[Python] Why does numpy indexing with a colon vs. range result in weird behavior?

Could someone explain the following behavior?

test = np.random.rand(1000, 10)
test.shape
(1000, 10)

idxs = np.random.randint(0, 10, 1000)
idxs.shape
(1000,)

test[:, idxs].shape
(1000, 1000)

test[range(test.shape[0]), idxs].shape
(1000,)


I would expect these to be equivalent. What am I missing?

Continue reading...