How to use multiprocessing for accelerate the following function?

Question

I have the following for loop:

for j in range(len(a_nested_list_of_ints)):
    arr_1_, arr_2_, arr_3_ = foo(a_nested_list_of_ints[j])
    arr_1[j,:] = arr_1_.data.numpy()
    arr_2[j,:] = arr_2_.data.numpy()
    arr_3[j,:] = arr_3_.data.numpy()

Where a_nested_list_of_ints is a nested list of ints. However it is taking a lot of time to finish. How can I optimize it through multiprocessing? So far I tried to use multiprocessing

p = Pool(5)
for j in range(len(a_nested_list_of_ints)):
    arr_1_, arr_2_, arr_3_ = p.map(foo,a_nested_list_of_ints[j])
    arr_1[j,:] = arr_1_.data.numpy()
    arr_2[j,:] = arr_2_.data.numpy()
    arr_3[j,:] = arr_3_.data.numpy()

However, I am getting:

ValueError: not enough values to unpack (expected 3, got 2)

here:

    arr_1_, arr_2_, arr_3_ = p.map(foo,a_nested_list_of_ints[j])

Any idea of how to make the above operation faster? I also even tried with starmap but it aint working.

Answer 1

Here's a pool demo that works:

In [11]: def foo(i): 
    ...:     return np.arange(i), np.arange(10-i) 
    ...:                                                                        
In [12]: with multiprocessing.Pool(processes=2) as pool: 
    ...:     x = pool.map(foo, range(10)) 
    ...:                                                                        
In [13]: x                                                                      
Out[13]: 
[(array([], dtype=int64), array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])),
 (array([0]), array([0, 1, 2, 3, 4, 5, 6, 7, 8])),
 (array([0, 1]), array([0, 1, 2, 3, 4, 5, 6, 7])),
 (array([0, 1, 2]), array([0, 1, 2, 3, 4, 5, 6])),
 (array([0, 1, 2, 3]), array([0, 1, 2, 3, 4, 5])),
 (array([0, 1, 2, 3, 4]), array([0, 1, 2, 3, 4])),
 (array([0, 1, 2, 3, 4, 5]), array([0, 1, 2, 3])),
 (array([0, 1, 2, 3, 4, 5, 6]), array([0, 1, 2])),
 (array([0, 1, 2, 3, 4, 5, 6, 7]), array([0, 1])),
 (array([0, 1, 2, 3, 4, 5, 6, 7, 8]), array([0]))]

pool.map is doing the iteration, not some external for loop.

And to get a little closer to your example:

In [14]: def foo(alist): 
    ...:     return np.arange(*alist), np.zeros(alist,int) 
    ...:      
    ...:                                                                        
In [15]: alists=[(0,3),(1,4),(1,6,2)]                                           
In [16]: with multiprocessing.Pool(processes=2) as pool: 
    ...:     x = pool.map(foo, alists) 
    ...:                                                                        
In [17]: x                                                                      
Out[17]: 
[(array([0, 1, 2]), array([], shape=(0, 3), dtype=int64)),
 (array([1, 2, 3]), array([[0, 0, 0, 0]])),
 (array([1, 3, 5]), array([[[0, 0],
          [0, 0],
          [0, 0],
          [0, 0],
          [0, 0],
          [0, 0]]]))]

Note that pool.map returns a list, with all cases generated from alists. It doesn't make sense to unpack that x.

 x,y = pool.map(...)   # too many values to pack error

I can unpack the x using the zip* idiom:

In [21]: list(zip(*x))                                                          
Out[21]: 
[(array([0, 1, 2]), array([1, 2, 3]), array([1, 3, 5])),
 (array([], shape=(0, 3), dtype=int64), array([[0, 0, 0, 0]]), array([[[0, 0],
          [0, 0],
          [0, 0],
          [0, 0],
          [0, 0],
          [0, 0]]]))]

This a list of 2 tuples; in effect a list version of transpose. This can be unpacked:

In [23]: y,z = zip(*x)                                                          
In [24]: y                                                                      
Out[24]: (array([0, 1, 2]), array([1, 2, 3]), array([1, 3, 5]))
In [25]: z                                                                      
Out[25]: 
(array([], shape=(0, 3), dtype=int64), array([[0, 0, 0, 0]]), array([[[0, 0],
         [0, 0],
         [0, 0],
         [0, 0],
         [0, 0],
         [0, 0]]]))