使用python joblib的并行类函数调用

Question

可以使用 joblib 对 python 中的函数进行多次调用。

from joblib import Parallel, delayed 

def normal(x):
    print "Normal", x
    return x**2

if  __name__ == '__main__':

    results = Parallel(n_jobs=2)(delayed(normal)(x) for x in range(20))
    print results

给出： [0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100, 121, 144, 169, 196, 225, 256, 289, 324, 361]

但是，我真正想要的是并行调用类实例列表上的类函数。 该函数只是存储一个类变量。 然后稍后我将访问这个变量。

from joblib import Parallel, delayed 

class A(object):
    def __init__(self, x):
        self.x = x
    def p(self):
        self.y = self.x**2

if  __name__ == '__main__':

    runs = [A(x) for x in range(20)]
    Parallel(n_jobs=4)(delayed(run.p() for run in runs))
    for run in runs:
        print run.y

这给出了一个错误：

回溯（最近一次调用最后一次）：

File "", line 1, in runfile('G:/My Drive/CODE/stackoverflow/parallel_classfunc/parallel_classfunc.py', wdir='G:/My Drive/CODE/stackoverflow/parallel_classfunc')

文件“C:\\ProgramData\\Anaconda2\\lib\\site-packages\\spyder\\utils\\site\\sitecustomize.py”，第 710 行，在运行文件 execfile(filename, namespace) 中

文件“C:\\ProgramData\\Anaconda2\\lib\\site-packages\\spyder\\utils\\site\\sitecustomize.py”，第 86 行，在 execfile exec(compile(scripttext, filename, 'exec'), glob, loc)

文件“G:/My Drive/CODE/stackoverflow/parallel_classfunc/parallel_classfunc.py”，第 12 行，在 Parallel(n_jobs=4)(delayed(run.p() for run in running))

文件“C:\\ProgramData\\Anaconda2\\lib\\site-packages\\joblib\\parallel.py”，第 183 行，在延迟的pickle.dumps（函数）中

文件“C:\\ProgramData\\Anaconda2\\lib\\copy_reg.py”，第 70 行，在 _reduce_ex 中引发 TypeError，“无法腌制 %s 个对象”% base。 名称

类型错误：不能pickle 生成器对象

如何将 joblib 与这样的类一起使用？ 或者有更好的方法吗？

Answer 1

像这样的类 怎么 可能使用joblib ？

让我们首先提出一些代码完善：

并非所有事情都适合joblib.Parallel()( delayed() )调用签名功能可以吞咽：

# >>> type( runs )                        <type 'list'>
# >>> type( runs[0] )                     <class '__main__.A'>
# >>> type( run.p() for run in runs )     <type 'generator'>

因此，让我们使DEMO对象通过aContainerFUN()传递：

StackOverflow_DEMO_joblib.Parallel.py ：

from sklearn.externals.joblib import Parallel, delayed
import time

class A( object ):

    def __init__( self, x ):
        self.x = x
        self.y = "Defined on .__init__()"

    def p(        self ):
        self.y = self.x**2

def aNormalFUN( aValueOfX ):
    time.sleep( float( aValueOfX ) / 10. )
    print ": aNormalFUN() has got aValueOfX == {0:} to process.".format( aValueOfX )
    return aValueOfX * aValueOfX

def aContainerFUN( aPayloadOBJECT ):
    time.sleep( float( aPayloadOBJECT.x ) / 10. )
    # try: except: finally:
    pass;  aPayloadOBJECT.p()
    print  "| aContainerFUN: has got aPayloadOBJECT.id({0:}) to process. [ Has made .y == {1:}, given .x == {2: } ]".format( id( aPayloadOBJECT ), aPayloadOBJECT.y, aPayloadOBJECT.x )
    time.sleep( 1 )

if __name__ == '__main__':
     # ------------------------------------------------------------------
     results = Parallel( n_jobs = 2
                         )(       delayed( aNormalFUN )( aParameterX )
                         for                             aParameterX in range( 11, 21 )
                         )
     print results
     print '.'
     # ------------------------------------------------------------------
     pass;       runs = [ A( x ) for x in range( 11, 21 ) ]
     # >>> type( runs )                        <type 'list'>
     # >>> type( runs[0] )                     <class '__main__.A'>
     # >>> type( run.p() for run in runs )     <type 'generator'>

     Parallel( verbose = 10,
               n_jobs  = 2
               )(        delayed( aContainerFUN )( run )
               for                                 run in runs
               )

---

结果呢？ 魅力十足！

 C:\\Python27.anaconda> python StackOverflow_DEMO_joblib.Parallel.py 

: aNormalFUN() has got aValueOfX == 11 to process.
: aNormalFUN() has got aValueOfX == 12 to process.
: aNormalFUN() has got aValueOfX == 13 to process.
: aNormalFUN() has got aValueOfX == 14 to process.
: aNormalFUN() has got aValueOfX == 15 to process.
: aNormalFUN() has got aValueOfX == 16 to process.
: aNormalFUN() has got aValueOfX == 17 to process.
: aNormalFUN() has got aValueOfX == 18 to process.
: aNormalFUN() has got aValueOfX == 19 to process.
: aNormalFUN() has got aValueOfX == 20 to process.
[121, 144, 169, 196, 225, 256, 289, 324, 361, 400]
.
| aContainerFUN: has got aPayloadOBJECT.id(50369168) to process. [ Has made .y == 121, given .x ==  11 ]
| aContainerFUN: has got aPayloadOBJECT.id(50369168) to process. [ Has made .y == 144, given .x ==  12 ]
[Parallel(n_jobs=2)]: Done   1 tasks      | elapsed:    2.4s
| aContainerFUN: has got aPayloadOBJECT.id(12896752) to process. [ Has made .y == 169, given .x ==  13 ]
| aContainerFUN: has got aPayloadOBJECT.id(12896752) to process. [ Has made .y == 196, given .x ==  14 ]
[Parallel(n_jobs=2)]: Done   4 tasks      | elapsed:    4.9s
| aContainerFUN: has got aPayloadOBJECT.id(12856464) to process. [ Has made .y == 225, given .x ==  15 ]
| aContainerFUN: has got aPayloadOBJECT.id(12856464) to process. [ Has made .y == 256, given .x ==  16 ]
| aContainerFUN: has got aPayloadOBJECT.id(50368592) to process. [ Has made .y == 289, given .x ==  17 ]
| aContainerFUN: has got aPayloadOBJECT.id(50368592) to process. [ Has made .y == 324, given .x ==  18 ]
| aContainerFUN: has got aPayloadOBJECT.id(12856528) to process. [ Has made .y == 361, given .x ==  19 ]
| aContainerFUN: has got aPayloadOBJECT.id(12856528) to process. [ Has made .y == 400, given .x ==  20 ]
[Parallel(n_jobs=2)]: Done  10 out of  10 | elapsed:   13.3s finished

Answer 2

让第一个调整类 a/c 到第一个函数：

class A(object):
    def __init__(self, x):
        self.x = x
    def p(self):
        self.y = self.x**2
        return self.y

现在要并行运行上述类，只需使用 lambda 函数而不是直接调用它（run.p()）。

from joblib import Parallel, delayed 

class A(object):
    def __init__(self, x):
        self.x = x
    def p(self):
        self.y = self.x**2
        return self.y

if  __name__ == '__main__':
    runs = [A(x) for x in range(20)]
    with Parallel(n_jobs=6, verbose=5) as parallel:
        delayed_funcs = [delayed(lambda x:x.p())(run) for run in runs]
        run_A = parallel(delayed_funcs)

    print(run_A)

您的输出如下所示：

[0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100, 121, 144, 169, 196, 225, 256, 289, 324, 361]
[Parallel(n_jobs=6)]: Using backend LokyBackend with 6 concurrent workers.
[Parallel(n_jobs=6)]: Done   6 tasks      | elapsed:    0.0s
[Parallel(n_jobs=6)]: Done  14 out of  20 | elapsed:    0.0s remaining:    0.0s
[Parallel(n_jobs=6)]: Done  20 out of  20 | elapsed:    0.0s finished