lambda：Python的匿名函数

参考１
参考２
lambda函数也叫匿名函数，即，函数没有具体的名称。先来看一个最简单例子：

def f(x):return x**2
print f(4)

Python中使用lambda的话，写成这样

g = lambda x : x**2
print g(4)

lambda表达式在很多编程语言都有对应的实现。比如C#：

var g = x => x**2
Console.WriteLine(g(4))

那么，lambda表达式有什么用处呢？很多人提出了质疑，lambda和普通的函数相比，就是省去了函数名称而已，同时这样的匿名函数，又不能共享在别的地方调用。其实说的没错，lambda在Python这种动态的语言中确实没有起到什么惊天动地的作用，因为有很多别的方法能够代替lambda。同时，使用lambda的写法有时显得并没有那么pythonic。甚至有人提出之后的Python版本要取消lambda。

回过头来想想，Python中的lambda真的没有用武之地吗？其实不是的，至少我能想到的点，主要有：

使用Python写一些执行脚本时，使用lambda可以省去定义函数的过程，让代码更加精简。
对于一些抽象的，不会别的地方再复用的函数，有时候给函数起个名字也是个难题，使用lambda不需要考虑命名的问题。
使用lambda在某些时候让代码更容易理解。

lambda基础

lambda语句中，冒号前是参数，可以有多个，用逗号隔开，冒号右边的返回值。lambda语句构建的其实是一个函数对象，见证一下：

g = lambda x : x**2
print g<function <lambda> at 0x00AFAAF0>>>> def make_incrementor (n): return lambda x: x + n
>>>
>>> f = make_incrementor(2)
>>> g = make_incrementor(6)
>>>
>>> print f(42), g(42)
44 48
>>>
>>> print make_incrementor(22)(33)
55

C#3.0开始，也有了lambda表达式，省去了使用delegate的麻烦写法。C#中的lambda表达式关键字是=>，看下面的一个例子：

var array = new int[] {2, 3, 5, 7, 9};
var result = array.Where(n => n > 3); // [5, 6, 9]

C#使用了扩展方法，才使得数组对象拥有了像Where,Sum之类方便的方法。Python中，也有几个定义好的全局函数方便使用的，他们就是filter, map, reduce。
复制代码

>>> foo = [2, 18, 9, 22, 17, 24, 8, 12, 27]
>>>
>>> print filter(lambda x: x % 3 == 0, foo)
[18, 9, 24, 12, 27]
>>>
>>> print map(lambda x: x * 2 + 10, foo)
[14, 46, 28, 54, 44, 58, 26, 34, 64]
>>>
>>> print reduce(lambda x, y: x + y, foo)
139

非lambda不可？

上面例子中的map的作用，和C#的Where扩展方法一样，非常简单方便。但是，Python是否非要使用lambda才能做到这样的简洁程度呢？在对象遍历处理方面，其实Python的for..in..if语法已经很强大，并且在易读上胜过了lambda。比如上面map的例子，可以写成：

print [x * 2 + 10 for x in foo]

非常的简洁，易懂。filter的例子可以写成：

print [x for x in foo if x % 3 == 0]

同样也是比lambda的方式更容易理解。

所以，什么时候使用lambda，什么时候不用，需要具体情况具体分析，只要表达的意图清晰就好。一般情况下，如果for..in..if能做的，我都不会选择lambda。

lambda broken?

在数学教学中，经常会使用到lambda，比如有一位老兄就遇到这样一个问题。他想创建一个函数数组fs=[f0,…,f9] where fi(n)=i+n. 于是乎，就定义了这么一个lambda函数：

fs = [(lambda n: i + n) for i in range(10)]

但是，奇怪的是，

>>> fs[3](4)
13
>>> fs[4](4)
13
>>> fs[5](4)
13

结果并没有达到这位老兄的预期，预期的结果应该是：
复制代码

>>> fs[3](4)
7
>>> fs[4](4)
8
>>> fs[5](4)
9

问题其实出在变量i上。上面的代码换个简单的不使用lambda的缩减版本：

i = 1
def fs(n):return n + i
print fs(1) # 2i = 2
print fs(1) # 3

可见，上面没有达到预期的原因是lambda中的i使用的是匿名函数外的全局变量。修改一下：
复制代码

fs = [(lambda n, i=i : i + n) for i in range(10)]
>>> fs[3](4)
7
>>> fs[4](4)
8
>>> fs[5](4)
9

应用
The following example is one way to compute prime numbers in Python (not the most efficient one, though)

>>> nums = range(2, 50)
>>> for i in range(2, 8):
...     nums = filter(lambda x: x == i or x % i, nums)
...
>>> print nums
[2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47]

In the following example, a sentence is split up into a list of words, then a list is created that contains the length of each word.

>>> sentence = 'It is raining cats and dogs'
>>> words = sentence.split()
>>> print words
['It', 'is', 'raining', 'cats', 'and', 'dogs']
>>>
>>> lengths = map(lambda word: len(word), words)
>>> print lengths
[2, 2, 7, 4, 3, 4]

I think it doesn’t need any further explanation, the code is practically self-documenting.

Of course, it could all be written in one single statement. Admittedly, this is somewhat less readable (not much, though).

>>> print map(lambda w: len(w), 'It is raining cats and dogs'.split())[2, 2, 7, 4, 3, 4]

Here’s an example from the UNIX scripting world: We want to find all mount points in our file system. To do that, we execute the external “mount” command and parse the output.

>>> import commands
>>>
>>> mount = commands.getoutput('mount -v')
>>> lines = mount.splitlines()
>>> points = map(lambda line: line.split()[2], lines)
>>>
>>> print points
['/', '/var', '/usr', '/usr/local', '/tmp', '/proc']

The getoutput function from the commands module (which is part of the Python standard library) runs the given command and returns its output as a single string. Therefore, we split it up into separate lines first. Finally we use “map” with a lambda function that splits each line (on whitespace, which is the default) and returns just the third element of the result, which is the mountpoint.

Again, we could write all of that in one single statement, which increases compactness but reduces readability:

When writing “real-world” scripts, it is recommended to split up complex statements so that it is easier to see what it does. Also, it is easier to make changes.

However, the task of splitting up the output of a command into a list of lines is very common. You need it all the time when parsing the output of external commands. Therefore, it is common practice to include the split operation on the getoutput line, but do the rest separately. This is a good trade-off between compactness and readability:

>>> lines = commands.getoutput('mount -v').splitlines()
>>>
>>> points = map(lambda line: line.split()[2], lines)
>>> print points
['/', '/var', '/usr', '/usr/local', '/tmp', '/proc']

An even better idea is probably to write a small function for that task, which encapsulates the job of running the command and splitting the output.

On a related note, you can also use so-called list comprehensions to construct lists from other lists. Sometimes this is preferable because of efficiency or readability. The previous example could very well be rewritten using a list comprehension:

>>> lines = commands.getoutput('mount -v').splitlines()
>>>
>>> points = [line.split()[2] for line in lines]
>>> print points
['/', '/var', '/usr', '/usr/local', '/tmp', '/proc']

In many cases, you can use list comprehensions instead of map() or filter(). It depends on the situation which one should be preferred.

Note: The commands module is deprecated in newer versions of Python (though it still works in all 2.x versions). Instead, the subprocess module should be used which is available since Python 2.4. The commands.getoutput() function can be replaced by subprocess.check_output(). Please refer to the documentation for details.