policy

sealed abstract class ConvolutionalLayerPolicy extends AnyRef

A parameter to direct the implementation of a convolutional layer.
trait Distribution extends Serializable

Case classes for picking randomly from different types of distributions
case class FCInit(inputLen: Int) extends WeightInitPolicy with Product with Serializable
trait LearningRule extends AnyRef
case class NormalDistribution(mu: Float, sigma: Float) extends Distribution with Product with Serializable
case class StandardLearningRule(learningRate: Float, momentum: Float, weightDecay: Float) extends LearningRule with Product with Serializable
case class UniformDistribution(scale: Float, offset: Float) extends Distribution with Product with Serializable
trait WeightInitPolicy extends AnyRef

object Best extends ConvolutionalLayerPolicy with Product with Serializable

Let the platform choose the implementation, guided by such parameters as border policy and downsampling use.
object ConvInit extends WeightInitPolicy with Product with Serializable
object Freq extends ConvolutionalLayerPolicy with Product with Serializable

Use a frequency-domain implementation of convolution (i.e.
Use a frequency-domain implementation of convolution (i.e. FFT - multiply - InverseFFT).
object Space extends ConvolutionalLayerPolicy with Product with Serializable

Use a space-domain implementation of convolution (i.e.
Use a space-domain implementation of convolution (i.e. standard convolution).
object StubInit extends WeightInitPolicy with Product with Serializable
object ZeroInit extends WeightInitPolicy with Product with Serializable