Train a two-layer NumPy classifier with inverted dropout and report held-out accuracy after every epoch.
