In numpy is there a fast way of calculating the mean across multiple axis? I am calculating the mean on all but the 0 axis of an n-dimensional array.
I am currently doing this;
for i in range(d.ndim - 1):
d = d.mean(axis=1)
I'm wondering if there is a solution that doesn't use a python loop.
In numpy 1.7 you can give multiple axis to np.mean:
d.mean(axis=tuple(range(1, d.ndim)))
I am guessing this will perform similarly to the other proposed solutions, unless reshaping the array to flatten all dimensions triggers a copy of the data, in which case this should be much faster. So this is probably going to give a more consistent performance.
np.mean, that rely on np.add. It wasn't added to the docs of mean and other similar functions, although there is an open PR to fix that. But you can take a look at the docs of np.sum which were updated and have the relevant change tagged as New in version 1.7.0.list of integers, rather than a tuple, inspired by your example that uses range (which in Python 2.7 returns a list). In my opinion, the numpy functions should be indifferent to the type, supporting any iterable (list, tuple, range...). Until then, you should change your example so it actually works. =)My approach would be to reshape the array to flatten all of the higher dimensions and then run the mean on axis 1. Is this what your looking for?
In [14]: x = np.array([[[1,2],[3,4]],[[5,6],[7,8]]])
In [16]: x.reshape((x.shape[0], -1)).mean(axis=1)
Out[16]: array([ 2.5, 6.5])
(step 2 just calculates the product of the lengths of the higher dims)
You can also use numpy.apply_over_axes:
import numpy as np
x = np.array([[[1,2],[3,4]],[[5,6],[7,8]]])
y = np.apply_over_axes(np.mean, x, (1, 2))
y = array([[[ 2.5]],[[ 6.5]]])
Following on from the suggestion of @dsg101, is this the sort of thing you want?
>>> import numpy as np
>>> d=np.reshape(np.arange(5*4*3),[5,4,3])
>>> d
array([[[ 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]]])
>>> np.mean(np.reshape(d,[d.shape[0],np.product(d.shape[1:])]),axis=1)
array([ 5.5, 17.5, 29.5, 41.5, 53.5])
