Package edu.gsu.cs.dmlab.sparse.dictionary

Class OnlineDictionaryLearner

java.lang.Object

edu.gsu.cs.dmlab.sparse.dictionary.OnlineDictionaryLearner

All Implemented Interfaces:

ISparseDictionaryLearner

public class OnlineDictionaryLearner
extends Object
implements ISparseDictionaryLearner

Implements the online dictionary learning algorithm of Mairal et al., 2010. The algorithm can be found in "Sparse Modeling, Theory, Algorithms, and Applications" by Irina Rish and Genady Ya. Grabarnik. Published by CRC Press 2015, on page 172.

Author:

Dustin Kempton, Data Mining Lab, Georgia State University

Constructor Summary

Constructors

Constructor	Description
OnlineDictionaryLearner(ISparseVectorApproximator leqSolver, IDictionaryUpdater dictionaryUpdater, IDictionaryCleaner cleaner, double fractionOfInputSize, int batchSize, int numThreads)	Constructor that specifies the number of dictionary elements to contain in the returned dictionary as a fraction of the input matrix size.
OnlineDictionaryLearner(ISparseVectorApproximator leqSolver, IDictionaryUpdater dictionaryUpdater, IDictionaryCleaner cleaner, int dictionaryElements, int batchSize, int numThreads)	Constructor that specifies the number of dictionary elements to contain in the returned dictionary.

Method Summary

Modifier and Type	Method	Description
void	finalize()
void	rank1Update(smile.math.matrix.JMatrix mat, double[] vec1, double[] vec2)
smile.math.matrix.DenseMatrix	train(List<double[]> inputSamples)	Trains the sparse dictionary for the passed Data.

Methods inherited from class java.lang.Object

clone, equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

OnlineDictionaryLearner

public OnlineDictionaryLearner(ISparseVectorApproximator leqSolver,
                               IDictionaryUpdater dictionaryUpdater,
                               IDictionaryCleaner cleaner,
                               double fractionOfInputSize,
                               int batchSize,
                               int numThreads)

Constructor that specifies the number of dictionary elements to contain in the returned dictionary as a fraction of the input matrix size.

Parameters:

leqSolver - System of Linear Equations Solver used by this object.

dictionaryUpdater - Dictionary updated used by this object.

cleaner - Dictionary cleaner used by this object.

fractionOfInputSize - The fraction of the input size that is wished as an output number of dictionary elements. Must be 0.025 (2.5%) or greater.

batchSize - The number of elements to select from the input between dictionary updates.

numThreads - The number of threads to use when processing batch. -1 for all available > 0 for a specific number.

OnlineDictionaryLearner

public OnlineDictionaryLearner(ISparseVectorApproximator leqSolver,
                               IDictionaryUpdater dictionaryUpdater,
                               IDictionaryCleaner cleaner,
                               int dictionaryElements,
                               int batchSize,
                               int numThreads)

Constructor that specifies the number of dictionary elements to contain in the returned dictionary.

Parameters:

leqSolver - System of Linear Equations Solver used by this object.

dictionaryUpdater - Dictionary updated used by this object.

cleaner - Dictionary cleaner used by this object.

dictionaryElements - The number of dictionary elements to include in the returned dictionary.

batchSize - The number of elements to select from the input between dictionary updates.

numThreads - The number of threads to use when processing batch. -1 for all available > 0 for a specific number.

Method Detail

finalize

public void finalize()
              throws Throwable

Overrides:

finalize in class Object

Throws:

Throwable

train

public smile.math.matrix.DenseMatrix train(List<double[]> inputSamples)
                                    throws MatrixDimensionMismatch,
                                           VectorDimensionMismatch

Description copied from interface: ISparseDictionaryLearner

Trains the sparse dictionary for the passed Data.

Specified by:

train in interface ISparseDictionaryLearner

Parameters:

inputSamples - The data to train on. Each element in the list is a column vector in the input matrix.

Returns:

The sparse dictionary.

Throws:

MatrixDimensionMismatch - When matrix dimensions are mismatched for this operation.

VectorDimensionMismatch - If this is thrown then there was something wrong internally.

rank1Update

public void rank1Update(smile.math.matrix.JMatrix mat,
                        double[] vec1,
                        double[] vec2)