Class DenseMatrix
- All Implemented Interfaces:
Serializable,Iterable<MatrixTuple>,Matrix,Tensor,ProtoSerializable<org.tribuo.math.protos.TensorProto>
- Direct Known Subclasses:
ShrinkingMatrix
-
Nested Class Summary
Nested ClassesModifier and TypeClassDescriptionstatic final classThe output of a successful Cholesky factorization.static final classThe output of a successful eigen decomposition.static final classThe output of a successful LU factorization.Nested classes/interfaces inherited from interface org.tribuo.math.la.Matrix
Matrix.Factorization -
Field Summary
FieldsModifier and TypeFieldDescriptionstatic final intProtobuf serialization version.protected final intThe number of rows.protected final intThe number of columns.static final doubleTolerance for non-zero diagonal values in the factorizations.protected final double[][]The value array.Fields inherited from interface org.tribuo.protos.ProtoSerializable
DESERIALIZATION_METHOD_NAME, PROVENANCE_SERIALIZER -
Constructor Summary
ConstructorsConstructorDescriptionDenseMatrix(int dim1, int dim2) Creates a dense matrix full of zeros.DenseMatrix(DenseMatrix other) Copies the supplied matrix.DenseMatrix(Matrix other) Copies the supplied matrix, densifying it if it's sparse. -
Method Summary
Modifier and TypeMethodDescriptionvoidadd(int i, int j, double value) Adds the argument value to the value at the supplied index.add(DenseMatrix other) Addsotherto this matrix, producing a newDenseMatrix.voidaddAcrossDim1(int[] indices, double value) Adds the specified value to the specified elements across dimension 1.voidaddAcrossDim2(int[] indices, double value) Adds the specified value to the specified elements across dimension 2.voidbroadcastIntersectAndAddInPlace(SGDVector input, boolean broadcastOverDim1) Broadcasts the input vector and adds it to each row/column of the matrix.Computes the Cholesky factorization of a positive definite matrix.Returns a dense vector containing each column sum.doublecolumnSum(int columnIndex) Calculates the sum of the specified column.copy()Copies the matrix.static DenseMatrixcreateDenseMatrix(double[][] values) Defensively copies the values before construction.static DenseMatrixcreateDenseMatrix(SGDVector[] vectors) Constructs a new DenseMatrix copying the values from the supplied vectors.static DenseMatrixcreateIdentity(int dimension) Creates an identity matrix of the specified size.static DenseMatrixdeserializeFromProto(int version, String className, com.google.protobuf.Any message) Deserialization factory.Eigen decomposition of a symmetric matrix.booleanvoidApplies aDoubleUnaryOperatorelementwise to thisTensor.gatherAcrossDim1(int[] elements) Constructs a dense vector by gathering values across dimension 1.gatherAcrossDim2(int[] elements) Constructs a dense vector by gathering values across dimension 2.doubleget(int i, int j) Gets an element from thisMatrix.getColumn(int index) Returns a copy of the specified column.intThe size of the first dimension.intThe size of the second dimension.getRow(int i) AnSGDVectorview of the row.int[]getShape()Returns an int array specifying the shape of thisTensor.voidhadamardProductInPlace(Tensor other, DoubleUnaryOperator f) Updates thisTensorwith the Hadamard product (i.e., a term by term multiply) of this andother.inthashCode()voidintersectAndAddInPlace(Tensor other, DoubleUnaryOperator f) Updates thisTensorby adding all the values from the intersection withother.booleanisSquare()Is this a square matrix?booleanReturns true if this matrix is square and symmetric.iterator()leftMultiply(SGDVector input) Multiplies this Matrix by aSGDVectorreturning a vector of the appropriate size.Computes the LU factorization of a square matrix.matrixMultiply(Matrix other) Multiplies this Matrix by anotherMatrixreturning a matrix of the appropriate size.matrixMultiply(Matrix other, boolean transposeThis, boolean transposeOther) Multiplies this Matrix by anotherMatrixreturning a matrix of the appropriate size.voidnormalizeRows(VectorNormalizer normalizer) Normalizes each row using the supplied normalizer in place.intnumActiveElements(int row) The number of non-zero elements in that row.reshape(int[] newShape) Reshapes the Tensor to the supplied shape.rightMultiply(SGDVector input) Multiplies this Matrix by aSGDVectorreturning a vector of the appropriate size.voidrowScaleInPlace(DenseVector scalingCoefficients) Scales each row by the appropriate value in theDenseVector.rowSum()Generates aDenseVectorrepresenting the sum of each row.doublerowSum(int rowIndex) Calculates the sum of the specified row.selectColumns(int[] columnIndices) Returns a new DenseMatrix containing a copy of the selected columns.selectColumns(List<Integer> columnIndices) Returns a new DenseMatrix containing a copy of the selected columns.org.tribuo.math.protos.TensorProtoSerializes this object to a protobuf.voidset(int i, int j, double value) Sets an element at the supplied location.voidSets the column to the supplied vector value.double[][]toArray()Returns a copy of this matrix as a 2d array.toString()Returns a transposed copy of this matrix.doubletwoNorm()Calculates the euclidean norm for this vector.protected static DenseMatrixunpackProto(org.tribuo.math.protos.DenseTensorProto proto) Unpacks aDenseTensorProtointo aDenseMatrix.Methods inherited from class java.lang.Object
clone, finalize, getClass, notify, notifyAll, wait, wait, waitMethods inherited from interface java.lang.Iterable
forEach, spliteratorMethods inherited from interface org.tribuo.math.la.Tensor
hadamardProductInPlace, intersectAndAddInPlace, scalarAddInPlace, scaleInPlace
-
Field Details
-
CURRENT_VERSION
public static final int CURRENT_VERSIONProtobuf serialization version.- See Also:
-
FACTORIZATION_TOLERANCE
public static final double FACTORIZATION_TOLERANCETolerance for non-zero diagonal values in the factorizations.- See Also:
-
values
protected final double[][] valuesThe value array. -
dim1
protected final int dim1The number of rows. -
dim2
protected final int dim2The number of columns.
-
-
Constructor Details
-
DenseMatrix
public DenseMatrix(int dim1, int dim2) Creates a dense matrix full of zeros.- Parameters:
dim1- The first dimension.dim2- The second dimension.
-
DenseMatrix
Copies the supplied matrix.- Parameters:
other- The matrix to copy.
-
DenseMatrix
Copies the supplied matrix, densifying it if it's sparse.- Parameters:
other- The matrix to copy.
-
-
Method Details
-
createDenseMatrix
Defensively copies the values before construction.Throws IllegalArgumentException if the supplied values are a ragged array.
- Parameters:
values- The values of this dense matrix.- Returns:
- A new dense matrix.
-
createDenseMatrix
Constructs a new DenseMatrix copying the values from the supplied vectors.Throws
IllegalArgumentExceptionif the supplied vectors are ragged (i.e., are not all the same size).- Parameters:
vectors- The vectors to coalesce.- Returns:
- A new dense matrix.
-
createIdentity
Creates an identity matrix of the specified size.- Parameters:
dimension- The matrix dimension.- Returns:
- The identity matrix.
-
deserializeFromProto
public static DenseMatrix 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 themessage.
-
unpackProto
Unpacks aDenseTensorProtointo aDenseMatrix.- Parameters:
proto- The proto to unpack.- Returns:
- The dense matrix.
-
serialize
public org.tribuo.math.protos.TensorProto serialize()Description copied from interface:ProtoSerializableSerializes this object to a protobuf.- Specified by:
serializein interfaceProtoSerializable<org.tribuo.math.protos.TensorProto>- Returns:
- The protobuf.
-
getShape
-
reshape
Description copied from interface:TensorReshapes the Tensor to the supplied shape. ThrowsIllegalArgumentExceptionif the shape isn't compatible. -
copy
Copies the matrix. -
get
-
add
Addsotherto this matrix, producing a newDenseMatrix.- Parameters:
other- The matrix to add.- Returns:
- A new
DenseMatrixwhere each element value = this.get(i,j) + other.get(i,j).
-
gatherAcrossDim1
Constructs a dense vector by gathering values across dimension 1.- Parameters:
elements- The indices to gather.- Returns:
- A dense vector.
-
gatherAcrossDim2
Constructs a dense vector by gathering values across dimension 2.- Parameters:
elements- The indices to gather.- Returns:
- A dense vector.
-
transpose
Returns a transposed copy of this matrix.- Returns:
- A transposed copy.
-
equals
-
hashCode
-
set
public void set(int i, int j, double value) Description copied from interface:MatrixSets an element at the supplied location. -
getDimension1Size
public int getDimension1Size()Description copied from interface:MatrixThe size of the first dimension.- Specified by:
getDimension1Sizein interfaceMatrix- Returns:
- The size of the first dimension.
-
getDimension2Size
public int getDimension2Size()Description copied from interface:MatrixThe size of the second dimension.- Specified by:
getDimension2Sizein interfaceMatrix- Returns:
- The size of the second dimension.
-
leftMultiply
Description copied from interface:MatrixMultiplies this Matrix by aSGDVectorreturning a vector of the appropriate size.The input must have dimension equal to
Matrix.getDimension2Size().- Specified by:
leftMultiplyin interfaceMatrix- Parameters:
input- The input vector.- Returns:
- A new
SGDVectorof sizeMatrix.getDimension1Size().
-
rightMultiply
Description copied from interface:MatrixMultiplies this Matrix by aSGDVectorreturning a vector of the appropriate size.The input must have dimension equal to
Matrix.getDimension1Size().- Specified by:
rightMultiplyin interfaceMatrix- Parameters:
input- The input vector.- Returns:
- A new
SGDVectorof sizeMatrix.getDimension2Size().
-
matrixMultiply
Description copied from interface:MatrixMultiplies this Matrix by anotherMatrixreturning a matrix of the appropriate size.The input must have dimension 1 equal to
Matrix.getDimension2Size().- Specified by:
matrixMultiplyin interfaceMatrix- Parameters:
other- The input matrix.- Returns:
- A new
Matrixof sizeMatrix.getDimension1Size(),input.getDimension2Size().
-
matrixMultiply
Description copied from interface:MatrixMultiplies this Matrix by anotherMatrixreturning a matrix of the appropriate size.Must obey the rules of matrix multiplication after the transposes are applied.
- Specified by:
matrixMultiplyin interfaceMatrix- Parameters:
other- The input matrix.transposeThis- Implicitly transposes this matrix just for the multiplication.transposeOther- Implicitly transposes other just for the multiplication.- Returns:
- A new
Matrix.
-
rowSum
Description copied from interface:MatrixGenerates aDenseVectorrepresenting the sum of each row.- Specified by:
rowSumin interfaceMatrix- Returns:
- A new
DenseVectorof sizeMatrix.getDimension1Size().
-
rowScaleInPlace
Description copied from interface:MatrixScales each row by the appropriate value in theDenseVector.- Specified by:
rowScaleInPlacein interfaceMatrix- Parameters:
scalingCoefficients- ADenseVectorwith sizeMatrix.getDimension1Size().
-
add
public void add(int i, int j, double value) Description copied from interface:MatrixAdds the argument value to the value at the supplied index. -
addAcrossDim1
public void addAcrossDim1(int[] indices, double value) Adds the specified value to the specified elements across dimension 1.- Parameters:
indices- The indices to update.value- The value to add.
-
addAcrossDim2
public void addAcrossDim2(int[] indices, double value) Adds the specified value to the specified elements across dimension 2.- Parameters:
indices- The indices to update.value- The value to add.
-
intersectAndAddInPlace
Description copied from interface:TensorUpdates thisTensorby adding all the values from the intersection withother.The function
fis applied to all values fromotherbefore the addition.Each value is updated as value += f(otherValue).
- Specified by:
intersectAndAddInPlacein interfaceTensor- Parameters:
other- The otherTensor.f- A function to apply.
-
hadamardProductInPlace
Description copied from interface:TensorUpdates thisTensorwith the Hadamard product (i.e., a term by term multiply) of this andother.The function
fis applied to all values fromotherbefore the addition.Each value is updated as value *= f(otherValue).
- Specified by:
hadamardProductInPlacein interfaceTensor- Parameters:
other- The otherTensor.f- A function to apply.
-
foreachInPlace
Description copied from interface:TensorApplies aDoubleUnaryOperatorelementwise to thisTensor.- Specified by:
foreachInPlacein interfaceTensor- Parameters:
f- The function to apply.
-
broadcastIntersectAndAddInPlace
Broadcasts the input vector and adds it to each row/column of the matrix.- Parameters:
input- The input vector.broadcastOverDim1- If true broadcasts over the first dimension, else broadcasts over the second.
-
numActiveElements
public int numActiveElements(int row) Description copied from interface:MatrixThe number of non-zero elements in that row.An element could be active and zero, if it was active on construction.
- Specified by:
numActiveElementsin interfaceMatrix- Parameters:
row- The index of the row.- Returns:
- The number of non-zero elements.
-
getRow
-
getColumn
Returns a copy of the specified column. -
setColumn
Sets the column to the supplied vector value.- Parameters:
index- The column to set.vector- The vector to write.
-
rowSum
public double rowSum(int rowIndex) Calculates the sum of the specified row.- Parameters:
rowIndex- The index of the row to sum.- Returns:
- The row sum.
-
columnSum
public double columnSum(int columnIndex) Calculates the sum of the specified column.- Parameters:
columnIndex- The index of the column to sum.- Returns:
- The column sum.
-
twoNorm
-
toArray
public double[][] toArray()Returns a copy of this matrix as a 2d array.- Returns:
- A copy of this matrix.
-
isSquare
public boolean isSquare()Is this a square matrix?- Returns:
- True if the matrix is square.
-
isSymmetric
public boolean isSymmetric()Returns true if this matrix is square and symmetric.- Returns:
- True if the matrix is symmetric.
-
choleskyFactorization
Computes the Cholesky factorization of a positive definite matrix.If the matrix is not symmetric or positive definite then it returns an empty optional.
- Returns:
- The Cholesky factorization or an empty optional.
-
luFactorization
Computes the LU factorization of a square matrix.If the matrix is singular or not square it returns an empty optional.
- Returns:
- The LU factorization or an empty optional.
-
eigenDecomposition
Eigen decomposition of a symmetric matrix.Non-symmetric matrices return an empty Optional as they may have complex eigenvalues, and any matrix which exceeds the default number of QL iterations in the decomposition also returns an empty Optional.
- Returns:
- The eigen decomposition of a symmetric matrix, or an empty optional if it's not symmetric.
-
toString
-
iterator
- Specified by:
iteratorin interfaceIterable<MatrixTuple>
-
normalizeRows
Normalizes each row using the supplied normalizer in place.- Parameters:
normalizer- The vector normalizer to use.
-
columnSum
Returns a dense vector containing each column sum.- Returns:
- The column sums.
-
selectColumns
Returns a new DenseMatrix containing a copy of the selected columns.Throws
IllegalArgumentExceptionif any column index is invalid or the array is null/empty.- Parameters:
columnIndices- The column indices- Returns:
- The submatrix comprising the selected columns.
-
selectColumns
Returns a new DenseMatrix containing a copy of the selected columns.Throws
IllegalArgumentExceptionif any column index is invalid or the list is null/empty.- Parameters:
columnIndices- The column indices- Returns:
- The submatrix comprising the selected columns.
-