no.uib.cipr.matrix.sparse.LinkedSparseMatrix Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of mt-java Show documentation
Show all versions of mt-java Show documentation
Matrix data structures, linear solvers, least squares methods, eigenvalue,
and singular value decompositions. For larger random dense matrices (above ~ 350 x 350)
matrix-matrix multiplication C = A.B is about 50% faster than MTJ.
package no.uib.cipr.matrix.sparse;
import no.uib.cipr.matrix.AbstractMatrix;
import no.uib.cipr.matrix.Matrix;
import no.uib.cipr.matrix.MatrixEntry;
import no.uib.cipr.matrix.Vector;
import no.uib.cipr.matrix.io.MatrixInfo;
import no.uib.cipr.matrix.io.MatrixSize;
import no.uib.cipr.matrix.io.MatrixVectorReader;
import java.io.IOException;
import java.util.Iterator;
/**
* A Linked List (with shortcuts to important nodes) implementation of an
* {@code n x m} Matrix with {@code z} elements that has a typical
* {@code O(z / m)} insertion / lookup cost and an iterator that traverses
* columns then rows: a good fit for unstructured sparse matrices. A secondary
* link maintains fast transpose iteration.
*
* However, memory requirements (
* {@code 1 instance (8 bytes), 2 int (16 bytes), 2 ref (16 bytes), 1 double (8 bytes) = 48 bytes}
* per matrix element, plus {@code 8 x numcol + 8 x numrow bytes}s for the
* cache) are slightly higher than structured sparse matrix storage. Note that
* on 32 bit JVMs, or on 64 bit JVMs with CompressedOops enabled, references and ints only cost 4 bytes each,
* bringing the cost to 28 bytes per element.
*
* @author Sam Halliday
*/
public class LinkedSparseMatrix extends AbstractMatrix {
// java.util.LinkedList is doubly linked and therefore too heavy-weight.
static class Node {
final int row, col;
double val;
Node rowTail, colTail;
Node(int row, int col, double val, Node rowTail, Node colTail) {
this.row = row;
this.col = col;
this.val = val;
this.rowTail = rowTail;
this.colTail = colTail;
}
@Override
public String toString() {
return (new StringBuilder()).append("LinkedSparseMatrix.Node(row=")
.append(row).append(", col=").append(col).append(", val=")
.append(val).append(")").toString();
}
}
// There is a lot of duplicated code in this class between
// row and col linkages, but subtle differences make it
// extremely difficult to factor away.
class Linked {
final Node head = new Node(0, 0, 0.0, null, null);
Node[] rows = new Node[numRows], cols = new Node[numColumns];
private boolean isHead(int row, int col) {
return head.row == row && head.col == col;
}
// true if node exists, it's row tail exists, and has this row/col
private boolean isNextByRow(Node node, int row, int col) {
return node != null && node.rowTail != null
&& node.rowTail.row == row && node.rowTail.col == col;
}
// true if node exists, it's col tail exists, and has this row/col
private boolean isNextByCol(Node node, int row, int col) {
return node != null && node.colTail != null
&& node.colTail.col == col && node.colTail.row == row;
}
public double get(int row, int col) {
if (isHead(row, col))
return head.val;
if (col <= row) {
Node node = findPreceedingByRow(row, col);
if (isNextByRow(node, row, col))
return node.rowTail.val;
} else {
Node node = findPreceedingByCol(row, col);
if (isNextByCol(node, row, col))
return node.colTail.val;
}
return 0;
}
public void set(int row, int col, double val) {
if (val == 0) {
delete(row, col);
return;
}
if (isHead(row, col))
head.val = val;
else {
Node prevRow = findPreceedingByRow(row, col);
if (isNextByRow(prevRow, row, col))
prevRow.rowTail.val = val;
else {
Node prevCol = findPreceedingByCol(row, col);
Node nextCol = findNextByCol(row, col);
prevRow.rowTail = new Node(row, col, val, prevRow.rowTail,
nextCol);
prevCol.colTail = prevRow.rowTail;
updateCache(prevRow.rowTail);
}
}
}
private Node findNextByCol(int row, int col) {
Node cur = cachedByCol(col - 1);
while (cur != null) {
if (row < cur.row && col <= cur.col || col < cur.col)
return cur;
cur = cur.colTail;
}
return cur;
}
private void updateCache(Node inserted) {
if (rows[inserted.row] == null
|| inserted.col > rows[inserted.row].col)
rows[inserted.row] = inserted;
if (cols[inserted.col] == null
|| inserted.row > cols[inserted.col].row)
cols[inserted.col] = inserted;
}
private void delete(int row, int col) {
if (isHead(row, col)) {
head.val = 0;
return;
}
Node precRow = findPreceedingByRow(row, col);
Node precCol = findPreceedingByCol(row, col);
if (isNextByRow(precRow, row, col)) {
if (rows[row] == precRow.rowTail)
rows[row] = precRow.row == row ? precRow : null;
precRow.rowTail = precRow.rowTail.rowTail;
}
if (isNextByCol(precCol, row, col)) {
if (cols[col] == precCol.colTail)
cols[col] = precCol.col == col ? precCol : null;
precCol.colTail = precCol.colTail.colTail;
}
}
// returns the node that either references this
// index, or should reference it if inserted.
Node findPreceedingByRow(int row, int col) {
Node last = cachedByRow(row - 1);
Node cur = last;
while (cur != null && cur.row <= row) {
if (cur.row == row && cur.col >= col)
return last;
last = cur;
cur = cur.rowTail;
}
return last;
}
// helper for findPreceeding
private Node cachedByRow(int row) {
for (int i = row; i >= 0; i--)
if (rows[i] != null)
return rows[i];
return head;
}
Node findPreceedingByCol(int row, int col) {
Node last = cachedByCol(col - 1);
Node cur = last;
while (cur != null && cur.col <= col) {
if (cur.col == col && cur.row >= row)
return last;
last = cur;
cur = cur.colTail;
}
return last;
}
private Node cachedByCol(int col) {
for (int i = col; i >= 0; i--)
if (cols[i] != null)
return cols[i];
return head;
}
Node startOfRow(int row) {
if (row == 0)
return head;
Node prec = findPreceedingByRow(row, 0);
if (prec.rowTail != null && prec.rowTail.row == row)
return prec.rowTail;
return null;
}
Node startOfCol(int col) {
if (col == 0)
return head;
Node prec = findPreceedingByCol(0, col);
if (prec != null && prec.colTail != null && prec.colTail.col == col)
return prec.colTail;
return null;
}
}
Linked links;
public LinkedSparseMatrix(int numRows, int numColumns) {
super(numRows, numColumns);
links = new Linked();
}
public LinkedSparseMatrix(Matrix A) {
super(A);
links = new Linked();
set(A);
}
public LinkedSparseMatrix(MatrixVectorReader r) throws IOException {
super(0, 0);
try {
MatrixInfo info = r.readMatrixInfo();
if (info.isComplex())
throw new IllegalArgumentException(
"complex matrices not supported");
if (!info.isCoordinate())
throw new IllegalArgumentException(
"only coordinate matrices supported");
MatrixSize size = r.readMatrixSize(info);
numRows = size.numRows();
numColumns = size.numColumns();
links = new Linked();
int nz = size.numEntries();
int[] row = new int[nz];
int[] column = new int[nz];
double[] entry = new double[nz];
r.readCoordinate(row, column, entry);
r.add(-1, row);
r.add(-1, column);
for (int i = 0; i < nz; ++i)
set(row[i], column[i], entry[i]);
} finally {
r.close();
}
}
@Override
public Matrix zero() {
links = new Linked();
return this;
}
@Override
public double get(int row, int column) {
return links.get(row, column);
}
@Override
public void set(int row, int column, double value) {
check(row, column);
links.set(row, column, value);
}
// avoids object creation
static class ReusableMatrixEntry implements MatrixEntry {
int row, col;
double val;
@Override
public int column() {
return col;
}
@Override
public int row() {
return row;
}
@Override
public double get() {
return val;
}
@Override
public void set(double value) {
throw new UnsupportedOperationException();
}
@Override
public String toString() {
return row + "," + col + "=" + val;
}
}
@Override
public Iterator iterator() {
return new Iterator() {
Node cur = links.head;
ReusableMatrixEntry entry = new ReusableMatrixEntry();
@Override
public boolean hasNext() {
return cur != null;
}
@Override
public MatrixEntry next() {
entry.row = cur.row;
entry.col = cur.col;
entry.val = cur.val;
cur = cur.rowTail;
return entry;
}
@Override
public void remove() {
throw new UnsupportedOperationException();
}
};
}
@Override
public Matrix scale(double alpha) {
if (alpha == 0)
zero();
else if (alpha != 1)
for (MatrixEntry e : this)
set(e.row(), e.column(), e.get() * alpha);
return this;
}
@Override
public Matrix copy() {
return new LinkedSparseMatrix(this);
}
@Override
public Matrix transpose() {
Linked old = links;
numRows = numColumns;
numColumns = old.rows.length;
links = new Linked();
Node node = old.head;
while (node != null) {
set(node.col, node.row, node.val);
node = node.rowTail;
}
return this;
}
@Override
public Vector multAdd(double alpha, Vector x, Vector y) {
checkMultAdd(x, y);
if (alpha == 0)
return y;
Node node = links.head;
while (node != null) {
y.add(node.row, alpha * node.val * x.get(node.col));
node = node.rowTail;
}
return y;
}
@Override
public Vector transMultAdd(double alpha, Vector x, Vector y) {
checkTransMultAdd(x, y);
if (alpha == 0)
return y;
Node node = links.head;
while (node != null) {
y.add(node.col, alpha * node.val * x.get(node.row));
node = node.colTail;
}
return y;
}
// TODO: optimise matrix mults based on RHS Matrix
@Override
public Matrix multAdd(double alpha, Matrix B, Matrix C) {
checkMultAdd(B, C);
if (alpha == 0)
return C;
for (int i = 0; i < numRows; i++) {
Node row = links.startOfRow(i);
if (row != null)
for (int j = 0; j < B.numColumns(); j++) {
Node node = row;
double v = 0;
while (node != null && node.row == i) {
v += (B.get(node.col, j) * node.val);
node = node.rowTail;
}
if (v != 0)
C.add(i, j, alpha * v);
}
}
return C;
}
@Override
public Matrix transBmultAdd(double alpha, Matrix B, Matrix C) {
checkTransBmultAdd(B, C);
if (alpha == 0)
return C;
for (int i = 0; i < numRows; i++) {
Node row = links.startOfRow(i);
if (row != null)
for (int j = 0; j < B.numRows(); j++) {
Node node = row;
double v = 0;
while (node != null && node.row == i) {
v += (B.get(j, node.col) * node.val);
node = node.rowTail;
}
if (v != 0)
C.add(i, j, alpha * v);
}
}
return C;
}
@Override
public Matrix transAmultAdd(double alpha, Matrix B, Matrix C) {
checkTransAmultAdd(B, C);
if (alpha == 0)
return C;
for (int i = 0; i < numColumns; i++) {
Node row = links.startOfCol(i);
if (row != null)
for (int j = 0; j < B.numColumns(); j++) {
Node node = row;
double v = 0;
while (node != null && node.col == i) {
v += (B.get(node.row, j) * node.val);
node = node.colTail;
}
if (v != 0)
C.add(i, j, alpha * v);
}
}
return C;
}
@Override
public Matrix transABmultAdd(double alpha, Matrix B, Matrix C) {
checkTransABmultAdd(B, C);
if (alpha == 0)
return C;
for (int i = 0; i < numColumns; i++) {
Node row = links.startOfCol(i);
if (row != null)
for (int j = 0; j < B.numRows(); j++) {
Node node = row;
double v = 0;
while (node != null && node.col == i) {
v += (B.get(j, node.row) * node.val);
node = node.colTail;
}
if (v != 0)
C.add(i, j, alpha * v);
}
}
return C;
}
}
© 2015 - 2024 Weber Informatics LLC | Privacy Policy