GradBasedBackPropagation

abstract val backPropagate: MetaPipe[LayerP, Stream[(I, I)], Stream[I]]

A meta pipeline which for a particular value of the layer parameters, returns a data pipe which takes as input Stream of Tuple2 consisting of delta's and gradients of activation function with respect to their local fields (calculated via Activation.grad).

abstract val computeOutputDelta: StreamMapPipe[(I, I, I), (I, Double)]

A data pipeline which takes Tuple3 consisting of output layer activations, targets and gradients of output activations with respect to their local fields, respectively and returns the output layer delta values and the loss.

abstract val gradCompute: DataPipe[Stream[(I, I)], LayerP]

A data pipeline which takes as input a Stream of Tuple2 whose first element is the activation and second element the delta value and outputs the gradient of the layer parameters.

abstract val stackFactory: NeuralStackFactory[LayerP, I]

abstract val updater: MomentumUpdater[Seq[LayerP]]

Performs the actual update to the layer parameters after all the gradients have been calculated.

final def !=(arg0: Any): Boolean

final def ##(): Int

final def ==(arg0: Any): Boolean

def _momentum: Double

final def asInstanceOf[T0]: T0

def clone(): AnyRef

final def eq(arg0: AnyRef): Boolean

def equals(arg0: Any): Boolean

def finalize(): Unit

final def getClass(): Class[_]

def hashCode(): Int

final def isInstanceOf[T0]: Boolean

var miniBatchFraction: Double

var momentum: Double

def momentum_(m: Double): GradBasedBackPropagation[LayerP, I]

final def ne(arg0: AnyRef): Boolean

final def notify(): Unit

final def notifyAll(): Unit

var numIterations: Int

def optimize(nPoints: Long, data: Stream[(I, I)], initialStack: NeuralStack[LayerP, I]): NeuralStack[LayerP, I]

Solve the optimization problem of determining NeuralStack weights, from training data.

var regParam: Double

def setMiniBatchFraction(fraction: Double): GradBasedBackPropagation.this.type

Set fraction of data to be used for each SGD iteration.

def setNumIterations(iters: Int): GradBasedBackPropagation.this.type

Set the number of iterations for SGD.

def setRegParam(regParam: Double): GradBasedBackPropagation.this.type

Set the regularization parameter.

def setStepSize(step: Double): GradBasedBackPropagation.this.type

Set the initial step size of SGD for the first step.

var stepSize: Double

final def synchronized[T0](arg0: ⇒ T0): T0

def toString(): String

final def wait(): Unit

final def wait(arg0: Long, arg1: Int): Unit

final def wait(arg0: Long): Unit