Class BernoulliFromLogOddsOp

Provides outgoing messages for BernoulliFromLogOdds(Double), given random arguments to the function.

Inheritance

Object

BernoulliFromLogOddsOp

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), "BernoulliFromLogOdds", new Type[]{})]
[Quality(QualityBand.Experimental)]
public static class BernoulliFromLogOddsOp

Remarks

Performs KL minimization using gradient matching, a distributed gradient descent algorithm.

Fields

ForceProper

Declaration

public static bool ForceProper

Field Value

Type	Description
Boolean

Methods

AverageLogFactor(Bernoulli, Gaussian)

Evidence message for VMP.

Declaration

public static double AverageLogFactor(Bernoulli sample, Gaussian logOdds)

Parameters

Type	Name	Description
Bernoulli	sample	Incoming message from sample.
Gaussian	logOdds	Incoming message from logOdds. Must be a proper distribution. If uniform, the result will be uniform.

Returns

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

Remarks

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

Exceptions

Type	Condition
ImproperMessageException	logOdds is not a proper distribution.

AverageLogFactor(Bernoulli, Double)

Evidence message for VMP.

Declaration

public static double AverageLogFactor(Bernoulli sample, double logOdds)

Parameters

Type	Name	Description
Bernoulli	sample	Incoming message from sample.
Double	logOdds	Constant value for logOdds.

Returns

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

Remarks

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

AverageLogFactor(Boolean, Gaussian)

Evidence message for VMP.

Declaration

public static double AverageLogFactor(bool sample, Gaussian logOdds)

Parameters

Type	Name	Description
Boolean	sample	Constant value for sample.
Gaussian	logOdds	Incoming message from logOdds. Must be a proper distribution. If uniform, the result will be uniform.

Returns

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

Remarks

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

Exceptions

Type	Condition
ImproperMessageException	logOdds is not a proper distribution.

AverageLogFactor(Boolean, Double)

Evidence message for VMP.

Declaration

public static double AverageLogFactor(bool sample, double logOdds)

Parameters

Type	Name	Description
Boolean	sample	Constant value for sample.
Double	logOdds	Constant value for logOdds.

Returns

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

Remarks

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

LogAverageFactor(Boolean, Gaussian)

Evidence message for EP.

Declaration

public static double LogAverageFactor(bool sample, Gaussian logOdds)

Parameters

Type	Name	Description
Boolean	sample	Constant value for sample.
Gaussian	logOdds	Incoming message from logOdds. Must be a proper distribution. If uniform, the result will be uniform.

Returns

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

Remarks

The formula for the result is log(sum_(logOdds) p(logOdds) factor(sample,logOdds)).

Exceptions

Type	Condition
ImproperMessageException	logOdds is not a proper distribution.

LogAverageFactor(Boolean, Double)

Evidence message for EP.

Declaration

public static double LogAverageFactor(bool sample, double logOdds)

Parameters

Type	Name	Description
Boolean	sample	Constant value for sample.
Double	logOdds	Constant value for logOdds.

Returns

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

Remarks

The formula for the result is log(factor(sample,logOdds)).

LogEvidenceRatio(Bernoulli)

Evidence message for EP.

Declaration

public static double LogEvidenceRatio(Bernoulli sample)

Parameters

Type	Name	Description
Bernoulli	sample	Incoming message from sample.

Returns

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

Remarks

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

LogEvidenceRatio(Boolean, Gaussian)

Evidence message for EP.

Declaration

public static double LogEvidenceRatio(bool sample, Gaussian logOdds)

Parameters

Type	Name	Description
Boolean	sample	Constant value for sample.
Gaussian	logOdds	Incoming message from logOdds.

Returns

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

Remarks

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

LogEvidenceRatio(Boolean, Double)

Evidence message for EP.

Declaration

public static double LogEvidenceRatio(bool sample, double logOdds)

Parameters

Type	Name	Description
Boolean	sample	Constant value for sample.
Double	logOdds	Constant value for logOdds.

Returns

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

Remarks

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

LogOddsAverageConditional(Bernoulli, Gaussian)

EP message to logOdds.

Declaration

public static Gaussian LogOddsAverageConditional(Bernoulli sample, Gaussian logOdds)

Parameters

Type	Name	Description
Bernoulli	sample	Incoming message from sample.
Gaussian	logOdds	Incoming message from logOdds. Must be a proper distribution. If uniform, the result will be uniform.

Returns

Type	Description
Gaussian	The outgoing EP message to the logOdds argument.

Remarks

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

Exceptions

Type	Condition
ImproperMessageException	logOdds is not a proper distribution.

LogOddsAverageConditional(Boolean, Gaussian)

EP message to logOdds.

Declaration

public static Gaussian LogOddsAverageConditional(bool sample, Gaussian logOdds)

Parameters

Type	Name	Description
Boolean	sample	Constant value for sample.
Gaussian	logOdds	Incoming message from logOdds. Must be a proper distribution. If uniform, the result will be uniform.

Returns

Type	Description
Gaussian	The outgoing EP message to the logOdds argument.

Remarks

The outgoing message is the factor viewed as a function of logOdds conditioned on the given values.

Exceptions

Type	Condition
ImproperMessageException	logOdds is not a proper distribution.

LogOddsAverageLogarithm(Bernoulli, Gaussian, Gaussian)

Gradient matching VMP message from factor to logOdds variable

Declaration

public static Gaussian LogOddsAverageLogarithm(Bernoulli sample, Gaussian logOdds, Gaussian result)

Parameters

Type	Name	Description
Bernoulli	sample	Incoming message from 'sample'.
Gaussian	logOdds	Incoming message. Must be a proper distribution. If uniform, the result will be uniform.
Gaussian	result	Previous message sent, used for damping

Returns

Type	Description
Gaussian	The outgoing VMP message.

Remarks

The outgoing message is the Gaussian approximation to the factor which results in the same derivatives of the KL(q||p) divergence with respect to the parameters of the posterior as if the true factor had been used.

Exceptions

Type	Condition
ImproperMessageException	logOdds is not a proper distribution

LogOddsAverageLogarithm(Boolean, Gaussian, Gaussian)

Gradient matching VMP message from factor to logOdds variable

Declaration

public static Gaussian LogOddsAverageLogarithm(bool sample, Gaussian logOdds, Gaussian to_LogOdds)

Parameters

Type	Name	Description
Boolean	sample	Constant value for 'sample'.
Gaussian	logOdds	Incoming message. Must be a proper distribution. If uniform, the result will be uniform.
Gaussian	to_LogOdds	Previous message sent, used for damping

Returns

Type	Description
Gaussian	The outgoing VMP message.

Remarks

The outgoing message is the Gaussian approximation to the factor which results in the same derivatives of the KL(q||p) divergence with respect to the parameters of the posterior as if the true factor had been used.

Exceptions

Type	Condition
ImproperMessageException	logOdds is not a proper distribution

SampleAverageConditional(Gaussian)

EP message to sample.

Declaration

public static Bernoulli SampleAverageConditional(Gaussian logOdds)

Parameters

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

Returns

Type	Description
Bernoulli	The outgoing EP message to the sample argument.

Remarks

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

Exceptions

Type	Condition
ImproperMessageException	logOdds is not a proper distribution.

SampleAverageLogarithm(Gaussian)

VMP message to sample.

Declaration

public static Bernoulli SampleAverageLogarithm(Gaussian logOdds)

Parameters

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

Returns

Type	Description
Bernoulli	The outgoing VMP message to the sample argument.

Remarks

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

Exceptions

Type	Condition
ImproperMessageException	logOdds is not a proper distribution.

SampleAverageLogarithm(Double)

VMP message to sample.

Declaration

public static Bernoulli SampleAverageLogarithm(double logOdds)

Parameters

Type	Name	Description
Double	logOdds	Constant value for logOdds.

Returns

Type	Description
Bernoulli	The outgoing VMP message to the sample argument.

Remarks

The outgoing message is the factor viewed as a function of sample conditioned on the given values.