gd

Documentation

Overview

Package gd provides gradient descent implementation.

Index

• type Converger
  • func NewConverger(ct options.ConvergenceType, ci float64, c CostFunc) (Converger, error)
• type ConvergerFunc
  • func (f ConvergerFunc) Converge(ctx context.Context, s Stepper) ([]float64, error)
• type CostFunc
• type GradientDescent
  • func New(h Hyphothesis, c CostFunc) GradientDescent
  • func (g GradientDescent) Run(ctx context.Context, o options.Options, x [][]float64, y []float64) ([]float64, error)
• type Hyphothesis
• type Stepper
  • func NewStepper(gdv options.GradientDescentVariant, h Hyphothesis, x [][]float64, y []float64, ...) (Stepper, error)

Types

type Converger

type Converger interface {
	Converge(context.Context, Stepper) ([]float64, error)
}

Converger is the interface that wraps the basic Converge method.

func NewConverger

func NewConverger(ct options.ConvergenceType, ci float64, c CostFunc) (Converger, error)

NewConverger returns a new converger. If unsupported ConvergenceType is passed, an error is returned.

type ConvergerFunc

type ConvergerFunc func(context.Context, Stepper) ([]float64, error)

ConvergerFunc is an adapter to allow the use of plain functions as convergers.

func (ConvergerFunc) Converge

func (f ConvergerFunc) Converge(ctx context.Context, s Stepper) ([]float64, error)

type CostFunc

type CostFunc func(x [][]float64, y []float64, coeffs []float64) (float64, error)

CostFunc is a function template for cost function used in gradient descent algorithm.

type GradientDescent

type GradientDescent struct {
	// contains filtered or unexported fields
}

GradientDescent holds the hyphothesis and cost functions needed by gradient descent algorithm.

func New

func New(h Hyphothesis, c CostFunc) GradientDescent

New creates new gradient descent.

func (GradientDescent) Run

func (g GradientDescent) Run(ctx context.Context, o options.Options, x [][]float64, y []float64) ([]float64, error)

Run runs the gradient descent algorithm.

type Hyphothesis

type Hyphothesis func(x []float64, coeffs []float64) (float64, error)

Hyphothesis is a function template for a hyphothesis function used in gradient descent algorithm.

type Stepper

type Stepper interface {
	// TakeStep takes single step towards cost function minimum.
	TakeStep() error
	// CurrentCoefficients returns current coefficients calculated by stepper.
	CurrentCoefficients() []float64
	// X returns design matrix used in calculations.
	X() [][]float64
	// Y returns target vector used in calculations.
	Y() []float64
}

A Stepper wraps logic around taking steps (calculating new coefficients' values).

func NewStepper

func NewStepper(gdv options.GradientDescentVariant, h Hyphothesis, x [][]float64, y []float64, lr float64) (Stepper, error)