当查看in -operator的TimeComplexity时，在最坏的情况下它与O(n)一起工作。即使是集合。

因此，当比较两个数组时，最好情况下的TimeComplexity为O(n)，最坏情况下为O(n²)。

另一种(但不幸的是更复杂)解决方案，在最好和最坏的情况下都适用于O(n):

# Compares the difference of list a and b
# uses a callback function to compare items
def diff(a, b, callback):
  a_missing_in_b = []
  ai = 0
  bi = 0

  a = sorted(a, callback)
  b = sorted(b, callback)

  while (ai < len(a)) and (bi < len(b)):

    cmp = callback(a[ai], b[bi])
    if cmp < 0:
      a_missing_in_b.append(a[ai])
      ai += 1
    elif cmp > 0:
      # Item b is missing in a
      bi += 1
    else:
      # a and b intersecting on this item
      ai += 1
      bi += 1

  # if a and b are not of same length, we need to add the remaining items
  for ai in xrange(ai, len(a)):
    a_missing_in_b.append(a[ai])


  return a_missing_in_b

e.g.

>>> a=[1,2,3]
>>> b=[2,4,6]
>>> diff(a, b, cmp)
[1, 3]

有三种选择，其中两种是可以接受的，另一种不应该这样做。

在较高的级别上，这3个选项是:

减去两组(有时最好) 检查每个列表项是否存在于一个集合中(大多数情况下最好) 检查每个列表项是否存在于列表中(不做)

选项3)永远不应该超过选项2)。根据应用程序的需要，您可能更喜欢选项1)或2)，而在大多数用例中，2)可能是首选方法。2)与1)的性能非常相似，因为两者都具有O(m + n)个时间复杂度。相比之下，2)在空间复杂度上比1)有边际优势，并且既保持了原始列表的顺序，又保持了原始列表中的任何重复。

如果你想删除重复，不关心顺序，那么1)可能是最适合你的。

import time

def fun1(l1, l2):
    # Order and duplications in l1 are both lost, O(m) + O(n)
    return set(l1) - set(l2)

def fun2(l1, l2):
    # Order and duplications in l1 are both preserved, O(m) + O(n)
    l2_set = set(l2)
    return [item for item in l1 if item not in l2_set]

def fun3(l1, l2):
    # Order and duplications in l1 are both preserved, O(m * n)
    # Don't do
    return [item for item in l1 if item not in l2]

A = list(range(7500))
B = list(range(5000, 10000))

loops = 100

start = time.time()
for _ in range(loops):
    fun1(A, B)
print(f"fun1 time: {time.time() - start}")

start = time.time()
for _ in range(loops):
    fun2(A, B)
print(f"fun2 time: {time.time() - start}")

start = time.time()
for _ in range(loops):
    fun3(A, B)
print(f"fun3 time: {time.time() - start}")

fun1 time: 0.03749704360961914
fun2 time: 0.04109621047973633
fun3 time: 32.55076885223389

如果你想要递归地深入到列表中的项目，我已经为python编写了一个包:https://github.com/erasmose/deepdiff

安装

从PyPi安装:

pip install deepdiff

如果你是Python3，你还需要安装:

pip install future six

示例使用

>>> from deepdiff import DeepDiff
>>> from pprint import pprint
>>> from __future__ import print_function

同一对象返回空

>>> t1 = {1:1, 2:2, 3:3}
>>> t2 = t1
>>> ddiff = DeepDiff(t1, t2)
>>> print (ddiff.changes)
    {}

项目类型发生变化

>>> t1 = {1:1, 2:2, 3:3}
>>> t2 = {1:1, 2:"2", 3:3}
>>> ddiff = DeepDiff(t1, t2)
>>> print (ddiff.changes)
    {'type_changes': ["root[2]: 2=<type 'int'> vs. 2=<type 'str'>"]}

某项的值已更改

>>> t1 = {1:1, 2:2, 3:3}
>>> t2 = {1:1, 2:4, 3:3}
>>> ddiff = DeepDiff(t1, t2)
>>> print (ddiff.changes)
    {'values_changed': ['root[2]: 2 ====>> 4']}

项目添加和/或删除

>>> t1 = {1:1, 2:2, 3:3, 4:4}
>>> t2 = {1:1, 2:4, 3:3, 5:5, 6:6}
>>> ddiff = DeepDiff(t1, t2)
>>> pprint (ddiff.changes)
    {'dic_item_added': ['root[5, 6]'],
     'dic_item_removed': ['root[4]'],
     'values_changed': ['root[2]: 2 ====>> 4']}

字符串的区别

>>> t1 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":"world"}}
>>> t2 = {1:1, 2:4, 3:3, 4:{"a":"hello", "b":"world!"}}
>>> ddiff = DeepDiff(t1, t2)
>>> pprint (ddiff.changes, indent = 2)
    { 'values_changed': [ 'root[2]: 2 ====>> 4',
                          "root[4]['b']:\n--- \n+++ \n@@ -1 +1 @@\n-world\n+world!"]}
>>>
>>> print (ddiff.changes['values_changed'][1])
    root[4]['b']:
    --- 
    +++ 
    @@ -1 +1 @@
    -world
    +world!

字符串差2

>>> t1 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":"world!\nGoodbye!\n1\n2\nEnd"}}
>>> t2 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":"world\n1\n2\nEnd"}}
>>> ddiff = DeepDiff(t1, t2)
>>> pprint (ddiff.changes, indent = 2)
    { 'values_changed': [ "root[4]['b']:\n--- \n+++ \n@@ -1,5 +1,4 @@\n-world!\n-Goodbye!\n+world\n 1\n 2\n End"]}
>>>
>>> print (ddiff.changes['values_changed'][0])
    root[4]['b']:
    --- 
    +++ 
    @@ -1,5 +1,4 @@
    -world!
    -Goodbye!
    +world
     1
     2
     End

类型变化

>>> t1 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":[1, 2, 3]}}
>>> t2 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":"world\n\n\nEnd"}}
>>> ddiff = DeepDiff(t1, t2)
>>> pprint (ddiff.changes, indent = 2)
    { 'type_changes': [ "root[4]['b']: [1, 2, 3]=<type 'list'> vs. world\n\n\nEnd=<type 'str'>"]}

列表的区别

>>> t1 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":[1, 2, 3]}}
>>> t2 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":[1, 2]}}
>>> ddiff = DeepDiff(t1, t2)
>>> pprint (ddiff.changes, indent = 2)
    { 'list_removed': ["root[4]['b']: [3]"]}

区别2:注意它不考虑顺序

>>> # Note that it DOES NOT take order into account
... t1 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":[1, 2, 3]}}
>>> t2 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":[1, 3, 2]}}
>>> ddiff = DeepDiff(t1, t2)
>>> pprint (ddiff.changes, indent = 2)
    { }

包含字典的列表:

>>> t1 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":[1, 2, {1:1, 2:2}]}}
>>> t2 = {1:1, 2:2, 3:3, 4:{"a":"hello", "b":[1, 2, {1:3}]}}
>>> ddiff = DeepDiff(t1, t2)
>>> pprint (ddiff.changes, indent = 2)
    { 'dic_item_removed': ["root[4]['b'][2][2]"],
      'values_changed': ["root[4]['b'][2][1]: 1 ====>> 3"]}

如果你的顺序不重要，两个集合都可以散列，你可以在两个集合之间使用一个对称差分。

这将返回集合A或集合B中出现的值，但不会同时出现。

例如，问题显示了在列表A和列表B上执行的差值的返回值。

如果我们要(将两个列表转换为集合并)执行对称差分，我们将在一次操作中得到两者的合并结果。

A = [1,2,3,4]
B = [2,5]
print(set(A) ^ set(B)

# {1, 3, 4, 5}

加上这个答案，因为我还没有看到现有答案中提供的对称差异

在字典列表的情况下，当集合解引发时，完整列表理解解工作

TypeError: unhashable type: 'dict'

测试用例

def diff(a, b):
    return [aa for aa in a if aa not in b]

d1 = {"a":1, "b":1}
d2 = {"a":2, "b":2}
d3 = {"a":3, "b":3}

>>> diff([d1, d2, d3], [d2, d3])
[{'a': 1, 'b': 1}]
>>> diff([d1, d2, d3], [d1])
[{'a': 2, 'b': 2}, {'a': 3, 'b': 3}]

Python 2.7.3(默认，2014年2月27日，19:58:35)- IPython 1.1.0 - timeit:(github gist)

def diff(a, b):
  b = set(b)
  return [aa for aa in a if aa not in b]

def set_diff(a, b):
  return list(set(a) - set(b))

diff_lamb_hension = lambda l1,l2: [x for x in l1 if x not in l2]

diff_lamb_filter = lambda l1,l2: filter(lambda x: x not in l2, l1)

from difflib import SequenceMatcher
def squeezer(a, b):
  squeeze = SequenceMatcher(None, a, b)
  return reduce(lambda p,q: p+q, map(
    lambda t: squeeze.a[t[1]:t[2]],
      filter(lambda x:x[0]!='equal',
        squeeze.get_opcodes())))

结果:

# Small
a = range(10)
b = range(10/2)

timeit[diff(a, b)]
100000 loops, best of 3: 1.97 µs per loop

timeit[set_diff(a, b)]
100000 loops, best of 3: 2.71 µs per loop

timeit[diff_lamb_hension(a, b)]
100000 loops, best of 3: 2.1 µs per loop

timeit[diff_lamb_filter(a, b)]
100000 loops, best of 3: 3.58 µs per loop

timeit[squeezer(a, b)]
10000 loops, best of 3: 36 µs per loop

# Medium
a = range(10**4)
b = range(10**4/2)

timeit[diff(a, b)]
1000 loops, best of 3: 1.17 ms per loop

timeit[set_diff(a, b)]
1000 loops, best of 3: 1.27 ms per loop

timeit[diff_lamb_hension(a, b)]
1 loops, best of 3: 736 ms per loop

timeit[diff_lamb_filter(a, b)]
1 loops, best of 3: 732 ms per loop

timeit[squeezer(a, b)]
100 loops, best of 3: 12.8 ms per loop

# Big
a = xrange(10**7)
b = xrange(10**7/2)

timeit[diff(a, b)]
1 loops, best of 3: 1.74 s per loop

timeit[set_diff(a, b)]
1 loops, best of 3: 2.57 s per loop

timeit[diff_lamb_filter(a, b)]
# too long to wait for

timeit[diff_lamb_filter(a, b)]
# too long to wait for

timeit[diff_lamb_filter(a, b)]
# TypeError: sequence index must be integer, not 'slice'

@roman-bodnarchuk列表推导函数def diff(a, b)似乎更快。