http://www.python.org/pycon/2005/pap...c05_bla_dp.pdf
include this gem:
Functions 'r descriptors
def adder(x, y): return x + y
add23 = adder.__get__(2 3)
add42 = adder.__get__(4 2)
print add23(100), add42(1000)
123 1042
This means that you can do (left) currying without a separate curry function
(Of course, google reveals that the idea has been discussed before,
http://mail.python.org/pipermail/pyt...er/038933.html)
Although it's less flexible than a general curry function, 'method currying' is
much faster, e.g., compare two functions for tail-filtering an iterator:
def filtertail(op, iterable):
"""Recursiv ely filter the tail of an iterator, based on its head
Useful for succinct (though not very fast) implementations
of sieve of eratosthenes among other"""
iterator = iter(iterable)
while 1:
head = iterator.next()
yield head
iterator = it.ifilter(curr y(op,Missing,he ad), iterator)
def filtertail2(op, iterable):
"""An alternative to filtertail, using Alex Martelli's observation
that functions are descriptors. Will not work for built-in
functions that lack a __get__ method"""
iterator = iter(iterable)
opcurry = op.__get__
while 1:
head = iterator.next()
yield head
iterator = it.ifilter(opcu rry(head), iterator)
using these generator functions, a Sieve of Eratosthenes can be written as:
primes = list(filtertail (operator.mod, xrange(2,N)))
or
primes = list(filtertail 2(lambda head, tail: tail % head, xrange(2,N)))
but the second version, using 'method currying' is 4 times the speed, despite
not using the stdlib operator.mod function
def timethem(N):
import time
t1 = time.clock()
p = list(filtertail (op.mod, xrange(2,N)))
t2 = time.clock()
p = list(filtertail 2(lambda head, tail: tail % head, xrange(2,N)))
t3 = time.clock()
return t2-t1, t3-t2
timethem(10000) (3.833199750247 5588, 0.7960575994993 6328) timethem(100000 ) (240.6815100801 9186, 61.818026872130 304)
of course, neither version is anywhere near the most efficient Python
implementation - this is a comparison of currying, not sieving.
BTW, here's the curry function I used (it could probably be faster; I'm not sure
what/where the future stdlib version is)
Missing = Ellipsis
def curry(*cargs, **ckwargs):
fn, cargs = cargs[0], cargs[1:]
if cargs[0] is Missing:
while cargs[0] is Missing: # rightcurry
cargs = cargs[1:]
def call_fn(*fargs, **fkwargs):
d = ckwargs.copy()
d.update(fkwarg s)
return fn(*(fargs+carg s),**d)
name = "%s(...,%s) " % (fn.__name__, ",".join(repr(i ) for i in cargs))
else:
def call_fn(*fargs, **fkwargs):
d = ckwargs.copy()
d.update(fkwarg s)
return fn(*(cargs + fargs), **d)
name = "%s(%s,...) " % (fn.__name__, ",".join(repr(i ) for i in cargs))
call_fn.func_na me = name
call_fn.curry = True
return call_fn
Michael
