Class DampBackwardOp

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

Inheritance

Object

DampBackwardOp

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(Damp), "Backward<>", new Type[]{})]
[Quality(QualityBand.Preview)]
public static class DampBackwardOp

Methods

AverageLogFactor()

Evidence message for VMP.

Declaration

public static double AverageLogFactor()

Returns

Type	Description
Double	Zero.

Remarks

The formula for the result is log(factor(backward,value,stepsize)). Adding up these values across all factors and variables gives the log-evidence estimate for VMP.

BackwardAverageConditional<Distribution>(Distribution)

EP message to backward.

Declaration

public static Distribution BackwardAverageConditional<Distribution>(Distribution value)

Parameters

Type	Name	Description
Distribution	value	Incoming message from value.

Returns

Type	Description
Distribution	The outgoing EP message to the backward argument.

Type Parameters

Name	Description
Distribution	The type of the distribution over the damped variable.

Remarks

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

BackwardAverageLogarithm<Distribution>(Distribution)

VMP message to backward.

Declaration

public static Distribution BackwardAverageLogarithm<Distribution>(Distribution value)

Parameters

Type	Name	Description
Distribution	value	Incoming message from value.

Returns

Type	Description
Distribution	The outgoing VMP message to the backward argument.

Type Parameters

Name	Description
Distribution	The type of the distribution over the damped variable.

Remarks

The outgoing message is a distribution matching the moments of backward as the random arguments are varied. The formula is proj[sum_(value) p(value) factor(backward,value,stepsize)].

LogEvidenceRatio()

Evidence message for EP.

Declaration

public static double LogEvidenceRatio()

Returns

Type	Description
Double	Logarithm of the factor's contribution the EP model evidence.

Remarks

The formula for the result is log(factor(backward,value,stepsize)). Adding up these values across all factors and variables gives the log-evidence estimate for EP.

ValueAverageConditional<Distribution>(Distribution, Double, Distribution)

EP message to value.

Declaration

public static Distribution ValueAverageConditional<Distribution>(Distribution backward, double stepsize, Distribution to_value)
    where Distribution : SettableToPower<Distribution>, SettableToProduct<Distribution>

Parameters

Type	Name	Description
Distribution	backward	Incoming message from backward. Must be a proper distribution. If uniform, the result will be uniform.
Double	stepsize	Constant value for stepsize.
Distribution	to_value	Previous outgoing message to value.

Returns

Type	Description
Distribution	The outgoing EP message to the value argument.

Type Parameters

Name	Description
Distribution	The type of the distribution over the damped variable.

Remarks

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

Exceptions

Type	Condition
ImproperMessageException	backward is not a proper distribution.

ValueAverageLogarithm<Distribution>(Distribution, Double, Distribution)

VMP message to value.

Declaration

public static Distribution ValueAverageLogarithm<Distribution>(Distribution backward, double stepsize, Distribution to_value)
    where Distribution : SettableToPower<Distribution>, SettableToProduct<Distribution>

Parameters

Type	Name	Description
Distribution	backward	Incoming message from backward. Must be a proper distribution. If uniform, the result will be uniform.
Double	stepsize	Constant value for stepsize.
Distribution	to_value	Previous outgoing message to value.

Returns

Type	Description
Distribution	The outgoing VMP message to the value argument.

Type Parameters

Name	Description
Distribution	The type of the distribution over the damped variable.

Remarks

The outgoing message is the factor viewed as a function of value with backward integrated out. The formula is sum_backward p(backward) factor(backward,value,stepsize).

Exceptions

Type	Condition
ImproperMessageException	backward is not a proper distribution.