上面的答案对于上面给出的具体例子是有用的，但对于更普遍的自然排序问题，却遗漏了几个有用的例子。我刚刚被其中一个案例咬了一口，所以想出了一个更彻底的解决方案:

def natural_sort_key(string_or_number):
    """
    by Scott S. Lawton <scott@ProductArchitect.com> 2014-12-11; public domain and/or CC0 license

    handles cases where simple 'int' approach fails, e.g.
        ['0.501', '0.55'] floating point with different number of significant digits
        [0.01, 0.1, 1]    already numeric so regex and other string functions won't work (and aren't required)
        ['elm1', 'Elm2']  ASCII vs. letters (not case sensitive)
    """

    def try_float(astring):
        try:
            return float(astring)
        except:
            return astring

    if isinstance(string_or_number, basestring):
        string_or_number = string_or_number.lower()

        if len(re.findall('[.]\d', string_or_number)) <= 1:
            # assume a floating point value, e.g. to correctly sort ['0.501', '0.55']
            # '.' for decimal is locale-specific, e.g. correct for the Anglosphere and Asia but not continental Europe
            return [try_float(s) for s in re.split(r'([\d.]+)', string_or_number)]
        else:
            # assume distinct fields, e.g. IP address, phone number with '.', etc.
            # caveat: might want to first split by whitespace
            # TBD: for unicode, replace isdigit with isdecimal
            return [int(s) if s.isdigit() else s for s in re.split(r'(\d+)', string_or_number)]
    else:
        # consider: add code to recurse for lists/tuples and perhaps other iterables
        return string_or_number

测试代码和几个链接(在StackOverflow上和关闭)在这里: http://productarchitect.com/code/better-natural-sort.py

欢迎您的反馈。这并不是一个明确的解决方案;只是向前迈出了一步。

在@Mark Byers的回答之后，这里有一个接受关键参数的适应，并且更符合pep8。

def natsorted(seq, key=None):
    def convert(text):
        return int(text) if text.isdigit() else text

    def alphanum(obj):
        if key is not None:
            return [convert(c) for c in re.split(r'([0-9]+)', key(obj))]
        return [convert(c) for c in re.split(r'([0-9]+)', obj)]

    return sorted(seq, key=alphanum)

我还做了一个Gist

很可能functools.cmp_to_key()与python的sort的底层实现密切相关。此外，cmp参数是遗留的。现代的方法是将输入项转换为支持所需的丰富比较操作的对象。

在CPython 2下。X，即使没有实现各自的富比较操作符，也可以对不同类型的对象排序。在CPython 3下。X，不同类型的对象必须显式地支持比较。参见Python如何比较字符串和int?链接到官方文件。大多数答案都依赖于这种隐含的顺序。切换到Python 3。X将需要一个新的类型来实现和统一数字和字符串之间的比较。

Python 2.7.12 (default, Sep 29 2016, 13:30:34) 
>>> (0,"foo") < ("foo",0)
True  

Python 3.5.2 (default, Oct 14 2016, 12:54:53) 
>>> (0,"foo") < ("foo",0)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  TypeError: unorderable types: int() < str()

有三种不同的方法。第一种使用嵌套类来利用Python的Iterable比较算法。第二个函数将这个嵌套展开到单个类中。第三种方法放弃了str的子类化，专注于性能。所有都是有时间的;第二辆快了一倍，第三辆快了近六倍。对str进行子类化并不是必需的，而且首先可能是一个坏主意，但它确实带来了某些便利。

排序字符被复制以强制按大小写排序，并交换大小写以强制小写字母优先排序;这就是“自然排序”的典型定义。我无法决定分组的类型;有些人可能更喜欢以下选项，这也会带来显著的性能优势:

d = lambda s: s.lower()+s.swapcase()

在使用的地方，比较运算符被设置为object的运算符，这样它们就不会被functools. total_orders忽略。

import functools
import itertools


@functools.total_ordering
class NaturalStringA(str):
    def __repr__(self):
        return "{}({})".format\
            ( type(self).__name__
            , super().__repr__()
            )
    d = lambda c, s: [ c.NaturalStringPart("".join(v))
                        for k,v in
                       itertools.groupby(s, c.isdigit)
                     ]
    d = classmethod(d)
    @functools.total_ordering
    class NaturalStringPart(str):
        d = lambda s: "".join(c.lower()+c.swapcase() for c in s)
        d = staticmethod(d)
        def __lt__(self, other):
            if not isinstance(self, type(other)):
                return NotImplemented
            try:
                return int(self) < int(other)
            except ValueError:
                if self.isdigit():
                    return True
                elif other.isdigit():
                    return False
                else:
                    return self.d(self) < self.d(other)
        def __eq__(self, other):
            if not isinstance(self, type(other)):
                return NotImplemented
            try:
                return int(self) == int(other)
            except ValueError:
                if self.isdigit() or other.isdigit():
                    return False
                else:
                    return self.d(self) == self.d(other)
        __le__ = object.__le__
        __ne__ = object.__ne__
        __gt__ = object.__gt__
        __ge__ = object.__ge__
    def __lt__(self, other):
        return self.d(self) < self.d(other)
    def __eq__(self, other):
        return self.d(self) == self.d(other)
    __le__ = object.__le__
    __ne__ = object.__ne__
    __gt__ = object.__gt__
    __ge__ = object.__ge__

import functools
import itertools


@functools.total_ordering
class NaturalStringB(str):
    def __repr__(self):
        return "{}({})".format\
            ( type(self).__name__
            , super().__repr__()
            )
    d = lambda s: "".join(c.lower()+c.swapcase() for c in s)
    d = staticmethod(d)
    def __lt__(self, other):
        if not isinstance(self, type(other)):
            return NotImplemented
        groups = map(lambda i: itertools.groupby(i, type(self).isdigit), (self, other))
        zipped = itertools.zip_longest(*groups)
        for s,o in zipped:
            if s is None:
                return True
            if o is None:
                return False
            s_k, s_v = s[0], "".join(s[1])
            o_k, o_v = o[0], "".join(o[1])
            if s_k and o_k:
                s_v, o_v = int(s_v), int(o_v)
                if s_v == o_v:
                    continue
                return s_v < o_v
            elif s_k:
                return True
            elif o_k:
                return False
            else:
                s_v, o_v = self.d(s_v), self.d(o_v)
                if s_v == o_v:
                    continue
                return s_v < o_v
        return False
    def __eq__(self, other):
        if not isinstance(self, type(other)):
            return NotImplemented
        groups = map(lambda i: itertools.groupby(i, type(self).isdigit), (self, other))
        zipped = itertools.zip_longest(*groups)
        for s,o in zipped:
            if s is None or o is None:
                return False
            s_k, s_v = s[0], "".join(s[1])
            o_k, o_v = o[0], "".join(o[1])
            if s_k and o_k:
                s_v, o_v = int(s_v), int(o_v)
                if s_v == o_v:
                    continue
                return False
            elif s_k or o_k:
                return False
            else:
                s_v, o_v = self.d(s_v), self.d(o_v)
                if s_v == o_v:
                    continue
                return False
        return True
    __le__ = object.__le__
    __ne__ = object.__ne__
    __gt__ = object.__gt__
    __ge__ = object.__ge__

import functools
import itertools
import enum


class OrderingType(enum.Enum):
    PerWordSwapCase         = lambda s: s.lower()+s.swapcase()
    PerCharacterSwapCase    = lambda s: "".join(c.lower()+c.swapcase() for c in s)


class NaturalOrdering:
    @classmethod
    def by(cls, ordering):
        def wrapper(string):
            return cls(string, ordering)
        return wrapper
    def __init__(self, string, ordering=OrderingType.PerCharacterSwapCase):
        self.string = string
        self.groups = [ (k,int("".join(v)))
                            if k else
                        (k,ordering("".join(v)))
                            for k,v in
                        itertools.groupby(string, str.isdigit)
                      ]
    def __repr__(self):
        return "{}({})".format\
            ( type(self).__name__
            , self.string
            )
    def __lesser(self, other, default):
        if not isinstance(self, type(other)):
            return NotImplemented
        for s,o in itertools.zip_longest(self.groups, other.groups):
            if s is None:
                return True
            if o is None:
                return False
            s_k, s_v = s
            o_k, o_v = o
            if s_k and o_k:
                if s_v == o_v:
                    continue
                return s_v < o_v
            elif s_k:
                return True
            elif o_k:
                return False
            else:
                if s_v == o_v:
                    continue
                return s_v < o_v
        return default
    def __lt__(self, other):
        return self.__lesser(other, default=False)
    def __le__(self, other):
        return self.__lesser(other, default=True)
    def __eq__(self, other):
        if not isinstance(self, type(other)):
            return NotImplemented
        for s,o in itertools.zip_longest(self.groups, other.groups):
            if s is None or o is None:
                return False
            s_k, s_v = s
            o_k, o_v = o
            if s_k and o_k:
                if s_v == o_v:
                    continue
                return False
            elif s_k or o_k:
                return False
            else:
                if s_v == o_v:
                    continue
                return False
        return True
    # functools.total_ordering doesn't create single-call wrappers if both
    # __le__ and __lt__ exist, so do it manually.
    def __gt__(self, other):
        op_result = self.__le__(other)
        if op_result is NotImplemented:
            return op_result
        return not op_result
    def __ge__(self, other):
        op_result = self.__lt__(other)
        if op_result is NotImplemented:
            return op_result
        return not op_result
    # __ne__ is the only implied ordering relationship, it automatically
    # delegates to __eq__

>>> import natsort
>>> import timeit
>>> l1 = ['Apple', 'corn', 'apPlE', 'arbour', 'Corn', 'Banana', 'apple', 'banana']
>>> l2 = list(map(str, range(30)))
>>> l3 = ["{} {}".format(x,y) for x in l1 for y in l2]
>>> print(timeit.timeit('sorted(l3+["0"], key=NaturalStringA)', number=10000, globals=globals()))
362.4729259099986
>>> print(timeit.timeit('sorted(l3+["0"], key=NaturalStringB)', number=10000, globals=globals()))
189.7340817489967
>>> print(timeit.timeit('sorted(l3+["0"], key=NaturalOrdering.by(OrderingType.PerCharacterSwapCase))', number=10000, globals=globals()))
69.34636392899847
>>> print(timeit.timeit('natsort.natsorted(l3+["0"], alg=natsort.ns.GROUPLETTERS | natsort.ns.LOWERCASEFIRST)', number=10000, globals=globals()))
98.2531585780016

自然排序既相当复杂，又定义模糊。不要忘记事先运行unicodedata.normalize(…)，并考虑使用str.casefold()而不是str.lower()。可能有一些微妙的编码问题我还没有考虑到。因此，我尝试性地推荐natsort库。我快速浏览了一下github存储库;代码维护非常出色。

All the algorithms I've seen depend on tricks such as duplicating and lowering characters, and swapping case. While this doubles the running time, an alternative would require a total natural ordering on the input character set. I don't think this is part of the unicode specification, and since there are many more unicode digits than [0-9], creating such a sorting would be equally daunting. If you want locale-aware comparisons, prepare your strings with locale.strxfrm per Python's Sorting HOW TO.

>>> import re
>>> sorted(lst, key=lambda x: int(re.findall(r'\d+$', x)[0]))
['elm0', 'elm1', 'Elm2', 'elm9', 'elm10', 'Elm11', 'Elm12', 'elm13']

上面的答案对于上面给出的具体例子是有用的，但对于更普遍的自然排序问题，却遗漏了几个有用的例子。我刚刚被其中一个案例咬了一口，所以想出了一个更彻底的解决方案:

def natural_sort_key(string_or_number):
    """
    by Scott S. Lawton <scott@ProductArchitect.com> 2014-12-11; public domain and/or CC0 license

    handles cases where simple 'int' approach fails, e.g.
        ['0.501', '0.55'] floating point with different number of significant digits
        [0.01, 0.1, 1]    already numeric so regex and other string functions won't work (and aren't required)
        ['elm1', 'Elm2']  ASCII vs. letters (not case sensitive)
    """

    def try_float(astring):
        try:
            return float(astring)
        except:
            return astring

    if isinstance(string_or_number, basestring):
        string_or_number = string_or_number.lower()

        if len(re.findall('[.]\d', string_or_number)) <= 1:
            # assume a floating point value, e.g. to correctly sort ['0.501', '0.55']
            # '.' for decimal is locale-specific, e.g. correct for the Anglosphere and Asia but not continental Europe
            return [try_float(s) for s in re.split(r'([\d.]+)', string_or_number)]
        else:
            # assume distinct fields, e.g. IP address, phone number with '.', etc.
            # caveat: might want to first split by whitespace
            # TBD: for unicode, replace isdigit with isdecimal
            return [int(s) if s.isdigit() else s for s in re.split(r'(\d+)', string_or_number)]
    else:
        # consider: add code to recurse for lists/tuples and perhaps other iterables
        return string_or_number

测试代码和几个链接(在StackOverflow上和关闭)在这里: http://productarchitect.com/code/better-natural-sort.py

欢迎您的反馈。这并不是一个明确的解决方案;只是向前迈出了一步。

我写了一个基于http://www.codinghorror.com/blog/2007/12/sorting-for-humans-natural-sort-order.html的函数，它增加了传递自己的“键”参数的能力。我需要这样才能执行包含更复杂对象(不仅仅是字符串)的列表的自然排序。

import re

def natural_sort(list, key=lambda s:s):
    """
    Sort the list into natural alphanumeric order.
    """
    def get_alphanum_key_func(key):
        convert = lambda text: int(text) if text.isdigit() else text 
        return lambda s: [convert(c) for c in re.split('([0-9]+)', key(s))]
    sort_key = get_alphanum_key_func(key)
    list.sort(key=sort_key)

例如:

my_list = [{'name':'b'}, {'name':'10'}, {'name':'a'}, {'name':'1'}, {'name':'9'}]
natural_sort(my_list, key=lambda x: x['name'])
print my_list
[{'name': '1'}, {'name': '9'}, {'name': '10'}, {'name': 'a'}, {'name': 'b'}]