np one hot encoding

1  labels = (np.arange(num_labels) == labels[:,None]).astype(np.float32)

1>>> a = np.array([1, 0, 3])
2>>> b = np.zeros((a.size, a.max()+1))
3>>> b[np.arange(a.size),a] = 1
4>>> b
5array([[ 0.,  1.,  0.,  0.],
6       [ 1.,  0.,  0.,  0.],
7       [ 0.,  0.,  0.,  1.]])