def test_mapping_and_ordering(self):
wf_name = "Montage_25"
wf = WF(wf_name)
manager = ExperimentResourceManager(ResourceGenerator.r([10, 15, 15, 25]))
estimator = ExperimentEstimator(None, 20, 1.0, transfer_time=100)
config = default_config(wf, manager, estimator)
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(wf, estimator, manager, solution)
for k, items in schedule.mapping.items():
for item in items:
item.state = ScheduleItem.FINISHED
## TODO: refactor this
ExecutorRunner.extract_result(schedule, True, wf)
config_2 = deepcopy(config)
config_2["interact_individuals_count"] = 4
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(wf, estimator, manager, solution)
for k, items in schedule.mapping.items():
for item in items:
item.state = ScheduleItem.FINISHED
## TODO: refactor this
ExecutorRunner.extract_result(schedule, True, wf)
pass
def test_mapping_and_ordering(self):
wf_name = "Montage_25"
wf = WF(wf_name)
manager = ExperimentResourceManager(
ResourceGenerator.r([10, 15, 15, 25]))
estimator = ExperimentEstimator(None, 20, 1.0, transfer_time=100)
config = default_config(wf, manager, estimator)
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(wf, estimator, manager, solution)
for k, items in schedule.mapping.items():
for item in items:
item.state = ScheduleItem.FINISHED
## TODO: refactor this
ExecutorRunner.extract_result(schedule, True, wf)
config_2 = deepcopy(config)
config_2["interact_individuals_count"] = 4
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(wf, estimator, manager, solution)
for k, items in schedule.mapping.items():
for item in items:
item.state = ScheduleItem.FINISHED
## TODO: refactor this
ExecutorRunner.extract_result(schedule, True, wf)
pass
def do_experiment(saver, config, _wf, rm, estimator):
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(_wf, estimator, rm, solution)
m = Utility.makespan(schedule)
data = {
"metainfo": {
"interact_individuals_count": config["interact_individuals_count"],
"generations": config["generations"],
"species": [s.name for s in config["species"]],
"pop_sizes": [s.pop_size for s in config["species"]],
"nodes": {n.name: n.flops
for n in config["env"].rm.get_nodes()},
"ideal_inds": {
MAPPING_SPECIE: ms_str_repr,
ORDERING_SPECIE: os_ideal_ind
},
"wf_name": _wf.name
},
"initial_pops": initial_pops,
"final_solution": solution,
"final_makespan": m,
"iterations": logbook
}
saver(data)
return m
def test_hall_of_fame(self):
wf_name = "Montage_25"
wf = WF(wf_name)
manager = ExperimentResourceManager(ResourceGenerator.r([10, 15, 15, 25]))
estimator = ExperimentEstimator(None, 20, 1.0, transfer_time=100)
config = default_config(wf, manager, estimator)
config["hall_of_fame_size"] = 3
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(wf, estimator, manager, solution)
for k, items in schedule.mapping.items():
for item in items:
item.state = ScheduleItem.FINISHED
## TODO: refactor this
ExecutorRunner.extract_result(schedule, True, wf)
pass
def test_hall_of_fame(self):
wf_name = "Montage_25"
wf = WF(wf_name)
manager = ExperimentResourceManager(
ResourceGenerator.r([10, 15, 15, 25]))
estimator = ExperimentEstimator(None, 20, 1.0, transfer_time=100)
config = default_config(wf, manager, estimator)
config["hall_of_fame_size"] = 3
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(wf, estimator, manager, solution)
for k, items in schedule.mapping.items():
for item in items:
item.state = ScheduleItem.FINISHED
## TODO: refactor this
ExecutorRunner.extract_result(schedule, True, wf)
pass
def test_predefined_init_pop(self):
## TODO: make loading from save file too
wf_name = "Montage_25"
_wf = WF(wf_name)
manager = ExperimentResourceManager(
ResourceGenerator.r([10, 15, 15, 25]))
estimator = ExperimentEstimator(None, 20, 1.0, transfer_time=100)
config = default_config(_wf, manager, estimator)
## TODO: need to hide low level details
import os
path = os.path.join(
TEMP_PATH,
"cga_exp_for_example/5760f488-932e-4224-942f-1f5ac68709bf.json")
with open(path, "r") as f:
data = json.load(f)
initial_pops = data["initial_pops"]
mps = [
ListBasedIndividual([tuple(g) for g in ind])
for ind in initial_pops[MAPPING_SPECIE]
]
os = [
ListBasedIndividual(ind) for ind in initial_pops[ORDERING_SPECIE]
]
for s in config["species"]:
if s.name == MAPPING_SPECIE:
s.initialize = lambda ctx, size: mps
elif s.name == ORDERING_SPECIE:
s.initialize = lambda ctx, size: os
else:
raise Exception("Unexpected specie: " + s.name)
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(_wf, estimator, manager, solution)
for k, items in schedule.mapping.items():
for item in items:
item.state = ScheduleItem.FINISHED
## TODO: refactor this
ExecutorRunner.extract_result(schedule, True, _wf)
pass
def do_exp():
config = {
"interact_individuals_count": 500,
"generations": 1000,
"env": Env(_wf, rm, estimator),
"species": [Specie(name=MAPPING_SPECIE, pop_size=500,
cxb=0.9, mb=0.9,
mate=lambda env, child1, child2: tools.cxOnePoint(child1, child2),
# mutate=mapping_default_mutate,
# mutate=lambda ctx, mutant: mapping_k_mutate(ctx, 3, mutant)
mutate=mapping_all_mutate,
# mutate=mapping_improving_mutation,
select=selector,
initialize=mapping_default_initialize,
# initialize=lambda ctx, pop: mapping_heft_based_initialize(ctx, pop, heft_mapping, 3),
stat=lambda pop: {"hamming_distances": hamming_distances([to_seq(p) for p in pop], to_seq(ms_ideal_ind)),
"unique_inds_count": unique_individuals(pop),
"pcm": pcm(pop),
"gdm": gdm(pop)}
),
Specie(name=ORDERING_SPECIE, fixed=True,
representative_individual=ListBasedIndividual(os_representative))
],
"solstat": lambda sols: {"best_components": hamming_for_best_components(sols, ms_ideal_ind, os_ideal_ind),
"best_components_itself": best_components_itself(sols),
"best": -1*Utility.makespan(build_schedule(_wf, estimator, rm, max(sols, key=lambda x: x.fitness)))
},
"operators": {
# "choose": default_choose,
"build_solutions": default_build_solutions,
# "fitness": fitness_mapping_and_ordering,
"fitness": overhead_fitness_mapping_and_ordering,
# "assign_credits": default_assign_credits
# "assign_credits": max_assign_credits
"assign_credits": assign_from_transfer_overhead
}
}
return do_experiment(saver, config, _wf, rm, estimator)
def test_fixed_specie(self):
wf_name = "Montage_25"
wf = WF(wf_name)
manager = ExperimentResourceManager(ResourceGenerator.r([10, 15, 15, 25]))
estimator = ExperimentEstimator(None, 20, 1.0, transfer_time=100)
config = default_config(wf, manager, estimator)
ms_ideal_ind = build_ms_ideal_ind(wf, manager)
config["species"][0] = Specie(name=MAPPING_SPECIE, fixed=True,
representative_individual=ListBasedIndividual(ms_ideal_ind))
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(wf, estimator, manager, solution)
for k, items in schedule.mapping.items():
for item in items:
item.state = ScheduleItem.FINISHED
## TODO: refactor this
ExecutorRunner.extract_result(schedule, True, wf)
pass
def test_predefined_init_pop(self):
## TODO: make loading from save file too
wf_name = "Montage_25"
_wf = WF(wf_name)
manager = ExperimentResourceManager(ResourceGenerator.r([10, 15, 15, 25]))
estimator = ExperimentEstimator(None, 20, 1.0, transfer_time=100)
config = default_config(_wf, manager, estimator)
## TODO: need to hide low level details
import os
path = os.path.join(TEMP_PATH, "cga_exp_for_example/5760f488-932e-4224-942f-1f5ac68709bf.json")
with open(path, "r") as f:
data = json.load(f)
initial_pops = data["initial_pops"]
mps = [ListBasedIndividual([tuple(g) for g in ind]) for ind in initial_pops[MAPPING_SPECIE]]
os = [ListBasedIndividual(ind) for ind in initial_pops[ORDERING_SPECIE]]
for s in config["species"]:
if s.name == MAPPING_SPECIE:
s.initialize = lambda ctx, size: mps
elif s.name == ORDERING_SPECIE:
s.initialize = lambda ctx, size: os
else:
raise Exception("Unexpected specie: " + s.name)
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(_wf, estimator, manager, solution)
for k, items in schedule.mapping.items():
for item in items:
item.state = ScheduleItem.FINISHED
## TODO: refactor this
ExecutorRunner.extract_result(schedule, True, _wf)
pass
def test_fixed_specie(self):
wf_name = "Montage_25"
wf = WF(wf_name)
manager = ExperimentResourceManager(
ResourceGenerator.r([10, 15, 15, 25]))
estimator = ExperimentEstimator(None, 20, 1.0, transfer_time=100)
config = default_config(wf, manager, estimator)
ms_ideal_ind = build_ms_ideal_ind(wf, manager)
config["species"][0] = Specie(
name=MAPPING_SPECIE,
fixed=True,
representative_individual=ListBasedIndividual(ms_ideal_ind))
solution, pops, logbook, initial_pops = run_cooperative_ga(**config)
schedule = build_schedule(wf, estimator, manager, solution)
for k, items in schedule.mapping.items():
for item in items:
item.state = ScheduleItem.FINISHED
## TODO: refactor this
ExecutorRunner.extract_result(schedule, True, wf)
pass
def cost_mapping_and_ordering(ctx, solution):
env = ctx['env']
schedule = build_schedule(env.wf, env.estimator, env.rm, solution)
all_cost = cost(schedule, env.estimator)
return -all_cost
stat=lambda pop: {
"hamming_distances": hamming_distances(pop, os_ideal_ind),
"unique_inds_count": unique_individuals(pop),
"pcm": pcm(pop),
"gdm": gdm(pop)
})
],
"solstat":
lambda sols: {
"best_components":
hamming_for_best_components(sols, ms_ideal_ind, os_ideal_ind),
"best_components_itself":
best_components_itself(sols),
"best":
-1 * Utility.makespan(
build_schedule(_wf, estimator, rm,
max(sols, key=lambda x: x.fitness)))
},
"analyzers": [mapping_mut_reg.analyze],
"operators": {
# "choose": default_choose,
# "build_solutions": default_build_solutions,
"build_solutions": one_to_one_build_solutions,
"fitness": fitness_mapping_and_ordering,
# "fitness": overhead_fitness_mapping_and_ordering,
# "assign_credits": default_assign_credits
# "assign_credits": bonus2_assign_credits
"assign_credits": max_assign_credits
}
}
),
Specie(name=ORDERING_SPECIE, pop_size=50,
cxb=0.8, mb=0.5,
mate=ordering_default_crossover,
mutate=ordering_default_mutate,
select=ordering_selector,
initialize=ordering_default_initialize,
stat=lambda pop: {"hamming_distances": hamming_distances(pop, os_ideal_ind),
"unique_inds_count": unique_individuals(pop),
"pcm": pcm(pop),
"gdm": gdm(pop)}
)
],
"solstat": lambda sols: {"best_components": hamming_for_best_components(sols, ms_ideal_ind, os_ideal_ind),
"best_components_itself": best_components_itself(sols),
"best": -1* Utility.makespan(build_schedule(_wf, estimator, rm, max(sols, key=lambda x: x.fitness)))
},
"operators": {
# "choose": default_choose,
"build_solutions": default_build_solutions,
# "fitness": fitness_mapping_and_ordering,
"fitness": overhead_fitness_mapping_and_ordering,
# "assign_credits": default_assign_credits
# "assign_credits": max_assign_credits
"assign_credits": assign_from_transfer_overhead
}
}
saver = UniqueNameSaver("../../temp/cga_heft_mixin")
def do_exp():
Specie(name=ORDERING_SPECIE, pop_size=50,
cxb=0.9, mb=0.9,
mate=ordering_default_crossover,
mutate=ordering_default_mutate,
select=ordering_selector,
# initialize=ordering_default_initialize,
initialize=lambda ctx, pop: ordering_heft_based_initialize(ctx, pop, heft_ordering, 3),
stat=lambda pop: {"hamming_distances": hamming_distances(pop, os_ideal_ind),
"unique_inds_count": unique_individuals(pop),
"pcm": pcm(pop),
"gdm": gdm(pop)}
)
],
"solstat": lambda sols: {"best_components": hamming_for_best_components(sols, ms_ideal_ind, os_ideal_ind),
"best_components_itself": best_components_itself(sols),
"best": -1*Utility.makespan(build_schedule(_wf, estimator, rm, max(sols, key=lambda x: x.fitness)))},
"analyzers": [mapping_mut_reg.analyze],
"use_credit_inheritance": True,
"operators": {
# "choose": default_choose,
# "build_solutions": default_build_solutions,
"build_solutions": one_to_one_build_solutions,
"fitness": fitness_mapping_and_ordering,
# "fitness": overhead_fitness_mapping_and_ordering,
# "assign_credits": default_assign_credits
# "assign_credits": bonus2_assign_credits
"assign_credits": max_assign_credits
}
def do_exp():
config = {
"interact_individuals_count":
100,
"generations":
300,
"env":
Env(_wf, rm, estimator),
"species": [
Specie(
name=MAPPING_SPECIE,
pop_size=50,
cxb=0.9,
mb=0.9,
mate=lambda env, child1, child2: tools.cxOnePoint(
child1, child2),
# mutate=mapping_default_mutate,
# mutate=lambda ctx, mutant: mapping_k_mutate(ctx, 3, mutant)
mutate=mapping_all_mutate,
# mutate=OnlyUniqueMutant()(mapping_all_mutate),
select=selector,
# initialize=mapping_default_initialize,
initialize=lambda ctx, pop: mapping_heft_based_initialize(
ctx, pop, heft_mapping, 3),
stat=lambda pop: {
"hamming_distances":
hamming_distances([to_seq(p)
for p in pop], to_seq(ms_ideal_ind)),
"unique_inds_count":
unique_individuals(pop),
"pcm":
pcm(pop),
"gdm":
gdm(pop)
}),
Specie(name=ORDERING_SPECIE,
fixed=True,
representative_individual=ListBasedIndividual(
os_representative))
],
"solstat":
lambda sols: {
"best_components":
hamming_for_best_components(sols, ms_ideal_ind, os_ideal_ind),
"best_components_itself":
best_components_itself(sols),
"best":
-1 * Utility.makespan(
build_schedule(_wf, estimator, rm,
max(sols, key=lambda x: x.fitness)))
},
"operators": {
# "choose": default_choose,
"build_solutions": default_build_solutions,
"fitness": fitness_mapping_and_ordering,
# "fitness": overhead_fitness_mapping_and_ordering,
# "assign_credits": default_assign_credits
# "assign_credits": max_assign_credits
"assign_credits": assign_from_transfer_overhead
}
}
return do_experiment(saver, config, _wf, rm, estimator)