Geek Answers Handbook

Numpy 2D Array Mapping Functions

I have a function foothat takes an NxM numpy array as an argument and returns a scalar value. I have a numpx AxNxM array datathat I would like to map over footo give me a numpy resulting array of length A.

Now I do this:

result = numpy.array([foo(x) for x in data])

This works, but it looks like I'm not using numpy magic (and speed). Is there a better way?

I looked through numpy.vectorizeand numpy.apply_along_axisbut does not work for the function of 2D arrays.

EDIT: I am doing enhanced regression on 24x24 images, so my AxNxM is something like 1000x24x24. What I have mentioned fooabove applies a haradike function to the patch (therefore, do not calculate too intensively).

+5
source share
2 answers

If NxM is large (say, 100), then the cost of iterating over A will be amortized basically.

Say the array is 1000 X 100 X 100.

The iteration is O (1000), but the cumulative cost of the internal function is O (1000 X 100 X 100) - 10,000 times slower. (Note, my terminology is a little fascinating, but I know what I'm talking about)

I'm not sure, but you can try the following:

result = numpy.empty(data.shape[0])
for i in range(len(data)):
    result[i] = foo(data[i])

You would save a lot of memory when creating the list ... but the overhead of the loop would be greater.

Or you could write a parallel version of the loop and split it into several processes. This can be much faster, depending on how intense it is foo(since this would compensate for data processing).

+3

, 3D- 2D- foo , 1D , , foo. ( trace foo):

from numpy import *

def apply2d_along_first(func2d, arr3d):
    a, n, m = arr3d.shape
    def func1d(arr1d):
        return func2d(arr1d.reshape((n,m)))
    arr2d = arr3d.reshape((a,n*m))
    return apply_along_axis(func1d, -1, arr2d)

A, N, M = 3, 4, 5
data = arange(A*N*M).reshape((A,N,M))

print data
print apply2d_along_first(trace, data)

:

[[[ 0  1  2  3  4]
  [ 5  6  7  8  9]
  [10 11 12 13 14]
  [15 16 17 18 19]]

 [[20 21 22 23 24]
  [25 26 27 28 29]
  [30 31 32 33 34]
  [35 36 37 38 39]]

 [[40 41 42 43 44]
  [45 46 47 48 49]
  [50 51 52 53 54]
  [55 56 57 58 59]]]
[ 36 116 196]
+1

More articles:


All Articles