Multiprocessing: first step

[paul18fr]

All

As ever evoqued in a previous post, I'm interested in using multiprocessing to speed-up some parts of my code:

• so far I'm using vectorization to avoid loops whenever it's possible (single thread used accordingly)
  
• I've tested Numba, but from my feedback, it's imited to 1) pure calculation, and 2) not all of numpy functions have been available (following snippet does not work using Numba)
  

So multiprocessing package is the next step, but there are too many docs and I'm deluged under so much information (process, pool, map, and so on).

I've built the following example as a test case: how would you proceed using multiprocessing (I should better say "multithreading" in that case)

Thanks for your advices

Paul

import numpy as np

def ReplaceLine(M, r):
    for i in range(r):
        ind=np.where(M[:, 0]==i)
        if (len(ind) !=0): M[ind, 1:]=[i, i] 
    return M


if __name__ == '__main__':
    #  Test case building
    n=100
    Mat=np.random.random((n, 3))
    i=np.random.randint(0, n, size=n)
    Mat[:, 0]=i
    
    # Run without multiprocessing
    Mat2 = np.copy(Mat)                 #  with "if __name__ == '__main__':", it seems to be a shallow copy
    Without=ReplaceLine(M=Mat2, r=n)
    
    
    # test with multiprocessing : synchronous or asynchronous