app.py

import os
import numpy as np
import pandas as pd
import gradio as gr
import tensorflow as tf
import joblib

# 1. CARGA DE MODELOS Y ARTEFACTOS
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
MODELS_DIR = os.path.join(BASE_DIR, "models")

def load_artifacts():
    try:
        model_bin = tf.keras.models.load_model(os.path.join(MODELS_DIR, "modelo_base.keras"))
        model_multi = tf.keras.models.load_model(os.path.join(MODELS_DIR, "modelo_transfer.keras"))
        scaler = joblib.load(os.path.join(MODELS_DIR, "scaler.joblib"))
        encoders = joblib.load(os.path.join(MODELS_DIR, "encoders.joblib"))
        return model_bin, model_multi, scaler, encoders
    except Exception as e:
        print(f"Error cargando modelos: {e}")
        return None, None, None, None

model_bin, model_multi, scaler, encoders = load_artifacts()

# Mapeo de categorías multiclase
MULTI_CLASSES = {0: "Normal", 1: "DoS", 2: "Probe", 3: "R2L", 4: "U2R"}

def predict_batch(file):
    if file is None: return None, "Cargue un archivo CSV."
    
    try:
        # Leer archivo (primeras 41 columnas)
        df_raw = pd.read_csv(file.name, header=None).iloc[:, :41]
        df_proc = df_raw.copy()

        # Preprocesamiento Batch
        if encoders:
            for col, pos in [('protocol_type', 1), ('service', 2), ('flag', 3)]:
                df_proc[pos] = encoders[col].transform(df_proc[pos].astype(str))

        X = df_proc.values.astype(float)
        if scaler:
            X = scaler.transform(X)

        # Predicciones
        preds_bin = model_bin.predict(X, verbose=0)
        preds_multi = model_multi.predict(X, verbose=0)

        results = []
        ataques_count = 0

        for i in range(len(X)):
            # Lógica Binaria (Umbral 0.5)
            # Nota: Si la probabilidad es > 0.5 es Ataque
            prob_bin = float(preds_bin[i][0])
            is_attack = prob_bin > 0.5
            
            label_bin = "🛑 Ataque" if is_attack else "✅ Normal"
            conf_bin = f"{prob_bin:.2%}" if is_attack else f"{(1-prob_bin):.2%}"
            
            if is_attack:
                ataques_count += 1

            # Lógica Multiclase
            idx_multi = np.argmax(preds_multi[i])
            label_multi = MULTI_CLASSES.get(idx_multi, "Otros")
            conf_multi = f"{np.max(preds_multi[i]):.2%}"

            results.append([i+1, label_bin, conf_bin, label_multi, conf_multi])

        # Resumen unificado
        total = len(results)
        normal_count = total - ataques_count
        summary = (f"📊 Análisis Completado\n"
                   f"-------------------\n"
                   f"Total registros: {total}\n"
                   f"✅ Tráfico Normal: {normal_count}\n"
                   f"🛑 Ataques: {ataques_count}")
        
        return results, summary

    except Exception as e:
        return None, f"Error en el proceso: {str(e)}"

# 3. INTERFAZ
with gr.Blocks(theme=gr.themes.Soft(), title="IDS Demo") as demo:
    gr.Markdown("# 🛡️ IDS con Redes Neuronales (NSL-KDD)")
    
    with gr.Row():
        file_input = gr.File(label="Subir test_heterogeneo.csv", file_types=[".csv"])
        run_btn = gr.Button("🚀 Analizar en Batch", variant="primary")
    
    with gr.Row():
        summary_out = gr.Textbox(label="Resumen del Análisis", lines=6)
        table_out = gr.DataFrame(
            headers=["#", "Clasif. Binaria", "Confianza", "Categoría", "Confianza"],
            label="Resultados Detallados"
        )

    run_btn.click(fn=predict_batch, inputs=file_input, outputs=[table_out, summary_out])

if __name__ == "__main__":
    demo.launch()