Class InferenceEngine

An inference engine, used to perform inference tasks in Infer.NET.

Inheritance

Object

InferenceEngine

Implements

SettableTo<InferenceEngine>

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.Models

Assembly: Microsoft.ML.Probabilistic.Compiler.dll

Syntax

public class InferenceEngine : SettableTo<InferenceEngine>

Remarks

The Debug class may be used to get debug messages for the inference engine. For example, use

Debug.Listeners.Add(new TextWriterTraceListener(Console.Out));

to get debug information for when compiled models and marginals are re-used.

Constructors

InferenceEngine()

Creates an inference engine which uses the default inference algorithm (currently this is expectation propagation).

Declaration

public InferenceEngine()

InferenceEngine(IAlgorithm)

Creates an inference engine which uses the specified inference algorithm.

Declaration

public InferenceEngine(IAlgorithm algorithm)

Parameters

Type	Name	Description
IAlgorithm	algorithm

Fields

compiler

The Compiler used to compile MSL into a compiled algorithm.

Declaration

protected ModelCompiler compiler

Field Value

Type	Description
ModelCompiler

DefaultEngine

Default inference engine whose settings will be copied onto newly created engines.

Declaration

public static readonly InferenceEngine DefaultEngine

Field Value

Type	Description
InferenceEngine

Properties

Algorithm

The default inference algorithm to use. This can be overridden for individual variables or factors using the Algorithm attribute.

Declaration

public IAlgorithm Algorithm { get; set; }

Property Value

Type	Description
IAlgorithm

Compiler

The model compiler that this inference engine uses.

Declaration

public ModelCompiler Compiler { get; }

Property Value

Type	Description
ModelCompiler

Groups

List of groups

Declaration

public IList<VariableGroup> Groups { get; }

Property Value

Type	Description
IList<VariableGroup>

ModelName

Model name, used when naming generated classes.

Declaration

public string ModelName { get; set; }

Property Value

Type	Description
String

ModelNamespace

Model namespace, used when naming generated classes.

Declaration

public string ModelNamespace { get; set; }

Property Value

Type	Description
String

Name

The full name of the inference engine, including version

Declaration

public static string Name { get; }

Property Value

Type	Description
String

NumberOfIterations

The number of iterations to use when executing the compiled inference algorithm.

Declaration

public int NumberOfIterations { get; set; }

Property Value

Type	Description
Int32

OptimiseForVariables

The variables to optimize the engine to infer. If set to a list of variables, only the specified variables can be inferred by this engine. If set to null, any variable can be inferred by this engine.

Declaration

public IList<IVariable> OptimiseForVariables { get; set; }

Property Value

Type	Description
IList<IVariable>

Remarks

Setting this property to a list of variables can improve performance by removing redundent computation and storage needed to infer marginals for variables which are not on the list.

ResetOnObservedValueChanged

If true (default), Infer resets messages to their initial values if an observed value has changed.

Declaration

public bool ResetOnObservedValueChanged { get; set; }

Property Value

Type	Description
Boolean

SaveFactorGraphToFolder

If not null, the factor graph will be saved (in DGML format) to a file in the specified folder (created if necessary) under the model name and the extension ".dgml"

Declaration

public string SaveFactorGraphToFolder { get; set; }

Property Value

Type	Description
String

ShowFactorGraph

If true, displays the factor graph for the model, prior to compiling it.

Declaration

public bool ShowFactorGraph { get; set; }

Property Value

Type	Description
Boolean

ShowMsl

If true, prints the model definition in Model Specification Language (MSL), prior to compiling the model.

Declaration

public bool ShowMsl { get; set; }

Property Value

Type	Description
Boolean

ShowProgress

If true, prints progress information to the console during inference.

Declaration

public bool ShowProgress { get; set; }

Property Value

Type	Description
Boolean

ShowSchedule

If true, displays the schedule for the model, after the scheduler has run.

Declaration

public bool ShowSchedule { get; set; }

Property Value

Type	Description
Boolean

ShowTimings

If true, prints timing information to the console during inference.

Declaration

public bool ShowTimings { get; set; }

Property Value

Type	Description
Boolean

ShowWarnings

If true, any warnings encountered during model compilation will be printed to the console.

Declaration

public bool ShowWarnings { get; set; }

Property Value

Type	Description
Boolean

Visualizer

Provides the implementation of ShowFactorGraph, ShowSchedule, and BrowserMode.

Declaration

public static Visualizer Visualizer { get; set; }

Property Value

Type	Description
Visualizer

Methods

GetBuiltInAlgorithms()

Returns an array of the built-in inference algorithms.

Declaration

public static IAlgorithm[] GetBuiltInAlgorithms()

Returns

Type	Description
IAlgorithm[]

GetCodeToInfer(IVariable)

Get the abstract syntax tree for the generated code.

Declaration

public List<ITypeDeclaration> GetCodeToInfer(IVariable var)

Parameters

Type	Name	Description
IVariable	var

Returns

Type	Description
List<ITypeDeclaration>	A list of type declaration objects.

GetCompiledInferenceAlgorithm(IVariable[])

Returns a compiled algorithm which can later be used to infer marginal distributions for the specified variables. This method allows more fine-grained control over the inference procedure.

Declaration

public IGeneratedAlgorithm GetCompiledInferenceAlgorithm(params IVariable[] vars)

Parameters

Type	Name	Description
IVariable[]	vars	The variables whose marginals are to be computed by the returned algorithm.

Returns

Type	Description
IGeneratedAlgorithm	An IGeneratedAlgorithm object

Remarks

This method should not be used unless fine-grained control over the inference is required. Infer.NET will cache the last compiled algorithm and re-use it if possible.

GetOutputMessage<Distribution>(IVariable)

Computes the output message (message to the prior) for the specified variable.

Declaration

public Distribution GetOutputMessage<Distribution>(IVariable var)

Parameters

Type	Name	Description
IVariable	var	The variable whose output message is to be inferred

Returns

Type	Description
Distribution	The output message (or an approximation to it)

Type Parameters

Name	Description
Distribution	Desired distribution type

GetRelevantModelExpressions(IVariable[])

For advanced use. Returns all the model expressions that are relevant to inferring the set of variables provided. This may be useful for constructing visualisations of the model.

Declaration

public IReadOnlyCollection<IModelExpression> GetRelevantModelExpressions(params IVariable[] vars)

Parameters

Type	Name	Description
IVariable[]	vars	The variables to build a model for

Returns

Type	Description
IReadOnlyCollection<IModelExpression>	A collection of model expressions

Remarks

The returned collection includes Variable and VariableArray objects which the engine has determined are relevant to inferring marginals over the variables provided. This will at least include the provided variables, but may include other relevant variables as well. It will also include MethodInvoke objects which act as priors, constraints or factors in the model.

Group(Variable[])

Add a variable group

Declaration

public VariableGroup Group(params Variable[] variables)

Parameters

Type	Name	Description
Variable[]	variables

Returns

Type	Description
VariableGroup

Infer(IVariable)

Infers the marginal distribution for the specified variable.

Declaration

public object Infer(IVariable var)

Parameters

Type	Name	Description
IVariable	var	The variable whose marginal is to be inferred

Returns

Type	Description
Object	The marginal distribution (or an approximation to it)

Infer(IVariable, QueryType)

Performs an inference query for the specified variable, given a query type.

Declaration

public object Infer(IVariable var, QueryType queryType)

Parameters

Type	Name	Description
IVariable	var	The variable whose marginal is to be inferred
QueryType	queryType	The type of query

Returns

Type	Description
Object	The marginal distribution (or an approximation to it)

Infer<TReturn>(IVariable)

Infers the marginal distribution for the specified variable.

Declaration

public TReturn Infer<TReturn>(IVariable var)

Parameters

Type	Name	Description
IVariable	var	The variable whose marginal is to be inferred

Returns

Type	Description
TReturn	The marginal distribution (or an approximation to it)

Type Parameters

Name	Description
TReturn	Desired return type which may be a distribution type or an array type if the argument is a VariableArray

Infer<TReturn>(IVariable, QueryType)

Infers the marginal distribution for the specified variable, and the specified query type

Declaration

public TReturn Infer<TReturn>(IVariable var, QueryType queryType)

Parameters

Type	Name	Description
IVariable	var	The variable whose marginal is to be inferred
QueryType	queryType	The query type

Returns

Type	Description
TReturn	The marginal distribution (or an approximation to it)

Type Parameters

Name	Description
TReturn	Desired return type

InferAll(Boolean, IVariable)

Declaration

protected IGeneratedAlgorithm InferAll(bool inferOnlySpecifiedVars, IVariable var)

Parameters

Type	Name	Description
Boolean	inferOnlySpecifiedVars
IVariable	var

Returns

Type	Description
IGeneratedAlgorithm

InferAll(Boolean, IEnumerable<IVariable>)

Declaration

protected IGeneratedAlgorithm InferAll(bool inferOnlySpecifiedVars, IEnumerable<IVariable> vars)

Parameters

Type	Name	Description
Boolean	inferOnlySpecifiedVars
IEnumerable<IVariable>	vars

Returns

Type	Description
IGeneratedAlgorithm

Reset()

For message passing algorithms, reset all messages to their initial values.

Declaration

protected void Reset()

SetTo(InferenceEngine)

Configures this inference engine by copying the settings from the supplied inference engine.

Declaration

public void SetTo(InferenceEngine engine)

Parameters

Type	Name	Description
InferenceEngine	engine

ShowFactorManager(Boolean)

Shows the factor manager, indicating which factors are available in Infer.NET and which are supported for each built-in inference algorithm.

Declaration

public static void ShowFactorManager(bool showMissingEvidences)

Parameters

Type	Name	Description
Boolean	showMissingEvidences

ShowFactorManager(Boolean, IAlgorithm[])

Shows the factor manager, indicating which factors are available in Infer.NET and which are supported for the supplied list of inference algorithms.

Declaration

public static void ShowFactorManager(bool showMissingEvidences, params IAlgorithm[] algorithms)

Parameters

Type	Name	Description
Boolean	showMissingEvidences
IAlgorithm[]	algorithms

Events

MessageUpdated

Event that is fired when a message that has been marked with ListenToMessages has been updated.

Declaration

public event MessageUpdatedEventHandler MessageUpdated

Event Type

Type	Description
MessageUpdatedEventHandler

ProgressChanged

Event that is fired when the progress of inference changes, typically at the end of one iteration of the inference algorithm.

Declaration

public event InferenceProgressEventHandler ProgressChanged

Event Type

Type	Description
InferenceProgressEventHandler

Implements

SettableTo<T>