Interface IPredictor<TInstanceSource, TInstance, TLabelSource, TResult, TResultDist>

Interface to a learner that acts on some data to predict a label.

Inherited Members
Namespace: Microsoft.ML.Probabilistic.Learners
Assembly: Microsoft.ML.Probabilistic.Learners.dll
Syntax
public interface IPredictor<in TInstanceSource, in TInstance, in TLabelSource, out TResult, out TResultDist> : ILearner
Type Parameters
Name Description
TInstanceSource

The type of a source of instances.
TInstance

The type of an instance.
TLabelSource

The type of a source of labels.
TResult

The type of a prediction for a single instance.
TResultDist

The type of an uncertain prediction for a single instance.
Remarks

Intended usage:

An instance refers to the data that can be attributed to a label. It may include the label itself if it is known. Typically, an instance provides the feature values which characterize the label. In spam detection, for example, an instance might refer to an email which may be labeled 'spam' or 'no spam'. In finance, an instance might refer to a commodity and the label may be its price.

An instance source provides all instances of interest.

A label source provides the known labels of some given instances.

Labels may hence be provided by label source or instance source.

Properties

Capabilities

Gets the capabilities of the predictor.

Declaration
IPredictorCapabilities Capabilities { get; }
Property Value
Type Description
IPredictorCapabilities

Methods

Predict(TInstanceSource)

Makes predictions for the specified instances. Uncertainty in the predictions is discarded and the "best" predictions are returned.

Declaration
IEnumerable<TResult> Predict(TInstanceSource instanceSource)
Parameters
Type Name Description
TInstanceSource instanceSource

The source of instances to make predictions for.
Returns
Type Description
IEnumerable<TResult>

The "best" prediction for every instance in instanceSource, discarding uncertainty.
Remarks

The definition of "best" depends on the particular implementation of the predictor and its settings.

Predict(TInstance, TInstanceSource)

Makes a prediction for the specified instance. Uncertainty in the prediction is discarded and a "best" prediction is returned.

Declaration
TResult Predict(TInstance instance, TInstanceSource instanceSource = null)
Parameters
Type Name Description
TInstance instance

The instance to make predictions for.
TInstanceSource instanceSource

The source of instances which provides instance.
Returns
Type Description
TResult

The "best" prediction for instance, discarding uncertainty.
Remarks

The definition of "best" depends on the particular implementation of the predictor and its settings.

PredictDistribution(TInstanceSource)

Makes predictions for the specified instances. Uncertainty in the predictions is characterized by the returned distribution type.

Declaration
IEnumerable<TResultDist> PredictDistribution(TInstanceSource instanceSource)
Parameters
Type Name Description
TInstanceSource instanceSource

The source of instances to make predictions for.
Returns
Type Description
IEnumerable<TResultDist>

The predictions of all instances provided by instanceSource, with uncertainty.

PredictDistribution(TInstance, TInstanceSource)

Makes a prediction for the specified instance. Uncertainty in the prediction is characterized by the returned distribution type.

Declaration
TResultDist PredictDistribution(TInstance instance, TInstanceSource instanceSource = null)
Parameters
Type Name Description
TInstance instance

The instance to make predictions for.
TInstanceSource instanceSource

The source of instances which provides instance.
Returns
Type Description
TResultDist

The prediction for instance, with uncertainty.

Train(TInstanceSource, TLabelSource)

Trains the predictor on the specified instances.

Declaration
void Train(TInstanceSource instanceSource, TLabelSource labelSource = null)
Parameters
Type Name Description
TInstanceSource instanceSource

The source of instances.
TLabelSource labelSource

The source of labels.

Extension Methods