How to optimize MAPE code in Python?

问题:

I need to have a MAPE function, however I was not able to find it in standard packages ... Below, my implementation of this function.

def mape(actual, predict): 
    tmp, n = 0.0, 0
    for i in range(0, len(actual)):
        if actual[i] <> 0:
            tmp += math.fabs(actual[i]-predict[i])/actual[i]
            n += 1
    return (tmp/n)

I don't like it, it's super not optimal in terms of speed. How to rewrite the code to be more Pythonic way and boost the speed?

回答1:

Here's one vectorized approach with masking -

def mape_vectorized(a, b): 
    mask = a <> 0
    return (np.fabs(a[mask] - b[mask])/a[mask]).mean()

Probably a faster one with masking after division computation -

def mape_vectorized_v2(a, b): 
    mask = a <> 0
    return (np.fabs(a - b)/a)[mask].mean() 

Runtime test -

In [217]: a = np.random.randint(-10,10,(10000))
     ...: b = np.random.randint(-10,10,(10000))
     ...: 

In [218]: %timeit mape(a,b)
100 loops, best of 3: 11.7 ms per loop

In [219]: %timeit mape_vectorized(a,b)
1000 loops, best of 3: 273 µs per loop

In [220]: %timeit mape_vectorized_v2(a,b)
1000 loops, best of 3: 220 µs per loop