def __init__(self, genome, owner, seed=None, interactiveMode=True):
""" Initializator of GSimpleGA """
if seed: random.seed(seed)
if type(interactiveMode) != BooleanType:
Util.raiseException(
"Interactive Mode option must be True or False", TypeError)
if not isinstance(genome, GenomeBase):
Util.raiseException("The genome must be a GenomeBase subclass",
TypeError)
self.internalPop = GPopulation(genome)
self.nGenerations = Consts.CDefGAGenerations
self.pMutation = Consts.CDefGAMutationRate
self.pCrossover = Consts.CDefGACrossoverRate
self.nElitismReplacement = Consts.CDefGAElitismReplacement
self.setPopulationSize(Consts.CDefGAPopulationSize)
self.minimax = Consts.minimaxType["maximize"]
self.elitism = True
self.owner = owner ## added 12/15 by Peter Graf, so GA can evaluate constraints
## and, now (4/16), so we can save let the owner save the state
# Adapters
self.dbAdapter = None
self.migrationAdapter = None
self.time_init = None
self.interactiveMode = interactiveMode
self.interactiveGen = -1
self.GPMode = False
self.selector = FunctionSlot("Selector")
self.stepCallback = FunctionSlot("Generation Step Callback")
self.terminationCriteria = FunctionSlot("Termination Criteria")
self.selector.set(Consts.CDefGASelector)
self.allSlots = [
self.selector, self.stepCallback, self.terminationCriteria
]
self.internalParams = {}
self.currentGeneration = 0
# GP Testing
for classes in Consts.CDefGPGenomes:
if isinstance(self.internalPop.oneSelfGenome, classes):
self.setGPMode(True)
break
logging.debug("A GA Engine was created, nGenerations=%d",
self.nGenerations)
def __init__(self, genome, seed=None, interactiveMode=True):
""" Initializator of GSimpleGA """
if seed: random.seed(seed)
if type(interactiveMode) != BooleanType:
Util.raiseException("Interactive Mode option must be True or False", TypeError)
if not isinstance(genome, GenomeBase):
Util.raiseException("The genome must be a GenomeBase subclass", TypeError)
self.internalPop = GPopulation(genome)
self.nGenerations = Consts.CDefGAGenerations
self.pMutation = Consts.CDefGAMutationRate
self.pCrossover = Consts.CDefGACrossoverRate
self.nElitismReplacement = Consts.CDefGAElitismReplacement
self.setPopulationSize(Consts.CDefGAPopulationSize)
self.minimax = Consts.minimaxType["maximize"]
self.elitism = True
# Adapters
self.dbAdapter = None
self.migrationAdapter = None
self.time_init = None
self.interactiveMode = interactiveMode
self.interactiveGen = -1
self.GPMode = False
self.selector = FunctionSlot("Selector")
self.stepCallback = FunctionSlot("Generation Step Callback")
self.terminationCriteria = FunctionSlot("Termination Criteria")
self.selector.set(Consts.CDefGASelector)
self.allSlots = [ self.selector, self.stepCallback, self.terminationCriteria ]
self.internalParams = {}
####
#####
self.currentGeneration = 0
# GP Testing
for classes in Consts.CDefGPGenomes:
if isinstance(self.internalPop.oneSelfGenome, classes):
self.setGPMode(True)
break
logging.debug("A GA Engine was created, nGenerations=%d", self.nGenerations)
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
popsize = len(self.internalPop)
numAdded = 0
maxTries = 1000
numTries = 0
crossover_empty = self.select(
popID=self.currentGeneration).crossover.isEmpty()
###TODO: enforce constraints!###
while numAdded < popsize:
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=2):
(sister, brother) = it
else:
if not crossover_empty and Util.randomFlipCoin(
self.pCrossover):
for it in genomeMom.crossover.applyFunctions(
mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
# logging.debug("done cloning")
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
if (numTries > maxTries
or self.owner.eval_constraints(sister)):
newPop.internalPop.append(sister)
numAdded += 1
print "successfully added sister"
if (numAdded < popsize
and (numTries > maxTries
or self.owner.eval_constraints(brother))):
newPop.internalPop.append(brother)
print "successfully added brother"
numAdded += 1
numTries += 1
#end rank0 onlye
# print "rank %d start eval pop" % MPI.COMM_WORLD.Get_rank()
logging.debug("Evaluating the newly created population.")
newPop.evaluate()
# print "rank %d done eval pop" % MPI.COMM_WORLD.Get_rank()
# if (MPI.COMM_WORLD.Get_rank() == 0):
# print "after eval, new pop's positions are:"
# for p in newPop:
# print p.wt_positions
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
if self.elitism:
logging.debug("Doing elitism, %d" % self.nElitismReplacement)
if self.getMinimax() == Consts.minimaxType["maximize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
# self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score > newPop.bestRaw(
i).score:
newPop[len(newPop) - 1 -
i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
# self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score < newPop.bestRaw(
i).score:
newPop[len(newPop) - 1 -
i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
self.internalPop.sort()
# if (MPI.COMM_WORLD.Get_rank() == 0):
# print "after sort, internal pop's positions are:"
# for p in self.internalPop:
# print p.wt_positions
logging.debug("The generation %d was finished.",
self.currentGeneration)
self.currentGeneration += 1
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
self.saveState()
return (self.currentGeneration == self.nGenerations)
class GSimpleGA:
""" GA Engine Class - The Genetic Algorithm Core
Example:
>>> ga = GSimpleGA.GSimpleGA(genome)
>>> ga.selector.set(Selectors.GRouletteWheel)
>>> ga.setGenerations(120)
>>> ga.terminationCriteria.set(GSimpleGA.ConvergenceCriteria)
:param genome: the :term:`Sample Genome`
:param interactiveMode: this flag enables the Interactive Mode, the default is True
:param seed: the random seed value
.. note:: if you use the same random seed, all the runs of algorithm will be the same
"""
selector = None
""" This is the function slot for the selection method
if you want to change the default selector, you must do this: ::
ga_engine.selector.set(Selectors.GRouletteWheel) """
stepCallback = None
""" This is the :term:`step callback function` slot,
if you want to set the function, you must do this: ::
def your_func(ga_engine):
# Here you have access to the GA Engine
return False
ga_engine.stepCallback.set(your_func)
now *"your_func"* will be called every generation.
When this function returns True, the GA Engine will stop the evolution and show
a warning, if is False, the evolution continues.
"""
terminationCriteria = None
""" This is the termination criteria slot, if you want to set one
termination criteria, you must do this: ::
ga_engine.terminationCriteria.set(GSimpleGA.ConvergenceCriteria)
Now, when you run your GA, it will stop when the population converges.
There are those termination criteria functions: :func:`GSimpleGA.RawScoreCriteria`, :func:`GSimpleGA.ConvergenceCriteria`, :func:`GSimpleGA.RawStatsCriteria`, :func:`GSimpleGA.FitnessStatsCriteria`
But you can create your own termination function, this function receives
one parameter which is the GA Engine, follows an example: ::
def ConvergenceCriteria(ga_engine):
pop = ga_engine.getPopulation()
return pop[0] == pop[len(pop)-1]
When this function returns True, the GA Engine will stop the evolution and show
a warning, if is False, the evolution continues, this function is called every
generation.
"""
def __init__(self, genome, owner, seed=None, interactiveMode=True):
""" Initializator of GSimpleGA """
if seed: random.seed(seed)
if type(interactiveMode) != BooleanType:
Util.raiseException(
"Interactive Mode option must be True or False", TypeError)
if not isinstance(genome, GenomeBase):
Util.raiseException("The genome must be a GenomeBase subclass",
TypeError)
self.internalPop = GPopulation(genome)
self.nGenerations = Consts.CDefGAGenerations
self.pMutation = Consts.CDefGAMutationRate
self.pCrossover = Consts.CDefGACrossoverRate
self.nElitismReplacement = Consts.CDefGAElitismReplacement
self.setPopulationSize(Consts.CDefGAPopulationSize)
self.minimax = Consts.minimaxType["maximize"]
self.elitism = True
self.owner = owner ## added 12/15 by Peter Graf, so GA can evaluate constraints
## and, now (4/16), so we can save let the owner save the state
# Adapters
self.dbAdapter = None
self.migrationAdapter = None
self.time_init = None
self.interactiveMode = interactiveMode
self.interactiveGen = -1
self.GPMode = False
self.selector = FunctionSlot("Selector")
self.stepCallback = FunctionSlot("Generation Step Callback")
self.terminationCriteria = FunctionSlot("Termination Criteria")
self.selector.set(Consts.CDefGASelector)
self.allSlots = [
self.selector, self.stepCallback, self.terminationCriteria
]
self.internalParams = {}
self.currentGeneration = 0
# GP Testing
for classes in Consts.CDefGPGenomes:
if isinstance(self.internalPop.oneSelfGenome, classes):
self.setGPMode(True)
break
logging.debug("A GA Engine was created, nGenerations=%d",
self.nGenerations)
def setGPMode(self, bool_value):
""" Sets the Genetic Programming mode of the GA Engine
:param bool_value: True or False
"""
self.GPMode = bool_value
def getGPMode(self):
""" Get the Genetic Programming mode of the GA Engine
:rtype: True or False
"""
return self.GPMode
def __call__(self, *args, **kwargs):
""" A method to implement a callable object
Example:
>>> ga_engine(freq_stats=10)
.. versionadded:: 0.6
The callable method.
"""
if kwargs.get("freq_stats", None):
return self.evolve(kwargs.get("freq_stats"))
else:
return self.evolve()
def setParams(self, **args):
""" Set the internal params
Example:
>>> ga.setParams(gp_terminals=['x', 'y'])
:param args: params to save
..versionaddd:: 0.6
Added the *setParams* method.
"""
self.internalParams.update(args)
def getParam(self, key, nvl=None):
""" Gets an internal parameter
Example:
>>> ga.getParam("gp_terminals")
['x', 'y']
:param key: the key of param
:param nvl: if the key doesn't exist, the nvl will be returned
..versionaddd:: 0.6
Added the *getParam* method.
"""
return self.internalParams.get(key, nvl)
def setInteractiveGeneration(self, generation):
""" Sets the generation in which the GA must enter in the
Interactive Mode
:param generation: the generation number, use "-1" to disable
.. versionadded::0.6
The *setInteractiveGeneration* method.
"""
if generation < -1:
Util.raiseException("Generation must be >= -1", ValueError)
self.interactiveGen = generation
def getInteractiveGeneration(self):
""" returns the generation in which the GA must enter in the
Interactive Mode
:rtype: the generation number or -1 if not set
.. versionadded::0.6
The *getInteractiveGeneration* method.
"""
return self.interactiveGen
def setElitismReplacement(self, numreplace):
""" Set the number of best individuals to copy to the next generation on the elitism
:param numreplace: the number of individuals
.. versionadded:: 0.6
The *setElitismReplacement* method.
"""
if numreplace < 1:
Util.raiseException("Replacement number must be >= 1", ValueError)
self.nElitismReplacement = numreplace
def setInteractiveMode(self, flag=True):
""" Enable/disable the interactive mode
:param flag: True or False
.. versionadded: 0.6
The *setInteractiveMode* method.
"""
if type(flag) != BooleanType:
Util.raiseException(
"Interactive Mode option must be True or False", TypeError)
self.interactiveMode = flag
def __repr__(self):
""" The string representation of the GA Engine """
ret = "- GSimpleGA\n"
ret += "\tGP Mode:\t\t %s\n" % self.getGPMode()
ret += "\tPopulation Size:\t %d\n" % (self.internalPop.popSize, )
ret += "\tGenerations:\t\t %d\n" % (self.nGenerations, )
ret += "\tCurrent Generation:\t %d\n" % (self.currentGeneration, )
ret += "\tMutation Rate:\t\t %.2f\n" % (self.pMutation, )
ret += "\tCrossover Rate:\t\t %.2f\n" % (self.pCrossover, )
ret += "\tMinimax Type:\t\t %s\n" % (Consts.minimaxType.keys()[
Consts.minimaxType.values().index(self.minimax)].capitalize(), )
ret += "\tElitism:\t\t %s\n" % (self.elitism, )
ret += "\tElitism Replacement:\t %d\n" % (self.nElitismReplacement, )
ret += "\tDB Adapter:\t\t %s\n" % (self.dbAdapter, )
for slot in self.allSlots:
ret += "\t" + slot.__repr__()
ret += "\n"
return ret
def setMultiProcessing(self, flag=True, full_copy=False):
""" Sets the flag to enable/disable the use of python multiprocessing module.
Use this option when you have more than one core on your CPU and when your
evaluation function is very slow.
Pyevolve will automaticly check if your Python version has **multiprocessing**
support and if you have more than one single CPU core. If you don't have support
or have just only one core, Pyevolve will not use the **multiprocessing**
feature.
Pyevolve uses the **multiprocessing** to execute the evaluation function over
the individuals, so the use of this feature will make sense if you have a
truly slow evaluation function (which is commom in GAs).
The parameter "full_copy" defines where the individual data should be copied back
after the evaluation or not. This parameter is useful when you change the
individual in the evaluation function.
:param flag: True (default) or False
:param full_copy: True or False (default)
.. warning:: Use this option only when your evaluation function is slow, so you'll
get a good tradeoff between the process communication speed and the
parallel evaluation. The use of the **multiprocessing** doesn't means
always a better performance.
.. note:: To enable the multiprocessing option, you **MUST** add the *__main__* check
on your application, otherwise, it will result in errors. See more on the
`Python Docs <http://docs.python.org/library/multiprocessing.html#multiprocessing-programming>`__
site.
.. versionadded:: 0.6
The `setMultiProcessing` method.
"""
if type(flag) != BooleanType:
Util.raiseException("Multiprocessing option must be True or False",
TypeError)
if type(full_copy) != BooleanType:
Util.raiseException(
"Multiprocessing 'full_copy' option must be True or False",
TypeError)
self.internalPop.setMultiProcessing(flag, full_copy)
def setMigrationAdapter(self, migration_adapter=None):
""" Sets the Migration Adapter
.. versionadded:: 0.6
The `setMigrationAdapter` method.
"""
self.migrationAdapter = migration_adapter
if self.migrationAdapter is not None:
self.migrationAdapter.setGAEngine(self)
def setDBAdapter(self, dbadapter=None):
""" Sets the DB Adapter of the GA Engine
:param dbadapter: one of the :mod:`DBAdapters` classes instance
.. warning:: the use the of a DB Adapter can reduce the speed performance of the
Genetic Algorithm.
"""
if (dbadapter
is not None) and (not isinstance(dbadapter, DBBaseAdapter)):
Util.raiseException(
"The DB Adapter must be a DBBaseAdapter subclass", TypeError)
self.dbAdapter = dbadapter
def setPopulationSize(self, size):
""" Sets the population size, calls setPopulationSize() of GPopulation
:param size: the population size
.. note:: the population size must be >= 2
"""
if size < 2:
Util.raiseException("population size must be >= 2", ValueError)
self.internalPop.setPopulationSize(size)
def setSortType(self, sort_type):
""" Sets the sort type, Consts.sortType["raw"]/Consts.sortType["scaled"]
Example:
>>> ga_engine.setSortType(Consts.sortType["scaled"])
:param sort_type: the Sort Type
"""
if sort_type not in Consts.sortType.values():
Util.raiseException("sort type must be a Consts.sortType type",
TypeError)
self.internalPop.sortType = sort_type
def setMutationRate(self, rate):
""" Sets the mutation rate, between 0.0 and 1.0
:param rate: the rate, between 0.0 and 1.0
"""
if (rate > 1.0) or (rate < 0.0):
Util.raiseException("Mutation rate must be >= 0.0 and <= 1.0",
ValueError)
self.pMutation = rate
def setCrossoverRate(self, rate):
""" Sets the crossover rate, between 0.0 and 1.0
:param rate: the rate, between 0.0 and 1.0
"""
if (rate > 1.0) or (rate < 0.0):
Util.raiseException("Crossover rate must be >= 0.0 and <= 1.0",
ValueError)
self.pCrossover = rate
def setGenerations(self, num_gens):
""" Sets the number of generations to evolve
:param num_gens: the number of generations
"""
if num_gens < 1:
Util.raiseException("Number of generations must be >= 1",
ValueError)
self.nGenerations = num_gens
def getGenerations(self):
""" Return the number of generations to evolve
:rtype: the number of generations
.. versionadded:: 0.6
Added the *getGenerations* method
"""
return self.nGenerations
def getMinimax(self):
""" Gets the minimize/maximize mode
:rtype: the Consts.minimaxType type
"""
return self.minimax
def setMinimax(self, mtype):
""" Sets the minimize/maximize mode, use Consts.minimaxType
:param mtype: the minimax mode, from Consts.minimaxType
"""
if mtype not in Consts.minimaxType.values():
Util.raiseException("Minimax must be maximize or minimize",
TypeError)
self.minimax = mtype
def getCurrentGeneration(self):
""" Gets the current generation
:rtype: the current generation
"""
return self.currentGeneration
def setElitism(self, flag):
""" Sets the elitism option, True or False
:param flag: True or False
"""
if type(flag) != BooleanType:
Util.raiseException("Elitism option must be True or False",
TypeError)
self.elitism = flag
def getDBAdapter(self):
""" Gets the DB Adapter of the GA Engine
:rtype: a instance from one of the :mod:`DBAdapters` classes
"""
return self.dbAdapter
def bestIndividual(self):
""" Returns the population best individual
:rtype: the best individual
"""
return self.internalPop.bestRaw()
def __gp_catch_functions(self, prefix):
""" Internally used to catch functions with some specific prefix
as non-terminals of the GP core """
import __main__ as mod_main
function_set = {}
main_dict = mod_main.__dict__
for obj, addr in main_dict.items():
if obj[0:len(prefix)] == prefix:
try:
op_len = addr.func_code.co_argcount
except:
continue
function_set[obj] = op_len
if len(function_set) <= 0:
Util.raiseException(
"No function set found using function prefix '%s' !" % prefix,
ValueError)
self.setParams(gp_function_set=function_set)
def initialize(self):
""" Initializes the GA Engine. Create and initialize population """
self.internalPop.create(minimax=self.minimax)
self.internalPop.initialize(ga_engine=self)
logging.debug("The GA Engine was initialized !")
def getPopulation(self):
""" Return the internal population of GA Engine
:rtype: the population (:class:`GPopulation.GPopulation`)
"""
return self.internalPop
def getStatistics(self):
""" Gets the Statistics class instance of current generation
:rtype: the statistics instance (:class:`Statistics.Statistics`)
"""
return self.internalPop.getStatistics()
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
popsize = len(self.internalPop)
numAdded = 0
maxTries = 1000
numTries = 0
crossover_empty = self.select(
popID=self.currentGeneration).crossover.isEmpty()
###TODO: enforce constraints!###
while numAdded < popsize:
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=2):
(sister, brother) = it
else:
if not crossover_empty and Util.randomFlipCoin(
self.pCrossover):
for it in genomeMom.crossover.applyFunctions(
mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
# logging.debug("done cloning")
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
if (numTries > maxTries
or self.owner.eval_constraints(sister)):
newPop.internalPop.append(sister)
numAdded += 1
print "successfully added sister"
if (numAdded < popsize
and (numTries > maxTries
or self.owner.eval_constraints(brother))):
newPop.internalPop.append(brother)
print "successfully added brother"
numAdded += 1
numTries += 1
#end rank0 onlye
# print "rank %d start eval pop" % MPI.COMM_WORLD.Get_rank()
logging.debug("Evaluating the newly created population.")
newPop.evaluate()
# print "rank %d done eval pop" % MPI.COMM_WORLD.Get_rank()
# if (MPI.COMM_WORLD.Get_rank() == 0):
# print "after eval, new pop's positions are:"
# for p in newPop:
# print p.wt_positions
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
if self.elitism:
logging.debug("Doing elitism, %d" % self.nElitismReplacement)
if self.getMinimax() == Consts.minimaxType["maximize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
# self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score > newPop.bestRaw(
i).score:
newPop[len(newPop) - 1 -
i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
# self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score < newPop.bestRaw(
i).score:
newPop[len(newPop) - 1 -
i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
self.internalPop.sort()
# if (MPI.COMM_WORLD.Get_rank() == 0):
# print "after sort, internal pop's positions are:"
# for p in self.internalPop:
# print p.wt_positions
logging.debug("The generation %d was finished.",
self.currentGeneration)
self.currentGeneration += 1
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
self.saveState()
return (self.currentGeneration == self.nGenerations)
def saveState(self):
self.owner.saveState(self.internalPop.internalPop,
self.currentGeneration)
def restoreState(self, gen):
self.owner.restoreState(self.internalPop.internalPop, gen)
self.currentGeneration = gen
def printStats(self):
""" Print generation statistics
:rtype: the printed statistics as string
.. versionchanged:: 0.6
The return of *printStats* method.
"""
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
percent = self.currentGeneration * 100 / float(self.nGenerations)
message = "Gen. %d (%.2f%%):" % (self.currentGeneration, percent)
logging.info(message)
print message,
sys_stdout.flush()
self.internalPop.statistics()
stat_ret = self.internalPop.printStats()
return message + stat_ret
else:
return ""
def printTimeElapsed(self):
""" Shows the time elapsed since the begin of evolution """
total_time = time() - self.time_init
print "Total time elapsed: %.3f seconds." % total_time
return total_time
def dumpStatsDB(self):
""" Dumps the current statistics to database adapter """
logging.debug("Dumping stats to the DB Adapter")
self.internalPop.statistics()
self.dbAdapter.insert(self)
def evolve(self, freq_stats=0, restore=-1):
""" Do all the generations until the termination criteria, accepts
the freq_stats (default is 0) to dump statistics at n-generation
Example:
>>> ga_engine.evolve(freq_stats=10)
(...)
:param freq_stats: if greater than 0, the statistics will be
printed every freq_stats generation.
:rtype: returns the best individual of the evolution
.. versionadded:: 0.6
the return of the best individual
"""
stopFlagCallback = False
stopFlagTerminationCriteria = False
self.time_init = time()
logging.debug(
"Starting the DB Adapter and the Migration Adapter if any")
if self.dbAdapter: self.dbAdapter.open(self)
if self.migrationAdapter: self.migrationAdapter.start()
if self.getGPMode():
gp_function_prefix = self.getParam("gp_function_prefix")
if gp_function_prefix is not None:
self.__gp_catch_functions(gp_function_prefix)
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
self.initialize()
if (restore >= 0):
self.restoreState(restore)
self.internalPop.evaluate()
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
self.internalPop.sort()
# from mpi4py import MPI
# if (MPI.COMM_WORLD.Get_rank() == 0):
# print "after eval, new pop's positions are:"
# for p in self.internalPop:
# print p.wt_positions
logging.debug("Starting loop over evolutionary algorithm.")
while True:
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
if self.migrationAdapter:
logging.debug("Migration adapter: exchange")
self.migrationAdapter.exchange()
self.internalPop.clearFlags()
self.internalPop.sort()
if not self.stepCallback.isEmpty():
for it in self.stepCallback.applyFunctions(self):
stopFlagCallback = it
if not self.terminationCriteria.isEmpty():
for it in self.terminationCriteria.applyFunctions(self):
stopFlagTerminationCriteria = it
if freq_stats:
if (self.currentGeneration % freq_stats
== 0) or (self.getCurrentGeneration() == 0):
self.printStats()
if self.dbAdapter:
if self.currentGeneration % self.dbAdapter.getStatsGenFreq(
) == 0:
self.dumpStatsDB()
if stopFlagTerminationCriteria:
logging.debug(
"Evolution stopped by the Termination Criteria !")
if freq_stats:
print "\n\tEvolution stopped by Termination Criteria function !\n"
break
if stopFlagCallback:
logging.debug(
"Evolution stopped by Step Callback function !")
if freq_stats:
print "\n\tEvolution stopped by Step Callback function !\n"
break
if self.interactiveMode:
if sys_platform[:3] == "win":
if msvcrt.kbhit():
if ord(msvcrt.getch()) == Consts.CDefESCKey:
print "Loading modules for Interactive Mode...",
logging.debug(
"Windows Interactive Mode key detected ! generation=%d",
self.getCurrentGeneration())
from pyevolve import Interaction
print " done !"
interact_banner = "## Pyevolve v.%s - Interactive Mode ##\nPress CTRL-Z to quit interactive mode." % (
pyevolve.__version__, )
session_locals = {
"ga_engine": self,
"population": self.getPopulation(),
"pyevolve": pyevolve,
"it": Interaction
}
print
code.interact(interact_banner,
local=session_locals)
if (self.getInteractiveGeneration() >=
0) and (self.getInteractiveGeneration()
== self.getCurrentGeneration()):
print "Loading modules for Interactive Mode...",
logging.debug(
"Manual Interactive Mode key detected ! generation=%d",
self.getCurrentGeneration())
from pyevolve import Interaction
print " done !"
interact_banner = "## Pyevolve v.%s - Interactive Mode ##" % (
pyevolve.__version__, )
session_locals = {
"ga_engine": self,
"population": self.getPopulation(),
"pyevolve": pyevolve,
"it": Interaction
}
print
code.interact(interact_banner, local=session_locals)
## end, rank0 only
if self.step(): break
if (MPI.COMM_WORLD.Get_rank() == 0
): ### assumes WE are on top of hierarchy!
if freq_stats != 0:
self.printStats()
self.printTimeElapsed()
if self.dbAdapter:
logging.debug("Closing the DB Adapter")
if not (self.currentGeneration %
self.dbAdapter.getStatsGenFreq() == 0):
self.dumpStatsDB()
self.dbAdapter.commitAndClose()
if self.migrationAdapter:
logging.debug("Closing the Migration Adapter")
if freq_stats: print "Stopping the migration adapter... ",
self.migrationAdapter.stop()
if freq_stats: print "done !"
return self.bestIndividual()
else:
return None
def select(self, **args):
""" Select one individual from population
:param args: this parameters will be sent to the selector
"""
for it in self.selector.applyFunctions(self.internalPop, **args):
return it
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
logging.debug("Population was cloned.")
size_iterate = len(self.internalPop)
# Odd population size
if size_iterate % 2 != 0: size_iterate -= 1
crossover_empty = self.select(
popID=self.currentGeneration).crossover.isEmpty()
for i in xrange(0, size_iterate, 2):
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=2):
(sister, brother) = it
else:
if not crossover_empty and Util.randomFlipCoin(
self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
newPop.internalPop.append(brother)
if len(self.internalPop) % 2 != 0:
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=1):
(sister, brother) = it
else:
sister = random.choice([genomeMom, genomeDad])
sister = sister.clone()
sister.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
logging.debug("Evaluating the new created population.")
newPop.evaluate()
if self.elitism:
logging.debug("Doing elitism.")
if self.getMinimax() == Consts.minimaxType["maximize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score > newPop.bestRaw(
i).score:
newPop[len(newPop) - 1 -
i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score < newPop.bestRaw(
i).score:
newPop[len(newPop) - 1 -
i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
self.internalPop.sort()
logging.debug("The generation %d was finished.",
self.currentGeneration)
self.currentGeneration += 1
return (self.currentGeneration == self.nGenerations)
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
logging.debug("Population was cloned.")
size_iterate = len(self.internalPop)
# Odd population size
if size_iterate % 2 != 0: size_iterate -= 1
crossover_empty = self.select(popID=self.currentGeneration).crossover.isEmpty()
for i in xrange(0, size_iterate, 2):
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
if not crossover_empty and Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
newPop.internalPop.append(brother)
if len(self.internalPop) % 2 != 0:
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=1):
(sister, brother) = it
else:
sister = random.choice([genomeMom, genomeDad])
sister = sister.clone()
sister.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
logging.debug("Evaluating the new created population.")
newPop.evaluate()
if self.elitism:
logging.debug("Doing elitism.")
if self.getMinimax() == Consts.minimaxType["maximize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score > newPop.bestRaw(i).score:
newPop[len(newPop)-1-i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score < newPop.bestRaw(i).score:
newPop[len(newPop)-1-i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
self.internalPop.sort()
logging.debug("The generation %d was finished.", self.currentGeneration)
self.currentGeneration += 1
return (self.currentGeneration == self.nGenerations)
class GSimpleGA:
""" GA Engine Class - The Genetic Algorithm Core
Example:
>>> ga = GSimpleGA.GSimpleGA(genome)
>>> ga.selector.set(Selectors.GRouletteWheel)
>>> ga.setGenerations(120)
>>> ga.terminationCriteria.set(GSimpleGA.ConvergenceCriteria)
:param genome: the :term:`Sample Genome`
:param interactiveMode: this flag enables the Interactive Mode, the default is True
:param seed: the random seed value
.. note:: if you use the same random seed, all the runs of algorithm will be the same
"""
selector = None
""" This is the function slot for the selection method
if you want to change the default selector, you must do this: ::
ga_engine.selector.set(Selectors.GRouletteWheel) """
stepCallback = None
""" This is the :term:`step callback function` slot,
if you want to set the function, you must do this: ::
def your_func(ga_engine):
# Here you have access to the GA Engine
return False
ga_engine.stepCallback.set(your_func)
now *"your_func"* will be called every generation.
When this function returns True, the GA Engine will stop the evolution and show
a warning, if is False, the evolution continues.
"""
terminationCriteria = None
""" This is the termination criteria slot, if you want to set one
termination criteria, you must do this: ::
ga_engine.terminationCriteria.set(GSimpleGA.ConvergenceCriteria)
Now, when you run your GA, it will stop when the population converges.
There are those termination criteria functions: :func:`GSimpleGA.RawScoreCriteria`, :func:`GSimpleGA.ConvergenceCriteria`, :func:`GSimpleGA.RawStatsCriteria`, :func:`GSimpleGA.FitnessStatsCriteria`
But you can create your own termination function, this function receives
one parameter which is the GA Engine, follows an example: ::
def ConvergenceCriteria(ga_engine):
pop = ga_engine.getPopulation()
return pop[0] == pop[len(pop)-1]
When this function returns True, the GA Engine will stop the evolution and show
a warning, if is False, the evolution continues, this function is called every
generation.
"""
def __init__(self, genome, seed=None, interactiveMode=True):
""" Initializator of GSimpleGA """
if seed: random.seed(seed)
if type(interactiveMode) != BooleanType:
Util.raiseException("Interactive Mode option must be True or False", TypeError)
if not isinstance(genome, GenomeBase):
Util.raiseException("The genome must be a GenomeBase subclass", TypeError)
self.internalPop = GPopulation(genome)
self.nGenerations = Consts.CDefGAGenerations
self.pMutation = Consts.CDefGAMutationRate
self.pCrossover = Consts.CDefGACrossoverRate
self.nElitismReplacement = Consts.CDefGAElitismReplacement
self.setPopulationSize(Consts.CDefGAPopulationSize)
self.minimax = Consts.minimaxType["maximize"]
self.elitism = True
# Adapters
self.dbAdapter = None
self.migrationAdapter = None
self.time_init = None
self.interactiveMode = interactiveMode
self.interactiveGen = -1
self.GPMode = False
self.selector = FunctionSlot("Selector")
self.stepCallback = FunctionSlot("Generation Step Callback")
self.terminationCriteria = FunctionSlot("Termination Criteria")
self.selector.set(Consts.CDefGASelector)
self.allSlots = [ self.selector, self.stepCallback, self.terminationCriteria ]
self.internalParams = {}
self.currentGeneration = 0
# GP Testing
for classes in Consts.CDefGPGenomes:
if isinstance(self.internalPop.oneSelfGenome, classes):
self.setGPMode(True)
break
logging.debug("A GA Engine was created, nGenerations=%d", self.nGenerations)
def setGPMode(self, bool_value):
""" Sets the Genetic Programming mode of the GA Engine
:param bool_value: True or False
"""
self.GPMode = bool_value
def getGPMode(self):
""" Get the Genetic Programming mode of the GA Engine
:rtype: True or False
"""
return self.GPMode
def __call__(self, *args, **kwargs):
""" A method to implement a callable object
Example:
>>> ga_engine(freq_stats=10)
.. versionadded:: 0.6
The callable method.
"""
if kwargs.get("freq_stats", None):
return self.evolve(kwargs.get("freq_stats"))
else:
return self.evolve()
def setParams(self, **args):
""" Set the internal params
Example:
>>> ga.setParams(gp_terminals=['x', 'y'])
:param args: params to save
..versionaddd:: 0.6
Added the *setParams* method.
"""
self.internalParams.update(args)
def getParam(self, key, nvl=None):
""" Gets an internal parameter
Example:
>>> ga.getParam("gp_terminals")
['x', 'y']
:param key: the key of param
:param nvl: if the key doesn't exist, the nvl will be returned
..versionaddd:: 0.6
Added the *getParam* method.
"""
return self.internalParams.get(key, nvl)
def setInteractiveGeneration(self, generation):
""" Sets the generation in which the GA must enter in the
Interactive Mode
:param generation: the generation number, use "-1" to disable
.. versionadded::0.6
The *setInteractiveGeneration* method.
"""
if generation < -1:
Util.raiseException("Generation must be >= -1", ValueError)
self.interactiveGen = generation
def getInteractiveGeneration(self):
""" returns the generation in which the GA must enter in the
Interactive Mode
:rtype: the generation number or -1 if not set
.. versionadded::0.6
The *getInteractiveGeneration* method.
"""
return self.interactiveGen
def setElitismReplacement(self, numreplace):
""" Set the number of best individuals to copy to the next generation on the elitism
:param numreplace: the number of individuals
.. versionadded:: 0.6
The *setElitismReplacement* method.
"""
if numreplace < 1:
Util.raiseException("Replacement number must be >= 1", ValueError)
self.nElitismReplacement = numreplace
def setInteractiveMode(self, flag=True):
""" Enable/disable the interactive mode
:param flag: True or False
.. versionadded: 0.6
The *setInteractiveMode* method.
"""
if type(flag) != BooleanType:
Util.raiseException("Interactive Mode option must be True or False", TypeError)
self.interactiveMode = flag
def __repr__(self):
""" The string representation of the GA Engine """
ret = "- GSimpleGA\n"
ret += "\tGP Mode:\t\t %s\n" % self.getGPMode()
ret += "\tPopulation Size:\t %d\n" % (self.internalPop.popSize,)
ret += "\tGenerations:\t\t %d\n" % (self.nGenerations,)
ret += "\tCurrent Generation:\t %d\n" % (self.currentGeneration,)
ret += "\tMutation Rate:\t\t %.2f\n" % (self.pMutation,)
ret += "\tCrossover Rate:\t\t %.2f\n" % (self.pCrossover,)
ret += "\tMinimax Type:\t\t %s\n" % (Consts.minimaxType.keys()[Consts.minimaxType.values().index(self.minimax)].capitalize(),)
ret += "\tElitism:\t\t %s\n" % (self.elitism,)
ret += "\tElitism Replacement:\t %d\n" % (self.nElitismReplacement,)
ret += "\tDB Adapter:\t\t %s\n" % (self.dbAdapter,)
for slot in self.allSlots:
ret+= "\t" + slot.__repr__()
ret+="\n"
return ret
def setMultiProcessing(self, flag=True, full_copy=False):
""" Sets the flag to enable/disable the use of python multiprocessing module.
Use this option when you have more than one core on your CPU and when your
evaluation function is very slow.
Pyevolve will automaticly check if your Python version has **multiprocessing**
support and if you have more than one single CPU core. If you don't have support
or have just only one core, Pyevolve will not use the **multiprocessing**
feature.
Pyevolve uses the **multiprocessing** to execute the evaluation function over
the individuals, so the use of this feature will make sense if you have a
truly slow evaluation function (which is commom in GAs).
The parameter "full_copy" defines where the individual data should be copied back
after the evaluation or not. This parameter is useful when you change the
individual in the evaluation function.
:param flag: True (default) or False
:param full_copy: True or False (default)
.. warning:: Use this option only when your evaluation function is slow, so you'll
get a good tradeoff between the process communication speed and the
parallel evaluation. The use of the **multiprocessing** doesn't means
always a better performance.
.. note:: To enable the multiprocessing option, you **MUST** add the *__main__* check
on your application, otherwise, it will result in errors. See more on the
`Python Docs <http://docs.python.org/library/multiprocessing.html#multiprocessing-programming>`__
site.
.. versionadded:: 0.6
The `setMultiProcessing` method.
"""
if type(flag) != BooleanType:
Util.raiseException("Multiprocessing option must be True or False", TypeError)
if type(full_copy) != BooleanType:
Util.raiseException("Multiprocessing 'full_copy' option must be True or False", TypeError)
self.internalPop.setMultiProcessing(flag, full_copy)
def setMigrationAdapter(self, migration_adapter=None):
""" Sets the Migration Adapter
.. versionadded:: 0.6
The `setMigrationAdapter` method.
"""
self.migrationAdapter = migration_adapter
if self.migrationAdapter is not None:
self.migrationAdapter.setGAEngine(self)
def setDBAdapter(self, dbadapter=None):
""" Sets the DB Adapter of the GA Engine
:param dbadapter: one of the :mod:`DBAdapters` classes instance
.. warning:: the use the of a DB Adapter can reduce the speed performance of the
Genetic Algorithm.
"""
if (dbadapter is not None) and (not isinstance(dbadapter, DBBaseAdapter)):
Util.raiseException("The DB Adapter must be a DBBaseAdapter subclass", TypeError)
self.dbAdapter = dbadapter
def setPopulationSize(self, size):
""" Sets the population size, calls setPopulationSize() of GPopulation
:param size: the population size
.. note:: the population size must be >= 2
"""
if size < 2:
Util.raiseException("population size must be >= 2", ValueError)
self.internalPop.setPopulationSize(size)
def setSortType(self, sort_type):
""" Sets the sort type, Consts.sortType["raw"]/Consts.sortType["scaled"]
Example:
>>> ga_engine.setSortType(Consts.sortType["scaled"])
:param sort_type: the Sort Type
"""
if sort_type not in Consts.sortType.values():
Util.raiseException("sort type must be a Consts.sortType type", TypeError)
self.internalPop.sortType = sort_type
def setMutationRate(self, rate):
""" Sets the mutation rate, between 0.0 and 1.0
:param rate: the rate, between 0.0 and 1.0
"""
if (rate>1.0) or (rate<0.0):
Util.raiseException("Mutation rate must be >= 0.0 and <= 1.0", ValueError)
self.pMutation = rate
def setCrossoverRate(self, rate):
""" Sets the crossover rate, between 0.0 and 1.0
:param rate: the rate, between 0.0 and 1.0
"""
if (rate>1.0) or (rate<0.0):
Util.raiseException("Crossover rate must be >= 0.0 and <= 1.0", ValueError)
self.pCrossover = rate
def setGenerations(self, num_gens):
""" Sets the number of generations to evolve
:param num_gens: the number of generations
"""
if num_gens < 1:
Util.raiseException("Number of generations must be >= 1", ValueError)
self.nGenerations = num_gens
def getGenerations(self):
""" Return the number of generations to evolve
:rtype: the number of generations
.. versionadded:: 0.6
Added the *getGenerations* method
"""
return self.nGenerations
def getMinimax(self):
""" Gets the minimize/maximize mode
:rtype: the Consts.minimaxType type
"""
return self.minimax
def setMinimax(self, mtype):
""" Sets the minimize/maximize mode, use Consts.minimaxType
:param mtype: the minimax mode, from Consts.minimaxType
"""
if mtype not in Consts.minimaxType.values():
Util.raiseException("Minimax must be maximize or minimize", TypeError)
self.minimax = mtype
def getCurrentGeneration(self):
""" Gets the current generation
:rtype: the current generation
"""
return self.currentGeneration
def setElitism(self, flag):
""" Sets the elitism option, True or False
:param flag: True or False
"""
if type(flag) != BooleanType:
Util.raiseException("Elitism option must be True or False", TypeError)
self.elitism = flag
def getDBAdapter(self):
""" Gets the DB Adapter of the GA Engine
:rtype: a instance from one of the :mod:`DBAdapters` classes
"""
return self.dbAdapter
def bestIndividual(self):
""" Returns the population best individual
:rtype: the best individual
"""
return self.internalPop.bestRaw()
def worstIndividual(self):
""" Returns the population worst individual
:rtype: the best individual
"""
return self.internalPop.worstRaw()
def __gp_catch_functions(self, prefix):
""" Internally used to catch functions with some specific prefix
as non-terminals of the GP core """
import __main__ as mod_main
function_set = {}
main_dict = mod_main.__dict__
for obj, addr in main_dict.items():
if obj[0:len(prefix)] == prefix:
try:
op_len = addr.func_code.co_argcount
except:
continue
function_set[obj] = op_len
if len(function_set) <= 0:
Util.raiseException("No function set found using function prefix '%s' !" % prefix, ValueError)
self.setParams(gp_function_set=function_set)
def initialize(self):
""" Initializes the GA Engine. Create and initialize population """
self.internalPop.create(minimax=self.minimax)
self.internalPop.initialize(ga_engine=self)
logging.debug("The GA Engine was initialized !")
def getPopulation(self):
""" Return the internal population of GA Engine
:rtype: the population (:class:`GPopulation.GPopulation`)
"""
return self.internalPop
def getStatistics(self):
""" Gets the Statistics class instance of current generation
:rtype: the statistics instance (:class:`Statistics.Statistics`)
"""
return self.internalPop.getStatistics()
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
logging.debug("Population was cloned.")
size_iterate = len(self.internalPop)
# Odd population size
if size_iterate % 2 != 0: size_iterate -= 1
crossover_empty = self.select(popID=self.currentGeneration).crossover.isEmpty()
for i in xrange(0, size_iterate, 2):
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
if not crossover_empty and Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
newPop.internalPop.append(brother)
if len(self.internalPop) % 2 != 0:
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=1):
(sister, brother) = it
else:
sister = random.choice([genomeMom, genomeDad])
sister = sister.clone()
sister.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
logging.debug("Evaluating the new created population.")
newPop.evaluate()
if self.elitism:
logging.debug("Doing elitism.")
if self.getMinimax() == Consts.minimaxType["maximize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score > newPop.bestRaw(i).score:
newPop[len(newPop)-1-i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score < newPop.bestRaw(i).score:
newPop[len(newPop)-1-i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
self.internalPop.sort()
logging.debug("The generation %d was finished.", self.currentGeneration)
self.currentGeneration += 1
return (self.currentGeneration == self.nGenerations)
def printStats(self):
""" Print generation statistics
:rtype: the printed statistics as string
.. versionchanged:: 0.6
The return of *printStats* method.
"""
percent = self.currentGeneration * 100 / float(self.nGenerations)
message = "Gen. %d (%.2f%%):" % (self.currentGeneration, percent)
logging.info(message)
print message,
sys_stdout.flush()
self.internalPop.statistics()
stat_ret = self.internalPop.printStats()
return message + stat_ret
def printTimeElapsed(self):
""" Shows the time elapsed since the begin of evolution """
total_time = time()-self.time_init
print "Total time elapsed: %.3f seconds." % total_time
return total_time
def dumpStatsDB(self):
""" Dumps the current statistics to database adapter """
logging.debug("Dumping stats to the DB Adapter")
self.internalPop.statistics()
self.dbAdapter.insert(self)
def evolve(self, freq_stats=0):
""" Do all the generations until the termination criteria, accepts
the freq_stats (default is 0) to dump statistics at n-generation
Example:
>>> ga_engine.evolve(freq_stats=10)
(...)
:param freq_stats: if greater than 0, the statistics will be
printed every freq_stats generation.
:rtype: returns the best individual of the evolution
.. versionadded:: 0.6
the return of the best individual
"""
stopFlagCallback = False
stopFlagTerminationCriteria = False
self.time_init = time()
logging.debug("Starting the DB Adapter and the Migration Adapter if any")
if self.dbAdapter: self.dbAdapter.open(self)
if self.migrationAdapter: self.migrationAdapter.start()
if self.getGPMode():
gp_function_prefix = self.getParam("gp_function_prefix")
if gp_function_prefix is not None:
self.__gp_catch_functions(gp_function_prefix)
self.initialize()
self.internalPop.evaluate()
self.internalPop.sort()
logging.debug("Starting loop over evolutionary algorithm.")
def stop_evolution(s, f):
#print signal, frame
if s == signal.SIGINT:
if self.internalPop.multiProcessing[0]:
logging.debug("CTRL-C detected, finishing evolution (stopping parallel jobs).")
self.internalPop.proc_pool.terminate()
self.internalPop.proc_pool.join()
else:
logging.debug("CTRL-C detected, finishing evolution.")
if freq_stats: print "\n\tA break was detected, you have interrupted the evolution !\n"
if freq_stats != 0:
self.printStats()
self.printTimeElapsed()
if self.dbAdapter:
logging.debug("Closing the DB Adapter")
if not (self.currentGeneration % self.dbAdapter.getStatsGenFreq() == 0):
self.dumpStatsDB()
self.dbAdapter.commitAndClose()
if self.migrationAdapter:
logging.debug("Closing the Migration Adapter")
if freq_stats: print "Stopping the migration adapter... ",
self.migrationAdapter.stop()
if freq_stats: print "done !"
if s == signal.SIGINT:
print self.bestIndividual()
exit(0)
else:
return self.bestIndividual()
signal.signal(signal.SIGINT, stop_evolution)
while True:
if self.migrationAdapter:
logging.debug("Migration adapter: exchange")
self.migrationAdapter.exchange()
self.internalPop.clearFlags()
self.internalPop.sort()
if not self.stepCallback.isEmpty():
for it in self.stepCallback.applyFunctions(self):
stopFlagCallback = it
if not self.terminationCriteria.isEmpty():
for it in self.terminationCriteria.applyFunctions(self):
stopFlagTerminationCriteria = it
if freq_stats:
if (self.currentGeneration % freq_stats == 0) or (self.getCurrentGeneration() == 0):
self.printStats()
#print self.bestIndividual()
if self.dbAdapter:
if self.currentGeneration % self.dbAdapter.getStatsGenFreq() == 0:
self.dumpStatsDB()
if stopFlagTerminationCriteria:
logging.debug("Evolution stopped by the Termination Criteria !")
if freq_stats:
print "\n\tEvolution stopped by Termination Criteria function !\n"
break
if stopFlagCallback:
logging.debug("Evolution stopped by Step Callback function !")
if freq_stats:
print "\n\tEvolution stopped by Step Callback function !\n"
break
if self.interactiveMode:
if sys_platform[:3] == "win":
if msvcrt.kbhit():
if ord(msvcrt.getch()) == Consts.CDefESCKey:
print "Loading modules for Interactive Mode...",
logging.debug("Windows Interactive Mode key detected ! generation=%d", self.getCurrentGeneration())
from pyevolve import Interaction
print " done !"
interact_banner = "## Pyevolve v.%s - Interactive Mode ##\nPress CTRL-Z to quit interactive mode." % (pyevolve.__version__,)
session_locals = { "ga_engine" : self,
"population" : self.getPopulation(),
"pyevolve" : pyevolve,
"it" : Interaction}
print
code.interact(interact_banner, local=session_locals)
if (self.getInteractiveGeneration() >= 0) and (self.getInteractiveGeneration() == self.getCurrentGeneration()):
print "Loading modules for Interactive Mode...",
logging.debug("Manual Interactive Mode key detected ! generation=%d", self.getCurrentGeneration())
from pyevolve import Interaction
print " done !"
interact_banner = "## Pyevolve v.%s - Interactive Mode ##" % (pyevolve.__version__,)
session_locals = { "ga_engine" : self,
"population" : self.getPopulation(),
"pyevolve" : pyevolve,
"it" : Interaction}
print
code.interact(interact_banner, local=session_locals)
if self.step(): break
return stop_evolution(signal.SIGUSR1, None)
def select(self, **args):
""" Select one individual from population
:param args: this parameters will be sent to the selector
"""
for it in self.selector.applyFunctions(self.internalPop, **args):
return it
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
logging.debug("Population was cloned.")
size_iterate = len(self.internalPop)
# Odd population size
if size_iterate % 2 != 0: size_iterate -= 1
crossover_empty = self.select(popID=self.currentGeneration).crossover.isEmpty()
from utils import delog
#############
delog.decache("mutate and check...")
#################
for i in xrange(0, size_iterate, 2):
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
if not crossover_empty and Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
newPop.internalPop.append(brother)
delog.deprint_string("over.")
if len(self.internalPop) % 2 != 0:
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=1):
(sister, brother) = it
else:
sister = random.choice([genomeMom, genomeDad])
sister = sister.clone()
sister.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
logging.debug("Evaluating the new created population.")
newPop.evaluate()
newPop.sort()
#Niching methods- Petrowski's clearing
self.clear()
if self.elitism:
logging.debug("Doing elitism.")
if self.getMinimax() == Consts.minimaxType["maximize"]:
# in ecoc, max value is expected.
for i in xrange(self.nElitismReplacement):
if self.internalPop.bestRaw(i).score > newPop.bestRaw(i).score:
# check duplicate to avoid repeat indivadual
duplicate = False
for j in xrange(len(newPop)-self.nElitismReplacement, len(newPop)):
if self.internalPop.bestRaw(i).score == newPop.bestRaw(j).score:
duplicate = True
break
if duplicate: continue
newPop[len(newPop)-1-i] = newPop[i]
newPop[i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
if self.internalPop.bestRaw(i).score < newPop.bestRaw(i).score:
newPop[len(newPop)-1-i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
self.internalPop.sort()
logging.debug("The generation %d was finished.", self.currentGeneration)
self.currentGeneration += 1
return (self.currentGeneration == self.nGenerations)
class GSimpleGA:
""" GA Engine Class - The Genetic Algorithm Core
Example:
>>> ga = GSimpleGA.GSimpleGA(genome)
>>> ga.selector.set(Selectors.GRouletteWheel)
>>> ga.setGenerations(120)
>>> ga.terminationCriteria.set(GSimpleGA.ConvergenceCriteria)
:param genome: the :term:`Sample Genome`
:param interactiveMode: this flag enables the Interactive Mode, the default is True
:param seed: the random seed value
.. note:: if you use the same random seed, all the runs of algorithm will be the same
"""
selector = None
""" This is the function slot for the selection method
if you want to change the default selector, you must do this: ::
ga_engine.selector.set(Selectors.GRouletteWheel) """
stepCallback = None
""" This is the :term:`step callback function` slot,
if you want to set the function, you must do this: ::
def your_func(ga_engine):
# Here you have access to the GA Engine
return False
ga_engine.stepCallback.set(your_func)
now *"your_func"* will be called every generation.
When this function returns True, the GA Engine will stop the evolution and show
a warning, if is False, the evolution continues.
"""
terminationCriteria = None
""" This is the termination criteria slot, if you want to set one
termination criteria, you must do this: ::
ga_engine.terminationCriteria.set(GSimpleGA.ConvergenceCriteria)
Now, when you run your GA, it will stop when the population converges.
There are those termination criteria functions: :func:`GSimpleGA.RawScoreCriteria`, :func:`GSimpleGA.ConvergenceCriteria`, :func:`GSimpleGA.RawStatsCriteria`, :func:`GSimpleGA.FitnessStatsCriteria`
But you can create your own termination function, this function receives
one parameter which is the GA Engine, follows an example: ::
def ConvergenceCriteria(ga_engine):
pop = ga_engine.getPopulation()
return pop[0] == pop[len(pop)-1]
When this function returns True, the GA Engine will stop the evolution and show
a warning, if is False, the evolution continues, this function is called every
generation.
"""
def __init__(self, genome, seed=None, interactiveMode=True):
""" Initializator of GSimpleGA """
if seed: random.seed(seed)
if type(interactiveMode) != BooleanType:
gp.Util.raiseException("Interactive Mode option must be True or False", TypeError)
if not isinstance(genome, GenomeBase):
gp.Util.raiseException("The genome must be a GenomeBase subclass", TypeError)
self.internalPop = GPopulation(genome)
self.nGenerations = Consts.CDefGAGenerations
self.pMutation = Consts.CDefGAMutationRate
self.pCrossover = Consts.CDefGACrossoverRate
self.nElitismReplacement = Consts.CDefGAElitismReplacement
self.setPopulationSize(Consts.CDefGAPopulationSize)
self.minimax = Consts.minimaxType["maximize"]
self.elitism = True
# Adapters
self.dbAdapter = None
self.migrationAdapter = None
self.time_init = None
self.interactiveMode = interactiveMode
self.interactiveGen = -1
self.GPMode = False
self.selector = FunctionSlot("Selector")
self.stepCallback = FunctionSlot("Generation Step Callback")
self.terminationCriteria = FunctionSlot("Termination Criteria")
self.selector.set(Consts.CDefGASelector)
self.allSlots = [ self.selector, self.stepCallback, self.terminationCriteria ]
self.internalParams = {}
self.currentGeneration = 0
# gp Testing
CDefGPGenomes = [GTreeGP]
for classes in CDefGPGenomes:
if isinstance(self.internalPop.oneSelfGenome, classes):
self.setGPMode(True)
break
logging.debug("A GA Engine was created, nGenerations=%d", self.nGenerations)
def setGPMode(self, bool_value):
""" Sets the Genetic Programming mode of the GA Engine
:param bool_value: True or False
"""
self.GPMode = bool_value
def getGPMode(self):
""" Get the Genetic Programming mode of the GA Engine
:rtype: True or False
"""
return self.GPMode
def __call__(self, *args, **kwargs):
""" A method to implement a callable object
Example:
>>> ga_engine(freq_stats=10)
.. versionadded:: 0.6
The callable method.
"""
if kwargs.get("freq_stats", None):
return self.evolve(kwargs.get("freq_stats"))
else:
return self.evolve()
def setParams(self, **args):
""" Set the internal params
Example:
>>> ga.setParams(gp_terminals=['x', 'y'])
:param args: params to save
..versionaddd:: 0.6
Added the *setParams* method.
"""
self.internalParams.update(args)
def getParam(self, key, nvl=None):
""" Gets an internal parameter
Example:
>>> ga.getParam("gp_terminals")
['x', 'y']
:param key: the key of param
:param nvl: if the key doesn't exist, the nvl will be returned
..versionaddd:: 0.6
Added the *getParam* method.
"""
return self.internalParams.get(key, nvl)
def setInteractiveGeneration(self, generation):
""" Sets the generation in which the GA must enter in the
Interactive Mode
:param generation: the generation number, use "-1" to disable
.. versionadded::0.6
The *setInteractiveGeneration* method.
"""
if generation < -1:
Util.raiseException("Generation must be >= -1", ValueError)
self.interactiveGen = generation
def getInteractiveGeneration(self):
""" returns the generation in which the GA must enter in the
Interactive Mode
:rtype: the generation number or -1 if not set
.. versionadded::0.6
The *getInteractiveGeneration* method.
"""
return self.interactiveGen
def setElitismReplacement(self, numreplace):
""" Set the number of best individuals to copy to the next generation on the elitism
:param numreplace: the number of individuals
.. versionadded:: 0.6
The *setElitismReplacement* method.
"""
if numreplace < 1:
Util.raiseException("Replacement number must be >= 1", ValueError)
self.nElitismReplacement = numreplace
def setInteractiveMode(self, flag=True):
""" Enable/disable the interactive mode
:param flag: True or False
.. versionadded: 0.6
The *setInteractiveMode* method.
"""
if type(flag) != BooleanType:
Util.raiseException("Interactive Mode option must be True or False", TypeError)
self.interactiveMode = flag
def __repr__(self):
""" The string representation of the GA Engine """
ret = "- GSimpleGA\n"
ret += "\tgp Mode:\t\t %s\n" % self.getGPMode()
ret += "\tPopulation Size:\t %d\n" % (self.internalPop.popSize,)
ret += "\tGenerations:\t\t %d\n" % (self.nGenerations,)
ret += "\tCurrent Generation:\t %d\n" % (self.currentGeneration,)
ret += "\tMutation Rate:\t\t %.2f\n" % (self.pMutation,)
ret += "\tCrossover Rate:\t\t %.2f\n" % (self.pCrossover,)
ret += "\tMinimax Type:\t\t %s\n" % (
Consts.minimaxType.keys()[Consts.minimaxType.values().index(self.minimax)].capitalize(),)
ret += "\tElitism:\t\t %s\n" % (self.elitism,)
ret += "\tElitism Replacement:\t %d\n" % (self.nElitismReplacement,)
ret += "\tDB Adapter:\t\t %s\n" % (self.dbAdapter,)
for slot in self.allSlots:
ret+= "\t" + slot.__repr__()
ret+="\n"
return ret
def setPopulationSize(self, size):
""" Sets the population size, calls setPopulationSize() of GPopulation
:param size: the population size
.. note:: the population size must be >= 2
"""
if size < 2:
Util.raiseException("population size must be >= 2", ValueError)
self.internalPop.setPopulationSize(size)
def setSortType(self, sort_type):
""" Sets the sort type, Consts.sortType["raw"]/Consts.sortType["scaled"]
Example:
>>> ga_engine.setSortType(Consts.sortType["scaled"])
:param sort_type: the Sort Type
"""
if sort_type not in Consts.sortType.values():
Util.raiseException("sort type must be a Consts.sortType type", TypeError)
self.internalPop.sortType = sort_type
def setMutationRate(self, rate):
""" Sets the mutation rate, between 0.0 and 1.0
:param rate: the rate, between 0.0 and 1.0
"""
if (rate>1.0) or (rate<0.0):
Util.raiseException("Mutation rate must be >= 0.0 and <= 1.0", ValueError)
self.pMutation = rate
def setCrossoverRate(self, rate):
""" Sets the crossover rate, between 0.0 and 1.0
:param rate: the rate, between 0.0 and 1.0
"""
if (rate>1.0) or (rate<0.0):
Util.raiseException("Crossover rate must be >= 0.0 and <= 1.0", ValueError)
self.pCrossover = rate
def setGenerations(self, num_gens):
""" Sets the number of generations to evolve
:param num_gens: the number of generations
"""
if num_gens < 1:
Util.raiseException("Number of generations must be >= 1", ValueError)
self.nGenerations = num_gens
def getGenerations(self):
""" Return the number of generations to evolve
:rtype: the number of generations
.. versionadded:: 0.6
Added the *getGenerations* method
"""
return self.nGenerations
def getMinimax(self):
""" Gets the minimize/maximize mode
:rtype: the Consts.minimaxType type
"""
return self.minimax
def setMinimax(self, mtype):
""" Sets the minimize/maximize mode, use Consts.minimaxType
:param mtype: the minimax mode, from Consts.minimaxType
"""
if mtype not in Consts.minimaxType.values():
Util.raiseException("Minimax must be maximize or minimize", TypeError)
self.minimax = mtype
def getCurrentGeneration(self):
""" Gets the current generation
:rtype: the current generation
"""
return self.currentGeneration
def setElitism(self, flag):
""" Sets the elitism option, True or False
:param flag: True or False
"""
if type(flag) != BooleanType:
Util.raiseException("Elitism option must be True or False", TypeError)
self.elitism = flag
def bestIndividual(self):
""" Returns the population best individual
:rtype: the best individual
"""
return self.internalPop.bestRaw()
def __gp_catch_functions(self, prefix):
""" Internally used to catch functions with some specific prefix
as non-terminals of the gp core """
import __main__ as mod_main
function_set = {}
main_dict = mod_main.__dict__
for obj, addr in main_dict.items():
if obj[0:len(prefix)] == prefix:
try:
op_len = addr.func_code.co_argcount
except:
continue
function_set[obj] = op_len
if len(function_set) <= 0:
Util.raiseException("No function set found using function prefix '%s' !" % prefix, ValueError)
self.setParams(gp_function_set=function_set)
def initialize(self):
""" Initializes the GA Engine. Create and initialize population """
self.internalPop.create(minimax=self.minimax)
self.internalPop.initialize(ga_engine=self)
logging.debug("The GA Engine was initialized !")
def getPopulation(self):
""" Return the internal population of GA Engine
:rtype: the population (:class:`GPopulation.GPopulation`)
"""
return self.internalPop
def getStatistics(self):
""" Gets the Statistics class instance of current generation
:rtype: the statistics instance (:class:`Statistics.Statistics`)
"""
return self.internalPop.getStatistics()
def clear(self):
""" Petrowski's Clearing Method """
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
logging.debug("Population was cloned.")
size_iterate = len(self.internalPop)
# Odd population size
if size_iterate % 2 != 0: size_iterate -= 1
crossover_empty = self.select(popID=self.currentGeneration).crossover.isEmpty()
from utils import delog
#############
delog.decache("mutate and check...")
#################
for i in xrange(0, size_iterate, 2):
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
if not crossover_empty and Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
newPop.internalPop.append(brother)
delog.deprint_string("over.")
if len(self.internalPop) % 2 != 0:
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=1):
(sister, brother) = it
else:
sister = random.choice([genomeMom, genomeDad])
sister = sister.clone()
sister.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
logging.debug("Evaluating the new created population.")
newPop.evaluate()
newPop.sort()
#Niching methods- Petrowski's clearing
self.clear()
if self.elitism:
logging.debug("Doing elitism.")
if self.getMinimax() == Consts.minimaxType["maximize"]:
# in ecoc, max value is expected.
for i in xrange(self.nElitismReplacement):
if self.internalPop.bestRaw(i).score > newPop.bestRaw(i).score:
# check duplicate to avoid repeat indivadual
duplicate = False
for j in xrange(len(newPop)-self.nElitismReplacement, len(newPop)):
if self.internalPop.bestRaw(i).score == newPop.bestRaw(j).score:
duplicate = True
break
if duplicate: continue
newPop[len(newPop)-1-i] = newPop[i]
newPop[i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
if self.internalPop.bestRaw(i).score < newPop.bestRaw(i).score:
newPop[len(newPop)-1-i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
self.internalPop.sort()
logging.debug("The generation %d was finished.", self.currentGeneration)
self.currentGeneration += 1
return (self.currentGeneration == self.nGenerations)
def printStats(self):
""" Print generation statistics
:rtype: the printed statistics as string
.. versionchanged:: 0.6
The return of *printStats* method.
"""
percent = self.currentGeneration * 100 / float(self.nGenerations)
#message = "Gen. %d (%.2f%%):" % (self.currentGeneration, percent)
message = "Gen. %3d: " % (self.currentGeneration)
logging.info(message)
print message,
sys_stdout.flush()
self.internalPop.statistics()
stat_ret = self.internalPop.printStats()
return message + stat_ret
def printTimeElapsed(self):
""" Shows the time elapsed since the begin of evolution """
total_time = time()-self.time_init
print "Total time elapsed: %.3f seconds." % total_time
return total_time
def evolve(self, freq_stats=0):
""" Do all the generations until the termination criteria, accepts
the freq_stats (default is 0) to dump statistics at n-generation
Example:
>>> ga_engine.evolve(freq_stats=10)
(...)
:param freq_stats: if greater than 0, the statistics will be
printed every freq_stats generation.
:rtype: returns the best individual of the evolution
.. versionadded:: 0.6
the return of the best individual
"""
stopFlagCallback = False
stopFlagTerminationCriteria = False
self.time_init = time()
if self.getGPMode():
gp_function_prefix = self.getParam("gp_function_prefix")
if gp_function_prefix is not None:
self.__gp_catch_functions(gp_function_prefix)
self.initialize()
logging.debug("Starting loop over evolutionary algorithm.")
try:
while True:
if not self.stepCallback.isEmpty():
for it in self.stepCallback.applyFunctions(self):
stopFlagCallback = it
if not self.terminationCriteria.isEmpty():
for it in self.terminationCriteria.applyFunctions(self):
stopFlagTerminationCriteria = it
if freq_stats:
if (self.currentGeneration % freq_stats == 0) or (self.getCurrentGeneration() == 0):
self.printStats()
if stopFlagTerminationCriteria:
logging.debug("Evolution stopped by the Termination Criteria !")
if freq_stats:
print "\n\tEvolution stopped by Termination Criteria function !\n"
break
if stopFlagCallback:
logging.debug("Evolution stopped by Step Callback function !")
if freq_stats:
print "\n\tEvolution stopped by Step Callback function !\n"
break
if self.interactiveMode:
if sys_platform[:3] == "win":
if msvcrt.kbhit():
if ord(msvcrt.getch()) == Consts.CDefESCKey:
print "Loading modules for Interactive Mode...",
logging.debug("Windows Interactive Mode key detected ! generation=%d", self.getCurrentGeneration())
from gp import Interaction
print " done !"
interact_banner = "## Pyevolve v.%s - Interactive Mode ##\nPress CTRL-Z to quit interactive mode." % (__version__,)
session_locals = { "ga_engine" : self,
"population" : self.getPopulation(),
"pyevolve" : pyevolve,
"it" : Interaction}
print
code.interact(interact_banner, local=session_locals)
if (self.getInteractiveGeneration() >= 0) and (self.getInteractiveGeneration() == self.getCurrentGeneration()):
print "Loading modules for Interactive Mode...",
logging.debug("Manual Interactive Mode key detected ! generation=%d", self.getCurrentGeneration())
from gp import Interaction
print " done !"
interact_banner = "## Pyevolve v.%s - Interactive Mode ##" % (__version__,)
session_locals = { "ga_engine" : self,
"population" : self.getPopulation(),
"pyevolve" : pyevolve,
"it" : Interaction}
print
code.interact(interact_banner, local=session_locals)
if self.step(): break #exit if the number of generations is equal to the max. number of gens.
except KeyboardInterrupt:
logging.debug("CTRL-C detected, finishing evolution.")
if freq_stats: print "\n\tA break was detected, you have interrupted the evolution !\n"
if freq_stats != 0:
self.printStats()
self.printTimeElapsed()
return self.bestIndividual()
def select(self, **args):
""" Select one individual from population
:param args: this parameters will be sent to the selector
"""
for it in self.selector.applyFunctions(self.internalPop, **args):
return it
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
if (MPI.COMM_WORLD.Get_rank() == 0): ### assumes WE are on top of hierarchy!
popsize = len(self.internalPop)
numAdded = 0
maxTries = 1000
numTries = 0
crossover_empty = self.select(popID=self.currentGeneration).crossover.isEmpty()
###TODO: enforce constraints!###
while numAdded < popsize:
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
if not crossover_empty and Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
# logging.debug("done cloning")
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
if (numTries > maxTries or self.owner.eval_constraints(sister)):
newPop.internalPop.append(sister)
numAdded += 1
print "successfully added sister"
if (numAdded < popsize and (numTries > maxTries or self.owner.eval_constraints(brother))):
newPop.internalPop.append(brother)
print "successfully added brother"
numAdded += 1
numTries += 1
#end rank0 onlye
# print "rank %d start eval pop" % MPI.COMM_WORLD.Get_rank()
logging.debug("Evaluating the newly created population.")
newPop.evaluate()
# print "rank %d done eval pop" % MPI.COMM_WORLD.Get_rank()
# if (MPI.COMM_WORLD.Get_rank() == 0):
# print "after eval, new pop's positions are:"
# for p in newPop:
# print p.wt_positions
if (MPI.COMM_WORLD.Get_rank() == 0): ### assumes WE are on top of hierarchy!
if self.elitism:
logging.debug("Doing elitism, %d" % self.nElitismReplacement)
if self.getMinimax() == Consts.minimaxType["maximize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
# self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score > newPop.bestRaw(i).score:
newPop[len(newPop)-1-i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
#re-evaluate before being sure this is the best
# self.internalPop.bestRaw(i).evaluate()
if self.internalPop.bestRaw(i).score < newPop.bestRaw(i).score:
newPop[len(newPop)-1-i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
if (MPI.COMM_WORLD.Get_rank() == 0): ### assumes WE are on top of hierarchy!
self.internalPop.sort()
# if (MPI.COMM_WORLD.Get_rank() == 0):
# print "after sort, internal pop's positions are:"
# for p in self.internalPop:
# print p.wt_positions
logging.debug("The generation %d was finished.", self.currentGeneration)
self.currentGeneration += 1
if (MPI.COMM_WORLD.Get_rank() == 0): ### assumes WE are on top of hierarchy!
self.saveState ()
return (self.currentGeneration == self.nGenerations)
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
logging.debug("Population was cloned.")
size_iterate = len(self.internalPop)
# Odd population size
if size_iterate % 2 != 0: size_iterate -= 1
#Check on the crossover function by picking a random individual - is it empty?
crossover_empty = self.select(popID=self.currentGeneration).crossover.isEmpty()
for i in xrange(0, size_iterate, 2):
#Ok, we select 2 parents using the selector (RouletteWheel, etc.)
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
#Crossover all of them
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
#Filp a coin each time to determine if you should crossover
if not crossover_empty and Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
#And "pre" mutate them
sister.premutate(pmut=self.pPreMutation, ga_engine=self)
brother.premutate(pmut=self.pPreMutation, ga_engine=self)
#Now each offspring is mutated
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
newPop.internalPop.append(brother)
if len(self.internalPop) % 2 != 0:
#Odd-numbered population
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom, dad=genomeDad, count=1):
(sister, brother) = it
else:
sister = random.choice([genomeMom, genomeDad])
sister = sister.clone()
#Do the 2 mutations
sister.premutate(pmut=self.pPreMutation, ga_engine=self)
sister.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
#---- Evaluate fitness ------
logging.debug("Evaluating the new created population.")
newPop.evaluate()
#Niching methods- Petrowski's clearing
self.clear()
if self.elitism:
#Avoid too much elitism
if self.nElitismReplacement >= len(self.internalPop):
self.nElitismReplacement = len(self.internalPop)-1
logging.debug("Doing elitism.")
if self.getMinimax() == Consts.minimaxType["maximize"]:
#Replace the n-th worst new ones with the nth best old ones
for i in xrange(self.nElitismReplacement):
if self.internalPop.bestRaw(i).score > newPop.bestRaw(i).score:
newPop[len(newPop)-1-i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
if self.internalPop.bestRaw(i).score < newPop.bestRaw(i).score:
newPop[len(newPop)-1-i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
self.internalPop.sort()
logging.debug("The generation %d was finished.", self.currentGeneration)
self.currentGeneration += 1
return (self.currentGeneration >= self.nGenerations)
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
logging.debug("Population was cloned.")
size_iterate = len(self.internalPop)
# Odd population size
if size_iterate % 2 != 0: size_iterate -= 1
#Check on the crossover function by picking a random individual - is it empty?
crossover_empty = self.select(
popID=self.currentGeneration).crossover.isEmpty()
for i in xrange(0, size_iterate, 2):
#Ok, we select 2 parents using the selector (RouletteWheel, etc.)
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
#Crossover all of them
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=2):
(sister, brother) = it
else:
#Filp a coin each time to determine if you should crossover
if not crossover_empty and Util.randomFlipCoin(
self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
#And "pre" mutate them
sister.premutate(pmut=self.pPreMutation, ga_engine=self)
brother.premutate(pmut=self.pPreMutation, ga_engine=self)
#Now each offspring is mutated
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
newPop.internalPop.append(brother)
if len(self.internalPop) % 2 != 0:
#Odd-numbered population
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=1):
(sister, brother) = it
else:
sister = random.choice([genomeMom, genomeDad])
sister = sister.clone()
#Do the 2 mutations
sister.premutate(pmut=self.pPreMutation, ga_engine=self)
sister.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
#---- Evaluate fitness ------
logging.debug("Evaluating the new created population.")
newPop.evaluate()
#Niching methods- Petrowski's clearing
self.clear()
if self.elitism:
#Avoid too much elitism
if self.nElitismReplacement >= len(self.internalPop):
self.nElitismReplacement = len(self.internalPop) - 1
logging.debug("Doing elitism.")
if self.getMinimax() == Consts.minimaxType["maximize"]:
#Replace the n-th worst new ones with the nth best old ones
for i in xrange(self.nElitismReplacement):
if self.internalPop.bestRaw(i).score > newPop.bestRaw(
i).score:
newPop[len(newPop) - 1 -
i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
if self.internalPop.bestRaw(i).score < newPop.bestRaw(
i).score:
newPop[len(newPop) - 1 -
i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
self.internalPop.sort()
logging.debug("The generation %d was finished.",
self.currentGeneration)
self.currentGeneration += 1
return (self.currentGeneration >= self.nGenerations)
class GSimpleGA:
""" GA Engine Class - The Genetic Algorithm Core
Example:
>>> ga = GSimpleGA.GSimpleGA(genome)
>>> ga.selector.set(Selectors.GRouletteWheel)
>>> ga.setGenerations(120)
>>> ga.terminationCriteria.set(GSimpleGA.ConvergenceCriteria)
:param genome: the :term:`Sample Genome`
:param interactiveMode: this flag enables the Interactive Mode, the default is True
:param seed: the random seed value
.. note:: if you use the same random seed, all the runs of algorithm will be the same
"""
selector = None
""" This is the function slot for the selection method
if you want to change the default selector, you must do this: ::
ga_engine.selector.set(Selectors.GRouletteWheel) """
stepCallback = None
""" This is the :term:`step callback function` slot,
if you want to set the function, you must do this: ::
def your_func(ga_engine):
# Here you have access to the GA Engine
return False
ga_engine.stepCallback.set(your_func)
now *"your_func"* will be called every generation.
When this function returns True, the GA Engine will stop the evolution and show
a warning, if is False, the evolution continues.
"""
terminationCriteria = None
""" This is the termination criteria slot, if you want to set one
termination criteria, you must do this: ::
ga_engine.terminationCriteria.set(GSimpleGA.ConvergenceCriteria)
Now, when you run your GA, it will stop when the population converges.
There are those termination criteria functions: :func:`GSimpleGA.RawScoreCriteria`, :func:`GSimpleGA.ConvergenceCriteria`, :func:`GSimpleGA.RawStatsCriteria`, :func:`GSimpleGA.FitnessStatsCriteria`
But you can create your own termination function, this function receives
one parameter which is the GA Engine, follows an example: ::
def ConvergenceCriteria(ga_engine):
pop = ga_engine.getPopulation()
return pop[0] == pop[len(pop)-1]
When this function returns True, the GA Engine will stop the evolution and show
a warning, if is False, the evolution continues, this function is called every
generation.
"""
def __init__(self, genome, seed=None, interactiveMode=True):
""" Initializator of GSimpleGA """
if seed: random.seed(seed)
if type(interactiveMode) != BooleanType:
Util.raiseException(
"Interactive Mode option must be True or False", TypeError)
if not isinstance(genome, GenomeBase):
Util.raiseException("The genome must be a GenomeBase subclass",
TypeError)
self.internalPop = GPopulation(genome)
self.nGenerations = Consts.CDefGAGenerations
self.pMutation = Consts.CDefGAMutationRate
self.pCrossover = Consts.CDefGACrossoverRate
self.nElitismReplacement = Consts.CDefGAElitismReplacement
self.setPopulationSize(Consts.CDefGAPopulationSize)
self.minimax = Consts.minimaxType["maximize"]
self.elitism = True
# Adapters
self.dbAdapter = None
self.migrationAdapter = None
self.time_init = None
self.interactiveMode = interactiveMode
self.interactiveGen = -1
self.GPMode = False
self.selector = FunctionSlot("Selector")
self.stepCallback = FunctionSlot("Generation Step Callback")
self.terminationCriteria = FunctionSlot("Termination Criteria")
self.selector.set(Consts.CDefGASelector)
self.allSlots = [
self.selector, self.stepCallback, self.terminationCriteria
]
self.internalParams = {}
self.currentGeneration = 0
# GP Testing
for classes in Consts.CDefGPGenomes:
if isinstance(self.internalPop.oneSelfGenome, classes):
self.setGPMode(True)
break
logging.debug("A GA Engine was created, nGenerations=%d",
self.nGenerations)
def setGPMode(self, bool_value):
""" Sets the Genetic Programming mode of the GA Engine
:param bool_value: True or False
"""
self.GPMode = bool_value
def getGPMode(self):
""" Get the Genetic Programming mode of the GA Engine
:rtype: True or False
"""
return self.GPMode
def __call__(self, *args, **kwargs):
""" A method to implement a callable object
Example:
>>> ga_engine(freq_stats=10)
.. versionadded:: 0.6
The callable method.
"""
if kwargs.get("freq_stats", None):
return self.evolve(kwargs.get("freq_stats"))
else:
return self.evolve()
def setParams(self, **args):
""" Set the internal params
Example:
>>> ga.setParams(gp_terminals=['x', 'y'])
:param args: params to save
..versionaddd:: 0.6
Added the *setParams* method.
"""
self.internalParams.update(args)
def getParam(self, key, nvl=None):
""" Gets an internal parameter
Example:
>>> ga.getParam("gp_terminals")
['x', 'y']
:param key: the key of param
:param nvl: if the key doesn't exist, the nvl will be returned
..versionaddd:: 0.6
Added the *getParam* method.
"""
return self.internalParams.get(key, nvl)
def setInteractiveGeneration(self, generation):
""" Sets the generation in which the GA must enter in the
Interactive Mode
:param generation: the generation number, use "-1" to disable
.. versionadded::0.6
The *setInteractiveGeneration* method.
"""
if generation < -1:
Util.raiseException("Generation must be >= -1", ValueError)
self.interactiveGen = generation
def getInteractiveGeneration(self):
""" returns the generation in which the GA must enter in the
Interactive Mode
:rtype: the generation number or -1 if not set
.. versionadded::0.6
The *getInteractiveGeneration* method.
"""
return self.interactiveGen
def setElitismReplacement(self, numreplace):
""" Set the number of best individuals to copy to the next generation on the elitism
:param numreplace: the number of individuals
.. versionadded:: 0.6
The *setElitismReplacement* method.
"""
if numreplace < 1:
Util.raiseException("Replacement number must be >= 1", ValueError)
self.nElitismReplacement = numreplace
def setInteractiveMode(self, flag=True):
""" Enable/disable the interactive mode
:param flag: True or False
.. versionadded: 0.6
The *setInteractiveMode* method.
"""
if type(flag) != BooleanType:
Util.raiseException(
"Interactive Mode option must be True or False", TypeError)
self.interactiveMode = flag
def __repr__(self):
""" The string representation of the GA Engine """
ret = "- GSimpleGA\n"
ret += "\tGP Mode:\t\t %s\n" % self.getGPMode()
ret += "\tPopulation Size:\t %d\n" % (self.internalPop.popSize, )
ret += "\tGenerations:\t\t %d\n" % (self.nGenerations, )
ret += "\tCurrent Generation:\t %d\n" % (self.currentGeneration, )
ret += "\tMutation Rate:\t\t %.2f\n" % (self.pMutation, )
ret += "\tCrossover Rate:\t\t %.2f\n" % (self.pCrossover, )
ret += "\tMinimax Type:\t\t %s\n" % (Consts.minimaxType.keys()[
Consts.minimaxType.values().index(self.minimax)].capitalize(), )
ret += "\tElitism:\t\t %s\n" % (self.elitism, )
ret += "\tElitism Replacement:\t %d\n" % (self.nElitismReplacement, )
ret += "\tDB Adapter:\t\t %s\n" % (self.dbAdapter, )
for slot in self.allSlots:
ret += "\t" + slot.__repr__()
ret += "\n"
return ret
def setMultiProcessing(self,
flag=True,
full_copy=False,
number_of_processes=-1):
""" Sets the flag to enable/disable the use of python multiprocessing module.
Use this option when you have more than one core on your CPU and when your
evaluation function is very slow.
Pyevolve will automaticly check if your Python version has **multiprocessing**
support and if you have more than one single CPU core. If you don't have support
or have just only one core, Pyevolve will not use the **multiprocessing**
feature.
Pyevolve uses the **multiprocessing** to execute the evaluation function over
the individuals, so the use of this feature will make sense if you have a
truly slow evaluation function (which is commom in GAs).
The parameter "full_copy" defines where the individual data should be copied back
after the evaluation or not. This parameter is useful when you change the
individual in the evaluation function.
:param flag: True (default) or False
:param full_copy: True or False (default)
:param number_of_processes: None = use the default, or specify the number
.. warning:: Use this option only when your evaluation function is slow, so you'll
get a good tradeoff between the process communication speed and the
parallel evaluation. The use of the **multiprocessing** doesn't means
always a better performance.
.. note:: To enable the multiprocessing option, you **MUST** add the *__main__* check
on your application, otherwise, it will result in errors. See more on the
`Python Docs <http://docs.python.org/library/multiprocessing.html#multiprocessing-programming>`__
site.
.. versionadded:: 0.6
The `setMultiProcessing` method.
"""
if type(flag) != BooleanType:
Util.raiseException("Multiprocessing option must be True or False",
TypeError)
if type(full_copy) != BooleanType:
Util.raiseException(
"Multiprocessing 'full_copy' option must be True or False",
TypeError)
self.internalPop.setMultiProcessing(flag, full_copy,
number_of_processes)
def setMigrationAdapter(self, migration_adapter=None):
""" Sets the Migration Adapter
.. versionadded:: 0.6
The `setMigrationAdapter` method.
"""
if (migration_adapter is not None) and (not isinstance(
migration_adapter, MigrationScheme)):
Util.raiseException(
"The Migration Adapter must be a MigrationScheme subclass",
TypeError)
self.migrationAdapter = migration_adapter
if self.migrationAdapter is not None:
self.migrationAdapter.setGAEngine(self)
def setDBAdapter(self, dbadapter=None):
""" Sets the DB Adapter of the GA Engine
:param dbadapter: one of the :mod:`DBAdapters` classes instance
.. warning:: the use the of a DB Adapter can reduce the speed performance of the
Genetic Algorithm.
"""
if (dbadapter
is not None) and (not isinstance(dbadapter, DBBaseAdapter)):
Util.raiseException(
"The DB Adapter must be a DBBaseAdapter subclass", TypeError)
self.dbAdapter = dbadapter
def setPopulationSize(self, size):
""" Sets the population size, calls setPopulationSize() of GPopulation
:param size: the population size
.. note:: the population size must be >= 2
"""
if size < 2:
Util.raiseException("population size must be >= 2", ValueError)
self.internalPop.setPopulationSize(size)
def setSortType(self, sort_type):
""" Sets the sort type, Consts.sortType["raw"]/Consts.sortType["scaled"]
Example:
>>> ga_engine.setSortType(Consts.sortType["scaled"])
:param sort_type: the Sort Type
"""
if sort_type not in Consts.sortType.values():
Util.raiseException("sort type must be a Consts.sortType type",
TypeError)
self.internalPop.sortType = sort_type
def setMutationRate(self, rate):
""" Sets the mutation rate, between 0.0 and 1.0
:param rate: the rate, between 0.0 and 1.0
"""
if (rate > 1.0) or (rate < 0.0):
Util.raiseException("Mutation rate must be >= 0.0 and <= 1.0",
ValueError)
self.pMutation = rate
def setPreMutationRate(self, rate):
""" Sets the pre-mutation rate, > 0.0.
Premutation is done only when no cross-over occurs.
:param rate: the rate, >= 0.0
"""
if (rate < 0.0):
Util.raiseException("Pre-mutation rate must be >= 0.0", ValueError)
self.pPreMutation = rate
def setCrossoverRate(self, rate):
""" Sets the crossover rate, between 0.0 and 1.0
:param rate: the rate, between 0.0 and 1.0
"""
if (rate > 1.0) or (rate < 0.0):
Util.raiseException("Crossover rate must be >= 0.0 and <= 1.0",
ValueError)
self.pCrossover = rate
def setGenerations(self, num_gens):
""" Sets the number of generations to evolve
:param num_gens: the number of generations
"""
if num_gens < 1:
Util.raiseException("Number of generations must be >= 1",
ValueError)
self.nGenerations = num_gens
def getGenerations(self):
""" Return the number of generations to evolve
:rtype: the number of generations
.. versionadded:: 0.6
Added the *getGenerations* method
"""
return self.nGenerations
def getMinimax(self):
""" Gets the minimize/maximize mode
:rtype: the Consts.minimaxType type
"""
return self.minimax
def setMinimax(self, mtype):
""" Sets the minimize/maximize mode, use Consts.minimaxType
:param mtype: the minimax mode, from Consts.minimaxType
"""
if mtype not in Consts.minimaxType.values():
Util.raiseException("Minimax must be maximize or minimize",
TypeError)
self.minimax = mtype
def getCurrentGeneration(self):
""" Gets the current generation
:rtype: the current generation
"""
return self.currentGeneration
def setElitism(self, flag):
""" Sets the elitism option, True or False
:param flag: True or False
"""
if type(flag) != BooleanType:
Util.raiseException("Elitism option must be True or False",
TypeError)
self.elitism = flag
def getDBAdapter(self):
""" Gets the DB Adapter of the GA Engine
:rtype: a instance from one of the :mod:`DBAdapters` classes
"""
return self.dbAdapter
def bestIndividual(self):
""" Returns the population best individual
:rtype: the best individual
"""
return self.internalPop.bestRaw()
def __gp_catch_functions(self, prefix):
""" Internally used to catch functions with some specific prefix
as non-terminals of the GP core """
import __main__ as mod_main
function_set = {}
main_dict = mod_main.__dict__
for obj, addr in main_dict.items():
if obj[0:len(prefix)] == prefix:
try:
op_len = addr.func_code.co_argcount
except:
continue
function_set[obj] = op_len
if len(function_set) <= 0:
Util.raiseException(
"No function set found using function prefix '%s' !" % prefix,
ValueError)
self.setParams(gp_function_set=function_set)
def initialize(self):
""" Initializes the GA Engine. Create and initialize population """
self.internalPop.create(minimax=self.minimax)
self.internalPop.initialize(ga_engine=self)
logging.debug("The GA Engine was initialized !")
def getPopulation(self):
""" Return the internal population of GA Engine
:rtype: the population (:class:`GPopulation.GPopulation`)
"""
return self.internalPop
def getStatistics(self):
""" Gets the Statistics class instance of current generation
:rtype: the statistics instance (:class:`Statistics.Statistics`)
"""
return self.internalPop.getStatistics()
def clear(self):
""" Petrowski's Clearing Method """
def step(self):
""" Just do one step in evolution, one generation """
genomeMom = None
genomeDad = None
newPop = GPopulation(self.internalPop)
logging.debug("Population was cloned.")
size_iterate = len(self.internalPop)
# Odd population size
if size_iterate % 2 != 0: size_iterate -= 1
#Check on the crossover function by picking a random individual - is it empty?
crossover_empty = self.select(
popID=self.currentGeneration).crossover.isEmpty()
for i in xrange(0, size_iterate, 2):
#Ok, we select 2 parents using the selector (RouletteWheel, etc.)
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if not crossover_empty and self.pCrossover >= 1.0:
#Crossover all of them
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=2):
(sister, brother) = it
else:
#Filp a coin each time to determine if you should crossover
if not crossover_empty and Util.randomFlipCoin(
self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=2):
(sister, brother) = it
else:
sister = genomeMom.clone()
brother = genomeDad.clone()
#And "pre" mutate them
sister.premutate(pmut=self.pPreMutation, ga_engine=self)
brother.premutate(pmut=self.pPreMutation, ga_engine=self)
#Now each offspring is mutated
sister.mutate(pmut=self.pMutation, ga_engine=self)
brother.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
newPop.internalPop.append(brother)
if len(self.internalPop) % 2 != 0:
#Odd-numbered population
genomeMom = self.select(popID=self.currentGeneration)
genomeDad = self.select(popID=self.currentGeneration)
if Util.randomFlipCoin(self.pCrossover):
for it in genomeMom.crossover.applyFunctions(mom=genomeMom,
dad=genomeDad,
count=1):
(sister, brother) = it
else:
sister = random.choice([genomeMom, genomeDad])
sister = sister.clone()
#Do the 2 mutations
sister.premutate(pmut=self.pPreMutation, ga_engine=self)
sister.mutate(pmut=self.pMutation, ga_engine=self)
newPop.internalPop.append(sister)
#---- Evaluate fitness ------
logging.debug("Evaluating the new created population.")
newPop.evaluate()
#Niching methods- Petrowski's clearing
self.clear()
if self.elitism:
#Avoid too much elitism
if self.nElitismReplacement >= len(self.internalPop):
self.nElitismReplacement = len(self.internalPop) - 1
logging.debug("Doing elitism.")
if self.getMinimax() == Consts.minimaxType["maximize"]:
#Replace the n-th worst new ones with the nth best old ones
for i in xrange(self.nElitismReplacement):
if self.internalPop.bestRaw(i).score > newPop.bestRaw(
i).score:
newPop[len(newPop) - 1 -
i] = self.internalPop.bestRaw(i)
elif self.getMinimax() == Consts.minimaxType["minimize"]:
for i in xrange(self.nElitismReplacement):
if self.internalPop.bestRaw(i).score < newPop.bestRaw(
i).score:
newPop[len(newPop) - 1 -
i] = self.internalPop.bestRaw(i)
self.internalPop = newPop
self.internalPop.sort()
logging.debug("The generation %d was finished.",
self.currentGeneration)
self.currentGeneration += 1
return (self.currentGeneration >= self.nGenerations)
def printStats(self):
""" Print generation statistics
:rtype: the printed statistics as string
.. versionchanged:: 0.6
The return of *printStats* method.
"""
percent = self.currentGeneration * 100 / float(self.nGenerations)
message = "Gen. %d (%.2f%%):" % (self.currentGeneration, percent)
logging.info(message)
print message,
sys_stdout.flush()
self.internalPop.statistics()
stat_ret = self.internalPop.printStats()
return message + stat_ret
def printTimeElapsed(self):
""" Shows the time elapsed since the begin of evolution """
total_time = time() - self.time_init
print "Total time elapsed: %.3f seconds." % total_time
return total_time
def dumpStatsDB(self):
""" Dumps the current statistics to database adapter """
logging.debug("Dumping stats to the DB Adapter")
self.internalPop.statistics()
self.dbAdapter.insert(self)
#----------------------------------------------------------
def evolve(self, freq_stats=0, skip_initialize=False):
""" Do all the generations until the termination criteria, accepts
the freq_stats (default is 0) to dump statistics at n-generation
Example:
>>> ga_engine.evolve(freq_stats=10)
(...)
:param freq_stats: if greater than 0, the statistics will be
printed every freq_stats generation.
:rtype: returns a tuple:
best: the best individual of the evolution
stopFlagCallback: stopped by the step_callback
stopFlagTerminationCriteria: stopped by reching the criterion
.. versionadded:: 0.6
the return of the best individual
"""
stopFlagCallback = False
stopFlagTerminationCriteria = False
self.time_init = time()
logging.debug(
"Starting the DB Adapter and the Migration Adapter if any")
if self.dbAdapter: self.dbAdapter.open(self)
if self.migrationAdapter: self.migrationAdapter.start()
if self.getGPMode():
gp_function_prefix = self.getParam("gp_function_prefix")
if gp_function_prefix is not None:
self.__gp_catch_functions(gp_function_prefix)
if not skip_initialize:
#Create the population
self.initialize()
#Initial fitness evaluation
self.internalPop.evaluate()
self.internalPop.sort()
logging.debug("Starting loop over evolutionary algorithm.")
try:
while True:
if self.migrationAdapter:
logging.debug("Migration adapter: exchange")
self.migrationAdapter.exchange()
self.internalPop.clearFlags()
self.internalPop.sort()
#The step callback is called before each step
if not self.stepCallback.isEmpty():
for it in self.stepCallback.applyFunctions(self):
stopFlagCallback = it
if not self.terminationCriteria.isEmpty():
for it in self.terminationCriteria.applyFunctions(self):
stopFlagTerminationCriteria = it
if freq_stats:
if (self.currentGeneration % freq_stats
== 0) or (self.getCurrentGeneration() == 0):
self.printStats()
if self.dbAdapter:
if self.currentGeneration % self.dbAdapter.getStatsGenFreq(
) == 0:
self.dumpStatsDB()
if stopFlagTerminationCriteria:
logging.debug(
"Evolution stopped by the Termination Criteria !")
if freq_stats:
print "\n\tEvolution stopped by Termination Criteria function !\n"
break
if stopFlagCallback:
logging.debug(
"Evolution stopped by Step Callback function !")
if freq_stats:
print "\n\tEvolution stopped by Step Callback function !\n"
break
#(Here was interactive mode code, removed)
if self.step():
break #exit if the number of generations is equal to the max. number of gens.
#(end While True)
except KeyboardInterrupt:
logging.debug("CTRL-C detected, finishing evolution.")
if freq_stats:
print "\n\tA break was detected, you have interrupted the evolution !\n"
#Finished. Clean up the multiprocessing pool
self.getPopulation().cleanupMultiProcessing()
if freq_stats != 0:
self.printStats()
self.printTimeElapsed()
if self.dbAdapter:
logging.debug("Closing the DB Adapter")
if not (self.currentGeneration % self.dbAdapter.getStatsGenFreq()
== 0):
self.dumpStatsDB()
self.dbAdapter.commitAndClose()
if self.migrationAdapter:
logging.debug("Closing the Migration Adapter")
if freq_stats: print "Stopping the migration adapter... ",
self.migrationAdapter.stop()
if freq_stats: print "done !"
return (self.bestIndividual(), stopFlagCallback,
stopFlagTerminationCriteria)
def select(self, **args):
""" Select one individual from population
:param args: this parameters will be sent to the selector
"""
for it in self.selector.applyFunctions(self.internalPop, **args):
return it
