Update app.py

app.py

@@ -9,14 +9,14 @@ from sklearn.metrics.pairwise import cosine_similarity
 # ==========================================
 # 1. SETUP & DATA LOADING
 # ==========================================
-# File paths (Assuming files are in the same root directory as app.py on Hugging Face)
+# Assuming files are in the same root directory as app.py
 CSV_PATH = "cleaned_dataset_10k.csv"
 PKL_PATH = "final_embeddings_10k.pkl"
 
 if not os.path.exists(CSV_PATH) or not os.path.exists(PKL_PATH):
-    raise FileNotFoundError("Missing required data files. Please ensure CSV and PKL are uploaded.")
+    raise FileNotFoundError("Missing required data files. Ensure CSV and PKL are uploaded.")
 
-# Load the dataset
+# Load the restaurant dataset
 df = pd.read_csv(CSV_PATH)
 df.columns = [c.strip().lower().replace(' ', '_') for c in df.columns]
 
@@ -25,10 +25,10 @@ with open(PKL_PATH, 'rb') as f:
     embedding_data = pickle.load(f)
     dataset_embeddings = embedding_data['embeddings'] if isinstance(embedding_data, dict) else embedding_data
 
-# Load the semantic model (MPNet for high accuracy)
+# Load the semantic transformer model (MPNet for high fidelity)
 model = SentenceTransformer('sentence-transformers/all-mpnet-base-v2')
 
-# Pre-calculate Persona Taste Profiles (Mean vectors)
+# Pre-calculate Persona Taste Profiles (Mean Vectors)
 persona_profiles = {}
 for persona in df['reviewer_persona'].unique():
     if pd.isna(persona): continue
@@ -41,38 +41,37 @@ for persona in df['reviewer_persona'].unique():
 # ==========================================
 def run_ven_engine(budget, dietary, company, purpose, noise):
     """
-    Finds the best restaurant match using a combination of 
-    Persona Matching and Contextual Semantic Similarity.
+    Finds the best restaurant match using Hybrid Scoring:
+    70% Semantic Contextual Fit + 30% User Rating.
     """
-    # Create the user's context string
+    # Construct the user's semantic context
     user_context = f"Searching for a {budget} experience, {dietary} friendly. Group: {company}. Occasion: {purpose}. Atmosphere: {noise}."
     query_vec = model.encode([user_context])
     
-    # Step A: Identify the closest Persona Profile
+    # Step A: Identify the closest overall Persona Profile
     persona_sims = {p: cosine_similarity(query_vec, v.reshape(1, -1))[0][0] 
                     for p, v in persona_profiles.items()}
     closest_persona = max(persona_sims, key=persona_sims.get)
     
-    # Step B: Filter reviews by that persona
+    # Step B: Filter reviews belonging to that persona
     persona_indices = df[df['reviewer_persona'] == closest_persona].index
     persona_embeddings = dataset_embeddings[persona_indices]
     
     # Step C: Calculate Contextual Similarity for specific reviews within that persona
-    # This prevents getting the same result every time
     sub_similarities = cosine_similarity(query_vec, persona_embeddings)[0]
     
     persona_df = df.loc[persona_indices].copy()
     persona_df['semantic_fit'] = sub_similarities
     persona_df['norm_rating'] = persona_df['rating_score'] / 5.0
     
-    # HYBRID SCORE: 70% Contextual Fit + 30% Rating Quality
+    # CALCULATE FINAL SCORE
     persona_df['final_score'] = (persona_df['semantic_fit'] * 0.7) + (persona_df['norm_rating'] * 0.3)
     
-    # Pick the top result after re-ranking
+    # Select the top re-ranked result
     top_match = persona_df.sort_values(by='final_score', ascending=False).iloc[0]
     match_pct = int(top_match['final_score'] * 100)
     
-    # Format Recommendation Card (HTML)
+    # Return Styled HTML Result Card
     return f"""
     <div style="background: white; border-radius: 20px; padding: 25px; color: #0f172a !important; text-align: left; border-left: 10px solid #f97316; box-shadow: 0 10px 25px rgba(0,0,0,0.2);">
         <div style="display:flex; justify-content:space-between; align-items: flex-start;">
@@ -97,20 +96,28 @@ def run_ven_engine(budget, dietary, company, purpose, noise):
     """
 
 # ==========================================
-# 3. GRADIO UI SETUP (HF OPTIMIZED)
+# 3. GRADIO UI SETUP (FINAL VISIBILITY FIX)
 # ==========================================
-# CSS Fixes for Visibility and Theme consistency
 ven_css = """
+/* 1. Dark background for the whole app */
 .gradio-container { background-color: #0f172a !important; }
 h1 { color: white !important; text-align: center; font-weight: 900 !important; font-size: 2.5rem !important; margin-bottom: 20px !important; }
 
-/* Force input labels to be white and readable */
-label span { color: white !important; font-weight: 700 !important; font-size: 15px !important; margin-bottom: 5px !important; }
+/* 2. Main Labels (e.g., "3. Social Context") -> MUST BE WHITE */
+label span { 
+    color: white !important; 
+    font-weight: 700 !important; 
+    font-size: 15px !important; 
+}
 
-/* Force Radio button text options to be white and bold */
-.gr-radio label span { color: white !important; font-weight: 600 !important; }
+/* 3. RADIO CHOICE TEXT (e.g., "Solo", "Date") -> MUST BE DARK SATE */
+/* Since radio buttons have white backgrounds, white text is invisible. We force it to dark blue/grey. */
+.gr-radio label span, .gr-radio span { 
+    color: #1e293b !important; 
+    font-weight: 600 !important; 
+}
 
-/* Style the primary orange button */
+/* 4. Orange primary button styling */
 .ven-button { 
     background-color: #f97316 !important; 
     color: white !important; 
@@ -121,7 +128,7 @@ label span { color: white !important; font-weight: 700 !important; font-size: 15
     border-radius: 12px !important;
 }
 
-/* Fix example table colors */
+/* 5. Quick Vibe Starters table colors */
 .gr-samples-table { background-color: #1e293b !important; color: white !important; }
 """