def main() -> None:
if len(sys.argv) < 4:
print("Usage: python main_minimal.py <seed> <config-file> "
"<num-generations> <tsp-instance-file>")
sys.exit(1)
########################################
# Read the command-line arguments and the instance
########################################
seed = int(sys.argv[1])
configuration_file = sys.argv[2]
num_generations = int(sys.argv[3])
instance_file = sys.argv[4]
print("Reading data...")
instance = TSPInstance(instance_file)
########################################
# Read algorithm parameters
########################################
print("Reading parameters...")
brkga_params, _ = load_configuration(configuration_file)
########################################
# Build the BRKGA data structures and initialize
########################################
print("Building BRKGA data and initializing...")
decoder = TSPDecoder(instance)
brkga = BrkgaMpIpr(decoder=decoder,
sense=Sense.MINIMIZE,
seed=seed,
chromosome_size=instance.num_nodes,
params=brkga_params)
# NOTE: don't forget to initialize the algorithm.
brkga.initialize()
########################################
# Find good solutions / evolve
########################################
print(f"Evolving {num_generations} generations...")
brkga.evolve(num_generations)
best_cost = brkga.get_best_fitness()
print(f"Best cost: {best_cost}")
def test_evolve(self):
"""
Tests evolve() method.
"""
param_values = deepcopy(self.default_param_values)
brkga_params = param_values["params"]
brkga = BrkgaMpIpr(**param_values)
# Not initialized
with self.assertRaises(RuntimeError) as context:
brkga.evolve()
self.assertEqual(
str(context.exception).strip(),
"The algorithm hasn't been initialized. "
"Call 'initialize()' before 'evolve()'")
brkga.initialize()
with self.assertRaises(ValueError) as context:
brkga.evolve(-1)
self.assertEqual(
str(context.exception).strip(),
"Number of generations must be large than one. "
"Given -1")
with open(os.path.join(STATE_DIR, "state5.pickle"), "rb") as hd:
brkga = pickle.load(hd)
with open(os.path.join(SOLUTION_DIR, "best_solution5.pickle"),
"rb") as hd:
results = pickle.load(hd)
brkga.evolve()
self.assertEqual(brkga.get_best_fitness(), results["fitness1"])
self.assertEqual(brkga.get_best_chromosome(), results["chromosome1"])
print(f"Elapsed time: {time() - self.start_time :.2f}")
brkga.evolve(10)
self.assertEqual(brkga.get_best_fitness(), results["fitness10"])
self.assertEqual(brkga.get_best_chromosome(), results["chromosome10"])
print(f"Elapsed time: {time() - self.start_time :.2f}")
brkga.evolve(100)
self.assertEqual(brkga.get_best_fitness(), results["fitness100"])
self.assertEqual(brkga.get_best_chromosome(), results["chromosome100"])
print(f"Elapsed time: {time() - self.start_time :.2f}")
def main() -> None:
"""
Proceeds with the optimization. Create to avoid spread `global` keywords
around the code.
"""
args = docopt.docopt(__doc__)
# print(args)
configuration_file = args["--config_file"]
instance_file = args["--instance_file"]
seed = int(args["--seed"])
stop_rule = StopRule(args["--stop_rule"])
if stop_rule == StopRule.TARGET:
stop_argument = float(args["--stop_arg"])
else:
stop_argument = int(args["--stop_arg"])
maximum_time = float(args["--max_time"])
if maximum_time <= 0.0:
raise RuntimeError(f"Maximum time must be larger than 0.0. "
f"Given {maximum_time}.")
perform_evolution = not args["--no_evolution"]
########################################
# Load config file and show basic info.
########################################
brkga_params, control_params = load_configuration(configuration_file)
print(f"""------------------------------------------------------
> Experiment started at {datetime.now()}
> Instance: {instance_file}
> Configuration: {configuration_file}
> Algorithm Parameters:""",
end="")
if not perform_evolution:
print("> - Simple multi-start: on (no evolutionary operators)")
else:
output_string = ""
for name, value in vars(brkga_params).items():
output_string += f"\n> -{name} {value}"
for name, value in vars(control_params).items():
output_string += f"\n> -{name} {value}"
print(output_string)
print(f"""> Seed: {seed}
> Stop rule: {stop_rule}
> Stop argument: {stop_argument}
> Maximum time (s): {maximum_time}
------------------------------------------------------""")
########################################
# Load instance and adjust BRKGA parameters
########################################
print(f"\n[{datetime.now()}] Reading TSP data...")
instance = TSPInstance(instance_file)
print(f"Number of nodes: {instance.num_nodes}")
print(f"\n[{datetime.now()}] Generating initial tour...")
# Generate a greedy solution to be used as warm start for BRKGA.
initial_cost, initial_tour = greedy_tour(instance)
print(f"Initial cost: {initial_cost}")
########################################
# Build the BRKGA data structures and initialize
########################################
print(f"\n[{datetime.now()}] Building BRKGA data...")
# Usually, it is a good idea to set the population size
# proportional to the instance size.
brkga_params.population_size = min(brkga_params.population_size,
10 * instance.num_nodes)
print(f"New population size: {brkga_params.population_size}")
# Build a decoder object.
decoder = TSPDecoder(instance)
# Chromosome size is the number of nodes.
# Each chromosome represents a permutation of nodes.
brkga = BrkgaMpIpr(decoder=decoder,
sense=Sense.MINIMIZE,
seed=seed,
chromosome_size=instance.num_nodes,
params=brkga_params,
evolutionary_mechanism_on=perform_evolution)
# To inject the initial tour, we need to create chromosome representing
# that solution. First, we create a set of keys to be used in the
# chromosome.
random.seed(seed)
keys = sorted([random.random() for _ in range(instance.num_nodes)])
# Then, we visit each node in the tour and assign to it a key.
initial_chromosome = [0] * instance.num_nodes
for i in range(instance.num_nodes):
initial_chromosome[initial_tour[i]] = keys[i]
# Inject the warm start solution in the initial population.
brkga.set_initial_population([initial_chromosome])
# NOTE: don't forget to initialize the algorithm.
print(f"\n[{datetime.now()}] Initializing BRKGA data...")
brkga.initialize()
########################################
# Warm up the script/code
########################################
# To make sure we are timing the runs correctly, we run some warmup
# iterations with bogus data. Warmup is always recommended for script
# languages. Here, we call the most used methods.
print(f"\n[{datetime.now()}] Warming up...")
bogus_alg = deepcopy(brkga)
bogus_alg.evolve(2)
# TODO (ceandrade): warm up path relink functions.
# bogus_alg.path_relink(brkga_params.pr_type, brkga_params.pr_selection,
# (x, y) -> 1.0, (x, y) -> true, 0, 0.5, 1, 10.0, 1.0)
bogus_alg.get_best_fitness()
bogus_alg.get_best_chromosome()
bogus_alg = None
########################################
# Evolving
########################################
print(f"\n[{datetime.now()}] Evolving...")
print("* Iteration | Cost | CurrentTime")
best_cost = initial_cost * 2
best_chromosome = initial_chromosome
iteration = 0
last_update_time = 0.0
last_update_iteration = 0
large_offset = 0
# TODO (ceandrade): enable the following when path relink is ready.
# path_relink_time = 0.0
# num_path_relink_calls = 0
# num_homogenities = 0
# num_best_improvements = 0
# num_elite_improvements = 0
run = True
# Main optimization loop. We evolve one generation at time,
# keeping track of all changes during such process.
start_time = time.time()
while run:
iteration += 1
# Evolves one iteration.
brkga.evolve()
# Checks the current results and holds the best.
fitness = brkga.get_best_fitness()
if fitness < best_cost:
last_update_time = time.time() - start_time
update_offset = iteration - last_update_iteration
if large_offset < update_offset:
large_offset = update_offset
last_update_iteration = iteration
best_cost = fitness
best_chromosome = brkga.get_best_chromosome()
print(f"* {iteration} | {best_cost:.0f} | {last_update_time:.2f}")
# end if
# TODO (ceandrade): implement path relink calls here.
# Please, see Julia version for that.
iter_without_improvement = iteration - last_update_iteration
# Check stop criteria.
run = not (
(time.time() - start_time > maximum_time) or
(stop_rule == StopRule.GENERATIONS and iteration == stop_argument)
or (stop_rule == StopRule.IMPROVEMENT
and iter_without_improvement >= stop_argument) or
(stop_rule == StopRule.TARGET and best_cost <= stop_argument))
# end while
total_elapsed_time = time.time() - start_time
total_num_iterations = iteration
print(f"[{datetime.now()}] End of optimization\n")
print(f"Total number of iterations: {total_num_iterations}")
print(f"Last update iteration: {last_update_iteration}")
print(f"Total optimization time: {total_elapsed_time:.2f}")
print(f"Last update time: {last_update_time:.2f}")
print(f"Large number of iterations between improvements: {large_offset}")
# TODO (ceandrade): enable when path relink is ready.
# print(f"\nTotal path relink time: {path_relink_time:.2f}")
# print(f"\nTotal path relink calls: {num_path_relink_calls}")
# print(f"\nNumber of homogenities: {num_homogenities}")
# print(f"\nImprovements in the elite set: {num_elite_improvements}")
# print(f"\nBest individual improvements: {num_best_improvements}")
########################################
# Extracting the best tour
########################################
tour = []
for (index, key) in enumerate(best_chromosome):
tour.append((key, index))
tour.sort()
print(f"\n% Best tour cost: {best_cost:.2f}")
print("% Best tour: ", end="")
for _, node in tour:
print(node, end=" ")
print(
"\n\nInstance,Seed,NumNodes,TotalIterations,TotalTime,"
#"TotalPRTime,PRCalls,NumHomogenities,NumPRImprovElite,"
#"NumPrImprovBest,"
"LargeOffset,LastUpdateIteration,LastUpdateTime,"
"Cost")
print(
f"{basename(instance_file)},"
f"{seed},{instance.num_nodes},{total_num_iterations},"
f"{total_elapsed_time:.2f},"
# f"{path_relink_time:.2f},{num_path_relink_calls},"
# f"{num_homogenities},{num_elite_improvements},{num_best_improvements},"
f"{large_offset},{last_update_iteration},"
f"{last_update_time:.2f},{best_cost:.0f}")
def gen_conf5():
param_values = deepcopy(default_param_values)
brkga_params = param_values["params"]
chromosome_size = 100
instance = Instance(chromosome_size)
brkga_params.population_size = 100
brkga_params.elite_percentage = 0.30
brkga_params.mutants_percentage = 0.20
brkga_params.num_elite_parents = 2
brkga_params.total_parents = 3
brkga_params.bias_type = BiasFunctionType.LOGINVERSE
brkga_params.num_independent_populations = 3
brkga_params.pr_number_pairs = 0
brkga_params.pr_minimum_distance = 0.0
brkga_params.pr_type = PathRelinkingType.DIRECT
brkga_params.pr_selection = PathRelinkingSelection.BESTSOLUTION
brkga_params.alpha_block_size = 1.0
brkga_params.pr_percentage = 1.0
param_values["decoder"] = SumDecode(instance)
param_values["sense"] = Sense.MINIMIZE
param_values["seed"] = 4659930950303
param_values["chromosome_size"] = chromosome_size
param_values["evolutionary_mechanism_on"] = True
print("\n> Building configuration 5")
brkga = BrkgaMpIpr(**param_values)
brkga.initialize()
print("> Writing configuration 5")
with open(os.path.join(STATE_DIR, "state5.pickle"), "wb") as hd:
pickle.dump(brkga, hd)
print("> Evolving one generation...")
brkga.evolve()
fitness1 = brkga.get_best_fitness()
chromosome1 = brkga.get_best_chromosome()
print("> Evolving 10 generations...")
brkga.evolve(10)
fitness10 = brkga.get_best_fitness()
chromosome10 = brkga.get_best_chromosome()
print("> Evolving 100 generations...")
brkga.evolve(100)
fitness100 = brkga.get_best_fitness()
chromosome100 = brkga.get_best_chromosome()
with open(os.path.join(SOLUTION_DIR, "best_solution5.pickle"), "wb") as hd:
pickle.dump(
{
"fitness1": fitness1,
"chromosome1": chromosome1,
"fitness10": fitness10,
"chromosome10": chromosome10,
"fitness100": fitness100,
"chromosome100": chromosome100
},
hd
)
print("fitness", fitness1)
print("fitness", fitness10)
print("fitness", fitness100)
