Created
February 18, 2017 08:15
-
-
Save berenoguz/f8bd037a82a23737a560d695cc9d6a0e to your computer and use it in GitHub Desktop.
The Fast Fourier Transform and The Fast Polynomial Multiplication Algorithms in Python 3
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
from cmath import exp | |
from math import pi | |
# A simple class to simulate n-th root of unity | |
# This class is by no means complete and is implemented | |
# merely for FFT and FPM algorithms | |
class NthRootOfUnity: | |
def __init__(self, n, k = 1): | |
self.k = k | |
self.n = n | |
def __pow__(self, other): | |
if type(other) is int: | |
n = NthRootOfUnity(self.n, self.k * other) | |
return n | |
def __eq__(self, other): | |
if other == 1: | |
return abs(self.n) == abs(self.k) | |
def __mul__(self, other): | |
return exp(2*1j*pi*self.k/self.n)*other | |
def __repr__(self): | |
return str(self.n) + "-th root of unity to the " + str(self.k) | |
@property | |
def th(self): | |
return abs(self.n // self.k) | |
# The Fast Fourier Transform Algorithm | |
# | |
# Input: A, An array of integers of size n representing a polynomial | |
# omega, a root of unity | |
# Output: [A(omega), A(omega^2), ..., A(omega^(n-1))] | |
# Complexity: O(n logn) | |
def FFT(A, omega): | |
if omega == 1: | |
return [sum(A)] | |
B = [[],[]] | |
i = 0 | |
for a in A: | |
B[i%2].append(a) | |
i+=1 | |
o2 = omega**2 | |
C1 = FFT(B[0], o2) | |
C2 = FFT(B[1], o2) | |
C3 = [None]*omega.th | |
for i in range(omega.th//2): | |
C3[i] = C1[i] + omega**i * C2[i] | |
C3[i+omega.th//2] = C1[i] - omega**i * C2[i] | |
return C3 | |
# The Fast Polynomial Multiplication Algorithm | |
# | |
# Input: A,B, two arrays of integers representing polynomials | |
# their length is in O(n) | |
# Output: Coefficient representation of AB | |
# Complexity: O(n logn) | |
def FPM(A,B): | |
n = 1<<(len(A)+len(B)-2).bit_length() | |
o = NthRootOfUnity(n) | |
AT = FFT(A, o) | |
BT = FFT(B, o) | |
C = [AT[i]*BT[i] for i in range(n)] | |
nm = (len(A)+len(B)-1) | |
D = [int((a/n).real) for a in FFT(C, o ** -1)] | |
while True: | |
if D[-1] != 0: | |
return D | |
else: | |
del D[-1] |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Nice code, but there is bug in line 68, cast to int should be replaced with round().
It's fixed in my fork: https://gist.github.com/ksenobojca/dc492206f8a8c7e9c75b155b5bd7a099
For anyone who need faster code - you would need to precalculate results of exp in mul, but that is a major refactor...