# app.py
import gradio as gr
import joblib
import pandas as pd
import numpy as np
import os
from sklearn.preprocessing import StandardScaler # Importado para evitar errores de unpickling

# --- CORRECCIÓN: Listas hardcodeadas para la interfaz (resuelve NameError) ---
UFC_LOCATIONS = [
    'Las Vegas, NV', 'Rio de Janeiro, Brazil', 'Abu Dhabi, UAE',
    'London, England', 'New York, NY', 'Outro' # Incluimos 'Outro' por si hay locations no vistas
]

# --- 1. Cargar objetos serializados ---
try:
    # Carga segura de artefactos
    model = joblib.load(os.path.join(os.path.dirname(__file__), 'model.pkl'))
    preprocessor = joblib.load(os.path.join(os.path.dirname(__file__), 'preprocessor.pkl'))
    model_columns = joblib.load(os.path.join(os.path.dirname(__file__), 'model_columns.pkl'))
except Exception as e:
    # Este error capturará el problema de compatibilidad de Scikit-learn
    print(f"Error al cargar artefactos: {e}")
    model = None
    preprocessor = None
    model_columns = []

# --- 2. Función de Predicción (Núcleo de la API) ---
def predict_ko_tko(F1_KD, F2_KD, F1_STR, F2_STR, F1_TD, F2_TD, F1_SUB, F2_SUB, Round, 
                   F1_acc, F2_acc, KD_diff, STR_diff, TD_diff, SUB_diff, Location, 
                   wc_B, wc_C, wc_F, wc_Fl, wc_H, wc_LH, wc_L, wc_M, wc_O, wc_SH, wc_W, 
                   wc_WB, wc_WF, wc_WFl, wc_WS):

    if not model or not preprocessor:
        # Devuelve un mensaje claro si la carga falló por la versión de Scikit-learn
        return "ERROR", "Fallo al cargar modelo. Revisa el log de versiones."

    # 1. Crear el DataFrame de entrada (31 columnas originales de X)
    input_data = pd.DataFrame({
        'Fighter_1_KD': [F1_KD], 'Fighter_2_KD': [F2_KD], 'Fighter_1_STR': [F1_STR], 'Fighter_2_STR': [F2_STR],
        'Fighter_1_TD': [F1_TD], 'Fighter_2_TD': [F2_TD], 'Fighter_1_SUB': [F1_SUB], 'Fighter_2_SUB': [F2_SUB],
        'Round': [Round], 'Fighter_1_accuracy': [F1_acc], 'Fighter_2_accuracy': [F2_acc],
        'KD_diff': [KD_diff], 'STR_diff': [STR_diff], 'TD_diff': [TD_diff], 'SUB_diff': [SUB_diff],
        'Location': [Location],
        # Columnas OHE ya existentes en el dataset (passthrough)
        'weight_class_Bantamweight': [wc_B], 'weight_class_Catch Weight': [wc_C], 'weight_class_Featherweight': [wc_F],
        'weight_class_Flyweight': [wc_Fl], 'weight_class_Heavyweight': [wc_H], 'weight_class_Light Heavyweight': [wc_LH],
        'weight_class_Lightweight': [wc_L], 'weight_class_Middleweight': [wc_M], 'weight_class_Open Weight': [wc_O],
        'weight_class_Super Heavyweight': [wc_SH], 'weight_class_Welterweight': [wc_W], 
        "weight_class_Women's Bantamweight": [wc_WB], "weight_class_Women's Featherweight": [wc_WF], 
        "weight_class_Women's Flyweight": [wc_WFl], "weight_class_Women's Strawweight": [wc_WS]
    })
    
    # 2. Preprocesamiento: Utilizar el ColumnTransformer ajustado.
    X_processed = preprocessor.transform(input_data)
    
    # 3. Convertir a DataFrame y asegurar el orden de las columnas
    X_final = pd.DataFrame(X_processed, columns=model_columns)
    
    # 4. Predicción
    prediction_proba = model.predict_proba(X_final)[0][1] # Probabilidad de 1 (KO/TKO)
    
    # 5. Formato de Salida
    prob_str = f"{prediction_proba*100:.2f}%"
    result_str = 'KO/TKO (¡Alta probabilidad de finalización!)' if prediction_proba > 0.5 else 'DECISIÓN/SUMISIÓN (Pelea a las tarjetas)'
        
    return prob_str, result_str

# --- 3. Creación de la Interfaz Gradio ---
inputs = [
    gr.Slider(0, 5, value=1, step=1, label="KD P1"), gr.Slider(0, 5, value=0, step=1, label="KD P2"),
    gr.Slider(0, 300, value=70, label="STR P1"), gr.Slider(0, 300, value=50, label="STR P2"),
    gr.Slider(0, 20, value=5, label="TD P1"), gr.Slider(0, 20, value=2, label="TD P2"),
    gr.Slider(0, 5, value=0, step=1, label="SUB P1"), gr.Slider(0, 5, value=0, step=1, label="SUB P2"),
    gr.Slider(1, 5, value=3, step=1, label="Ronda actual (Round)"),
    gr.Slider(0, 1, value=0.4, label="Precisión STR P1 (F1_acc)"), gr.Slider(0, 1, value=0.3, label="Precisión STR P2 (F2_acc)"),
    gr.Slider(-5, 5, value=1, label="Diferencia de KD"), gr.Slider(-300, 300, value=20, label="Diferencia de STR"),
    gr.Slider(-20, 20, value=3, label="Diferencia de TD"), gr.Slider(-5, 5, value=0, label="Diferencia de SUB"),
    gr.Dropdown(UFC_LOCATIONS, value='Las Vegas, NV', label="Ubicación"), # Usa la lista corregida
    gr.Checkbox(value=True, label="weight_class_Lightweight"), gr.Checkbox(value=False, label="weight_class_Bantamweight"), 
    gr.Checkbox(value=False, label="weight_class_Catch Weight"), gr.Checkbox(value=False, label="weight_class_Featherweight"),
    gr.Checkbox(value=False, label="weight_class_Flyweight"), gr.Checkbox(value=False, label="weight_class_Heavyweight"),
    gr.Checkbox(value=False, label="weight_class_Light Heavyweight"), gr.Checkbox(value=False, label="weight_class_Middleweight"),
    gr.Checkbox(value=False, label="weight_class_Open Weight"), gr.Checkbox(value=False, label="weight_class_Super Heavyweight"),
    gr.Checkbox(value=False, label="weight_class_Welterweight"), gr.Checkbox(value=False, label="weight_class_Women's Bantamweight"),
    gr.Checkbox(value=False, label="weight_class_Women's Featherweight"), gr.Checkbox(value=False, label="weight_class_Women's Flyweight"),
    gr.Checkbox(value=False, label="weight_class_Women's Strawweight")
]

outputs = [gr.Textbox(label="Probabilidad de KO/TKO (%)"), gr.Textbox(label="Resultado Más Probable")]

gr.Interface(
    fn=predict_ko_tko,
    inputs=inputs,
    outputs=outputs,
    title="🥊 Predictor de KO/TKO en Combates UFC (Despliegue ML)",
    description="Modelo Random Forest para predecir la finalización de un combate."
).launch(server_name="0.0.0.0", server_port=7860)