I have a Python function, f, which takes a pair of numbers and return a calculation result on them, say, x+y
def f(x,y):
return x+y
How can I vectorize f, so for given vectors X=(x1,...,xn) and Y=(y1,...,yn)
f_vectorized(X,Y) returns the array of f(x1,y1), f(x2,y2)...?
EDIT
Above, x+y is an example, but the actual calculation is more complex, so by regarding x and y as numpy vector does not necessarily work out-of-box.
What about using numpy as input?
import numpy as np
def f(x,y):
return x+y
a = np.array([0,1,2,3])
b = np.array([1,2,3,4])
In [430]: f(a,b)
Out[430]: array([1, 3, 5, 7])
EDIT
For more complex function you could use list comprehension and zip:
In [451]: [f(*par) for par in zip(a, b)]
Out[451]: [1, 3, 5, 7]
EDIT2
Or you could use np.vectorize as you mentioned in the comment:
f_vec = np.vectorize(f)
In [470]: f_vec([0, 1, 2, 3], [1, 2, 3, 4])
Out[470]: array([1, 3, 5, 7])
Perfomance:
In [471]: %timeit f_vec([0, 1, 2, 3], [1, 2, 3, 4])
10000 loops, best of 3: 38.3 µs per loop
In [472]: %timeit [f(*par) for par in zip([0, 1, 2, 3], [1, 2, 3, 4])]
1000000 loops, best of 3: 1.8 µs per loop
In [476]: %timeit list(map(f, [0, 1, 2, 3], [1, 2, 3, 4]))
1000000 loops, best of 3: 1.51 µs per loop
So if you are interested in performance you should use zip and list comprehension or map solution suggested by @tglaria
list(map(...)) when map(...) already returns a list?
map return iterator and if you'd like to convert it to list you need to pass the result to list functionWell, isn't map supposed to be used for this?
def f(x,y):
return (x,y, x+y)
print map(f,[0,1,2], [-1,0,-2])
>[(0, -1, -1), (1, 0, 1), (2, -2, 0)]
np.vectorizewill sacrifice all the performance benefits of NumPy, which is likely to be around a 1000x slowdown. Unless your code involves loops that fundamentally need to execute for a number of times that depends on the input, something like this should be fairly straightforward to vectorize.