org.carrot2.matrix.factorization
Class IterationNumberGuesser

java.lang.Object
  org.carrot2.matrix.factorization.IterationNumberGuesser

public class IterationNumberGuesser
extends Object

This class helps to guesstimate the number of iterations for iterative factorization algorithms. Note: for the time being it uses a very simple linear model with lots of dodgy assumptions.

Nested Class Summary

static class

IterationNumberGuesser.FactorizationQuality Factorization quality.

Constructor Summary

IterationNumberGuesser()

Method Summary

static boolean

setEstimatedIterationsNumber(IterativeMatrixFactorizationFactory factory, org.apache.mahout.math.matrix.DoubleMatrix2D A, IterationNumberGuesser.FactorizationQuality qualityLevel) Sets the guesstimated iterations number in the factory.

Methods inherited from class java.lang.Object

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

Constructor Detail

IterationNumberGuesser

public IterationNumberGuesser()

Method Detail

setEstimatedIterationsNumber

public static boolean setEstimatedIterationsNumber(IterativeMatrixFactorizationFactory factory,
                                                   org.apache.mahout.math.matrix.DoubleMatrix2D A,
                                                   IterationNumberGuesser.FactorizationQuality qualityLevel)

Sets the guesstimated iterations number in the factory. Different models are used depending on the actual factorization algorithm, the seeding strategy and the qualityLevel. For the time being a crude linear model is assumed based on:

• the number of columns in A. The number of rows is assumed to be 2.8 bigger than the number of rows (2.8 is pretty much random, the value is taken from an application to search results clustering). In case matrix has different proportions, the size will be scaled so that the cardinality of the matrix (total number of elements) is maintained. The model supports column counts (after scaling) ranging from 50 to 400.
• the number of requested base vectors (read from the factorization factory). The model supports k = 5 ... 50.

A new maximum number of iterations will be set in the factory only if the requested parameters fall within the ranges supported by the model. In this case true will be returned.