import streamlit as st import pandas as pd import plotly.express as px import matplotlib.pyplot as plt import numpy as np import lightgbm as lgb from sklearn.feature_extraction.text import TfidfVectorizer from sklearn.metrics.pairwise import cosine_similarity # Page configuration st.set_page_config(page_title="Customer Insights App", page_icon=":bar_chart:") # Load CSV files at the top, only once df = pd.read_csv("df_clean.csv") nombres_proveedores = pd.read_csv("nombres_proveedores.csv", sep=';') euros_proveedor = pd.read_csv("euros_proveedor.csv", sep=',') ventas_clientes = pd.read_csv("ventas_clientes.csv", sep=',') customer_clusters = pd.read_csv('predicts/customer_clusters.csv') # Load the customer clusters here # Ensure customer codes are strings df['CLIENTE'] = df['CLIENTE'].astype(str) nombres_proveedores['codigo'] = nombres_proveedores['codigo'].astype(str) euros_proveedor['CLIENTE'] = euros_proveedor['CLIENTE'].astype(str) customer_clusters['cliente_id'] = customer_clusters['cliente_id'].astype(str) # Ensure customer IDs are strings fieles_df = pd.read_csv("clientes_relevantes.csv") cestas = pd.read_csv("cestas.csv") productos = pd.read_csv("productos.csv") # Convert all columns except 'CLIENTE' to float in euros_proveedor for col in euros_proveedor.columns: if col != 'CLIENTE': euros_proveedor[col] = pd.to_numeric(euros_proveedor[col], errors='coerce') # Check for NaN values after conversion if euros_proveedor.isna().any().any(): st.warning("Some values in euros_proveedor couldn't be converted to numbers. Please review the input data.") # Ignore the last two columns of df df = df.iloc[:, :-2] # Function to get supplier name def get_supplier_name(code): code = str(code) # Ensure code is a string name = nombres_proveedores[nombres_proveedores['codigo'] == code]['nombre'].values return name[0] if len(name) > 0 else code # Function to create radar chart with square root transformation def radar_chart(categories, values, amounts, title): N = len(categories) angles = [n / float(N) * 2 * np.pi for n in range(N)] angles += angles[:1] fig, ax = plt.subplots(figsize=(12, 12), subplot_kw=dict(projection='polar')) # Apply square root transformation sqrt_values = np.sqrt(values) sqrt_amounts = np.sqrt(amounts) max_sqrt_value = max(sqrt_values) normalized_values = [v / max_sqrt_value for v in sqrt_values] # Adjust scaling for spend values max_sqrt_amount = max(sqrt_amounts) scaling_factor = 0.7 # Adjust this value to control how much the spend values are scaled up normalized_amounts = [min((a / max_sqrt_amount) * scaling_factor, 1.0) for a in sqrt_amounts] normalized_values += normalized_values[:1] ax.plot(angles, normalized_values, 'o-', linewidth=2, color='#FF69B4', label='% Units (sqrt)') ax.fill(angles, normalized_values, alpha=0.25, color='#FF69B4') normalized_amounts += normalized_amounts[:1] ax.plot(angles, normalized_amounts, 'o-', linewidth=2, color='#4B0082', label='% Spend (sqrt)') ax.fill(angles, normalized_amounts, alpha=0.25, color='#4B0082') ax.set_xticks(angles[:-1]) ax.set_xticklabels(categories, size=8, wrap=True) ax.set_ylim(0, 1) circles = np.linspace(0, 1, 5) for circle in circles: ax.plot(angles, [circle]*len(angles), '--', color='gray', alpha=0.3, linewidth=0.5) ax.set_yticklabels([]) ax.spines['polar'].set_visible(False) plt.title(title, size=16, y=1.1) plt.legend(loc='upper right', bbox_to_anchor=(1.3, 1.1)) return fig # Main page design st.title("Welcome to Customer Insights App") st.markdown(""" This app helps businesses analyze customer behaviors and provide personalized recommendations based on purchase history. Use the tools below to dive deeper into your customer data. """) # Navigation menu page = st.selectbox("Select the tool you want to use", ["", "Customer Analysis", "Articles Recommendations"]) # Home Page if page == "": st.markdown("## Welcome to the Customer Insights App") st.write("Use the dropdown menu to navigate between the different sections.") # Customer Analysis Page # elif page == "Customer Analysis": # st.title("Customer Analysis") # st.markdown("Use the tools below to explore your customer data.") # partial_code = st.text_input("Enter part of Customer Code (or leave empty to see all)") # if partial_code: # filtered_customers = df[df['CLIENTE'].str.contains(partial_code)] # else: # filtered_customers = df # customer_list = filtered_customers['CLIENTE'].unique() # customer_code = st.selectbox("Select Customer Code", customer_list) # if st.button("Calcular"): # if customer_code: # # Find Customer's Cluster # customer_match = customer_clusters[customer_clusters['cliente_id'] == customer_code] # if not customer_match.empty: # cluster = customer_match['cluster_id'].values[0] # st.write(f"Customer {customer_code} belongs to cluster {cluster}") # # Load the Corresponding Model # model_path = f'models/modelo_cluster_{cluster}.txt' # gbm = lgb.Booster(model_file=model_path) # st.write(f"Loaded model for cluster {cluster}") # # Load X_predict for that cluster # X_predict_cluster = pd.read_csv(f'predicts/X_predict_cluster_{cluster}.csv') # # Filter for the specific customer # X_cliente = X_predict_cluster[X_predict_cluster['cliente_id'] == customer_code] # if not X_cliente.empty: # # Prepare data for prediction # features_for_prediction = X_cliente.drop(columns=['cliente_id', 'fecha_mes']) # # Make Prediction for the selected customer # y_pred = gbm.predict(features_for_prediction, num_iteration=gbm.best_iteration) # # Reassemble the results # results = X_cliente[['cliente_id', 'marca_id_encoded', 'fecha_mes']].copy() # results['ventas_predichas'] = y_pred # st.write(f"Predicted total sales for Customer {customer_code}: {results['ventas_predichas'].sum():.2f}") # # Load actual data # df_agg_2024 = pd.read_csv('predicts/df_agg_2024.csv') # actual_sales = df_agg_2024[df_agg_2024['cliente_id'] == customer_code] # if not actual_sales.empty: # results = results.merge(actual_sales[['cliente_id', 'marca_id_encoded', 'fecha_mes', 'precio_total']], # on=['cliente_id', 'marca_id_encoded', 'fecha_mes'], # how='left') # results.rename(columns={'precio_total': 'ventas_reales'}, inplace=True) # # Calculate metrics only for non-null actual sales # valid_results = results.dropna(subset=['ventas_reales']) # if not valid_results.empty: # mae = mean_absolute_error(valid_results['ventas_reales'], valid_results['ventas_predichas']) # mape = np.mean(np.abs((valid_results['ventas_reales'] - valid_results['ventas_predichas']) / valid_results['ventas_reales'])) * 100 # rmse = np.sqrt(mean_squared_error(valid_results['ventas_reales'], valid_results['ventas_predichas'])) # st.write(f"Actual total sales for Customer {customer_code}: {valid_results['ventas_reales'].sum():.2f}") # st.write(f"MAE: {mae:.2f}") # st.write(f"MAPE: {mape:.2f}%") # st.write(f"RMSE: {rmse:.2f}") # # Analysis of results # threshold_good = 100 # You may want to adjust this threshold # if mae < threshold_good: # st.success(f"Customer {customer_code} is performing well based on the predictions.") # else: # st.warning(f"Customer {customer_code} is not performing well based on the predictions.") # else: # st.warning(f"No actual sales data found for customer {customer_code} in df_agg_2024.") # # Show the radar chart # all_manufacturers = customer_data.iloc[:, 1:].T # Exclude CLIENTE column # all_manufacturers.index = all_manufacturers.index.astype(str) # sales_data = customer_euros.iloc[:, 1:].T # Exclude CLIENTE column # sales_data.index = sales_data.index.astype(str) # sales_data_filtered = sales_data.drop(index='CLIENTE', errors='ignore') # sales_data_filtered = sales_data_filtered.apply(pd.to_numeric, errors='coerce') # top_units = all_manufacturers.sort_values(by=all_manufacturers.columns[0], ascending=False).head(10) # top_sales = sales_data_filtered.sort_values(by=sales_data_filtered.columns[0], ascending=False).head(10) # combined_top = pd.concat([top_units, top_sales]).index.unique()[:20] # combined_top = [m for m in combined_top if m in all_manufacturers.index and m in sales_data_filtered.index] # combined_data = pd.DataFrame({ # 'units': all_manufacturers.loc[combined_top, all_manufacturers.columns[0]], # 'sales': sales_data_filtered.loc[combined_top, sales_data_filtered.columns[0]] # }).fillna(0) # combined_data_sorted = combined_data.sort_values(by=['units', 'sales'], ascending=False) # non_zero_manufacturers = combined_data_sorted[combined_data_sorted['units'] > 0] # if len(non_zero_manufacturers) < 3: # zero_manufacturers = combined_data_sorted[combined_data_sorted['units'] == 0].head(3 - len(non_zero_manufacturers)) # manufacturers_to_show = pd.concat([non_zero_manufacturers, zero_manufacturers]) # else: # manufacturers_to_show = non_zero_manufacturers # values = manufacturers_to_show['units'].tolist() # amounts = manufacturers_to_show['sales'].tolist() # manufacturers = [get_supplier_name(m) for m in manufacturers_to_show.index] # st.write(f"### Results for top {len(manufacturers)} manufacturers:") # for manufacturer, value, amount in zip(manufacturers, values, amounts): # st.write(f"{manufacturer} = {value:.2f}% of units, €{amount:.2f} total sales") # if manufacturers: # fig = radar_chart(manufacturers, values, amounts, f'Radar Chart for Top {len(manufacturers)} Manufacturers of Customer {customer_code}') # st.pyplot(fig) # else: # st.warning("No data available to create the radar chart.") # # Show sales over the years graph # sales_columns = ['VENTA_2021', 'VENTA_2022', 'VENTA_2023'] # if all(col in ventas_clientes.columns for col in sales_columns): # years = ['2021', '2022', '2023'] # customer_sales = ventas_clientes[ventas_clientes['codigo_cliente'] == customer_code][sales_columns].values[0] # fig_sales = px.line(x=years, y=customer_sales, markers=True, title=f'Sales Over the Years for Customer {customer_code}') # fig_sales.update_layout(xaxis_title="Year", yaxis_title="Sales") # st.plotly_chart(fig_sales) # else: # st.warning("Sales data for 2021-2023 not available.") # else: # st.warning(f"No prediction data found for customer {customer_code}.") # else: # st.warning(f"No data found for customer {customer_code}. Please check the code.") # else: # st.warning("Please select a customer.") elif page == "Customer Analysis": st.title("Customer Analysis") st.markdown("Use the tools below to explore your customer data.") partial_code = st.text_input("Enter part of Customer Code (or leave empty to see all)") if partial_code: filtered_customers = df[df['CLIENTE'].str.contains(partial_code)] else: filtered_customers = df customer_list = filtered_customers['CLIENTE'].unique() customer_code = st.selectbox("Select Customer Code", customer_list) if st.button("Calcular"): if customer_code: # Find Customer's Cluster customer_match = customer_clusters[customer_clusters['cliente_id'] == customer_code] if not customer_match.empty: cluster = customer_match['cluster_id'].values[0] st.write(f"Customer {customer_code} belongs to cluster {cluster}") # Load the Corresponding Model model_path = f'models/modelo_cluster_{cluster}.txt' gbm = lgb.Booster(model_file=model_path) st.write(f"Loaded model for cluster {cluster}") # Inspect the model st.write("### Model Information:") st.write(f"Number of trees: {gbm.num_trees()}") st.write(f"Number of features: {gbm.num_feature()}") st.write("Feature names:") st.write(gbm.feature_name()) # Load X_predict for that cluster X_predict_cluster = pd.read_csv(f'predicts/X_predict_cluster_{cluster}.csv') # Convert cliente_id to string X_predict_cluster['cliente_id'] = X_predict_cluster['cliente_id'].astype(str) st.write("### X_predict_cluster DataFrame:") st.write(X_predict_cluster.head()) st.write(f"Shape: {X_predict_cluster.shape}") # Filter for the specific customer customer_code_str = str(customer_code) X_cliente = X_predict_cluster[X_predict_cluster['cliente_id'] == customer_code_str] # Add debug statements st.write(f"Unique customer IDs in X_predict_cluster: {X_predict_cluster['cliente_id'].unique()}") st.write(f"Customer code we're looking for: {customer_code_str}") st.write("### X_cliente DataFrame:") st.write(X_cliente.head()) st.write(f"Shape: {X_cliente.shape}") if not X_cliente.empty: # Prepare data for prediction features_for_prediction = X_cliente.drop(columns=['cliente_id', 'fecha_mes']) st.write("### Features for Prediction:") st.write(features_for_prediction.head()) st.write(f"Shape: {features_for_prediction.shape}") # Make Prediction for the selected customer y_pred = gbm.predict(features_for_prediction, num_iteration=gbm.best_iteration) st.write("### Prediction Results:") st.write(f"Type of y_pred: {type(y_pred)}") st.write(f"Shape of y_pred: {y_pred.shape}") st.write("First few predictions:") st.write(y_pred[:5]) # Reassemble the results results = X_cliente[['cliente_id', 'marca_id_encoded', 'fecha_mes']].copy() results['ventas_predichas'] = y_pred st.write("### Results DataFrame:") st.write(results.head()) st.write(f"Shape: {results.shape}") st.write(f"Predicted total sales for Customer {customer_code}: {results['ventas_predichas'].sum():.2f}") # Load actual data df_agg_2024 = pd.read_csv('predicts/df_agg_2024.csv') actual_sales = df_agg_2024[df_agg_2024['cliente_id'] == customer_code] st.write("### Actual Sales DataFrame:") st.write(actual_sales.head()) st.write(f"Shape: {actual_sales.shape}") if not actual_sales.empty: results = results.merge(actual_sales[['cliente_id', 'marca_id_encoded', 'fecha_mes', 'precio_total']], on=['cliente_id', 'marca_id_encoded', 'fecha_mes'], how='left') results.rename(columns={'precio_total': 'ventas_reales'}, inplace=True) st.write("### Final Results DataFrame:") st.write(results.head()) st.write(f"Shape: {results.shape}") # Calculate metrics only for non-null actual sales valid_results = results.dropna(subset=['ventas_reales']) if not valid_results.empty: mae = mean_absolute_error(valid_results['ventas_reales'], valid_results['ventas_predichas']) mape = np.mean(np.abs((valid_results['ventas_reales'] - valid_results['ventas_predichas']) / valid_results['ventas_reales'])) * 100 rmse = np.sqrt(mean_squared_error(valid_results['ventas_reales'], valid_results['ventas_predichas'])) st.write(f"Actual total sales for Customer {customer_code}: {valid_results['ventas_reales'].sum():.2f}") st.write(f"MAE: {mae:.2f}") st.write(f"MAPE: {mape:.2f}%") st.write(f"RMSE: {rmse:.2f}") # Analysis of results threshold_good = 100 # You may want to adjust this threshold if mae < threshold_good: st.success(f"Customer {customer_code} is performing well based on the predictions.") else: st.warning(f"Customer {customer_code} is not performing well based on the predictions.") else: st.warning(f"No actual sales data found for customer {customer_code} in df_agg_2024.") # Show the radar chart all_manufacturers = customer_data.iloc[:, 1:].T # Exclude CLIENTE column all_manufacturers.index = all_manufacturers.index.astype(str) sales_data = customer_euros.iloc[:, 1:].T # Exclude CLIENTE column sales_data.index = sales_data.index.astype(str) sales_data_filtered = sales_data.drop(index='CLIENTE', errors='ignore') sales_data_filtered = sales_data_filtered.apply(pd.to_numeric, errors='coerce') top_units = all_manufacturers.sort_values(by=all_manufacturers.columns[0], ascending=False).head(10) top_sales = sales_data_filtered.sort_values(by=sales_data_filtered.columns[0], ascending=False).head(10) combined_top = pd.concat([top_units, top_sales]).index.unique()[:20] combined_top = [m for m in combined_top if m in all_manufacturers.index and m in sales_data_filtered.index] combined_data = pd.DataFrame({ 'units': all_manufacturers.loc[combined_top, all_manufacturers.columns[0]], 'sales': sales_data_filtered.loc[combined_top, sales_data_filtered.columns[0]] }).fillna(0) combined_data_sorted = combined_data.sort_values(by=['units', 'sales'], ascending=False) non_zero_manufacturers = combined_data_sorted[combined_data_sorted['units'] > 0] if len(non_zero_manufacturers) < 3: zero_manufacturers = combined_data_sorted[combined_data_sorted['units'] == 0].head(3 - len(non_zero_manufacturers)) manufacturers_to_show = pd.concat([non_zero_manufacturers, zero_manufacturers]) else: manufacturers_to_show = non_zero_manufacturers values = manufacturers_to_show['units'].tolist() amounts = manufacturers_to_show['sales'].tolist() manufacturers = [get_supplier_name(m) for m in manufacturers_to_show.index] st.write(f"### Results for top {len(manufacturers)} manufacturers:") for manufacturer, value, amount in zip(manufacturers, values, amounts): st.write(f"{manufacturer} = {value:.2f}% of units, €{amount:.2f} total sales") if manufacturers: fig = radar_chart(manufacturers, values, amounts, f'Radar Chart for Top {len(manufacturers)} Manufacturers of Customer {customer_code}') st.pyplot(fig) else: st.warning("No data available to create the radar chart.") # Show sales over the years graph sales_columns = ['VENTA_2021', 'VENTA_2022', 'VENTA_2023'] if all(col in ventas_clientes.columns for col in sales_columns): years = ['2021', '2022', '2023'] customer_sales = ventas_clientes[ventas_clientes['codigo_cliente'] == customer_code][sales_columns].values[0] fig_sales = px.line(x=years, y=customer_sales, markers=True, title=f'Sales Over the Years for Customer {customer_code}') fig_sales.update_layout(xaxis_title="Year", yaxis_title="Sales") st.plotly_chart(fig_sales) else: st.warning("Sales data for 2021-2023 not available.") else: st.warning(f"No prediction data found for customer {customer_code}.") else: st.warning(f"No data found for customer {customer_code}. Please check the code.") else: st.warning("Please select a customer.") # Customer Recommendations Page elif page == "Articles Recommendations": st.title("Articles Recommendations") st.markdown(""" Get tailored recommendations for your customers based on their basket. """) # Campo input para cliente partial_code = st.text_input("Enter part of Customer Code for Recommendations (or leave empty to see all)") if partial_code: filtered_customers = df[df['CLIENTE'].str.contains(partial_code)] else: filtered_customers = df customer_list = filtered_customers['CLIENTE'].unique() customer_code = st.selectbox("Select Customer Code for Recommendations", [""] + list(customer_list)) # Definición de la función recomienda def recomienda(new_basket): # Calcular la matriz TF-IDF tfidf = TfidfVectorizer() tfidf_matrix = tfidf.fit_transform(cestas['Cestas']) # Convertir la nueva cesta en formato TF-IDF new_basket_str = ' '.join(new_basket) new_basket_tfidf = tfidf.transform([new_basket_str]) # Comparar la nueva cesta con las anteriores similarities = cosine_similarity(new_basket_tfidf, tfidf_matrix) # Obtener los índices de las cestas más similares similar_indices = similarities.argsort()[0][-3:] # Las 3 más similares # Crear un diccionario para contar las recomendaciones recommendations_count = {} total_similarity = 0 # Recomendar productos de cestas similares for idx in similar_indices: sim_score = similarities[0][idx] total_similarity += sim_score products = cestas.iloc[idx]['Cestas'].split() for product in products: if product.strip() not in new_basket: # Evitar recomendar lo que ya está en la cesta if product.strip() in recommendations_count: recommendations_count[product.strip()] += sim_score else: recommendations_count[product.strip()] = sim_score # Calcular la probabilidad relativa de cada producto recomendado recommendations_with_prob = [] if total_similarity > 0: # Verificar que total_similarity no sea cero recommendations_with_prob = [(product, score / total_similarity) for product, score in recommendations_count.items()] else: print("No se encontraron similitudes suficientes para calcular probabilidades.") recommendations_with_prob.sort(key=lambda x: x[1], reverse=True) # Ordenar por puntuación # Crear un nuevo DataFrame para almacenar las recomendaciones con descripciones y probabilidades recommendations_df = pd.DataFrame(columns=['ARTICULO', 'DESCRIPCION', 'PROBABILIDAD']) # Agregar las recomendaciones al DataFrame usando pd.concat for product, prob in recommendations_with_prob: # Buscar la descripción en el DataFrame de productos description = productos.loc[productos['ARTICULO'] == product, 'DESCRIPCION'] if not description.empty: # Crear un nuevo DataFrame temporal para la recomendación temp_df = pd.DataFrame({ 'ARTICULO': [product], 'DESCRIPCION': [description.values[0]], # Obtener el primer valor encontrado 'PROBABILIDAD': [prob] }) # Concatenar el DataFrame temporal al DataFrame de recomendaciones recommendations_df = pd.concat([recommendations_df, temp_df], ignore_index=True) return recommendations_df # Comprobar si el cliente está en el CSV de fieles is_fiel = customer_code in fieles_df['Cliente'].astype(str).values if customer_code: if is_fiel: st.write(f"### Customer {customer_code} is a loyal customer.") option = st.selectbox("Select Recommendation Type", ["Select an option", "By Purchase History", "By Current Basket"]) if option == "By Purchase History": st.warning("Option not available... aún") elif option == "By Current Basket": st.write("Select the items and assign quantities for the basket:") # Mostrar lista de artículos disponibles available_articles = productos['ARTICULO'].unique() selected_articles = st.multiselect("Select Articles", available_articles) # Crear inputs para ingresar las cantidades de cada artículo seleccionado quantities = {} for article in selected_articles: quantities[article] = st.number_input(f"Quantity for {article}", min_value=0, step=1) if st.button("Calcular"): # Añadimos el botón "Calcular" # Crear una lista de artículos basada en la selección new_basket = [f"{article} x{quantities[article]}" for article in selected_articles if quantities[article] > 0] if new_basket: # Procesar la lista para recomendar recommendations_df = recomienda(new_basket) if not recommendations_df.empty: st.write("### Recommendations based on the current basket:") st.dataframe(recommendations_df) else: st.warning("No recommendations found for the provided basket.") else: st.warning("Please select at least one article and set its quantity.") else: st.write(f"### Customer {customer_code} is not a loyal customer.") st.write("Select items and assign quantities for the basket:") # Mostrar lista de artículos disponibles available_articles = productos['ARTICULO'].unique() selected_articles = st.multiselect("Select Articles", available_articles) # Crear inputs para ingresar las cantidades de cada artículo seleccionado quantities = {} for article in selected_articles: quantities[article] = st.number_input(f"Quantity for {article}", min_value=0, step=1) if st.button("Calcular"): # Añadimos el botón "Calcular" # Crear una lista de artículos basada en la selección new_basket = [f"{article} x{quantities[article]}" for article in selected_articles if quantities[article] > 0] if new_basket: # Procesar la lista para recomendar recommendations_df = recomienda(new_basket) if not recommendations_df.empty: st.write("### Recommendations based on the current basket:") st.dataframe(recommendations_df) else: st.warning("No recommendations found for the provided basket.") else: st.warning("Please select at least one article and set its quantity.")