python random.getstate() and random.setstate()

Python’s default generator is a Mersenne Twister with a state space that is 19937 bits, much larger than what you think of as the seed.

You can think of it conceptually as three functions:

  • f(seed) -> state0
  • g(statei) -> statei+1
  • h(statei) -> outcomei

When you start with a seed value using random.seed(), it generates a full state value of 19937 bits one time using function f(). Each time you use the generator, it advances to the next 19937 bit state using g() and returns the output found by collapsing the updated state down a single integer using h().

Normally you don’t actually see the internal state which is at the core of the generator. getstate() bypasses the collapsing function h(), and setstate() bypasses the seeding function f(), so that you can reproduce your sequence from any point without having to go all the way back to the beginning and reproduce the entire sequence to that point.

Most people don’t need to (and shouldn’t) use the get/setstate capability, but it can be useful for pulling some clever mathematical tricks to reduce variability of Monte Carlo estimators.

Leave a Comment