Class ShiftAlphaFromFactorOp

Provides outgoing messages for FromFactor<T>(T, Double, Double), given random arguments to the function.

Inheritance

Object

ShiftAlphaFromFactorOp

Inherited Members

Object.Equals(Object)

Object.Equals(Object, Object)

Object.GetHashCode()

Object.GetType()

Object.MemberwiseClone()

Object.ReferenceEquals(Object, Object)

Object.ToString()

Namespace: Microsoft.ML.Probabilistic.Factors

Assembly: Microsoft.ML.Probabilistic.dll

Syntax

[FactorMethod(typeof(ShiftAlpha), "FromFactor<>", new Type[]{})]
[Quality(QualityBand.Experimental)]
public static class ShiftAlphaFromFactorOp

Methods

FactorAverageConditional<T>(T, T, Double, Double, T)

EP message to factor.

Declaration

public static T FactorAverageConditional<T>(T factor, T variable, double factorAlpha, double variableAlpha, T result)
    where T : SettableToPower<T>, SettableToProduct<T>

Parameters

Type	Name	Description
T	factor	Incoming message from factor.
T	variable	Incoming message from variable. Must be a proper distribution. If uniform, the result will be uniform.
Double	factorAlpha	Constant value for factorAlpha.
Double	variableAlpha	Constant value for variableAlpha.
T	result	Modified to contain the outgoing message.

Returns

Type	Description
T	result

Type Parameters

Name	Description
T	The type of the messages.

Remarks

The outgoing message is a distribution matching the moments of factor as the random arguments are varied. The formula is proj[p(factor) sum_(variable) p(variable) factor(variable,factor,factorAlpha,variableAlpha)]/p(factor).

Exceptions

Type	Condition
ImproperMessageException	variable is not a proper distribution.

LogEvidenceRatioOld<T>(T, T, Double, Double)

Evidence message for EP.

Declaration

public static double LogEvidenceRatioOld<T>(T factor, T variable, double variableAlpha, double factorAlpha)
    where T : ICloneable, CanGetAverageLog<T>, SettableToPower<T>, SettableToProduct<T>

Parameters

Type	Name	Description
T	factor	Incoming message from 'factor'.
T	variable	Incoming message from 'variable'.
Double	variableAlpha	Constant value for 'variableAlpha'.
Double	factorAlpha	Constant value for 'factorAlpha'.

Returns

Type	Description
Double	log(int f(x) qnotf(x) dx / int ftilde(x) qnotf(x) dx)

Type Parameters

Name	Description
T

Remarks

The formula for the result is log(int f(x) qnotf(x) dx / int ftilde(x) qnotf(x) dx) where x = (variable,factor,factorAlpha,variableAlpha).

VariableAverageConditional<T>(T, T)

EP message to variable.

Declaration

public static T VariableAverageConditional<T>(T factor, T result)
    where T : SettableTo<T>

Parameters

Type	Name	Description
T	factor	Incoming message from factor. Must be a proper distribution. If uniform, the result will be uniform.
T	result	Modified to contain the outgoing message.

Returns

Type	Description
T	result

Type Parameters

Name	Description
T	The type of the messages.

Remarks

The outgoing message is a distribution matching the moments of variable as the random arguments are varied. The formula is proj[p(variable) sum_(factor) p(factor) factor(variable,factor,factorAlpha,variableAlpha)]/p(variable).

Exceptions

Type	Condition
ImproperMessageException	factor is not a proper distribution.