Reranker Framework (ReFr)
Reranking framework for structure prediction and discriminative language modeling
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Pages
List of all members
reranker::PerceptronModel::DefaultUpdater Class Reference

The default update function for perceptron models. More...

#include <perceptron-model.H>

Inheritance diagram for reranker::PerceptronModel::DefaultUpdater:
reranker::Model::Updater reranker::FactoryConstructible

Additional Inherited Members

- Public Member Functions inherited from reranker::FactoryConstructible
virtual ~FactoryConstructible ()
 
virtual void RegisterInitializers (Initializers &initializers)
 
virtual void Init (const Environment *env, const string &arg)
 

Detailed Description

The default update function for perceptron models.

The implementation here does the following operations:

  1. Computes which features to update using the PerceptronModel::ComputeFeaturesToUpdate method.
  2. Updates the gold and top-scoring candidate feature averages using the TrainingVectorSet::UpdateGoldAndCandidateFeatureAverages method.
  3. Computes the step size using the PerceptronModel::ComputeStepSize method.
  4. Updates weights, biasing the model toward the gold candidate’s features and away from the current top-scoring (but “incorrect”) candidate’s features, using the TrainingVectorSet::UpdateWeights method.

Definition at line 171 of file perceptron-model.H.


The documentation for this class was generated from the following files: