forked from kscience/kmath
NDStructures redone
This commit is contained in:
parent
f894175897
commit
f950313f41
2
.gitignore
vendored
2
.gitignore
vendored
@ -1,5 +1,5 @@
|
||||
.gradle
|
||||
/build/
|
||||
**/build/
|
||||
/.idea/
|
||||
|
||||
# Avoid ignoring Gradle wrapper jar file (.jar files are usually ignored)
|
||||
|
||||
@ -1,5 +1,5 @@
|
||||
buildscript {
|
||||
ext.kotlin_version = '1.3.0-rc-146'
|
||||
ext.kotlin_version = '1.3.0-rc-190'
|
||||
|
||||
repositories {
|
||||
jcenter()
|
||||
|
||||
@ -1,6 +1,6 @@
|
||||
plugins {
|
||||
id 'kotlin-multiplatform'// version '1.3.0-rc-116'
|
||||
id "me.champeau.gradle.jmh" version "0.4.5"
|
||||
id "me.champeau.gradle.jmh" version "0.4.7"
|
||||
}
|
||||
|
||||
repositories {
|
||||
@ -8,9 +8,14 @@ repositories {
|
||||
mavenCentral()
|
||||
}
|
||||
|
||||
dependencies{
|
||||
jmh 'org.jetbrains.kotlin:kotlin-stdlib-jdk8'
|
||||
}
|
||||
|
||||
kotlin {
|
||||
targets {
|
||||
fromPreset(presets.jvm, 'jvm')
|
||||
fromPreset(presets.jvmWithJava, 'jvm')
|
||||
//fromPreset(presets.jvm, 'jvm')
|
||||
fromPreset(presets.js, 'js')
|
||||
// For ARM, preset should be changed to presets.iosArm32 or presets.iosArm64
|
||||
// For Linux, preset should be changed to e.g. presets.linuxX64
|
||||
|
||||
@ -1,18 +1,19 @@
|
||||
package scientifik.kmath.linear
|
||||
|
||||
import scientifik.kmath.structures.MutableNDArray
|
||||
import scientifik.kmath.structures.NDArray
|
||||
import scientifik.kmath.structures.NDArrays
|
||||
import scientifik.kmath.structures.MutableNDStructure
|
||||
import scientifik.kmath.structures.NDStructure
|
||||
import scientifik.kmath.structures.genericNdStructure
|
||||
import scientifik.kmath.structures.get
|
||||
import kotlin.math.absoluteValue
|
||||
|
||||
/**
|
||||
* Implementation copier from Apache common-maths
|
||||
* Implementation based on Apache common-maths LU-decomposition
|
||||
*/
|
||||
abstract class LUDecomposition<T : Comparable<T>>(val matrix: Matrix<T>) {
|
||||
|
||||
private val field get() = matrix.context.field
|
||||
/** Entries of LU decomposition. */
|
||||
internal val lu: NDArray<T>
|
||||
internal val lu: NDStructure<T>
|
||||
/** Pivot permutation associated with LU decomposition. */
|
||||
internal val pivot: IntArray
|
||||
/** Parity of the permutation associated with the LU decomposition. */
|
||||
@ -85,26 +86,18 @@ abstract class LUDecomposition<T : Comparable<T>>(val matrix: Matrix<T>) {
|
||||
}
|
||||
}
|
||||
|
||||
// /**
|
||||
// * Get a solver for finding the A X = B solution in exact linear
|
||||
// * sense.
|
||||
// * @return a solver
|
||||
// */
|
||||
// val solver: DecompositionSolver
|
||||
// get() = Solver(lu, pivot, singular)
|
||||
|
||||
/**
|
||||
* In-place transformation for [MutableNDArray], using given transformation for each element
|
||||
*/
|
||||
operator fun <T> MutableNDArray<T>.set(i: Int, j: Int, value: T) {
|
||||
this[listOf(i, j)] = value
|
||||
operator fun <T> MutableNDStructure<T>.set(i: Int, j: Int, value: T) {
|
||||
this[intArrayOf(i, j)] = value
|
||||
}
|
||||
|
||||
abstract fun isSingular(value: T): Boolean
|
||||
|
||||
private fun abs(value: T) = if (value > matrix.context.field.zero) value else with(matrix.context.field) { -value }
|
||||
|
||||
private fun calculateLU(): Pair<NDArray<T>, IntArray> {
|
||||
private fun calculateLU(): Pair<NDStructure<T>, IntArray> {
|
||||
if (matrix.rows != matrix.columns) {
|
||||
error("LU decomposition supports only square matrices")
|
||||
}
|
||||
@ -112,7 +105,7 @@ abstract class LUDecomposition<T : Comparable<T>>(val matrix: Matrix<T>) {
|
||||
val m = matrix.columns
|
||||
val pivot = IntArray(matrix.rows)
|
||||
//TODO fix performance
|
||||
val lu: MutableNDArray<T> = NDArrays.createMutable(matrix.context.field, listOf(matrix.rows, matrix.columns)) { index -> matrix[index[0], index[1]] }
|
||||
val lu: MutableNDStructure<T> = genericNdStructure(intArrayOf(matrix.rows, matrix.columns)) { index -> matrix[index[0], index[1]] }
|
||||
|
||||
|
||||
with(matrix.context.field) {
|
||||
@ -203,44 +196,6 @@ class RealLUDecomposition(matrix: Matrix<Double>, private val singularityThresho
|
||||
/** Specialized solver. */
|
||||
object RealLUSolver : LinearSolver<Double> {
|
||||
|
||||
//
|
||||
// /** {@inheritDoc} */
|
||||
// override fun solve(b: RealVector): RealVector {
|
||||
// val m = pivot.size
|
||||
// if (b.getDimension() != m) {
|
||||
// throw DimensionMismatchException(b.getDimension(), m)
|
||||
// }
|
||||
// if (singular) {
|
||||
// throw SingularMatrixException()
|
||||
// }
|
||||
//
|
||||
// val bp = DoubleArray(m)
|
||||
//
|
||||
// // Apply permutations to b
|
||||
// for (row in 0 until m) {
|
||||
// bp[row] = b.getEntry(pivot[row])
|
||||
// }
|
||||
//
|
||||
// // Solve LY = b
|
||||
// for (col in 0 until m) {
|
||||
// val bpCol = bp[col]
|
||||
// for (i in col + 1 until m) {
|
||||
// bp[i] -= bpCol * lu[i][col]
|
||||
// }
|
||||
// }
|
||||
//
|
||||
// // Solve UX = Y
|
||||
// for (col in m - 1 downTo 0) {
|
||||
// bp[col] /= lu[col][col]
|
||||
// val bpCol = bp[col]
|
||||
// for (i in 0 until col) {
|
||||
// bp[i] -= bpCol * lu[i][col]
|
||||
// }
|
||||
// }
|
||||
//
|
||||
// return ArrayRealVector(bp, false)
|
||||
// }
|
||||
|
||||
|
||||
fun decompose(mat: Matrix<Double>, threshold: Double = 1e-11): RealLUDecomposition = RealLUDecomposition(mat, threshold)
|
||||
|
||||
|
||||
@ -4,10 +4,10 @@ import scientifik.kmath.operations.DoubleField
|
||||
import scientifik.kmath.operations.Field
|
||||
import scientifik.kmath.operations.Space
|
||||
import scientifik.kmath.operations.SpaceElement
|
||||
import scientifik.kmath.structures.GenericNDField
|
||||
import scientifik.kmath.structures.NDArray
|
||||
import scientifik.kmath.structures.NDArrays.createFactory
|
||||
import scientifik.kmath.structures.NDFieldFactory
|
||||
import scientifik.kmath.structures.realNDFieldFactory
|
||||
import scientifik.kmath.structures.NDField
|
||||
import scientifik.kmath.structures.get
|
||||
|
||||
/**
|
||||
* The space for linear elements. Supports scalar product alongside with standard linear operations.
|
||||
@ -193,6 +193,11 @@ interface Vector<T : Any> : SpaceElement<Vector<T>, VectorSpace<T>> {
|
||||
}
|
||||
}
|
||||
|
||||
typealias NDFieldFactory<T> = (IntArray) -> NDField<T>
|
||||
|
||||
internal fun <T : Any> genericNDFieldFactory(field: Field<T>): NDFieldFactory<T> = { index -> GenericNDField(index, field) }
|
||||
internal val realNDFieldFactory: NDFieldFactory<Double> = { index -> GenericNDField(index, DoubleField) }
|
||||
|
||||
|
||||
/**
|
||||
* NDArray-based implementation of vector space. By default uses slow [SimpleNDField], but could be overridden with custom [NDField] factory.
|
||||
@ -201,11 +206,11 @@ class ArrayMatrixSpace<T : Any>(
|
||||
rows: Int,
|
||||
columns: Int,
|
||||
field: Field<T>,
|
||||
val ndFactory: NDFieldFactory<T> = createFactory(field)
|
||||
val ndFactory: NDFieldFactory<T> = genericNDFieldFactory(field)
|
||||
) : MatrixSpace<T>(rows, columns, field) {
|
||||
|
||||
val ndField by lazy {
|
||||
ndFactory(listOf(rows, columns))
|
||||
ndFactory(intArrayOf(rows, columns))
|
||||
}
|
||||
|
||||
override fun produce(initializer: (Int, Int) -> T): Matrix<T> = ArrayMatrix(this, initializer)
|
||||
@ -218,10 +223,10 @@ class ArrayMatrixSpace<T : Any>(
|
||||
class ArrayVectorSpace<T : Any>(
|
||||
size: Int,
|
||||
field: Field<T>,
|
||||
val ndFactory: NDFieldFactory<T> = createFactory(field)
|
||||
val ndFactory: NDFieldFactory<T> = genericNDFieldFactory(field)
|
||||
) : VectorSpace<T>(size, field) {
|
||||
val ndField by lazy {
|
||||
ndFactory(listOf(size))
|
||||
ndFactory(intArrayOf(size))
|
||||
}
|
||||
|
||||
override fun produce(initializer: (Int) -> T): Vector<T> = ArrayVector(this, initializer)
|
||||
@ -306,6 +311,6 @@ fun <T : Any> Vector<T>.toMatrix(): Matrix<T> {
|
||||
// //Generic vector
|
||||
// matrix(size, 1, context.field) { i, j -> get(i) }
|
||||
// }
|
||||
return Matrix.of(size, 1, context.field) { i, j -> get(i) }
|
||||
return Matrix.of(size, 1, context.field) { i, _ -> get(i) }
|
||||
}
|
||||
|
||||
|
||||
@ -1,112 +0,0 @@
|
||||
package scientifik.kmath.structures
|
||||
|
||||
import scientifik.kmath.operations.Field
|
||||
|
||||
|
||||
/**
|
||||
* A generic buffer for both primitives and objects
|
||||
*/
|
||||
interface Buffer<T> {
|
||||
operator fun get(index: Int): T
|
||||
operator fun set(index: Int, value: T)
|
||||
|
||||
/**
|
||||
* A shallow copy of the buffer
|
||||
*/
|
||||
fun copy(): Buffer<T>
|
||||
}
|
||||
|
||||
/**
|
||||
* Generic implementation of NDField based on continuous buffer
|
||||
*/
|
||||
abstract class BufferNDField<T>(shape: List<Int>, field: Field<T>) : NDField<T>(shape, field) {
|
||||
|
||||
/**
|
||||
* Strides for memory access
|
||||
*/
|
||||
private val strides: List<Int> by lazy {
|
||||
ArrayList<Int>(shape.size).apply {
|
||||
var current = 1
|
||||
add(1)
|
||||
shape.forEach {
|
||||
current *= it
|
||||
add(current)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
protected fun offset(index: List<Int>): Int {
|
||||
return index.mapIndexed { i, value ->
|
||||
if (value < 0 || value >= shape[i]) {
|
||||
throw RuntimeException("Index out of shape bounds: ($i,$value)")
|
||||
}
|
||||
value * strides[i]
|
||||
}.sum()
|
||||
}
|
||||
|
||||
//TODO introduce a fast way to calculate index of the next element?
|
||||
protected fun index(offset: Int): List<Int> {
|
||||
return sequence {
|
||||
var current = offset
|
||||
var strideIndex = strides.size - 2
|
||||
while (strideIndex >= 0) {
|
||||
yield(current / strides[strideIndex])
|
||||
current %= strides[strideIndex]
|
||||
strideIndex--
|
||||
}
|
||||
}.toList().reversed()
|
||||
}
|
||||
|
||||
private val capacity: Int
|
||||
get() = strides[shape.size]
|
||||
|
||||
|
||||
protected abstract fun createBuffer(capacity: Int, initializer: (Int) -> T): Buffer<T>
|
||||
|
||||
override fun produce(initializer: (List<Int>) -> T): NDArray<T> {
|
||||
val buffer = createBuffer(capacity) { initializer(index(it)) }
|
||||
return BufferNDArray(this, buffer)
|
||||
}
|
||||
|
||||
/**
|
||||
* Produce mutable NDArray instance
|
||||
*/
|
||||
fun produceMutable(initializer: (List<Int>) -> T): MutableNDArray<T> {
|
||||
val buffer = createBuffer(capacity) { initializer(index(it)) }
|
||||
return MutableBufferedNDArray(this, buffer)
|
||||
}
|
||||
|
||||
|
||||
private open class BufferNDArray<T>(override val context: BufferNDField<T>, val data: Buffer<T>) : NDArray<T> {
|
||||
|
||||
override fun get(vararg index: Int): T {
|
||||
return data[context.offset(index.asList())]
|
||||
}
|
||||
|
||||
override fun equals(other: Any?): Boolean {
|
||||
if (this === other) return true
|
||||
if (other !is BufferNDArray<*>) return false
|
||||
|
||||
if (context != other.context) return false
|
||||
if (data != other.data) return false
|
||||
|
||||
return true
|
||||
}
|
||||
|
||||
override fun hashCode(): Int {
|
||||
var result = context.hashCode()
|
||||
result = 31 * result + data.hashCode()
|
||||
return result
|
||||
}
|
||||
|
||||
override val self: NDArray<T> get() = this
|
||||
}
|
||||
|
||||
private class MutableBufferedNDArray<T>(context: BufferNDField<T>, data: Buffer<T>): BufferNDArray<T>(context,data), MutableNDArray<T>{
|
||||
override operator fun set(index: List<Int>, value: T){
|
||||
data[context.offset(index)] = value
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@ -0,0 +1,74 @@
|
||||
package scientifik.kmath.structures
|
||||
|
||||
|
||||
/**
|
||||
* A generic linear buffer for both primitives and objects
|
||||
*/
|
||||
interface Buffer<T> {
|
||||
|
||||
val size: Int
|
||||
|
||||
operator fun get(index: Int): T
|
||||
|
||||
/**
|
||||
* A shallow copy of the buffer
|
||||
*/
|
||||
fun copy(): Buffer<T>
|
||||
}
|
||||
|
||||
interface MutableBuffer<T> : Buffer<T> {
|
||||
operator fun set(index: Int, value: T)
|
||||
|
||||
/**
|
||||
* A shallow copy of the buffer
|
||||
*/
|
||||
override fun copy(): MutableBuffer<T>
|
||||
}
|
||||
|
||||
inline class ListBuffer<T>(private val list: MutableList<T>) : MutableBuffer<T> {
|
||||
|
||||
override val size: Int
|
||||
get() = list.size
|
||||
|
||||
override fun get(index: Int): T = list[index]
|
||||
|
||||
override fun set(index: Int, value: T) {
|
||||
list[index] = value
|
||||
}
|
||||
|
||||
override fun copy(): MutableBuffer<T> = ListBuffer(ArrayList(list))
|
||||
}
|
||||
|
||||
class ArrayBuffer<T>(private val array: Array<T>) : MutableBuffer<T> {
|
||||
override val size: Int
|
||||
get() = array.size
|
||||
|
||||
override fun get(index: Int): T = array[index]
|
||||
|
||||
override fun set(index: Int, value: T) {
|
||||
array[index] = value
|
||||
}
|
||||
|
||||
override fun copy(): MutableBuffer<T> = ArrayBuffer(array.copyOf())
|
||||
}
|
||||
|
||||
class DoubleBuffer(private val array: DoubleArray) : MutableBuffer<Double> {
|
||||
override val size: Int
|
||||
get() = array.size
|
||||
|
||||
override fun get(index: Int): Double = array[index]
|
||||
|
||||
override fun set(index: Int, value: Double) {
|
||||
array[index] = value
|
||||
}
|
||||
|
||||
override fun copy(): MutableBuffer<Double> = DoubleBuffer(array.copyOf())
|
||||
}
|
||||
|
||||
inline fun <reified T : Any> buffer(size: Int, noinline initializer: (Int) -> T): Buffer<T> {
|
||||
return ArrayBuffer(Array(size, initializer))
|
||||
}
|
||||
|
||||
inline fun <reified T : Any> mutableBuffer(size: Int, noinline initializer: (Int) -> T): MutableBuffer<T> {
|
||||
return ArrayBuffer(Array(size, initializer))
|
||||
}
|
||||
@ -1,72 +0,0 @@
|
||||
package scientifik.kmath.structures
|
||||
|
||||
import scientifik.kmath.operations.Field
|
||||
|
||||
typealias NDFieldFactory<T> = (shape: List<Int>) -> NDField<T>
|
||||
|
||||
/**
|
||||
* The factory class for fast platform-dependent implementation of NDField of doubles
|
||||
*/
|
||||
expect val realNDFieldFactory: NDFieldFactory<Double>
|
||||
|
||||
|
||||
class SimpleNDField<T : Any>(field: Field<T>, shape: List<Int>) : BufferNDField<T>(shape, field) {
|
||||
override fun createBuffer(capacity: Int, initializer: (Int) -> T): Buffer<T> {
|
||||
val array = ArrayList<T>(capacity)
|
||||
(0 until capacity).forEach {
|
||||
array.add(initializer(it))
|
||||
}
|
||||
|
||||
return BufferOfObjects(array)
|
||||
}
|
||||
|
||||
private class BufferOfObjects<T>(val array: ArrayList<T>) : Buffer<T> {
|
||||
override fun get(index: Int): T = array[index]
|
||||
|
||||
override fun set(index: Int, value: T) {
|
||||
array[index] = value
|
||||
}
|
||||
|
||||
override fun copy(): Buffer<T> = BufferOfObjects(ArrayList(array))
|
||||
}
|
||||
}
|
||||
|
||||
object NDArrays {
|
||||
/**
|
||||
* Create a platform-optimized NDArray of doubles
|
||||
*/
|
||||
fun realNDArray(shape: List<Int>, initializer: (List<Int>) -> Double = { 0.0 }): NDArray<Double> {
|
||||
return realNDFieldFactory(shape).produce(initializer)
|
||||
}
|
||||
|
||||
fun real1DArray(dim: Int, initializer: (Int) -> Double = { _ -> 0.0 }): NDArray<Double> {
|
||||
return realNDArray(listOf(dim)) { initializer(it[0]) }
|
||||
}
|
||||
|
||||
fun real2DArray(dim1: Int, dim2: Int, initializer: (Int, Int) -> Double = { _, _ -> 0.0 }): NDArray<Double> {
|
||||
return realNDArray(listOf(dim1, dim2)) { initializer(it[0], it[1]) }
|
||||
}
|
||||
|
||||
fun real3DArray(dim1: Int, dim2: Int, dim3: Int, initializer: (Int, Int, Int) -> Double = { _, _, _ -> 0.0 }): NDArray<Double> {
|
||||
return realNDArray(listOf(dim1, dim2, dim3)) { initializer(it[0], it[1], it[2]) }
|
||||
}
|
||||
|
||||
/**
|
||||
* Simple boxing NDField
|
||||
*/
|
||||
fun <T : Any> createFactory(field: Field<T>): NDFieldFactory<T> = { shape -> SimpleNDField(field, shape) }
|
||||
|
||||
/**
|
||||
* Simple boxing NDArray
|
||||
*/
|
||||
fun <T : Any> create(field: Field<T>, shape: List<Int>, initializer: (List<Int>) -> T): NDArray<T> {
|
||||
return SimpleNDField(field, shape).produce { initializer(it) }
|
||||
}
|
||||
|
||||
/**
|
||||
* Mutable boxing NDArray
|
||||
*/
|
||||
fun <T : Any> createMutable(field: Field<T>, shape: List<Int>, initializer: (List<Int>) -> T): MutableNDArray<T> {
|
||||
return SimpleNDField(field, shape).produceMutable { initializer(it) }
|
||||
}
|
||||
}
|
||||
@ -1,12 +1,13 @@
|
||||
package scientifik.kmath.structures
|
||||
|
||||
import scientifik.kmath.operations.DoubleField
|
||||
import scientifik.kmath.operations.Field
|
||||
import scientifik.kmath.operations.FieldElement
|
||||
|
||||
/**
|
||||
* An exception is thrown when the expected ans actual shape of NDArray differs
|
||||
*/
|
||||
class ShapeMismatchException(val expected: List<Int>, val actual: List<Int>) : RuntimeException()
|
||||
class ShapeMismatchException(val expected: IntArray, val actual: IntArray) : RuntimeException()
|
||||
|
||||
/**
|
||||
* Field for n-dimensional arrays.
|
||||
@ -14,13 +15,15 @@ class ShapeMismatchException(val expected: List<Int>, val actual: List<Int>) : R
|
||||
* @param field - operations field defined on individual array element
|
||||
* @param T the type of the element contained in NDArray
|
||||
*/
|
||||
abstract class NDField<T>(val shape: List<Int>, val field: Field<T>) : Field<NDArray<T>> {
|
||||
abstract class NDField<T>(val shape: IntArray, val field: Field<T>) : Field<NDArray<T>> {
|
||||
|
||||
abstract fun produceStructure(initializer: (IntArray) -> T): NDStructure<T>
|
||||
|
||||
/**
|
||||
* Create new instance of NDArray using field shape and given initializer
|
||||
* The producer takes list of indices as argument and returns contained value
|
||||
*/
|
||||
abstract fun produce(initializer: (List<Int>) -> T): NDArray<T>
|
||||
fun produce(initializer: (IntArray) -> T): NDArray<T> = NDArray(this, produceStructure(initializer))
|
||||
|
||||
override val zero: NDArray<T> by lazy {
|
||||
produce { this.field.zero }
|
||||
@ -31,7 +34,7 @@ abstract class NDField<T>(val shape: List<Int>, val field: Field<T>) : Field<NDA
|
||||
*/
|
||||
private fun checkShape(vararg arrays: NDArray<T>) {
|
||||
arrays.forEach {
|
||||
if (shape != it.shape) {
|
||||
if (!shape.contentEquals(it.shape)) {
|
||||
throw ShapeMismatchException(shape, it.shape)
|
||||
}
|
||||
}
|
||||
@ -71,76 +74,44 @@ abstract class NDField<T>(val shape: List<Int>, val field: Field<T>) : Field<NDA
|
||||
checkShape(a)
|
||||
return produce { with(field) { a[it] / b[it] } }
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Many-dimensional array
|
||||
*/
|
||||
interface NDArray<T> : FieldElement<NDArray<T>, NDField<T>> {
|
||||
|
||||
/**
|
||||
* The list of dimensions of this NDArray
|
||||
* Reverse sum operation
|
||||
*/
|
||||
val shape: List<Int>
|
||||
get() = context.shape
|
||||
operator fun <T> T.plus(arg: NDArray<T>): NDArray<T> = arg + this
|
||||
|
||||
/**
|
||||
* The number of dimentsions for this array
|
||||
* Reverse minus operation
|
||||
*/
|
||||
val dimension: Int
|
||||
get() = shape.size
|
||||
|
||||
/**
|
||||
* Get the element with given indexes. If number of indexes is different from {@link dimension}, throws exception.
|
||||
*/
|
||||
operator fun get(vararg index: Int): T
|
||||
|
||||
operator fun get(index: List<Int>): T {
|
||||
return get(*index.toIntArray())
|
||||
}
|
||||
|
||||
operator fun iterator(): Iterator<Pair<List<Int>, T>> {
|
||||
return iterateIndexes(shape).map { Pair(it, this[it]) }.iterator()
|
||||
operator fun <T> T.minus(arg: NDArray<T>): NDArray<T> = arg.transform { _, value ->
|
||||
with(arg.context.field) {
|
||||
this@minus - value
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Generate new NDArray, using given transformation for each element
|
||||
* Reverse product operation
|
||||
*/
|
||||
fun transform(action: (List<Int>, T) -> T): NDArray<T> = context.produce { action(it, this[it]) }
|
||||
operator fun <T> T.times(arg: NDArray<T>): NDArray<T> = arg * this
|
||||
|
||||
companion object {
|
||||
/**
|
||||
* Iterate over all indexes in the nd-shape
|
||||
*/
|
||||
fun iterateIndexes(shape: List<Int>): Sequence<List<Int>> {
|
||||
return if (shape.size == 1) {
|
||||
(0 until shape[0]).asSequence().map { listOf(it) }
|
||||
} else {
|
||||
val tailShape = ArrayList(shape).apply { removeAt(0) }
|
||||
val tailSequence: List<List<Int>> = iterateIndexes(tailShape).toList()
|
||||
(0 until shape[0]).asSequence().map { firstIndex ->
|
||||
//adding first element to each of provided index lists
|
||||
tailSequence.map { listOf(firstIndex) + it }.asSequence()
|
||||
}.flatten()
|
||||
}
|
||||
/**
|
||||
* Reverse division operation
|
||||
*/
|
||||
operator fun <T> T.div(arg: NDArray<T>): NDArray<T> = arg.transform { _, value ->
|
||||
with(arg.context.field) {
|
||||
this@div / value
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* In-place mutable [NDArray]
|
||||
* NDStructure coupled to the context. Emulates Python ndarray
|
||||
*/
|
||||
interface MutableNDArray<T> : NDArray<T> {
|
||||
operator fun set(index: List<Int>, value: T)
|
||||
}
|
||||
data class NDArray<T>(override val context: NDField<T>, private val structure: NDStructure<T>) : FieldElement<NDArray<T>, NDField<T>>, NDStructure<T> by structure {
|
||||
override val self: NDArray<T>
|
||||
get() = this
|
||||
|
||||
/**
|
||||
* In-place transformation for [MutableNDArray], using given transformation for each element
|
||||
*/
|
||||
fun <T> MutableNDArray<T>.transformInPlace(action: (List<Int>, T) -> T) {
|
||||
for ((index, oldValue) in this) {
|
||||
this[index] = action(index, oldValue)
|
||||
}
|
||||
fun transform(action: (IntArray, T) -> T): NDArray<T> = context.produce { action(it, get(*it)) }
|
||||
}
|
||||
|
||||
/**
|
||||
@ -159,11 +130,6 @@ operator fun <T> NDArray<T>.plus(arg: T): NDArray<T> = transform { _, value ->
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Reverse sum operation
|
||||
*/
|
||||
operator fun <T> T.plus(arg: NDArray<T>): NDArray<T> = arg + this
|
||||
|
||||
/**
|
||||
* Subtraction operation between [NDArray] and single element
|
||||
*/
|
||||
@ -173,15 +139,6 @@ operator fun <T> NDArray<T>.minus(arg: T): NDArray<T> = transform { _, value ->
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Reverse minus operation
|
||||
*/
|
||||
operator fun <T> T.minus(arg: NDArray<T>): NDArray<T> = arg.transform { _, value ->
|
||||
with(arg.context.field) {
|
||||
this@minus - value
|
||||
}
|
||||
}
|
||||
|
||||
/* prod and div */
|
||||
|
||||
/**
|
||||
@ -193,11 +150,6 @@ operator fun <T> NDArray<T>.times(arg: T): NDArray<T> = transform { _, value ->
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Reverse product operation
|
||||
*/
|
||||
operator fun <T> T.times(arg: NDArray<T>): NDArray<T> = arg * this
|
||||
|
||||
/**
|
||||
* Division operation between [NDArray] and single element
|
||||
*/
|
||||
@ -207,12 +159,41 @@ operator fun <T> NDArray<T>.div(arg: T): NDArray<T> = transform { _, value ->
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Reverse division operation
|
||||
*/
|
||||
operator fun <T> T.div(arg: NDArray<T>): NDArray<T> = arg.transform { _, value ->
|
||||
with(arg.context.field) {
|
||||
this@div / value
|
||||
}
|
||||
class GenericNDField<T : Any>(shape: IntArray, field: Field<T>) : NDField<T>(shape, field) {
|
||||
override fun produceStructure(initializer: (IntArray) -> T): NDStructure<T> = genericNdStructure(shape, initializer)
|
||||
}
|
||||
|
||||
//typealias NDFieldFactory<T> = (IntArray)->NDField<T>
|
||||
|
||||
object NDArrays {
|
||||
/**
|
||||
* Create a platform-optimized NDArray of doubles
|
||||
*/
|
||||
fun realNDArray(shape: IntArray, initializer: (IntArray) -> Double = { 0.0 }): NDArray<Double> {
|
||||
return GenericNDField(shape, DoubleField).produce(initializer)
|
||||
}
|
||||
|
||||
fun real1DArray(dim: Int, initializer: (Int) -> Double = { _ -> 0.0 }): NDArray<Double> {
|
||||
return realNDArray(intArrayOf(dim)) { initializer(it[0]) }
|
||||
}
|
||||
|
||||
fun real2DArray(dim1: Int, dim2: Int, initializer: (Int, Int) -> Double = { _, _ -> 0.0 }): NDArray<Double> {
|
||||
return realNDArray(intArrayOf(dim1, dim2)) { initializer(it[0], it[1]) }
|
||||
}
|
||||
|
||||
fun real3DArray(dim1: Int, dim2: Int, dim3: Int, initializer: (Int, Int, Int) -> Double = { _, _, _ -> 0.0 }): NDArray<Double> {
|
||||
return realNDArray(intArrayOf(dim1, dim2, dim3)) { initializer(it[0], it[1], it[2]) }
|
||||
}
|
||||
|
||||
// /**
|
||||
// * Simple boxing NDField
|
||||
// */
|
||||
// fun <T : Any> fieldFactory(field: Field<T>): NDFieldFactory<T> = { shape -> GenericNDField(shape, field) }
|
||||
|
||||
/**
|
||||
* Simple boxing NDArray
|
||||
*/
|
||||
fun <T : Any> create(field: Field<T>, shape: IntArray, initializer: (IntArray) -> T): NDArray<T> {
|
||||
return GenericNDField(shape, field).produce { initializer(it) }
|
||||
}
|
||||
}
|
||||
@ -0,0 +1,174 @@
|
||||
package scientifik.kmath.structures
|
||||
|
||||
|
||||
interface NDStructure<T> : Iterable<Pair<IntArray, T>> {
|
||||
|
||||
val shape: IntArray
|
||||
|
||||
val dimension
|
||||
get() = shape.size
|
||||
|
||||
operator fun get(index: IntArray): T
|
||||
}
|
||||
|
||||
operator fun <T> NDStructure<T>.get(vararg index: Int): T = get(index)
|
||||
|
||||
interface MutableNDStructure<T> : NDStructure<T> {
|
||||
operator fun set(index: IntArray, value: T)
|
||||
}
|
||||
|
||||
fun <T> MutableNDStructure<T>.transformInPlace(action: (IntArray, T) -> T) {
|
||||
for ((index, oldValue) in this) {
|
||||
this[index] = action(index, oldValue)
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* A way to convert ND index to linear one and back
|
||||
*/
|
||||
interface Strides {
|
||||
/**
|
||||
* Shape of NDstructure
|
||||
*/
|
||||
val shape: IntArray
|
||||
|
||||
/**
|
||||
* Array strides
|
||||
*/
|
||||
val strides: List<Int>
|
||||
|
||||
/**
|
||||
* Get linear index from multidimensional index
|
||||
*/
|
||||
fun offset(index: IntArray): Int
|
||||
|
||||
/**
|
||||
* Get multidimensional from linear
|
||||
*/
|
||||
fun index(offset: Int): IntArray
|
||||
|
||||
val linearSize: Int
|
||||
|
||||
/**
|
||||
* Iterate over ND indices in a natural order
|
||||
*/
|
||||
fun indices(): Sequence<IntArray> {
|
||||
//TODO introduce a fast way to calculate index of the next element?
|
||||
return (0 until linearSize).asSequence().map { index(it) }
|
||||
}
|
||||
}
|
||||
|
||||
class DefaultStrides(override val shape: IntArray) : Strides {
|
||||
/**
|
||||
* Strides for memory access
|
||||
*/
|
||||
override val strides by lazy {
|
||||
sequence {
|
||||
var current = 1
|
||||
yield(1)
|
||||
shape.forEach {
|
||||
current *= it
|
||||
yield(current)
|
||||
}
|
||||
}.toList()
|
||||
}
|
||||
|
||||
override fun offset(index: IntArray): Int {
|
||||
return index.mapIndexed { i, value ->
|
||||
if (value < 0 || value >= shape[i]) {
|
||||
throw RuntimeException("Index out of shape bounds: ($i,$value)")
|
||||
}
|
||||
value * strides[i]
|
||||
}.sum()
|
||||
}
|
||||
|
||||
override fun index(offset: Int): IntArray {
|
||||
return sequence {
|
||||
var current = offset
|
||||
var strideIndex = strides.size - 2
|
||||
while (strideIndex >= 0) {
|
||||
yield(current / strides[strideIndex])
|
||||
current %= strides[strideIndex]
|
||||
strideIndex--
|
||||
}
|
||||
}.toList().reversed().toIntArray()
|
||||
}
|
||||
|
||||
override val linearSize: Int
|
||||
get() = strides[shape.size]
|
||||
}
|
||||
|
||||
abstract class GenericNDStructure<T, B : Buffer<T>> : NDStructure<T> {
|
||||
protected abstract val buffer: B
|
||||
protected abstract val strides: Strides
|
||||
|
||||
override fun get(index: IntArray): T = buffer[strides.offset(index)]
|
||||
|
||||
override val shape: IntArray
|
||||
get() = strides.shape
|
||||
|
||||
override fun iterator(): Iterator<Pair<IntArray, T>> =
|
||||
strides.indices().map { it to this[it] }.iterator()
|
||||
}
|
||||
|
||||
/**
|
||||
* Boxing generic [NDStructure]
|
||||
*/
|
||||
class BufferNDStructure<T>(
|
||||
override val strides: Strides,
|
||||
override val buffer: Buffer<T>
|
||||
) : GenericNDStructure<T, Buffer<T>>() {
|
||||
|
||||
init {
|
||||
if (strides.linearSize != buffer.size) {
|
||||
error("Expected buffer side of ${strides.linearSize}, but found ${buffer.size}")
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
inline fun <reified T: Any> ndStructure(strides: Strides, noinline initializer: (IntArray) -> T) =
|
||||
BufferNDStructure<T>(strides, buffer(strides.linearSize){ i-> initializer(strides.index(i))})
|
||||
|
||||
inline fun <reified T: Any> ndStructure(shape: IntArray, noinline initializer: (IntArray) -> T) =
|
||||
ndStructure(DefaultStrides(shape), initializer)
|
||||
|
||||
|
||||
/**
|
||||
* Mutable ND buffer based on linear [Buffer]
|
||||
*/
|
||||
class MutableBufferNDStructure<T>(
|
||||
override val strides: Strides,
|
||||
override val buffer: MutableBuffer<T>
|
||||
) : GenericNDStructure<T, MutableBuffer<T>>(), MutableNDStructure<T> {
|
||||
|
||||
init {
|
||||
if (strides.linearSize != buffer.size) {
|
||||
error("Expected buffer side of ${strides.linearSize}, but found ${buffer.size}")
|
||||
}
|
||||
}
|
||||
|
||||
override fun set(index: IntArray, value: T) = buffer.set(strides.offset(index), value)
|
||||
}
|
||||
|
||||
/**
|
||||
* Create optimized mutable structure for given type
|
||||
*/
|
||||
inline fun <reified T: Any> mutableNdStructure(strides: Strides, noinline initializer: (IntArray) -> T) =
|
||||
MutableBufferNDStructure(strides, mutableBuffer(strides.linearSize) { i -> initializer(strides.index(i)) })
|
||||
|
||||
inline fun <reified T: Any> mutableNdStructure(shape: IntArray, noinline initializer: (IntArray) -> T) =
|
||||
mutableNdStructure(DefaultStrides(shape), initializer)
|
||||
|
||||
/**
|
||||
* Create universal mutable structure
|
||||
*/
|
||||
fun <T> genericNdStructure(shape: IntArray, initializer: (IntArray) -> T): MutableBufferNDStructure<T>{
|
||||
val strides = DefaultStrides(shape)
|
||||
val sequence = sequence{
|
||||
strides.indices().forEach{
|
||||
yield(initializer(it))
|
||||
}
|
||||
}
|
||||
val buffer = ListBuffer<T>(sequence.toMutableList())
|
||||
return MutableBufferNDStructure(strides, buffer)
|
||||
}
|
||||
@ -7,15 +7,15 @@ class ArrayMatrixTest {
|
||||
|
||||
@Test
|
||||
fun testSum() {
|
||||
val vector1 = realVector(5) { it.toDouble() }
|
||||
val vector2 = realVector(5) { 5 - it.toDouble() }
|
||||
val vector1 = Vector.ofReal(5) { it.toDouble() }
|
||||
val vector2 = Vector.ofReal(5) { 5 - it.toDouble() }
|
||||
val sum = vector1 + vector2
|
||||
assertEquals(5.0, sum[2])
|
||||
}
|
||||
|
||||
@Test
|
||||
fun testVectorToMatrix() {
|
||||
val vector = realVector(5) { it.toDouble() }
|
||||
val vector = Vector.ofReal(5) { it.toDouble() }
|
||||
val matrix = vector.toMatrix()
|
||||
assertEquals(4.0, matrix[4, 0])
|
||||
}
|
||||
@ -23,8 +23,8 @@ class ArrayMatrixTest {
|
||||
|
||||
@Test
|
||||
fun testDot() {
|
||||
val vector1 = realVector(5) { it.toDouble() }
|
||||
val vector2 = realVector(5) { 5 - it.toDouble() }
|
||||
val vector1 = Vector.ofReal(5) { it.toDouble() }
|
||||
val vector2 = Vector.ofReal(5) { 5 - it.toDouble() }
|
||||
val product = vector1.toMatrix() dot (vector2.toMatrix().transpose())
|
||||
|
||||
|
||||
|
||||
@ -6,10 +6,10 @@ import kotlin.test.Test
|
||||
import kotlin.test.assertEquals
|
||||
|
||||
|
||||
class SimpleNDFieldTest{
|
||||
class GenericNDFieldTest{
|
||||
@Test
|
||||
fun testStrides(){
|
||||
val ndArray = create(DoubleField, listOf(10,10)){(it[0]+it[1]).toDouble()}
|
||||
val ndArray = create(DoubleField, intArrayOf(10,10)){(it[0]+it[1]).toDouble()}
|
||||
assertEquals(ndArray[5,5], 10.0)
|
||||
}
|
||||
|
||||
@ -1,12 +1,11 @@
|
||||
package scietifik.kmath.structures
|
||||
|
||||
import org.openjdk.jmh.annotations.*
|
||||
import java.nio.IntBuffer
|
||||
|
||||
|
||||
@Fork(1)
|
||||
@Warmup(iterations = 2)
|
||||
@Measurement(iterations = 50)
|
||||
@Measurement(iterations = 5)
|
||||
@State(Scope.Benchmark)
|
||||
open class ArrayBenchmark {
|
||||
|
||||
|
||||
@ -1,8 +0,0 @@
|
||||
package scientifik.kmath.structures
|
||||
|
||||
import scientifik.kmath.operations.DoubleField
|
||||
|
||||
/**
|
||||
* Using boxing implementation for js
|
||||
*/
|
||||
actual val realNDFieldFactory: NDFieldFactory<Double> = NDArrays.createFactory(DoubleField)
|
||||
@ -1,26 +0,0 @@
|
||||
package scientifik.kmath.structures
|
||||
|
||||
import scientifik.kmath.operations.DoubleField
|
||||
import java.nio.DoubleBuffer
|
||||
|
||||
private class RealNDField(shape: List<Int>) : BufferNDField<Double>(shape, DoubleField) {
|
||||
override fun createBuffer(capacity: Int, initializer: (Int) -> Double): Buffer<Double> {
|
||||
val array = DoubleArray(capacity, initializer)
|
||||
val buffer = DoubleBuffer.wrap(array)
|
||||
return BufferOfDoubles(buffer)
|
||||
}
|
||||
|
||||
private class BufferOfDoubles(val buffer: DoubleBuffer): Buffer<Double>{
|
||||
override fun get(index: Int): Double = buffer.get(index)
|
||||
|
||||
override fun set(index: Int, value: Double) {
|
||||
buffer.put(index, value)
|
||||
}
|
||||
|
||||
override fun copy(): Buffer<Double> {
|
||||
return BufferOfDoubles(buffer)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
actual val realNDFieldFactory: NDFieldFactory<Double> = { shape -> RealNDField(shape) }
|
||||
Loading…
Reference in New Issue
Block a user