def __init__(self, model, activationfn=None):
if isinstance(model, Model):
self.Model = model
self.Output_Options = []
self.Input_Options = []
self.Callback = None
self.LastLayerId = None
self.FirstLayerId = None
self.LastNodes = []
self.ParallelExecute = False
self.previous_accuracy = 0.0
self.current_accuracy = 0.0
self.counter_ep = 1
self.optimum_pass = 10
self.activationfn = activationfn()
self.setActivationFunction(activationfn)
#save a ref copy of all nodes of model.
self.SaveTensors = SaveTensors(
allTNodes=self.Model.get_AllTNodes())
Optimizer.__init__(self,
model=model,
SavedTensors=self.SaveTensors,
optimizer='gradientdecent',
activationfn=self.activationfn)
else:
raise RuntimeError("passed model is not of type Model")
def __init__(self, chromosome, resEval, penEval, fitnessEval, popSize, elitism):
Optimizer.__init__(self)
self.Chromosome = chromosome
self.ResEval = resEval
self.PenEval = penEval
self.FitnessEval = fitnessEval
self.PopulationSize = popSize
self.Elitism = elitism
self.SelectionOp = TournamentSelectionOperator(2, 2)
self.RecombinationOp = DrunkRecombinationOperator()
self.MutationOp = None
self.Population = None
self.PenEval.ConstraintPenalties.append(Penalty("LOWER_BOUND", "__SUCCESS__", 1.0, 1.0, 0.5))
def __init__(self, **kargs):
Optimizer.__init__(self, **kargs)
self.epochs = 0
self.npop = 10
if "population" in kargs:
self.npop = kargs["population"]
self.mutationrate = 0.2
if "mutationrate" in kargs:
self.mutationrate = kargs["mutationrate"]
self.threads = 4
if "threads" in kargs:
self.threads = kargs["threads"]
self.population = []
self.MAX_INT = 2**32 if not self.maximize else -2**32
for _ in range(0, self.npop):
self.population.append([self.getInstance(), self.MAX_INT])
self.lastscore = self.MAX_INT
def __init__(self, **kargs):
Optimizer.__init__(self, **kargs)
randomobj = self.getInstance()
ev = randomobj.evaluate(self.objectives, self.penalties, self.maximize)
self.bestFitness = ev
self.bestObj = randomobj
def __init__(self, log_level=0):
'''Constructor'''
Optimizer.__init__(self, log_level)
