strange numpy fft performance

Question

During testing I have noticed something strange.

I’m FFT’ing a lot of vectors, and from time to time the numpy FFT function seemed to crash.

I briefly debu

Answer 1

I think that power 2 padding of array some times has several disadvantages:

1. If you cut array to power 2 length it will produce a big loses of the data for big arrays
2. If you pad array with zeros it will produce so called "edge effect"

I have found in this topic that fft performance depends on the array size prime factorization. If the array length is prime number it lead to long calculation time. So I propose following code that decreases array length looking for best it factorization.

from sympy.ntheory import factorint

FACTOR_LIMIT = 100

def bestFFTlength(n):
    while max(factorint(n)) >= FACTOR_LIMIT:
        n -= 1
    return n

a = np.zeros(166400)
audio_fft = np.fft.fft(a,bestFFTlength(len(a)))