import numpy as np import pandas as pd import joblib import streamlit as st from sklearn.metrics.pairwise import cosine_similarity st.set_page_config(page_title="Restaurant Recommendation System", page_icon="đŸŊī¸", layout="wide") st.title("đŸŊī¸ Restaurant Recommendation System") st.write("Select a restaurant and get similar restaurant recommendations using **TF-IDF + Cosine Similarity**.") # ---------------------------- # Load artifacts (cached) # ---------------------------- @st.cache_resource def load_artifacts(): # If your files are in repo root, remove 'src/' from these paths tfidf = joblib.load("src/tfidf.pkl") df = pd.read_pickle("src/restaurants_df.pkl") # Ensure required columns exist required_cols = {"Name", "Type", "Location", "Price_Range", "Comments", "Trip_advisor Url"} missing = required_cols - set(df.columns) if missing: raise ValueError(f"Missing required columns in DataFrame: {missing}") # Ensure combined text column exists if "combined_columns" not in df.columns: df["Type"] = df["Type"].fillna("Unknown") df["Comments"] = df["Comments"].fillna("") df["combined_columns"] = ( df["Type"].astype(str) + " " + df["Location"].astype(str) + " " + df["Price_Range"].astype(str) + " " + df["Comments"].astype(str) ) # Normalize text text = df["combined_columns"].astype(str).str.lower() # Build similarity matrix at runtime (lightweight deployment) X = tfidf.transform(text) sim = cosine_similarity(X) return df, sim # ---------------------------- # Recommender with scores # ---------------------------- def recommend_with_scores(name, df, sim, top_n=5, same_location_only=False, price_filter="All"): # Safety: if list is empty or name not found matches = df.index[df["Name"] == name].tolist() if not matches: return pd.DataFrame({"Message": [f"'{name}' not found in dataset."]}) idx = matches[0] scores = sim[idx] # Sort indices by similarity (desc), exclude itself ranked_idx = np.argsort(scores)[::-1] ranked_idx = ranked_idx[ranked_idx != idx] # Build candidates dataframe candidates = df.loc[ ranked_idx, ["Name", "Type", "Location", "Price_Range", "Trip_advisor Url"] ].copy() candidates["Similarity"] = scores[ranked_idx] # Optional filters if same_location_only: selected_loc = df.loc[idx, "Location"] candidates = candidates[candidates["Location"] == selected_loc] if price_filter and price_filter != "All": candidates = candidates[candidates["Price_Range"] == price_filter] # Take top_n result = candidates.head(top_n).copy() result["Similarity"] = result["Similarity"].round(3) # Clickable link column (Markdown) result["TripAdvisor Link"] = result["Trip_advisor Url"].apply( lambda u: f"[Open]({u})" if isinstance(u, str) and u.startswith("http") else "" ) # Keep clean columns result = result[["Name", "Type", "Location", "Price_Range", "Similarity", "TripAdvisor Link"]] result.reset_index(drop=True, inplace=True) result.index += 1 return result # ---------------------------- # Main app # ---------------------------- try: df, sim_matrix = load_artifacts() except Exception as e: st.error("❌ Could not load artifacts. Please check file paths and required columns.") st.exception(e) st.stop() # ✅ DEBUG (damit du sofort siehst, ob df geladen wurde) with st.expander("🔍 Debug (open if app looks empty)"): st.write("Number of rows:", len(df)) st.write("Columns:", list(df.columns)) st.dataframe(df.head(5), use_container_width=True) # Sidebar settings (optional but stable) st.sidebar.header("âš™ī¸ Settings") top_n = st.sidebar.slider("Number of recommendations", 3, 20, 5, 1) same_location_only = st.sidebar.checkbox("Recommend only from the same location", value=False) price_options = ["All"] + sorted(df["Price_Range"].dropna().astype(str).unique().tolist()) price_filter = st.sidebar.selectbox("Price range filter", price_options, index=0) # Dropdown (IMPORTANT: ensure list not empty) restaurant_list = df["Name"].dropna().astype(str).unique().tolist() restaurant_list.sort() if len(restaurant_list) == 0: st.error("No restaurant names found. Please check your dataset.") st.stop() selected_restaurant = st.selectbox("Select a restaurant", restaurant_list) if st.button("Recommend"): st.subheader(f"Recommendations for: {selected_restaurant}") recs = recommend_with_scores( selected_restaurant, df, sim_matrix, top_n=top_n, same_location_only=same_location_only, price_filter=price_filter ) # Use markdown table so links are visible st.dataframe(recs, use_container_width=True) st.caption("Similarity closer to 1.0 means stronger similarity.") st.divider() with st.expander("â„šī¸ How it works"): st.write( """ - This is a **content-based recommendation system**. - Restaurant data (type, location, price range, comments) is combined into one text feature. - Text is vectorized using **TF-IDF**. - Similar restaurants are found with **Cosine Similarity**. - The similarity matrix is computed at runtime to keep the Space lightweight. """ )