Class UnaryOp<DomainType>

Provides outgoing messages for Random<DomainType>(Sampleable<DomainType>), given random arguments to the function.

Inheritance

Object

UnaryOp<DomainType>

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(Factor), "Random<>", new Type[]{})]
[Quality(QualityBand.Mature)]
public static class UnaryOp<DomainType>

Type Parameters

Name	Description
DomainType	The type of the sampled variable.

Methods

AverageLogFactor<T>(T, T)

Evidence message for VMP.

Declaration

public static double AverageLogFactor<T>(T random, T dist)
    where T : CanGetAverageLog<T>

Parameters

Type	Name	Description
T	random	Incoming message from random.
T	dist	Incoming message from dist.

Returns

Type	Description
Double	Average of the factor's log-value across the given argument distributions.

Type Parameters

Name	Description
T	The type of the distribution over the sampled variable.

Remarks

The formula for the result is sum_(random,dist) p(random,dist) log(factor(random,dist)). Adding up these values across all factors and variables gives the log-evidence estimate for VMP.

AverageLogFactor<T>(DomainType, T)

Evidence message for VMP.

Declaration

public static double AverageLogFactor<T>(DomainType random, T dist)
    where T : CanGetLogProb<DomainType>

Parameters

Type	Name	Description
DomainType	random	Incoming message from random.
T	dist	Incoming message from dist.

Returns

Type	Description
Double	Average of the factor's log-value across the given argument distributions.

Type Parameters

Name	Description
T	The type of the distribution over the sampled variable.

Remarks

The formula for the result is sum_(random,dist) p(random,dist) log(factor(random,dist)). Adding up these values across all factors and variables gives the log-evidence estimate for VMP.

LogAverageFactor<T>(T, T)

Evidence message for EP.

Declaration

public static double LogAverageFactor<T>(T random, T dist)
    where T : CanGetLogAverageOf<T>

Parameters

Type	Name	Description
T	random	Incoming message from random.
T	dist	Incoming message from dist.

Returns

Type	Description
Double	Logarithm of the factor's average value across the given argument distributions.

Type Parameters

Name	Description
T	The type of the distribution over the sampled variable.

Remarks

The formula for the result is log(sum_(random,dist) p(random,dist) factor(random,dist)).

LogAverageFactor<T>(DomainType, T)

Evidence message for EP.

Declaration

public static double LogAverageFactor<T>(DomainType random, T dist)
    where T : CanGetLogProb<DomainType>

Parameters

Type	Name	Description
DomainType	random	Incoming message from random.
T	dist	Incoming message from dist.

Returns

Type	Description
Double	Logarithm of the factor's average value across the given argument distributions.

Type Parameters

Name	Description
T	The type of the distribution over the sampled variable.

Remarks

The formula for the result is log(sum_(random,dist) p(random,dist) factor(random,dist)).

LogEvidenceRatio<T>(T, T)

Evidence message for EP.

Declaration

public static double LogEvidenceRatio<T>(T random, T dist)

Parameters

Type	Name	Description
T	random	Incoming message from random.
T	dist	Incoming message from dist.

Returns

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

Type Parameters

Name	Description
T	The type of the distribution over the sampled variable.

Remarks

The formula for the result is log(sum_(random,dist) p(random,dist) factor(random,dist) / sum_random p(random) messageTo(random)). Adding up these values across all factors and variables gives the log-evidence estimate for EP.

LogEvidenceRatio<T>(DomainType, T)

Evidence message for EP.

Declaration

public static double LogEvidenceRatio<T>(DomainType random, T dist)
    where T : CanGetLogProb<DomainType>

Parameters

Type	Name	Description
DomainType	random	Incoming message from random.
T	dist	Incoming message from dist.

Returns

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

Type Parameters

Name	Description
T	The type of the distribution over the sampled variable.

Remarks

The formula for the result is log(sum_(random,dist) p(random,dist) factor(random,dist) / sum_random p(random) messageTo(random)). Adding up these values across all factors and variables gives the log-evidence estimate for EP.

RandomAverageConditional<T>(T)

EP message to random.

Declaration

public static T RandomAverageConditional<T>(T dist)

Parameters

Type	Name	Description
T	dist	Incoming message from dist.

Returns

Type	Description
T	The outgoing EP message to the random argument.

Type Parameters

Name	Description
T	The type of the distribution over the sampled variable.

Remarks

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

RandomAverageLogarithm<T>(T)

VMP message to random.

Declaration

public static T RandomAverageLogarithm<T>(T dist)

Parameters

Type	Name	Description
T	dist	Incoming message from dist.

Returns

Type	Description
T	The outgoing VMP message to the random argument.

Type Parameters

Name	Description
T	The type of the distribution over the sampled variable.

Remarks

The outgoing message is the exponential of the average log-factor value, where the average is over all arguments except random. The formula is exp(sum_(dist) p(dist) log(factor(random,dist))).

RandomMaxConditional(Bernoulli)

Declaration

public static Bernoulli RandomMaxConditional(Bernoulli dist)

Parameters

Type	Name	Description
Bernoulli	dist	Incoming message from dist.

Returns

Type	Description
Bernoulli

Remarks

RandomMaxConditional(Discrete)

Declaration

public static UnnormalizedDiscrete RandomMaxConditional(Discrete dist)

Parameters

Type	Name	Description
Discrete	dist	Incoming message from dist. Must be a proper distribution. If uniform, the result will be uniform.

Returns

Type	Description
UnnormalizedDiscrete

Remarks

Exceptions

Type	Condition
ImproperMessageException	dist is not a proper distribution.