用掩码平均的numpy数组

Question

I'm looking to do some basic clustering on a boolean numpy array and I'm basically just trying to do 2d averaging with a mask, but I feel like there must be a better solution than what I've got, since it's slow and inelegant:

def grab_window(location, array, window=(3,3)):
    minimums = [min(0, i-a) for i, a in zip(location, window)]
    maximums = [(i + a) for i, a in zip(location, window)]
    answer = array
    for i, _ in enumerate(location):
        answer = answer[slice(minimums[i],maximums[i])]
    return answer

And then I basically just iterate through the original array, multiplying each window by a kernel, and returning the mean of the modified window.

It seems like there must be a filter or something similar that would have the same effect, but I haven't been able to find one thus far.

edit: location is a tuple of a form similar to window.

For instance, if we were to do the simplest version of this, with a uniform 1-ply mask I would be looking for something along these lines:

import numpy as np
test = np.arange(0,24).reshape(6, 4)
footprint = [
[1,1,1],
[1,0,1],
[1,1,1]
]
some_function(test, footprint)
array([[ 1,  2,  3,  4],
   [ 4,  5,  6,  6],
   [ 8,  9, 10, 10],
   [12, 13, 14, 14],
   [16, 17, 18, 18],
   [18, 19, 20, 21]])

Answer 1

Turns out scipy totally has a function that already does this. generic_filter actually does exactly this in a much more stable way as mentioned in How to apply ndimage.generic_filter()

Example:

def some_avg(values):
    return values.mean()

footprint = np.array([
    [1,1,1],
    [1,0,1],
    [1,1,1]
])

test = test = np.arange(0,24).reshape(6, 4)

scipy.ndimage.filters.generic_filter(test, some_avg, footprint=footprint)

array([[ 1,  2,  3,  4],
   [ 4,  5,  6,  6],
   [ 8,  9, 10, 10],
   [12, 13, 14, 14],
   [16, 17, 18, 18],
   [18, 19, 20, 21]])