我想在 Python (2.7) 中的可迭代对象上运行几个 reduce 函数。一个例子是在一个可迭代的整数上调用 min 和 max。但是当然你不能在同一个迭代器上调用 reduce(min, it) 和 reduce(max, it),因为它在第一次调用后就用完了。所以你可能会想做这样的事情:
reduce(lambda a, b: (min(a[0], b[0]), max(a[1], b[1])), ((x, x) for x in it))
然后您认为,嘿,这非常漂亮,所以您将其概括为如下内容:
from itertools import izip
def multireduce(iterable, *funcs):
""":Return: The tuple resulting from calling ``reduce(func, iterable)`` for each `func` in `funcs`."""
return reduce(lambda a, b: tuple(func(aa, bb) for func, aa, bb in izip(funcs, a, b)), ((item,) * len(funcs) for item in iterable))
(而且你喜欢单元测试,所以你包括这样的东西:)
import unittest
class TestMultireduce(unittest.TestCase):
def test_multireduce(self):
vecs = (
((1,), (min,), (1,)),
(xrange(10), (min, max), (0, 9)),
(xrange(101), (min, max, lambda x, y: x + y,), (0, 100, (100 * 101) // 2))
)
for iterable, funcs, expected in vecs:
self.assertSequenceEqual(tuple(multireduce(iterable, *funcs)), expected)
但随后您尝试了一下,您发现它真的很慢:
%timeit reduce(min, xrange(1000000)) ; reduce(max, xrange(1000000))
10 loops, best of 3: 140 ms per loop
%timeit reduce(lambda a, b: (min(a[0], b[0]), max(a[1], b[1])), ((x, x) for x in xrange(1000000)))
1 loop, best of 3: 682 ms per loop
%timeit multireduce(xrange(1000000), min, max)
1 loop, best of 3: 1.99 s per loop
哎呀。于是你来到 Stack Overflow 寻找 Python 优化智慧......
请您参考如下方法:
好吧,就是这样,这在某种程度上破坏了可迭代对象的意义......
def multireduce(iterable, *funcs):
""":Return: The tuple resulting from calling ``reduce(func, iterable)`` for each `func` in `funcs`."""
return tuple(imap(reduce, funcs, tee(iterable, len(funcs))))
但是对于我的测试用例来说它是相当快的:
%timeit multireduce(xrange(1000000), min, max)
10 loops, best of 3: 166 ms per loop
