If I train a Keras model (e.g. Conv1D) on a masked numpy array, does the Keras model actually treat the masked values like none/NaN values? Or does the Keras model ignore th
