More basic functionality, tests to come

This commit is contained in:
Roland Grinis 2021-05-03 19:49:23 +01:00
parent 7f8914d8ea
commit b59e48410f
6 changed files with 202 additions and 231 deletions

View File

@ -5,14 +5,107 @@
package space.kscience.kmath.tensors.api
/**
* Element-wise analytic operations on [Tensor].
* Analytic operations on [Tensor].
*
* @param T the type of items closed under analytic functions in the tensors.
*/
public interface AnalyticTensorAlgebra<T> :
TensorPartialDivisionAlgebra<T> {
/**
* @return the minimum value of all elements in the input tensor.
*/
public fun Tensor<T>.min(): T
/**
* Returns the minimum value of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the minimum value of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.min(dim: Int, keepDim: Boolean): Tensor<T>
/**
* @return the maximum value of all elements in the input tensor.
*/
public fun Tensor<T>.max(): T
/**
* Returns the maximum value of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the maximum value of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.max(dim: Int, keepDim: Boolean): Tensor<T>
/**
* @return the mean of all elements in the input tensor.
*/
public fun Tensor<T>.mean(): T
/**
* Returns the mean of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the mean of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.mean(dim: Int, keepDim: Boolean): Tensor<T>
/**
* @return the standard deviation of all elements in the input tensor.
*/
public fun Tensor<T>.std(): T
/**
* Returns the standard deviation of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the standard deviation of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.std(dim: Int, keepDim: Boolean): Tensor<T>
/**
* @return the variance of all elements in the input tensor.
*/
public fun Tensor<T>.variance(): T
/**
* Returns the variance of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the variance of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.variance(dim: Int, keepDim: Boolean): Tensor<T>
//For information: https://pytorch.org/docs/stable/generated/torch.exp.html
public fun Tensor<T>.exp(): Tensor<T>

View File

@ -1,124 +0,0 @@
/*
* Copyright 2018-2021 KMath contributors.
* Use of this source code is governed by the Apache 2.0 license that can be found in the license/LICENSE.txt file.
*/
package space.kscience.kmath.tensors.api
import space.kscience.kmath.tensors.core.DoubleTensor
/**
* Common algebra with statistics methods. Operates on [Tensor].
*/
public interface StatisticTensorAlgebra<T>: TensorAlgebra<T> {
/**
* Returns the minimum value of all elements in the input tensor.
*/
public fun Tensor<T>.min(): Double
/**
* Returns the minimum value of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the minimum value of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.min(dim: Int, keepDim: Boolean): DoubleTensor
/**
* Returns the maximum value of all elements in the input tensor.
*/
public fun Tensor<T>.max(): Double
/**
* Returns the maximum value of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the maximum value of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.max(dim: Int, keepDim: Boolean): DoubleTensor
/**
* Returns the sum of all elements in the input tensor.
*/
public fun Tensor<T>.sum(): Double
/**
* Returns the sum of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the sum of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.sum(dim: Int, keepDim: Boolean): DoubleTensor
/**
* Returns the mean of all elements in the input tensor.
*/
public fun Tensor<T>.mean(): Double
/**
* Returns the mean of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the mean of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.mean(dim: Int, keepDim: Boolean): DoubleTensor
/**
* Returns the standard deviation of all elements in the input tensor.
*/
public fun Tensor<T>.std(): Double
/**
* Returns the standard deviation of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the standard deviation of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.std(dim: Int, keepDim: Boolean): DoubleTensor
/**
* Returns the variance of all elements in the input tensor.
*/
public fun Tensor<T>.variance(): Double
/**
* Returns the variance of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the variance of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.variance(dim: Int, keepDim: Boolean): DoubleTensor
}

View File

@ -251,4 +251,24 @@ public interface TensorAlgebra<T>: Algebra<Tensor<T>> {
dim2: Int = -1
): Tensor<T>
/**
* @return the sum of all elements in the input tensor.
*/
public fun Tensor<T>.sum(): T
/**
* Returns the sum of each row of the input tensor in the given dimension [dim].
*
* If [keepDim] is true, the output tensor is of the same size as
* input except in the dimension [dim] where it is of size 1.
* Otherwise, [dim] is squeezed, resulting in the output tensor having 1 fewer dimension.
*
* @param dim the dimension to reduce.
* @param keepDim whether the output tensor has [dim] retained or not.
* @return the sum of each row of the input tensor in the given dimension [dim].
*/
public fun Tensor<T>.sum(dim: Int, keepDim: Boolean): Tensor<T>
}

View File

@ -14,6 +14,60 @@ import kotlin.math.*
public object DoubleAnalyticTensorAlgebra :
AnalyticTensorAlgebra<Double>,
DoubleTensorAlgebra() {
override fun Tensor<Double>.min(): Double = this.fold { it.minOrNull()!! }
override fun Tensor<Double>.min(dim: Int, keepDim: Boolean): DoubleTensor =
foldDim({ x -> x.minOrNull()!! }, dim, keepDim)
override fun Tensor<Double>.max(): Double = this.fold { it.maxOrNull()!! }
override fun Tensor<Double>.max(dim: Int, keepDim: Boolean): DoubleTensor =
foldDim({ x -> x.maxOrNull()!! }, dim, keepDim)
override fun Tensor<Double>.mean(): Double = this.fold { it.sum() / tensor.numElements }
override fun Tensor<Double>.mean(dim: Int, keepDim: Boolean): DoubleTensor =
foldDim(
{ arr ->
check(dim < dimension) { "Dimension $dim out of range $dimension" }
arr.sum() / shape[dim]
},
dim,
keepDim
)
override fun Tensor<Double>.std(): Double = this.fold { arr ->
val mean = arr.sum() / tensor.numElements
sqrt(arr.sumOf { (it - mean) * (it - mean) } / (tensor.numElements - 1))
}
override fun Tensor<Double>.std(dim: Int, keepDim: Boolean): DoubleTensor = foldDim(
{ arr ->
check(dim < dimension) { "Dimension $dim out of range $dimension" }
val mean = arr.sum() / shape[dim]
sqrt(arr.sumOf { (it - mean) * (it - mean) } / (shape[dim] - 1))
},
dim,
keepDim
)
override fun Tensor<Double>.variance(): Double = this.fold { arr ->
val mean = arr.sum() / tensor.numElements
arr.sumOf { (it - mean) * (it - mean) } / (tensor.numElements - 1)
}
override fun Tensor<Double>.variance(dim: Int, keepDim: Boolean): DoubleTensor = foldDim(
{ arr ->
check(dim < dimension) { "Dimension $dim out of range $dimension" }
val mean = arr.sum() / shape[dim]
arr.sumOf { (it - mean) * (it - mean) } / (shape[dim] - 1)
},
dim,
keepDim
)
override fun Tensor<Double>.exp(): DoubleTensor = tensor.map(::exp)
override fun Tensor<Double>.log(): DoubleTensor = tensor.map(::ln)

View File

@ -1,104 +0,0 @@
/*
* Copyright 2018-2021 KMath contributors.
* Use of this source code is governed by the Apache 2.0 license that can be found in the license/LICENSE.txt file.
*/
package space.kscience.kmath.tensors.core.algebras
import kotlin.math.sqrt
import space.kscience.kmath.tensors.api.*
import space.kscience.kmath.tensors.core.*
import space.kscience.kmath.tensors.core.algebras.DoubleStatisticTensorAlgebra.max
import space.kscience.kmath.tensors.core.algebras.DoubleStatisticTensorAlgebra.mean
import space.kscience.kmath.tensors.core.algebras.DoubleStatisticTensorAlgebra.min
import space.kscience.kmath.tensors.core.algebras.DoubleStatisticTensorAlgebra.sum
import space.kscience.kmath.tensors.core.algebras.DoubleStatisticTensorAlgebra.variance
public object DoubleStatisticTensorAlgebra : StatisticTensorAlgebra<Double>, DoubleTensorAlgebra() {
private fun Tensor<Double>.fold(foldFunction: (DoubleArray) -> Double): Double =
foldFunction(this.tensor.toDoubleArray())
private fun Tensor<Double>.foldDim(
foldFunction: (DoubleArray) -> Double,
dim: Int,
keepDim: Boolean
): DoubleTensor {
check(dim < dimension) { "Dimension $dim out of range $dimension" }
val resShape = if (keepDim) {
shape.take(dim).toIntArray() + intArrayOf(1) + shape.takeLast(dimension - dim - 1).toIntArray()
} else {
shape.take(dim).toIntArray() + shape.takeLast(dimension - dim - 1).toIntArray()
}
val resNumElements = resShape.reduce(Int::times)
val resTensor = DoubleTensor(resShape, DoubleArray(resNumElements) { 0.0 }, 0)
for (index in resTensor.linearStructure.indices()) {
val prefix = index.take(dim).toIntArray()
val suffix = index.takeLast(dimension - dim - 1).toIntArray()
resTensor[index] = foldFunction(DoubleArray(shape[dim]) { i ->
this[prefix + intArrayOf(i) + suffix]
})
}
return resTensor
}
override fun Tensor<Double>.min(): Double = this.fold { it.minOrNull()!! }
override fun Tensor<Double>.min(dim: Int, keepDim: Boolean): DoubleTensor =
foldDim({ x -> x.minOrNull()!! }, dim, keepDim)
override fun Tensor<Double>.max(): Double = this.fold { it.maxOrNull()!! }
override fun Tensor<Double>.max(dim: Int, keepDim: Boolean): DoubleTensor =
foldDim({ x -> x.maxOrNull()!! }, dim, keepDim)
override fun Tensor<Double>.sum(): Double = this.fold { it.sum() }
override fun Tensor<Double>.sum(dim: Int, keepDim: Boolean): DoubleTensor =
foldDim({ x -> x.sum() }, dim, keepDim)
override fun Tensor<Double>.mean(): Double = this.fold { it.sum() / tensor.numElements }
override fun Tensor<Double>.mean(dim: Int, keepDim: Boolean): DoubleTensor =
foldDim(
{ arr ->
check(dim < dimension) { "Dimension $dim out of range $dimension" }
arr.sum() / shape[dim]
},
dim,
keepDim
)
override fun Tensor<Double>.std(): Double = this.fold { arr ->
val mean = arr.sum() / tensor.numElements
sqrt(arr.sumOf { (it - mean) * (it - mean) } / (tensor.numElements - 1))
}
override fun Tensor<Double>.std(dim: Int, keepDim: Boolean): DoubleTensor = foldDim(
{ arr ->
check(dim < dimension) { "Dimension $dim out of range $dimension" }
val mean = arr.sum() / shape[dim]
sqrt(arr.sumOf { (it - mean) * (it - mean) } / (shape[dim] - 1))
},
dim,
keepDim
)
override fun Tensor<Double>.variance(): Double = this.fold { arr ->
val mean = arr.sum() / tensor.numElements
arr.sumOf { (it - mean) * (it - mean) } / (tensor.numElements - 1)
}
override fun Tensor<Double>.variance(dim: Int, keepDim: Boolean): DoubleTensor = foldDim(
{ arr ->
check(dim < dimension) { "Dimension $dim out of range $dimension" }
val mean = arr.sum() / shape[dim]
arr.sumOf { (it - mean) * (it - mean) } / (shape[dim] - 1)
},
dim,
keepDim
)
}

View File

@ -284,7 +284,7 @@ public open class DoubleTensorAlgebra : TensorPartialDivisionAlgebra<Double> {
val m1 = newThis.shape[newThis.shape.size - 1]
val m2 = newOther.shape[newOther.shape.size - 2]
val n = newOther.shape[newOther.shape.size - 1]
check (m1 == m2) {
check(m1 == m2) {
throw RuntimeException("Tensors dot operation dimension mismatch: ($l, $m1) x ($m2, $n)")
}
@ -403,7 +403,7 @@ public open class DoubleTensorAlgebra : TensorPartialDivisionAlgebra<Double> {
public fun stack(tensors: List<DoubleTensor>): DoubleTensor {
val shape = tensors.firstOrNull()?.shape
check(shape != null) { "Collection must have at least 1 element" }
check(tensors.all { it.shape contentEquals shape }) {"Stacking tensors must have same shapes"}
check(tensors.all { it.shape contentEquals shape }) { "Stacking tensors must have same shapes" }
val resShape = intArrayOf(tensors.size) + shape
val resBuffer = tensors.flatMap {
it.tensor.mutableBuffer.array().drop(it.bufferStart).take(it.numElements)
@ -415,4 +415,36 @@ public open class DoubleTensorAlgebra : TensorPartialDivisionAlgebra<Double> {
public fun Tensor<Double>.rowsByIndices(indices: IntArray): DoubleTensor {
return stack(indices.map { this[it] })
}
internal fun Tensor<Double>.fold(foldFunction: (DoubleArray) -> Double): Double =
foldFunction(tensor.toDoubleArray())
internal fun Tensor<Double>.foldDim(
foldFunction: (DoubleArray) -> Double,
dim: Int,
keepDim: Boolean
): DoubleTensor {
check(dim < dimension) { "Dimension $dim out of range $dimension" }
val resShape = if (keepDim) {
shape.take(dim).toIntArray() + intArrayOf(1) + shape.takeLast(dimension - dim - 1).toIntArray()
} else {
shape.take(dim).toIntArray() + shape.takeLast(dimension - dim - 1).toIntArray()
}
val resNumElements = resShape.reduce(Int::times)
val resTensor = DoubleTensor(resShape, DoubleArray(resNumElements) { 0.0 }, 0)
for (index in resTensor.linearStructure.indices()) {
val prefix = index.take(dim).toIntArray()
val suffix = index.takeLast(dimension - dim - 1).toIntArray()
resTensor[index] = foldFunction(DoubleArray(shape[dim]) { i ->
tensor[prefix + intArrayOf(i) + suffix]
})
}
return resTensor
}
override fun Tensor<Double>.sum(): Double = tensor.fold { it.sum() }
override fun Tensor<Double>.sum(dim: Int, keepDim: Boolean): DoubleTensor =
foldDim({ x -> x.sum() }, dim, keepDim)
}