Overview > klyn.math.nn.optimizers > SGD
Version 0.2.1 Overview Index Tree Deprecated
SGD
classin packageklyn.math.nn.optimizers
public class SGD<T extends AbstractFloat = Float> implements Optimizer<T>
SGD
All Implemented Interfaces: Optimizer

Stochastic gradient descent without momentum or hidden optimizer state. Gradients must already include the loss reduction (sum or mean). Updates traverse packed storage without allocating a scaled gradient tensor.

Example

optimizer = SGD<Float>(learningRate=0.01)

Properties
Modifier and Type Member Description
public readonly property learningRate
learningRate as Double
 Finite positive step size.
Constructors
Modifier and Type Member Description
public SGD
SGD(learningRate as Double = 0.01)
 Configures a finite, positive learning rate.
Methods
Modifier and Type Member Description
public override apply
apply(parameter as NDArray<T>, gradient as NDArray<T>, gradientScale as Double = 1.0) as Void
 Updates a parameter tensor in place: parameter -= learningRate * gradient.
Inherited Methods
methodInherited Methods from Optimizer: apply