0

Outputting a set of numbers does not give the same sequence even after using random.seed(myseed). This only occurs in Python3, not in Python2 (both on a Debian stable system). Is it a bug or something wrong with my code ? 

import random
seed=20.0
random.seed(seed)
print("seed: {}".format(seed))
test = [str(random.randint(0,1000)) for _ in range(10)]
print(', '.join(test))
ss = set(test)
print(', '.join(ss))

Below Python3 gives a different sequences at each run, but Python2 gives similar sequences across all runs(as expected). 

$ python3 --version
Python 3.4.2
$ python2 --version
Python 2.7.9

#same sequences
$ python2 randtest.py 
seed: 20.0
906, 686, 767, 905, 260, 636, 905, 873, 573, 169
906, 636, 905, 573, 767, 873, 260, 169, 686

$ python2 randtest.py 
seed: 20.0
906, 686, 767, 905, 260, 636, 905, 873, 573, 169
906, 636, 905, 573, 767, 873, 260, 169, 686

$ python2 randtest.py 
seed: 20.0
906, 686, 767, 905, 260, 636, 905, 873, 573, 169
906, 636, 905, 573, 767, 873, 260, 169, 686

#diff sequences
$ python3 randtest.py 
seed: 20.0
927, 740, 702, 805, 784, 901, 926, 154, 266, 690
926, 690, 784, 702, 740, 927, 266, 154, 901, 805

$ python3 randtest.py 
seed: 20.0
927, 740, 702, 805, 784, 901, 926, 154, 266, 690
702, 926, 784, 901, 154, 266, 805, 690, 740, 927

$ python3 randtest.py 
seed: 20.0
927, 740, 702, 805, 784, 901, 926, 154, 266, 690
805, 926, 901, 784, 740, 927, 154, 690, 266, 702
Vu Nguyen
  • 987
  • 1
  • 9
  • 20

1 Answers1

5

You are actually incorrect. Python 3 is returning the same set of numbers. You are making an assumption that set an unordered container will have the same order each time you execute python which is incorrect.

For example, for the last two python3 tests:

>>> a = set([702, 926, 784, 901, 154, 266, 805, 690, 740, 927])
>>> b = set([805, 926, 901, 784, 740, 927, 154, 690, 266, 702])
>>> a == b
True

You can ensure that your sets are correctly ordered by using sorted

print(', '.join(sorted(test)))
GWW
  • 43,129
  • 11
  • 115
  • 108
  • I know that set is an unordered container. However I would expect if you set the random seed then iterating that set would be in some random, but specific order according to the seed. In my example, this is shown with Python2, but not Python3. – Vu Nguyen Jun 25 '16 at 17:20
  • The order of the set is not dependent on your random seed so that assumption is not correct. Maybe it is the case with python2 for your examples but that does not imply it is always true nor does it imply it will be the case with python3. – GWW Jun 25 '16 at 17:22
  • There is a comment [here](http://stackoverflow.com/questions/3848091/python-set-iteration-order-varies-from-run-to-run) that explains why in more detail. Essentially, in some cases the objects `id()` which depends on its location in memory is used in the hash function. – GWW Jun 25 '16 at 17:25
  • I see, so it seems to preserve the order I need to set the PYTHONHASHSEED to some specific value. https://docs.python.org/3.3/using/cmdline.html#envvar-PYTHONHASHSEED – Vu Nguyen Jun 25 '16 at 17:30
  • No that probably won't work either if the memory address of the object is being used. It is better to just sort the set. – GWW Jun 25 '16 at 17:32
  • Ok, sorting will do too. – Vu Nguyen Jun 25 '16 at 17:36