parameters

Documentation

Index

• Constants
• func Bind(e Parametrizable, p *P)
• func Exists(p Parametrizable) bool
• func Hardness(t int) int
• func Method(t int) int
• func Unbind(e Parametrizable)
• type P
  • func Default() *P
  • func New(norm, hw bool, sm float64, t int, eta, eps float64, bs int, l float64, i int) *P
  • func Retrieve(e Parametrizable) (*P, bool)
• type Parametrizable

Constants

const (
	HardGD = iota // Hard gradient descent key.
	SoftGD        // Soft gradient descent key.
	HardEM        // Hard expectation-maximization key.
	SoftEM        // Soft expectation-maximization key.
)

Constants to be used for P.LearningType.

const (
	GD = iota
	EM
)

Constants to be used with Method(P.LearningType).

const (
	Hard = iota
	Soft
)

Constants to be used with Hardness(P.LearningType).

Functions

func Bind

func Bind(e Parametrizable, p *P)

func Exists

func Exists(p Parametrizable) bool

func Hardness

func Hardness(t int) int

Hardness returns whether P.LearningType uses soft or hard inference for learning.

func Method

func Method(t int) int

Method returns what super-type of learning method P.LearningType is.

func Unbind

func Unbind(e Parametrizable)

Types

type P

type P struct {
	Normalize    bool    // Normalize on weight update.
	HardWeight   bool    // Hard weights (true) or soft weights (false).
	SmoothSum    float64 // Constant for smoothing sum counts when hard weights is true.
	LearningType int     // Soft or hard EM or GD (only applies to weight learning functions).
	Eta          float64 // Learning rate.
	Epsilon      float64 // Epsilon convergence criterion (in logspace).
	BatchSize    int     // Batch size if mini-batch. If bs <= 1, then no batching.
	Lambda       float64 // Regularization constant.
	Iterations   int     // Number of iterations for gradient descent.
}

P is a collection of available parameters for learning algorithms.

Disclaimer: Parameters do not work on inline methods (e.g. S.Value(E)) since that would require GoSPN storing a P pointer in each Node.

func Default

func Default() *P

Default returns a P instance with the following default options:

Normalize    = true
HardWeight   = false
SmoothSum    = 0.01
LearningType = parameters.SoftGD
Eta          = 0.1
Epsilon      = 1.0
BatchSize    = 0
Lambda 			 = 0.01
Iterations 	 = 4

func New

func New(norm, hw bool, sm float64, t int, eta, eps float64, bs int, l float64, i int) *P

New returns a P instance with the given parameters as option values.

func Retrieve

func Retrieve(e Parametrizable) (*P, bool)

type Parametrizable

type Parametrizable interface {
	// Parameters returns the parameters of this object.
	Parameters() *P
}

Parametriable defines a type that has parameters.