numpy数组转换中的性能改进 [英] Performance improvement in numpy array transformation

问题描述

给出三个 numpy 一维数组，我想按如下方式转换它们:

Given three numpy 1D arrays, I want to transform them as follows:

import numpy as np

Xd = np.asarray([0, 0,   1,   1,   0.5])
Yd = np.asarray([0, 0,   0,   2.5, 2.5])
Zd = np.asarray([0, 1.5, 1.5, 1.5, 1.5])

points = np.stack([Xd, Yd, Zd], axis=1).reshape(-1, 1, 3)
segments = np.concatenate([points[:-1], points[1:]], axis = 1)    

print(segments.shape)
print(segments)

输出:

(4, 2, 3)
[[[0.  0.  0. ]
  [0.  0.  1.5]]

 [[0.  0.  1.5]
  [1.  0.  1.5]]

 [[1.  0.  1.5]
  [1.  2.5 1.5]]

 [[1.  2.5 1.5]
  [0.5 2.5 1.5]]]

有没有办法提高这种转换的性能?

Is there a way to improve the performance of this transformation?

背景

此转换对于将 matplotlib 中的 XYZ 坐标与数千个坐标或插值数据以获得更好的分辨率，优化方法是必要的.

This transformation is necessary to use the XYZ coordinates in matplotlib with Line3DCollection. So far, I have only seen variations of the above code but with thousands of coordinates or interpolated data for better resolution, an optimized approach is necessary.

摘要

由于 @Mercury ，可以得出结论，对于较短的数组(长度小于1k)， @Miguel的答案效果更好，但当数组变长时，@mathfux 的方法可以更好地扩展.

Thanks to @Mercury, it can be concluded that for shorter arrays (<1k in length) the answer by @Miguel performs better but the approach by @mathfux scales way better when the arrays get longer.

推荐答案

您似乎正在尝试在二维数组中滚动形状为 (2, 3) 的窗口.这类似于图像卷积，可以非常有效地使用 np.lib.stride_tricks 进行卷积.方式.

It seems like you're trying to roll a window of shape (2, 3) in a 2D array. This is similar to convolution of image which can be done with np.lib.stride_tricks in a very efficient way.

a = np.transpose([Xd, Yd, Zd])
window = (2, 3)
view_shape = (len(a) - window[0] + 1,) + window # (4,2,3) if len(a) == 5
sub_matrix = np.lib.stride_tricks.as_strided(a, shape = view_shape, strides = (a.itemsize,) + a.strides)
>>> sub_matrix
array([[[0. , 0. , 0. ],
        [0. , 0. , 1.5]],

       [[0. , 0. , 1.5],
        [1. , 0. , 1.5]],

       [[1. , 0. , 1.5],
        [1. , 2.5, 1.5]],

       [[1. , 2.5, 1.5],
        [0.5, 2.5, 1.5]]])

请注意， np.lib.stride_tricks 与其他方法相比，性能非常出色.

Note that np.lib.stride_tricks is very performant against any alternative ways.

这篇关于numpy数组转换中的性能改进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！