app.py

# app.py
# Gradio app for coffee yield prediction on Hugging Face Spaces

import gradio as gr
import pandas as pd
import joblib

# Load the model, scaler, and feature names
lasso_model = joblib.load('lasso_model.joblib')
scaler = joblib.load('scaler.joblib')
features = joblib.load('feature_names.joblib')

# Define the prediction function
def predict_yield(hectares, *feature_inputs):
    try:
        # Create a dictionary of inputs
        input_dict = {feature: value for feature, value in zip(features, feature_inputs)}
        
        # Convert to DataFrame
        input_df = pd.DataFrame([input_dict], columns=features)
        
        # Scale the input data
        input_scaled = scaler.transform(input_df)
        input_scaled = pd.DataFrame(input_scaled, columns=features)
        
        # Predict yield per hectare (kg/ha)
        yield_kg_ha = lasso_model.predict(input_scaled)[0]
        
        # Scale by hectares to get total yield (kg)
        total_yield_kg = yield_kg_ha * hectares
        
        return f"Predicted Total Yield: {total_yield_kg:.2f} kg for {hectares} hectares (Yield per hectare: {yield_kg_ha:.2f} kg/ha)"
    except Exception as e:
        return f"Error: {str(e)}"

# Create feature input fields dynamically based on feature names
inputs = [
    gr.Number(label="Total Hectares of Farmland", value=1.0, minimum=0.1, step=0.1)
] + [gr.Number(label=feature, value=0.0, step=0.01) for feature in features]

# Create the Gradio interface
demo = gr.Interface(
    fn=predict_yield,
    inputs=inputs,
    outputs=gr.Textbox(label="Prediction Result"),
    title="Coffee Yield Predictor for Sidama",
    description="Enter the total hectares of farmland and environmental/vegetation features to predict coffee yield. Features are based on annual means for Sidama, Ethiopia (2019–2023).",
    allow_flagging="never"
)

# Launch the app (Hugging Face will handle this automatically)
demo.launch(share=False)