A numpy/multiprocessing surprise

bad.py

import numpy as np
from multiprocessing import Pool

def test(top):
    return np.random.random_integers(0,100,5)

def main():
    # set this to a large CPU number
    pool = Pool(4)
    results = pool.map(test, [1,1,1,1,1,1])
    print results

if __name__ == '__main__':
    main()

good.py

from numpy.random import RandomState
from multiprocessing import Pool

def test(top):
    prng = RandomState()
    nums = prng.randint(0, 100, size=5)
    return nums

def main():
    pool = Pool(4)
    results = pool.map(test, [1,1,1,1,1,1])
    print results

if __name__ == '__main__':
    main()

links.md

More info:

• https://groups.google.com/forum/#!topic/briansupport/9ErDidIBBFM
• http://numpy-discussion.10968.n7.nabble.com/numpy-random-and-multiprocessing-td16862.html
• http://stackoverflow.com/questions/5836335/consistenly-create-same-random-numpy-array

Output from bad.py:

array([46, 38, 54, 65, 65]) 
array([46, 38, 54, 65, 65]) 
array([46, 38, 54, 65, 65]) 
array([ 1, 12, 75, 2, 22]) 
array([ 1, 12, 75, 2, 22]) 
array([44, 54, 65, 84, 86])] 

Output from good.py:

array([70, 84, 58, 97, 67]) 
array([12, 97, 89, 63, 90]) 
array([23, 72, 31, 72, 63]) 
array([38, 71, 29, 72, 9]) 
array([55, 32, 46, 42, 5]) 
array([50, 91, 91, 44, 81])]