零（行方向）“numpy”数组的元素小于给定向量的值

I have a numpy array, e.g.,

我有一个numpy数组,例如,

import numpy as np    
A = np.exp(np.random.randn(3,10))

i.e., the array

即阵列

array([[ 1.17164655,  1.39153953,  0.68628548,  0.1051013 ],
       [ 0.45604269,  2.21059251,  1.79624195,  0.37553947],
       [ 1.03063907,  0.28035114,  1.70371105,  3.66090236]])

and I compute the maximum of row as follows

我按如下方式计算行的最大值

np.max(A, axis=1)
array([ 1.39153953,  2.21059251,  3.66090236])

I want to zero the elements of A, whose values are less than a fraction of the maximum value of the corresponding row. For instance, for the above example, if we set this fraction to 0.9, I would like to zero the following elements:

我想将A的元素归零,其值小于相应行的最大值的一小部分。例如,对于上面的示例,如果我们将此分数设置为0.9,我想将以下元素归零:

1st row: Zero the elements that are less than 0.9 * maximum = 1.25238557

第1行:将小于0.9 *的元素归零= 1.25238557

2nd row: Zero the elements that are less than 0.9 * maximum = 1.98953326

第2行:将小于0.9 *的元素归零= 1.98953326

3rd row: Zero the elements that are less than 0.9 * maximum = 3.29481212

第3行:将小于0.9 *的元素归零= 3.29481212

I took a look at numpy's documentation, but I had no luck. I also tried

我看了一下numpy的文档,但我没有运气。我也试过了

A < np.max(A, axis=1)

which I'd expect to work, but it doesn't.

我希望它可以工作,但事实并非如此。

1 个解决方案

#1

Use the keepdims argument to keep a length-1 axis instead of removing the collapsed axis, so the axes line up with the original shape for broadcasting:

使用keepdims参数保持长度为1的轴而不是删除折叠的轴,因此轴与原始形状对齐以进行广播:

A[A < 0.9*np.amax(A, axis=1, keepdims=True)] = 0