Can't pickle static method - Multiprocessing -

问题:

I'm applying some parallelization to my code, in which I use classes. I knew that is not possible to pick a class method without any other approach different of what Python provides. I found a solution here. In my code, I have to parts that should be parallelized, both using class. Here, I'm posting a very simple code just representing the structure of mine (is the same, but I deleted the methods content, which was a lot of math calculus, insignificant for the output that I'm getting). The problem is 'cause I can pickle one method (shepard_interpolation), but with the other one (calculate_orientation_uncertainty) I got the pickle error. I don't know why this is happing, or why it works partly.

def _pickle_method(method):
    func_name = method.im_func.__name__
    obj = method.im_self
    cls = method.im_class
    if func_name.startswith('__') and not func_name.endswith('__'): #deal with mangled names
        cls_name = cls.__name__.lstrip('_')
        func_name = '_' + cls_name + func_name
    print cls
    return _unpickle_method, (func_name, obj, cls)


def _unpickle_method(func_name, obj, cls):
    for cls in cls.__mro__:
        try:
            func = cls.__dict__[func_name]
        except KeyError:
            pass
        else:
            break
    return func.__get__(obj, cls)

class ImageData(object):

    def __init__(self, width=60, height=60):
        self.width = width
        self.height = height
        self.data = []
        for i in range(width):
            self.data.append([0] * height)

    def shepard_interpolation(self, seeds=20):
        print "ImD - Sucess"       

import copy_reg
import types
from itertools import product
from multiprocessing import Pool

copy_reg.pickle(types.MethodType, _pickle_method, _unpickle_method)

class VariabilityOfGradients(object):
    def __init__(self):
        pass

    @staticmethod
    def aux():
        return "VoG - Sucess" 

    @staticmethod
    def calculate_orientation_uncertainty():
        results = []
        pool = Pool()
        for x, y in product(range(1, 5), range(1, 5)):
            result = pool.apply_async(VariabilityOfGradients.aux) 
        results.append(result.get())
        pool.close()
        pool.join()        


if __name__ == '__main__':  
    results = []
    pool = Pool()
    for _ in range(3):
        result = pool.apply_async(ImageData.shepard_interpolation, args=[ImageData()])
        results.append(result.get())
    pool.close()
    pool.join()

    VariabilityOfGradients.calculate_orientation_uncertainty()   

When running, I got "PicklingError: Can't pickle : attribute lookup builtin.function failed". And this is almost the same found here. The only difference that I see is that my methods are static.

EDIT:

I noticed that in my calculate_orientation_uncertainty, when I call the function as result = pool.apply_async(VariabilityOfGradients.aux()), i.e., with the parenthesis (in the doc examples I never saw this), it seems to work. But, when I try to get the result, I receive "TypeError: 'int' object is not callable"...

Any help would be appreciated. Thank you in advance.

回答1:

You could define a plain function at the module level and a staticmethod as well. This preserves the calling syntax, introspection and inheritability features of a staticmethod, while avoiding the pickling problem:

def aux():
    return "VoG - Sucess" 

class VariabilityOfGradients(object):
    aux = staticmethod(aux)

---

For example,

import copy_reg
import types
from itertools import product
import multiprocessing as mp

def _pickle_method(method):
    """
    Author: Steven Bethard (author of argparse)
    http://bytes.com/topic/python/answers/552476-why-cant-you-pickle-instancemethods
    """
    func_name = method.im_func.__name__
    obj = method.im_self
    cls = method.im_class
    cls_name = ''
    if func_name.startswith('__') and not func_name.endswith('__'):
        cls_name = cls.__name__.lstrip('_')
    if cls_name:
        func_name = '_' + cls_name + func_name
    return _unpickle_method, (func_name, obj, cls)


def _unpickle_method(func_name, obj, cls):
    """
    Author: Steven Bethard
    http://bytes.com/topic/python/answers/552476-why-cant-you-pickle-instancemethods
    """
    for cls in cls.mro():
        try:
            func = cls.__dict__[func_name]
        except KeyError:
            pass
        else:
            break
    return func.__get__(obj, cls)

copy_reg.pickle(types.MethodType, _pickle_method, _unpickle_method)

class ImageData(object):

    def __init__(self, width=60, height=60):
        self.width = width
        self.height = height
        self.data = []
        for i in range(width):
            self.data.append([0] * height)

    def shepard_interpolation(self, seeds=20):
        print "ImD - Success"       

def aux():
    return "VoG - Sucess" 

class VariabilityOfGradients(object):
    aux = staticmethod(aux)

    @staticmethod
    def calculate_orientation_uncertainty():
        pool = mp.Pool()
        results = []
        for x, y in product(range(1, 5), range(1, 5)):
            # result = pool.apply_async(aux) # this works too
            result = pool.apply_async(VariabilityOfGradients.aux, callback=results.append)
        pool.close()
        pool.join()
        print(results)


if __name__ == '__main__':  
    results = []
    pool = mp.Pool()
    for _ in range(3):
        result = pool.apply_async(ImageData.shepard_interpolation, args=[ImageData()])
        results.append(result.get())
    pool.close()
    pool.join()

    VariabilityOfGradients.calculate_orientation_uncertainty()   

yields

ImD - Success
ImD - Success
ImD - Success
['VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess', 'VoG - Sucess']

---

By the way, result.get() blocks the calling process until the function called by pool.apply_async (e.g. ImageData.shepard_interpolation) is completed. So

for _ in range(3):
    result = pool.apply_async(ImageData.shepard_interpolation, args=[ImageData()])
    results.append(result.get())

is really calling ImageData.shepard_interpolation sequentially, defeating the purpose of the pool.

Instead you could use

for _ in range(3):
    pool.apply_async(ImageData.shepard_interpolation, args=[ImageData()],
                     callback=results.append)

The callback function (e.g. results.append) is called in a thread of the calling process when the function is completed. It is sent one argument -- the return value of the function. Thus nothing blocks the three pool.apply_async calls from being made quickly, and the work done by the three calls to ImageData.shepard_interpolation will be performed concurrently.

Alternatively, it might be simpler to just use pool.map here.

results = pool.map(ImageData.shepard_interpolation, [ImageData()]*3)

回答2:

If you use a fork of multiprocessing called pathos.multiprocesssing, you can directly use classes and class methods in multiprocessing's map functions. This is because dill is used instead of pickle or cPickle, and dill can serialize almost anything in python.

pathos.multiprocessing also provides an asynchronous map function… and it can map functions with multiple arguments (e.g. map(math.pow, [1,2,3], [4,5,6]))

See: What can multiprocessing and dill do together?

and: http://matthewrocklin.com/blog/work/2013/12/05/Parallelism-and-Serialization/

>>> from pathos.multiprocessing import ProcessingPool as Pool
>>> 
>>> p = Pool(4)
>>> 
>>> def add(x,y):
...   return x+y
... 
>>> x = [0,1,2,3]
>>> y = [4,5,6,7]
>>> 
>>> p.map(add, x, y)
[4, 6, 8, 10]
>>> 
>>> class Test(object):
...   def plus(self, x, y): 
...     return x+y
... 
>>> t = Test()
>>> 
>>> p.map(Test.plus, [t]*4, x, y)
[4, 6, 8, 10]
>>> 
>>> p.map(t.plus, x, y)
[4, 6, 8, 10]

Get the code here: https://github.com/uqfoundation/pathos

pathos also has an asynchronous map (amap), as well as imap.