def init():
computer = Computer("Zong's Computer")
admin = Admin("Zong", computer)
products = []
products.append(Product("14314", "锤子"))
products.append(Product("241432", "螺丝刀"))
products.append(Product("441413", "剪刀"))
products.append(Product("534134", "钳子"))
products.append(Product("534134", "挖掘机", expensive=True))
converger = Converger("永不罢工的传送带")
robots = []
robots.append(Robot("1号机器人", converger))
robots.append(Robot("2号机器人", converger))
robots.append(Robot("3号机器人", converger))
computer.addRobot(robots[0])
computer.addRobot(robots[1])
computer.addRobot(robots[2])
return TWS(admin, computer, robots,products, converger)
def __init__(self, name='', con=Converger()):
self.name = name
self.converger = con
self.state = 'Fine'
self.use = False
self.request_queue = Queue(maxsize=10)
def run(self):
"""
Run the optimization algorithm
@return: The history from the run
@rtype: History
"""
# Initialize the population
parent_pop = self.initialize()
# Evaluate the population
self.evaluate_population(parent_pop)
# Perform selection to get the crowding distances
parent_pop = self.selection(parent_pop, self._pop_size)
pareto_front = self.nondominated_sort(parent_pop, len(parent_pop), first_front_only=True)
self._converger = Converger(pareto_front[0], self._pop_size, self._true_pareto_front)
# Initialize the archive
if self._write_history is True:
self._history.add_population(parent_pop)
gen = 0
converged = 0
convergence_history = []
while (gen < self._generations) and converged < 10:
self._logger.info('Starting generation ' + repr(gen+1) + ' of ' + repr(self._generations))
# Apply crossover to generate a child population
child_pop = self.tournament_select(parent_pop, self._pop_size)[:]
# Apply crossover and mutation
for ind1, ind2 in zip(child_pop[::2], child_pop[1::2]):
if self._rnd.random() <= self._p_cross:
self.mate(ind1, ind2)
self.mutate(ind1)
self.mutate(ind2)
# Evaluate the child population
self.evaluate_population(child_pop)
# Update ID's
for ind in child_pop:
ind.id += self._pop_size
# Perform NSGA-2 selection to pick the next generation
parent_pop = self.selection(parent_pop + child_pop, self._pop_size)
pareto_front = self.nondominated_sort(parent_pop, len(parent_pop), first_front_only=True)
self._history.add_population(parent_pop)
self._logger.info('Pareto Front Size: ' + repr(len(pareto_front[0])))
# Check convergence
convergence_metrics = \
self._converger.check_convergence(pareto_front[0], self._pop_size)
if self._write_history is True:
convergence_history.append(convergence_metrics)
self._logger.info('Frontier Goodness: ' + repr(convergence_metrics.fg))
self._logger.info('Expansion Metric: ' + repr(convergence_metrics.expansion))
self._logger.info('Density Metric: ' + repr(convergence_metrics.dm))
self._logger.info('Distance Metric: ' + repr(convergence_metrics.distance))
self._logger.info('Volume Ratio: ' + repr(convergence_metrics.volume_ratio))
if self._conv_tol != 0:
if max([convergence_metrics.fg, convergence_metrics.expansion, convergence_metrics.dm]) <= \
self._conv_tol:
converged += 1
self._logger.info('Converged Generations: ' + repr(converged))
else:
converged = 0
# Update
gen += 1
# Final solution
solution = self.nondominated_sort(parent_pop, len(parent_pop), first_front_only=True)
return self._history, convergence_history, solution[0]
class Serial(AlgorithmBase):
"""
The serial implementation of NSGA-II
"""
def __init__(self):
"""
Initialize the Serial object
@return: None
"""
super(Serial, self).__init__()
def initialize(self):
"""
Randomly initialize a population
@return: The initialize population
@rtype: list(Individual)
"""
self._logger.info('Initializing the population')
pop = []
# Randomly initialize the population
for i in xrange(self._pop_size):
ind = Individual()
ind.id = i
x_set = [None]*self._ndim
for j in xrange(self._ndim):
x_set[j] = self._rnd.uniform(self._lower_bound[j], self._upper_bound[j])
ind.x = x_set
pop.append(ind)
return pop
def run(self):
"""
Run the optimization algorithm
@return: The history from the run
@rtype: History
"""
# Initialize the population
parent_pop = self.initialize()
# Evaluate the population
self.evaluate_population(parent_pop)
# Perform selection to get the crowding distances
parent_pop = self.selection(parent_pop, self._pop_size)
pareto_front = self.nondominated_sort(parent_pop, len(parent_pop), first_front_only=True)
self._converger = Converger(pareto_front[0], self._pop_size, self._true_pareto_front)
# Initialize the archive
if self._write_history is True:
self._history.add_population(parent_pop)
gen = 0
converged = 0
convergence_history = []
while (gen < self._generations) and converged < 10:
self._logger.info('Starting generation ' + repr(gen+1) + ' of ' + repr(self._generations))
# Apply crossover to generate a child population
child_pop = self.tournament_select(parent_pop, self._pop_size)[:]
# Apply crossover and mutation
for ind1, ind2 in zip(child_pop[::2], child_pop[1::2]):
if self._rnd.random() <= self._p_cross:
self.mate(ind1, ind2)
self.mutate(ind1)
self.mutate(ind2)
# Evaluate the child population
self.evaluate_population(child_pop)
# Update ID's
for ind in child_pop:
ind.id += self._pop_size
# Perform NSGA-2 selection to pick the next generation
parent_pop = self.selection(parent_pop + child_pop, self._pop_size)
pareto_front = self.nondominated_sort(parent_pop, len(parent_pop), first_front_only=True)
self._history.add_population(parent_pop)
self._logger.info('Pareto Front Size: ' + repr(len(pareto_front[0])))
# Check convergence
convergence_metrics = \
self._converger.check_convergence(pareto_front[0], self._pop_size)
if self._write_history is True:
convergence_history.append(convergence_metrics)
self._logger.info('Frontier Goodness: ' + repr(convergence_metrics.fg))
self._logger.info('Expansion Metric: ' + repr(convergence_metrics.expansion))
self._logger.info('Density Metric: ' + repr(convergence_metrics.dm))
self._logger.info('Distance Metric: ' + repr(convergence_metrics.distance))
self._logger.info('Volume Ratio: ' + repr(convergence_metrics.volume_ratio))
if self._conv_tol != 0:
if max([convergence_metrics.fg, convergence_metrics.expansion, convergence_metrics.dm]) <= \
self._conv_tol:
converged += 1
self._logger.info('Converged Generations: ' + repr(converged))
else:
converged = 0
# Update
gen += 1
# Final solution
solution = self.nondominated_sort(parent_pop, len(parent_pop), first_front_only=True)
return self._history, convergence_history, solution[0]
class Islands(AlgorithmBase):
"""
The Islands optimization algorithm.
"""
def __init__(self):
"""
Initialize the Islands algorithm.
@return: None
"""
super(Islands, self).__init__()
def initialize(self):
"""
Randomly initialize the island populations
@return: The initialized islands
@rtype: list(list(Individual))
"""
self._logger.info('Initializing the islands')
islands = []
# TODO (JLD): Use _global_popsize where appropriate
self._global_popsize = self._islands * self._pop_size
# Randomly initialize each sub-population
for k in xrange(self._islands):
pop = []
for i in xrange(self._pop_size):
ind = Individual()
ind.id = i
x_set = [None]*self._ndim
for j in xrange(self._ndim):
x_set[j] = self._rnd.uniform(self._lower_bound[j], self._upper_bound[j])
ind.x = x_set
pop.append(ind)
islands.append(pop)
return islands
def run(self):
"""
Run the optimization algorithm
@return: The history from the run
@rtype: History
"""
# Initialize the islands
islands = self.initialize()
# Evaluate the individuals
for island in islands:
self.evaluate_population(island)
# Perform selection to get the crowding distances
parent_islands = []
pareto_fronts = []
for island in islands:
parent_pop = self.selection(island, self._pop_size)
parent_islands.append(parent_pop)
# Get the global pareto front
pareto_front = self.nondominated_sort(self.coalesce_populations(parent_islands),
len(parent_islands)*self._pop_size, first_front_only=True)
self._converger = Converger(pareto_front[0], len(pareto_front))
# Initialize the archive
self._history.add_population(parent_islands)
gen = 0
converged = 0
# TODO (JLD): Make epoch a configuration property
epoch = 5
# TODO (JLD): Make number_of_migrants a configuration property
number_of_migrants = 3
while (gen < self._generations) and converged < 10:
self._logger.info('Starting generation ' + repr(gen+1) + ' of ' + repr(self._generations))
# Migration
if (gen % epoch) == 0:
self._logger.info('Epoch occurred, migrating...')
self.migration(parent_islands, number_of_migrants)
# Apply crossover to generate a child population
child_islands = []
for parent_pop in parent_islands:
child_pop = self.tournament_select(parent_pop, self._pop_size)
child_islands.append(child_pop)
# Apply crossover and mutation
for child_pop in child_islands:
for ind1, ind2 in zip(child_pop[::2], child_pop[1::2]):
if self._rnd.random() <= self._p_cross:
self.mate(ind1, ind2)
self.mutate(ind1)
self.mutate(ind2)
# Evaluate the child population
for child_pop in child_islands:
self.evaluate_population(child_pop)
# Update ID's
for child_pop in child_islands:
for ind in child_pop:
ind.id += self._pop_size
# Perform NSGA-2 selection to pick the next generation
for parent_pop, child_pop in zip(parent_islands, child_islands):
parent_pop = self.selection(parent_pop + child_pop, self._pop_size)
pareto_front = self.nondominated_sort(self.coalesce_populations(parent_islands),
len(parent_islands)*self._pop_size, first_front_only=True)
for parent_pop in parent_islands:
self._history.add_population(parent_pop)
convergence_metrics = self._converger.check_convergence(pareto_front[0], self._pop_size*self._islands)
self._logger.info('Frontier Goodness: ' + repr(convergence_metrics.fg))
self._logger.info('Expansion Metric: ' + repr(convergence_metrics.expansion))
self._logger.info('Density Metric: ' + repr(convergence_metrics.dm))
if max([convergence_metrics.fg, convergence_metrics.expansion, convergence_metrics.dm]) <= self._conv_tol:
converged += 1
self._logger.info('Converged Generations: ' + repr(converged))
else:
converged = 0
# Update
gen += 1
# Final solution
solution = []
for parent_pop in parent_islands:
for ind in parent_pop:
solution.append(ind)
# TODO (JLD): Add convergence history, and solution to outputs
return self._history,
# TODO (JLD): Add in different migration schemes
def migration(self, islands, number_of_migrants):
"""
Perform the migration operation in place on the sub populations
@param islands: The sub populations to include in the migration
@type islands: list(list(Individual))
@param number_of_migrants: The number of migrants
@type number_of_migrants: int
@return: None
"""
# Generate a migration pool, currently just selecting a random individual
migrant_pool = []
for island in islands:
for i in xrange(number_of_migrants):
random_index = self._rnd.randrange(0, len(island))
migrant = island.pop(random_index)
migrant_pool.append(migrant)
# assign the migrants to an island
for island in islands:
for i in xrange(number_of_migrants):
random_index = self._rnd.randrange(0, len(migrant_pool))
immigrant = migrant_pool.pop(random_index)
island.append(immigrant)
def run(self):
"""
Run the optimization algorithm
@return: The history from the run
@rtype: History
"""
# Initialize the islands
islands = self.initialize()
# Evaluate the individuals
for island in islands:
self.evaluate_population(island)
# Perform selection to get the crowding distances
parent_islands = []
pareto_fronts = []
for island in islands:
parent_pop = self.selection(island, self._pop_size)
parent_islands.append(parent_pop)
# Get the global pareto front
pareto_front = self.nondominated_sort(self.coalesce_populations(parent_islands),
len(parent_islands)*self._pop_size, first_front_only=True)
self._converger = Converger(pareto_front[0], len(pareto_front))
# Initialize the archive
self._history.add_population(parent_islands)
gen = 0
converged = 0
# TODO (JLD): Make epoch a configuration property
epoch = 5
# TODO (JLD): Make number_of_migrants a configuration property
number_of_migrants = 3
while (gen < self._generations) and converged < 10:
self._logger.info('Starting generation ' + repr(gen+1) + ' of ' + repr(self._generations))
# Migration
if (gen % epoch) == 0:
self._logger.info('Epoch occurred, migrating...')
self.migration(parent_islands, number_of_migrants)
# Apply crossover to generate a child population
child_islands = []
for parent_pop in parent_islands:
child_pop = self.tournament_select(parent_pop, self._pop_size)
child_islands.append(child_pop)
# Apply crossover and mutation
for child_pop in child_islands:
for ind1, ind2 in zip(child_pop[::2], child_pop[1::2]):
if self._rnd.random() <= self._p_cross:
self.mate(ind1, ind2)
self.mutate(ind1)
self.mutate(ind2)
# Evaluate the child population
for child_pop in child_islands:
self.evaluate_population(child_pop)
# Update ID's
for child_pop in child_islands:
for ind in child_pop:
ind.id += self._pop_size
# Perform NSGA-2 selection to pick the next generation
for parent_pop, child_pop in zip(parent_islands, child_islands):
parent_pop = self.selection(parent_pop + child_pop, self._pop_size)
pareto_front = self.nondominated_sort(self.coalesce_populations(parent_islands),
len(parent_islands)*self._pop_size, first_front_only=True)
for parent_pop in parent_islands:
self._history.add_population(parent_pop)
convergence_metrics = self._converger.check_convergence(pareto_front[0], self._pop_size*self._islands)
self._logger.info('Frontier Goodness: ' + repr(convergence_metrics.fg))
self._logger.info('Expansion Metric: ' + repr(convergence_metrics.expansion))
self._logger.info('Density Metric: ' + repr(convergence_metrics.dm))
if max([convergence_metrics.fg, convergence_metrics.expansion, convergence_metrics.dm]) <= self._conv_tol:
converged += 1
self._logger.info('Converged Generations: ' + repr(converged))
else:
converged = 0
# Update
gen += 1
# Final solution
solution = []
for parent_pop in parent_islands:
for ind in parent_pop:
solution.append(ind)
# TODO (JLD): Add convergence history, and solution to outputs
return self._history,
from converger import Converger
#convergence in 'total energy', 'geometry' or 'single phonon' calculations
#can be studied with respect to the following variables:
# 'nbnd' : 'system',
# 'degauss' : 'system',
# 'ecutwfc' : 'system',
# 'ecutrho' : 'system',
# 'conv_thr' : 'electrons',
# 'etot_conv_thr': 'control',
# 'forc_conv_thr': 'control',
# 'path_thr' : 'ions',
# 'kpoints' : 'k_points'
opt = Converger('config.ini', 'total energy', tolerance=0.1)
conv_thr = opt.converge(what='conv_thr', startValue=1e-4, multiply=0.1)
ecut = opt.converge(what='ecutwfc', startValue=18, step=4)
ecutrho = opt.converge('ecutrho', ecut * 4, 16)
#opt.converge('kpoints',[12,12,12],[2,2,2])
#print opt.getForces()
#print opt.getStress()
from converger import Converger
#convergence in 'total energy', 'geometry' or 'single phonon' calculations
#can be studied with respect to the following variables:
# 'nbnd' : 'system',
# 'degauss' : 'system',
# 'ecutwfc' : 'system',
# 'ecutrho' : 'system',
# 'conv_thr' : 'electrons',
# 'etot_conv_thr': 'control',
# 'forc_conv_thr': 'control',
# 'path_thr' : 'ions',
# 'kpoints' : 'k_points'
opt = Converger('config.ini','total energy', tolerance = 0.1)
conv_thr = opt.converge(what = 'conv_thr', startValue = 1e-4, multiply = 0.1)
ecut = opt.converge(what = 'ecutwfc', startValue = 18, step = 4)
ecutrho = opt.converge('ecutrho', ecut*4, 16)
#opt.converge('kpoints',[12,12,12],[2,2,2])
#print opt.getForces()
#print opt.getStress()
