Class SharedVariable<DomainType>

Abstract base class for shared variables. Shared variables allow a model to be split into submodels in which variables are shared. Each submodel can have many copies.

Inheritance

Object

SharedVariable<DomainType>

Implements

ISharedVariable

Inherited Members

Object.Equals(Object)

Object.Equals(Object, Object)

Object.GetHashCode()

Object.GetType()

Object.MemberwiseClone()

Object.ReferenceEquals(Object, Object)

Namespace: Microsoft.ML.Probabilistic.Models

Assembly: Microsoft.ML.Probabilistic.Compiler.dll

Syntax

public abstract class SharedVariable<DomainType> : ISharedVariable

Type Parameters

Name	Description
DomainType	Domian type of the variable

Remarks

A typical use of this is for large data sets where the likelihood parts of the model cannot all fit in memory. The solution is to divide the data into chunks (or 'batches'), and specify a single submodel which includes the likelihood factors and variables for one chunk, along with the shared parameters; the number of copies of the submodel is set to the number of chunks. In a related pattern, there are one or more additional submodels for defining the parameter variables.

Fields

Name

Name of the shared variable.

Declaration

public string Name

Field Value

Type	Description
String

Properties

IsEvidenceVariable

Marks this shared variable as one that calculates evidence

Declaration

public bool IsEvidenceVariable { get; set; }

Property Value

Type	Description
Boolean

Methods

GetCopyFor(Model)

Gets a copy of the variable for the specified model.

Declaration

public abstract Variable<DomainType> GetCopyFor(Model model)

Parameters

Type	Name	Description
Model	model	Model id.

Returns

Type	Description
Variable<DomainType>

GetEvidenceCorrection()

Gets the evidence correction for this shared variable.

Declaration

public abstract double GetEvidenceCorrection()

Returns

Type	Description
Double

InferOutput(IGeneratedAlgorithm, Model, Int32)

Infer the shared variable's output message for the given model and batch number.

Declaration

public abstract void InferOutput(IGeneratedAlgorithm ca, Model modelNumber, int batchNumber)

Parameters

Type	Name	Description
IGeneratedAlgorithm	ca	The compiled algorithm.
Model	modelNumber	The model id.
Int32	batchNumber	The batch number.

InferOutput(InferenceEngine, Model, Int32)

Infer the shared variable's output message for the given model and batch number.

Declaration

public abstract void InferOutput(InferenceEngine engine, Model modelNumber, int batchNumber)

Parameters

Type	Name	Description
InferenceEngine	engine	The inference engine.
Model	modelNumber	The model id.
Int32	batchNumber	The batch number.

Marginal<T>()

Get the marginal distribution for the shared variable, converted to type T.

Declaration

public abstract T Marginal<T>()

Returns

Type	Description
T

Type Parameters

Name	Description
T	The desired type

Named(String)

Inline method for naming a shared variable.

Declaration

public SharedVariable<DomainType> Named(string name)

Parameters

Type	Name	Description
String	name	The name

Returns

Type	Description
SharedVariable<DomainType>	this

Random<DistributionType>(DistributionType, Boolean)

Creates a shared random variable with the specified prior distribution.

Declaration

public static SharedVariable<DomainType> Random<DistributionType>(DistributionType prior, bool divideMessages = true)
    where DistributionType : IDistribution<DomainType>, Sampleable<DomainType>, SettableToProduct<DistributionType>, ICloneable, SettableToUniform, SettableTo<DistributionType>, SettableToRatio<DistributionType>, CanGetLogAverageOf<DistributionType>

Parameters

Type	Name	Description
DistributionType	prior	Prior
Boolean	divideMessages	Use division (the faster default) for calculating messages to batches

Returns

Type	Description
SharedVariable<DomainType>

Type Parameters

Name	Description
DistributionType	Distribution type

Random<DistributionArrayType>(Range, DistributionArrayType, Boolean)

Creates a 1D array of shared random variables of size given by the specified range.

Declaration

public static SharedVariableArray<DomainType> Random<DistributionArrayType>(Range range, DistributionArrayType prior, bool divideMessages = true)
    where DistributionArrayType : IDistribution<DomainType[]>, Sampleable<DomainType[]>, SettableToProduct<DistributionArrayType>, ICloneable, SettableToUniform, SettableTo<DistributionArrayType>, SettableToRatio<DistributionArrayType>, CanGetLogAverageOf<DistributionArrayType>

Parameters

Type	Name	Description
Range	range	Range.
DistributionArrayType	prior	A distribution over an array, to use as the prior.
Boolean	divideMessages	Use division (the faster default) for calculating messages to batches

Returns

Type	Description
SharedVariableArray<DomainType>

Type Parameters

Name	Description
DistributionArrayType	The type of the supplied prior

Random<DistributionArrayType>(VariableArray<DomainType>, Range, DistributionArrayType, Boolean)

Creates a 1D jagged array of shared random variables.

Declaration

public static ISharedVariableArray<VariableArray<DomainType>, DomainType[][]> Random<DistributionArrayType>(VariableArray<DomainType> itemPrototype, Range range, DistributionArrayType prior, bool divideMessages = true)
    where DistributionArrayType : IDistribution<DomainType[][]>, Sampleable<DomainType[][]>, SettableToProduct<DistributionArrayType>, ICloneable, SettableToUniform, SettableTo<DistributionArrayType>, SettableToRatio<DistributionArrayType>, CanGetLogAverageOf<DistributionArrayType>

Parameters

Type	Name	Description
VariableArray<DomainType>	itemPrototype	A fresh variable object representing an array element.
Range	range	Outer range.
DistributionArrayType	prior	Prior for the array.
Boolean	divideMessages	Use division (the faster default) for calculating messages to batches

Returns

Type	Description
ISharedVariableArray<VariableArray<DomainType>, DomainType[][]>

Type Parameters

Name	Description
DistributionArrayType

Random<ItemType, DistributionArrayType>(ItemType, Range, DistributionArrayType, Boolean)

Creates a generic jagged array of shared random variables.

Declaration

public static ISharedVariableArray<ItemType, DomainType> Random<ItemType, DistributionArrayType>(ItemType itemPrototype, Range range, DistributionArrayType prior, bool divideMessages = true)
    where ItemType : Variable, SettableTo<ItemType>, ICloneable where DistributionArrayType : IDistribution<DomainType>, Sampleable<DomainType>, SettableToProduct<DistributionArrayType>, ICloneable, SettableToUniform, SettableTo<DistributionArrayType>, SettableToRatio<DistributionArrayType>, CanGetLogAverageOf<DistributionArrayType>

Parameters

Type	Name	Description
ItemType	itemPrototype	A fresh variable object representing an array element.
Range	range	Outer range
DistributionArrayType	prior	Prior for the array.
Boolean	divideMessages	Use division (the faster default) for calculating messages to batches

Returns

Type	Description
ISharedVariableArray<ItemType, DomainType>

Type Parameters

Name	Description
ItemType	Item type
DistributionArrayType	Distribution array type

SetDefinitionTo(Model, Variable<DomainType>)

Sets the definition of the shared variable.

Declaration

public abstract void SetDefinitionTo(Model model, Variable<DomainType> definition)

Parameters

Type	Name	Description
Model	model	Model id.
Variable<DomainType>	definition	Defining variable.

Remarks

Use this method if the model is defining the shared variable rather than using one defined in this or another model.

SetInput(Model, Int32)

Sets the shared variable's inbox for a given model and batch.

Declaration

public abstract void SetInput(Model modelNumber, int batchNumber)

Parameters

Type	Name	Description
Model	modelNumber
Int32	batchNumber

ToString()

ToString override.

Declaration

public override string ToString()

Returns

Type	Description
String

Overrides

Object.ToString()

Implements

ISharedVariable