你能得到的用户生成字符串的最快速度是:

if '  ' in text:
    while '  ' in text:
        text = text.replace('  ', ' ')

短路使它比pythonlarry的综合回答略快。如果你追求效率，并严格要求去除单个空格的额外空白，那么可以使用这种方法。

你也可以在Pandas DataFrame中使用字符串分割技术，而不需要使用.apply(..)，如果你需要对大量字符串快速执行操作，这是非常有用的。这是一行话:

df['message'] = (df['message'].str.split()).str.join(' ')

你能得到的用户生成字符串的最快速度是:

if '  ' in text:
    while '  ' in text:
        text = text.replace('  ', ' ')

短路使它比pythonlarry的综合回答略快。如果你追求效率，并严格要求去除单个空格的额外空白，那么可以使用这种方法。

类似于前面的解决方案，但更具体:用一个空格替换两个或多个空格:

>>> import re
>>> s = "The   fox jumped   over    the log."
>>> re.sub('\s{2,}', ' ', s)
'The fox jumped over the log.'

因为@pythonlarry问这里缺少基于生成器的版本

groupby连接很简单。Groupby将对具有相同键的连续元素进行分组。并返回每个组的键对和元素列表。所以当键是空格空格是返回整个组。

from itertools import groupby
def group_join(string):
  return ''.join(' ' if chr==' ' else ''.join(times) for chr,times in groupby(string))

由变体组成的组很简单，但是很慢。现在来看发电机变体。在这里，我们使用了一个迭代器，即字符串，并生成除字符后面的字符外的所有字符。

def generator_join_generator(string):
  last=False
  for c in string:
    if c==' ':
      if not last:
        last=True
        yield ' '
    else:
      last=False
    yield c

def generator_join(string):
  return ''.join(generator_join_generator(string))

所以我用其他的方法测量了时间。

while_replace 0.015868543065153062 re_replace 0.22579886706080288 proper_join 0.40058281796518713 group_join 5.53206754301209 generator_join 1.6673167790286243

Hello和World由64KB的空格分隔

while_replace 2.991308711003512 re_replace 0.08232860406860709 proper_join 6.294375243945979 group_join 2.4320066600339487 generator_join 6.329648651066236

不忘原话

while_replace 0.002160938922315836 re_replace 0.008620491018518806 proper_join 0.005650000995956361 group_join 0.028368217987008393 generator_join 0.009435956948436797

有趣的是，这里几乎空间只有字符串组连接不是那么糟糕 计时显示的中位数总是七次，每次一千次。

使用带有“\s”的正则表达式并执行简单的string.split()也将删除其他空白-如换行符、回车符、制表符。除非需要使用多个空格，否则我将给出这些示例。

我使用了11段，1000个单词，6665字节的Lorem Ipsum来进行真实的时间测试，并在整个过程中使用了随机长度的额外空格:

original_string = ''.join(word + (' ' * random.randint(1, 10)) for word in lorem_ipsum.split(' '))

一行程序实际上是删除任何前导/尾随空格，并保留一个前导/尾随空格(但只有ONE;-)。

# setup = '''

import re

def while_replace(string):
    while '  ' in string:
        string = string.replace('  ', ' ')

    return string

def re_replace(string):
    return re.sub(r' {2,}' , ' ', string)

def proper_join(string):
    split_string = string.split(' ')

    # To account for leading/trailing spaces that would simply be removed
    beg = ' ' if not split_string[ 0] else ''
    end = ' ' if not split_string[-1] else ''

    # versus simply ' '.join(item for item in string.split(' ') if item)
    return beg + ' '.join(item for item in split_string if item) + end

original_string = """Lorem    ipsum        ... no, really, it kept going...          malesuada enim feugiat.         Integer imperdiet    erat."""

assert while_replace(original_string) == re_replace(original_string) == proper_join(original_string)

#'''

# while_replace_test
new_string = original_string[:]

new_string = while_replace(new_string)

assert new_string != original_string

# re_replace_test
new_string = original_string[:]

new_string = re_replace(new_string)

assert new_string != original_string

# proper_join_test
new_string = original_string[:]

new_string = proper_join(new_string)

assert new_string != original_string

NOTE: The "while version" made a copy of the original_string, as I believe once modified on the first run, successive runs would be faster (if only by a bit). As this adds time, I added this string copy to the other two so that the times showed the difference only in the logic. Keep in mind that the main stmt on timeit instances will only be executed once; the original way I did this, the while loop worked on the same label, original_string, thus the second run, there would be nothing to do. The way it's set up now, calling a function, using two different labels, that isn't a problem. I've added assert statements to all the workers to verify we change something every iteration (for those who may be dubious). E.g., change to this and it breaks:

# while_replace_test
new_string = original_string[:]

new_string = while_replace(new_string)

assert new_string != original_string # will break the 2nd iteration

while '  ' in original_string:
    original_string = original_string.replace('  ', ' ')

Tests run on a laptop with an i5 processor running Windows 7 (64-bit).

timeit.Timer(stmt = test, setup = setup).repeat(7, 1000)

test_string = 'The   fox jumped   over\n\t    the log.' # trivial

Python 2.7.3, 32-bit, Windows
                test |      minum |    maximum |    average |     median
---------------------+------------+------------+------------+-----------
  while_replace_test |   0.001066 |   0.001260 |   0.001128 |   0.001092
     re_replace_test |   0.003074 |   0.003941 |   0.003357 |   0.003349
    proper_join_test |   0.002783 |   0.004829 |   0.003554 |   0.003035

Python 2.7.3, 64-bit, Windows
                test |      minum |    maximum |    average |     median
---------------------+------------+------------+------------+-----------
  while_replace_test |   0.001025 |   0.001079 |   0.001052 |   0.001051
     re_replace_test |   0.003213 |   0.004512 |   0.003656 |   0.003504
    proper_join_test |   0.002760 |   0.006361 |   0.004626 |   0.004600

Python 3.2.3, 32-bit, Windows
                test |      minum |    maximum |    average |     median
---------------------+------------+------------+------------+-----------
  while_replace_test |   0.001350 |   0.002302 |   0.001639 |   0.001357
     re_replace_test |   0.006797 |   0.008107 |   0.007319 |   0.007440
    proper_join_test |   0.002863 |   0.003356 |   0.003026 |   0.002975

Python 3.3.3, 64-bit, Windows
                test |      minum |    maximum |    average |     median
---------------------+------------+------------+------------+-----------
  while_replace_test |   0.001444 |   0.001490 |   0.001460 |   0.001459
     re_replace_test |   0.011771 |   0.012598 |   0.012082 |   0.011910
    proper_join_test |   0.003741 |   0.005933 |   0.004341 |   0.004009

test_string = lorem_ipsum
# Thanks to http://www.lipsum.com/
# "Generated 11 paragraphs, 1000 words, 6665 bytes of Lorem Ipsum"

Python 2.7.3, 32-bit
                test |      minum |    maximum |    average |     median
---------------------+------------+------------+------------+-----------
  while_replace_test |   0.342602 |   0.387803 |   0.359319 |   0.356284
     re_replace_test |   0.337571 |   0.359821 |   0.348876 |   0.348006
    proper_join_test |   0.381654 |   0.395349 |   0.388304 |   0.388193    

Python 2.7.3, 64-bit
                test |      minum |    maximum |    average |     median
---------------------+------------+------------+------------+-----------
  while_replace_test |   0.227471 |   0.268340 |   0.240884 |   0.236776
     re_replace_test |   0.301516 |   0.325730 |   0.308626 |   0.307852
    proper_join_test |   0.358766 |   0.383736 |   0.370958 |   0.371866    

Python 3.2.3, 32-bit
                test |      minum |    maximum |    average |     median
---------------------+------------+------------+------------+-----------
  while_replace_test |   0.438480 |   0.463380 |   0.447953 |   0.446646
     re_replace_test |   0.463729 |   0.490947 |   0.472496 |   0.468778
    proper_join_test |   0.397022 |   0.427817 |   0.406612 |   0.402053    

Python 3.3.3, 64-bit
                test |      minum |    maximum |    average |     median
---------------------+------------+------------+------------+-----------
  while_replace_test |   0.284495 |   0.294025 |   0.288735 |   0.289153
     re_replace_test |   0.501351 |   0.525673 |   0.511347 |   0.508467
    proper_join_test |   0.422011 |   0.448736 |   0.436196 |   0.440318

对于普通字符串，while循环似乎是最快的，其次是Pythonic字符串分割/连接，regex在后面拉。

对于非平凡的字符串，似乎有更多的考虑。32位2.7 ?它是正则表达式的救星!2.7 64位?while循环是最好的。32位3.2，使用“适当的”连接。64位3.3，执行while循环。一次。

最后，如果/何地/何时需要，就可以提高性能，但最好记住这句咒语:

让它起作用 纠正错误 快一点

IANAL, ymv