[英]iterate over slices of an ndarray
Say I have a 3D numpy.array, eg with dimensions xyz, is there a way to iterate over slices along a particular axis?假设我有一个 3D numpy.array,例如尺寸为 xyz,有没有办法沿特定轴迭代切片? Something like:
就像是:
for layer in data.slices(dim=2):
# do something with layer
Edit: To clarify, the example is a dim=3 array, ie shape=(len_x, len_y, len_z).编辑:为了澄清,这个例子是一个dim = 3数组,即shape =(len_x,len_y,len_z)。 Elazar and equivalently kamjagin's solutions work, but aren't that general - you have to construct the
[:, :, i]
by hand, which means you need to know the dimensions, and the code isn't general enough to handle arrays of arbitrary dimensions. Elazar 和等效 kamjagin 的解决方案有效,但不是那么通用 - 您必须手动构建
[:, :, i]
,这意味着您需要知道尺寸,并且代码不够通用,无法处理数组任意维度。 You can fill missing dimension by using something like [..., :]
, but again you still have to construct this yourself.您可以使用
[..., :]
之类的东西来填充缺失的维度,但是您仍然必须自己构建它。
Sorry, should have been clearer, the example was a bit too simple!对不起,应该更清楚,例子有点太简单了!
Iterating over the first dimension is very easy, see below. 迭代第一维很容易,见下文。 To iterate over the others, roll that dimension to the front and do the same:
要迭代其他维度,请将该维度滚动到前面并执行相同操作:
>>> data = np.arange(24).reshape(2, 3, 4)
>>> for dim_0_slice in data: # the first dimension is easy
... print dim_0_slice
...
[[ 0 1 2 3]
[ 4 5 6 7]
[ 8 9 10 11]]
[[12 13 14 15]
[16 17 18 19]
[20 21 22 23]]
>>> for dim_1_slice in np.rollaxis(data, 1): # for the others, roll it to the front
... print dim_1_slice
...
[[ 0 1 2 3]
[12 13 14 15]]
[[ 4 5 6 7]
[16 17 18 19]]
[[ 8 9 10 11]
[20 21 22 23]]
>>> for dim_2_slice in np.rollaxis(data, 2):
... print dim_2_slice
...
[[ 0 4 8]
[12 16 20]]
[[ 1 5 9]
[13 17 21]]
[[ 2 6 10]
[14 18 22]]
[[ 3 7 11]
[15 19 23]]
EDIT Some timings, to compare different methods for largish arrays: 编辑一些时间,比较大型阵列的不同方法:
In [7]: a = np.arange(200*100*300).reshape(200, 100, 300)
In [8]: %timeit for j in xrange(100): a[:, j]
10000 loops, best of 3: 60.2 us per loop
In [9]: %timeit for j in xrange(100): a[:, j, :]
10000 loops, best of 3: 82.8 us per loop
In [10]: %timeit for j in np.rollaxis(a, 1): j
10000 loops, best of 3: 28.2 us per loop
In [11]: %timeit for j in np.swapaxes(a, 0, 1): j
10000 loops, best of 3: 26.7 us per loop
This could probably be solved more elegantly than this, but one way of doing it if you know dim beforehand(eg 2) is: 这可能比这更优雅地解决,但如果你事先知道昏暗(例如2),那么这样做的一种方法是:
for i in range(data.shape[dim]):
layer = data[:,:,i]
or if dim=0 或者如果dim = 0
for i in range(data.shape[dim]):
layer = data[i,:,:]
etc. 等等
something like that? 这样的事情?
>>> data = np.array([[1,2,3],[4,5,6],[7,8,9]])
>>> for layer in [data[:,i] for i in range(3)]:
... print layer
...
[1 4 7]
[2 5 8]
[3 6 9]
Correct me if I am wrong but it seems to me that your 3D array is gonna look like : 如果我错了,请纠正我,但在我看来,你的3D阵列看起来像是:
>>> my_array.shape
(3,N)
Where N is the size of your array. 其中N是数组的大小。 So if you want to iterate over one dimension, you can just do :
因此,如果您想迭代一个维度,您可以这样做:
>>> for item in my_array[1,:]:
This will iterate on the second dimension. 这将迭代第二个维度。
I think the original question was ambiguous, and so is the title:我认为最初的问题是模棱两可的,标题也是如此:
x[k,:,:,...]
for all k yields as many items as x.shape[0]
, so I would call this iterating over a dimension ;x[k,:,:,...]
会产生与x.shape[0]
一样多的项目,因此我将其称为对维度的迭代;x[:,i,j,k,...]
for all i,j,k.x[:,i,j,k,...]
上对所有 i,j,k。 For example, iterating over all columns in an ndarray. Although that's not what the OP asked after clarification, people looking for a solution to the second case above might find the following code useful:尽管这不是 OP 在澄清后所要求的,但寻找上述第二种情况的解决方案的人可能会发现以下代码很有用:
from itertools import product
def iterslice(x,axis=0):
sub = [ range(s) for s in x.shape ]
sub[axis] = (slice(None),)
for p in product(*sub):
yield x[p]
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.