Python Performance Tips

Ignore this page! It's over 20 years old and nearly worthless.

"Pyrex" is now Cython. "Psycho" provided an early basis for PyPy. cProfile is a C-based implementation of the pure-Python 'profile' module. Support has been dropped for the "hotshot" module.

The 'sort' method takes a 'key' parameter which implements 'the so-called Schwartzian Transform' (decorate-sort-undecorate). There's no need to implement it manually.

map() now returns an iterator instead of list, so the timings, as given, won't work.

Given:

  python -mtimeit -s 'import random' -s 'data = [str(random.random()) for _ in range(10000)]'

then

  '[x.upper() for x in data]'

is significantly faster than

  'z=[]' 'for x in data: z.append(x.upper())'

(889 usec per loop instead of 1.22 msec per loop).

Use collections.defaultdict to have an initial default value for missing dictionary entries.

The following appears to be wrong:

  dict = {}
  nil = []
  for s in list:
      dict[s] = nil

as all dict values will share the same list object.

The timings have changed so that there is no longer "a significant advantage in Python to adding a number to itself instead of multiplying it by two":

  % python -m timeit -s 'x = 47' 'x * 2'
  10000000 loops, best of 5: 26.1 nsec per loop
  % python -m timeit -s 'x = 47' 'x << 1'
  5000000 loops, best of 5: 44.3 nsec per loop
  % python -m timeit -s 'x = 47' 'x + x'
  10000000 loops, best of 5: 25.5 nsec per loop