试试你的Numpy的FFT比Matlab的FFT快多少？

mayaview

不试一下绝对想不到，按照FFTW的宣传他们比fftpack快三倍。Numpy用的是fftpack，Matlab用的是FFTW，为什么Numpy不用更快的FFTW呢？事实是这样吗？大家都测试测试。

Matlab 用这个代码吧，改里面的n，可以试试10的幂次和2的幂次
n=10000;
a=rand(n,1);
tic;fft(a);toc;
tic;fft(a);toc

Python用这个代码吧，改里面的n (在IPython里跑了，懒得写timeit的代码)

import numpy as np
n=10000;
a=np.random.random((n,1))
%timeit np.fft.fft(a)

以下是我的机器的结果
Matlab：
n=10000;
Elapsed time is 0.000147 seconds.
n=100000;
Elapsed time is 0.001076 seconds.
n=1000000;
Elapsed time is 0.012248 seconds.
n=10000000;
Elapsed time is 0.301267 seconds.
n=2^10;
Elapsed time is 0.000087 seconds.
n=2^15;
Elapsed time is 0.000859 seconds.
n=2^20;
Elapsed time is 0.024616 seconds.
n=2^25;
Elapsed time is 0.787241 seconds.

Python：
n=10000;
10000 loops, best of 3: 40.5 us per loop
n=100000;
1000 loops, best of 3: 681 us per loop
n=1000000;
100 loops, best of 3: 7.66 ms per loop
n=10000000;
10 loops, best of 3: 76 ms per loop
n=2**10
100000 loops, best of 3: 7.75 us per loop
n=2**15
10000 loops, best of 3: 124 us per loop
n=2**20
100 loops, best of 3: 7.81 ms per loop
n=2**25
1 loops, best of 3: 245 ms per loop

结论：虽然不知道怎么回事，但是在我机器上Numpy的FFT速度完胜Matlab，和FFTW官网声称的速度相差甚远。请大家在自己的机器上测试一下FFT的速度。

saaya

没试过python

epistemer

试了一下，ipython程序计时比较方便

epistemer

n=10000, 10000 loops, best of 3: 65.5 μs per loop

n=100000, 1000 loops, best of 3: 1.28 ms per loop

n=1000000，100 loops, best of 3: 14.3 ms per loop

n=10000000，  10 loops, best of 3: 148 ms per loop

epistemer

我的耗时比你的多啊，不过我是笔记本
matlab
n=10000;
Elapsed time is 0.250556 seconds.
Elapsed time is 0.000190 seconds.

n=100000;
Elapsed time is 0.043305 seconds.
Elapsed time is 0.002512 seconds.