src/train.py

import tensorflow as tf
from tensorflow.keras.applications import MobileNetV2
from tensorflow.keras.layers import Dense, GlobalAveragePooling2D
from tensorflow.keras.models import Model
from tensorflow.keras.callbacks import ModelCheckpoint
from utils import load_dataset

# 1. Cargar datos
X, y = load_dataset("data")
print(f"✅ Datos cargados: {X.shape[0]} audios (formato {X.shape[1:]}")

# 2. Dividir en entrenamiento/validación
from sklearn.model_selection import train_test_split
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.2, random_state=42)

# 3. Crear modelo (MobileNetV2 + fine-tuning)
base_model = MobileNetV2(
    weights='imagenet',
    include_top=False,
    input_shape=(128, 128, 3)
)

# Congelar capas base (opcional para pocos datos)
for layer in base_model.layers:
    layer.trainable = False

# Añadir capas personalizadas
x = GlobalAveragePooling2D()(base_model.output)
x = Dense(128, activation='relu')(x)
predictions = Dense(1, activation='sigmoid')(x)  # Salida binaria

model = Model(inputs=base_model.input, outputs=predictions)

# 4. Compilar (énfasis en Recall)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy', tf.keras.metrics.Recall()]
)

# 5. Callbacks (guardar mejor modelo)
checkpoint = ModelCheckpoint(
    "model.h5",
    monitor='val_recall',  # Priorizar recall en validación
    mode='max',
    save_best_only=True,
    verbose=1
)

# 6. Entrenar
history = model.fit(
    X_train, y_train,
    epochs=15,
    batch_size=32,
    validation_data=(X_val, y_val),
    callbacks=[checkpoint],
    class_weight={0: 1, 1: 4}  # Ajustar si hay desbalance
)

print("✅ Entrenamiento completado. Modelo guardado como 'model.h5'")