def __init__(self):
self.__layer = []
self.__input_symbol = Variable(name='InputSymbol')
self.__current_symbol = self.__input_symbol
self.__current_output = None
self.__current_weight = None
self.__current_bias = None
self.__variables = []
self.__data = None
self.__optimizer = None
self.__loss = None
self.__train_engine = Engine()
self.__predict_engine = Engine()
def end_iteration(self):
for layer_tuple in self.__batch_normalization_layer:
layer_mean_symbol, layer_variance_symbol = layer_tuple[0].normalization_symbol()
normalization_engine = Engine(layer_variance_symbol)
normalization_engine.bind = self.network.engine.bind
layer_tuple[2].append(normalization_engine.value())
layer_tuple[1].append(normalization_engine.value_cache[layer_mean_symbol])
def __init__(self, symbol: Symbol = None, variables=None):
self.__engine: Engine = None
self.__variables = []
self.__value_cache = {}
self.__gradient_cache = {}
self.__gradient_engines = {}
self.__symbol_dependence_variables = {}
self.set_engine(Engine(symbol, variables))
def __init__(self):
self.epoch = None
self.iteration = None
self.epochs = None
self.batch_size = None
self.engine = Engine()
self.__layer = []
self.__input_symbol = Variable(name='InputSymbol')
self.__current_symbol = self.__input_symbol
self.__current_output = None
self.__current_weight = None
self.__current_bias = None
self.__variables = []
self.__data = None
self.__optimizer = None
self.__loss = None
self.__predict_engine = Engine()
self.__plugin = collections.OrderedDict()
self.load_default_plugin()
def __init__(self,
rate: float,
decay: float,
square_decay: float,
consistent: bool = False):
self.__rate = rate
self.__decay = decay
self.__square_decay = square_decay
self.__consistent = consistent
self.__gradient_engine = Engine()
self.__estimation_map = {}
self.__square_estimation_map = {}
self.__step = 1
def __init__(self, rate: float, consistent: bool = False):
self.__rate = rate
self.__consistent = consistent
self.__gradient_engine = Engine()
self.__accumulate_gradient_map = {}
def __init__(self, rate: float, factor: float, consistent: bool = False):
self.__rate = rate
self.__factor = factor
self.__consistent = consistent
self.__old_gradient_map = {}
self.__gradient_engine = Engine()
def __init__(self, rate: float, consistent: bool = False):
self.__rate = rate
self.__consistent = consistent
self.__gradient_engine = Engine()
self.__mean_map = {}
self.__step = 1
def __init__(self, decay: float, consistent: bool = False):
self.__decay = decay
self.__consistent = consistent
self.__gradient_engine = Engine()
self.__accumulate_gradient_map = {}
self.__expectation_map = {}
def maximize(self, engine: Engine):
engine.differentiate()
variables = engine.variables
for variable in variables:
variable.value += self.__rate * Engine(
engine.gradient(variable)).value()
def __init__(self, rate: float):
self.__rate = rate
self.__gradient_engine = Engine()
def initialization(self):
for symbol in self.__variables:
gradient_symbol = self.__engine.gradient(symbol)
self.__gradient_engines[symbol] = Engine(gradient_symbol)
self.__symbol_dependence_variables[
symbol] = gradient_symbol.dependenc(SymbolCategory.variable)