用于python 2.7的memoization库


memoization library for python 2.7

我看到python 3.2在functools库中有一个装饰器memoization。http://docs.python.org/py3k/library/functools.html functools.lru缓存

不幸的是，它还没有恢复到2.7。2.7中没有具体原因吗？是否有第三方库提供相同的功能，或者我应该自己写？

相关讨论

Is there any specific reason as why it is not available in 2.7?

@NIRK已经提供了原因：不幸的是，2.x行只接收错误修复，新功能只针对3.x开发。

Is there any 3rd party library providing the same feature?

repoze.lru是针对python 2.6、python2.7和python 3.2的一个lru缓存实现。

文档和源代码在Github上可用。

简单用法：

1
2
3
4
5
6
7
from repoze.lru import lru_cache

@lru_cache(maxsize=500)
def fib(n):
    if n < 2:
        return n
    return fib(n-1) + fib(n-2)
相关讨论

python 3.2.3中的functools模块有一个端口，用于python 2.7和pypy:functools32。

它包括lru_cache装饰器。

相关讨论

我也处于同样的情况，被迫自己执行。python 3.x实现还存在一些其他问题：

  • 主要问题是没有为每个实例启用单独的缓存(如果正在缓存的函数是实例方法)。也就是说，如果我将MaxSize设置为100，并且我有100个实例，如果所有实例都处于相同的活动状态，那么缓存将有效地不做任何事情。
    • 另外，如果运行clear_cache，它会清除所有实例的缓存。
  • 第二件主要的事情是，我需要一个超时特性来每隔X秒清除一次缓存。

python 2.7的函数lru缓存实现：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
import time
import functools
import collections

def lru_cache(maxsize = 255, timeout = None):
   """lru_cache(maxsize = 255, timeout = None) --> returns a decorator which returns an instance (a descriptor).

        Purpose         - This decorator factory will wrap a function / instance method and will supply a caching mechanism to the function.
                            For every given input params it will store the result in a queue of maxsize size, and will return a cached ret_val
                            if the same parameters are passed.

        Params          - maxsize - int, the cache size limit, anything added above that will delete the first values enterred (FIFO).
                            This size is per instance, thus 1000 instances with maxsize of 255, will contain at max 255K elements.
                        - timeout - int / float / None, every n seconds the cache is deleted, regardless of usage. If None - cache will never be refreshed.

        Notes           - If an instance method is wrapped, each instance will have it's own cache and it's own timeout.
                        - The wrapped function will have a cache_clear variable inserted into it and may be called to clear it's specific cache.
                        - The wrapped function will maintain the original function's docstring and name (wraps)
                        - The type of the wrapped function will no longer be that of a function but either an instance of _LRU_Cache_class or a functool.partial type.

        On Error        - No error handling is done, in case an exception is raised - it will permeate up.
   """

    class _LRU_Cache_class(object):
        def __init__(self, input_func, max_size, timeout):
            self._input_func        = input_func
            self._max_size          = max_size
            self._timeout           = timeout

            # This will store the cache for this function, format - {caller1 : [OrderedDict1, last_refresh_time1], caller2 : [OrderedDict2, last_refresh_time2]}.
            #   In case of an instance method - the caller is the instance, in case called from a regular function - the caller is None.
            self._caches_dict        = {}

        def cache_clear(self, caller = None):
            # Remove the cache for the caller, only if exists:
            if caller in self._caches_dict:
                del self._caches_dict[caller]
                self._caches_dict[caller] = [collections.OrderedDict(), time.time()]

        def __get__(self, obj, objtype):
           """ Called for instance methods"""
            return_func = functools.partial(self._cache_wrapper, obj)
            return_func.cache_clear = functools.partial(self.cache_clear, obj)
            # Return the wrapped function and wraps it to maintain the docstring and the name of the original function:
            return functools.wraps(self._input_func)(return_func)

        def __call__(self, *args, **kwargs):
           """ Called for regular functions"""
            return self._cache_wrapper(None, *args, **kwargs)
        # Set the cache_clear function in the __call__ operator:
        __call__.cache_clear = cache_clear


        def _cache_wrapper(self, caller, *args, **kwargs):
            # Create a unique key including the types (in order to differentiate between 1 and '1'):
            kwargs_key ="".join(map(lambda x : str(x) + str(type(kwargs[x])) + str(kwargs[x]), sorted(kwargs)))
            key ="".join(map(lambda x : str(type(x)) + str(x) , args)) + kwargs_key

            # Check if caller exists, if not create one:
            if caller not in self._caches_dict:
                self._caches_dict[caller] = [collections.OrderedDict(), time.time()]
            else:
                # Validate in case the refresh time has passed:
                if self._timeout != None:
                    if time.time() - self._caches_dict[caller][1] > self._timeout:
                        self.cache_clear(caller)

            # Check if the key exists, if so - return it:
            cur_caller_cache_dict = self._caches_dict[caller][0]
            if key in cur_caller_cache_dict:
                return cur_caller_cache_dict[key]

            # Validate we didn't exceed the max_size:
            if len(cur_caller_cache_dict) >= self._max_size:
                # Delete the first item in the dict:
                cur_caller_cache_dict.popitem(False)

            # Call the function and store the data in the cache (call it with the caller in case it's an instance function - Ternary condition):
            cur_caller_cache_dict[key] = self._input_func(caller, *args, **kwargs) if caller != None else self._input_func(*args, **kwargs)
            return cur_caller_cache_dict[key]


    # Return the decorator wrapping the class (also wraps the instance to maintain the docstring and the name of the original function):
    return (lambda input_func : functools.wraps(input_func)(_LRU_Cache_class(input_func, maxsize, timeout)))

单元测试代码：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
#!/usr/bin/python
# -*- coding: utf-8 -*-
import time
import random
import unittest
import lru_cache

class Test_Decorators(unittest.TestCase):
    def test_decorator_lru_cache(self):
        class LRU_Test(object):
           """class"""
            def __init__(self):
                self.num = 0

            @lru_cache.lru_cache(maxsize = 10, timeout = 3)
            def test_method(self, num):
               """test_method_doc"""
                self.num += num
                return self.num

        @lru_cache.lru_cache(maxsize = 10, timeout = 3)
        def test_func(num):
           """test_func_doc"""
            return num

        @lru_cache.lru_cache(maxsize = 10, timeout = 3)
        def test_func_time(num):
           """test_func_time_doc"""
            return time.time()

        @lru_cache.lru_cache(maxsize = 10, timeout = None)
        def test_func_args(*args, **kwargs):
            return random.randint(1,10000000)



        # Init vars:
        c1 = LRU_Test()
        c2 = LRU_Test()
        m1 = c1.test_method
        m2 = c2.test_method
        f1 = test_func

        # Test basic caching functionality:
        self.assertEqual(m1(1), m1(1))
        self.assertEqual(c1.num, 1)     # c1.num now equals 1 - once cached, once real
        self.assertEqual(f1(1), f1(1))

        # Test caching is different between instances - once cached, once not cached:
        self.assertNotEqual(m1(2), m2(2))
        self.assertNotEqual(m1(2), m2(2))

        # Validate the cache_clear funcionality only on one instance:
        prev1 = m1(1)
        prev2 = m2(1)
        prev3 = f1(1)
        m1.cache_clear()
        self.assertNotEqual(m1(1), prev1)
        self.assertEqual(m2(1), prev2)
        self.assertEqual(f1(1), prev3)

        # Validate the docstring and the name are set correctly:
        self.assertEqual(m1.__doc__,"test_method_doc")
        self.assertEqual(f1.__doc__,"test_func_doc")
        self.assertEqual(m1.__name__,"test_method")
        self.assertEqual(f1.__name__,"test_func")

        # Test the limit of the cache, cache size is 10, fill 15 vars, the first 5 will be overwritten for each and the other 5 are untouched. Test that:
        c1.num = 0
        c2.num = 10
        m1.cache_clear()
        m2.cache_clear()
        f1.cache_clear()
        temp_list = map(lambda i : (test_func_time(i), m1(i), m2(i)), range(15))

        for i in range(5, 10):
            self.assertEqual(temp_list[i], (test_func_time(i), m1(i), m2(i)))
        for i in range(0, 5):
            self.assertNotEqual(temp_list[i], (test_func_time(i), m1(i), m2(i)))
        # With the last run the next 5 vars were overwritten, now it should have only 0..4 and 10..14:
        for i in range(5, 10):
            self.assertNotEqual(temp_list[i], (test_func_time(i), m1(i), m2(i)))

        # Test different vars don't collide:
        self.assertNotEqual(test_func_args(1), test_func_args('1'))
        self.assertNotEqual(test_func_args(1.0), test_func_args('1.0'))
        self.assertNotEqual(test_func_args(1.0), test_func_args(1))
        self.assertNotEqual(test_func_args(None), test_func_args('None'))
        self.assertEqual(test_func_args(test_func), test_func_args(test_func))
        self.assertEqual(test_func_args(LRU_Test), test_func_args(LRU_Test))
        self.assertEqual(test_func_args(object), test_func_args(object))
        self.assertNotEqual(test_func_args(1, num = 1), test_func_args(1, num = '1'))
        # Test the sorting of kwargs:
        self.assertEqual(test_func_args(1, aaa = 1, bbb = 2), test_func_args(1, bbb = 2, aaa = 1))
        self.assertNotEqual(test_func_args(1, aaa = '1', bbb = 2), test_func_args(1, bbb = 2, aaa = 1))


        # Sanity validation of values
        c1.num = 0
        c2.num = 10
        m1.cache_clear()
        m2.cache_clear()
        f1.cache_clear()
        self.assertEqual((f1(0), m1(0), m2(0)), (0, 0, 10))
        self.assertEqual((f1(0), m1(0), m2(0)), (0, 0, 10))
        self.assertEqual((f1(1), m1(1), m2(1)), (1, 1, 11))
        self.assertEqual((f1(2), m1(2), m2(2)), (2, 3, 13))
        self.assertEqual((f1(2), m1(2), m2(2)), (2, 3, 13))
        self.assertEqual((f1(3), m1(3), m2(3)), (3, 6, 16))
        self.assertEqual((f1(3), m1(3), m2(3)), (3, 6, 16))
        self.assertEqual((f1(4), m1(4), m2(4)), (4, 10, 20))
        self.assertEqual((f1(4), m1(4), m2(4)), (4, 10, 20))

        # Test timeout - sleep, it should refresh cache, and then check it was cleared:
        prev_time = test_func_time(0)
        self.assertEqual(test_func_time(0), prev_time)
        self.assertEqual(m1(4), 10)
        self.assertEqual(m2(4), 20)
        time.sleep(3.5)
        self.assertNotEqual(test_func_time(0), prev_time)
        self.assertNotEqual(m1(4), 10)
        self.assertNotEqual(m2(4), 20)


if __name__ == '__main__':
    unittest.main()
相关讨论

http://www.python.org/download/releases/3.2.3/

Since the final release of Python 2.7, the 2.x line will only receive bugfixes, and new features are developed for 3.x only.

python 2.7有一些来自3.1的特性，但是在3.2中添加了lru缓存

如评论中所述，http://code.activestate.com/recipes/578078-py26-and-py30-backport-of-python-33s-lru-cache/是一个潜在的解决方案