generate sequence by indices / one-hot encoding

Question

I have a sequence s = [4,3,1,0,5] and num_classes = 6 and I want to generate a Numpy matrix m of shape (len(s), num_classes)

Answer 1

Since you want a single 1 per row, you can fancy-index using arange(len(s)) along the first axis, and using s along the second:

s = [4,3,1,0,5]
n = len(s)
k = 6
m = np.zeros((n, k))
m[np.arange(n), s] = 1
m
=> 
array([[ 0.,  0.,  0.,  0.,  1.,  0.],
       [ 0.,  0.,  0.,  1.,  0.,  0.],
       [ 0.,  1.,  0.,  0.,  0.,  0.],
       [ 1.,  0.,  0.,  0.,  0.,  0.],
       [ 0.,  0.,  0.,  0.,  0.,  1.]])

m.nonzero()
=> (array([0, 1, 2, 3, 4]), array([4, 3, 1, 0, 5]))

This can be thought of as using index (0,4), then (1,3), then (2,1), (3,0), (4,5).