Class SGD

Stochastic gradient descent optimizer

Hierarchy

Optimizer
- SGD

Index

Constructors

constructor

Properties

Methods

Constructors

constructor

new SGD(model: Module, lr?: number): SGD

Inherited from Optimizer.constructor
- Defined in model/optimizer/SGD.ts:7
New SGD optimizer for a particular model.

Parameters
- model: Module
- lr: number = 0.001
  
  Learning rate, the step size for each update step
Returns SGD

Properties

lr

lr: number = 0.001

model

model: Module

parameters

parameters: Variable<any>[]

Methods

step

step(): void

Overrides Optimizer.step
- Defined in model/optimizer/SGD.ts:17
Returns void

zeroGrads

zeroGrads(): void

Inherited from Optimizer.zeroGrads
- Defined in model/optimizer/optimizer.ts:27
Zeros all gradients of the model parameters. This should be called after each optimization step, when the gradients are no longer needed.

Returns void

Generated using TypeDoc