Importance Learning: New Methods To Efficiently Model Heterogeneous Catalysts With Quenched Disorder