英文:
Conditional replacement of column in numpy array
问题
你可以使用NumPy的函数来实现这个灵活的功能。以下是一个示例代码，可以根据指定的轴来过滤数组：
import numpy as npdef filter_array(arr, axis=2):nonzero_counts = np.count_nonzero(arr, axis=axis)mask = nonzero_counts <= 1return arr * mask[..., np.newaxis]# 对列进行过滤filtered_columns = filter_array(arr, axis=2)print(filtered_columns)# 对行进行过滤filtered_rows = filter_array(arr, axis=1)print(filtered_rows)
这个函数可以根据传入的轴参数来过滤数组的行或列，使代码更加灵活和可复用。
英文:
I`m currently stuck on writing some script in numpy, which main goal is to be efficient (so, vectorization is mandatory).
Let`s assume 3-d array:
arr = [[[0, 0, 0, 0],[0, 0, 3, 4],[0, 0, 3, 0],[0, 2, 3, 0]],[[0, 0, 3, 0],[0, 0, 0, 0],[1, 0, 3, 0],[0, 0, 0, 0]],[[0, 2, 3, 4],[0, 0, 0, 0],[0, 0, 3, 4],[0, 0, 3, 0]],[[0, 0, 3, 4],[0, 0, 3, 4],[0, 0, 0, 0],[0, 0, 0, 0]]]
My goal is to set to dismiss every column which have more than one number other than zero. So, having above matrix the result should be something like:
filtered = [[[0, 0, 0, 0],[0, 0, 0, 4],[0, 0, 0, 0],[0, 2, 0, 0]],[[0, 0, 0, 0],[0, 0, 0, 0],[1, 0, 0, 0],[0, 0, 0, 0]],[[0, 2, 0, 0],[0, 0, 0, 0],[0, 0, 0, 0],[0, 0, 0, 0]],[[0, 0, 0, 0],[0, 0, 0, 0],[0, 0, 0, 0],[0, 0, 0, 0]]]
I`ve managed to work this around by set of np.count_nonzero, np.repeat and reshape:
indices = np.repeat(np.count_nonzero(a=arr, axis=1), repeats=4, axis=0).reshape(4, 4, 4)result = indices * a
Which produces good results but looks like missing the point (there is a lot of cryptic matrix shape manipulation only to slice array properly). Furthermore, I`d wish this function to be flexible enough to work out with other axes too (for rows e.g.), resulting:
rows_fil = [[[0, 0, 0, 0],[0, 0, 0, 0],[0, 0, 3, 0],[0, 0, 0, 0]],[[0, 0, 3, 0],[0, 0, 0, 0],[0, 0, 0, 0],[0, 0, 0, 0]],[[0, 0, 0, 0],[0, 0, 0, 0],[0, 0, 0, 0],[0, 0, 3, 0]],[[0, 0, 0, 0],[0, 0, 0, 0],[0, 0, 0, 0],[0, 0, 0, 0]]
Is there any "numpy" way to achieve such flexible function?
答案1
得分: 1
以下是已翻译的内容：
这里有一个涵盖通用轴参数的解决方案 -
def mask_nnzcount(a, axis):# a是输入数组mask = (a != 0).sum(axis=axis, keepdims=True) > 1return np.where(mask, 0, a)
关键在于 keepdims = True，它允许我们拥有一个通用的解决方案。
对于一个3D数组，对于列填充，使用 axis=1，对于行填充，使用 axis=2。
对于通用的ndarray，您可能想要使用 axis=-2 进行列填充，使用 axis=-1 进行行填充。
或者，我们还可以在最后一步使用元素级别的乘法来获得输出，即 a*(~mask)。或者获得一个反转的掩码，即说 inv_mask = (a != 0).sum(axis=axis, keepdims=True) <= 1，然后执行 a*inv_mask。
英文:
Here's a solution to cover a generic axis param -
def mask_nnzcount(a, axis):# a is input arraymask = (a!=0).sum(axis=axis, keepdims=True)>1return np.where(mask, 0, a)
The trick really is at keepdims = True which allows us to have a generic solution.
With a 3D array, for your column-fill, that's with axis=1 and for row-fill it's axis=2.
For a generic ndarray, you might want to use axis=-2 for column-fill and axis=-1 for row-fill.
Alternatively, we could also use element-wise multiplication instead at the last step to get the output with a*(~mask). Or get an inverted mask i.e. say inv_mask = (a!=0).sum(axis=axis, keepdims=True)<=1 and then do a*inv_mask.
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库，让每个人都能够通过互相帮助和分享经验来进步。
评论