Update app.py

Updated with MIA Analysis

app.py

@@ -20,7 +20,7 @@ from PIL import Image
 import torch
 import torch.nn as nn
 
-# Dynamic import from enhanced IBOM backend 
+# Dynamic import from enhanced IBOM backend (no hardcoded paths)
 from IBOM import DetailedIBOMGenerator, get_paths_with_smart_fallback
 
 # Page configuration
@@ -118,17 +118,17 @@ def load_tbom_data(file_path="TBOM.json"):
                 with open(tbom_path, 'r') as f:
                     return json.load(f)
             except Exception:
-                st.error("DBOM-T file not found. Please ensure TBOM.json is available.")
+                st.error("TBOM file not found. Please ensure TBOM.json is available.")
                 return None
     except Exception as e:
-        st.error(f"Error loading DBOM_T data: {str(e)}")
+        st.error(f"Error loading TBOM data: {str(e)}")
         return None
 
 #Enhanced IBOM Interface
 
 def create_enhanced_ibom_interface(tbom_data):
     """Complete enhanced IBOM interface with all advanced features"""
-    st.markdown('<h2 class="sub-header">🔍 Interactive Decision Bill of Materials (DBOM-I)</h2>', unsafe_allow_html=True)
+    st.markdown('<h2 class="sub-header">🔍 Enhanced Interactive Inference Bill of Materials (IBOM)</h2>', unsafe_allow_html=True)
 
     # Feature overview
     with st.expander("Enhanced Features Overview", expanded=False):
@@ -148,7 +148,7 @@ def create_enhanced_ibom_interface(tbom_data):
         with col2:
             st.markdown("""
             **📥 Export Functionality:**
-            - Download DBOM-I JSON files
+            - Download IBOM JSON files
             - CSV exports for spreadsheet analysis
             - Detailed analysis reports
 
@@ -164,12 +164,8 @@ def create_enhanced_ibom_interface(tbom_data):
         "Upload a mushroom image for comprehensive analysis",
         type=['jpg', 'jpeg', 'png', 'bmp'],
         help="Upload a clear image showing cap, gills, and stalk for detailed AI analysis"
-        )
+    )
 
-    # Display the disclaimer and caption AFTER the file uploader
-    st.info("Disclaimer: **🚨 NEVER consume any mushroom based solely on AI analysis.**")
-    st.caption("This tool is for educational and research purposes only.")
-   
     # Main layout: Image on the left, results on the right
     img_col, results_col = st.columns([1, 2.5])
 
@@ -195,7 +191,7 @@ def create_enhanced_ibom_interface(tbom_data):
                     st.warning("⚠️ Low resolution - results may be less accurate")
 
             # Enhanced analysis trigger
-            if st.button(" Analyze for Detailed DBOM-I", type="primary"):
+            if st.button(" Analyze with Enhanced IBOM", type="primary"):
                 perform_enhanced_ibom_analysis(uploaded_image)
         else:
             st.info("Upload an image to begin analysis")
@@ -208,13 +204,22 @@ def create_enhanced_ibom_interface(tbom_data):
                 - Show any bruising or discoloration
                 - Minimal background distractions
                 """)
-
+    # Results Section
     with results_col:
-        # Display results if available
         if 'enhanced_ibom_data' in st.session_state:
             display_comprehensive_ibom_results()
-            create_concept_score_modification_interface()
             
+            # Add MIA analysis section
+            st.markdown("---")
+            st.markdown("### 🔐 Membership Inference Attack (MIA) Analysis")
+            st.markdown("**Description:** Analysis to distinguish members from non-members using Loss and Confidence metrics.")
+            
+            # Display MIA results for uploaded image
+            if uploaded_image is not None:
+                display_mia_analysis_for_image(st.session_state.enhanced_ibom_data['ibom_generator'], 
+                                            Image.open(uploaded_image))
+            
+            create_concept_score_modification_interface()
 
 
 def perform_enhanced_ibom_analysis(uploaded_file):
@@ -286,7 +291,7 @@ def display_comprehensive_ibom_results():
     # --- Add download button for Original IBOM ---
     original_json_data = json.dumps(original_result, indent=2)
     st.download_button(
-        label="📄 Download Original DBOM-I.json",
+        label="📄 Download Original IBOM.json",
         data=original_json_data,
         file_name=f"IBOM_{data['analysis_timestamp'].strftime('%Y%m%d_%H%M%S')}.json",
         mime="application/json"
@@ -577,7 +582,7 @@ def display_enhanced_export_options():
 
     with col1:
         st.download_button(
-            label="📄 Original DBOM-I JSON",
+            label="📄 Original IBOM JSON",
             data=json.dumps(data['original_result'], indent=2),
             file_name=f"original_IBOM_{timestamp}.json",
             mime="application/json"
@@ -585,7 +590,7 @@ def display_enhanced_export_options():
     with col2:
         if 'modified_result' in data:
             st.download_button(
-                label="📄 Modified DBOM-I JSON",
+                label="📄 Modified IBOM JSON",
                 data=json.dumps(data['modified_result'], indent=2),
                 file_name=f"modified_IBOM_{timestamp}.json",
                 mime="application/json"
@@ -612,7 +617,7 @@ def create_performance_overview(tbom_data):
     test_results = perf_metrics.get('final_test_results', {})
     cv_results = perf_metrics.get('cross_validation_results', {})
 
-    col1, col2, col3 = st.columns(3)
+    col1, col2, col3, col4 = st.columns(4)
 
     with col1:
         test_acc = test_results.get('accuracy', 0) * 100
@@ -638,6 +643,14 @@ def create_performance_overview(tbom_data):
             delta=f"CV: {cv_results.get('mean_pr_auc', 0)*100:.2f}%"
         )
 
+    with col4:
+        training_time = tbom_data.get('generation_details', {}).get('training_time_seconds', 0)
+        st.metric(
+            label="⏱️ Training Time",
+            value=f"{training_time/60:.1f} min",
+            delta=f"{training_time:.0f}s total"
+        )
+
 def create_confusion_matrix_viz(tbom_data):
     """Create interactive confusion matrix visualization (UNCHANGED)"""
     test_results = tbom_data.get('performance_metrics', {}).get('final_test_results', {})
@@ -1068,7 +1081,7 @@ def create_technical_details(tbom_data):
 
         st.markdown(f"""
         <div class="metric-card">
-            <h4>📋 DBOM-T Information</h4>
+            <h4>📋 TBOM Information</h4>
             <p><strong>Version:</strong> {tbom_version}</p>
             <p><strong>Generated:</strong> {timestamp}</p>
             <p><strong>Training Time:</strong> {training_time/3600:.2f} hours</p>
@@ -1097,19 +1110,126 @@ def create_technical_details(tbom_data):
 
     st.markdown("---")
     # --- Add download button for TBOM.json ---
-    st.subheader("Download Full DBOM-T")
+    st.subheader("Download Full TBOM")
     tbom_json_data = json.dumps(tbom_data, indent=2)
     st.download_button(
-        label="📥 Download DBOM-T.json",
+        label="📥 Download TBOM.json",
         data=tbom_json_data,
         file_name="TBOM.json",
         mime="application/json"
     )
 
     # Expander for raw data
-    with st.expander("🔍 View Raw DBOM-T Data"):
+    with st.expander("🔍 View Raw TBOM Data"):
         st.json(tbom_data)
-
+        
+# Membership Inference Attack Analysis tab
+def create_mia_analysis(tbom_data):
+    """Display Membership Inference Attack analysis results"""
+    st.markdown("### 🔐 Membership Inference Attack (MIA) Analysis")
+    
+    mia_data = tbom_data.get('membership_inference_analysis', {})
+    if not mia_data:
+        st.warning("No MIA analysis data available")
+        return
+        
+    # Display description and method
+    st.markdown(f"**Description:** {mia_data.get('description', 'N/A')}")
+    
+    # Create metrics display
+    col1, col2 = st.columns(2)
+    with col1:
+        threshold = mia_data.get('threshold_finding_summary', {}).get('decision_threshold', 0)
+        st.metric(
+            "Decision Threshold",
+            f"{threshold:.4f}",
+            help="Confidence score threshold for member/non-member classification"
+        )
+    
+    with col2:
+        auc_score = mia_data.get('threshold_finding_summary', {}).get('attack_auc_score_on_confidence', 0)
+        st.metric(
+            "Attack AUC Score",
+            f"{auc_score:.4f}",
+            help="Area Under Curve score for the membership inference attack"
+        )
+    
+    # Display visualization if available
+    viz_path = mia_data.get('visualization_artifact')
+    if viz_path and os.path.exists(viz_path):
+        st.markdown("### Confidence Score Distribution")
+        st.image(viz_path, caption="Member vs Non-member Confidence Distributions")
+    
+    # Display detailed statistics
+    st.markdown("### Detailed Statistics")
+    
+    stats_tabs = st.tabs(["Member Statistics", "Non-member Statistics"])
+    
+    with stats_tabs[0]:
+        member_stats = mia_data.get('metric_statistics', {}).get('members', {})
+        if member_stats:
+            st.dataframe(pd.DataFrame(member_stats))
+        else:
+            st.info("No member statistics available")
+            
+    with stats_tabs[1]:
+        non_member_stats = mia_data.get('metric_statistics', {}).get('non_members', {})
+        if non_member_stats:
+            st.dataframe(pd.DataFrame(non_member_stats))
+        else:
+            st.info("No non-member statistics available")
+    
+ # Display MIA analysis for uploaded image  
+def display_mia_analysis_for_image(model_wrapper, image):
+    """Display MIA analysis results for an uploaded image"""
+    st.markdown("### 🔐 MIA Analysis for Uploaded Image")
+    
+    # Get MIA metrics for single image
+    transform = model_wrapper.preprocess
+    img_tensor = transform(image).unsqueeze(0).to(model_wrapper.device)
+    
+    with torch.no_grad():
+        # Feature extraction
+        img_features = model_wrapper.clip_model.encode_image(img_tensor).float()
+        img_features_norm = img_features / img_features.norm(dim=1, keepdim=True)
+        concept_scores = img_features_norm @ model_wrapper.text_embeddings.T
+        hybrid_features = torch.cat([img_features, concept_scores], dim=1)
+        
+        # Get predictions
+        logits = model_wrapper.model(hybrid_features)
+        probabilities = torch.sigmoid(logits)
+        confidence = torch.max(probabilities, 1 - probabilities).item()
+        
+        # Load threshold from TBOM
+        tbom_path = "TBOM.json"
+        with open(tbom_path, 'r') as f:
+            tbom_data = json.load(f)
+        
+        mia_data = tbom_data.get('membership_inference_analysis', {})
+        threshold = mia_data.get('threshold_finding_summary', {}).get('decision_threshold', 0.92)
+        
+        # Display results
+        col1, col2 = st.columns(2)
+        with col1:
+            st.metric(
+                "Confidence Score",
+                f"{confidence:.4f}",
+                help="Model's confidence in its prediction"
+            )
+        
+        with col2:
+            is_member = confidence > threshold
+            st.metric(
+                "MIA Classification",
+                "Training Set Member" if is_member else "Non-Member",
+                help=f"Threshold: {threshold:.4f}"
+            )
+        
+        # Add interpretation
+        if is_member:
+            st.success("✅ This image is likely a member of the training data")
+        else:
+            st.warning("⚠️ This image is likely NOT a member of the training data")
 
 #Main function to run the Streamlit app
 
@@ -1117,13 +1237,13 @@ def main():
     """Main Streamlit application with enhanced IBOM integration"""
     # Header (keeping your original style)
     st.markdown('<h1 class="main-header">🍄 Dependable FungAI</h1>', unsafe_allow_html=True)
-    st.markdown('<h2 class="sub-header">Descision Bill of Materials DBOM-T & DBOM-I</h2>', unsafe_allow_html=True)
+    st.markdown('<h2 class="sub-header">Training Bill of Materials (TBOM) & Enhanced Inference Analysis</h2>', unsafe_allow_html=True)
 
     # Load TBOM data
     tbom_data = load_tbom_data()
 
     if tbom_data is None:
-        st.error("Could not load DBOM-T data. Please ensure the TBOM.json file is available.")
+        st.error("Could not load TBOM data. Please ensure the TBOM.json file is available.")
         st.stop()
 
     # Sidebar information (keeping your original sidebar)
@@ -1145,7 +1265,7 @@ def main():
         st.markdown('</div>', unsafe_allow_html=True)
 
         # File info
-        st.markdown("### 📄 DBOM-T Info")
+        st.markdown("### 📄 TBOM Info")
         tbom_version = tbom_data.get('tbom_version', 'Unknown')
         timestamp = tbom_data.get('generation_details', {}).get('timestamp_utc', 'Unknown')
         st.markdown(f"**Version:** {tbom_version}")
@@ -1153,16 +1273,17 @@ def main():
 
     # Main content tabs
     tabs = st.tabs([
-        "🔍 Detailed DBOM-I",
-        "🎯 Overview",
-        "📈 Performance",
-        "🏗️ Architecture",
-        "🧠 Concepts",
-        "📊 Data Summary",
-        "💻 Technical"
+        "🔍 Enhanced IBOM",
+        "Overview",
+        "Performance",
+        "Architecture",
+        "Concepts",
+        "Data Summary",
+        "Technical",
+        "MIA Analysis"
     ])
 
-    # NEW ENHANCED IBOM TAB
+    # NEW ENHANCED IBOM TAB WITH ALL REQUESTED FEATURES
     with tabs[0]:
         create_enhanced_ibom_interface(tbom_data)
 
@@ -1194,6 +1315,8 @@ def main():
 
     with tabs[6]:
         create_technical_details(tbom_data)
+    with tabs[7]:
+        create_mia_analysis(tbom_data)
 
 if __name__ == "__main__":
     main()