View on GitHub
Install in Dash
SGDOptimizer Class Reference
-
Learning rate. Should
>= 0.Declaration
Swift
public var learningRate: Float -
Initial leanring rate
Declaration
Swift
public var initLearningRate: Float -
LR Decay method
Declaration
Swift
public var decayMethod: LearningRateDecayMethod -
Learning rate decay per epoch. Before each epoch’s parmaeter updating, do
learningRate -= learningRate * decay. Should>= 0.Declaration
Swift
public var decay: Float -
Momentum. Should
>= 0.Declaration
Swift
public var momentum: Float -
Whether to turn on Nesterov momentum. Default is
false.Declaration
Swift
public var nesterov: Bool -
Initial leanring rate
Declaration
Swift
internal var initialLR: Float
-
Undocumented
Declaration
Swift
public init(learningRate: Float = 0.001, momentum: Float = 0.0, decayMethod: LearningRateDecayMethod = LearningRateDecayMethod.Step, decay: Float = 0.0, nesterov: Bool = false)
-
Do some preparing work before each epoch training
Declaration
Swift
public func prepare(_ graph: Graph)Parameters
graphtarget graph
-
Update a data symbol’s updated value
/ - Parameters:
- dataSymbol: target symbol
- gradValue: gradvalue fot this time updating
Declaration
Swift
public func updateParameter(_ dataSymbol: DataSymbol, gradValue: DataSymbolSupportedDataType)