def __init__(self, netArch,
maxIterations,
patterns,
probability = 0.01,
monitor = None,
learningRate = 0.5,
momentum = 0.0,
batch = False,
randomize = True ):
"""
Build the Backpropagation mutator
@author Thiago F Pappacena
@param tuple netArch The network archicteture (returned by Network.getDimension method)
@param integer maxIterations The max number of iterations
@param list patterns The list of patterns to use in training
@param float probability The probability of operating
@see pyann.mlp.trainer.backprop.BackpropagationTrainer
"""
super(BackpropMutation, self).__init__(probability = probability)
self.patterns = patterns
self.maxIterations = maxIterations
self.stopMaxIter = MaxIterationsStopCondition(self.maxIterations)
self.trainer = BackpropagationTrainer( monitor = monitor,
learningRate = learningRate,
momentum = momentum,
batch = batch,
randomize = randomize )
self.trainer.setStopConditions( (self.stopMaxIter, ) )
self.netArch = netArch
class BackpropMutation(Mutation):
"""
A backpropagation mutation object
@author Thiago F Pappacena
"""
def __init__(self, netArch,
maxIterations,
patterns,
probability = 0.01,
monitor = None,
learningRate = 0.5,
momentum = 0.0,
batch = False,
randomize = True ):
"""
Build the Backpropagation mutator
@author Thiago F Pappacena
@param tuple netArch The network archicteture (returned by Network.getDimension method)
@param integer maxIterations The max number of iterations
@param list patterns The list of patterns to use in training
@param float probability The probability of operating
@see pyann.mlp.trainer.backprop.BackpropagationTrainer
"""
super(BackpropMutation, self).__init__(probability = probability)
self.patterns = patterns
self.maxIterations = maxIterations
self.stopMaxIter = MaxIterationsStopCondition(self.maxIterations)
self.trainer = BackpropagationTrainer( monitor = monitor,
learningRate = learningRate,
momentum = momentum,
batch = batch,
randomize = randomize )
self.trainer.setStopConditions( (self.stopMaxIter, ) )
self.netArch = netArch
# end :: __init__
def modify(self, population):
"""
Apply the backpropagation mutation in someone in population
@author Thiago F Pappacena
@param object population The population object
@return void
"""
if random.random() <= self.probability:
self.stopMaxIter.setIterations( random.randrange(1, self.maxIterations) )
chromo = random.choice(population.getChromossomes())
genes = chromo.getGenes()
weights = tuple(x.getValue() for x in genes)
# rebuild the net and train
net = Network.fromPlanifiedWeights(self.netArch, weights)
self.trainer.setNetwork(net)
self.trainer.train(self.patterns)
self.netArch = net.getDimension()
# copy values to genes
for gene, weight in zip(genes, net.planify()):
gene.setValue(weight)
chromo.setFitness(None)
