I've spent a while trying to implement Karatsuba's algorithm in Python and I'm getting close though when I try to multiply two larger numbers (over ~10^15), my result starts to get inaccurate. I can't figure out why.
Side question: Would there be a way for my base case to be "both (instead of either) x and y are strictly less (instead of less) than 10"
def karatsuba(x, y):
# 1. Split ints
if x <= 10 or y <= 10:
#Base case
return x * y
n_x = ceil(log(x, 10)) # Nb of digits in x
n_y = ceil(log(y, 10))
n = max(n_x, n_y)
b = int(x % (10 ** (n // 2)))
a = int(x / (10 ** (n // 2)))
d = int(y % (10 ** (n // 2)))
c = int(y / (10 ** (n // 2)))
# 2. Recursive calls
ac = karatsuba(a, c)
bd = karatsuba(b, d)
kara = karatsuba((a b), (c d))
res = ac * (10 ** (2*(n//2))) (kara - ac - bd) * (10 ** (n//2)) bd
return res
Example :
x = 151222321858446622145369417738339374
y = 875336699541236667457869597252254524
karatsuba(x, y)
returns:
132370448112535269852891372864998437604548273605778561898354233338827976
instead of:
132370448112535277024334963430875927265604725663292579898354233338827976
CodePudding user response:
You lose precision by going through float
due to your /
divisions. Use //
instead. Then you also don't need to convert back to int
. Better yet, use divmod
:
N = 10 ** (n // 2)
a, b = divmod(x, N)
c, d = divmod(y, N)