Class ShrinkingMatrix

java.lang.Object

org.tribuo.math.la.DenseMatrix

org.tribuo.math.optimisers.util.ShrinkingMatrix

All Implemented Interfaces:

Serializable, Iterable<MatrixTuple>, Matrix, Tensor, ShrinkingTensor, ProtoSerializable<org.tribuo.math.protos.TensorProto>

public class ShrinkingMatrix
extends DenseMatrix
implements ShrinkingTensor

A subclass of DenseMatrix which shrinks the value every time a new value is added.

Be careful when modifying this or DenseMatrix.

See Also:

• Serialized Form

Nested Class Summary

Nested classes/interfaces inherited from class org.tribuo.math.la.DenseMatrix

DenseMatrix.CholeskyFactorization, DenseMatrix.EigenDecomposition, DenseMatrix.LUFactorization

Nested classes/interfaces inherited from interface org.tribuo.math.la.Matrix

Matrix.Factorization

Field Summary

Fields inherited from class org.tribuo.math.la.DenseMatrix

CURRENT_VERSION, dim1, dim2, FACTORIZATION_TOLERANCE, values

Fields inherited from interface org.tribuo.protos.ProtoSerializable

DESERIALIZATION_METHOD_NAME, PROVENANCE_SERIALIZER

Fields inherited from interface org.tribuo.math.optimisers.util.ShrinkingTensor

tolerance

Constructor Summary

Constructors

Constructor	Description
ShrinkingMatrix(DenseMatrix v, double baseRate, boolean scaleShrinking)	Constructs a shrinking matrix copy of the supplied dense matrix.
ShrinkingMatrix(DenseMatrix v, double baseRate, double lambda)	Constructs a shrinking matrix copy of the supplied dense matrix.

Method Summary

Modifier and Type	Method	Description
DenseMatrix	convertToDense()	Converts the tensor into a dense tensor.
static ShrinkingMatrix	deserializeFromProto(int version, String className, com.google.protobuf.Any message)	Deserialization factory.
double	get(int i, int j)	Gets an element from this Matrix.
void	intersectAndAddInPlace(Tensor other, DoubleUnaryOperator f)	Updates this Tensor by adding all the values from the intersection with other.
MatrixIterator	iterator()
DenseVector	leftMultiply(SGDVector input)	Multiplies this Matrix by a SGDVector returning a vector of the appropriate size.
void	scaleInPlace(double value)	Scales each element of this Tensor by coefficient.
org.tribuo.math.protos.TensorProto	serialize()	Serializes this object to a protobuf.
double	twoNorm()	Calculates the euclidean norm for this vector.

Methods inherited from class org.tribuo.math.la.DenseMatrix

add, add, addAcrossDim1, addAcrossDim2, broadcastIntersectAndAddInPlace, choleskyFactorization, columnSum, columnSum, copy, createDenseMatrix, createDenseMatrix, createIdentity, eigenDecomposition, equals, foreachInPlace, gatherAcrossDim1, gatherAcrossDim2, getColumn, getDimension1Size, getDimension2Size, getRow, getShape, hadamardProductInPlace, hashCode, isSquare, isSymmetric, luFactorization, matrixMultiply, matrixMultiply, normalizeRows, numActiveElements, reshape, rightMultiply, rowScaleInPlace, rowSum, rowSum, selectColumns, selectColumns, set, setColumn, toArray, toString, transpose, unpackProto

Methods inherited from class java.lang.Object

clone, finalize, getClass, notify, notifyAll, wait, wait, wait

Methods inherited from interface java.lang.Iterable

forEach, spliterator

Methods inherited from interface org.tribuo.math.la.Tensor

hadamardProductInPlace, intersectAndAddInPlace, scalarAddInPlace

Constructor Details

ShrinkingMatrix

public ShrinkingMatrix(DenseMatrix v,
 double baseRate,
 boolean scaleShrinking)

Constructs a shrinking matrix copy of the supplied dense matrix.

This matrix shrinks during each call to intersectAndAddInPlace(Tensor, DoubleUnaryOperator).

Parameters:

v - The matrix to copy.

baseRate - The base amount of shrinking to apply after each update.

scaleShrinking - If true reduce the shrinking value over time proportionally to the number of updates.

ShrinkingMatrix

public ShrinkingMatrix(DenseMatrix v,
 double baseRate,
 double lambda)

Constructs a shrinking matrix copy of the supplied dense matrix.

This matrix shrinks during each call to intersectAndAddInPlace(Tensor, DoubleUnaryOperator), and then reprojects the matrix so it has the same twoNorm.

Parameters:

v - The matrix to copy.

baseRate - The base rate of shrinkage.

lambda - The lambda value (see Pegasos).

Method Details

deserializeFromProto

public static ShrinkingMatrix deserializeFromProto(int version,
 String className,
 com.google.protobuf.Any message)
                                            throws com.google.protobuf.InvalidProtocolBufferException

Deserialization factory.

Parameters:

version - The serialized object version.

className - The class name.

message - The serialized data.

Returns:

The deserialized object.

Throws:

com.google.protobuf.InvalidProtocolBufferException - If the protobuf could not be parsed from the message.

serialize

public org.tribuo.math.protos.TensorProto serialize()

Description copied from interface: ProtoSerializable

Serializes this object to a protobuf.

Specified by:

serialize in interface ProtoSerializable<org.tribuo.math.protos.TensorProto>

Overrides:

serialize in class DenseMatrix

Returns:

The protobuf.

convertToDense

public DenseMatrix convertToDense()

Description copied from interface: ShrinkingTensor

Converts the tensor into a dense tensor.

Specified by:

convertToDense in interface ShrinkingTensor

Returns:

A dense tensor copy of this shrinking tensor.

leftMultiply

public DenseVector leftMultiply(SGDVector input)

Description copied from interface: Matrix

Multiplies this Matrix by a SGDVector returning a vector of the appropriate size.

The input must have dimension equal to Matrix.getDimension2Size().

Specified by:

leftMultiply in interface Matrix

Overrides:

leftMultiply in class DenseMatrix

Parameters:

input - The input vector.

Returns:

A new SGDVector of size Matrix.getDimension1Size().

intersectAndAddInPlace

public void intersectAndAddInPlace(Tensor other,
 DoubleUnaryOperator f)

Description copied from interface: Tensor

Updates this Tensor by adding all the values from the intersection with other.

The function f is applied to all values from other before the addition.

Each value is updated as value += f(otherValue).

Specified by:

intersectAndAddInPlace in interface Tensor

Overrides:

intersectAndAddInPlace in class DenseMatrix

Parameters:

other - The other Tensor.

f - A function to apply.

get

public double get(int i,
 int j)

Description copied from interface: Matrix

Gets an element from this Matrix.

Specified by:

get in interface Matrix

Overrides:

get in class DenseMatrix

Parameters:

i - The index for the first dimension.

j - The index for the second dimension.

Returns:

The value at matrix[i][j].

scaleInPlace

public void scaleInPlace(double value)

Description copied from interface: Tensor

Scales each element of this Tensor by coefficient.

Specified by:

scaleInPlace in interface Tensor

Parameters:

value - The coefficient of scaling.

twoNorm

public double twoNorm()

Description copied from interface: Tensor

Calculates the euclidean norm for this vector.

Specified by:

twoNorm in interface Tensor

Overrides:

twoNorm in class DenseMatrix

Returns:

The euclidean norm.

iterator

public MatrixIterator iterator()

Specified by:

iterator in interface Iterable<MatrixTuple>

Overrides:

iterator in class DenseMatrix