fastrand

Fast random number generation in Python using PCG: Up to 10x faster than random.randint.

Blog post: Ranged random-number generation is slow in Python

Usage... (don't forget to type the above lines in your shell!)

import fastrand

print("generate an integer in [0,1001)")
fastrand.pcg32bounded(1001) 
print("generate an integer in [100,1000]")
fastrand.pcg32randint(100,1000) # requires Python 3.7 or better
print("Generate a random 32-bit integer.")
fastrand.pcg32()

It is nearly an order of magnitude faster than the alternatives:

python3 -m timeit -s 'import fastrand' 'fastrand.pcg32bounded(1001)'
10000000 loops, best of 5: 23.6 nsec per loop

python3 -m timeit -s 'import fastrand' 'fastrand.pcg32randint(100,1000)'
10000000 loops, best of 5: 24.6 nsec per loop

python3 -m timeit -s 'import random' 'random.randint(0,1000)'
1000000 loops, best of 5: 216 nsec per loop

python3 -m timeit -s 'import numpy' 'numpy.random.randint(0, 1000)'
500000 loops, best of 5: 955 nsec per loop

If you have Linux, macOS or Windows, you should be able to do just pip install...

pip install fastrand

You may need root access (sudo on macOS and Linux).

It is sometimes useful to install a specific version, you can do so as follows;

pip install fastrand==1.2.4

Generally, you can build the library as follows (if you have root):

python setup.py build
python setup.py install 

or

python setup.py build
python setup.py install --home=$HOME
export PYTHONPATH=$PYTHONPATH:~/lib/python

Changing the seed and multiple streams

• You can change the seed with a function like pcg32_seed. The seed determines the random values you get. Be mindful that naive seeds (e.g., int(time.time())) can deliver poor initial randomness. A few calls to pcg32() may help to boost the improve the randomness. Or else you may try a better seed.

• If you need to produce multiple streams of random numbers, merely changing the seed is not enough. You are better off using different increments by calling the pcg32inc. The increments should all be distinct. Note that the least significant bit of the increment is always set to 1 no matter which value you pass: so make sure your increments are distinct 31-bit values (ignoring the least significant bit).

future work

Also look at https://github.com/rkern/line_profiler

Reference

• http://www.pcg-random.org
• Daniel Lemire, Fast Random Integer Generation in an Interval, ACM Transactions on Modeling and Computer Simulation, Volume 29 Issue 1, February 2019