How to set the fixed random seed in numpy?

Question

I was believing that setting a seed always gives the same result. But I got different result each times.How to set the seed so that we get the same result each time?

Here is the MWE:

import numpy as np
import pandas as pd


random_state = 100
np.random.state = random_state
np.random.seed = random_state

mu, sigma = 0, 0.25
eps = np.random.normal(mu,sigma,size=100)
print(eps[0])

I get different result each times.

Update:

I can not use np.random.seed(xxx)

Possible duplicate of [How to specify a random seed while using Python's numpy random choice?](https://stackoverflow.com/questions/52991675/how-to-specify-a-random-seed-while-using-pythons-numpy-random-choice) — SpghttCd, Oct 24 '19 at 15:56
@SpghttCd, that answer uses np.random.seed(42), but in my case the function is not working. — BhishanPoudel, Oct 24 '19 at 15:57
@update, because you already overwrote `np.random.seed` by setting it to `100`. Close everything and start again without overwriting. Or perhaps try `del np.random.seed` — SpghttCd, Oct 24 '19 at 15:58
Another dup: https://stackoverflow.com/questions/26309911/making-numpy-random-draws-consistent-for-reproducability — Warren Weckesser, Oct 24 '19 at 16:02
I _knew_ there would be a user with nickname @update - sorry for disturbing... :) — SpghttCd, Oct 24 '19 at 16:04
@warrenweckesser then not only mention it but flag as dupe, please, so that this here can be closed — SpghttCd, Oct 24 '19 at 16:07

score 20 · Accepted Answer · answered Oct 24 '19 at 15:52

20

np.random.seed is a function, which you need to call, not assign to it. E.g.:

np.random.seed(42)

answered Oct 24 '19 at 15:52

orlp

112,504
36
218
315

Fei Yao · Answer 2 · 2020-06-01T15:00:49.123

np.random.seed is function that sets the random state globally. As an alternative, you can also use np.random.RandomState(x) to instantiate a random state class to obtain reproducibility locally. Adapted from your code, I provide an alternative option as follows.

import numpy as np
random_state = 100
rng=np.random.RandomState(random_state )
mu, sigma = 0, 0.25
eps = rng.normal(mu,sigma,size=100) # Difference here
print(eps[0])

More details on np.random.seed and np.random.RandomState can be found here.

How to set the fixed random seed in numpy?

Update:

2 Answers2