model

Documentation

Index

• func RandomBipartiteGraph(leftDegrees, rightDegrees []int, maxTries int) (edges [][2]int, err error)
• type MultipleBSCModel
  • func NewMultipleBSCModel() *MultipleBSCModel
  • func (model *MultipleBSCModel) AddChannel(name string, noise float64)
  • func (model *MultipleBSCModel) AddObservation(input, channel string, value bool)
  • func (model *MultipleBSCModel) EM(maxRounds int, tolerance float64, ...)
  • func (model MultipleBSCModel) HasChannel(name string) bool
  • func (model MultipleBSCModel) HasInput(name string) bool
  • func (model MultipleBSCModel) Score() float64
• type MultipleBSCPairModel
  • func NewMultipleBSCPairModel() *MultipleBSCPairModel
  • func (model *MultipleBSCPairModel) AddChannel(name1 string, noise1 float64, name2 string, noise2 float64)
  • func (model *MultipleBSCPairModel) AddObservation(input, channel1, channel2 string, value bool)
  • func (model *MultipleBSCPairModel) EM(maxRounds int, tolerance float64, ...)
  • func (model MultipleBSCPairModel) HasChannel(name1, name2 string) bool
  • func (model MultipleBSCPairModel) HasInput(name string) bool
  • func (model MultipleBSCPairModel) Score() float64

Functions

func RandomBipartiteGraph

func RandomBipartiteGraph(leftDegrees, rightDegrees []int, maxTries int) (edges [][2]int, err error)

Generates a random bipartite graph with the specified node degrees. The graph is selected approximately uniformly at random from the space of all graphs with the specified node degrees. Implements the algorithm in:

M. Bayati, J. H. Kim, and A. Saberi, "A Sequential Algorithm for Generating Random Graphs," Algorithmica, vol. 58, no. 4, pp. 860–910, 2010.

If successful, returns a list of edges. Each edge contains an index from the left nodes and an index from the right nodes.

Types

type MultipleBSCModel

type MultipleBSCModel struct {

	// The variables sent over the noisy channels
	Inputs map[string]*variable.DiscreteRV

	// The posterior probability that each latent input variable is true
	InputScores map[string]float64

	// Whether to use soft or hard assignments to inputs during inference
	SoftInputs bool

	// The noisy channels
	Channels map[string]*bsc.BSC

	// A factor graph relating inputs to outputs
	FactorGraph *factor.FactorGraph
}

A noisy channel model which explains a data set as passing through one of a set of BSC channels with unknown noise rates. For instance, this can be used to infer the answers of independent binary-valued crowdsourcing questions.

func NewMultipleBSCModel

func NewMultipleBSCModel() *MultipleBSCModel

Create a new MultipleBSCModel

func (*MultipleBSCModel) AddChannel

func (model *MultipleBSCModel) AddChannel(name string, noise float64)

Adds a new BSC to the model with the given name and noise rate.

func (*MultipleBSCModel) AddObservation

func (model *MultipleBSCModel) AddObservation(input, channel string, value bool)

Adds a new observation to the model for the given channel and input. If the input is new, it will be created automatically.

func (*MultipleBSCModel) EM

func (model *MultipleBSCModel) EM(maxRounds int, tolerance float64,
	callback func(model *MultipleBSCModel, round int, stage string))

Train noise rates and input values using expectation maximization.

func (MultipleBSCModel) HasChannel

func (model MultipleBSCModel) HasChannel(name string) bool

Ask whether a given channel exists

func (MultipleBSCModel) HasInput

func (model MultipleBSCModel) HasInput(name string) bool

Ask whether a given input exists

func (MultipleBSCModel) Score

func (model MultipleBSCModel) Score() float64

Score the model, using the current parameter values

type MultipleBSCPairModel

type MultipleBSCPairModel struct {

	// The noise rates
	Noise1Rates, Noise2Rates map[string]*variable.ContinuousRV

	// The variables sent over the noisy channels
	Inputs map[string]*variable.DiscreteRV

	// The posterior probability that each latent input variable is true
	InputScores map[string]float64

	// Whether to use soft or hard assignments to inputs during inference
	SoftInputs bool

	// The noisy channels
	Channels map[string]map[string]*bsc.BSCPair

	// A factor graph relating inputs to outputs
	FactorGraph *factor.FactorGraph
}

A noisy channel model which explains a data set as passing through one of a set of BSC channels with unknown noise rates. For instance, this can be used to infer the answers of independent binary-valued crowdsourcing questions.

func NewMultipleBSCPairModel

func NewMultipleBSCPairModel() *MultipleBSCPairModel

Create a new MultipleBSCPairModel

func (*MultipleBSCPairModel) AddChannel

func (model *MultipleBSCPairModel) AddChannel(name1 string, noise1 float64,
	name2 string, noise2 float64)

Adds a new BSCPair to the model with the given layer names and noise rates. If a given noise rate has already been added, it will not be changed.

func (*MultipleBSCPairModel) AddObservation

func (model *MultipleBSCPairModel) AddObservation(input, channel1, channel2 string, value bool)

Adds a new observation to the model for the given channel and input. If the input is new, it will be created automatically.

func (*MultipleBSCPairModel) EM

func (model *MultipleBSCPairModel) EM(maxRounds int, tolerance float64,
	callback func(model *MultipleBSCPairModel, round int, stage string))

Train noise rates and input values using expectation maximization.

func (MultipleBSCPairModel) HasChannel

func (model MultipleBSCPairModel) HasChannel(name1, name2 string) bool

Ask whether a given channel exists

func (MultipleBSCPairModel) HasInput

func (model MultipleBSCPairModel) HasInput(name string) bool

Ask whether a given input exists

func (MultipleBSCPairModel) Score

func (model MultipleBSCPairModel) Score() float64

Score the model, using the current parameter values