def __init__(self, dataManager):
'''
Constructor
'''
TFMapping.__init__(self, dataManager)
SettingsClient.__init__(self)
self.iterations = 0
def __init__(self,
dataManager,
inputArguments,
outputArguments,
name='Function'):
TFMapping.__init__(self,
dataManager,
inputArguments,
outputArguments,
name=name)
def __init__(self,
dataManager,
inputArguments,
outputArguments,
name='Classifier'):
TFMapping.__init__(self,
dataManager,
inputArguments,
outputArguments,
name=name)
def _addTensorToDictionary(self, tensor):
from pypost.mappings import TFMapping
tensorMapping = TFMapping(self.dataManager,
tensorNode=tensor,
name='data_tfmapping')
self.tensorDictionary[tensor] = tensorMapping
return tensorMapping
def addFeatureMapping(self, mapping):
if isinstance(mapping, tf.Tensor):
from pypost.mappings import TFMapping
mapping = TFMapping(self, tensorNode=mapping)
outputVariable = mapping.getOutputVariables()[0]
if (not outputVariable in self.dataEntries):
if (self.subDataManager):
self.subDataManager.addFeatureMapping(mapping)
else:
raise ValueError(
'Can only add Feature Mapping for existing data entries (aliases are not supported). Current Entry %s does not exist'
% outputVariable)
else:
self.dataEntries[outputVariable].isFeature = True
self.dataEntries[outputVariable].callBackGetter = mapping
self.addDataEntry(outputVariable + '_validFlag', 1)
def __init__(self, dataManager, lossFunction, variables_list = None, name = None, printIterations=False):
super().__init__(dataManager)
self.loss = lossFunction
self.variables_list = variables_list
if (name is None):
self.name = ''
else:
self.name = name + '_'
self.linkPropertyToSettings('tfOptimizerType', globalName = self.name + 'tfOptimizerType', defaultValue=TFOptimizerType.Adam)
self.linkPropertyToSettings('tfOptimizerNumIterations', globalName = self.name + 'tfOptimizerNumIterations', defaultValue=1000)
self.linkPropertyToSettings('tfOptimizerBatchSize', globalName=self.name + 'tfOptimizerBatchSize', defaultValue = -1)
if self.tfOptimizerType == TFOptimizerType.Adam:
self.linkPropertyToSettings('tfAdamLearningRate', globalName = self.name + 'tfAdamLearningRate', defaultValue = 0.001)
self.linkPropertyToSettings('tfAdamBeta1', globalName=self.name + 'tfAdamBeta1',
defaultValue=0.9)
self.linkPropertyToSettings('tfAdamBeta2', globalName=self.name + 'tfAdamBeta2',
defaultValue=0.999)
self.linkPropertyToSettings('tfAdamEpsilon', globalName=self.name + 'tfAdamEpsilon',
defaultValue=10**-8)
self.optimizer = tf.train.AdamOptimizer(learning_rate = self.tfAdamLearningRate, beta1 = self.tfAdamBeta1,
beta2 = self.tfAdamBeta2, epsilon = self.tfAdamEpsilon)
elif self.tfOptimizerType == TFOptimizerType.GradientDescent:
self.linkPropertyToSettings('tfGradientLearningRate', globalName=self.name + 'tfGradientLearningRate', defaultValue=0.001)
self.optimizer = tf.train.GradientDescentOptimizer(learning_rate=self.tfGradientLearningRate)
self.minimize = self.optimizer.minimize(self.loss, var_list=self.variables_list)
self.tm_minimize = TFMapping(dataManager, tensorNode = self.minimize)
self.tm_loss = TFMapping(dataManager, tensorNode = self.loss)
self._printIterations = printIterations
self.lossLogger = []
def __init__(self, dataManager, inputArguments, outputArguments, meanFunction, name = 'FullGaussian'):
TFMapping.__init__(self, dataManager, inputArguments, outputArguments, name = name)
self.meanFunction = meanFunction
self.additionalScopes.append(meanFunction.name)
def __init__(self, dataManager, inputArguments, outputArguments, name = 'NaturalFullGaussian'):
TFMapping.__init__(self, dataManager, inputArguments, outputArguments, name = name)