Apply an array of functions to another array

Question

I have an numpy array of functions and I'd like to apply to an array of floats using numpy's brodcasting rules. Aside from using a for loop, is there any standard way of achive this?

funcs = np.array([f1, f2, f3]) # each f is of type f(x: float) -> float
data = np.array([[1,2,3],
                 [4,5,6]])

# Expected result
np.array([[f1(1),f2(2),f3(3)],
          [f1(4),f2(5),f3(6)]])

Edit

Below there is a better example of the expected behaviour. It is important to be able to broadcast, as like summing two arrays:

>>> np.array([0,1]) + 1
np.array([0+1, 1+1])

>>> np.array([0,1]) + np.array([1,2])
np.array([0+1, 1+2])

>>> np.array([0,1]) + np.array([[1,2], [3,4]])
np.array([[0+1, 1+2], [0+3, 1+4]])

## Expected behaviour (Imagine % means function application operator)
>>> np.array([0,1]) % f
np.array([f(0), f(1)])

>>> np.array([0,1]) % np.array([f1,f2])
np.array([f1(0), f2(1)])

>>> np.array([0,1]) % np.array([[f1,f2], [f3,f4]])
np.array([[f1(0), f2(1)], [f3(0), f4(1)]])

@NilsWerner Yes. Those functions can accept the whole axis for example — lsmor
– lsmor, Commented Feb 28, 2020 at 9:39

hpaulj · Accepted Answer · 2020-02-28 16:43:33Z

1

Use frompyfunc

f = np.frompyfunc(lambda fx,arg: fx(arg), 2, 1)
res = f(fns, arg)

Result should be object dtype, with full broadcasting.

In [416]: f = np.frompyfunc(lambda fn,x:fn(x),2,1)                                             
In [420]: f1=lambda x:x+1 
     ...: f2=lambda x: x*3                                                                     
In [421]: f([[f1],[f2]], np.arange(4))                                                         
Out[421]: 
array([[1, 2, 3, 4],
       [0, 3, 6, 9]], dtype=object)

edited Feb 28, 2020 at 16:43

answered Feb 28, 2020 at 12:19

hpaulj

233k14 gold badges260 silver badges392 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

Jan · Accepted Answer · 2020-02-28 09:39:08Z

0

Using a comprehension you could go for:

import numpy as np

def f1(input):
    return input + 1

def f2(input):
    return input + 2

def f3(input):
    return input + 3

funcs = np.array([f1, f2, f3])
data = np.array([[1,2,3],
                 [4,5,6]])

result = np.array(
    [[fun(item) for fun, item in zip(funcs, lst)]
    for lst in data])

print(result)

Which yields

[[2 4 6]
 [5 7 9]]

answered Feb 28, 2020 at 9:39

Jan

43.3k11 gold badges57 silver badges87 bronze badges

1 Comment

lsmor Over a year ago

I've tried this, but it doesn't broadcast. Let me edit the question so it is a little more clear

Serge Ballesta · Accepted Answer · 2020-02-28 10:42:05Z

0

If the functions can accept a numpy array, I would use 2 transpositions:

resul = np.transpose(np.array([f[i](x[i]) for i, x in enumerate(np.transpose(data)]))

answered Feb 28, 2020 at 10:42

Serge Ballesta

150k13 gold badges137 silver badges267 bronze badges

Collectives™ on Stack Overflow

Apply an array of functions to another array

Edit

3 Answers 3

Comments

1 Comment

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

Edit

3 Answers 3

Comments

1 Comment

Comments

Your Answer

Sign up or log in

Post as a guest

Related