In [1]:

    

import sys
sys.path.append('..')


    

In [2]:

    

import utils
import imp
imp.reload(utils)


    

    
Out[2]:





<module 'utils' from '../utils.py'>

In [3]:

    

import numpy as np
from joblib import Parallel, delayed
from itertools import product
from time import sleep


    

In [8]:

    

def power_it(number, n=2):
    # We add a sleep in here to simulate heavy work that 
    # needs the parallelizing:
    sleep(0.1)
    return number ** n


    

In [9]:

    

my_array = np.arange(100).reshape(10, 10)


    

In [10]:

    

in_list = list(my_array.ravel())


    

In [12]:

    

results = utils.parfor(power_it, in_list)


    

In [15]:

    

results_array = np.array(results).reshape(my_array.shape)


    

In [18]:

    

i, j = np.random.randint(0, 9, 2)
assert results_array[i, j] == power_it(my_array[i, j])


    

In [21]:

    

%timeit power_it(my_array[i, j])


    

    




10 loops, best of 3: 102 ms per loop

In [22]:

    

%timeit utils.parfor(power_it, in_list, n=3, n_jobs=1)


    

    




1 loop, best of 3: 10.2 s per loop

In [23]:

    

%timeit utils.parfor(power_it, in_list, n=3, n_jobs=2)


    

    




1 loop, best of 3: 5.22 s per loop

In [24]:

    

%timeit utils.parfor(power_it, in_list, n=3, n_jobs=8)


    

    




1 loop, best of 3: 1.45 s per loop

In [25]:

    

import multiprocessing


    

In [26]:

    

multiprocessing.cpu_count()


    

    
Out[26]:





4

In [ ]: