def test_create_sample_pop(self):
logger.debug("HEFT makespan {0}".format(heft(self.wf).makespan))
pop = generate_population(self.wf, size=25, rng=self.rng, skip_limit=5)
for soln in pop:
self.assertEqual(soln.execution_order[-1].task.aft, soln.makespan)
logger.debug("GA Initial Population")
logger.debug("########################")
for soln in pop:
logger.debug(("Execution order: {0}".format(soln.execution_order)))
logger.debug("Allocations: {0}".format(
soln.list_all_allocations()))
logger.debug("Makespan (s): {0}".format(
calculate_fitness(['time'], soln)))
logger.debug("Cost ($){0}".format(calculate_fitness(['cost'],
soln)))
soln.fitness = calculate_fitness(['time', 'cost'], soln)
fig, ax = plt.subplots()
ax.set_xlim([90, 200])
ax.set_ylim([100, 250])
x = [soln.fitness['time'] for soln in pop]
y = [soln.fitness['cost'] for soln in pop]
ax.scatter(x, y, c='red')
ax.set_axisbelow(True)
ax.legend()
ax.grid(True)
plt.xlabel('Solution Runtime')
plt.ylabel('Solution execution cost')
plt.show()
def ga(workflow,
objectives,
seed,
mutation_probability,
crossover_probability,
generations=100,
popsize=100):
"""
This function runs a standard genetic algorithm, with basic elitism, and
basic diversification. The algorithm:
* Generates a population of Solutions, based on the workflow provided
* For a given number of generations:
* Create a child population (newgen)
* While the newgen < pop
* Select parents using binary tournament seleciton
* Crossover the parents to children
* mutate the children
* calculate the fitness of each child
* add new children to newgen
* Keep as many children as there are that outperform the parents
:param mutation_probability:
:param crossover_probability:
:param workflow:
:param objectives:
:param seed:
:param generations:
:param popsize:
:return:
"""
pop = generate_population(workflow, seed, generations, popsize)
for soln in pop:
soln.fitness = fitness.calculate_fitness(objectives, soln)
# binary_tournament(pop)
generation = 0
limit = False
while generation < generations or limit:
newgen = []
while len(newgen) > len(pop):
p1 = binary_tournament(pop)
p2 = binary_tournament(pop)
c1, c2 = crossover(p1, p2, seed)
for c in [c1, c2]:
mutation(c)
fitness.calculate_fitness(c, objectives)
newgen.append(c)
generation += 1
continue
pass
def test_crossover(self):
pop = generate_population(self.wf, size=25, rng=self.rng, skip_limit=5)
for soln in pop:
soln.fitness = calculate_fitness(['time', 'cost'], soln)
random.seed(self.rng)
p1 = binary_tournament(pop)
self.assertSequenceEqual([0, 5, 3, 2, 1, 4, 7, 8, 6, 9],
[t.task.tid for t in p1.execution_order])
p2 = binary_tournament(pop)
self.assertSequenceEqual([0, 5, 4, 1, 3, 2, 8, 6, 7, 9],
[t.task.tid for t in p2.execution_order])
c1, c2 = crossover(p1, p2, self.wf)
p1order = [t.task.tid for t in p1.execution_order]
c1order = [t.task.tid for t in c1.execution_order]
self.assertSequenceEqual(p1order, c1order)
for m in c1.machines:
for allocation in c1.list_machine_allocations(m):
if allocation.task.tid == 4:
self.assertEqual('cat1_m1', allocation.machine.id)
for m in c2.machines:
for allocation in c2.list_machine_allocations(m):
if allocation.task.tid == 4:
self.assertEqual('cat0_m0', allocation.machine.id)
self.assertNotEqual(p2.makespan, c2.makespan)
fig, ax = plt.subplots()
c1.fitness = calculate_fitness(['time', 'cost'], c1)
c2.fitness = calculate_fitness(['time', 'cost'], c2)
crossx = [c1.fitness['time'], c2.fitness['time']]
crossy = [c1.fitness['cost'], c2.fitness['cost']]
ax.scatter(crossx, crossy, c='green')
x = [soln.fitness['time'] for soln in pop]
y = [soln.fitness['cost'] for soln in pop]
ax.grid(True)
ax.set_xlim([90, 200])
ax.set_ylim([100, 170])
ax.scatter(x, y, c='red')
ax.set_axisbelow(True)
selectedx = [p1.fitness['time'], p2.fitness['time']]
selectedy = [p1.fitness['cost'], p2.fitness['cost']]
ax.scatter(selectedx, selectedy, c='blue')
ax.legend()
plt.xlabel('Solution Runtime')
plt.ylabel('Solution execution cost')
plt.show()
def test_binary_tournament(self):
pop = generate_population(self.wf, size=25, rng=self.rng, skip_limit=5)
for soln in pop:
soln.fitness = calculate_fitness(['time', 'cost'], soln)
compare_prob = 0.5
parent1 = binary_tournament(pop, compare_prob, self.rng)
logger.debug(parent1.execution_order)
compare_prob = 0.7
parent2 = binary_tournament(pop, compare_prob, self.rng)
# parent2 = binary_tournament(pop, self.rng)
logger.debug(parent2.execution_order)
self.assertSequenceEqual([0, 5, 3, 4, 2, 1, 6, 8, 7, 9],
[t.task.tid for t in parent1.execution_order])
logger.debug("Fitness: {0}".format(parent1.fitness))
self.assertSequenceEqual([0, 5, 4, 1, 3, 2, 8, 6, 7, 9],
[t.task.tid for t in parent2.execution_order])
logger.debug("Fitness: {0}".format(parent2.fitness))
def test_mutation(self):
pop = generate_population(self.wf, size=25, rng=self.rng, skip_limit=5)
for soln in pop:
soln.fitness = calculate_fitness(['time', 'cost'], soln)
random.seed(self.rng)
p1 = binary_tournament(pop, self.rng, 0.6)
self.assertSequenceEqual([0, 5, 3, 2, 1, 4, 7, 8, 6, 9],
[t.task.tid for t in p1.execution_order])
mutated_child = mutation(p1, self.wf, 'swapping', rng=self.rng)
mutated_order_swapped = [0, 5, 3, 1, 7, 4, 2, 8, 6, 9]
self.assertSequenceEqual(
mutated_order_swapped,
[alloc.task.tid for alloc in mutated_child.execution_order])
selected_machine = None
for machine in mutated_child.machines:
if machine.id == 'cat2_m2':
selected_machine = machine
mutated_alloc = mutated_child.list_machine_allocations(
selected_machine)
self.assertSequenceEqual([7, 2, 9],
[alloc.task.tid for alloc in mutated_alloc])
def test_overall(self):
total_generations = 25
crossover_probability = 0.5
mutation_probability = 0.4
popsize = 25
pop = generate_population(self.wf,
size=popsize,
rng=self.rng,
skip_limit=5)
for soln in pop:
soln.fitness = calculate_fitness(['time', 'cost'], soln)
generations = []
x = [soln.fitness['time'] for soln in pop]
y = [soln.fitness['cost'] for soln in pop]
weights = [200 * i for i in Counter(x).values() for j in range(i)]
generations.append((x, y))
parents1 = []
parents2 = []
for gen in range(total_generations):
new_pop = []
parent1 = None
parent2 = None
while len(new_pop) < len(pop):
p1 = binary_tournament(pop, 0.5, self.rng)
p2 = binary_tournament(pop, 0.5, self.rng)
parent1 = p1
parent2 = p2
if random.random() < crossover_probability:
c1, c2 = crossover(p1, p2, self.wf)
new_pop.append(c1)
new_pop.append(c2)
elif random.random() < mutation_probability:
c1 = mutation(p1, self.wf, 'swapping', rng=self.rng)
if c1 is None:
# The mutation didn't occur due to selection issue
continue
new_pop.append(c1)
else:
new_pop.append(p1)
new_pop.append(p2)
continue
tmp_pop = pop + new_pop
for soln in tmp_pop:
soln.fitness = calculate_fitness(['time', 'cost'], soln)
# soln.total_fitness = soln.calc_total_fitness()
tmp_pop.sort(key=lambda x: (x.fitness['time'], x.fitness['cost']))
# tmp_pop.sort(key=lambda solution: solution.total_fitness)
pop = tmp_pop[0:popsize]
weights = [10 * i for i in Counter(x).values() for j in range(i)]
x = [soln.fitness['time'] for soln in pop]
y = [soln.fitness['cost'] for soln in pop]
parent1_x = [parent1.fitness['time']]
parent1_y = [parent1.fitness['cost']]
parent2_x = [parent2.fitness['time']]
parent2_y = [parent2.fitness['cost']]
generations.append((x, y))
parents1.append((parent1_x, parent1_y))
parents2.append((parent2_x, parent2_y))
plt.draw()
plt.show()
for soln in pop:
logger.info(soln.fitness)
fig, ax = plt.subplots()
ax.set_xlim([80, 250])
ax.set_ylim([100, 250])
scatter = ax.scatter(generations[0][0],
generations[0][1],
c='blue',
alpha=.2,
label='New Pop')
parent_scatter = ax.scatter(parents1[0][0],
parents1[0][1],
c='red',
alpha=0.8,
label='Parent 1')
parent2_scatter = ax.scatter(parents2[0][0],
parents1[0][1],
c='red',
alpha=0.8,
label='Parent 2')
ax.legend()
def animate(i):
scatter.set_offsets(np.c_[generations[i][0], generations[i][1]])
parent_scatter.set_offsets(np.c_[parents1[i][0], parents1[i][1]])
parent2_scatter.set_offsets(np.c_[parents2[i][0], parents2[i][1]])
ax.set_xlabel('Runtime (s) \n Generation {0}'.format(i))
ax.set_ylabel('Cost ($)')
anim = FuncAnimation(fig, animate, interval=100, frames=25)
plt.draw()
anim.save('filename.mp4', fps=1)