Seedable CSPRNG for python?

With the random module, you are able to seed it to get the same values every time.

import random

random.seed(1)
print(random.randint(1,100)) # outputs 18 every time

lst = [1,2,3]
random.shuffle(lst)
print(lst) # [2, 3, 1] every time

Is there a CSPRNG that can do this? For example, according to this question How can I create a random number that is cryptographically secure in python?, random.SystemRandom is secure. But seeding it doesn’t return the same thing.

from random import SystemRandom

s = SystemRandom()
s.seed(1)
print(s.randint(1,100)) # 81, 16, 100, 58

lst = [1,2,3]
s.shuffle(lst)
print(lst) # [1, 3, 2], [3, 2, 1]

Does such a CSPRNG exist? or does this negate the security aspect?

Answer

The randomgen package provides NumPy compatible CSPRNGs, e.g. ChaCha which can be used as:

import numpy as np
from randomgen import ChaCha

rg = np.random.Generator(ChaCha(seed=1234, rounds=8))
rg.integers(1, 100)

Notes:

• I’m using a reduced round variant, 20 rounds is more normal but 8 seems to be enough for most purposes
• randomgen provides its own Generator, but it’s deprecated and moving into numpy