Numpy用另一个数组的值总结一个数组 [英] Numpy summarize one array by values of another
问题描述
我正在尝试找到一种向量化方法来完成以下工作:
I am trying to find a vectorized way to accomplish the follwing:
说我有一个x和y值的数组.请注意,x值并不总是ints且可以为负:
Say I have an array of x and y values. Note that the x values are not always ints and CAN be negative:
import numpy as np
x = np.array([-1,-1,-1,3,2,2,2,5,4,4], dtype=float)
y = np.array([0,1,0,1,0,1,0,1,0,1])
我想按x数组的排序后的唯一值对y数组进行分组,并汇总每个y类的计数.因此,上面的示例将如下所示:
I want to group the y array by the sorted, unique values of the x array and summarize the counts for each y class. So the example above would look like this:
array([[ 2., 1.],
[ 2., 1.],
[ 0., 1.],
[ 1., 1.],
[ 0., 1.]])
第一列代表x的每个唯一值的"0"值计数,第二列代表x的每个唯一值的"1"值计数.
Where the first column represents the count of '0' values for each unique value of x and the second column represents the count of '1' values for each unique value of x.
我当前的实现如下:
x_sorted, y_sorted = x[x.argsort()], y[x.argsort()]
def collapse(x_sorted, y_sorted):
uniq_ids = np.unique(x_sorted, return_index=True)[1]
y_collapsed = np.zeros((len(uniq_ids), 2))
x_collapsed = x_sorted[uniq_ids]
for idx, y in enumerate(np.split(y_sorted, uniq_ids[1:])):
y_collapsed[idx,0] = (y == 0).sum()
y_collapsed[idx,1] = (y == 1).sum()
return (x_collapsed, y_collapsed)
collapse(x_sorted, y_sorted)
(array([-1, 2, 3, 4, 5]),
array([[ 2., 1.],
[ 2., 1.],
[ 0., 1.],
[ 1., 1.],
[ 0., 1.]]))
这似乎不是numpy的精髓,但是我希望这种操作可以使用一些矢量化方法.我正在尝试不使用熊猫来做到这一点.我知道该库具有非常方便的groupby操作.
This doesn't seem very much in the spirit of numpy, however, and I'm hoping some vectorized method exists for this kind of operation. I am trying to do this without resorting to pandas. I know that library has a very convenient groupby operation.
推荐答案
由于x
是float
.我会这样做:
In [136]:
np.array([(x[y==0]==np.unique(x)[..., np.newaxis]).sum(axis=1),
(x[y==1]==np.unique(x)[..., np.newaxis]).sum(axis=1)]).T
Out[136]:
array([[2, 1],
[2, 1],
[0, 1],
[1, 1],
[0, 1]])
速度:
In [152]:
%%timeit
ux=np.unique(x)[..., np.newaxis]
np.array([(x[y==0]==ux).sum(axis=1),
(x[y==1]==ux).sum(axis=1)]).T
10000 loops, best of 3: 92.7 µs per loop
解决方案@seikichi
Solution @seikichi
In [151]:
%%timeit
>>> x = np.array([1.1, 1.1, 1.1, 3.3, 2.2, 2.2, 2.2, 5.5, 4.4, 4.4])
>>> y = np.array([0, 1, 0, 1, 0, 1, 0, 1, 0, 1])
>>> r = np.r_[np.unique(x), np.inf]
>>> np.concatenate([[np.histogram(x[y == v], r)[0]] for v in sorted(set(y))]).T
1000 loops, best of 3: 388 µs per loop
对于更常见的情况,如@askewchan指出的,y
不仅仅是{0,1}
:
For more general cases when y
is not just {0,1}
, as @askewchan pointed out:
In [155]:
%%timeit
ux=np.unique(x)[..., np.newaxis]
uy=np.unique(y)
np.asanyarray([(x[y==v]==ux).sum(axis=1) for v in uy]).T
10000 loops, best of 3: 116 µs per loop
要进一步解释广播,请参见以下示例:
To explain the broadcasting further, see this example:
In [5]:
np.unique(a)
Out[5]:
array([ 0. , 0.2, 0.4, 0.5, 0.6, 1.1, 1.5, 1.6, 1.7, 2. ])
In [8]:
np.unique(a)[...,np.newaxis] #what [..., np.newaxis] will do:
Out[8]:
array([[ 0. ],
[ 0.2],
[ 0.4],
[ 0.5],
[ 0.6],
[ 1.1],
[ 1.5],
[ 1.6],
[ 1.7],
[ 2. ]])
In [10]:
(a==np.unique(a)[...,np.newaxis]).astype('int') #then we can boardcast (converted to int for readability)
Out[10]:
array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0],
[1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1],
[0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 1, 0]])
In [11]:
(a==np.unique(a)[...,np.newaxis]).sum(axis=1) #getting the count of unique value becomes summing among the 2nd axis
Out[11]:
array([1, 3, 1, 1, 2, 1, 1, 1, 1, 3])
这篇关于Numpy用另一个数组的值总结一个数组的文章就介绍到这了,希望我们推荐的答案对大家有所帮助,也希望大家多多支持IT屋!