[英]Compute matrix of pairwise angles between two arrays of points
I have two vectors of points, x
and y
, shaped (n, p)
and (m, p)
respectively. 我有两个点矢量,
x
和y
,分别为(n, p)
和(m, p)
。 As an example: 举个例子:
x = np.array([[ 0. , -0.16341, 0.98656],
[-0.05937, -0.25205, 0.96589],
[ 0.05937, -0.25205, 0.96589],
[-0.11608, -0.33488, 0.93508],
[ 0. , -0.33416, 0.94252]])
y = np.array([[ 0. , -0.36836, 0.92968],
[-0.12103, -0.54558, 0.82928],
[ 0.12103, -0.54558, 0.82928]])
I want to compute an (n, m)
-sized matrix that contains the angles between the two points, a la this question. 我想计算一个
(n, m)
大小的矩阵,它包含两个点之间的角度, 这个问题。 That is, a vectorized version of: 也就是说,矢量化版本:
theta = np.array(
[ np.arccos(np.dot(i, j) / (la.norm(i) * la.norm(j)))
for i in x for j in y ]
).reshape((n, m))
Note: n
and m
can be of the order of ~10000 each. 注意:
n
和m
各为~10000。
There are multiple ways to do this: 有多种方法可以做到这一点:
import numpy.linalg as la
from scipy.spatial import distance as dist
# Manually
def method0(x, y):
dotprod_mat = np.dot(x, y.T)
costheta = dotprod_mat / la.norm(x, axis=1)[:, np.newaxis]
costheta /= la.norm(y, axis=1)
return np.arccos(costheta)
# Using einsum
def method1(x, y):
dotprod_mat = np.einsum('ij,kj->ik', x, y)
costheta = dotprod_mat / la.norm(x, axis=1)[:, np.newaxis]
costheta /= la.norm(y, axis=1)
return np.arccos(costheta)
# Using scipy.spatial.cdist (one-liner)
def method2(x, y):
costheta = 1 - dist.cdist(x, y, 'cosine')
return np.arccos(costheta)
# Realize that your arrays `x` and `y` are already normalized, meaning you can
# optimize method1 even more
def method3(x, y):
costheta = np.einsum('ij,kj->ik', x, y) # Directly gives costheta, since
# ||x|| = ||y|| = 1
return np.arccos(costheta)
Timing results for (n, m) = (1212, 252): (n,m)=(1212,252)的定时结果:
>>> %timeit theta = method0(x, y)
100 loops, best of 3: 11.1 ms per loop
>>> %timeit theta = method1(x, y)
100 loops, best of 3: 10.8 ms per loop
>>> %timeit theta = method2(x, y)
100 loops, best of 3: 12.3 ms per loop
>>> %timeit theta = method3(x, y)
100 loops, best of 3: 9.42 ms per loop
The difference in timing reduces as the number of elements increases. 随着元件数量的增加,时序差异减小。 For (n, m) = (6252, 1212):
对于(n,m)=(6252,1212):
>>> %timeit -n10 theta = method0(x, y)
10 loops, best of 3: 365 ms per loop
>>> %timeit -n10 theta = method1(x, y)
10 loops, best of 3: 358 ms per loop
>>> %timeit -n10 theta = method2(x, y)
10 loops, best of 3: 384 ms per loop
>>> %timeit -n10 theta = method3(x, y)
10 loops, best of 3: 314 ms per loop
However, if you leave out the np.arccos
step, ie, suppose you could manage with just costheta
, and didn't need theta
itself, then: 但是,如果你忽略了
np.arccos
步骤,即假设你只能用costheta
管理,并且不需要 theta
本身,那么:
>>> %timeit costheta = np.einsum('ij,kj->ik', x, y)
10 loops, best of 3: 61.3 ms per loop
>>> %timeit costheta = 1 - dist.cdist(x, y, 'cosine')
10 loops, best of 3: 124 ms per loop
>>> %timeit costheta = dist.cdist(x, y, 'cosine')
10 loops, best of 3: 112 ms per loop
This is for the case of (6252, 1212). 这是针对(6252,1212)的情况。 So actually
np.arccos
is taking up 80% of the time. 所以实际上
np.arccos
占80%的时间。 In this case I find that np.einsum
is much faster than dist.cdist
. 在这种情况下,我发现
np.einsum
比快得多 dist.cdist
。 So you definitely want to be using einsum
. 所以你肯定想要使用
einsum
。
Summary: Results for theta
are largely similar, but np.einsum
is fastest for me, especially when you're not extraneously computing the norms. 总结:
theta
结果大致相似,但np.einsum
对我来说最快,特别是当你没有无关地计算规范时。 Try to avoid computing theta
and working with just costheta
. 尽量避免计算
theta
并使用costheta
。
Note: An important point I didn't mention is that finiteness of floating-point precision can cause np.arccos
to give nan
values. 注意:我没有提到的一个重点是浮点精度的有限性会导致
np.arccos
给出nan
值。 method[0:3]
worked for values of x
and y
that hadn't been properly normalized, naturally. method[0:3]
自然地适用于未正确归一化的x
和y
值。 But method3
gave a few nan
s. 但
method3
给了几个nan
秒。 I fixed this with pre-normalization, which naturally destroys any gain in using method3
, unless you need to do this computation many many times for a small set of pre-normalized matrices (for whatever reason). 我用预标准化来修复它,这自然会破坏使用
method3
任何增益,除非你需要对一method3
预标准化矩阵进行多次计算(无论出于何种原因)。
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.