Nicholas P. Rougier, From Python To Numpy1, 2017
Introduction
def randomwalkfaster(n=1000): from itertools import accumulate
steps = random.choices([-1,+1], k=n) return [0]+list(accumulate(steps))
walk = randomwalkfaster(1000)
accumulate([1,2,3,4,5]) --> 1 3 6 10 152
Without using loops and instead vectorizing the problem we get a 85% increase in performance.
>>> from tools import timeit >>> timeit(“randomwalkfaster(n=10000)”, globals()) 10 loops, best of 3: 2.21 msec per loop
Translating in numpy we get:
def randomwalkfastest(n=1000):
steps = np.random.choice([-1,+1], n) return np.cumsum(steps)
walk = randomwalkfastest(1000)
>>> from tools import timeit >>> timeit(“randomwalkfastest(n=10000)”, globals()) 1000 loops, best of 3: 14 usec per loop
Readability vs Speed
The tradeoff for the massive speedups using numpy is often the readabily of the code: comment your code!
- future-self will thank you