Package de.labathome.trustconstr.matrix

Class Matrix

java.lang.Object

de.labathome.trustconstr.matrix.Matrix

Direct Known Subclasses:

DenseMatrix, SparseMatrix

public abstract class Matrix
extends Object

Abstract dense/sparse matrix base. Two concrete subclasses:

• DenseMatrix -- column-major double[] buffer; the BLAS and LAPACK kernels in LinAlg take DenseMatrix.data() directly with no copy.
• SparseMatrix -- Dictionary-of-Keys; cheap incremental setAsDouble(double, long, long) plus SparseMatrix.toCSR() to materialise a CSRMatrix for compute.

The base class exposes only the operations both layouts must support: shape queries, element get/set, arithmetic, norms, slicing, scalar extraction, and an in-place fill/absInPlace. The fast paths (BLAS dgemm / dsyr / dsyr2, LU/QR/SVD/Cholesky via LAPACK) live on DenseMatrix and LinAlg.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static final class	Matrix.Factory	Static constructors.

Constructor Summary

Constructors

Constructor	Description
Matrix()

Method Summary

Modifier and Type	Method	Description
abstract Matrix	abs()
Matrix	absInPlace()	In-place absolute value.
abstract Matrix	copy()
abstract Matrix	divide(double alpha)
abstract Matrix	divide(Matrix elementwise)
double	doubleValue()	Extract the single scalar value of a 1 x 1 matrix.
boolean	equalsContent(Matrix other)
Matrix	fill(double v)	In-place fill with the given scalar.
abstract double	getAsDouble(long row, long col)
abstract long	getColumnCount()
abstract long	getRowCount()
long[]	getSize()
boolean	isColumnVector()
boolean	isRowVector()	Lenient shape predicate: returns true if either dimension is 1 -- i.e. the matrix is shaped like a vector (row, column, or 1×1 scalar).
boolean	isSparse()
protected Matrix	mapAdd(double scalar)	Helper used by plus(double) and minus(double): returns a fresh dense matrix with scalar added to every entry.
Matrix	minus(double scalar)
abstract Matrix	minus(Matrix other)
abstract Matrix	mtimes(Matrix other)
abstract double	norm2()
double	normF()
abstract double	normInf()
Matrix	plus(double scalar)
abstract Matrix	plus(Matrix other)
long	rank()	Rank computed via SVD: count singular values strictly greater than the default LAPACK numerical-zero threshold max(m,n) * ulp(sigma_max).
Matrix	selectColumns(long col)
abstract void	setAsDouble(double v, long row, long col)
Matrix	solve(Matrix rhs)	Solve this * x = rhs via LAPACK dgesv (general LU).
Matrix	subMatrix(long r0, long c0, long r1, long c1)	Slice rows [r0..r1] and columns [c0..c1] (inclusive).
abstract Matrix	times(double alpha)
abstract Matrix	times(Matrix elementwise)
double[]	toColumnArray()
double[][]	toDoubleArray()
abstract Matrix	transpose()

Methods inherited from class java.lang.Object

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

Constructor Details

Matrix

public Matrix()

Method Details

getRowCount

public abstract long getRowCount()

Returns:

number of rows

getColumnCount

public abstract long getColumnCount()

Returns:

number of columns

getSize

public long[] getSize()

Returns:

shape as a length-2 array [rows, cols]

getAsDouble

public abstract double getAsDouble(long row,
 long col)

Parameters:

row - 0-based row index

col - 0-based column index

Returns:

entry at (row, col)

setAsDouble

public abstract void setAsDouble(double v,
 long row,
 long col)

Parameters:

v - value to store

row - 0-based row index

col - 0-based column index

solve

public Matrix solve(Matrix rhs)

Solve this * x = rhs via LAPACK dgesv (general LU). Routes through LinAlg.solve(DenseMatrix, DenseMatrix) after densifying both operands.

Parameters:

rhs - right-hand side

Returns:

solution x

isSparse

public boolean isSparse()

Returns:

true if this is a sparse matrix

toDoubleArray

public double[][] toDoubleArray()

Returns:

a fresh row-major double[][] copy of the contents

copy

public abstract Matrix copy()

Returns:

a deep copy of this matrix

toColumnArray

public double[] toColumnArray()

Returns:

contents of an n x 1 column vector as a fresh double[n]; throws if the matrix is not a column vector

mtimes

public abstract Matrix mtimes(Matrix other)

Parameters:

other - right operand

Returns:

matrix product this * other

plus

public abstract Matrix plus(Matrix other)

Parameters:

other - right operand

Returns:

element-wise sum this + other

minus

public abstract Matrix minus(Matrix other)

Parameters:

other - right operand

Returns:

element-wise difference this - other

plus

public Matrix plus(double scalar)

Parameters:

scalar - value to add elementwise

Returns:

this + scalar

minus

public Matrix minus(double scalar)

Parameters:

scalar - value to subtract elementwise

Returns:

this - scalar

times

public abstract Matrix times(double alpha)

Parameters:

alpha - scalar

Returns:

alpha * this

times

public abstract Matrix times(Matrix elementwise)

Parameters:

elementwise - same-shape factor

Returns:

Hadamard product this .* elementwise

divide

public abstract Matrix divide(double alpha)

Parameters:

alpha - scalar

Returns:

this / alpha

divide

public abstract Matrix divide(Matrix elementwise)

Parameters:

elementwise - same-shape divisor

Returns:

Hadamard division this ./ elementwise

transpose

public abstract Matrix transpose()

Returns:

the transpose of this matrix

abs

public abstract Matrix abs()

Returns:

element-wise absolute value as a fresh matrix

norm2

public abstract double norm2()

Returns:

Frobenius norm

normInf

public abstract double normInf()

Returns:

infinity-norm (largest absolute entry)

subMatrix

public Matrix subMatrix(long r0,
 long c0,
 long r1,
 long c1)

Slice rows [r0..r1] and columns [c0..c1] (inclusive). Always returns a fresh dense matrix.

Parameters:

r0 - first row (inclusive)

c0 - first column (inclusive)

r1 - last row (inclusive)

c1 - last column (inclusive)

Returns:

the (r1-r0+1) x (c1-c0+1) slice

selectColumns

public Matrix selectColumns(long col)

Parameters:

col - 0-based column index

Returns:

that single column as an n x 1 matrix

absInPlace

public Matrix absInPlace()

In-place absolute value.

Returns:

this after replacing every entry by its absolute value

rank

public long rank()

Rank computed via SVD: count singular values strictly greater than the default LAPACK numerical-zero threshold max(m,n) * ulp(sigma_max).

Returns:

numerical rank (zero for empty matrices)

isRowVector

public boolean isRowVector()

Lenient shape predicate: returns true if either dimension is 1 -- i.e. the matrix is shaped like a vector (row, column, or 1×1 scalar). Used by BFGS / SR1 / FullHessianUpdateStrategy as a shape precondition on their step / gradient inputs.

Returns:

true if either rows == 1 or cols == 1

isColumnVector

public boolean isColumnVector()

Returns:

true iff this matrix has exactly one column

normF

public double normF()

Returns:

the Frobenius norm; alias for norm2() kept for clarity

equalsContent

public boolean equalsContent(Matrix other)

Parameters:

other - matrix to compare against

Returns:

true iff other has the same shape and every entry compares bit-equal under ==

fill

public Matrix fill(double v)

In-place fill with the given scalar.

Parameters:

v - value to assign to every entry

Returns:

this after the mutation

doubleValue

public double doubleValue()

Extract the single scalar value of a 1 x 1 matrix.

Returns:

the only entry

Throws:

IllegalStateException - if the matrix is not 1 x 1

mapAdd

protected Matrix mapAdd(double scalar)

Helper used by plus(double) and minus(double): returns a fresh dense matrix with scalar added to every entry.

Parameters:

scalar - value to add elementwise

Returns:

a fresh DenseMatrix of the same shape