Update app.py

app.py

@@ -2,127 +2,98 @@
 
 import gradio as gr
 import pandas as pd
+import torch
 from sentence_transformers import SentenceTransformer, util
 
 # 1) Load your committed Excel database
 df = pd.read_excel("BuyWellDataBase.xlsx")
 
-# 2) Prepare for embedding
+# 2) Prepare the "combined" field and compute embeddings for each row
 df["combined"] = df["Company Name"] + " — " + df["Product or Services"]
 model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
-embeddings = model.encode(df["combined"].tolist(), convert_to_tensor=True)
+row_embeddings = model.encode(df["combined"].tolist(), convert_to_tensor=True)
 
-# 3) Define blocked‐brand → sector map (fully extended)
-BLOCKED = {
-    # Original “everyday” brands
-    "amazon":           "Online Store",
-    "booking":          "Booking",
-    "siemens":          "Home Appliances",
-
-    # BDS “Act Now” consumer‐boycott targets
-    "barclays":         "Banking",
-    "bae systems":      "Electronics",
-    "boeing":           "Electronics",
-    "caterpillar":      "Electronics",
-    "elbit systems":    "Electronics",
-    "lockheed martin":  "Electronics",
-    "general dynamics": "Electronics",
-    "israel aerospace industries": "Electronics",
-
-    # Your added consumer‐boycott targets
-    "hp":               "Electronics",
-    "hpe":              "Electronics",
-    "chevron":          "Energy Providers",
-    "caltex":           "Energy Providers",
-    "texaco":           "Energy Providers",
-    "carrefour":        "Food",
-    "axa":              "Insurance",
-    "sodastream":       "Home Appliances",
-    "ahava":            "Online Store",
-    "re/max":           "Booking",
-
-    # Divestment & exclusion targets
-    "intel":            "Electronics",
-    "hd hyundai":       "Electronics",
-    "volvo":            "Car",
-    "cat":              "Electronics",   # Caterpillar
-    "jcb":              "Electronics",
-    "caf":              "Electronics",
-    "noble energy":     "Energy Providers",
-
-    # Pressure targets
-    "google":           "Search Engines",
-    "airbnb":           "Booking",
-    "expedia":          "Booking",
-    "disney":           "Media Outlets",
-    "teva":             "Online Store",
-
-    # Grassroots organic boycott
-    "mcdonald":         "Food",
-    "burger king":      "Food",
-    "papa john":        "Food",
-    "pizza hut":        "Food",
-    "wix":              "Telecom & Internet",
-}
+# 3) Build a centroid embedding for each sector
+sectors = df["Sector"].unique().tolist()
+sector_embeddings = {}
+for sector in sectors:
+    # indices of rows in this sector
+    idxs = df.index[df["Sector"] == sector].tolist()
+    # mean of their embeddings
+    sector_embeddings[sector] = row_embeddings[idxs].mean(dim=0, keepdim=True)
 
 def recommend_store(query: str) -> str:
-    query_clean = query.strip().lower()
-    if not query_clean:
-        return "❗ Please enter what you’d like to buy."
-
-    # 1) Check if the query mentions any blocked brand
-    for brand, sector in BLOCKED.items():
-        if brand in query_clean:
-            # Recommend the top 5 sustainable alternatives in that sector
-            alts = (
-                df[df["Sector"] == sector]
-                .sort_values("Rating", ascending=False)
-                .head(5)
-            )
-            lines = [
-                f"- **{row['Company Name']}** — {row['Product or Services']}\n"
-                f"  Rating: {row['Rating']} | [Website]({row['Website']})"
-                for _, row in alts.iterrows()
-            ]
-            return (
-                f"### Oops—“{brand.title()}” isn’t in our sustainable list!  \n"
-                f"Here are 5 ethical alternatives in **{sector}**:\n\n"
-                + "\n".join(lines)
-            )
+    q = query.strip()
+    if not q:
+        return "<p style='color:red;'>❗ Please enter what you’d like to buy.</p>"
+
+    # 1) Embed the query
+    q_emb = model.encode(q, convert_to_tensor=True, normalize_embeddings=True)
+
+    # 2) Find the best-matching sector via cosine similarity
+    sector_sims = {
+        sector: util.cos_sim(q_emb, emb).item()
+        for sector, emb in sector_embeddings.items()
+    }
+    best_sector = max(sector_sims, key=sector_sims.get)
+
+    # 3) Within that sector, score all rows and pick top 5
+    mask = df["Sector"] == best_sector
+    sector_idxs = df.index[mask].tolist()
+    sims = util.cos_sim(q_emb, row_embeddings[sector_idxs])[0]
+    top5 = torch.topk(sims, k=min(5, len(sector_idxs))).indices.tolist()
+
+    # 4) Build an HTML table of the top 5
+    rows = []
+    for rank_idx in top5:
+        df_idx = sector_idxs[rank_idx]
+        row = df.loc[df_idx]
+        rows.append(f"""
+<tr>
+  <td><strong>{row['Company Name']}</strong></td>
+  <td>{row['Product or Services']}</td>
+  <td style="text-align:center;">{row['Rating']}</td>
+  <td>{row['Google Map Location']}</td>
+  <td><a href="{row['Website']}" target="_blank">Visit site</a></td>
+</tr>
+""")
+    table_html = f"""
+<h3>Top 5 alternatives in <em>{best_sector}</em></h3>
+<table border="1" cellpadding="4" cellspacing="0">
+  <thead style="background:#f0f0f0;">
+    <tr>
+      <th>Company</th>
+      <th>Product / Services</th>
+      <th>Rating</th>
+      <th>Location</th>
+      <th>Website</th>
+    </tr>
+  </thead>
+  <tbody>
+    {''.join(rows)}
+  </tbody>
+</table>
+"""
 
-    # 2) Otherwise, do the normal semantic search
-    q_emb = model.encode(query, convert_to_tensor=True)
-    scores = util.cos_sim(q_emb, embeddings)[0]
-    best = df.iloc[int(scores.argmax())]
-    return (
-        f"### {best['Company Name']}\n\n"
-        f"**Sector:** {best['Sector']}\n\n"
-        f"**Product / Services:** {best['Product or Services']}\n\n"
-        f"**Rating:** {best['Rating']}\n\n"
-        f"**Location:** {best['Google Map Location']}\n\n"
-        f"**Website:** [Visit site]({best['Website']})"
-    )
+    return table_html
 
-# 4) Build Gradio UI
+# 5) Build Gradio UI
 with gr.Blocks(title="GoodBuy Guide") as demo:
-    gr.Markdown(
-        """
-# 🌿 GoodBuy Guide
-
-Discover ethical, democratic & sustainable alternatives—one “good buy” at a time!
-
-*Feedback or issues?* 📧 **msalanfortuny@gmail.com**
-"""
-    )
+    gr.HTML("""
+    <h1 style="text-align:center; color: #2a7f4f;">🌿 GoodBuy Guide</h1>
+    <p style="text-align:center;">Discover ethical, democratic & sustainable alternatives—one “good buy” at a time!</p>
+    <p style="text-align:center; font-size:0.9em;">*Feedback or issues?* 📧 <a href="mailto:msalanfortuny@gmail.com">msalanfortuny@gmail.com</a></p>
+    """)
     with gr.Row():
         inp = gr.Textbox(
-            placeholder="e.g. organic coffee, Amazon Prime, Siemens fridge…",
-            label="What do you want to buy?"
+            placeholder="e.g. coffee beans, Amazon Prime, Siemens fridge…",
+            label="What do you want to buy?",
+            interactive=True
         )
-        btn = gr.Button("Find a Good Buy")
-    out = gr.Markdown()
-    btn.click(recommend_store, inp, out)
-    inp.submit(recommend_store, inp, out)
+        btn = gr.Button("Find a Good Buy", variant="primary")
+    out = gr.HTML()
+    btn.click(fn=recommend_store, inputs=inp, outputs=out)
+    inp.submit(fn=recommend_store, inputs=inp, outputs=out)
 
 if __name__ == "__main__":
     demo.launch()