使用TxK列索引数组从TxN numpy数组中选择TxK numpy数组 [英] pick TxK numpy array from TxN numpy array using TxK column index array
问题描述
这是间接索引问题。
可以使用列表解析来解决。
It can be solved with a list comprehension.
问题是，是否，或者如何在numpy内解决它，
The question is whether, or, how to solve it within numpy,
当
data.shape 是（T，N）
和
c .shape 是（T，K）
以及<$ c的每个元素$ c> c 是一个介于0和N-1之间的 int ，即
数据中的列号。
and each element of c is an int between 0 and N-1 inclusive, that is,
each element of c is intended to refer to a column number from data.
目标是获得 out 其中
out.shape = (T,K)
并且 i in 0 ..（T-1）
行 out [i] = [data [ i，c [i，0]]，...，data [i，c [i，K-1]]]
具体示例：
data = np.array([\
[ 0, 1, 2],\
[ 3, 4, 5],\
[ 6, 7, 8],\
[ 9, 10, 11],\
[12, 13, 14]])
c = np.array([
[0, 2],\
[1, 2],\
[0, 0],\
[1, 1],\
[2, 2]])
out should be out = [[0, 2], [4, 5], [6, 6], [10, 10], [14, 14]]
输出的第一行是[0,2]，因为选择的列由c的第0行给出，它们是0和2，第0列和第2列的数据[0]是0和2.
The first row of out is [0,2] because the columns chosen are given by c's row 0, they are 0 and 2, and data[0] at columns 0 and 2 are 0 and 2.
第二行输出是[4,5]因为列选择由c的第1行给出，它们是1和2，第1列和第2列的数据[1]是4和5.
The second row of out is [4,5] because the columns chosen are given by c's row 1, they are 1 and 2, and data[1] at columns 1 and 2 is 4 and 5.
Numpy花式索引不是似乎以一种显而易见的方式解决这个问题，因为用c索引数据（例如 data [c] ， np.take（data，c，axis = 1））总是生成一个三维数组。
Numpy fancy indexing doesn't seem to solve this in an obvious way because indexing data with c (e.g. data[c], np.take(data,c,axis=1) ) always produces a 3 dimensional array.
列表理解可以解决它：
out = [[data [rowidx] ，i1]，数据[rowidx，i2]]用于（rowidx，（i1，i2））枚举（c）]
如果K是2我认为这是勉强的。如果K是可变的，那就不太好了。
if K is 2 I suppose this is marginally OK. If K is variable, this is not so good.
必须为每个值K重写列表推导，因为它将从数据中选出的列展开每行 c 。它也违反了DRY。
The list comprehension has to be rewritten for each value K, because it unrolls the columns picked out of data by each row of c. It also violates DRY.
是否存在完全基于 numpy 的解决方案？
Is there a solution based entirely in numpy?
推荐答案
你可以用 np.choose ：
In [1]: %cpaste
Pasting code; enter '--' alone on the line to stop or use Ctrl-D.
data = np.array([\
[ 0, 1, 2],\
[ 3, 4, 5],\
[ 6, 7, 8],\
[ 9, 10, 11],\
[12, 13, 14]])
c = np.array([
[0, 2],\
[1, 2],\
[0, 0],\
[1, 1],\
[2, 2]])
--
In [2]: np.choose(c, data.T[:,:,np.newaxis])
Out[2]:
array([[ 0, 2],
[ 4, 5],
[ 6, 6],
[10, 10],
[14, 14]])
这篇关于使用TxK列索引数组从TxN numpy数组中选择TxK numpy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！
