在类中使用多处理

Question

我在混乱的代码配置中完美地使用了multiprocessing。我决定对我的代码进行一些排序，并将其作为一个类重新编写，这样我就可以轻松地更改输入，我的新代码如下：

class LikelihoodTest:
      def __init__(self,Xgal,Ygal):
          self.x=Xgal
          self.y=Ygal
          self.objPosition=gal_pos
          self.beta_s=beta
          self.RhoCrit_SigmaC=rho_c_over_sigma_c
          self.AngularDiameter=DA
          self.RhoCrit=rho_crit
          self.Reducedshear=observed_g
          self.ShearError=g_err
      #The 2D function
      def like2d(self,posx, posy):
          stuff=[self.objPosition, self.beta_s, self.RhoCrit_SigmaC , self.AngularDiameter, self.RhoCrit]
          m=4.447e14
          c=7.16
          param=[posx, posy, m, c]
          return reduced_shear( param, stuff, self.Reducedshear, self.ShearError)
      def ShearLikelihood(self):
          n=len(self.x)
          m=len(self.y)
          shared_array_base = multiprocessing.Array(ctypes.c_double, n*m)
          shared_array = np.ctypeslib.as_array(shared_array_base.get_obj())
          shared_array = shared_array.reshape( n,m)
          #Restructure the function before you create instance of Pool.
          # Parallel processing
          def my_func(self,i, def_param=shared_array):
              shared_array[i,:] = np.array([float(self.like2d(self.x[j],self.y[i])) for j in range(len(self.x))])
          while True:
                try:
                   print "processing to estimate likelihood in 2D grids......!!!"
                   start = time.time()
                   pool = multiprocessing.Pool(processes=10)
                   pool.map(my_func, range(len(self.y)))
                   print shared_array
                   end = time.time()
                   print "process time:\n",end - start
                   pool.close()
                except ValueError:
                   print "Oops! value error!"
          return shared_array
      def plotLikelihood(self,shared_array):
          #plotting on a mesh the likelihood function in order to see whether you have defined the inputs correctly and you can observe the maximum likelihood in 2D
          # Set up a regular grid of interpolation points
          xi, yi = np.linspace(self.x.min(), self.x.max(), 100), np.linspace(self.y.min(), self.y.max(), 100)
          # Interpolate
          rbf = scipy.interpolate.interp2d(self.x, self.y,shared_array , kind='linear')
          zi = rbf(xi, yi)
          fig, ax = plt.subplots()
          divider = make_axes_locatable(ax)
          im = ax.imshow(zi, vmin=shared_array.min(), vmax=shared_array.max(), origin='lower',
                        extent=[self.x.min(), self.x.max(), self.y.min(),self.y.max()])
          ax.set_xlabel(r"$Xpos$")
          ax.set_ylabel(r"$Ypos$")
          ax.xaxis.set_label_position('top')
          ax.xaxis.set_tick_params(labeltop='on')
          cax = divider.append_axes("right", size="5%", pad=0.05)
          cbar = fig.colorbar(im,cax=cax, ticks=list(np.linspace(shared_array.max(), shared_array.min(),20)),format='$%.2f$')
          cbar.ax.tick_params(labelsize=8) 
          plt.savefig('/users/Desktop/MassRecons/Likelihood2d_XY_Without_Shear_Uncertainty.pdf', transparent=True, bbox_inches='tight', pad_inches=0)
          plt.close()

尝试使用类配置运行它时出现以下错误：

^{pr2}$

有什么办法解决吗？在

Answer 1

不能使用Pickle将函数或方法传递给不同的进程，但可以传递字符串。在

您可以维护一个方法字典，并通过它们的字符串键引用这些方法。这不是很优雅，但解决了问题。在

编辑： 当您使用多处理时，有一个隐式的“fork”。这将创建多个没有共享资源的独立进程，因为这样，传递给另一个进程的每个东西都必须用Pickle序列化。问题是pickle不允许序列化可执行代码以将其发送到另一个进程。在

Answer 2

我终于可以想出如何在我的类中使用multiprocessing工作。我使用pathos.multiprocessing并将代码更改如下：

import numpy as np
import pathos.multiprocessing as multiprocessing 

class LikelihoodTest:
      def __init__(self,Xgal,Ygal):
          self.x=Xgal
          self.y=Ygal
          self.objPosition=gal_pos
          self.beta_s=beta
          self.RhoCrit_SigmaC=rho_c_over_sigma_c
          self.AngularDiameter=DA
          self.RhoCrit=rho_crit
          self.Reducedshear=observed_g
          self.ShearError=g_err
          #The 2D function
      def like2d(self,posx, posy):
          stuff=[self.objPosition, self.beta_s, self.RhoCrit_SigmaC , self.AngularDiameter, self.RhoCrit]
          m=4.447e14
          c=7.16
          param=[posx, posy, m, c]
          return reduced_shear( param, stuff, self.Reducedshear, self.ShearError)
      def ShearLikelihood(self,r):
          return [float(self.like2d(self.x[j],r)) for j in range(len(self.x))]
      def run(self):
          try:
              print "processing to estimate likelihood in 2D grids......!!!"
              start = time.time()
              pool = multiprocessing.Pool(processes=10)
              seq=[ self.y[i] for i in range( self.y.shape[0])]
              results=np.array( pool.map(self.ShearLikelihood, seq ))
              end = time.time()
              print "process time:\n",end - start
              pool.close()
          except ValueError:
              print "Oops! value error ....!"
          return results
      def plotLikelihood(self,shared_array):
          #plotting on a mesh the likelihood function in order to see whether you have defined the inputs correctly and you can observe the maximum likelihood in 2D
          # Set up a regular grid of interpolation points
          xi, yi = np.linspace(self.x.min(), self.x.max(), 100), np.linspace(self.y.min(), self.y.max(), 100)
          # Interpolate
          rbf = scipy.interpolate.interp2d(self.x, self.y,shared_array , kind='linear')
          zi = rbf(xi, yi)
          fig, ax = plt.subplots()
          divider = make_axes_locatable(ax)
          im = ax.imshow(zi, vmin=shared_array.min(), vmax=shared_array.max(), origin='lower',
                        extent=[self.x.min(), self.x.max(), self.y.min(),self.y.max()])
          ax.set_xlabel(r"$Xpos$")
          ax.set_ylabel(r"$Ypos$")
          ax.xaxis.set_label_position('top')
          ax.xaxis.set_tick_params(labeltop='on')
          cax = divider.append_axes("right", size="5%", pad=0.05)
          cbar = fig.colorbar(im,cax=cax, ticks=list(np.linspace(shared_array.max(), shared_array.min(),20)),format='$%.2f$')
          cbar.ax.tick_params(labelsize=8) 
          plt.savefig('/users/Desktop/MassRecons/Likelihood2d_XY_coordinate.pdf', transparent=True, bbox_inches='tight', pad_inches=0)
          plt.close()

if __name__ == '__main__':
     Xgal = np.linspace(Xgalaxy.min(), Xgalaxy.max(), 1000)
     Ygal = np.linspace(Ygalaxy.min(), Ygalaxy.max(), 1000)          
     Test=LikelihoodTest(Xgal,Ygal) 
     x=Test.run()
     Test.plotLikelihood(x)

现在它就像一个魔咒！：）