File predictor_common.hpp

template<typename T, typename FeatureMatrix, typename Model>
class PredictionDispatcher
#include <predictor_common.hpp>

Allows to obtain predictions for multiple query examples by delegating the prediction for individual examples to another class.

Template Parameters:

  • T – The type of the predictions

  • FeatureMatrix – The type of the feature matrix that provides row-wise access to the feature values of the query examples

  • Model – The type of the rule-based model that is used to obtain predictions

Public Functions

inline void predict(const IPredictionDelegate &delegate, const FeatureMatrix &featureMatrix, typename Model::const_iterator rulesBegin, typename Model::const_iterator rulesEnd, MultiThreadingSettings multiThreadingSettings) const

Obtains predictions for multiple query examples by delegating the prediction for individual examples to a given PredictionDelegate.

Parameters:

  • delegate – A reference to an object of type IPredictionDelegate, the prediction for individual examples should be delegated to

  • featureMatrix – A reference to an object of template type FeatureMatrix that provides row-wise access to the feature values of the query examples

  • rulesBegin – An iterator of type Model::const_iterator to the first rule that should be used for prediction

  • rulesEnd – An iterator of type Model::const_iterator to the last rule (exclusive) that should be used for prediction

  • multiThreadingSettings – An object of type MultiThreadingSettings that stores the settings to be used for making predictions for different query examples in parallel

class IPredictionDelegate
#include <predictor_common.hpp>

Defines an interface for all classes, the prediction for individual examples can be delegated to by a PredictionDispatcher.

Public Functions

inline virtual ~IPredictionDelegate()
virtual void predictForExample(const FeatureMatrix &featureMatrix, typename Model::const_iterator rulesBegin, typename Model::const_iterator rulesEnd, uint32 threadIndex, uint32 exampleIndex, uint32 predictionIndex) const = 0

Obtains predictions for a single query example.

Parameters:

  • featureMatrix – A reference to an object of template type FeatureMatrix that provides row-wise access to the feature values of the query examples

  • rulesBegin – An iterator of type Model::const_iterator to the first rule that should be used for prediction

  • rulesEnd – An iterator of type Model::const_iterator to the last rule (exclusive) that should be used for prediction

  • threadIndex – The index of the thread used for prediction

  • exampleIndex – The index of the query example to predict for

  • predictionIndex – The index of the row in the prediction matrix, where the predictions should be stored

template<typename FeatureMatrix, typename Model>
class BinarySparsePredictionDispatcher
#include <predictor_common.hpp>

Allows to obtain sparse binary predictions for multiple query examples by delegating the prediction for individual examples to another class.

Template Parameters:

  • FeatureMatrix – The type of the feature matrix that provides row-wise access to the feature values of the query examples

  • Model – The type of the rule-based model that is used to obtain predictions

Public Functions

inline uint32 predict(const IPredictionDelegate &delegate, const FeatureMatrix &featureMatrix, typename Model::const_iterator rulesBegin, typename Model::const_iterator rulesEnd, MultiThreadingSettings multiThreadingSettings) const

Obtains predictions for multiple query examples by delegating the prediction for individual examples to a given IPredictionDelegate.

Parameters:

  • delegate – A reference to an object of type IPredictionDelegate, the prediction for individual examples should be delegated to

  • featureMatrix – A reference to an object of template type FeatureMatrix that provides row-wise access to the feature values of the query examples

  • rulesBegin – An iterator of type Model::const_iterator to the first rule that should be used for prediction

  • rulesEnd – An iterator of type Model::const_iterator to the last rule (exclusive) that should be used for prediction

  • multiThreadingSettings – An object of type MultiThreadingSettings that stores the settings to be used for making predictions for different query examples in parallel

Returns:

The total number of dense predictions explicitly stored in the affected row of the prediction matrix

class IPredictionDelegate
#include <predictor_common.hpp>

Defines an interface for all classes, the prediction for individual examples can be delegated to by a BinarySparsePredictionDispatcher.

Public Functions

inline virtual ~IPredictionDelegate()
virtual uint32 predictForExample(const FeatureMatrix &featureMatrix, typename Model::const_iterator rulesBegin, typename Model::const_iterator rulesEnd, uint32 threadIndex, uint32 exampleIndex, uint32 predictionIndex) const = 0

Obtains predictions for a single query example.

Parameters:

  • featureMatrix – A reference to an object of template type FeatureMatrix that provides row-wise access to the feature values of the query examples

  • rulesBegin – An iterator of type Model::const_iterator to the first rule that should be used for prediction

  • rulesEnd – An iterator of type Model::const_iterator to the last rule (exclusive) that should be used for prediction

  • threadIndex – The index of the thread used for prediction

  • exampleIndex – The index of the query example to predict for

  • predictionIndex – The index of the row in the prediction matrix, where the predictions should be stored

Returns:

The number of dense elements explicitly stored in the affected row of the prediction matrix

template<typename FeatureMatrix, typename Model, typename PredictionMatrix>
class AbstractIncrementalPredictor : public IIncrementalPredictor<PredictionMatrix>
#include <predictor_common.hpp>

An abstract base class for all implementations of the class IIncrementalPredictor.

Template Parameters:

  • FeatureMatrix – The type of the feature matrix that provides row-wise access to the feature values of the query examples

  • Model – The type of the rule-based model that is used to obtain predictions

  • PredictionMatrix – The type of the matrix that is used to store the predictions

Public Functions

inline AbstractIncrementalPredictor(const FeatureMatrix &featureMatrix, const Model &model, MultiThreadingSettings multiThreadingSettings, uint32 maxRules)
Parameters:

  • featureMatrix – A reference to an object of template type FeatureMatrix that provides row-wise access to the feature values of the query examples

  • model – A reference to an object of template type Model that should be used for prediction

  • multiThreadingSettings – An object of type MultiThreadingSettings that stores the settings to be used for making predictions for different query examples in parallel

  • maxRules – The maximum number of rules to be used for prediction. Must be at least 1 or 0, if the number of rules should not be restricted

inline virtual ~AbstractIncrementalPredictor() override
inline virtual uint32 getNumNext() const final override

Returns the number of remaining ensemble members that have not been used yet.

Returns:

The number of remaining ensemble members

inline virtual PredictionMatrix &applyNext(uint32 stepSize) final override

Updates the current predictions by considering several of the remaining ensemble members. If not enough ensemble members are remaining, only the available ones will be used for updating the current predictions.

Parameters:

stepSize – The number of additional ensemble members to be considered for prediction

Returns:

A reference to an object of template type PredictionMatrix that stores the updated predictions

Protected Functions

virtual PredictionMatrix &applyNext(const FeatureMatrix &featureMatrix, MultiThreadingSettings multiThreadingSettings, typename Model::const_iterator rulesBegin, typename Model::const_iterator rulesEnd) = 0

Must be implemented by subclasses in order to obtain predictions.

Parameters:

  • featureMatrix – A reference to an object of template type FeatureMatrix that provides row-wise access to the feature values of the query examples

  • multiThreadingSettings – An object of type MultiThreadingSettings that stores the settings to be used for making predictions for different query examples in parallel

  • rulesBegin – An iterator of type Model::const_iterator to the first rule that should be used for prediction

  • rulesEnd – An iterator of type Model::const_iterator to the last rule (exclusive) that should be used for prediction

Returns:

A reference to an object of template type PredictionMatrix that stores the predictions

Private Members

const FeatureMatrix &featureMatrix_
const MultiThreadingSettings multiThreadingSettings_
Model::const_iterator current_
Model::const_iterator end_