Say I have the following labels and target values:

labels = np.array([[0, 0, 2],
                   [1, 1, 2],
                   [1, 1, 2]])

targets = np.array([9, 4, 7])

I want to generate a new array, where I replace value i in labels with targets[i]. Here is a possible solution:

n_labels = np.max(labels) + 1
out = np.empty(labels.shape)
for i in range(n_labels):
    mask = labels == i
    out[mask] = targets[i]

>>> out
np.array([[9, 9, 7],
          [4, 4, 7],
          [4, 4, 7]])

However, as the size of labels and targets grow, I see this solution as being inefficient. Each iteration, only a small number of values of out are being populated. However, each iteration, this approach computes the mask over all of labels (in the expression labels == i) and subsequently indexes over all of out (in the expression out[mask]). Is there a more efficient way to do this?