KMP library for tensors #300

Merged
grinisrit merged 215 commits from feature/tensor-algebra into dev 2021-05-08 09:48:04 +03:00
4 changed files with 190 additions and 190 deletions
Showing only changes of commit d281dfca3a - Show all commits

View File

@ -1,7 +1,5 @@
package space.kscience.kmath.tensors.core
import space.kscience.kmath.nd.MutableStructure1D
import space.kscience.kmath.nd.MutableStructure2D
import space.kscience.kmath.structures.*
import space.kscience.kmath.tensors.TensorStructure
@ -18,9 +16,6 @@ public open class BufferedTensor<T>(
public val numel: Int
get() = linearStructure.size
internal constructor(tensor: BufferedTensor<T>) :
this(tensor.shape, tensor.buffer, tensor.bufferStart)
override fun get(index: IntArray): T = buffer[bufferStart + linearStructure.offset(index)]
override fun set(index: IntArray, value: T) {
@ -35,39 +30,6 @@ public open class BufferedTensor<T>(
override fun hashCode(): Int = 0
internal fun vectorSequence(): Sequence<BufferedTensor<T>> = sequence {
val n = shape.size
val vectorOffset = shape[n - 1]
val vectorShape = intArrayOf(shape.last())
for (offset in 0 until numel step vectorOffset) {
val vector = BufferedTensor(vectorShape, buffer, offset)
yield(vector)
}
}
internal fun matrixSequence(): Sequence<BufferedTensor<T>> = sequence {
check(shape.size >= 2) { "todo" }
val n = shape.size
val matrixOffset = shape[n - 1] * shape[n - 2]
val matrixShape = intArrayOf(shape[n - 2], shape[n - 1])
for (offset in 0 until numel step matrixOffset) {
val matrix = BufferedTensor(matrixShape, buffer, offset)
yield(matrix)
}
}
internal inline fun forEachVector(vectorAction: (BufferedTensor<T>) -> Unit): Unit {
for (vector in vectorSequence()) {
vectorAction(vector)
}
}
internal inline fun forEachMatrix(matrixAction: (BufferedTensor<T>) -> Unit): Unit {
for (matrix in matrixSequence()) {
matrixAction(matrix)
}
}
}
public class IntTensor internal constructor(

View File

@ -1,11 +1,8 @@
package space.kscience.kmath.tensors.core
import space.kscience.kmath.nd.MutableStructure1D
import space.kscience.kmath.nd.MutableStructure2D
import space.kscience.kmath.tensors.LinearOpsTensorAlgebra
import space.kscience.kmath.nd.as1D
import space.kscience.kmath.nd.as2D
import space.kscience.kmath.tensors.LinearOpsTensorAlgebra
import kotlin.math.sqrt
public class DoubleLinearOpsTensorAlgebra :
LinearOpsTensorAlgebra<Double, DoubleTensor, IntTensor>,
@ -15,48 +12,6 @@ public class DoubleLinearOpsTensorAlgebra :
override fun DoubleTensor.det(): DoubleTensor = detLU()
private inline fun luHelper(lu: MutableStructure2D<Double>, pivots: MutableStructure1D<Int>, m: Int) {
for (row in 0 until m) pivots[row] = row
for (i in 0 until m) {
var maxVal = -1.0
var maxInd = i
for (k in i until m) {
val absA = kotlin.math.abs(lu[k, i])
if (absA > maxVal) {
maxVal = absA
maxInd = k
}
}
//todo check singularity
if (maxInd != i) {
val j = pivots[i]
pivots[i] = pivots[maxInd]
pivots[maxInd] = j
for (k in 0 until m) {
val tmp = lu[i, k]
lu[i, k] = lu[maxInd, k]
lu[maxInd, k] = tmp
}
pivots[m] += 1
}
for (j in i + 1 until m) {
lu[j, i] /= lu[i, i]
for (k in i + 1 until m) {
lu[j, k] -= lu[j, i] * lu[i, k]
}
}
}
}
override fun DoubleTensor.lu(): Pair<DoubleTensor, IntTensor> {
checkSquareMatrix(shape)
@ -80,37 +35,6 @@ public class DoubleLinearOpsTensorAlgebra :
}
private inline fun pivInit(
p: MutableStructure2D<Double>,
pivot: MutableStructure1D<Int>,
n: Int
) {
for (i in 0 until n) {
p[i, pivot[i]] = 1.0
}
}
private inline fun luPivotHelper(
l: MutableStructure2D<Double>,
u: MutableStructure2D<Double>,
lu: MutableStructure2D<Double>,
n: Int
) {
for (i in 0 until n) {
for (j in 0 until n) {
if (i == j) {
l[i, j] = 1.0
}
if (j < i) {
l[i, j] = lu[i, j]
}
if (j >= i) {
u[i, j] = lu[i, j]
}
}
}
}
override fun luPivot(
luTensor: DoubleTensor,
pivotsTensor: IntTensor
@ -139,27 +63,6 @@ public class DoubleLinearOpsTensorAlgebra :
}
private inline fun choleskyHelper(
a: MutableStructure2D<Double>,
l: MutableStructure2D<Double>,
n: Int
) {
for (i in 0 until n) {
for (j in 0 until i) {
var h = a[i, j]
for (k in 0 until j) {
h -= l[i, k] * l[j, k]
}
l[i, j] = h / l[j, j]
}
var h = a[i, i]
for (j in 0 until i) {
h -= l[i, j] * l[i, j]
}
l[i, i] = sqrt(h)
}
}
override fun DoubleTensor.cholesky(): DoubleTensor {
// todo checks
checkSquareMatrix(shape)
@ -185,14 +88,6 @@ public class DoubleLinearOpsTensorAlgebra :
TODO("ANDREI")
}
private fun luMatrixDet(luTensor: MutableStructure2D<Double>, pivotsTensor: MutableStructure1D<Int>): Double {
val lu = luTensor.as2D()
val pivots = pivotsTensor.as1D()
val m = lu.shape[0]
val sign = if ((pivots[m] - m) % 2 == 0) 1.0 else -1.0
return (0 until m).asSequence().map { lu[it, it] }.fold(sign) { left, right -> left * right }
}
public fun DoubleTensor.detLU(): DoubleTensor {
val (luTensor, pivotsTensor) = lu()
val n = shape.size
@ -213,33 +108,6 @@ public class DoubleLinearOpsTensorAlgebra :
return detTensor
}
private fun luMatrixInv(
lu: MutableStructure2D<Double>,
pivots: MutableStructure1D<Int>,
invMatrix: MutableStructure2D<Double>
) {
val m = lu.shape[0]
for (j in 0 until m) {
for (i in 0 until m) {
if (pivots[i] == j) {
invMatrix[i, j] = 1.0
}
for (k in 0 until i) {
invMatrix[i, j] -= lu[i, k] * invMatrix[k, j]
}
}
for (i in m - 1 downTo 0) {
for (k in i + 1 until m) {
invMatrix[i, j] -= lu[i, k] * invMatrix[k, j]
}
invMatrix[i, j] /= lu[i, i]
}
}
}
public fun DoubleTensor.invLU(): DoubleTensor {
val (luTensor, pivotsTensor) = lu()
val invTensor = luTensor.zeroesLike()

View File

@ -1,8 +1,7 @@
package space.kscience.kmath.tensors.core
import space.kscience.kmath.nd.MutableStructure2D
import space.kscience.kmath.nd.as2D
import space.kscience.kmath.tensors.TensorPartialDivisionAlgebra
import space.kscience.kmath.nd.as2D
import kotlin.math.abs
public open class DoubleTensorAlgebra : TensorPartialDivisionAlgebra<Double, DoubleTensor> {
@ -230,23 +229,6 @@ public open class DoubleTensorAlgebra : TensorPartialDivisionAlgebra<Double, Dou
return this.view(other.shape)
}
private inline fun dotHelper(
a: MutableStructure2D<Double>,
b: MutableStructure2D<Double>,
res: MutableStructure2D<Double>,
l: Int, m: Int, n: Int
) {
for (i in 0 until l) {
for (j in 0 until n) {
var curr = 0.0
for (k in 0 until m) {
curr += a[i, k] * b[k, j]
}
res[i, j] = curr
}
}
}
override fun DoubleTensor.dot(other: DoubleTensor): DoubleTensor {
if (this.shape.size == 1 && other.shape.size == 1) {
return DoubleTensor(intArrayOf(1), doubleArrayOf(this.times(other).buffer.array().sum()))

View File

@ -0,0 +1,188 @@
package space.kscience.kmath.tensors.core
import space.kscience.kmath.nd.MutableStructure1D
import space.kscience.kmath.nd.MutableStructure2D
import space.kscience.kmath.nd.as1D
import space.kscience.kmath.nd.as2D
import kotlin.math.sqrt
internal inline fun <T> BufferedTensor<T>.vectorSequence(): Sequence<BufferedTensor<T>> = sequence {
val n = shape.size
val vectorOffset = shape[n - 1]
val vectorShape = intArrayOf(shape.last())
for (offset in 0 until numel step vectorOffset) {
val vector = BufferedTensor(vectorShape, buffer, offset)
yield(vector)
}
}
internal inline fun <T> BufferedTensor<T>.matrixSequence(): Sequence<BufferedTensor<T>> = sequence {
check(shape.size >= 2) { "todo" }
val n = shape.size
val matrixOffset = shape[n - 1] * shape[n - 2]
val matrixShape = intArrayOf(shape[n - 2], shape[n - 1])
for (offset in 0 until numel step matrixOffset) {
val matrix = BufferedTensor(matrixShape, buffer, offset)
yield(matrix)
}
}
internal inline fun <T> BufferedTensor<T>.forEachVector(vectorAction: (BufferedTensor<T>) -> Unit): Unit {
for (vector in vectorSequence()) {
vectorAction(vector)
}
}
internal inline fun <T> BufferedTensor<T>.forEachMatrix(matrixAction: (BufferedTensor<T>) -> Unit): Unit {
for (matrix in matrixSequence()) {
matrixAction(matrix)
}
}
internal inline fun dotHelper(
a: MutableStructure2D<Double>,
b: MutableStructure2D<Double>,
res: MutableStructure2D<Double>,
l: Int, m: Int, n: Int
) {
for (i in 0 until l) {
for (j in 0 until n) {
var curr = 0.0
for (k in 0 until m) {
curr += a[i, k] * b[k, j]
}
res[i, j] = curr
}
}
}
internal inline fun luHelper(lu: MutableStructure2D<Double>, pivots: MutableStructure1D<Int>, m: Int) {
for (row in 0 until m) pivots[row] = row
for (i in 0 until m) {
var maxVal = -1.0
var maxInd = i
for (k in i until m) {
val absA = kotlin.math.abs(lu[k, i])
if (absA > maxVal) {
maxVal = absA
maxInd = k
}
}
//todo check singularity
if (maxInd != i) {
val j = pivots[i]
pivots[i] = pivots[maxInd]
pivots[maxInd] = j
for (k in 0 until m) {
val tmp = lu[i, k]
lu[i, k] = lu[maxInd, k]
lu[maxInd, k] = tmp
}
pivots[m] += 1
}
for (j in i + 1 until m) {
lu[j, i] /= lu[i, i]
for (k in i + 1 until m) {
lu[j, k] -= lu[j, i] * lu[i, k]
}
}
}
}
internal inline fun pivInit(
p: MutableStructure2D<Double>,
pivot: MutableStructure1D<Int>,
n: Int
) {
for (i in 0 until n) {
p[i, pivot[i]] = 1.0
}
}
internal inline fun luPivotHelper(
l: MutableStructure2D<Double>,
u: MutableStructure2D<Double>,
lu: MutableStructure2D<Double>,
n: Int
) {
for (i in 0 until n) {
for (j in 0 until n) {
if (i == j) {
l[i, j] = 1.0
}
if (j < i) {
l[i, j] = lu[i, j]
}
if (j >= i) {
u[i, j] = lu[i, j]
}
}
}
}
internal inline fun choleskyHelper(
a: MutableStructure2D<Double>,
l: MutableStructure2D<Double>,
n: Int
) {
for (i in 0 until n) {
for (j in 0 until i) {
var h = a[i, j]
for (k in 0 until j) {
h -= l[i, k] * l[j, k]
}
l[i, j] = h / l[j, j]
}
var h = a[i, i]
for (j in 0 until i) {
h -= l[i, j] * l[i, j]
}
l[i, i] = sqrt(h)
}
}
internal inline fun luMatrixDet(luTensor: MutableStructure2D<Double>, pivotsTensor: MutableStructure1D<Int>): Double {
val lu = luTensor.as2D()
val pivots = pivotsTensor.as1D()
val m = lu.shape[0]
val sign = if ((pivots[m] - m) % 2 == 0) 1.0 else -1.0
return (0 until m).asSequence().map { lu[it, it] }.fold(sign) { left, right -> left * right }
}
internal inline fun luMatrixInv(
lu: MutableStructure2D<Double>,
pivots: MutableStructure1D<Int>,
invMatrix: MutableStructure2D<Double>
) {
val m = lu.shape[0]
for (j in 0 until m) {
for (i in 0 until m) {
if (pivots[i] == j) {
invMatrix[i, j] = 1.0
}
for (k in 0 until i) {
invMatrix[i, j] -= lu[i, k] * invMatrix[k, j]
}
}
for (i in m - 1 downTo 0) {
for (k in i + 1 until m) {
invMatrix[i, j] -= lu[i, k] * invMatrix[k, j]
}
invMatrix[i, j] /= lu[i, i]
}
}
}