Update app.py

app.py

@@ -2,49 +2,53 @@ import pickle
 import streamlit as st
 import numpy as np
 
-# Load saved vectorizer, model, and binarizer
-with open("vectorizer (3).pkl", "rb") as f:
+# Load saved model, vectorizer, and binarizer
+with open("vectorizer.pkl", "rb") as f:
     vectorizer = pickle.load(f)
 
-with open("model (6).pkl", "rb") as f:
+with open("model.pkl", "rb") as f:
     model = pickle.load(f)
 
-with open("binarizer (3).pkl", "rb") as f:
+with open("binarizer.pkl", "rb") as f:
     mlb = pickle.load(f)
 
 st.title("🔖 Stack Overflow Tags Predictor")
-st.markdown("Enter a question title and description. Tags will be predicted automatically based on model confidence.")
+st.markdown("Enter a question title and description. The top 3 most relevant tags will be predicted automatically.")
 
+# Input fields
 title = st.text_input("📌 Enter Question Title")
 description = st.text_area("📝 Enter Question Description", height=150)
 
-# 🔧 Adjust this to control how many tags are returned
-threshold = 0.2  # Lower threshold means more tags predicted
-
-def predict_tags_auto(title, description, threshold=0.2):
+# Function to predict top N tags (e.g., top 3)
+def predict_tags_top_n(title, description, top_n=3):
     input_text = title + " " + description
     input_vector = vectorizer.transform([input_text])
 
-    # Get probabilities for each tag
+    # Get probability estimates from each classifier
     probas = model.predict_proba(input_vector)
 
-    # Get the probability for class=1 (relevant tag) for each classifier
-    probas_array = np.array([p[:, 1][0] for p in probas])  # (n_classes,)
+    # Extract positive class probabilities for each tag
+    probas_array = np.array([p[0][1] for p in probas])  # shape: (n_classes,)
+
+    # Get indices of top N tags
+    top_indices = probas_array.argsort()[-top_n:][::-1]
 
-    # Apply threshold
-    predicted_binary = (probas_array >= threshold).astype(int).reshape(1, -1)
+    # Build binary array for top tags
+    predicted_binary = np.zeros_like(probas_array, dtype=int)
+    predicted_binary[top_indices] = 1
+    predicted_binary = predicted_binary.reshape(1, -1)
 
-    # Convert binary vector to tags
+    # Convert binary to tag names
     tags = mlb.inverse_transform(predicted_binary)
     return tags[0] if tags else []
 
-
+# Predict and display
 if st.button("Predict Tags"):
     if not title.strip() or not description.strip():
         st.warning("⚠️ Please enter both title and description.")
     else:
-        tags = predict_tags_auto(title, description, threshold)
-        if tags:
-            st.success("✅ Predicted Tags: " + ", ".join(tags))
+        predicted_tags = predict_tags_top_n(title, description, top_n=3)
+        if predicted_tags:
+            st.success("✅ Predicted Tags: " + ", ".join(predicted_tags))
         else:
             st.info("ℹ️ No tags predicted. Try refining your question.")