Sampling from a set according to unnormalized log-probabilities in NumPy

Question

I have a 1-D np.ndarray filled with unnormalized log-probabilities that define a categorical distribution. I would like to sample an integer index from this distrib