如何删除实例方法对象，为了pickle，而不修改原始类

Question

我想持久地保留来自reverend.thomas.Bayes的对象。当然，如果我尝试直接pickle其中一个类，我会得到:

TypeError: can't pickle instancemethod objects

为了解决这个问题，我尝试声明两个函数:

import types
from itertools import chain
from copy import copy
from reverend.thomas import Bayes
def prepare_bayes_for_pickle(bayes_obj):
    dic = copy(bayes_obj.__dict__) #I also tried using deepcopy instead of copy
    for k in dic:
        if type(k) == types.MethodType:
            dic.pop(k)
    return dic
def reconstruct_bayes_from_pickle(bayes_dic):
    b = Bayes()
    # Merge b with bayes_dic, with bayes_dic taking precedence
    dic = dict(chain(bayes_dic, b))
    b.__dict__ = dic
    return b

基本上，我尝试复制对象的__dict__，并尝试通过测试types.MethodType的类型来删除instancemethod。

然后我会通过创建一个新的Bayes对象来重建对象，然后将它与bayes_dic合并在一起(在它被unpickle之后)。

但是，我还没有提到第二个方法，因为我仍然不能pickle从prepare_bayes_for_pickle返回的对象而不得到原始错误。

Answer 1

更好的解决方案是将__getstate__方法添加到Bayes类中(附带__setstate__):

import types
from reverend.thomas import Bayes
def Bayes__getstate__(self):
    state = {}
    for attr, value in self.__dict__.iteritems():
        if not isinstance(value, types.MethodType):
            state[attr] = value
        elif attr == 'combiner' and value.__name__ == 'robinson':
            # by default, self.combiner is set to self.robinson
            state['combiner'] = None
    return state
def Bayes__setstate__(self, state):
    self.__dict__.update(state)
    # support the default combiner (an instance method):
    if 'combiner' in state and state['combiner'] is None:
        self.combiner = self.robinson
Bayes.__getstate__ = Bayes__getstate__
Bayes.__setstate__ = Bayes__setstate__

现在Bayes类总是可以被pickle和取消pickle，而无需额外的处理。

我确实看到类有一个self.cache = {}映射;也许在腌制的时候应该把它排除在外?在__getstate__中忽略它，在__setstate__中调用self.buildCache()。

Answer 2

k是一个键，即属性/方法名。您需要测试属性本身:

    if type(dic[k]) == types.MethodType:
            ^~~~~~ here

我更喜欢使用理解式;你还应该使用isinstance:

dic = dict((k, v) for k, v in bayes_obj.__dict__
           if not isinstance(v, types.MethodType))

Answer 3

这听起来就像用方钉来适应圆孔。如何使用pickle来pickle参数，然后unpickle来重建reverand.Thomas.Bayes对象?

>>> from collections import namedtuple
>>> ArgList = namedtuple('your', 'arguments', 'for', 'the', 'reverand')
>>> def pickle_rtb(n):
...     return pickle.dumps(ArgList(*n.args))
... 
>>> def unpickle_rtb(s):
...     return reverand.Thomas.Bayes(*pickle.loads(s))
... 
>>> s = pickle_rtb(reverand.Thomas.Bayes(1, 2, 3, 4, 5)) # note arguments are a guess
>>> rtb = unpickle_norm(s)

受SO问题的启发