Class NativeND4JLinearAlgebraBackend

java.lang.Object

org.episteme.nativ.mathematics.linearalgebra.backends.NativeND4JLinearAlgebraBackend

All Implemented Interfaces:

AutoCloseable, LinearAlgebraProvider<Real>, AlgorithmProvider, Backend, ComputeBackend, CPUBackend, NativeBackend

@AutoService({Backend.class,ComputeBackend.class,NativeBackend.class,LinearAlgebraProvider.class,CPUBackend.class})
public class NativeND4JLinearAlgebraBackend
extends Object
implements LinearAlgebraProvider<Real>, NativeBackend, CPUBackend

NativeND4J Linear Algebra Backend (Dense).

When the ND4J library (org.nd4j:nd4j-native-platform) is on the classpath, this backend delegates to ND4J's optimized BLAS/LAPACK backends (Native/AVX/CUDA). Decompositions (eigen, SVD, LU) are implemented natively using ND4J array operations.

Since:

1.0

Author:

Silvere Martin-Michiellot

Constructor Summary

Constructors

Constructor	Description
NativeND4JLinearAlgebraBackend()

Method Summary

Modifier and Type	Method	Description
Matrix<Real>	add(Matrix<Real> a, Matrix<Real> b)
Vector<Real>	add(Vector<Real> a, Vector<Real> b)
CholeskyResult<Real>	cholesky(Matrix<Real> a)	Computes the Cholesky decomposition of the specified matrix.
ExecutionContext	createContext()	Creates an execution context for running operations.
Real	determinant(Matrix<Real> a)	Determinant via Gaussian elimination on ND4J arrays.
Real	dot(Vector<Real> a, Vector<Real> b)
EigenResult<Real>	eigen(Matrix<Real> a)	Computes the eigenvalue decomposition of the specified matrix.
HardwareAccelerator	getAcceleratorType()	Returns the type of hardware accelerator used by this backend.
String	getAlgorithmType()	Returns the unique category of the algorithm.
String	getEnvironmentInfo()	Returns a string describing the execution environment (e.g., "CPU (AVX2)", "GPU (CUDA 12.0)").
String	getId()	Returns the unique identifier for this backend (e.g., "javafx", "jmol", "cuda").
String	getName()	Returns the display name for UI presentation.
String	getNativeLibraryName()	Returns the name of the native library this backend depends on.
int	getPriority()	Priority of this provider (higher means more preferred).
String	getType()	Returns the backend type category (e.g., "plotting", "molecular", "tensor", "math").
Matrix<Real>	inverse(Matrix<Real> a)
boolean	isAvailable()	Checks if the provider is available in the current environment.
boolean	isCompatible(Ring<?> ring)	Checks if this provider is compatible with the given ring.
boolean	isExplicitlyDisabled()	Checks if this backend is specifically disabled via a system property: "episteme.backend.disable." + getId().
boolean	isLoaded()	Checks if the native library is loaded and available.
LUResult<Real>	lu(Matrix<Real> a)	LU decomposition via Gaussian elimination on ND4J arrays.
Matrix<Real>	multiply(Matrix<Real> a, Matrix<Real> b)
Vector<Real>	multiply(Matrix<Real> a, Vector<Real> b)
Vector<Real>	multiply(Vector<Real> vector, Real scalar)
Real	norm(Vector<Real> a)
QRResult<Real>	qr(Matrix<Real> a)	QR decomposition via ND4J's 'qr' custom op.
Matrix<Real>	scale(Real scalar, Matrix<Real> a)
double	score(OperationContext context)	Scores this provider for a specific operation context.
void	shutdown()	Called when the provider is no longer needed (e.g., application shutdown).
Vector<Real>	solve(Matrix<Real> a, Vector<Real> b)	Solve Ax=b via ND4J's LU decomposition.
Vector<Real>	solveTriangular(Matrix<Real> A, Vector<Real> b, boolean upper, boolean transpose, boolean conjugate, boolean unit)	Solves the triangular system Ax = b.
Matrix<Real>	subtract(Matrix<Real> a, Matrix<Real> b)
Vector<Real>	subtract(Vector<Real> a, Vector<Real> b)
SVDResult<Real>	svd(Matrix<Real> a)	SVD using ND4J's built-in Svd custom op.
Real	trace(Matrix<Real> a)
Matrix<Real>	transpose(Matrix<Real> a)

Methods inherited from class Object

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

Methods inherited from interface AlgorithmProvider

description, getMetadata

Methods inherited from interface Backend

getAlgorithmProviders, getStatusMessage

Methods inherited from interface ComputeBackend

createBackend, getDescription

Methods inherited from interface CPUBackend

supportsComplexNumbers, supportsFloatingPoint, supportsParallelOps

Methods inherited from interface LinearAlgebraProvider

acos, acosh, angle, asin, asinh, atan, atanh, cbrt, close, conditionNumber, configure, conjugateTranspose, cos, cosh, cross, exp, log, log10, normalize, pow, projection, rank, sin, sinh, solve, solve, solve, sqrt, tan, tanh

Constructor Details

NativeND4JLinearAlgebraBackend

public NativeND4JLinearAlgebraBackend()

Method Details

getPriority

public int getPriority()

Description copied from interface: LinearAlgebraProvider

Priority of this provider (higher means more preferred). Used for automatic backend selection.

Specified by:

getPriority in interface AlgorithmProvider

Specified by:

getPriority in interface Backend

Specified by:

getPriority in interface LinearAlgebraProvider<Real>

getName

public String getName()

Description copied from interface: Backend

Returns the display name for UI presentation.

Specified by:

getName in interface AlgorithmProvider

Specified by:

getName in interface Backend

Specified by:

getName in interface LinearAlgebraProvider<Real>

getAcceleratorType

public HardwareAccelerator getAcceleratorType()

Description copied from interface: ComputeBackend

Returns the type of hardware accelerator used by this backend.

Specified by:

getAcceleratorType in interface ComputeBackend

Specified by:

getAcceleratorType in interface CPUBackend

Returns:

the accelerator type

getEnvironmentInfo

public String getEnvironmentInfo()

Description copied from interface: LinearAlgebraProvider

Returns a string describing the execution environment (e.g., "CPU (AVX2)", "GPU (CUDA 12.0)").

Specified by:

getEnvironmentInfo in interface LinearAlgebraProvider<Real>

isLoaded

public boolean isLoaded()

Description copied from interface: NativeBackend

Checks if the native library is loaded and available.

Specified by:

isLoaded in interface NativeBackend

Returns:

true if the native library has been successfully loaded

createContext

public ExecutionContext createContext()

Description copied from interface: ComputeBackend

Creates an execution context for running operations.

The context should be used within a try-with-resources block to ensure proper cleanup of resources.

Specified by:

createContext in interface ComputeBackend

Returns:

new execution context

getNativeLibraryName

public String getNativeLibraryName()

Description copied from interface: NativeBackend

Returns the name of the native library this backend depends on.

Specified by:

getNativeLibraryName in interface NativeBackend

Returns:

library name (e.g., "openblas", "cuda", "fftw3")

getAlgorithmType

public String getAlgorithmType()

Description copied from interface: AlgorithmProvider

Returns the unique category of the algorithm.

Specified by:

getAlgorithmType in interface AlgorithmProvider

Specified by:

getAlgorithmType in interface LinearAlgebraProvider<Real>

isAvailable

public boolean isAvailable()

Description copied from interface: AlgorithmProvider

Checks if the provider is available in the current environment.

Specified by:

isAvailable in interface AlgorithmProvider

Specified by:

isAvailable in interface Backend

getId

public String getId()

Description copied from interface: Backend

Returns the unique identifier for this backend (e.g., "javafx", "jmol", "cuda").

Specified by:

getId in interface Backend

Specified by:

getId in interface ComputeBackend

isExplicitlyDisabled

public boolean isExplicitlyDisabled()

Description copied from interface: Backend

Checks if this backend is specifically disabled via a system property: "episteme.backend.disable." + getId().

It also checks specialized properties like "episteme.linearalgebra.disable." + getId().

Specified by:

isExplicitlyDisabled in interface Backend

Returns:

true if explicitly disabled

getType

public String getType()

Description copied from interface: Backend

Returns the backend type category (e.g., "plotting", "molecular", "tensor", "math").

Specified by:

getType in interface Backend

Specified by:

getType in interface ComputeBackend

Specified by:

getType in interface CPUBackend

shutdown

public void shutdown()

Description copied from interface: AlgorithmProvider

Called when the provider is no longer needed (e.g., application shutdown). Use this to release local resources or close native segments.

Specified by:

shutdown in interface AlgorithmProvider

Specified by:

shutdown in interface Backend

isCompatible

public boolean isCompatible(Ring<?> ring)

Description copied from interface: LinearAlgebraProvider

Checks if this provider is compatible with the given ring.

Specified by:

isCompatible in interface LinearAlgebraProvider<Real>

score

public double score(OperationContext context)

Description copied from interface: AlgorithmProvider

Scores this provider for a specific operation context.

Higher scores indicate better suitability. Used by ProviderSelector for context-aware selection.

Default implementation returns AlgorithmProvider.getPriority(), so existing providers behave identically without changes.

Specified by:

score in interface AlgorithmProvider

Specified by:

score in interface LinearAlgebraProvider<Real>

Parameters:

context - the operation context (data size, hints, etc.)

Returns:

suitability score (higher = better)

add

public Vector<Real> add(Vector<Real> a,
 Vector<Real> b)

Specified by:

add in interface LinearAlgebraProvider<Real>

subtract

public Vector<Real> subtract(Vector<Real> a,
 Vector<Real> b)

Specified by:

subtract in interface LinearAlgebraProvider<Real>

multiply

public Vector<Real> multiply(Vector<Real> vector,
 Real scalar)

Specified by:

multiply in interface LinearAlgebraProvider<Real>

dot

public Real dot(Vector<Real> a,
 Vector<Real> b)

Specified by:

dot in interface LinearAlgebraProvider<Real>

norm

public Real norm(Vector<Real> a)

Specified by:

norm in interface LinearAlgebraProvider<Real>

add

public Matrix<Real> add(Matrix<Real> a,
 Matrix<Real> b)

Specified by:

add in interface LinearAlgebraProvider<Real>

subtract

public Matrix<Real> subtract(Matrix<Real> a,
 Matrix<Real> b)

Specified by:

subtract in interface LinearAlgebraProvider<Real>

multiply

public Matrix<Real> multiply(Matrix<Real> a,
 Matrix<Real> b)

Specified by:

multiply in interface LinearAlgebraProvider<Real>

solveTriangular

public Vector<Real> solveTriangular(Matrix<Real> A,
 Vector<Real> b,
 boolean upper,
 boolean transpose,
 boolean conjugate,
 boolean unit)

Description copied from interface: LinearAlgebraProvider

Solves the triangular system Ax = b.

Specified by:

solveTriangular in interface LinearAlgebraProvider<Real>

Parameters:

A - the triangular matrix

b - the right-hand side vector

upper - true if A is upper triangular, false if lower

transpose - true if solving A^T x = b

conjugate - true if solving A^H x = b (only if transpose is true and complex)

unit - true if A is unit triangular (diagonal is all ones)

multiply

public Vector<Real> multiply(Matrix<Real> a,
 Vector<Real> b)

Specified by:

multiply in interface LinearAlgebraProvider<Real>

inverse

public Matrix<Real> inverse(Matrix<Real> a)

Specified by:

inverse in interface LinearAlgebraProvider<Real>

determinant

public Real determinant(Matrix<Real> a)

Determinant via Gaussian elimination on ND4J arrays. O(n^3), no external lib.

Specified by:

determinant in interface LinearAlgebraProvider<Real>

solve

public Vector<Real> solve(Matrix<Real> a,
 Vector<Real> b)

Solve Ax=b via ND4J's LU decomposition.

Specified by:

solve in interface LinearAlgebraProvider<Real>

transpose

public Matrix<Real> transpose(Matrix<Real> a)

Specified by:

transpose in interface LinearAlgebraProvider<Real>

qr

public QRResult<Real> qr(Matrix<Real> a)

QR decomposition via ND4J's 'qr' custom op.

Specified by:

qr in interface LinearAlgebraProvider<Real>

scale

public Matrix<Real> scale(Real scalar,
 Matrix<Real> a)

Specified by:

scale in interface LinearAlgebraProvider<Real>

svd

public SVDResult<Real> svd(Matrix<Real> a)

SVD using ND4J's built-in Svd custom op. Returns U, S-diagonal-vector, V.

Specified by:

svd in interface LinearAlgebraProvider<Real>

lu

public LUResult<Real> lu(Matrix<Real> a)

LU decomposition via Gaussian elimination on ND4J arrays. Returns L (lower triangular), U (upper triangular), and pivot vector P.

Specified by:

lu in interface LinearAlgebraProvider<Real>

cholesky

public CholeskyResult<Real> cholesky(Matrix<Real> a)

Description copied from interface: LinearAlgebraProvider

Computes the Cholesky decomposition of the specified matrix.

Specified by:

cholesky in interface LinearAlgebraProvider<Real>

eigen

public EigenResult<Real> eigen(Matrix<Real> a)

Description copied from interface: LinearAlgebraProvider

Computes the eigenvalue decomposition of the specified matrix.

Specified by:

eigen in interface LinearAlgebraProvider<Real>

trace

public Real trace(Matrix<Real> a)

Specified by:

trace in interface LinearAlgebraProvider<Real>