LUDecomposition finished. Not tested

This commit is contained in:
Alexander Nozik 2019-01-16 19:13:09 +03:00
parent 58e939e0cf
commit cedb8a816e
9 changed files with 121 additions and 82 deletions

View File

@ -1,5 +1,5 @@
buildscript {
extra["kotlinVersion"] = "1.3.20-eap-52"
extra["kotlinVersion"] = "1.3.20-eap-100"
extra["ioVersion"] = "0.1.2"
extra["coroutinesVersion"] = "1.1.0"

View File

@ -57,7 +57,7 @@ private class LUPDecomposition<T : Comparable<T>, R : Ring<T>>(
* U is an upper-triangular matrix
* @return the U matrix (or null if decomposed matrix is singular)
*/
override val u: Matrix<T> = VirtualMatrix(lu.shape[0], lu.shape[1]) { i, j ->
override val u: Matrix<T> = VirtualMatrix(lu.shape[0], lu.shape[1]) { i, j ->
if (j >= i) lu[i, j] else context.zero
}
@ -93,7 +93,11 @@ private class LUPDecomposition<T : Comparable<T>, R : Ring<T>>(
/**
* Implementation based on Apache common-maths LU-decomposition
*/
class LUPDecompositionBuilder<T : Comparable<T>, F : Field<T>>(val context: F, val bufferFactory: MutableBufferFactory<T> = ::boxing, val singularityCheck: (T) -> Boolean) {
class LUPDecompositionBuilder<T : Comparable<T>, F : Field<T>>(
val context: F,
val bufferFactory: MutableBufferFactory<T> = ::boxing,
val singularityCheck: (T) -> Boolean
) {
/**
* In-place transformation for [MutableNDStructure], using given transformation for each element
@ -105,23 +109,16 @@ class LUPDecompositionBuilder<T : Comparable<T>, F : Field<T>>(val context: F, v
private fun abs(value: T) = if (value > context.zero) value else with(context) { -value }
fun decompose(matrix: Matrix<T>): LUPDecompositionFeature<T> {
// Use existing decomposition if it is provided by matrix
matrix.features.find { it is LUPDecompositionFeature<*> }?.let {
@Suppress("UNCHECKED_CAST")
return it as LUPDecompositionFeature<T>
}
if (matrix.rowNum != matrix.colNum) {
error("LU decomposition supports only square matrices")
}
val m = matrix.colNum
val pivot = IntArray(matrix.rowNum)
//TODO replace by custom optimized 2d structure
val lu: MutableNDStructure<T> = mutableNdStructure(
intArrayOf(matrix.rowNum, matrix.colNum),
bufferFactory
) { index: IntArray -> matrix[index[0], index[1]] }
val lu = Mutable2DStructure.create(matrix.rowNum, matrix.colNum, bufferFactory) { i, j ->
matrix[i, j]
}
with(context) {
@ -188,57 +185,56 @@ class LUPDecompositionBuilder<T : Comparable<T>, F : Field<T>>(val context: F, v
}
class LUSolver<T : Comparable<T>, F : Field<T>>(val singularityCheck: (T) -> Boolean) : LinearSolver<T, F> {
class LUSolver<T : Comparable<T>, F : Field<T>>(
override val context: MatrixContext<T, F>,
val bufferFactory: MutableBufferFactory<T> = ::boxing,
val singularityCheck: (T) -> Boolean
) : LinearSolver<T, F> {
override fun solve(a: Matrix<T>, b: Matrix<T>): Matrix<T> {
val decomposition = LUPDecompositionBuilder(ring, singularityCheck).decompose(a)
if (b.rowNum != a.colNum) {
error("Matrix dimension mismatch expected ${a.rowNum}, but got ${b.colNum}")
}
// Use existing decomposition if it is provided by matrix
@Suppress("UNCHECKED_CAST")
val decomposition = a.features.find { it is LUPDecompositionFeature<*> }?.let {
it as LUPDecompositionFeature<T>
} ?: LUPDecompositionBuilder(context.elementContext, bufferFactory, singularityCheck).decompose(a)
// val bp = Array(a.rowNum) { Array<T>(b.colNum){ring.zero} }
// for (row in 0 until a.rowNum) {
// val bpRow = bp[row]
// val pRow = decomposition.pivot[row]
// for (col in 0 until b.colNum) {
// bpRow[col] = b[pRow, col]
// }
// }
// Apply permutations to b
val bp = produce(a.rowNum, a.colNum) { i, j -> b[decomposition.pivot[i], j] }
// Solve LY = b
for (col in 0 until a.rowNum) {
val bpCol = bp[col]
for (i in col + 1 until a.rowNum) {
val bpI = bp[i]
val luICol = decomposition.lu[i, col]
for (j in 0 until b.colNum) {
bpI[j] -= bpCol[j] * luICol
with(decomposition) {
with(context.elementContext) {
// Apply permutations to b
val bp = Mutable2DStructure.create(a.rowNum, a.colNum, bufferFactory) { i, j ->
b[pivot[i], j]
}
// Solve LY = b
for (col in 0 until a.rowNum) {
for (i in col + 1 until a.rowNum) {
for (j in 0 until b.colNum) {
bp[i, j] -= bp[col, j] * l[i, col]
}
}
}
// Solve UX = Y
for (col in a.rowNum - 1 downTo 0) {
for (i in 0 until col) {
for (j in 0 until b.colNum) {
bp[i, j] -= bp[col, j] / u[col, col] * u[i, col]
}
}
}
return context.produce(a.rowNum, a.colNum) { i, j -> bp[i, j] }
}
}
}
// Solve UX = Y
for (col in a.rowNum - 1 downTo 0) {
val bpCol = bp[col]
val luDiag = decomposition.lu[col, col]
for (j in 0 until b.colNum) {
bpCol[j] /= luDiag
}
for (i in 0 until col) {
val bpI = bp[i]
val luICol = decomposition.lu[i, col]
for (j in 0 until b.colNum) {
bpI[j] -= bpCol[j] * luICol
}
}
}
return produce(a.rowNum, a.colNum) { i, j -> bp[i][j] }
companion object {
val real = LUSolver(MatrixContext.real, MutableBuffer.Companion::auto) { it < 1e-11 }
}
}

View File

@ -4,26 +4,20 @@ import scientifik.kmath.operations.Field
import scientifik.kmath.operations.Norm
import scientifik.kmath.operations.RealField
import scientifik.kmath.operations.Ring
import scientifik.kmath.structures.VirtualBuffer
import scientifik.kmath.structures.asSequence
/**
* A group of methods to resolve equation A dot X = B, where A and B are matrices or vectors
*/
interface LinearSolver<T : Any, R : Ring<T>> : MatrixContext<T, R> {
/**
* Convert matrix to vector if it is possible
*/
fun Matrix<T>.toVector(): Point<T> =
if (this.colNum == 1) {
point(rowNum){ get(it, 0) }
} else error("Can't convert matrix with more than one column to vector")
interface LinearSolver<T : Any, R : Ring<T>> {
val context: MatrixContext<T,R>
fun Point<T>.toMatrix(): Matrix<T> = produce(size, 1) { i, _ -> get(i) }
fun solve(a: Matrix<T>, b: Matrix<T>): Matrix<T>
fun solve(a: Matrix<T>, b: Point<T>): Point<T> = solve(a, b.toMatrix()).toVector()
fun inverse(a: Matrix<T>): Matrix<T> = solve(a, one(a.rowNum, a.colNum))
fun inverse(a: Matrix<T>): Matrix<T> = solve(a, context.one(a.rowNum, a.colNum))
}
/**
@ -41,3 +35,15 @@ object VectorL2Norm : Norm<Point<out Number>, Double> {
typealias RealVector = Vector<Double, RealField>
typealias RealMatrix = Matrix<Double>
/**
* Convert matrix to vector if it is possible
*/
fun <T: Any> Matrix<T>.toVector(): Point<T> =
if (this.colNum == 1) {
VirtualBuffer(rowNum){ get(it, 0) }
} else error("Can't convert matrix with more than one column to vector")
fun <T: Any> Point<T>.toMatrix(): Matrix<T> = VirtualMatrix(size, 1) { i, _ -> get(i) }

View File

@ -11,7 +11,7 @@ interface MatrixContext<T : Any, R : Ring<T>> {
/**
* The ring context for matrix elements
*/
val ring: R
val elementContext: R
/**
* Produce a matrix with this context and given dimensions
@ -25,7 +25,7 @@ interface MatrixContext<T : Any, R : Ring<T>> {
fun scale(a: Matrix<T>, k: Number): Matrix<T> {
//TODO create a special wrapper class for scaled matrices
return produce(a.rowNum, a.colNum) { i, j -> ring.run { a[i, j] * k } }
return produce(a.rowNum, a.colNum) { i, j -> elementContext.run { a[i, j] * k } }
}
infix fun Matrix<T>.dot(other: Matrix<T>): Matrix<T> {
@ -34,7 +34,7 @@ interface MatrixContext<T : Any, R : Ring<T>> {
return produce(rowNum, other.colNum) { i, j ->
val row = rows[i]
val column = other.columns[j]
with(ring) {
with(elementContext) {
row.asSequence().zip(column.asSequence(), ::multiply).sum()
}
}
@ -45,27 +45,27 @@ interface MatrixContext<T : Any, R : Ring<T>> {
if (this.colNum != vector.size) error("Matrix dot vector operation dimension mismatch: ($rowNum, $colNum) x (${vector.size})")
return point(rowNum) { i ->
val row = rows[i]
with(ring) {
with(elementContext) {
row.asSequence().zip(vector.asSequence(), ::multiply).sum()
}
}
}
operator fun Matrix<T>.unaryMinus() =
produce(rowNum, colNum) { i, j -> ring.run { -get(i, j) } }
produce(rowNum, colNum) { i, j -> elementContext.run { -get(i, j) } }
operator fun Matrix<T>.plus(b: Matrix<T>): Matrix<T> {
if (rowNum != b.rowNum || colNum != b.colNum) error("Matrix operation dimension mismatch. [$rowNum,$colNum] + [${b.rowNum},${b.colNum}]")
return produce(rowNum, colNum) { i, j -> ring.run { get(i, j) + b[i, j] } }
return produce(rowNum, colNum) { i, j -> elementContext.run { get(i, j) + b[i, j] } }
}
operator fun Matrix<T>.minus(b: Matrix<T>): Matrix<T> {
if (rowNum != b.rowNum || colNum != b.colNum) error("Matrix operation dimension mismatch. [$rowNum,$colNum] - [${b.rowNum},${b.colNum}]")
return produce(rowNum, colNum) { i, j -> ring.run { get(i, j) + b[i, j] } }
return produce(rowNum, colNum) { i, j -> elementContext.run { get(i, j) + b[i, j] } }
}
operator fun Matrix<T>.times(number: Number): Matrix<T> =
produce(rowNum, colNum) { i, j -> ring.run { get(i, j) * number } }
produce(rowNum, colNum) { i, j -> elementContext.run { get(i, j) * number } }
operator fun Number.times(m: Matrix<T>): Matrix<T> = m * this
@ -157,7 +157,7 @@ fun <T : Any, R : Ring<T>> MatrixContext<T, R>.one(rows: Int, columns: Int): Mat
override val rowNum: Int get() = rows
override val colNum: Int get() = columns
override val features: Set<MatrixFeature> get() = setOf(DiagonalFeature, UnitFeature)
override fun get(i: Int, j: Int): T = if (i == j) ring.one else ring.zero
override fun get(i: Int, j: Int): T = if (i == j) elementContext.one else elementContext.zero
}
}
@ -169,7 +169,7 @@ fun <T : Any, R : Ring<T>> MatrixContext<T, R>.zero(rows: Int, columns: Int): Ma
override val rowNum: Int get() = rows
override val colNum: Int get() = columns
override val features: Set<MatrixFeature> get() = setOf(ZeroFeature)
override fun get(i: Int, j: Int): T = ring.zero
override fun get(i: Int, j: Int): T = elementContext.zero
}
}

View File

@ -0,0 +1,40 @@
package scientifik.kmath.linear
import scientifik.kmath.structures.MutableBuffer
import scientifik.kmath.structures.MutableBufferFactory
import scientifik.kmath.structures.MutableNDStructure
class Mutable2DStructure<T>(val rowNum: Int, val colNum: Int, val buffer: MutableBuffer<T>) : MutableNDStructure<T> {
override val shape: IntArray
get() = intArrayOf(rowNum, colNum)
operator fun get(i: Int, j: Int): T = buffer[i * colNum + j]
override fun get(index: IntArray): T = get(index[0], index[1])
override fun elements(): Sequence<Pair<IntArray, T>> = sequence {
for (i in 0 until rowNum) {
for (j in 0 until colNum) {
yield(intArrayOf(i, j) to get(i, j))
}
}
}
operator fun set(i: Int, j: Int, value: T) {
buffer[i * colNum + j] = value
}
override fun set(index: IntArray, value: T) = set(index[0], index[1], value)
companion object {
fun <T> create(
rowNum: Int,
colNum: Int,
bufferFactory: MutableBufferFactory<T>,
init: (i: Int, j: Int) -> T
): Mutable2DStructure<T> {
val buffer = bufferFactory(rowNum * colNum) { offset -> init(offset / colNum, offset % colNum) }
return Mutable2DStructure(rowNum, colNum, buffer)
}
}
}

View File

@ -10,7 +10,7 @@ import scientifik.kmath.structures.ndStructure
* Basic implementation of Matrix space based on [NDStructure]
*/
class StructureMatrixContext<T : Any, R : Ring<T>>(
override val ring: R,
override val elementContext: R,
private val bufferFactory: BufferFactory<T>
) : MatrixContext<T, R> {

View File

@ -124,8 +124,6 @@ inline class MutableListBuffer<T>(private val list: MutableList<T>) : MutableBuf
override fun copy(): MutableBuffer<T> = MutableListBuffer(ArrayList(list))
}
fun <T> MutableList<T>.asBuffer() = MutableListBuffer(this)
class ArrayBuffer<T>(private val array: Array<T>) : MutableBuffer<T> {
//Can't inline because array is invariant
override val size: Int

View File

@ -22,9 +22,8 @@ class MatrixTest {
@Test
fun testTranspose() {
val matrix = MatrixContext.real(3, 3).one
val matrix = MatrixContext.real.one(3, 3)
val transposed = matrix.transpose()
assertEquals(matrix.context, transposed.context)
assertEquals((matrix as StructureMatrix).structure, (transposed as StructureMatrix).structure)
assertEquals(matrix, transposed)
}
@ -37,7 +36,7 @@ class MatrixTest {
val matrix1 = vector1.toMatrix()
val matrix2 = vector2.toMatrix().transpose()
val product = matrix1 dot matrix2
val product = MatrixContext.real.run { matrix1 dot matrix2 }
assertEquals(5.0, product[1, 0])

View File

@ -6,8 +6,8 @@ import kotlin.test.assertEquals
class RealLUSolverTest {
@Test
fun testInvertOne() {
val matrix = MatrixContext.real(2, 2).one
val inverted = RealLUSolver.inverse(matrix)
val matrix = MatrixContext.real.one(2, 2)
val inverted = LUSolver.real.inverse(matrix)
assertEquals(matrix, inverted)
}