Difference between np.random.seed() and np.random.RandomState()

Question

I know that to seed the randomness of numpy.random, and be able to reproduce it, I should us:

import numpy as np
np.random.seed(1234)

but w

Answer 1

random.seed is a method to fill random.RandomState container.

from numpy docs:

numpy.random.seed(seed=None)

Seed the generator.

This method is called when RandomState is initialized. It can be called again to re-seed the generator. For details, see RandomState.

class numpy.random.RandomState

Container for the Mersenne Twister pseudo-random number generator.

Answer 2

If you want to set the seed that calls to np.random... will use, use np.random.seed:

np.random.seed(1234)
np.random.uniform(0, 10, 5)
#array([ 1.9151945 ,  6.22108771,  4.37727739,  7.85358584,  7.79975808])
np.random.rand(2,3)
#array([[ 0.27259261,  0.27646426,  0.80187218],
#       [ 0.95813935,  0.87593263,  0.35781727]])

Use the class to avoid impacting the global numpy state:

r = np.random.RandomState(1234)
r.uniform(0, 10, 5)
#array([ 1.9151945 ,  6.22108771,  4.37727739,  7.85358584,  7.79975808])

And it maintains the state just as before:

r.rand(2,3)
#array([[ 0.27259261,  0.27646426,  0.80187218],
#       [ 0.95813935,  0.87593263,  0.35781727]])

You can see the state of the sort of 'global' class with:

np.random.get_state()

and of your own class instance with:

r.get_state()

Answer 3

np.random.RandomState() constructs a random number generator. It does not have any effect on the freestanding functions in np.random, but must be used explicitly:

>>> rng = np.random.RandomState(42)
>>> rng.randn(4)
array([ 0.49671415, -0.1382643 ,  0.64768854,  1.52302986])
>>> rng2 = np.random.RandomState(42)
>>> rng2.randn(4)
array([ 0.49671415, -0.1382643 ,  0.64768854,  1.52302986])

Answer 4

np.random.RandomState() - a class that provides several methods based on different probability distributions.
np.random.RandomState.seed() - called when RandomState() is initialised.