Weird test accuracy for VGG-like training

Deep Learning
18

One more thing, you should put replace=False in the choice function, so that same element is not repeated in the mask.

mask = np.random.choice(length, length, replace=False)

–