Why is seeding the random generator not stable bet

问题:

I am trying to reproduce a random sequence from python's random.random() on a different system with a different python3 version installed.

This should be easy as the documentation says:

Most of the random module’s algorithms and seeding functions are subject to change across Python versions, but two aspects are guaranteed not to change:

• If a new seeding method is added, then a backward compatible seeder will be offered.
• The generator’s random() method will continue to produce the same sequence when the compatible seeder is given the same seed.

So I expect the following code to print always the same 10 numbers, no matter the specific python3 version:

import sys
print(sys.version)

from random import seed, random

seed(str(1))
for i in range(10):
    print(random())

However, testing it on two different machines:

3.2.3 (default, May  3 2012, 15:51:42) 
[GCC 4.6.3]
0.4782479962566343
0.044242767098090496
0.11703586901195051
0.8566892547933538
0.2926790185279551
0.0067328440779825804
0.0013279506360178717
0.22167546902173108
0.9864945747444945
0.5157002525757287

and

3.1.2 (release31-maint, Dec  9 2011, 20:59:40)  
[GCC 4.4.5]
0.0698436845523
0.27772471476
0.833036057868
0.35569897036
0.36366158783
0.722487971761
0.963133581734
0.263723867191
0.451002768569
0.0998765577881

Give different results.

Why is this? And is there any way to make this to work (i.e. get the same random sequence twice?)

回答1:

I was looking through What's New in Python 3.2 (because of this question), and I found:

The random.seed() function and method now salt string seeds with an sha512 hash function. To access the previous version of seed in order to reproduce Python 3.1 sequences, set the version argument to 1, random.seed(s, version=1).

It appears to be a breaking change (from 3.1 to 3.2) with a backwards compatibility option.

(As borrible pointed out, because a compatible seeder is offered the documentation contract has not been violated.)

回答2:

The docs for seed say that they use the hash function to convert strings to valid input seeds. When I tested various versions of Python2.X (don't have 3 installed at the moment), some versions gave different values for hash(str(1)) Note that the docs for seed say that, regardless of version, they use the hash value for the string. You might want to pass an int instead (in addition to @pst 's point about using the backwards-compatible version of seed).

Snippet from the random module docs for 3.2:

If x is an int, it is used directly.

With version 2 (the default), a str, bytes, or bytearray object gets converted to an int and all of its bits are used. With version 1, the hash() of x is used instead.

(x here is the initializer for seed)