app.py

import gradio as gr
import joblib
import pandas as pd

# Load pipeline components
scaler, pca, clf = joblib.load("tennis_model.pkl")

# All possible categorical values (must match training one-hot encoding)
outlook_options = ["Sunny", "Overcast", "Rain"]
temp_options = ["Hot", "Mild", "Cool"]
humidity_options = ["High", "Normal"]
wind_options = ["Weak", "Strong"]

def predict_play(outlook, temp, humidity, wind):
    # Build input row
    data = pd.DataFrame([[outlook, temp, humidity, wind]],
                        columns=["outlook", "temp", "humidity", "wind"])
    
    # One-hot encode to match training
    data_enc = pd.get_dummies(data)
    
    # Ensure all training columns exist
    for col in scaler.feature_names_in_:
        if col not in data_enc:
            data_enc[col] = 0
    
    data_enc = data_enc[scaler.feature_names_in_]  # reorder
    
    # Scale + PCA + Predict
    X_scaled = scaler.transform(data_enc)
    X_pca = pca.transform(X_scaled)
    pred = clf.predict(X_pca)[0]
    
    return f"Prediction: {pred}"

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("# 🎾 Play Tennis Predictor")
    outlook = gr.Dropdown(outlook_options, label="Outlook")
    temp = gr.Dropdown(temp_options, label="Temperature")
    humidity = gr.Dropdown(humidity_options, label="Humidity")
    wind = gr.Dropdown(wind_options, label="Wind")
    
    btn = gr.Button("Predict")
    output = gr.Textbox(label="Result")
    
    btn.click(fn=predict_play, inputs=[outlook, temp, humidity, wind], outputs=output)

demo.launch(server_name="0.0.0.0",server_port=7860,debug=True)