ai.libs.mlplan.metamining.similaritymeasures

Class F3Optimizer

java.lang.Object

ai.libs.mlplan.metamining.similaritymeasures.F3Optimizer

All Implemented Interfaces:

IHeterogenousSimilarityMeasureComputer

public class F3Optimizer
extends java.lang.Object
implements IHeterogenousSimilarityMeasureComputer

Constructor Summary

Constructors

Constructor and Description
F3Optimizer(double mu)

Method Summary

Modifier and Type	Method and Description
void	build(org.nd4j.linalg.api.ndarray.INDArray X, org.nd4j.linalg.api.ndarray.INDArray W, org.nd4j.linalg.api.ndarray.INDArray R) Learns a matrix U that minimizes F1 (W is ignored here)
double	computeSimilarity(org.nd4j.linalg.api.ndarray.INDArray x, org.nd4j.linalg.api.ndarray.INDArray w) Compute the 'quality of the match' of given feature values for a new problem instance based on the training.
double	getCost(org.nd4j.linalg.api.ndarray.INDArray U, org.nd4j.linalg.api.ndarray.INDArray V) This evaluates F1
float	getFirstDerivative(org.nd4j.linalg.api.ndarray.INDArray U, org.nd4j.linalg.api.ndarray.INDArray V, int s, int t, boolean deriveForU) This compute the derivative of F1 for the (k,l)-th element of the U matrix
org.nd4j.linalg.api.ndarray.INDArray	getGradientAsMatrix(org.nd4j.linalg.api.ndarray.INDArray U, org.nd4j.linalg.api.ndarray.INDArray V, boolean computeDerivationsOfU) This computes the gradient of F1 in matrix form
double	getSquaredFrobeniusNorm(org.nd4j.linalg.api.ndarray.INDArray matrix)
de.jungblut.math.DoubleVector	matrices2vector(org.nd4j.linalg.api.ndarray.INDArray... matrices)
de.jungblut.math.DoubleVector	matrix2vector(org.nd4j.linalg.api.ndarray.INDArray matrix) collapses a matrix of the Nd4j framework into a double vector of Thomas Jungblut's framework
ai.libs.jaicore.basic.sets.SetUtil.Pair<org.nd4j.linalg.api.ndarray.INDArray,org.nd4j.linalg.api.ndarray.INDArray>	vector2matrices(de.jungblut.math.DoubleVector vector, int n, int d, int m, int l)
org.nd4j.linalg.api.ndarray.INDArray	vector2matrix(de.jungblut.math.DoubleVector vector, int m, int n) creates a matrix of the Nd4j framework from a vector of Thomas Jungblut's math framework

Methods inherited from class java.lang.Object

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

Constructor Detail

F3Optimizer

public F3Optimizer(double mu)

Method Detail

build

public void build(org.nd4j.linalg.api.ndarray.INDArray X,
                  org.nd4j.linalg.api.ndarray.INDArray W,
                  org.nd4j.linalg.api.ndarray.INDArray R)

Learns a matrix U that minimizes F1 (W is ignored here)

Specified by:

build in interface IHeterogenousSimilarityMeasureComputer

Parameters:

X - Feature values for instances of the first measure (One row = features of one instance, e.g. meta features of a data set)

W - Feature values for instances of the second measure (One row = features of one instance, e.g. a characterization of a machine learning pipeline)

R - A matrix giving an indication of how good of a match a specific instance of the first measure is to a specific instance of the second measure, i.e. how well a pipeline performs on a data set

vector2matrix

public org.nd4j.linalg.api.ndarray.INDArray vector2matrix(de.jungblut.math.DoubleVector vector,
                                                          int m,
                                                          int n)

creates a matrix of the Nd4j framework from a vector of Thomas Jungblut's math framework

Parameters:

vector -

m -

n -

Returns:

vector2matrices

public ai.libs.jaicore.basic.sets.SetUtil.Pair<org.nd4j.linalg.api.ndarray.INDArray,org.nd4j.linalg.api.ndarray.INDArray> vector2matrices(de.jungblut.math.DoubleVector vector,
                                                                                                                                          int n,
                                                                                                                                          int d,
                                                                                                                                          int m,
                                                                                                                                          int l)

matrix2vector

public de.jungblut.math.DoubleVector matrix2vector(org.nd4j.linalg.api.ndarray.INDArray matrix)

collapses a matrix of the Nd4j framework into a double vector of Thomas Jungblut's framework

Parameters:

matrix -

Returns:

matrices2vector

public de.jungblut.math.DoubleVector matrices2vector(org.nd4j.linalg.api.ndarray.INDArray... matrices)

getCost

public double getCost(org.nd4j.linalg.api.ndarray.INDArray U,
                      org.nd4j.linalg.api.ndarray.INDArray V)

This evaluates F1

Parameters:

RRT -

U -

X -

Returns:

getSquaredFrobeniusNorm

public double getSquaredFrobeniusNorm(org.nd4j.linalg.api.ndarray.INDArray matrix)

getGradientAsMatrix

public org.nd4j.linalg.api.ndarray.INDArray getGradientAsMatrix(org.nd4j.linalg.api.ndarray.INDArray U,
                                                                org.nd4j.linalg.api.ndarray.INDArray V,
                                                                boolean computeDerivationsOfU)

This computes the gradient of F1 in matrix form

Parameters:

R -

U -

X -

Returns:

getFirstDerivative

public float getFirstDerivative(org.nd4j.linalg.api.ndarray.INDArray U,
                                org.nd4j.linalg.api.ndarray.INDArray V,
                                int s,
                                int t,
                                boolean deriveForU)

This compute the derivative of F1 for the (k,l)-th element of the U matrix

Parameters:

RRT -

U -

X -

k -

l -

Returns:

computeSimilarity

public double computeSimilarity(org.nd4j.linalg.api.ndarray.INDArray x,
                                org.nd4j.linalg.api.ndarray.INDArray w)

Description copied from interface: IHeterogenousSimilarityMeasureComputer

Compute the 'quality of the match' of given feature values for a new problem instance based on the training.

Specified by:

computeSimilarity in interface IHeterogenousSimilarityMeasureComputer

Parameters:

x - Feature values for the instance for the first measure (e.g. meta data of a new data set)

w - Feature values for the instance for the second measure (e.g. a characterization of machine learning pipeline)

Returns:

The quality of the match, or similarity for the given vectors