def printStats(self):
"""
Override printStats so we get an extended summary
"""
GSimpleGA.printStats(self)
self.printSimulationRate()
self.printShapeSummary()
def main():
genome = G2DList(7, 4)
genome.setParams(rangemin=0, rangemax=MAX_SLOTS)
genome.evaluator.set(eval_function)
genome.crossover.set(Crossovers.G2DListCrossoverSingleHPoint)
# genome.mutator.set(Mutators.G2DListMutatorSwap)
genome.mutator.set(noop)
genome.initializator.set(PlanningInitializator)
ga = GSimpleGA(genome)
ga.setGenerations(1000)
ga.setPopulationSize(10000)
ga.terminationCriteria.set(ConvergenceCriteria)
ga.evolve(freq_stats=500)
best = ga.bestIndividual()
if best.fitness == 1.:
print('!! SUCCESS !!')
else:
print("Sorry, no valid planning found :( Here is the best so far.")
display(best.genomeList)
print('fitness: %f' % best.fitness)
def __init__(self, genome, *args, **kwargs):
# Initialize original GA engine.
GSimpleGA.__init__(self, genome, *args, **kwargs)
self.internalPop = CustomGPopulation(genome)
# Set custom evolution parameters.
self.setMinimax(minimaxType["minimize"])
self.setMutationRate(0.05)
self.setCrossoverRate(1.0)
# Useful callbacks (stats, stopping evolution).
self.stepCallback.add(stats_step_callback)
self.terminationCriteria.add(stop_flag_criteria)
# Clean previous stop flag.
stop_file = self.getParam("stop_file", None)
if stop_file and os.path.isfile(stop_file):
os.remove(stop_file)
def __init__(self, genome, *args, **kwargs):
# Initialize original GA engine.
GSimpleGA.__init__(self, genome, *args, **kwargs)
self.internalPop = CustomGPopulation(genome)
# Set custom evolution parameters.
self.setMinimax(minimaxType["minimize"])
self.setMutationRate(0.05)
self.setCrossoverRate(1.0)
# Useful callbacks (stats, stopping evolution).
self.stepCallback.add(stats_step_callback)
self.terminationCriteria.add(stop_flag_criteria)
# Clean previous stop flag.
stop_file = self.getParam("stop_file", None)
if stop_file and os.path.isfile(stop_file):
os.remove(stop_file)
def call(settings, input_dict):
"""
This function ...
:param settings:
:param input_dict:
:return:
"""
# Inform the user
log.info("Calling Pyevolve ...")
# Import pyevolve modules
from pyevolve.G1DList import G1DList
from pyevolve.G2DList import G2DList
from pyevolve.G1DBinaryString import G1DBinaryString
from pyevolve.G2DBinaryString import G2DBinaryString
from pyevolve.GSimpleGA import GSimpleGA, RawScoreCriteria
from pyevolve.Initializators import G1DListInitializatorInteger, G1DListInitializatorReal
from pyevolve.Mutators import G1DListMutatorIntegerRange, G1DListMutatorIntegerGaussian, G1DListMutatorIntegerBinary, G1DListMutatorRealRange, G1DListMutatorRealGaussian
from pyevolve.Consts import minimaxType
# Get the settings
output = settings["output"]
genome_type = settings[
"genome_type"] if "genome_type" in settings else "list"
genome_dimension = settings[
"genome_dimension"] if "genome_dimension" in settings else 1
nparameters = settings["nparameters"] if "nparameters" in settings else None
nparameters2 = settings[
"nparameters2"] if "nparameters2" in settings else None
nindividuals = settings[
"nindividuals"] if "nindividuals" in settings else None
parameter_range = settings[
"parameter_range"] if "parameter_range" in settings else None
best_raw_score = settings[
"best_raw_score"] if "best_raw_score" in settings else None
round_decimal = settings[
"round_decimal"] if "round_decimal" in settings else None
ngenerations = settings[
"ngenerations"] if "ngenerations" in settings else None
mutation_rate = settings[
"mutation_rate"] if "mutation_rate" in settings else None
crossover_rate = settings[
"crossover_rate"] if "crossover_rate" in settings else None
stats_freq = settings["stats_freq"] if "stats_freq" in settings else None
mutation_method = settings[
"mutation_method"] if "mutation_method" in settings else None
min_or_max = settings["min_or_max"] if "min_or_max" in settings else None
# No messages
stats_freq = 0
# Get the input
evaluator = input_dict["evaluator"] # this is required!
genome = input_dict["genome"].clone() if "genome" in input_dict else None
initializator = input_dict[
"initializator"] if "initializator" in input_dict else None
mutator = input_dict["mutator"] if "mutator" in input_dict else None
crossover = input_dict["crossover"] if "crossover" in input_dict else None
callback = input_dict["callback"] if "callback" in input_dict else None
# Get kwargs for the functions
evaluator_kwargs = input_dict[
"evaluator_kwargs"] if "evaluator_kwargs" in input_dict else None
initializator_kwargs = input_dict[
"initializator_kwargs"] if "initializator_kwargs" in input_dict else None
mutator_kwargs = input_dict[
"mutator_kwargs"] if "mutator_kwargs" in input_dict else None
crossover_kwargs = input_dict[
"crossover_kwargs"] if "crossover_kwargs" in input_dict else None
callback_kwargs = input_dict[
"callback_kwargs"] if "callback_kwargs" in input_dict else None
# Create partial functions with kwargs set
if evaluator_kwargs is not None:
evaluator = partial(evaluator, **evaluator_kwargs)
if initializator is not None and initializator_kwargs is not None:
initializator = partial(initializator, **initializator_kwargs)
if mutator is not None and mutator_kwargs is not None:
mutator = partial(mutator, **mutator_kwargs)
if crossover is not None and crossover_kwargs is not None:
crossover = partial(crossover, **crossover_kwargs)
if callback is not None and callback_kwargs is not None:
callback = partial(callback, **callback_kwargs)
# Create genome instance
if genome is None:
# List genome
if genome_type == "list":
# Create 1D genome
if genome_dimension == 1: genome = G1DList(nparameters)
elif genome_dimension == 2:
genome = G2DList(nparameters, nparameters2)
else:
raise ValueError("Dimensions > 2 are not supported")
# Binary string genome
elif genome_type == "binary_string":
# 1D or 2D
if genome_dimension == 1: genome = G1DBinaryString(nparameters)
elif genome_dimension == 2:
genome = G2DBinaryString(nparameters, nparameters2)
else:
raise ValueError("Dimensions > 2 are not supported")
# Invalid option
else:
raise ValueError("Genome type must be 'list' or 'binary_string")
# Set genome properties
if parameter_range is not None:
genome.setParams(rangemin=parameter_range.min,
rangemax=parameter_range.max)
if best_raw_score is not None:
genome.setParams(bestrawscore=best_raw_score)
if round_decimal is not None: genome.setParams(rounddecimal=round_decimal)
# Set initializator
if initializator is not None: genome.initializator.set(initializator)
else:
if isinstance(parameter_range, IntegerRange):
genome.initializator.set(G1DListInitializatorInteger)
elif isinstance(parameter_range, RealRange):
genome.initializator.set(G1DListInitializatorReal)
else:
raise ValueError("Invalid parameter range")
# Set mutation method
if mutator is not None: genome.mutator.set(mutator)
else:
if isinstance(parameter_range, IntegerRange):
if mutation_method == "range":
genome.mutator.set(G1DListMutatorIntegerRange)
elif mutation_method == "gaussian":
genome.mutator.set(G1DListMutatorIntegerGaussian)
elif mutation_method == "binary":
genome.mutator.set(G1DListMutatorIntegerBinary)
else:
raise ValueError("Mutation method '" + mutation_method +
"' not recognized")
elif isinstance(parameter_range, RealRange):
if mutation_method == "range":
genome.mutator.set(G1DListMutatorRealRange)
elif mutation_method == "gaussian":
genome.mutator.set(G1DListMutatorRealGaussian)
else:
raise ValueError("Mutation method '" + mutation_method +
"' not valid for genome of real values")
else:
raise ValueError("Invalid parameter range")
# Set crossover
if crossover is not None: genome.crossover.set(crossover)
# The evaluator function (objective function)
genome.evaluator.set(evaluator)
# Genetic Algorithm Instance
ga = GSimpleGA(genome)
ga.terminationCriteria.set(RawScoreCriteria)
ga.setMinimax(minimaxType[min_or_max])
ga.setGenerations(ngenerations)
if crossover_rate is not None: ga.setCrossoverRate(crossover_rate)
ga.setPopulationSize(nindividuals)
ga.setMutationRate(mutation_rate)
# Set the callback function
if callback is not None: ga.stepCallback.set(callback)
# -----------------------------------------------------------------
# Do the evolution
with Bar(label='',
width=32,
hide=None,
empty_char=BAR_EMPTY_CHAR,
filled_char=BAR_FILLED_CHAR,
expected_size=ga.nGenerations,
every=1,
add_datetime=True) as bar:
# Create partial function for the callback
refresh_progress_bar_partial = partial(refresh_progress_bar,
**{"bar": bar})
# Set callback function to print generation
ga.stepCallback.set(refresh_progress_bar_partial)
# Evolve
ga.evolve(freq_stats=stats_freq)
# -----------------------------------------------------------------
# Get the best individual
best = ga.bestIndividual()
# Return the best individual
return best
# -----------------------------------------------------------------
# Genome instance
genome = G1DList(2)
genome.setParams(rangemin=0.0, rangemax=50.0, bestrawscore=0.00, rounddecimal=2)
genome.initializator.set(Initializators.G1DListInitializatorReal)
genome.mutator.set(Mutators.G1DListMutatorRealGaussian)
# Set the evaluator function
genome.evaluator.set(chi_squared_function)
# Genetic algorithm instance
# seed = 4357
ga = GSimpleGA(genome, seed=config.seed)
ga.terminationCriteria.set(RawScoreCriteria)
ga.setMinimax(Consts.minimaxType["minimize"])
ga.setGenerations(5)
ga.setCrossoverRate(0.5)
ga.setPopulationSize(100)
ga.setMutationRate(0.5)
# Evolve
ga.evolve(freq_stats=1)
print("Final generation:", ga.currentGeneration)
# -----------------------------------------------------------------
# Determine the path to the reference directory
if absError <= 1:
fitness += 1 - absError # Removes as many percent as the network is incorrect from the fitness
else:
fitness -= 1.0 / len(
DS) # If the network is too inaccurate it is instead punished
return fitness
genome = G1DList(4) # We'll change all 4 weights in the network
genome.setParams(rangemin=-10, rangemax=10)
genome.initializator.set(
G1DListInitializatorReal) # Initializes the genes with a Real value
genome.mutator.set(
G1DListMutatorRealGaussian) # Uses a Real value Gaussian mutator
genome.evaluator.set(testNetwork)
#genome.crossover.clear() # Remove the crossover function, not possible if our list only has 1 element, but it seems like crossover improves the results a lot if enabled
ga = GSimpleGA(genome)
ga.setElitism(
True
) # Uses Elitism, so a few of the successful inidividuals get carried over
ga.selector.set(
GRouletteWheel
) # Roulette wheel selector, see http://www.edc.ncl.ac.uk/assets/hilite_graphics/rhjan07g02.png
ga.nGenerations = 100 # Number of generations
ga.evolve(freq_stats=10) # Run 10 generations and then show stats
testNetwork(
ga.bestIndividual()) # Just fetch the network of the best individual
for i in range(len(DS)):
print "Target: %s, Result: %s" % (DS['target'][i],
net.activate(DS['input'][i]))
max_streak = max(max_streak, streak)
streak = 0
else:
streak += 1
prev = v
return max_streak
def initializer_shuffled(genome, **args):
l = range(genome.getListSize())
random.shuffle(l)
genome.setInternalList(l)
if __name__ == "__main__":
list_length = 20
genome = G1DList(list_length)
genome.evaluator.set(sortedness)
genome.initializator.set(initializer_shuffled)
genome.mutator.set(Mutators.G1DListMutatorSwap)
genome.crossover.clear()
ga = GSimpleGA(genome)
ga.setGenerations(100000)
ga.evolve(freq_stats=1000)
ga.setMultiProcessing()
print ga.bestIndividual()
return chi_squared
# -----------------------------------------------------------------
# Genome instance
genome = G1DList(2)
genome.setParams(rangemin=0., rangemax=50., bestrawscore=0.00, rounddecimal=2)
genome.initializator.set(Initializators.G1DListInitializatorReal)
genome.mutator.set(Mutators.G1DListMutatorRealGaussian)
# Set the evaluator function
genome.evaluator.set(chi_squared_function)
# Genetic algorithm instance
#seed = 4357
ga = GSimpleGA(genome, seed=config.seed)
ga.terminationCriteria.set(RawScoreCriteria)
ga.setMinimax(Consts.minimaxType["minimize"])
ga.setGenerations(5)
ga.setCrossoverRate(0.5)
ga.setPopulationSize(100)
ga.setMutationRate(0.5)
# Evolve
ga.evolve(freq_stats=1)
print("Final generation:", ga.currentGeneration)
# -----------------------------------------------------------------
# Determine the path to the reference directory
net.connections[net['in']][0].params[0] = eVar[0]; net.connections[net['hidden0']][0].params[0] = eVar[1]; net.connections[net['bias']][0].params[0] = eVar[2]; net.connections[net['bias']][1].params[0] = eVar[3]; res = net.activateOnDataset(DS) fitness = 1 for i in range(len(DS['target'])): absError = abs(calcError(res[i], DS['target'][i])) if absError <= 1: fitness += 1 - absError # Removes as many percent as the network is incorrect from the fitness else: fitness -= 1.0/len(DS) # If the network is too inaccurate it is instead punished return fitness genome = G1DList(4) # We'll change all 4 weights in the network genome.setParams(rangemin=-10,rangemax=10) genome.initializator.set(G1DListInitializatorReal) # Initializes the genes with a Real value genome.mutator.set(G1DListMutatorRealGaussian) # Uses a Real value Gaussian mutator genome.evaluator.set(testNetwork) #genome.crossover.clear() # Remove the crossover function, not possible if our list only has 1 element, but it seems like crossover improves the results a lot if enabled ga = GSimpleGA(genome) ga.setElitism(True) # Uses Elitism, so a few of the successful inidividuals get carried over ga.selector.set(GRouletteWheel) # Roulette wheel selector, see http://www.edc.ncl.ac.uk/assets/hilite_graphics/rhjan07g02.png ga.nGenerations = 100 # Number of generations ga.evolve(freq_stats=10) # Run 10 generations and then show stats testNetwork(ga.bestIndividual()) # Just fetch the network of the best individual for i in range(len(DS)): print "Target: %s, Result: %s" % (DS['target'][i], net.activate(DS['input'][i]))