analysis'
|
73 |
+
}
|
74 |
+
|
75 |
+
documents.append(Document(
|
76 |
+
page_content=doc_text,
|
77 |
+
metadata=metadata
|
78 |
+
))
|
79 |
|
80 |
+
logger.info(f"Created {len(documents)} documents from knowledge base")
|
81 |
+
return documents
|
82 |
+
except Exception as e:
|
83 |
+
logger.error(f"Error creating documents: {str(e)}")
|
84 |
+
raise
|
85 |
+
|
86 |
+
def initialize_knowledge_base(self, knowledge_base: Dict):
|
87 |
+
"""Initialize vector store with construction knowledge"""
|
88 |
+
try:
|
89 |
+
# Create documents
|
90 |
+
documents = self._create_documents(knowledge_base)
|
91 |
|
92 |
+
# Split documents into chunks
|
93 |
+
splits = self.text_splitter.split_documents(documents)
|
94 |
|
95 |
+
# Create vector store
|
96 |
+
self.vector_store = FAISS.from_documents(
|
97 |
+
documents=splits,
|
98 |
+
embedding=self.embeddings
|
|
|
|
|
|
|
|
|
99 |
)
|
100 |
+
|
101 |
+
logger.info("Knowledge base initialized successfully")
|
102 |
+
except Exception as e:
|
103 |
+
logger.error(f"Error initializing knowledge base: {str(e)}")
|
104 |
+
raise
|
105 |
+
|
106 |
+
def _format_response(self, docs: List[Document], damage_type: str, confidence: float) -> Dict[str, List[str]]:
|
107 |
+
"""Format retrieved documents into structured response"""
|
108 |
+
response = {
|
109 |
+
"technical_details": [],
|
110 |
+
"safety_considerations": [],
|
111 |
+
"expert_recommendations": []
|
112 |
+
}
|
113 |
+
|
114 |
+
try:
|
115 |
+
for doc in docs:
|
116 |
+
content = doc.page_content
|
117 |
+
# Parse technical details
|
118 |
+
if "Technical Analysis" in content:
|
119 |
+
response["technical_details"].append(
|
120 |
+
f"For {damage_type} (Confidence: {confidence:.1f}%):\n" +
|
121 |
+
content.split("Technical Analysis")[1].split("Repair")[0].strip()
|
122 |
+
)
|
123 |
+
|
124 |
+
# Parse safety considerations
|
125 |
+
if "Safety and Prevention" in content:
|
126 |
+
response["safety_considerations"].append(
|
127 |
+
content.split("Safety and Prevention")[1].strip()
|
128 |
+
)
|
129 |
+
|
130 |
+
# Parse repair recommendations
|
131 |
+
if "Repair and Maintenance" in content:
|
132 |
+
response["expert_recommendations"].append(
|
133 |
+
content.split("Repair and Maintenance")[1].split("Safety")[0].strip()
|
134 |
+
)
|
135 |
+
|
136 |
+
return response
|
137 |
except Exception as e:
|
138 |
+
logger.error(f"Error formatting response: {str(e)}")
|
139 |
+
raise
|
140 |
+
|
141 |
+
def get_enhanced_analysis(
|
142 |
+
self,
|
143 |
+
damage_type: str,
|
144 |
+
confidence: float,
|
145 |
+
custom_query: str = None
|
146 |
+
) -> Dict[str, List[str]]:
|
147 |
+
"""Get enhanced analysis with optional custom query support"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
148 |
try:
|
149 |
if not self.vector_store:
|
150 |
+
raise ValueError("Vector store not initialized")
|
151 |
|
152 |
+
# Prepare query
|
153 |
+
if custom_query:
|
154 |
+
query = f"{custom_query} for {damage_type} damage"
|
|
|
|
|
155 |
else:
|
156 |
+
query = f"""
|
157 |
+
Provide detailed analysis for {damage_type} damage with {confidence}% confidence level.
|
158 |
+
Include technical assessment, safety considerations, and repair recommendations.
|
159 |
+
"""
|
160 |
+
|
161 |
# Get relevant documents
|
162 |
+
docs = self.vector_store.similarity_search(
|
163 |
+
query=query,
|
164 |
+
k=3, # Get top 3 most relevant documents
|
165 |
+
fetch_k=5 # Fetch top 5 for better diversity
|
166 |
+
)
|
167 |
+
|
168 |
+
# Format and return response
|
169 |
+
return self._format_response(docs, damage_type, confidence)
|
170 |
+
|
171 |
+
except Exception as e:
|
172 |
+
logger.error(f"Error getting enhanced analysis: {str(e)}")
|
173 |
+
return {
|
174 |
+
"technical_details": [f"Error retrieving analysis: {str(e)}"],
|
175 |
+
"safety_considerations": ["Please try again or contact support."],
|
176 |
+
"expert_recommendations": ["System currently unavailable."]
|
177 |
}
|
178 |
+
|
179 |
+
def get_similar_cases(self, damage_type: str, confidence: float) -> List[Dict[str, Any]]:
|
180 |
+
"""Get similar damage cases for comparison"""
|
181 |
+
try:
|
182 |
+
if not self.vector_store:
|
183 |
+
raise ValueError("Vector store not initialized")
|
184 |
+
|
185 |
+
query = f"Find similar cases of {damage_type} damage"
|
186 |
+
docs = self.vector_store.similarity_search(query, k=3)
|
187 |
+
|
188 |
+
similar_cases = []
|
189 |
+
for doc in docs:
|
190 |
+
if doc.metadata['damage_type'] != damage_type: # Avoid same damage type
|
191 |
+
similar_cases.append({
|
192 |
+
'damage_type': doc.metadata['damage_type'],
|
193 |
+
'severity': doc.metadata['severity'],
|
194 |
+
'location': doc.metadata['location'],
|
195 |
+
'details': doc.page_content[:200] + '...' # First 200 chars
|
196 |
+
})
|
197 |
+
|
198 |
+
return similar_cases
|
199 |
except Exception as e:
|
200 |
+
logger.error(f"Error getting similar cases: {str(e)}")
|
201 |
+
return []
|