Using mock to write unit tests for random functions
Writing unit tests is important,
but this becomes difficult for functions which use random elements.
One possible approach would be to use known seeds for the random number generation,
allowing you to have a deterministic sequence of random numbers,
so that regression tests could be written.
Recently I’ve been writing unit tests for a function which uses a sequence of random.random() calls to determine what is returned
(this is for mutating individuals within a genetic algorithm).
Because of this, I want full control over what is returned by random.random() so that I can test all possible permutations of what happens inside the function.
We can use the mock package to replace the behaviour of any calls to random.random() with behaviour we define ourselves.
For my purposes, I wanted to make the calls to random be replaced with a sequence of predefined values.
This can be done through passing a generator to the side_effect keyword argument of a mock.Mock object.
This small test shows how calls to random.random are replaced by the list of numbers I provided.
import mock # in python 3: `from unittest import mock` will work
import random
def RandMock(randseq):
return mock.Mock(side_effect=(val for val in randseq))
def test_randomness():
with mock.patch('random.random', RandMock([1, 2, 3, 4])):
assert random.random() == 1
assert random.random() == 2
assert random.random() == 3
assert random.random() == 4
Moving on to a more real example,
I had a function which performed a blend crossover between parents
to produce offspring with values which are a mix of the two parents (or slightly outside).
This had a call to random.random() to first decide if a crossover is performed (90% chance),
and then if so, two more calls (one per child) to decide where in the range of values between the two parents the child value is
(1.0 would extend slightly beyond the highest value, 0.5 would choose a value directly between the parents’ values).
Here I’ve also used a pytest.mark.parametrize
decorator to iterate over many different test cases with a single function.
import mock
import random
import pytest
## Somewhere in my package....
bounds = ((10.0, 30.0),)
blend_alpha = 0.1
blend_probability = 0.9
def clamp(val, lower, upper):
"""He's champin' for a clampin'!"""
val = max(val, lower)
val = min(val, upper)
return val
def blend_crossover(candidates):
new_candidates = []
for mother, father in zip(candidates[::2], candidates[1::2]):
brother, sister = [], []
for a, b, (min_bound, max_bound) in zip(mother, father, bounds):
if random.random() < blend_probability:
smallest, largest = min(a, b), max(a, b)
# range between parents
width = largest - smallest
# amount we go out of bounds from natural range
extra = width * blend_alpha
# start point is (smallest - delta)
# we then move up to (width + 2 * extra) from this point
a = ((smallest - extra) +
random.random() * (width + 2 * extra))
b = ((smallest - extra) +
random.random() * (width + 2 * extra))
a = clamp(a, min_bound, max_bound)
b = clamp(b, min_bound, max_bound)
brother.append(a)
sister.append(b)
new_candidates.append(tuple(brother))
new_candidates.append(tuple(sister))
return new_candidates
## In my test suite....
def RandMock(randseq):
return mock.Mock(side_effect=(val for val in randseq))
@pytest.mark.parametrize('randseq,mum,dad,child1,child2', [
# randseq - (roll for performing crossover,
# child1 position, child2 position)
((0.0, 0.0, 0.0), (15.0,), (25.0,), (14.0,), (14.0,)), # check minimum crossover
((0.0, 0.0, 1.0), (15.0,), (25.0,), (14.0,), (26.0,)), # check other limit
((0.0, 0.5, 0.5), (15.0,), (25.0,), (20.0,), (20.0,)), # check no crossover
((1.0, 1.0, 1.0), (15.0,), (25.0,), (15.0,), (25.0,)), # check no crossover
])
def test_crossover(randseq, mum, dad, child1, child2):
with mock.patch('random.random', RandMock(randseq)):
ret = blend_crossover([mum, dad])
assert ret[0] == child1
assert ret[1] == child2