def _setInitEvaluable(self, evaluable):
evaluable = array(evaluable)
ContinuousOptimizer._setInitEvaluable(self, evaluable)
self.gradient.init(evaluable)
self.prevFitness = sys.float_info.max
self.currentFitness = sys.float_info.min
def _setInitEvaluable(self, evaluable):
ContinuousOptimizer._setInitEvaluable(self, evaluable)
self.current = self._initEvaluable
self.gd = GradientDescent()
self.gd.alpha = self.learningRate
if self.learningRateDecay is not None:
self.gd.alphadecay = self.learningRateDecay
self.gd.momentum = self.momentum
self.gd.rprop = self.rprop
self.gd.init(self._initEvaluable)
def _setInitEvaluable(self, evaluable):
ContinuousOptimizer._setInitEvaluable(self, evaluable)
self.current = self._initEvaluable
self.gd = GradientDescent()
self.gd.alpha = self.learningRate
if self.learningRateDecay is not None:
self.gd.alphadecay = self.learningRateDecay
self.gd.momentum = self.momentum
self.gd.rprop = self.rprop
self.gd.init(self._initEvaluable)
def _stoppingCriterion(self):
result = False
if ContinuousOptimizer._stoppingCriterion(self):
result = True
# Check if during the last step we made a significant progress
elif abs(self.prevFitness - self.currentFitness) < self.minChange:
result = True
self.prevFitness = self.currentFitness
return result
def __init__(self, evaluator=None, initEvaluable=None, gradientsCalculator=None, gradient = GradientDescent(), minChange = 1e-6, minimize = True):
"""
evaluator - the same as in superclass
initEvaluable - the same as in superclass
gradientsCalculator - function that calculates partial derivatives for each variable for specified coordinates.
It should accept the only parameter - list of current values of parameters and should return an array of partial derivatives values.
Value of partial derivative for parameter X should have the same position in the result array as parameter X has in parameters array.
gradient (GradientDescent) - class that changes values of parameters using current gradient
minChange -
"""
self.gradientsCalculator = gradientsCalculator
self.gradient = gradient
self.minChange = minChange
self.prevFitness = sys.float_info.max
self.currentFitness = sys.float_info.min
initEvaluable = array(initEvaluable)
ContinuousOptimizer.__init__(self, evaluator, initEvaluable)
self.minimize = minimize
def _setInitEvaluable(self, evaluable):
if evaluable is not None:
logging.warning("Initial point provided was ignored.")
ContinuousOptimizer._setInitEvaluable(self, evaluable)
def _setInitEvaluable(self, evaluable):
if evaluable is not None:
logging.warning("Initial point provided was ignored.")
ContinuousOptimizer._setInitEvaluable(self, evaluable)