I like Numba, but it's sometimes jarring writing C-like code in a language which almost explicitly advises against it. Once you get into the habit of "FOR LOOP

I like Numba, but it's sometimes jarring writing C-like code in a language which almost explicitly advises against it. Once you get into the habit of "FOR LOOP BAD" it's difficult to get out of! It's interesting seeing what the effect of Big-O is. Once you introduce a logarithmic solution, it doesn't really matter that you're using Python because it's asymptotically so much better than the n^2 brute force approach. Nearest neighbour is a particularly nice demonstration of this, because it's an algorithm which is frequently used in realtime and obviously you don't want a performance hit of 2 seconds for every query! Note Numpy has a builtin np.radians() which is a little cleaner than a one value look up table. In [2]: %timeit np.random.random((10000))*3.14 1000 loops, best of 3: 245 µs per loop In [3]: %timeit np.radians(np.random.random((10000))) 1000 loops, best of 3: 262 µs per loop