On Sun, Jul 21, 2013 at 8:31 PM, Steven D'Aprano <steve+comp.lang.pyt...@pearwood.info> wrote: > On Sun, 21 Jul 2013 03:19:24 -0700, pablobarhamalzas wrote: > >> Thank's for all the replies! I've tried some of the imporovements you >> suggested (using math.exp() and sum() or math.fsum()). None of that made >> the code faster, because they are functions you are calling lots of >> times, and function calling is quite time expensive (same as x**(1/2) is >> faster than math.sqrt(x)). > > You are *badly* mistaken. Not only is sqrt more accurate, but it is also > much faster. > > > [steve@ando ~]$ python3.3 -m timeit -s "x = 2.357e7" "x**0.5" > 1000000 loops, best of 3: 0.319 usec per loop > [steve@ando ~]$ python3.3 -m timeit -s "x = 2.357e7" -s "from math import > sqrt" "sqrt(x)" > 10000000 loops, best of 3: 0.172 usec per loop
Don't forget the cost of attribute lookup, which adds 50% to the sqrt() figure. Still faster than exponentiation. (Figures from Python 3.4 alpha, but unlikely to be materially different.) rosuav@sikorsky:~$ python3 -m timeit -s "x = 2.357e7" "x**0.5" 1000000 loops, best of 3: 0.239 usec per loop rosuav@sikorsky:~$ python3 -m timeit -s "x = 2.357e7" -s "from math import sqrt" "sqrt(x)" 10000000 loops, best of 3: 0.102 usec per loop rosuav@sikorsky:~$ python3 -m timeit -s "x = 2.357e7" -s "import math" "math.sqrt(x)" 10000000 loops, best of 3: 0.155 usec per loop ChrisA -- http://mail.python.org/mailman/listinfo/python-list
