/// <summary>
/// Safely retrieves the feature indexes for a specified instance.
/// </summary>
/// <typeparam name="TInstanceSource">The type of the instance source.</typeparam>
/// <typeparam name="TInstance">The type of an instance.</typeparam>
/// <typeparam name="TLabelSource">The type of the label source.</typeparam>
/// <typeparam name="TLabel">The type of a label.</typeparam>
/// <param name="mapping">The mapping.</param>
/// <param name="instance">The instance.</param>
/// <param name="instanceSource">An optional instance source.</param>
/// <returns>The feature indexes for the specified instance.</returns>
/// <exception cref="MappingException">Thrown if the feature indexes contain duplicates or negative indexes.</exception>
public static int[] GetSingleFeatureVectorIndexesSafe <TInstanceSource, TInstance, TLabelSource, TLabel>(
this IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, TLabel> mapping,
TInstance instance,
TInstanceSource instanceSource = default(TInstanceSource))
{
int[] featureIndexes = mapping.GetFeatureIndexes(instance, instanceSource);
if (featureIndexes == null)
{
return(null);
}
// Check that indexes are not negative
if (featureIndexes.Any(index => index < 0))
{
throw new MappingException("The feature indexes must not be negative.");
}
// Check for duplicate indexes
var featureIndexSet = new HashSet <int>(featureIndexes);
if (featureIndexSet.Count != featureIndexes.Length)
{
throw new MappingException("The feature indexes must not contain duplicates.");
}
return(featureIndexes);
}
/// <summary>
/// Safely retrieves the feature values of all instances from a specified batch of the instance source.
/// </summary>
/// <typeparam name="TInstanceSource">The type of the instance source.</typeparam>
/// <typeparam name="TInstance">The type of an instance.</typeparam>
/// <typeparam name="TLabelSource">The type of the label source.</typeparam>
/// <typeparam name="TLabel">The type of a label.</typeparam>
/// <param name="mapping">The mapping.</param>
/// <param name="instanceSource">The instance source.</param>
/// <param name="batchNumber">An optional batch number. Defaults to 0 and is used only if the instance source is divided into batches.</param>
/// <returns>The feature values provided by the specified batch of the instance source.</returns>
/// <exception cref="MappingException">
/// Thrown if the feature values are null, empty or contain infinite values or NaNs.
/// </exception>
public static double[][] GetAllFeatureVectorValuesSafe <TInstanceSource, TInstance, TLabelSource, TLabel>(
this IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, TLabel> mapping,
TInstanceSource instanceSource,
int batchNumber = 0)
{
if (batchNumber < 0)
{
throw new MappingException("The number of a batch must not be negative.");
}
double[][] featureValues = mapping.GetFeatureValues(instanceSource, batchNumber);
if (featureValues == null)
{
throw new MappingException("The feature values must not be null.");
}
if (featureValues.Length == 0)
{
throw new MappingException("The feature values must not be empty.");
}
if (featureValues.Any(values => values == null))
{
throw new MappingException("The feature values must not be null for any instance.");
}
if (featureValues.Any(values => values.Any(value => double.IsInfinity(value) || double.IsNaN(value))))
{
throw new MappingException("A feature value must not be infinite or NaN.");
}
return(featureValues);
}
/// <summary>
/// Initializes a new instance of the
/// <see cref="BinaryNativeDataFormatBayesPointMachineClassifier{TInstanceSource, TInstance, TLabelSource, TTrainingSettings}"/>
/// class from a reader of a binary stream.
/// </summary>
/// <param name="reader">The binary reader to read the state of the binary Bayes point machine classifier from.</param>
/// <param name="mapping">The mapping used for accessing data in the native format.</param>
protected BinaryNativeDataFormatBayesPointMachineClassifier(IReader reader, IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, bool> mapping)
: base(reader, mapping)
{
reader.ReadSerializationVersion(CustomSerializationVersion);
// Nothing to deserialize
}
/// <summary>
/// Initializes a new instance of the
/// <see cref="NativeDataFormatBayesPointMachineClassifier{TInstanceSource, TInstance, TLabelSource, TLabel, TLabelDistribution, TTrainingSettings, TPredictionSettings}"/>
/// class.
/// </summary>
/// <param name="mapping">The mapping used for accessing data in the native format.</param>
protected NativeDataFormatBayesPointMachineClassifier(IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, TLabel> mapping)
{
Debug.Assert(mapping != null, "The mapping must not be null.");
this.Mapping = mapping;
this.capabilities = new BayesPointMachineClassifierCapabilities();
this.IsTrained = false;
this.isIncrementallyTrained = false;
}
CreateMulticlassClassifier <TInstanceSource, TInstance, TLabelSource>(
IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, int> mapping)
{
if (mapping == null)
{
throw new ArgumentNullException("mapping");
}
return(new CompoundMulticlassNativeDataFormatBayesPointMachineClassifier <TInstanceSource, TInstance, TLabelSource>(mapping));
}
CreateGaussianPriorBinaryClassifier <TInstanceSource, TInstance, TLabelSource>(
IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, bool> mapping)
{
if (mapping == null)
{
throw new ArgumentNullException("mapping");
}
return(new GaussianBinaryNativeDataFormatBayesPointMachineClassifier <TInstanceSource, TInstance, TLabelSource>(mapping));
}
/// <summary>
/// Safely retrieves the total number of features for the specified instance source.
/// </summary>
/// <typeparam name="TInstanceSource">The type of the instance source.</typeparam>
/// <typeparam name="TInstance">The type of an instance.</typeparam>
/// <typeparam name="TLabelSource">The type of the label source.</typeparam>
/// <typeparam name="TLabel">The type of a label.</typeparam>
/// <param name="mapping">The mapping.</param>
/// <param name="instanceSource">An optional instance source.</param>
/// <returns>The total number of features.</returns>
/// <exception cref="MappingException">Thrown if the number of features is negative.</exception>
public static int GetFeatureCountSafe <TInstanceSource, TInstance, TLabelSource, TLabel>(
this IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, TLabel> mapping,
TInstanceSource instanceSource = default(TInstanceSource))
{
int featureCount = mapping.GetFeatureCount(instanceSource);
if (featureCount < 0)
{
throw new MappingException("The total number of features must not be negative.");
}
return(featureCount);
}
/// <summary>
/// Safely retrieves the number of classes that the Bayes point machine classifier is used for.
/// </summary>
/// <typeparam name="TInstanceSource">The type of the instance source.</typeparam>
/// <typeparam name="TInstance">The type of an instance.</typeparam>
/// <typeparam name="TLabelSource">The type of the label source.</typeparam>
/// <typeparam name="TLabel">The type of a label.</typeparam>
/// <param name="mapping">The mapping.</param>
/// <param name="instanceSource">An optional instance source.</param>
/// <param name="labelSource">An optional label source.</param>
/// <returns>The number of classes that the Bayes point machine classifier is used for.</returns>
/// <exception cref="MappingException">Thrown if there are less than three classes.</exception>
public static int GetClassCountSafe <TInstanceSource, TInstance, TLabelSource, TLabel>(
this IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, TLabel> mapping,
TInstanceSource instanceSource = default(TInstanceSource),
TLabelSource labelSource = default(TLabelSource))
{
int classCount = mapping.GetClassCount(instanceSource, labelSource);
if (classCount < 2)
{
throw new MappingException("There must be at least two classes.");
}
return(classCount);
}
/// <summary>
/// Safely retrieves the feature values for a specified instance.
/// </summary>
/// <typeparam name="TInstanceSource">The type of the instance source.</typeparam>
/// <typeparam name="TInstance">The type of an instance.</typeparam>
/// <typeparam name="TLabelSource">The type of the label source.</typeparam>
/// <typeparam name="TLabel">The type of a label.</typeparam>
/// <param name="mapping">The mapping.</param>
/// <param name="instance">The instance.</param>
/// <param name="instanceSource">An optional instance source.</param>
/// <returns>The feature values for the specified instance.</returns>
/// <exception cref="MappingException">Thrown if the feature values are null, infinite or NaN.</exception>
public static double[] GetSingleFeatureVectorValuesSafe <TInstanceSource, TInstance, TLabelSource, TLabel>(
this IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, TLabel> mapping,
TInstance instance,
TInstanceSource instanceSource = default(TInstanceSource))
{
double[] featureValues = mapping.GetFeatureValues(instance, instanceSource);
if (featureValues == null)
{
throw new MappingException("The feature values must not be null.");
}
if (featureValues.Any(value => double.IsInfinity(value) || double.IsNaN(value)))
{
throw new MappingException("A feature value must not be infinite or NaN.");
}
return(featureValues);
}
/// <summary>
/// Safely retrieves the feature indexes of all instances from a specified batch of the instance source.
/// </summary>
/// <typeparam name="TInstanceSource">The type of the instance source.</typeparam>
/// <typeparam name="TInstance">The type of an instance.</typeparam>
/// <typeparam name="TLabelSource">The type of the label source.</typeparam>
/// <typeparam name="TLabel">The type of a label.</typeparam>
/// <param name="mapping">The mapping.</param>
/// <param name="instanceSource">The instance source.</param>
/// <param name="batchNumber">An optional batch number. Defaults to 0 and is used only if the instance source is divided into batches.</param>
/// <returns>The feature indexes provided by the specified batch of the instance source.</returns>
/// <exception cref="MappingException">
/// Thrown if the feature indexes are empty or contain instances with duplicate, negative or null indexes.
/// </exception>
public static int[][] GetAllFeatureVectorIndexesSafe <TInstanceSource, TInstance, TLabelSource, TLabel>(
this IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, TLabel> mapping,
TInstanceSource instanceSource,
int batchNumber = 0)
{
if (batchNumber < 0)
{
throw new MappingException("The number of a batch must not be negative.");
}
int[][] featureIndexes = mapping.GetFeatureIndexes(instanceSource, batchNumber);
if (featureIndexes == null)
{
return(null);
}
if (featureIndexes.Length == 0)
{
throw new MappingException("The feature indexes must not be empty.");
}
if (featureIndexes.Any(indexes => indexes == null))
{
throw new MappingException("The feature indexes must not be null for any instance.");
}
for (int instance = 0; instance < featureIndexes.Length; instance++)
{
// Check that indexes are not negative
if (featureIndexes[instance].Any(index => index < 0))
{
throw new MappingException("The feature indexes must not be negative.");
}
// Check for duplicate indexes
var featureIndexSet = new HashSet <int>(featureIndexes[instance]);
if (featureIndexSet.Count != featureIndexes[instance].Length)
{
throw new MappingException("The feature indexes must not contain duplicates.");
}
}
return(featureIndexes);
}
/// <summary>
/// Initializes a new instance of the
/// <see cref="NativeDataFormatBayesPointMachineClassifier{TInstanceSource, TInstance, TLabelSource, TLabel, TLabelDistribution, TTrainingSettings, TPredictionSettings}"/>
/// class from a reader of a binary stream.
/// </summary>
/// <param name="reader">The binary reader to read the state of the Bayes point machine classifier from.</param>
/// <param name="mapping">The mapping used for accessing data in the native format.</param>
protected NativeDataFormatBayesPointMachineClassifier(
BinaryReader reader,
IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, TLabel> mapping)
{
Debug.Assert(reader != null, "The reader must not be null.");
Debug.Assert(mapping != null, "The mapping must not be null.");
this.Mapping = mapping;
this.capabilities = new BayesPointMachineClassifierCapabilities();
reader.VerifySerializationGuid(
this.customSerializationGuid, "The binary stream does not contain an Infer.NET Bayes point machine classifier.");
int deserializedVersion = reader.ReadSerializationVersion(CustomSerializationVersion);
if (deserializedVersion == CustomSerializationVersion)
{
this.IsTrained = reader.ReadBoolean();
this.isIncrementallyTrained = reader.ReadBoolean();
this.logModelEvidence = reader.ReadDouble();
}
}
/// <summary>
/// Safely retrieves an array of labels for the specified batch of the instance source.
/// </summary>
/// <typeparam name="TInstanceSource">The type of the instance source.</typeparam>
/// <typeparam name="TInstance">The type of an instance.</typeparam>
/// <typeparam name="TLabelSource">The type of the label source.</typeparam>
/// <typeparam name="TLabel">The type of a label.</typeparam>
/// <param name="mapping">The mapping.</param>
/// <param name="instanceSource">The instance source.</param>
/// <param name="labelSource">An optional label source.</param>
/// <param name="batchNumber">An optional batch number. Defaults to 0 and is used only if the instance and label sources are divided into batches.</param>
/// <returns>The labels provided by the specified batch of the sources.</returns>
/// <exception cref="MappingException">Thrown if the array of labels is null or empty.</exception>
public static TLabel[] GetAllLabelsSafe <TInstanceSource, TInstance, TLabelSource, TLabel>(
this IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, TLabel> mapping,
TInstanceSource instanceSource,
TLabelSource labelSource = default(TLabelSource),
int batchNumber = 0)
{
if (batchNumber < 0)
{
throw new MappingException("The number of a batch must not be negative.");
}
TLabel[] labels = mapping.GetLabels(instanceSource, labelSource, batchNumber);
if (labels == null)
{
throw new MappingException("The labels must not be null.");
}
if (labels.Length == 0)
{
throw new MappingException("The labels must not be empty.");
}
return(labels);
}
/// <summary>
/// Initializes a new instance of the
/// <see cref="GaussianMulticlassNativeDataFormatBayesPointMachineClassifier{TInstanceSource, TInstance, TLabelSource}"/>
/// class from a reader of a binary stream.
/// </summary>
/// <param name="reader">The binary reader to read the state of the multi-class Bayes point machine classifier from.</param>
/// <param name="mapping">The mapping used for accessing data in the native format.</param>
public GaussianMulticlassNativeDataFormatBayesPointMachineClassifier(BinaryReader reader, IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, int> mapping)
: base(reader, mapping)
{
int deserializedVersion = reader.ReadSerializationVersion(CustomSerializationVersion);
if (deserializedVersion == CustomSerializationVersion)
{
// Deserialize settings, inference algorithms
this.Settings = new GaussianMulticlassBayesPointMachineClassifierSettings <int>(reader, () => this.IsTrained);
if (this.IsTrained)
{
this.InferenceAlgorithms = new GaussianMulticlassFactorizedInferenceAlgorithms(reader);
}
}
}
/// <summary>
/// Initializes a new instance of the
/// <see cref="GaussianMulticlassNativeDataFormatBayesPointMachineClassifier{TInstanceSource, TInstance, TLabelSource}"/>
/// class.
/// </summary>
/// <param name="mapping">The mapping used for accessing data in the native format.</param>
public GaussianMulticlassNativeDataFormatBayesPointMachineClassifier(IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, int> mapping)
: base(mapping)
{
this.Settings = new GaussianMulticlassBayesPointMachineClassifierSettings <int>(() => this.IsTrained);
}
/// <summary>
/// Initializes a new instance of the
/// <see cref="MulticlassNativeDataFormatBayesPointMachineClassifier{TInstanceSource, TInstance, TLabelSource, TTrainingSettings}"/>
/// class from a reader of a binary stream.
/// </summary>
/// <param name="reader">The binary reader to read the state of the multi-class Bayes point machine classifier from.</param>
/// <param name="mapping">The mapping used for accessing data in the native format.</param>
protected MulticlassNativeDataFormatBayesPointMachineClassifier(IReader reader, IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, int> mapping)
: base(reader, mapping)
{
int deserializedVersion = reader.ReadSerializationVersion(CustomSerializationVersion);
if (deserializedVersion == CustomSerializationVersion)
{
this.ClassCount = reader.ReadInt32();
}
}
/// <summary>
/// Initializes a new instance of the
/// <see cref="MulticlassNativeDataFormatBayesPointMachineClassifier{TInstanceSource, TInstance, TLabelSource, TTrainingSettings}"/>
/// class.
/// </summary>
/// <param name="mapping">The mapping used for accessing data in the native format.</param>
protected MulticlassNativeDataFormatBayesPointMachineClassifier(IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, int> mapping)
: base(mapping)
{
}
/// <summary>
/// Initializes a new instance of the
/// <see cref="BinaryNativeDataFormatBayesPointMachineClassifier{TInstanceSource, TInstance, TLabelSource, TTrainingSettings}"/>
/// class.
/// </summary>
/// <param name="mapping">The mapping used for accessing data in the native format.</param>
protected BinaryNativeDataFormatBayesPointMachineClassifier(IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, bool> mapping)
: base(mapping)
{
}
/// <summary>
/// Initializes a new instance of the
/// <see cref="CompoundBinaryNativeDataFormatBayesPointMachineClassifier{TInstanceSource, TInstance, TLabelSource}"/>
/// class.
/// </summary>
/// <param name="mapping">The mapping used for accessing data in the native format.</param>
public CompoundBinaryNativeDataFormatBayesPointMachineClassifier(IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, bool> mapping)
: base(mapping)
{
this.Settings = new BinaryBayesPointMachineClassifierSettings <bool>(() => this.IsTrained);
}
/// <summary>
/// Initializes a new instance of the
/// <see cref="CompoundBinaryNativeDataFormatBayesPointMachineClassifier{TInstanceSource, TInstance, TLabelSource}"/>
/// class from a reader of a binary stream.
/// </summary>
/// <param name="reader">The binary reader to read the state of the binary Bayes point machine classifier from.</param>
/// <param name="mapping">The mapping used for accessing data in the native format.</param>
public CompoundBinaryNativeDataFormatBayesPointMachineClassifier(BinaryReader reader, IBayesPointMachineClassifierMapping <TInstanceSource, TInstance, TLabelSource, bool> mapping)
: base(reader, mapping)
{
int deserializedVersion = reader.ReadSerializationVersion(CustomSerializationVersion);
if (deserializedVersion == CustomSerializationVersion)
{
// Deserialize settings and inference algorithms
this.Settings = new BinaryBayesPointMachineClassifierSettings <bool>(reader, () => this.IsTrained);
if (this.IsTrained)
{
this.InferenceAlgorithms = new CompoundBinaryFactorizedInferenceAlgorithms(reader);
}
}
}
