def test_truncation_selection():
""" Basic truncation selection test"""
pop = [
Individual([0, 0, 0], decoder=IdentityDecoder(), problem=MaxOnes()),
Individual([0, 0, 1], decoder=IdentityDecoder(), problem=MaxOnes()),
Individual([1, 1, 0], decoder=IdentityDecoder(), problem=MaxOnes()),
Individual([1, 1, 1], decoder=IdentityDecoder(), problem=MaxOnes())
]
# We first need to evaluate all the individuals so that truncation selection has fitnesses to compare
pop = Individual.evaluate_population(pop)
truncated = ops.truncation_selection(pop, 2)
assert len(truncated) == 2
# Just to make sure, check that the two best individuals from the original population are in the selected population
assert pop[2] in truncated
assert pop[3] in truncated
def test_naive_cyclic_selection():
""" Test of the naive deterministic cyclic selection """
pop = [
Individual(np.array([0, 0]), problem=MaxOnes()),
Individual(np.array([0, 1]), problem=MaxOnes())
]
# This selection operator will deterministically cycle through the
# given population
selector = ops.naive_cyclic_selection(pop)
selected = next(selector)
assert np.all(selected.genome == [0, 0])
selected = next(selector)
assert np.all(selected.genome == [0, 1])
# And now we cycle back to the first individual
selected = next(selector)
assert np.all(selected.genome == [0, 0])
def _build_test_pop():
"""Construct a synthetic population for illustrating example operations."""
pop = [
Individual([1, 0, 1, 1, 0], IdentityDecoder(), MaxOnes()),
Individual([0, 0, 1, 0, 0], IdentityDecoder(), MaxOnes()),
Individual([0, 1, 1, 1, 1], IdentityDecoder(), MaxOnes()),
Individual([1, 0, 0, 0, 1], IdentityDecoder(), MaxOnes())
]
pop = Individual.evaluate_population(pop)
# Assign distinct values to an attribute on each individual
attrs = [('foo', ['GREEN', 15, 'BLUE', 72.81]),
('bar', ['Colorless', 'green', 'ideas', 'sleep']),
('baz', [['a', 'b', 'c'], [1, 2, 3], [None, None, None],
[0.1, 0.2, 0.3]])]
for attr, vals in attrs:
for (ind, val) in zip(pop, vals):
ind.__dict__[attr] = val
return pop
def test_proportional_selection_custom_key():
''' Test of proportional selection with custom evaluation '''
pop = [
Individual(np.array([0, 0, 0]), problem=MaxOnes()),
Individual(np.array([1, 1, 1]), problem=MaxOnes())
]
def custom_key(individual):
''' Returns fitness based on MaxZeros '''
return np.count_nonzero(individual.genome == 0)
pop = Individual.evaluate_population(pop)
selector = ops.proportional_selection(pop, key=custom_key)
# we expect the first individual to always be selected
# since its genome is all 0s
selected = next(selector)
assert np.all(selected.genome == [0, 0, 0])
selected = next(selector)
assert np.all(selected.genome == [0, 0, 0])
def test_proportional_selection_pop_min():
''' Test of proportional selection with pop-min offset '''
# Create a population of positive fitness individuals
# scaling the fitness by the population minimum makes it so the
# least fit member never gets selected.
pop = [
Individual(np.array([0, 1, 0]), problem=MaxOnes()),
Individual(np.array([1, 1, 1]), problem=MaxOnes())
]
pop = Individual.evaluate_population(pop)
selector = ops.proportional_selection(pop, offset='pop-min')
# we expect that the second individual is always selected
# since the new zero point will be at the minimum fitness
# of the population
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
def test_mutate_bitflip():
# Create a very simple individual with two binary genes of all ones.
ind = [Individual(np.array([1, 1]), problem=MaxOnes())]
# Now mutate the individual such that we *expect both bits to bitflip*
mutated_ind = next(ops.mutate_bitflip(iter(ind), expected_num_mutations=2))
assert np.all(mutated_ind.genome == [0, 0])
# Of course, since we didn't clone the original, well, that actually got
# zapped, too.
assert np.all(ind[0].genome == [0, 0])
def test_sus_selection_num_points():
''' Test of SUS selection with varying `n` random points '''
# the second individual should always be selected
pop = [
Individual(np.array([0, 0, 0]), problem=MaxOnes()),
Individual(np.array([1, 1, 1]), problem=MaxOnes())
]
pop = Individual.evaluate_population(pop)
# with negative points
with pytest.raises(ValueError):
selector = ops.sus_selection(pop, n=-1)
selected = next(selector)
# with n = None (default)
selector = ops.sus_selection(pop, n=None)
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
# with n less than len(population)
selector = ops.sus_selection(pop, n=1)
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
# with n greater than len(population)
selector = ops.sus_selection(pop, n=3)
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
def test_sus_selection1():
''' Test of a deterministic case of stochastic universal sampling '''
# Make a population where sus_selection has an obvious
# reproducible choice
pop = [
Individual(np.array([0, 0, 0]), problem=MaxOnes()),
Individual(np.array([1, 1, 1]), problem=MaxOnes())
]
pop = Individual.evaluate_population(pop)
# This selection operator will always choose the [1, 1, 1] individual
# since [0, 0, 0] has zero fitness
selector = ops.sus_selection(pop)
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
# run one more time to test shuffle
selected = next(selector)
assert np.all(selected.genome == [1, 1, 1])
def test_good_eval():
"""
This is for testing a plain ole good individual to ensure that
leap_ec.distrib.evaluate works for normal circumstances.
"""
# set up a basic dask local cluster
with Client() as client:
# hand craft an individual that should evaluate fine
# Let's try evaluating a single individual
individual = Individual(np.array([1, 1]), problem=MaxOnes())
future = client.submit(evaluate(context=context), individual)
evaluated_individual = future.result()
assert evaluated_individual.fitness == 2
def test_clone():
# We need an encoder and problem to ensure those float across during
# clones.
decoder = IdentityDecoder()
problem = MaxOnes()
original = Individual([1, 1], decoder=decoder, problem=problem)
cloned = next(ops.clone(iter([original])))
assert original == cloned
# Yes, but did the other state make it across OK?
print(original.__dict__)
print(cloned.__dict__)
assert original.fitness == cloned.fitness
assert original.decoder == cloned.decoder
assert original.problem == cloned.problem
# use this when comparing complex objects with arrays
np.testing.assert_equal(original.__dict__, cloned.__dict__)
def test_lexical_maximization():
"""
Tests the lexical_parsimony() for maximization problems
"""
problem = MaxOnes()
# fitness=3, len(genome)=6
pop = [Individual(np.array([0, 0, 0, 1, 1, 1]), problem=problem)]
# fitness=2, len(genome)=2
pop.append(Individual(np.array([1, 1]), problem=problem))
# fitness=3, len(genome)=3
pop.append(Individual(np.array([1, 1, 1]), problem=problem))
pop = Individual.evaluate_population(pop)
best = ops.truncation_selection(pop, size=1, key=lexical_parsimony)
# prefers the shorter of the 3 genomes
assert np.all(best[0].genome == [1, 1, 1])
def test_truncation_selection_with_nan1():
"""If truncation selection encounters a NaN and non-NaN fitness
while maximizing, the non-NaN wins.
"""
# Make a population where binary tournament_selection has an obvious
# reproducible choice
problem = MaxOnes()
pop = [
Individual(np.array([0, 0, 0]), problem=problem),
Individual(np.array([1, 1, 1]), problem=problem)
]
# We first need to evaluate all the individuals so that truncation
# selection has fitnesses to compare
pop = Individual.evaluate_population(pop)
# Now set the "best" to NaN
pop[1].fitness = nan
best = ops.truncation_selection(pop, size=1)
assert pop[0] == best[0]
from leap_ec.individual import Individual
from leap_ec.decoder import IdentityDecoder
from leap_ec.global_vars import context
import leap_ec.ops as ops
from leap_ec.binary_rep.problems import MaxOnes
from leap_ec.binary_rep.initializers import create_binary_sequence
from leap_ec.binary_rep.ops import mutate_bitflip
from leap_ec import util
# create initial rand population of 5 individuals
parents = Individual.create_population(5,
initialize=create_binary_sequence(4),
decoder=IdentityDecoder(),
problem=MaxOnes())
# Evaluate initial population
parents = Individual.evaluate_population(parents)
# print initial, random population
util.print_population(parents, generation=0)
# generation_counter is an optional convenience for generation tracking
generation_counter = util.inc_generation(context=context)
while generation_counter.generation() < 6:
offspring = pipe(parents, ops.tournament_selection, ops.clone,
mutate_bitflip(expected_num_mutations=1),
ops.uniform_crossover(p_swap=0.2), ops.evaluate,
ops.pool(size=len(parents))) # accumulate offspring
