Update app.py

app.py

@@ -1,73 +1,192 @@
-import pandas as pd
-from sentence_transformers import SentenceTransformer, util
 import gradio as gr
+import pandas as pd
+import numpy as np
 import pickle
 import os
+from sentence_transformers import SentenceTransformer
+from sklearn.metrics.pairwise import cosine_similarity
+
+# ==========================================
+# 1. INITIALIZATION & DATA LOADING
+# ==========================================
 
-# 1. Load the Dataset and Embeddings
-dataset_path = "cleaned_dataset_10k.csv"
-embeddings_path = "final_embeddings_10k.pkl"
+# NOTE: We use relative paths because the files are in the same Hugging Face Space
+csv_path = "cleaned_dataset_10k.csv"
+pkl_path = "final_embeddings_10k.pkl"
 
-# Check if files exist
-if not os.path.exists(dataset_path) or not os.path.exists(embeddings_path):
-    raise FileNotFoundError("Files not found. Please upload cleaned_dataset_10k.csv and final_embeddings_10k.pkl")
+if not os.path.exists(csv_path) or not os.path.exists(pkl_path):
+    raise FileNotFoundError("❌ Missing files! Please upload 'cleaned_dataset_10k.csv' and 'final_embeddings_10k.pkl' to the Files tab.")
 
 # Load Data
-df = pd.read_csv(dataset_path)
-
-# Load Embeddings
-with open(embeddings_path, "rb") as fIn:
-    stored_data = pickle.load(fIn)
-    stored_embeddings = stored_data['embeddings']
-
-# 2. Load the Model
-model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
-
-# 3. Define the Search Function
-def search_restaurant(query):
-    # Encode the user's query
-    query_embedding = model.encode(query, convert_to_tensor=True)
-
-    # Perform semantic search (find top 3 matches)
-    hits = util.semantic_search(query_embedding, stored_embeddings, top_k=3)
-    hits = hits[0]
-
-    results = []
-    for hit in hits:
-        row_id = hit['corpus_id']
-        row = df.iloc[row_id]
-
-        # Create a nice text output for each result
-        result_text = (
-            f"🍽️ **Name:** {row['Restaurant Name']}\n"
-            f"🥘 **Cuisine:** {row['Food Type']}\n"
-            f"⭐ **Rating:** {row['Rating']}\n"
-            f"📍 **Address:** {row['Address']}\n"
-            f"💬 **Review:** \"{row['Review']}\"\n"
-            f"----------------------------------------"
-        )
-        results.append(result_text)
-
-    return "\n\n".join(results)
-
-# 4. Create the App Interface
-# These examples satisfy instruction #7 ("3 Quick Starters")
-examples = [
-    ["I want a romantic italian dinner"],
-    ["Best sushi place with good service"],
-    ["Cheap fast food for lunch"]
-]
-
-interface = gr.Interface(
-    fn=search_restaurant,
-    inputs=gr.Textbox(lines=2, placeholder="Type here... (e.g., 'Spicy Mexican food')"),
-    outputs=gr.Markdown(label="Recommended Restaurants"),
-    title="Restaurant Recommendation System 🍔",
-    description="Describe what you want to eat, and I'll find the best match!",
-    examples=examples,
-    theme="default"
-)
-
-# 5. Launch
+df = pd.read_csv(csv_path)
+with open(pkl_path, 'rb') as f:
+    embedding_data = pickle.load(f)
+    dataset_embeddings = embedding_data['embeddings']
+
+# Load the model
+# NOTE: Using the model Gal specified. 
+# If you get a "dimension mismatch" error, change this back to 'sentence-transformers/all-MiniLM-L6-v2'
+model = SentenceTransformer('sentence-transformers/all-mpnet-base-v2')
+
+# Pre-calculate Persona Taste Centers (Mean vectors)
+# This finds the "average" taste for each type of reviewer in your data
+persona_profiles = {}
+if 'reviewer_persona' in df.columns:
+    for persona in df['reviewer_persona'].unique():
+        indices = df[df['reviewer_persona'] == persona].index
+        # We must ensure we only take embeddings that exist in the dataframe indices
+        valid_indices = [i for i in indices if i < len(dataset_embeddings)]
+        if valid_indices:
+            persona_vectors = dataset_embeddings[valid_indices]
+            persona_profiles[persona] = np.mean(persona_vectors, axis=0)
+else:
+    # Fallback if 'reviewer_persona' column is missing, just use global average
+    persona_profiles['Default'] = np.mean(dataset_embeddings, axis=0)
+
+# ==========================================
+# 2. DESIGN SYSTEM (VEN BRANDING)
+# ==========================================
+ven_css = """
+body {
+    background: radial-gradient(1200px 600px at 20% 0%, #eef6ff 0%, #f8fafc 45%, #ffffff 100%) !important;
+    font-family: 'Inter', system-ui, -apple-system, sans-serif !important;
+}
+.ven-card {
+    background: white; border: 1px solid rgba(15,23,42,0.08);
+    border-radius: 24px; box-shadow: 0 20px 40px rgba(2,6,23,0.08);
+    overflow: hidden; padding: 0; transition: transform 0.3s ease;
+}
+.ven-badge {
+    width: 48px; height: 48px; border-radius: 16px; display: grid; place-items: center;
+    background: linear-gradient(135deg, #006CE4, #3b82f6); color: white; font-weight: 900;
+}
+.ven-chip {
+    padding: 6px 14px; border-radius: 100px; font-size: 12px; font-weight: 700;
+    background: #f1f5f9; color: #475569; border: 1px solid #e2e8f0;
+}
+.ven-bar-bg { height: 8px; border-radius: 100px; background: #f1f5f9; margin-top: 8px; }
+.ven-bar-fill { height: 100%; border-radius: 100px; background: #006CE4; }
+.ven-btn {
+    background: #006CE4; color: white !important; border: none;
+    padding: 14px 28px; border-radius: 14px; font-weight: 800; cursor: pointer;
+    width: 100%; transition: opacity 0.2s;
+}
+.ven-btn:hover { opacity: 0.9; }
+"""
+
+# ==========================================
+# 3. COMPONENT GENERATORS
+# ==========================================
+def format_recommendation_ui(res_name, rating, persona, score, review):
+    match_pct = int(score * 100)
+    
+    # Safety check for review text
+    review_display = review[:160] + "..." if isinstance(review, str) else "Great place!"
+
+    return f"""
+    <div class="ven-card">
+        <div style="padding: 24px;">
+            <div style="display: flex; justify-content: space-between; align-items: flex-start;">
+                <div style="display: flex; gap: 16px;">
+                    <div class="ven-badge">✨</div>
+                    <div>
+                        <h2 style="margin:0; font-size:22px; font-weight:900; color:#0f172a;">{res_name}</h2>
+                        <div style="margin-top:8px; display:flex; gap:8px;">
+                            <span class="ven-chip" style="background:#fff7ed; color:#c2410c; border-color:#fed7aa;">Top pick for {persona}</span>
+                        </div>
+                    </div>
+                </div>
+                <div style="text-align: right;">
+                    <div style="font-size:28px; font-weight:900; color:#006CE4;">{rating:.1f}</div>
+                    <div style="font-size:12px; font-weight:700; color:#94a3b8;">RATING</div>
+                </div>
+            </div>
+
+            <div style="margin-top:24px;">
+                <div style="display:flex; justify-content:space-between; font-weight:800; font-size:14px;">
+                    <span>VEN Match Confidence</span>
+                    <span style="color:#006CE4;">{match_pct}%</span>
+                </div>
+                <div class="ven-bar-bg"><div class="ven-bar-fill" style="width:{match_pct}%"></div></div>
+            </div>
+
+            <div style="margin-top:24px; padding:16px; background:#f8fafc; border-radius:16px;">
+                <p style="margin:0; font-size:14px; line-height:1.6; color:#334155;">
+                    <b>Why it's a match:</b> Based on your context, this venue aligns with the preferences of our <b>{persona}</b> profile.
+                    Users said: "<i>{review_display}</i>"
+                </p>
+            </div>
+
+            <div style="margin-top:24px;">
+                <button class="ven-btn">Reserve with VEN Exclusive</button>
+            </div>
+        </div>
+    </div>
+    """
+
+# ==========================================
+# 4. LOGIC ENGINE
+# ==========================================
+def run_ven_engine(budget, dietary, company, purpose, noise):
+    # Construct descriptive bio
+    user_context = f"Searching for a {budget} experience, {dietary} friendly. Group: {company}. Occasion: {purpose}. Atmosphere: {noise}."
+
+    # Semantic Search
+    query_vec = model.encode([user_context])
+    
+    # Find closest persona
+    similarities = {p: cosine_similarity(query_vec, v.reshape(1, -1))[0][0]
+                    for p, v in persona_profiles.items()}
+    closest_persona = max(similarities, key=similarities.get)
+
+    # Filter data for that persona
+    persona_df = df[df['reviewer_persona'] == closest_persona]
+    
+    # Safety: If no restaurants found for this persona, pick from the whole list
+    if persona_df.empty:
+        persona_df = df
+        
+    # Get highest rated in that group
+    top_match = persona_df.sort_values(by='Rating', ascending=False).iloc[0]
+
+    return format_recommendation_ui(
+        top_match['Restaurant Name'],
+        top_match['Rating'],
+        closest_persona,
+        similarities[closest_persona],
+        top_match['Review']
+    )
+
+# ==========================================
+# 5. UI LAYOUT
+# ==========================================
+with gr.Blocks(css=ven_css, title="VEN — AI Matchmaker") as demo:
+    gr.HTML("<div style='text-align:center; padding: 40px 0;'><h1 style='font-size:36px; font-weight:950; color:#0f172a;'>VEN</h1><p style='color:#64748b; font-weight:600;'>Semantic Restaurant Discovery for Tel Aviv</p></div>")
+
+    with gr.Row():
+        with gr.Column(scale=1):
+            with gr.Group():
+                gr.Markdown("### 🔍 Filter your Vibe")
+                in_budget = gr.Dropdown(["Budget-friendly", "Mid-range", "Premium"], label="Budget", value="Mid-range")
+                in_diet = gr.Dropdown(["Anything", "Vegetarian", "Vegan", "Meat-lover"], label="Diet", value="Anything")
+                in_company = gr.Radio(["Solo", "Date/Couple", "Friends", "Business"], label="With who?", value="Date/Couple")
+                in_purpose = gr.Dropdown(["Casual dinner", "Special occasion", "Quick bite", "Professional meeting"], label="Occasion", value="Casual dinner")
+                in_noise = gr.Radio(["Quiet/Intimate", "Moderate/Social", "Lively/Music"], label="Environment", value="Moderate/Social")
+
+                search_btn = gr.Button("Find My Table", variant="primary")
+
+        with gr.Column(scale=1.5):
+            gr.Markdown("### 🎯 Your Personal Match")
+            output_ui = gr.HTML("<div style='text-align:center; padding:100px; color:#cbd5e1; font-weight:600; border:2px dashed #e2e8f0; border-radius:24px;'>Adjust the filters to generate your AI recommendation</div>")
+
+    search_btn.click(
+        fn=run_ven_engine,
+        inputs=[in_budget, in_diet, in_company, in_purpose, in_noise],
+        outputs=output_ui
+    )
+
+# ==========================================
+# 6. LAUNCH
+# ==========================================
 if __name__ == "__main__":
-    interface.launch()
+    demo.launch()