Python 对 numpy 数组的每 n 个元素求平均值
声明：本页面是StackOverFlow热门问题的中英对照翻译，遵循CC BY-SA 4.0协议，如果您需要使用它，必须同样遵循CC BY-SA许可，注明原文地址和作者信息，同时你必须将它归于原作者（不是我）：StackOverFlow
原文地址: http://stackoverflow.com/questions/15956309/
Warning: these are provided under cc-by-sa 4.0 license. You are free to use/share it, But you must attribute it to the original authors (not me):
StackOverFlow
Averaging over every n elements of a numpy array
提问by user1654183
I have a numpy array. I want to create a new array which is the average over every consecutive triplet of elements. So the new array will be a third of the size as the original.
我有一个 numpy 数组。我想创建一个新数组，它是每个连续三元组元素的平均值。所以新数组的大小将是原始数组的三分之一。
As an example:
举个例子：
np.array([1,2,3,1,2,3,1,2,3])
should return the array:
应该返回数组：
np.array([2,2,2])
Can anyone suggest an efficient way of doing this? I'm drawing blanks.
谁能提出一种有效的方法来做到这一点？我在画空白。
采纳答案by Jaime
If your array arrhas a length divisible by 3:
如果您的数组arr的长度可被 3 整除：
np.mean(arr.reshape(-1, 3), axis=1)
Reshaping to a higher dimensional array and then performing some form of reduce operation on one of the additional dimensions is a staple of numpy programming.
重构为更高维的数组，然后在其中一个附加维度上执行某种形式的归约操作是 numpy 编程的主要内容。
回答by L_W
For googlers looking for a simple generalisation for arrays with multiple dimensions: the function block_reducein the scikit-imagemodule (link to docs).
让Google寻找一个简单概括为具有多个尺寸的阵列：该函数block_reduce中scikit-image模块（链接到文档）。
It has a very simple interface to downsample arrays by applying a function such as numpy.mean, but can also use others (maximum, median, ...). The downsampling can be done by different factors for different axes by supplying a tuple with different sizes for the blocks. Here's an example with a 2D array; downsampling only axis 1 by 10 using the mean:
它有一个非常简单的接口，可以通过应用诸如 之类的函数来对数组进行下采样numpy.mean，但也可以使用其他函数（最大值、中值等）。通过为块提供具有不同大小的元组，可以通过不同轴的不同因素来完成下采样。这是一个二维数组的例子；使用平均值仅对轴 1 进行 10 次下采样：
import numpy as np
from skimage.measure import block_reduce
arr = np.stack((np.arange(1,20), np.arange(20,39)))
# array([[ 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]])
arr_reduced = block_reduce(arr, block_size=(1,5), func=np.mean, cval=np.mean(arr))
# array([[ 3. , 8. , 13. , 17.8],
# [22. , 27. , 32. , 33. ]])
As it was discussed in the comments to the other answer: if the array in the reduced dimension is not divisible by block size, padding values are provided by the argument cval(0 by default).
正如在另一个答案的评论中所讨论的那样：如果减少维度中的数组不能被块大小整除，则填充值由参数提供cval（默认为 0）。
