GradientDescent

Defined in: compat.js:47

Gradient Descent (batch; optional backtracking line search)

Extends

• Optimizer

Constructors

Constructor

new GradientDescent(options?): GradientDescent

Defined in: compat.js:48

Parameters

options?

Returns

GradientDescent

Overrides

Optimizer.constructor

Properties

batchSize

batchSize: any

Defined in: compat.js:51

---

learningRate

learningRate: any

Defined in: compat.js:17

Inherited from

Optimizer.learningRate

---

lineSearch

lineSearch: any

Defined in: compat.js:52

---

maxIter

maxIter: any

Defined in: compat.js:18

Inherited from

Optimizer.maxIter

---

stochastic

stochastic: any

Defined in: compat.js:50

---

tol

tol: any

Defined in: compat.js:19

Inherited from

Optimizer.tol

---

verbose

verbose: any

Defined in: compat.js:20

Inherited from

Optimizer.verbose

Methods

_options()

_options(options): object

Defined in: compat.js:34

Parameters

options

any

Returns

object

learningRate

learningRate: any

maxIter

maxIter: any

tol

tol: any

verbose

verbose: any

Inherited from

Optimizer._options

---

minimize()

minimize(lossFn, x0, options?): object

Defined in: compat.js:55

Minimize a loss function

Parameters

lossFn

any

Function that returns {loss, gradient}

x0

any

Initial parameters

options?

Additional options

Returns

object

{x, history}

history

history: any

x

x: any

Overrides

Optimizer.minimize