the situation is as following: If I have a numpy array of shape (64, 100, 300) and I want to transform it to (64, 1, 300) based on an array of indices of shape 64, what should I do? Say we have
a=np.random.randn(64, 100, 300)
indices = np.random.randint(low=0, high=100, size=64)
I currently do
a[:, indices, :]
which does not work. The returning array is of shape (64, 64, 300) since it tries to select the series of values from every batch.