python比matlab慢,在这个简单的例子中,为什么Matlab看起来比Python慢​​得多

我最近问了

a question on the optimization of a mask function in Matlab.我得到了两个对我有帮助的答案,但似乎根据我的时间,所有Matlab解决方案看起来都比Numpy解决方案慢得多.不同功能的代码可以在我之前的问题中找到,但为了了解我在做什么,我给出了Numpy“循环”解决方案,当然这不是最快的解决方案,但可能是最简单的解读&＃xff1a;

def dealiasing2d(where, data):

nk, n0, n1 &＃61; data.shape

for i0 in xrange(n0):

for i1 in xrange(n1):

if where[i0, i1]:

data[:, i0, i1] &＃61; 0.

我获得(使用Matlab R2014b和与Blas和Lapack相关联的“基本”Numpy 1.9.1)(n0 &＃61; n1 &＃61; N)&＃xff1a;

N &＃61; 500 ; nk &＃61; 500:

Method | time (s) | normalized

----------------|----------|------------

Numpy | 0.05 | 1.0

Numpy loop | 0.05 | 1.0

Matlab find | 0.74 | 14.8

Matlab bsxfun2 | 0.76 | 15.2

Matlab bsxfun | 0.78 | 15.6

Matlab loop | 0.78 | 15.6

Matlab repmat | 0.89 | 17.8

N &＃61; 500 ; nk &＃61; 100:

Method | time (s) | normalized

----------------|----------|------------

Numpy | 0.01 | 1.0

Numpy loop | 0.03 | 3.0

Matlab find | 0.15 | 13.6

Matlab bsxfun2 | 0.15 | 13.6

Matlab bsxfun | 0.16 | 14.5

Matlab loop | 0.16 | 14.5

Matlab repmat | 0.18 | 16.4

N &＃61; 2000 ; nk &＃61; 10:

Method | time (s) | normalized

----------------|----------|------------

Numpy | 0.02 | 1.0

Matlab find | 0.23 | 13.8

Matlab bsxfun2 | 0.23 | 13.8

Matlab bsxfun | 0.26 | 15.6

Matlab repmat | 0.28 | 16.8

Matlab loop | 0.34 | 20.4

Numpy loop | 0.42 | 25.1

在我看来,这些结果非常奇怪.对我来说,Numpy和Matlab在科学计算方面非常相似所以性能应该相似,而这里有10倍以上&＃xff01;所以我的第一个猜测是我比较两种语言的方式有问题.另一种可能是我的Matlab设置有问题,但我不明白为什么.或者Matlab和Numpy之间真正的深刻区别&＃xff1f;

有人可以用这些函数来验证这些结果吗&＃xff1f;你知道为什么Matlab在这个简单的例子中看起来比Python慢​​得多吗&＃xff1f;

为了计算Matlab函数的时间,我使用了一个文件&＃xff1a;

N &＃61; 500;

n0 &＃61; N;

n1 &＃61; N;

nk &＃61; 500;

disp([&＃39;N &＃61; &＃39;, num2str(N), &＃39; ; nk &＃61; &＃39;, num2str(nk)])

where &＃61; false([n1, n0]);

where(1:100, 1:100) &＃61; 1;

data &＃61; (5.&＃43;1i)*ones([n1, n0, nk]);

disp(&＃39;time dealiasing2d_loops:&＃39;)

time &＃61; timeit(&＃64;() dealiasing2d_loops(where, data));

disp([&＃39; &＃39;, num2str(time), &＃39; s&＃39;])

disp(&＃39;time dealiasing2d_find:&＃39;)

time &＃61; timeit(&＃64;() dealiasing2d_find(where, data));

disp([&＃39; &＃39;, num2str(time), &＃39; s&＃39;])

disp(&＃39;time dealiasing2d_bsxfun:&＃39;)

time &＃61; timeit(&＃64;() dealiasing2d_bsxfun(where, data));

disp([&＃39; &＃39;, num2str(time), &＃39; s&＃39;])

disp(&＃39;time dealiasing2d_bsxfun2:&＃39;)

time &＃61; timeit(&＃64;() dealiasing2d_bsxfun2(where, data));

disp([&＃39; &＃39;, num2str(time), &＃39; s&＃39;])

disp(&＃39;time dealiasing2d_repmat:&＃39;)

time &＃61; timeit(&＃64;() dealiasing2d_repmat(where, data));

disp([&＃39; &＃39;, num2str(time), &＃39; s&＃39;])

我测量Python函数的性能

from __future__ import print_function

import numpy as np

from timeit import timeit, repeat

import_lines &＃61; {

&＃39;numpy_bad&＃39;: &＃39;from dealiasing_numpy_bad import dealiasing2d as dealiasing&＃39;,

&＃39;numpy&＃39;: &＃39;from dealiasing_numpy import dealiasing&＃39;}

tools &＃61; import_lines.keys()

time_approx_one_bench &＃61; 5.

setup &＃61; """

import numpy as np

N &＃61; 500

n0, n1 &＃61; N, N

nk &＃61; 500

where &＃61; np.zeros((n0, n1), dtype&＃61;np.uint8)

where[0:100, 0:100] &＃61; 1

data &＃61; (5.&＃43;1j)*np.ones((nk, n0, n1), dtype&＃61;np.complex128)

"""

exec(setup)

print(&＃39;n0 &＃61; n1 &＃61; {}, nk &＃61; {}&＃39;.format(N, nk))

print(13*&＃39; &＃39; &＃43; &＃39;min mean&＃39;)

durations &＃61; np.empty([len(tools)])

for it, tool in enumerate(tools):

setup_tool &＃61; import_lines[tool] &＃43; setup

duration &＃61; timeit(setup_tool &＃43; &＃39;dealiasing(where, data)&＃39;, number&＃61;1)

nb_repeat &＃61; int(round((time_approx_one_bench - duration)/duration))

nb_repeat &＃61; max(1, nb_repeat)

ds &＃61; np.array(repeat(&＃39;dealiasing(where, data)&＃39;,

setup&＃61;setup_tool, number&＃61;1, repeat&＃61;nb_repeat))

duration &＃61; ds.min()

print(&＃39;{:11s} {:8.2e} s ; {:8.2e} s&＃39;.format(

tool.capitalize() &＃43; &＃39;:&＃39;, duration, ds.mean()))

durations[it] &＃61; duration

fastest &＃61; tools[durations.argmin()].capitalize()

durations &＃61; durations / durations.min()

print(&＃39;Durations normalized by the fastest method (&＃39; &＃43; fastest &＃43; &＃39;):&＃39;)

for it, tool in enumerate(tools):

print(&＃39;{:11s} {:8.3f}&＃39;.format(tool.capitalize() &＃43; &＃39;:&＃39;, durations[it]))