Python：for循环内的多线程

Question

I have to solve a system of ODE using a fix time step. At each time step, I must compute the output of a number of similar equations (I am programming a neural network and I must compute the value of each neuron. Those neurons are represented by a system of ODE). Thus I would like to parallel the computations inside the loop. I tried using the threading package but I should have done something wrong because it does not work... (In the following code, I skip the generation of the input for clarity)

class EBNMCPU(threading.Thread):
  def __init__(self, threadID, Ts, OnDir):
    self.threadID = threadID
    self.Neuron=NM.EBN(Ts,Dir=OnDir)
    threading.Thread.__init__(self)
  def run(self,Glu=[np.array([1.0],dtype='float'),0.0],GlyOPN=np.array([1.0],dtype='float'),GlyIBN=np.array([0.0],dtype='float')):
    threadLock.acquire()
    self.Neuron.NextStep(Glu, GlyOPN, GlyIBN)
    threadLock.release()

  def getOutput(self):
    return self.Neuron.getOutput()

if __name__ == '__main__':
  Ts=np.array( [0.005] , dtype=float)

  threadLock = threading.Lock()
  EBN=[]
  for ii in np.arange(0,10):
      EBN.append(EBNMCPU(ii,Ts,OnDir=0.0))

  for tc in EBN:
    tc.start() 

  for ii in np.arange(0,NSample):
    for tc in EBN:
      tc.run([np.array([AllY[ii,1]]),0.0],np.array([AllY[ii,2]]),np.array([AllY[ii,3]]))

    for tc in EBN:
      tc.join()

The problem is that I do not see any parallel activity in my system manager...

Thank you for your comments/help,

Pierre

Answer 1

The problem is that I do not see any parallel activity in my system manager

That's because Python threads do not distribute over multiple CPUs/cores due to the global interpreter lock . Use multiprocessing or joblib to write parallel Python programs.

Answer 2

You're calling run() directly, which run the code in the same thread. run() must have no parameter. You should pass every argument when constructing the object then call start() to run it in a new thread.