app.py

import gradio as gr
import requests
import pandas as pd
import os
from datasets import load_dataset

# FastAPI endpoint URL
API_URL = "https://molinari135-product-return-prediction-api.hf.space/predict/"

# Load the inventory dataset from Hugging Face
hf_token = os.getenv("inventory_data")
dataset = load_dataset("molinari135/armani-inventory", token=hf_token, data_files="inventory.tsv")
inventory = pd.DataFrame(dataset['train']).head(50)

# Gradio Interface function
def predict_return(selected_products, total_customer_purchases, total_customer_returns):
    # Input validation for returns (must be <= purchases)
    if total_customer_returns > total_customer_purchases:
        return "Error: Total returns cannot be greater than total purchases."

    # Prepare the request data
    models = []
    fabrics = []
    colours = []
    descriptions = []
    total_value = 0

    for selected_product in selected_products:
        # Split each selected product into model, fabric, and color
        model, fabric, color = selected_product.split("-")
        models.append(model)
        fabrics.append(fabric)
        colours.append(color)
        
        # Get the product details from the inventory
        product_details = inventory[(
            inventory['Item Brand Model'] == model) & 
            (inventory['Item Brand Fabric'] == fabric) & 
            (inventory['Item Brand Colour'] == color)
        ]
        
        if not product_details.empty:
            # Calculate the product value and add it to the total
            product_value = product_details['Net Sales (FA)'].values[0]
            total_value += product_value

            # Add description to the cart
            description = (
                f"Model: {model}, Fabric: {fabric}, Colour: {color}, "
                f"Product Type: {product_details['Product Type'].values[0]}, "
                f"Material: {product_details['Main Material'].values[0]}, "
                f"Sales Value: {product_value} USD"
            )
            descriptions.append(description)
        else:
            descriptions.append(f"{model}-{fabric}-{color}: Not Found")

    # Prepare the data to send to the API
    data = {
        "models": models,
        "fabrics": fabrics,
        "colours": colours,
        "total_customer_purchases": total_customer_purchases,
        "total_customer_returns": total_customer_returns
    }

    try:
        # Make the POST request to the FastAPI endpoint
        response = requests.post(API_URL, json=data)
        response.raise_for_status()  # Raise an error for bad responses

        # Get the predictions and return them
        result = response.json()
        predictions = result.get('predictions', [])

        if not predictions:
            return "Error: No predictions found."

        # Format the cart output
        cart_output = "\n".join(descriptions) + f"\nTotal Cart Value: {total_value} USD"

        # Format the prediction results
        formatted_result = "\n".join([f"Product: {pred['product']} \t Prediction: {pred['prediction']} \t Confidence: {pred['confidence']}%" for pred in predictions])

        return cart_output, formatted_result

    except requests.exceptions.RequestException as e:
        return f"Error: {str(e)}"

# Gradio interface elements
interface = gr.Interface(
    fn=predict_return,  # Funzione per la logica di predizione
    inputs=[
        gr.CheckboxGroup(
            choices=[f"{row['Item Brand Model']}-{row['Item Brand Fabric']}-{row['Item Brand Colour']}" 
                     for _, row in inventory.iterrows()], 
            label="Select Products"
        ),
        gr.Slider(0, 10, step=1, label="Total Customer Purchases", value=0),
        gr.Slider(0, 10, step=1, label="Total Customer Returns", value=0)
    ],
    outputs=[
        gr.Textbox(label="Cart Details"),  # Dettagli del carrello
        gr.Textbox(label="Prediction Results")  # Risultati della predizione
    ],
    live=True  # Interattività in tempo reale
)

# Lancio dell'interfaccia Gradio
interface.launch()