random.choice gives different results on Python 2 and 3 -

background

i want test code depends on random module.

the problematic pr https://github.com/axelrod-python/axelrod/pull/202 , code here https://github.com/axelrod-python/axelrod/blob/master/axelrod/strategies/qlearner.py

the problem

since random module produces pseudo-random numbers, set random.seed(x) known value x. works consecutive test runs. however, python 3 seems give different numbers python 2 when using random.choice([d, c])

following snippet:

import random random.seed(1)  in range(10):     print(random.choice(['c', 'd']), end=', ') 

gives different result python 2 , 3

$ python2 test.py                                                                                                                                                      c, d, d, c, c, c, d, d, c, c  $ python3 test.py c, c, d, c, d, d, d, d, c, c 

however, random.random method works same on 2.x , 3.x:

import random random.seed(1)  in range(10):     print(random.random())  $ python3 test.py 0.13436424411240122 0.8474337369372327 0.763774618976614 0.2550690257394217 0.49543508709194095 0.4494910647887381 0.651592972722763 0.7887233511355132 0.0938595867742349 0.02834747652200631  $ python2 test.py 0.134364244112 0.847433736937 0.763774618977 0.255069025739 0.495435087092 0.449491064789 0.651592972723 0.788723351136 0.0938595867742 0.028347476522 

workaround

i can mock output of random.choice, works simple test cases. however, complicated test cases, i'm not able mock output, because don't know how should like.

the question

have done wrong when calling random.choice method?

according https://docs.python.org/2/library/random.html, rng changed in python 2.4 , may use operating system resources. based on , other answer question, it's not reasonable expect random give same result on 2 different versions of python, 2 different operating systems, or 2 different computers. of knows, next version of python implement random function uses system's microphone generate random sequence.

short version: should never depend on rng give deterministic result. if need known sequence satisfy unit test, need either redesign method or unit test.

one way might split method 2 parts: 1 part generates random number. second part consumes value , acts on it. write 2 unit tests: 1 test coverage of generated values, , separate 1 test output of method based on specific inputs.

another way might change method output not result random number created result. can modify unit test compare 2 , pass or fail test based on expected output of known pairs.

or perhaps unit test can modified run test n number of times , spread confirms sort of randomness.