org.deeplearning4j.util

Class MatrixUtil

java.lang.Object

org.deeplearning4j.util.MatrixUtil

public class MatrixUtil
extends Object

Constructor Summary

Constructors

Constructor and Description
MatrixUtil()

Method Summary

Methods

Modifier and Type	Method and Description
static org.jblas.DoubleMatrix	add(org.jblas.DoubleMatrix a, org.jblas.DoubleMatrix b)
static void	assertIntMatrix(org.jblas.DoubleMatrix matrix)
static org.jblas.DoubleMatrix	avg(org.jblas.DoubleMatrix... matrices)
static org.jblas.DoubleMatrix	binomial(org.jblas.DoubleMatrix p, int n, org.apache.commons.math3.random.RandomGenerator rng) Generate a binomial distribution based on the given rng, a matrix of p values, and a max number.
static void	columnNormalizeBySum(org.jblas.DoubleMatrix x)
static org.jblas.DoubleMatrix	columnStd(org.jblas.DoubleMatrix m) Calculates the column wise standard deviations of the matrix
static org.jblas.DoubleMatrix	columnStdDeviation(org.jblas.DoubleMatrix m)
static org.jblas.DoubleMatrix	columnWiseMean(org.jblas.DoubleMatrix x, int axis)
static void	complainAboutMissMatchedMatrices(org.jblas.DoubleMatrix d1, org.jblas.DoubleMatrix d2)
static org.jblas.DoubleMatrix	convolution2D(org.jblas.DoubleMatrix input, int width, int height, org.jblas.DoubleMatrix kernel, int kernelWidth, int kernelHeight) Takes a 2D array of grey-levels and a kernel and applies the convolution over the area of the image specified by width and height.
static double	cosine(org.jblas.DoubleMatrix matrix)
static double	cosineSim(org.jblas.DoubleMatrix d1, org.jblas.DoubleMatrix d2)
static void	discretizeColumns(org.jblas.DoubleMatrix toDiscretize, int numBins)
static org.jblas.DoubleMatrix	divColumnsByStDeviation(org.jblas.DoubleMatrix m) Divides the given matrix's columns by each column's respective standard deviations
static org.jblas.DoubleMatrix	dot(org.jblas.DoubleMatrix a, org.jblas.DoubleMatrix b)
static void	ensureValidOutcomeMatrix(org.jblas.DoubleMatrix out)
static boolean	isInfinite(org.jblas.DoubleMatrix test)
static boolean	isNaN(org.jblas.DoubleMatrix test)
static boolean	isValidOutcome(org.jblas.DoubleMatrix out)
static org.jblas.DoubleMatrix	log(org.jblas.DoubleMatrix vals) A log impl that prevents numerical underflow Any number that's infinity or NaN is replaced by 1e-6.
static double	magnitude(org.jblas.DoubleMatrix vec)
static void	max(double minNumber, org.jblas.DoubleMatrix matrix) Cuts all numbers below a certain cut off
static double	max(org.jblas.DoubleMatrix matrix)
static int	maxIndex(org.jblas.DoubleMatrix matrix)
static org.jblas.DoubleMatrix	mean(org.jblas.DoubleMatrix input, int axis)
static double	meanSquaredError(org.jblas.DoubleMatrix input, org.jblas.DoubleMatrix other) Returns the mean squared error of the 2 matrices.
static double	min(org.jblas.DoubleMatrix matrix)
static org.jblas.DoubleMatrix	normal(org.apache.commons.math3.random.RandomGenerator rng, org.jblas.DoubleMatrix standardDeviations) Sample from a normal distribution given a mean of zero and a matrix of standard deviations.
static org.jblas.DoubleMatrix	normal(org.apache.commons.math3.random.RandomGenerator rng, org.jblas.DoubleMatrix mean, double sigma) A uniform sample ranging from 0 to sigma.
static org.jblas.DoubleMatrix	normal(org.apache.commons.math3.random.RandomGenerator rng, org.jblas.DoubleMatrix mean, org.jblas.DoubleMatrix variance) A uniform sample ranging from 0 to sigma.
static org.jblas.DoubleMatrix	normalize(org.jblas.DoubleMatrix input)
static org.jblas.DoubleMatrix	normalizeByColumnMeans(org.jblas.DoubleMatrix m) Subtracts by column mean.
static org.jblas.DoubleMatrix	normalizeByColumnSums(org.jblas.DoubleMatrix m)
static org.jblas.DoubleMatrix	normalizeByRowSums(org.jblas.DoubleMatrix m)
static void	normalizeMatrix(org.jblas.DoubleMatrix toNormalize)
static org.jblas.DoubleMatrix	oneDiv(org.jblas.DoubleMatrix ep)
static org.jblas.DoubleMatrix	oneMinus(org.jblas.DoubleMatrix ep)
static org.jblas.DoubleMatrix	out(org.jblas.DoubleMatrix a, org.jblas.DoubleMatrix b)
static org.jblas.DoubleMatrix	outcomes(org.jblas.DoubleMatrix d)
static org.jblas.DoubleMatrix	roundToTheNearest(org.jblas.DoubleMatrix d, int num)
static org.jblas.DoubleMatrix	rowStd(org.jblas.DoubleMatrix m) Calculates the column wise standard deviations of the matrix
static org.jblas.DoubleMatrix	scalarMinus(double scalar, org.jblas.DoubleMatrix ep)
static void	scaleByMax(org.jblas.DoubleMatrix toScale)
static org.jblas.DoubleMatrix	sigmoid(org.jblas.DoubleMatrix x)
static double	singlePixelConvolution(org.jblas.DoubleMatrix input, int x, int y, org.jblas.DoubleMatrix k, int kernelWidth, int kernelHeight) Takes an image (grey-levels) and a kernel and a position, applies the convolution at that position and returns the new pixel value.
static org.jblas.DoubleMatrix	softmax(org.jblas.DoubleMatrix input) Soft max function row_maxes is a row vector (max for each row) row_maxes = rowmaxes(input) diff = exp(input - max) / diff.rowSums()
static org.jblas.DoubleMatrix	sum(org.jblas.DoubleMatrix input, int axis)
static double	sumSquaredError(org.jblas.DoubleMatrix input, org.jblas.DoubleMatrix other) Returns the sum squared error of the 2 matrices.
static org.jblas.DoubleMatrix	toMatrix(int[] arr)
static org.jblas.DoubleMatrix	toMatrix(int[][] arr)
static org.jblas.DoubleMatrix	toOutcomeVector(int index, int numOutcomes)
static org.jblas.DoubleMatrix	uniform(org.apache.commons.math3.random.RandomGenerator rng, int rows, int columns) A uniform sample ranging from 0 to 1.
static org.jblas.DoubleMatrix	unitVec(org.jblas.DoubleMatrix toScale)
static org.jblas.DoubleMatrix	unroll(org.jblas.DoubleMatrix d)
static org.jblas.DoubleMatrix	variance(org.jblas.DoubleMatrix input)
static DataSet	xorData(int n)
static DataSet	xorData(int n, int columns)

Methods inherited from class java.lang.Object

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

Constructor Detail

MatrixUtil

public MatrixUtil()

Method Detail

complainAboutMissMatchedMatrices

public static void complainAboutMissMatchedMatrices(org.jblas.DoubleMatrix d1,
                                    org.jblas.DoubleMatrix d2)

max

public static void max(double minNumber,
       org.jblas.DoubleMatrix matrix)

Cuts all numbers below a certain cut off

Parameters:

minNumber - the min number to check

matrix - the matrix to max by

scaleByMax

public static void scaleByMax(org.jblas.DoubleMatrix toScale)

variance

public static org.jblas.DoubleMatrix variance(org.jblas.DoubleMatrix input)

singlePixelConvolution

public static double singlePixelConvolution(org.jblas.DoubleMatrix input,
                            int x,
                            int y,
                            org.jblas.DoubleMatrix k,
                            int kernelWidth,
                            int kernelHeight)

Takes an image (grey-levels) and a kernel and a position, applies the convolution at that position and returns the new pixel value.

Parameters:

input - The 2D double array representing the image.

x - The x coordinate for the position of the convolution.

y - The y coordinate for the position of the convolution.

k - The 2D array representing the kernel.

kernelWidth - The width of the kernel.

kernelHeight - The height of the kernel.

Returns:

The new pixel value after the convolution.

convolution2D

public static org.jblas.DoubleMatrix convolution2D(org.jblas.DoubleMatrix input,
                                   int width,
                                   int height,
                                   org.jblas.DoubleMatrix kernel,
                                   int kernelWidth,
                                   int kernelHeight)

Takes a 2D array of grey-levels and a kernel and applies the convolution over the area of the image specified by width and height.

Parameters:

input - the 2D double array representing the image

width - the width of the image

height - the height of the image

kernel - the 2D array representing the kernel

kernelWidth - the width of the kernel

kernelHeight - the height of the kernel

Returns:

the 2D array representing the new image

xorData

public static DataSet xorData(int n)

xorData

public static DataSet xorData(int n,
              int columns)

magnitude

public static double magnitude(org.jblas.DoubleMatrix vec)

unroll

public static org.jblas.DoubleMatrix unroll(org.jblas.DoubleMatrix d)

outcomes

public static org.jblas.DoubleMatrix outcomes(org.jblas.DoubleMatrix d)

cosineSim

public static double cosineSim(org.jblas.DoubleMatrix d1,
               org.jblas.DoubleMatrix d2)

normalize

public static org.jblas.DoubleMatrix normalize(org.jblas.DoubleMatrix input)

cosine

public static double cosine(org.jblas.DoubleMatrix matrix)

unitVec

public static org.jblas.DoubleMatrix unitVec(org.jblas.DoubleMatrix toScale)

uniform

public static org.jblas.DoubleMatrix uniform(org.apache.commons.math3.random.RandomGenerator rng,
                             int rows,
                             int columns)

A uniform sample ranging from 0 to 1.

Parameters:

rng - the rng to use

rows - the number of rows of the matrix

columns - the number of columns of the matrix

Returns:

a uniform sample of the given shape and size with numbers between 0 and 1

normal

public static org.jblas.DoubleMatrix normal(org.apache.commons.math3.random.RandomGenerator rng,
                            org.jblas.DoubleMatrix mean,
                            double sigma)

A uniform sample ranging from 0 to sigma.

Parameters:

rng - the rng to use

mean, - the matrix mean from which to generate values from

sigma - the standard deviation to use to generate the gaussian noise

Returns:

a uniform sample of the given shape and size with numbers between 0 and 1

normal

public static org.jblas.DoubleMatrix normal(org.apache.commons.math3.random.RandomGenerator rng,
                            org.jblas.DoubleMatrix mean,
                            org.jblas.DoubleMatrix variance)

A uniform sample ranging from 0 to sigma.

Parameters:

rng - the rng to use

mean, - the matrix mean from which to generate values from

variance - the variance matrix where each column is the variance for the respective columns of the matrix

Returns:

a uniform sample of the given shape and size with numbers between 0 and 1

normal

public static org.jblas.DoubleMatrix normal(org.apache.commons.math3.random.RandomGenerator rng,
                            org.jblas.DoubleMatrix standardDeviations)

Sample from a normal distribution given a mean of zero and a matrix of standard deviations.

Parameters:

rng - the rng to use

standard - deviations, the matrix standard deviations to use for the respective columns of the matrix

Returns:

a uniform sample of the given shape and size

isValidOutcome

public static boolean isValidOutcome(org.jblas.DoubleMatrix out)

min

public static double min(org.jblas.DoubleMatrix matrix)

max

public static double max(org.jblas.DoubleMatrix matrix)

ensureValidOutcomeMatrix

public static void ensureValidOutcomeMatrix(org.jblas.DoubleMatrix out)

assertIntMatrix

public static void assertIntMatrix(org.jblas.DoubleMatrix matrix)

isInfinite

public static boolean isInfinite(org.jblas.DoubleMatrix test)

isNaN

public static boolean isNaN(org.jblas.DoubleMatrix test)

discretizeColumns

public static void discretizeColumns(org.jblas.DoubleMatrix toDiscretize,
                     int numBins)

roundToTheNearest

public static org.jblas.DoubleMatrix roundToTheNearest(org.jblas.DoubleMatrix d,
                                       int num)

columnNormalizeBySum

public static void columnNormalizeBySum(org.jblas.DoubleMatrix x)

toOutcomeVector

public static org.jblas.DoubleMatrix toOutcomeVector(int index,
                                     int numOutcomes)

toMatrix

public static org.jblas.DoubleMatrix toMatrix(int[][] arr)

toMatrix

public static org.jblas.DoubleMatrix toMatrix(int[] arr)

add

public static org.jblas.DoubleMatrix add(org.jblas.DoubleMatrix a,
                         org.jblas.DoubleMatrix b)

softmax

public static org.jblas.DoubleMatrix softmax(org.jblas.DoubleMatrix input)

Soft max function row_maxes is a row vector (max for each row) row_maxes = rowmaxes(input) diff = exp(input - max) / diff.rowSums()

Parameters:

input - the input for the softmax

Returns:

the softmax output (a probability matrix) scaling each row to between 0 and 1

mean

public static org.jblas.DoubleMatrix mean(org.jblas.DoubleMatrix input,
                          int axis)

sum

public static org.jblas.DoubleMatrix sum(org.jblas.DoubleMatrix input,
                         int axis)

binomial

public static org.jblas.DoubleMatrix binomial(org.jblas.DoubleMatrix p,
                              int n,
                              org.apache.commons.math3.random.RandomGenerator rng)

Generate a binomial distribution based on the given rng, a matrix of p values, and a max number.

Parameters:

p - the p matrix to use

n - the n to use

rng - the rng to use

Returns:

a binomial distribution based on the one n, the passed in p values, and rng

columnWiseMean

public static org.jblas.DoubleMatrix columnWiseMean(org.jblas.DoubleMatrix x,
                                    int axis)

avg

public static org.jblas.DoubleMatrix avg(org.jblas.DoubleMatrix... matrices)

maxIndex

public static int maxIndex(org.jblas.DoubleMatrix matrix)

sigmoid

public static org.jblas.DoubleMatrix sigmoid(org.jblas.DoubleMatrix x)

dot

public static org.jblas.DoubleMatrix dot(org.jblas.DoubleMatrix a,
                         org.jblas.DoubleMatrix b)

out

public static org.jblas.DoubleMatrix out(org.jblas.DoubleMatrix a,
                         org.jblas.DoubleMatrix b)

scalarMinus

public static org.jblas.DoubleMatrix scalarMinus(double scalar,
                                 org.jblas.DoubleMatrix ep)

oneMinus

public static org.jblas.DoubleMatrix oneMinus(org.jblas.DoubleMatrix ep)

oneDiv

public static org.jblas.DoubleMatrix oneDiv(org.jblas.DoubleMatrix ep)

columnStd

public static org.jblas.DoubleMatrix columnStd(org.jblas.DoubleMatrix m)

Calculates the column wise standard deviations of the matrix

Parameters:

m - the matrix to use

Returns:

the standard deviations of each column in the matrix as a row matrix

rowStd

public static org.jblas.DoubleMatrix rowStd(org.jblas.DoubleMatrix m)

Calculates the column wise standard deviations of the matrix

Parameters:

m - the matrix to use

Returns:

the standard deviations of each column in the matrix as a row matrix

meanSquaredError

public static double meanSquaredError(org.jblas.DoubleMatrix input,
                      org.jblas.DoubleMatrix other)

Returns the mean squared error of the 2 matrices. Note that the matrices must be the same length or an IllegalArgumentException is thrown

Parameters:

input - the first one

other - the second one

Returns:

the mean square error of the matrices

log

public static org.jblas.DoubleMatrix log(org.jblas.DoubleMatrix vals)

A log impl that prevents numerical underflow Any number that's infinity or NaN is replaced by 1e-6.

Parameters:

vals - the vals to convert to log

Returns:

the log of the numbers or 1e-6 for anomalies

sumSquaredError

public static double sumSquaredError(org.jblas.DoubleMatrix input,
                     org.jblas.DoubleMatrix other)

Returns the sum squared error of the 2 matrices. Note that the matrices must be the same length or an IllegalArgumentException is thrown

Parameters:

input - the first one

other - the second one

Returns:

the sum square error of the matrices

normalizeMatrix

public static void normalizeMatrix(org.jblas.DoubleMatrix toNormalize)

normalizeByColumnSums

public static org.jblas.DoubleMatrix normalizeByColumnSums(org.jblas.DoubleMatrix m)

columnStdDeviation

public static org.jblas.DoubleMatrix columnStdDeviation(org.jblas.DoubleMatrix m)

divColumnsByStDeviation

public static org.jblas.DoubleMatrix divColumnsByStDeviation(org.jblas.DoubleMatrix m)

Divides the given matrix's columns by each column's respective standard deviations

Parameters:

m - the matrix to divide

Returns:

the column divided by the standard deviation

normalizeByColumnMeans

public static org.jblas.DoubleMatrix normalizeByColumnMeans(org.jblas.DoubleMatrix m)

Subtracts by column mean. This ensures a mean of zero. This is part of normalizing inputs for a neural net

Parameters:

m - the matrix to normalize

Returns:

the normalized matrix which each column subtracted by its mean

normalizeByRowSums

public static org.jblas.DoubleMatrix normalizeByRowSums(org.jblas.DoubleMatrix m)