Guillaume Redoulès

Optimized numpy random number generation on Intel CPU

  2. Python
  3. Other

Python Intel distribution

Make sure you have a python intel distribution. When you startup python you should see somethine like :

Python 3.6.2 |Intel Corporation| (default, Aug 15 2017, 11:34:02) [MSC v.1900 64 bit (AMD64)]
Type 'copyright', 'credits' or 'license' for more information
IPython 6.1.0 -- An enhanced Interactive Python. Type '?' for help.

If not, you can force the installation of the intel optimized python with :

conda update --all
conda config --add channels intel
conda install numpy --channel intel --override-channels

oh and by the way, make sure you a running an Intel CPU ;)

Comparing numpy.random with numpy.random_intel

Let's now test both the rand function with and without the Intel optimization

import numpy as np
from numpy import random, random_intel

%timeit np.random.rand(10**5)

1.06 ms ± 91.5 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

%timeit np.random_intel.rand(10**5)

225 µs ± 3.46 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)