diff --git a/.space.kts b/.space.kts index 9dda0cbf7..d70ad6d59 100644 --- a/.space.kts +++ b/.space.kts @@ -1 +1,3 @@ -job("Build") { gradlew("openjdk:11", "build") } +job("Build") { + gradlew("openjdk:11", "build") +} diff --git a/CHANGELOG.md b/CHANGELOG.md index 214730ecc..0652156b1 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -8,6 +8,11 @@ - Automatic README generation for features (#139) - Native support for `memory`, `core` and `dimensions` - `kmath-ejml` to supply EJML SimpleMatrix wrapper. +- A separate `Symbol` entity, which is used for global unbound symbol. +- A `Symbol` indexing scope. +- Basic optimization API for Commons-math. +- Chi squared optimization for array-like data in CM +- `Fitting` utility object in prob/stat ### Changed - Package changed from `scientifik` to `kscience.kmath`. @@ -16,6 +21,8 @@ - `Polynomial` secondary constructor made function. - Kotlin version: 1.3.72 -> 1.4.20-M1 - `kmath-ast` doesn't depend on heavy `kotlin-reflect` library. +- Full autodiff refactoring based on `Symbol` +- `kmath-prob` renamed to `kmath-stat` ### Deprecated diff --git a/README.md b/README.md index eadc3036b..afab32dcf 100644 --- a/README.md +++ b/README.md @@ -53,9 +53,7 @@ can be used for a wide variety of purposes from high performance calculations to * **Commons-math wrapper** It is planned to gradually wrap most parts of [Apache commons-math](http://commons.apache.org/proper/commons-math/) library in Kotlin code and maybe rewrite some parts to better suit the Kotlin programming paradigm, however there is no fixed roadmap for that. Feel free to submit a feature request if you want something to be done first. - -* **EJML wrapper** Provides EJML `SimpleMatrix` wrapper consistent with the core matrix structures. - + ## Planned features * **Messaging** A mathematical notation to support multi-language and multi-node communication for mathematical tasks. @@ -101,7 +99,7 @@ can be used for a wide variety of purposes from high performance calculations to > - [buffers](kmath-core/src/commonMain/kotlin/kscience/kmath/structures/Buffers.kt) : One-dimensional structure > - [expressions](kmath-core/src/commonMain/kotlin/kscience/kmath/expressions) : Functional Expressions > - [domains](kmath-core/src/commonMain/kotlin/kscience/kmath/domains) : Domains -> - [autodif](kmath-core/src/commonMain/kotlin/kscience/kmath/misc/AutoDiff.kt) : Automatic differentiation +> - [autodif](kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/SimpleAutoDiff.kt) : Automatic differentiation
@@ -117,6 +115,12 @@ can be used for a wide variety of purposes from high performance calculations to > **Maturity**: EXPERIMENTAL
+* ### [kmath-ejml](kmath-ejml) +> +> +> **Maturity**: EXPERIMENTAL +
+ * ### [kmath-for-real](kmath-for-real) > > @@ -147,7 +151,7 @@ can be used for a wide variety of purposes from high performance calculations to > **Maturity**: EXPERIMENTAL
-* ### [kmath-prob](kmath-prob) +* ### [kmath-stat](kmath-stat) > > > **Maturity**: EXPERIMENTAL @@ -178,8 +182,8 @@ repositories{ } dependencies{ - api("kscience.kmath:kmath-core:${kmathVersion}") - //api("scientifik:kmath-core:${kmathVersion}") for 0.1.3 and earlier + api("kscience.kmath:kmath-core:0.2.0-dev-2") + //api("kscience.kmath:kmath-core-jvm:0.2.0-dev-2") for jvm-specific version } ``` @@ -197,4 +201,4 @@ with the same artifact names. ## Contributing -The project requires a lot of additional work. Please feel free to contribute in any way and propose new features. +The project requires a lot of additional work. The most important thing we need is a feedback about what features are required the most. Feel free to open feature issues with requests. We are also welcome to code contributions, especially in issues marked as [waiting for a hero](https://github.com/mipt-npm/kmath/labels/waiting%20for%20a%20hero). \ No newline at end of file diff --git a/build.gradle.kts b/build.gradle.kts index 05e2d5979..acb9f3b68 100644 --- a/build.gradle.kts +++ b/build.gradle.kts @@ -2,7 +2,7 @@ plugins { id("ru.mipt.npm.project") } -val kmathVersion: String by extra("0.2.0-dev-2") +val kmathVersion: String by extra("0.2.0-dev-3") val bintrayRepo: String by extra("kscience") val githubProject: String by extra("kmath") @@ -25,3 +25,7 @@ subprojects { readme { readmeTemplate = file("docs/templates/README-TEMPLATE.md") } + +apiValidation{ + validationDisabled = true +} \ No newline at end of file diff --git a/docs/templates/README-TEMPLATE.md b/docs/templates/README-TEMPLATE.md index f451adb24..5117e0694 100644 --- a/docs/templates/README-TEMPLATE.md +++ b/docs/templates/README-TEMPLATE.md @@ -107,4 +107,4 @@ with the same artifact names. ## Contributing -The project requires a lot of additional work. Please feel free to contribute in any way and propose new features. +The project requires a lot of additional work. The most important thing we need is a feedback about what features are required the most. Feel free to open feature issues with requests. We are also welcome to code contributions, especially in issues marked as [waiting for a hero](https://github.com/mipt-npm/kmath/labels/waiting%20for%20a%20hero). \ No newline at end of file diff --git a/examples/build.gradle.kts b/examples/build.gradle.kts index 900da966b..9ba1ec5be 100644 --- a/examples/build.gradle.kts +++ b/examples/build.gradle.kts @@ -23,7 +23,7 @@ dependencies { implementation(project(":kmath-core")) implementation(project(":kmath-coroutines")) implementation(project(":kmath-commons")) - implementation(project(":kmath-prob")) + implementation(project(":kmath-stat")) implementation(project(":kmath-viktor")) implementation(project(":kmath-dimensions")) implementation(project(":kmath-ejml")) diff --git a/examples/src/main/kotlin/kscience/kmath/commons/prob/DistributionBenchmark.kt b/examples/src/main/kotlin/kscience/kmath/commons/prob/DistributionBenchmark.kt index 9c0a01961..ef554aeff 100644 --- a/examples/src/main/kotlin/kscience/kmath/commons/prob/DistributionBenchmark.kt +++ b/examples/src/main/kotlin/kscience/kmath/commons/prob/DistributionBenchmark.kt @@ -4,7 +4,7 @@ import kotlinx.coroutines.Dispatchers import kotlinx.coroutines.async import kotlinx.coroutines.runBlocking import kscience.kmath.chains.BlockingRealChain -import kscience.kmath.prob.* +import kscience.kmath.stat.* import org.apache.commons.rng.sampling.distribution.ZigguratNormalizedGaussianSampler import org.apache.commons.rng.simple.RandomSource import java.time.Duration diff --git a/examples/src/main/kotlin/kscience/kmath/commons/prob/DistributionDemo.kt b/examples/src/main/kotlin/kscience/kmath/commons/prob/DistributionDemo.kt index 7d53e5178..6146e17af 100644 --- a/examples/src/main/kotlin/kscience/kmath/commons/prob/DistributionDemo.kt +++ b/examples/src/main/kotlin/kscience/kmath/commons/prob/DistributionDemo.kt @@ -3,9 +3,9 @@ package kscience.kmath.commons.prob import kotlinx.coroutines.runBlocking import kscience.kmath.chains.Chain import kscience.kmath.chains.collectWithState -import kscience.kmath.prob.Distribution -import kscience.kmath.prob.RandomGenerator -import kscience.kmath.prob.normal +import kscience.kmath.stat.Distribution +import kscience.kmath.stat.RandomGenerator +import kscience.kmath.stat.normal private data class AveragingChainState(var num: Int = 0, var value: Double = 0.0) diff --git a/examples/src/main/kotlin/kscience/kmath/operations/ComplexDemo.kt b/examples/src/main/kotlin/kscience/kmath/operations/ComplexDemo.kt index 34b3c9981..e84fd8df3 100644 --- a/examples/src/main/kotlin/kscience/kmath/operations/ComplexDemo.kt +++ b/examples/src/main/kotlin/kscience/kmath/operations/ComplexDemo.kt @@ -6,8 +6,8 @@ import kscience.kmath.structures.complex fun main() { // 2d element - val element = NDElement.complex(2, 2) { index: IntArray -> - Complex(index[0].toDouble() - index[1].toDouble(), index[0].toDouble() + index[1].toDouble()) + val element = NDElement.complex(2, 2) { (i,j) -> + Complex(i.toDouble() - j.toDouble(), i.toDouble() + j.toDouble()) } println(element) diff --git a/kmath-ast/src/commonMain/kotlin/kscience/kmath/ast/MstExpression.kt b/kmath-ast/src/commonMain/kotlin/kscience/kmath/ast/MstExpression.kt index 483bc530c..5ca75e993 100644 --- a/kmath-ast/src/commonMain/kotlin/kscience/kmath/ast/MstExpression.kt +++ b/kmath-ast/src/commonMain/kotlin/kscience/kmath/ast/MstExpression.kt @@ -14,8 +14,8 @@ import kotlin.contracts.contract * @author Alexander Nozik */ public class MstExpression(public val algebra: Algebra, public val mst: MST) : Expression { - private inner class InnerAlgebra(val arguments: Map) : NumericAlgebra { - override fun symbol(value: String): T = arguments[value] ?: algebra.symbol(value) + private inner class InnerAlgebra(val arguments: Map) : NumericAlgebra { + override fun symbol(value: String): T = arguments[StringSymbol(value)] ?: algebra.symbol(value) override fun unaryOperation(operation: String, arg: T): T = algebra.unaryOperation(operation, arg) override fun binaryOperation(operation: String, left: T, right: T): T = @@ -27,7 +27,7 @@ public class MstExpression(public val algebra: Algebra, public val mst: MS error("Numeric nodes are not supported by $this") } - override operator fun invoke(arguments: Map): T = InnerAlgebra(arguments).evaluate(mst) + override operator fun invoke(arguments: Map): T = InnerAlgebra(arguments).evaluate(mst) } /** @@ -37,7 +37,7 @@ public class MstExpression(public val algebra: Algebra, public val mst: MS */ public inline fun , E : Algebra> A.mst( mstAlgebra: E, - block: E.() -> MST + block: E.() -> MST, ): MstExpression = MstExpression(this, mstAlgebra.block()) /** @@ -116,7 +116,7 @@ public inline fun > FunctionalExpressionField> FunctionalExpressionExtendedField.mstInExtendedField( - block: MstExtendedField.() -> MST + block: MstExtendedField.() -> MST, ): MstExpression { contract { callsInPlace(block, InvocationKind.EXACTLY_ONCE) } return algebra.mstInExtendedField(block) diff --git a/kmath-ast/src/jvmMain/kotlin/kscience/kmath/asm/internal/AsmBuilder.kt b/kmath-ast/src/jvmMain/kotlin/kscience/kmath/asm/internal/AsmBuilder.kt index 06f02a94d..a1e482103 100644 --- a/kmath-ast/src/jvmMain/kotlin/kscience/kmath/asm/internal/AsmBuilder.kt +++ b/kmath-ast/src/jvmMain/kotlin/kscience/kmath/asm/internal/AsmBuilder.kt @@ -25,7 +25,7 @@ internal class AsmBuilder internal constructor( private val classOfT: Class<*>, private val algebra: Algebra, private val className: String, - private val invokeLabel0Visitor: AsmBuilder.() -> Unit + private val invokeLabel0Visitor: AsmBuilder.() -> Unit, ) { /** * Internal classloader of [AsmBuilder] with alias to define class from byte array. @@ -379,22 +379,14 @@ internal class AsmBuilder internal constructor( * Loads a variable [name] from arguments [Map] parameter of [Expression.invoke]. The [defaultValue] may be * provided. */ - internal fun loadVariable(name: String, defaultValue: T? = null): Unit = invokeMethodVisitor.run { + internal fun loadVariable(name: String): Unit = invokeMethodVisitor.run { load(invokeArgumentsVar, MAP_TYPE) aconst(name) - if (defaultValue != null) - loadTConstant(defaultValue) - invokestatic( MAP_INTRINSICS_TYPE.internalName, "getOrFail", - - Type.getMethodDescriptor( - OBJECT_TYPE, - MAP_TYPE, - OBJECT_TYPE, - *OBJECT_TYPE.wrapToArrayIf { defaultValue != null }), + Type.getMethodDescriptor(OBJECT_TYPE, MAP_TYPE, STRING_TYPE), false ) @@ -429,7 +421,7 @@ internal class AsmBuilder internal constructor( method: String, descriptor: String, expectedArity: Int, - opcode: Int = INVOKEINTERFACE + opcode: Int = INVOKEINTERFACE, ) { run loop@{ repeat(expectedArity) { diff --git a/kmath-ast/src/jvmMain/kotlin/kscience/kmath/asm/internal/mapIntrinsics.kt b/kmath-ast/src/jvmMain/kotlin/kscience/kmath/asm/internal/mapIntrinsics.kt index 708b3c2b4..588b9611a 100644 --- a/kmath-ast/src/jvmMain/kotlin/kscience/kmath/asm/internal/mapIntrinsics.kt +++ b/kmath-ast/src/jvmMain/kotlin/kscience/kmath/asm/internal/mapIntrinsics.kt @@ -2,11 +2,12 @@ package kscience.kmath.asm.internal +import kscience.kmath.expressions.StringSymbol +import kscience.kmath.expressions.Symbol + /** - * Gets value with given [key] or throws [IllegalStateException] whenever it is not present. + * Gets value with given [key] or throws [NoSuchElementException] whenever it is not present. * * @author Iaroslav Postovalov */ -@JvmOverloads -internal fun Map.getOrFail(key: K, default: V? = null): V = - this[key] ?: default ?: error("Parameter not found: $key") +internal fun Map.getOrFail(key: String): V = getValue(StringSymbol(key)) diff --git a/kmath-ast/src/jvmTest/kotlin/kscience/kmath/asm/TestAsmAlgebras.kt b/kmath-ast/src/jvmTest/kotlin/kscience/kmath/asm/TestAsmAlgebras.kt index 0cf1307d1..5eebfe43d 100644 --- a/kmath-ast/src/jvmTest/kotlin/kscience/kmath/asm/TestAsmAlgebras.kt +++ b/kmath-ast/src/jvmTest/kotlin/kscience/kmath/asm/TestAsmAlgebras.kt @@ -1,6 +1,5 @@ package kscience.kmath.asm -import kscience.kmath.asm.compile import kscience.kmath.ast.mstInField import kscience.kmath.ast.mstInRing import kscience.kmath.ast.mstInSpace @@ -11,6 +10,7 @@ import kotlin.test.Test import kotlin.test.assertEquals internal class TestAsmAlgebras { + @Test fun space() { val res1 = ByteRing.mstInSpace { diff --git a/kmath-commons/build.gradle.kts b/kmath-commons/build.gradle.kts index ed6452ad8..6a44c92f2 100644 --- a/kmath-commons/build.gradle.kts +++ b/kmath-commons/build.gradle.kts @@ -6,7 +6,7 @@ description = "Commons math binding for kmath" dependencies { api(project(":kmath-core")) api(project(":kmath-coroutines")) - api(project(":kmath-prob")) -// api(project(":kmath-functions")) + api(project(":kmath-stat")) + api(project(":kmath-functions")) api("org.apache.commons:commons-math3:3.6.1") } diff --git a/kmath-commons/src/main/kotlin/kscience/kmath/commons/expressions/DerivativeStructureExpression.kt b/kmath-commons/src/main/kotlin/kscience/kmath/commons/expressions/DerivativeStructureExpression.kt new file mode 100644 index 000000000..c593f5103 --- /dev/null +++ b/kmath-commons/src/main/kotlin/kscience/kmath/commons/expressions/DerivativeStructureExpression.kt @@ -0,0 +1,115 @@ +package kscience.kmath.commons.expressions + +import kscience.kmath.expressions.* +import kscience.kmath.operations.ExtendedField +import org.apache.commons.math3.analysis.differentiation.DerivativeStructure + +/** + * A field over commons-math [DerivativeStructure]. + * + * @property order The derivation order. + * @property bindings The map of bindings values. All bindings are considered free parameters + */ +public class DerivativeStructureField( + public val order: Int, + private val bindings: Map +) : ExtendedField, ExpressionAlgebra { + public override val zero: DerivativeStructure by lazy { DerivativeStructure(bindings.size, order) } + public override val one: DerivativeStructure by lazy { DerivativeStructure(bindings.size, order, 1.0) } + + /** + * A class that implements both [DerivativeStructure] and a [Symbol] + */ + public inner class DerivativeStructureSymbol(symbol: Symbol, value: Double) : + DerivativeStructure(bindings.size, order, bindings.keys.indexOf(symbol), value), Symbol { + override val identity: String = symbol.identity + override fun toString(): String = identity + override fun equals(other: Any?): Boolean = this.identity == (other as? Symbol)?.identity + override fun hashCode(): Int = identity.hashCode() + } + + /** + * Identity-based symbol bindings map + */ + private val variables: Map = bindings.entries.associate { (key, value) -> + key.identity to DerivativeStructureSymbol(key, value) + } + + override fun const(value: Double): DerivativeStructure = DerivativeStructure(bindings.size, order, value) + + public override fun bindOrNull(symbol: Symbol): DerivativeStructureSymbol? = variables[symbol.identity] + + public fun bind(symbol: Symbol): DerivativeStructureSymbol = variables.getValue(symbol.identity) + + //public fun Number.const(): DerivativeStructure = const(toDouble()) + + public fun DerivativeStructure.derivative(parameter: Symbol, order: Int = 1): Double { + return derivative(mapOf(parameter to order)) + } + + public fun DerivativeStructure.derivative(orders: Map): Double { + return getPartialDerivative(*bindings.keys.map { orders[it] ?: 0 }.toIntArray()) + } + + public fun DerivativeStructure.derivative(vararg orders: Pair): Double = derivative(mapOf(*orders)) + public override fun add(a: DerivativeStructure, b: DerivativeStructure): DerivativeStructure = a.add(b) + + public override fun multiply(a: DerivativeStructure, k: Number): DerivativeStructure = when (k) { + is Double -> a.multiply(k) + is Int -> a.multiply(k) + else -> a.multiply(k.toDouble()) + } + + public override fun multiply(a: DerivativeStructure, b: DerivativeStructure): DerivativeStructure = a.multiply(b) + public override fun divide(a: DerivativeStructure, b: DerivativeStructure): DerivativeStructure = a.divide(b) + public override fun sin(arg: DerivativeStructure): DerivativeStructure = arg.sin() + public override fun cos(arg: DerivativeStructure): DerivativeStructure = arg.cos() + public override fun tan(arg: DerivativeStructure): DerivativeStructure = arg.tan() + public override fun asin(arg: DerivativeStructure): DerivativeStructure = arg.asin() + public override fun acos(arg: DerivativeStructure): DerivativeStructure = arg.acos() + public override fun atan(arg: DerivativeStructure): DerivativeStructure = arg.atan() + public override fun sinh(arg: DerivativeStructure): DerivativeStructure = arg.sinh() + public override fun cosh(arg: DerivativeStructure): DerivativeStructure = arg.cosh() + public override fun tanh(arg: DerivativeStructure): DerivativeStructure = arg.tanh() + public override fun asinh(arg: DerivativeStructure): DerivativeStructure = arg.asinh() + public override fun acosh(arg: DerivativeStructure): DerivativeStructure = arg.acosh() + public override fun atanh(arg: DerivativeStructure): DerivativeStructure = arg.atanh() + + public override fun power(arg: DerivativeStructure, pow: Number): DerivativeStructure = when (pow) { + is Double -> arg.pow(pow) + is Int -> arg.pow(pow) + else -> arg.pow(pow.toDouble()) + } + + public fun power(arg: DerivativeStructure, pow: DerivativeStructure): DerivativeStructure = arg.pow(pow) + public override fun exp(arg: DerivativeStructure): DerivativeStructure = arg.exp() + public override fun ln(arg: DerivativeStructure): DerivativeStructure = arg.log() + + public override operator fun DerivativeStructure.plus(b: Number): DerivativeStructure = add(b.toDouble()) + public override operator fun DerivativeStructure.minus(b: Number): DerivativeStructure = subtract(b.toDouble()) + public override operator fun Number.plus(b: DerivativeStructure): DerivativeStructure = b + this + public override operator fun Number.minus(b: DerivativeStructure): DerivativeStructure = b - this + + public companion object : AutoDiffProcessor { + override fun process(function: DerivativeStructureField.() -> DerivativeStructure): DifferentiableExpression { + return DerivativeStructureExpression(function) + } + } +} + +/** + * A constructs that creates a derivative structure with required order on-demand + */ +public class DerivativeStructureExpression( + public val function: DerivativeStructureField.() -> DerivativeStructure, +) : DifferentiableExpression { + public override operator fun invoke(arguments: Map): Double = + DerivativeStructureField(0, arguments).function().value + + /** + * Get the derivative expression with given orders + */ + public override fun derivativeOrNull(orders: Map): Expression = Expression { arguments -> + with(DerivativeStructureField(orders.values.maxOrNull() ?: 0, arguments)) { function().derivative(orders) } + } +} diff --git a/kmath-commons/src/main/kotlin/kscience/kmath/commons/expressions/DiffExpression.kt b/kmath-commons/src/main/kotlin/kscience/kmath/commons/expressions/DiffExpression.kt deleted file mode 100644 index c39f0d04c..000000000 --- a/kmath-commons/src/main/kotlin/kscience/kmath/commons/expressions/DiffExpression.kt +++ /dev/null @@ -1,131 +0,0 @@ -package kscience.kmath.commons.expressions - -import kscience.kmath.expressions.Expression -import kscience.kmath.expressions.ExpressionAlgebra -import kscience.kmath.operations.ExtendedField -import kscience.kmath.operations.Field -import kscience.kmath.operations.invoke -import org.apache.commons.math3.analysis.differentiation.DerivativeStructure -import kotlin.properties.ReadOnlyProperty - -/** - * A field over commons-math [DerivativeStructure]. - * - * @property order The derivation order. - * @property parameters The map of free parameters. - */ -public class DerivativeStructureField( - public val order: Int, - public val parameters: Map -) : ExtendedField { - public override val zero: DerivativeStructure by lazy { DerivativeStructure(parameters.size, order) } - public override val one: DerivativeStructure by lazy { DerivativeStructure(parameters.size, order, 1.0) } - - private val variables: Map = parameters.mapValues { (key, value) -> - DerivativeStructure(parameters.size, order, parameters.keys.indexOf(key), value) - } - - public val variable: ReadOnlyProperty = ReadOnlyProperty { _, property -> - variables[property.name] ?: error("A variable with name ${property.name} does not exist") - } - - public fun variable(name: String, default: DerivativeStructure? = null): DerivativeStructure = - variables[name] ?: default ?: error("A variable with name $name does not exist") - - public fun Number.const(): DerivativeStructure = DerivativeStructure(order, parameters.size, toDouble()) - - public fun DerivativeStructure.deriv(parName: String, order: Int = 1): Double { - return deriv(mapOf(parName to order)) - } - - public fun DerivativeStructure.deriv(orders: Map): Double { - return getPartialDerivative(*parameters.keys.map { orders[it] ?: 0 }.toIntArray()) - } - - public fun DerivativeStructure.deriv(vararg orders: Pair): Double = deriv(mapOf(*orders)) - public override fun add(a: DerivativeStructure, b: DerivativeStructure): DerivativeStructure = a.add(b) - - public override fun multiply(a: DerivativeStructure, k: Number): DerivativeStructure = when (k) { - is Double -> a.multiply(k) - is Int -> a.multiply(k) - else -> a.multiply(k.toDouble()) - } - - public override fun multiply(a: DerivativeStructure, b: DerivativeStructure): DerivativeStructure = a.multiply(b) - public override fun divide(a: DerivativeStructure, b: DerivativeStructure): DerivativeStructure = a.divide(b) - public override fun sin(arg: DerivativeStructure): DerivativeStructure = arg.sin() - public override fun cos(arg: DerivativeStructure): DerivativeStructure = arg.cos() - public override fun tan(arg: DerivativeStructure): DerivativeStructure = arg.tan() - public override fun asin(arg: DerivativeStructure): DerivativeStructure = arg.asin() - public override fun acos(arg: DerivativeStructure): DerivativeStructure = arg.acos() - public override fun atan(arg: DerivativeStructure): DerivativeStructure = arg.atan() - public override fun sinh(arg: DerivativeStructure): DerivativeStructure = arg.sinh() - public override fun cosh(arg: DerivativeStructure): DerivativeStructure = arg.cosh() - public override fun tanh(arg: DerivativeStructure): DerivativeStructure = arg.tanh() - public override fun asinh(arg: DerivativeStructure): DerivativeStructure = arg.asinh() - public override fun acosh(arg: DerivativeStructure): DerivativeStructure = arg.acosh() - public override fun atanh(arg: DerivativeStructure): DerivativeStructure = arg.atanh() - - public override fun power(arg: DerivativeStructure, pow: Number): DerivativeStructure = when (pow) { - is Double -> arg.pow(pow) - is Int -> arg.pow(pow) - else -> arg.pow(pow.toDouble()) - } - - public fun power(arg: DerivativeStructure, pow: DerivativeStructure): DerivativeStructure = arg.pow(pow) - public override fun exp(arg: DerivativeStructure): DerivativeStructure = arg.exp() - public override fun ln(arg: DerivativeStructure): DerivativeStructure = arg.log() - - public override operator fun DerivativeStructure.plus(b: Number): DerivativeStructure = add(b.toDouble()) - public override operator fun DerivativeStructure.minus(b: Number): DerivativeStructure = subtract(b.toDouble()) - public override operator fun Number.plus(b: DerivativeStructure): DerivativeStructure = b + this - public override operator fun Number.minus(b: DerivativeStructure): DerivativeStructure = b - this -} - -/** - * A constructs that creates a derivative structure with required order on-demand - */ -public class DiffExpression(public val function: DerivativeStructureField.() -> DerivativeStructure) : - Expression { - public override operator fun invoke(arguments: Map): Double = DerivativeStructureField( - 0, - arguments - ).function().value - - /** - * Get the derivative expression with given orders - * TODO make result [DiffExpression] - */ - public fun derivative(orders: Map): Expression = Expression { arguments -> - (DerivativeStructureField(orders.values.maxOrNull() ?: 0, arguments)) { function().deriv(orders) } - } - - //TODO add gradient and maybe other vector operators -} - -public fun DiffExpression.derivative(vararg orders: Pair): Expression = derivative(mapOf(*orders)) -public fun DiffExpression.derivative(name: String): Expression = derivative(name to 1) - -/** - * A context for [DiffExpression] (not to be confused with [DerivativeStructure]) - */ -public object DiffExpressionAlgebra : ExpressionAlgebra, Field { - public override val zero: DiffExpression = DiffExpression { 0.0.const() } - public override val one: DiffExpression = DiffExpression { 1.0.const() } - - public override fun variable(name: String, default: Double?): DiffExpression = - DiffExpression { variable(name, default?.const()) } - - public override fun const(value: Double): DiffExpression = DiffExpression { value.const() } - - public override fun add(a: DiffExpression, b: DiffExpression): DiffExpression = - DiffExpression { a.function(this) + b.function(this) } - - public override fun multiply(a: DiffExpression, k: Number): DiffExpression = DiffExpression { a.function(this) * k } - - public override fun multiply(a: DiffExpression, b: DiffExpression): DiffExpression = - DiffExpression { a.function(this) * b.function(this) } - - public override fun divide(a: DiffExpression, b: DiffExpression): DiffExpression = - DiffExpression { a.function(this) / b.function(this) } -} diff --git a/kmath-commons/src/main/kotlin/kscience/kmath/commons/optimization/CMOptimizationProblem.kt b/kmath-commons/src/main/kotlin/kscience/kmath/commons/optimization/CMOptimizationProblem.kt new file mode 100644 index 000000000..13f9af7bb --- /dev/null +++ b/kmath-commons/src/main/kotlin/kscience/kmath/commons/optimization/CMOptimizationProblem.kt @@ -0,0 +1,111 @@ +package kscience.kmath.commons.optimization + +import kscience.kmath.expressions.* +import kscience.kmath.stat.OptimizationFeature +import kscience.kmath.stat.OptimizationProblem +import kscience.kmath.stat.OptimizationProblemFactory +import kscience.kmath.stat.OptimizationResult +import org.apache.commons.math3.optim.* +import org.apache.commons.math3.optim.nonlinear.scalar.GoalType +import org.apache.commons.math3.optim.nonlinear.scalar.MultivariateOptimizer +import org.apache.commons.math3.optim.nonlinear.scalar.ObjectiveFunction +import org.apache.commons.math3.optim.nonlinear.scalar.ObjectiveFunctionGradient +import org.apache.commons.math3.optim.nonlinear.scalar.gradient.NonLinearConjugateGradientOptimizer +import org.apache.commons.math3.optim.nonlinear.scalar.noderiv.AbstractSimplex +import org.apache.commons.math3.optim.nonlinear.scalar.noderiv.NelderMeadSimplex +import org.apache.commons.math3.optim.nonlinear.scalar.noderiv.SimplexOptimizer +import kotlin.reflect.KClass + +public operator fun PointValuePair.component1(): DoubleArray = point +public operator fun PointValuePair.component2(): Double = value + +public class CMOptimizationProblem( + override val symbols: List, +) : OptimizationProblem, SymbolIndexer, OptimizationFeature { + private val optimizationData: HashMap, OptimizationData> = HashMap() + private var optimizatorBuilder: (() -> MultivariateOptimizer)? = null + public var convergenceChecker: ConvergenceChecker = SimpleValueChecker(DEFAULT_RELATIVE_TOLERANCE, + DEFAULT_ABSOLUTE_TOLERANCE, DEFAULT_MAX_ITER) + + public fun addOptimizationData(data: OptimizationData) { + optimizationData[data::class] = data + } + + init { + addOptimizationData(MaxEval.unlimited()) + } + + public fun exportOptimizationData(): List = optimizationData.values.toList() + + public override fun initialGuess(map: Map): Unit { + addOptimizationData(InitialGuess(map.toDoubleArray())) + } + + public override fun expression(expression: Expression): Unit { + val objectiveFunction = ObjectiveFunction { + val args = it.toMap() + expression(args) + } + addOptimizationData(objectiveFunction) + } + + public override fun diffExpression(expression: DifferentiableExpression): Unit { + expression(expression) + val gradientFunction = ObjectiveFunctionGradient { + val args = it.toMap() + DoubleArray(symbols.size) { index -> + expression.derivative(symbols[index])(args) + } + } + addOptimizationData(gradientFunction) + if (optimizatorBuilder == null) { + optimizatorBuilder = { + NonLinearConjugateGradientOptimizer( + NonLinearConjugateGradientOptimizer.Formula.FLETCHER_REEVES, + convergenceChecker + ) + } + } + } + + public fun simplex(simplex: AbstractSimplex) { + addOptimizationData(simplex) + //Set optimization builder to simplex if it is not present + if (optimizatorBuilder == null) { + optimizatorBuilder = { SimplexOptimizer(convergenceChecker) } + } + } + + public fun simplexSteps(steps: Map) { + simplex(NelderMeadSimplex(steps.toDoubleArray())) + } + + public fun goal(goalType: GoalType) { + addOptimizationData(goalType) + } + + public fun optimizer(block: () -> MultivariateOptimizer) { + optimizatorBuilder = block + } + + override fun update(result: OptimizationResult) { + initialGuess(result.point) + } + + override fun optimize(): OptimizationResult { + val optimizer = optimizatorBuilder?.invoke() ?: error("Optimizer not defined") + val (point, value) = optimizer.optimize(*optimizationData.values.toTypedArray()) + return OptimizationResult(point.toMap(), value, setOf(this)) + } + + public companion object : OptimizationProblemFactory { + public const val DEFAULT_RELATIVE_TOLERANCE: Double = 1e-4 + public const val DEFAULT_ABSOLUTE_TOLERANCE: Double = 1e-4 + public const val DEFAULT_MAX_ITER: Int = 1000 + + override fun build(symbols: List): CMOptimizationProblem = CMOptimizationProblem(symbols) + } +} + +public fun CMOptimizationProblem.initialGuess(vararg pairs: Pair): Unit = initialGuess(pairs.toMap()) +public fun CMOptimizationProblem.simplexSteps(vararg pairs: Pair): Unit = simplexSteps(pairs.toMap()) diff --git a/kmath-commons/src/main/kotlin/kscience/kmath/commons/optimization/cmFit.kt b/kmath-commons/src/main/kotlin/kscience/kmath/commons/optimization/cmFit.kt new file mode 100644 index 000000000..42475db6c --- /dev/null +++ b/kmath-commons/src/main/kotlin/kscience/kmath/commons/optimization/cmFit.kt @@ -0,0 +1,70 @@ +package kscience.kmath.commons.optimization + +import kscience.kmath.commons.expressions.DerivativeStructureField +import kscience.kmath.expressions.DifferentiableExpression +import kscience.kmath.expressions.Expression +import kscience.kmath.expressions.Symbol +import kscience.kmath.stat.Fitting +import kscience.kmath.stat.OptimizationResult +import kscience.kmath.stat.optimizeWith +import kscience.kmath.structures.Buffer +import kscience.kmath.structures.asBuffer +import org.apache.commons.math3.analysis.differentiation.DerivativeStructure +import org.apache.commons.math3.optim.nonlinear.scalar.GoalType + + +/** + * Generate a chi squared expression from given x-y-sigma data and inline model. Provides automatic differentiation + */ +public fun Fitting.chiSquared( + x: Buffer, + y: Buffer, + yErr: Buffer, + model: DerivativeStructureField.(x: DerivativeStructure) -> DerivativeStructure, +): DifferentiableExpression = chiSquared(DerivativeStructureField, x, y, yErr, model) + +/** + * Generate a chi squared expression from given x-y-sigma data and inline model. Provides automatic differentiation + */ +public fun Fitting.chiSquared( + x: Iterable, + y: Iterable, + yErr: Iterable, + model: DerivativeStructureField.(x: DerivativeStructure) -> DerivativeStructure, +): DifferentiableExpression = chiSquared( + DerivativeStructureField, + x.toList().asBuffer(), + y.toList().asBuffer(), + yErr.toList().asBuffer(), + model +) + + +/** + * Optimize expression without derivatives + */ +public fun Expression.optimize( + vararg symbols: Symbol, + configuration: CMOptimizationProblem.() -> Unit, +): OptimizationResult = optimizeWith(CMOptimizationProblem, symbols = symbols, configuration) + + +/** + * Optimize differentiable expression + */ +public fun DifferentiableExpression.optimize( + vararg symbols: Symbol, + configuration: CMOptimizationProblem.() -> Unit, +): OptimizationResult = optimizeWith(CMOptimizationProblem, symbols = symbols, configuration) + +public fun DifferentiableExpression.minimize( + vararg startPoint: Pair, + configuration: CMOptimizationProblem.() -> Unit = {}, +): OptimizationResult { + require(startPoint.isNotEmpty()) { "Must provide a list of symbols for optimization" } + val problem = CMOptimizationProblem(startPoint.map { it.first }).apply(configuration) + problem.diffExpression(this) + problem.initialGuess(startPoint.toMap()) + problem.goal(GoalType.MINIMIZE) + return problem.optimize() +} \ No newline at end of file diff --git a/kmath-commons/src/main/kotlin/kscience/kmath/commons/random/CMRandomGeneratorWrapper.kt b/kmath-commons/src/main/kotlin/kscience/kmath/commons/random/CMRandomGeneratorWrapper.kt index 58609deae..1eab5f2bd 100644 --- a/kmath-commons/src/main/kotlin/kscience/kmath/commons/random/CMRandomGeneratorWrapper.kt +++ b/kmath-commons/src/main/kotlin/kscience/kmath/commons/random/CMRandomGeneratorWrapper.kt @@ -1,9 +1,10 @@ package kscience.kmath.commons.random -import kscience.kmath.prob.RandomGenerator +import kscience.kmath.stat.RandomGenerator -public class CMRandomGeneratorWrapper(public val factory: (IntArray) -> RandomGenerator) : - org.apache.commons.math3.random.RandomGenerator { +public class CMRandomGeneratorWrapper( + public val factory: (IntArray) -> RandomGenerator, +) : org.apache.commons.math3.random.RandomGenerator { private var generator: RandomGenerator = factory(intArrayOf()) public override fun nextBoolean(): Boolean = generator.nextBoolean() diff --git a/kmath-commons/src/test/kotlin/kscience/kmath/commons/expressions/AutoDiffTest.kt b/kmath-commons/src/test/kotlin/kscience/kmath/commons/expressions/DerivativeStructureExpressionTest.kt similarity index 51% rename from kmath-commons/src/test/kotlin/kscience/kmath/commons/expressions/AutoDiffTest.kt rename to kmath-commons/src/test/kotlin/kscience/kmath/commons/expressions/DerivativeStructureExpressionTest.kt index f905e6818..8886e123f 100644 --- a/kmath-commons/src/test/kotlin/kscience/kmath/commons/expressions/AutoDiffTest.kt +++ b/kmath-commons/src/test/kotlin/kscience/kmath/commons/expressions/DerivativeStructureExpressionTest.kt @@ -1,6 +1,6 @@ package kscience.kmath.commons.expressions -import kscience.kmath.expressions.invoke +import kscience.kmath.expressions.* import kotlin.contracts.InvocationKind import kotlin.contracts.contract import kotlin.test.Test @@ -8,33 +8,37 @@ import kotlin.test.assertEquals internal inline fun diff( order: Int, - vararg parameters: Pair, - block: DerivativeStructureField.() -> R + vararg parameters: Pair, + block: DerivativeStructureField.() -> R, ): R { contract { callsInPlace(block, InvocationKind.EXACTLY_ONCE) } return DerivativeStructureField(order, mapOf(*parameters)).run(block) } internal class AutoDiffTest { + private val x by symbol + private val y by symbol + @Test fun derivativeStructureFieldTest() { - val res = diff(3, "x" to 1.0, "y" to 1.0) { - val x by variable - val y = variable("y") + val res: Double = diff(3, x to 1.0, y to 1.0) { + val x = bind(x)//by binding() + val y = symbol("y") val z = x * (-sin(x * y) + y) - z.deriv("x") + z.derivative(x) } + println(res) } @Test fun autoDifTest() { - val f = DiffExpression { - val x by variable - val y by variable + val f = DerivativeStructureExpression { + val x by binding() + val y by binding() x.pow(2) + 2 * x * y + y.pow(2) + 1 } - assertEquals(10.0, f("x" to 1.0, "y" to 2.0)) - assertEquals(6.0, f.derivative("x")("x" to 1.0, "y" to 2.0)) + assertEquals(10.0, f(x to 1.0, y to 2.0)) + assertEquals(6.0, f.derivative(x)(x to 1.0, y to 2.0)) } } diff --git a/kmath-commons/src/test/kotlin/kscience/kmath/commons/optimization/OptimizeTest.kt b/kmath-commons/src/test/kotlin/kscience/kmath/commons/optimization/OptimizeTest.kt new file mode 100644 index 000000000..4384a5124 --- /dev/null +++ b/kmath-commons/src/test/kotlin/kscience/kmath/commons/optimization/OptimizeTest.kt @@ -0,0 +1,66 @@ +package kscience.kmath.commons.optimization + +import kscience.kmath.commons.expressions.DerivativeStructureExpression +import kscience.kmath.expressions.symbol +import kscience.kmath.stat.Distribution +import kscience.kmath.stat.Fitting +import kscience.kmath.stat.RandomGenerator +import kscience.kmath.stat.normal +import kscience.kmath.structures.asBuffer +import org.junit.jupiter.api.Test +import kotlin.math.pow + +internal class OptimizeTest { + val x by symbol + val y by symbol + + val normal = DerivativeStructureExpression { + exp(-bind(x).pow(2) / 2) + exp(-bind(y).pow(2) / 2) + } + + @Test + fun testGradientOptimization() { + val result = normal.optimize(x, y) { + initialGuess(x to 1.0, y to 1.0) + //no need to select optimizer. Gradient optimizer is used by default because gradients are provided by function + } + println(result.point) + println(result.value) + } + + @Test + fun testSimplexOptimization() { + val result = normal.optimize(x, y) { + initialGuess(x to 1.0, y to 1.0) + simplexSteps(x to 2.0, y to 0.5) + //this sets simplex optimizer + } + println(result.point) + println(result.value) + } + + @Test + fun testCmFit() { + val a by symbol + val b by symbol + val c by symbol + + val sigma = 1.0 + val generator = Distribution.normal(0.0, sigma) + val chain = generator.sample(RandomGenerator.default(112667)) + val x = (1..100).map { it.toDouble() } + val y = x.map { it -> + it.pow(2) + it + 1 + chain.nextDouble() + } + val yErr = x.map { sigma } + val chi2 = Fitting.chiSquared(x.asBuffer(), y.asBuffer(), yErr.asBuffer()) { x -> + val cWithDefault = bindOrNull(c) ?: one + bind(a) * x.pow(2) + bind(b) * x + cWithDefault + } + + val result = chi2.minimize(a to 1.5, b to 0.9, c to 1.0) + println(result) + println("Chi2/dof = ${result.value / (x.size - 3)}") + } + +} \ No newline at end of file diff --git a/kmath-core/README.md b/kmath-core/README.md index 2cf7ed5dc..5501b1d7a 100644 --- a/kmath-core/README.md +++ b/kmath-core/README.md @@ -7,12 +7,12 @@ The core features of KMath: - [buffers](src/commonMain/kotlin/kscience/kmath/structures/Buffers.kt) : One-dimensional structure - [expressions](src/commonMain/kotlin/kscience/kmath/expressions) : Functional Expressions - [domains](src/commonMain/kotlin/kscience/kmath/domains) : Domains - - [autodif](src/commonMain/kotlin/kscience/kmath/misc/AutoDiff.kt) : Automatic differentiation + - [autodif](src/commonMain/kotlin/kscience/kmath/expressions/SimpleAutoDiff.kt) : Automatic differentiation > #### Artifact: > -> This module artifact: `kscience.kmath:kmath-core:0.2.0-dev-1`. +> This module artifact: `kscience.kmath:kmath-core:0.2.0-dev-2`. > > Bintray release version: [ ![Download](https://api.bintray.com/packages/mipt-npm/kscience/kmath-core/images/download.svg) ](https://bintray.com/mipt-npm/kscience/kmath-core/_latestVersion) > @@ -22,25 +22,28 @@ The core features of KMath: > > ```gradle > repositories { +> maven { url "https://dl.bintray.com/kotlin/kotlin-eap" } > maven { url 'https://dl.bintray.com/mipt-npm/kscience' } > maven { url 'https://dl.bintray.com/mipt-npm/dev' } > maven { url 'https://dl.bintray.com/hotkeytlt/maven' } + > } > > dependencies { -> implementation 'kscience.kmath:kmath-core:0.2.0-dev-1' +> implementation 'kscience.kmath:kmath-core:0.2.0-dev-2' > } > ``` > **Gradle Kotlin DSL:** > > ```kotlin > repositories { +> maven("https://dl.bintray.com/kotlin/kotlin-eap") > maven("https://dl.bintray.com/mipt-npm/kscience") > maven("https://dl.bintray.com/mipt-npm/dev") > maven("https://dl.bintray.com/hotkeytlt/maven") > } > > dependencies { -> implementation("kscience.kmath:kmath-core:0.2.0-dev-1") +> implementation("kscience.kmath:kmath-core:0.2.0-dev-2") > } > ``` diff --git a/kmath-core/build.gradle.kts b/kmath-core/build.gradle.kts index b56151abe..b0849eca5 100644 --- a/kmath-core/build.gradle.kts +++ b/kmath-core/build.gradle.kts @@ -41,6 +41,6 @@ readme { feature( id = "autodif", description = "Automatic differentiation", - ref = "src/commonMain/kotlin/kscience/kmath/misc/AutoDiff.kt" + ref = "src/commonMain/kotlin/kscience/kmath/expressions/SimpleAutoDiff.kt" ) } \ No newline at end of file diff --git a/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/DifferentiableExpression.kt b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/DifferentiableExpression.kt new file mode 100644 index 000000000..4fe73f283 --- /dev/null +++ b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/DifferentiableExpression.kt @@ -0,0 +1,39 @@ +package kscience.kmath.expressions + +/** + * An expression that provides derivatives + */ +public interface DifferentiableExpression : Expression{ + public fun derivativeOrNull(orders: Map): Expression? +} + +public fun DifferentiableExpression.derivative(orders: Map): Expression = + derivativeOrNull(orders) ?: error("Derivative with orders $orders not provided") + +public fun DifferentiableExpression.derivative(vararg orders: Pair): Expression = + derivative(mapOf(*orders)) + +public fun DifferentiableExpression.derivative(symbol: Symbol): Expression = derivative(symbol to 1) + +public fun DifferentiableExpression.derivative(name: String): Expression = + derivative(StringSymbol(name) to 1) + +/** + * A [DifferentiableExpression] that defines only first derivatives + */ +public abstract class FirstDerivativeExpression : DifferentiableExpression { + + public abstract fun derivativeOrNull(symbol: Symbol): Expression? + + public override fun derivativeOrNull(orders: Map): Expression? { + val dSymbol = orders.entries.singleOrNull { it.value == 1 }?.key ?: return null + return derivativeOrNull(dSymbol) + } +} + +/** + * A factory that converts an expression in autodiff variables to a [DifferentiableExpression] + */ +public interface AutoDiffProcessor> { + public fun process(function: A.() -> I): DifferentiableExpression +} \ No newline at end of file diff --git a/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/Expression.kt b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/Expression.kt index 5ade9e3ca..ab9ff0e72 100644 --- a/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/Expression.kt +++ b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/Expression.kt @@ -1,6 +1,25 @@ package kscience.kmath.expressions import kscience.kmath.operations.Algebra +import kotlin.jvm.JvmName +import kotlin.properties.ReadOnlyProperty + +/** + * A marker interface for a symbol. A symbol mus have an identity + */ +public interface Symbol { + /** + * Identity object for the symbol. Two symbols with the same identity are considered to be the same symbol. + */ + public val identity: String +} + +/** + * A [Symbol] with a [String] identity + */ +public inline class StringSymbol(override val identity: String) : Symbol { + override fun toString(): String = identity +} /** * An elementary function that could be invoked on a map of arguments @@ -12,30 +31,69 @@ public fun interface Expression { * @param arguments the map of arguments. * @return the value. */ - public operator fun invoke(arguments: Map): T - - public companion object + public operator fun invoke(arguments: Map): T } +/** + * Invoke an expression without parameters + */ +public operator fun Expression.invoke(): T = invoke(emptyMap()) +//This method exists to avoid resolution ambiguity of vararg methods + /** * Calls this expression from arguments. * * @param pairs the pair of arguments' names to values. * @return the value. */ -public operator fun Expression.invoke(vararg pairs: Pair): T = invoke(mapOf(*pairs)) +@JvmName("callBySymbol") +public operator fun Expression.invoke(vararg pairs: Pair): T = invoke(mapOf(*pairs)) + +@JvmName("callByString") +public operator fun Expression.invoke(vararg pairs: Pair): T = + invoke(mapOf(*pairs).mapKeys { StringSymbol(it.key) }) + /** * A context for expression construction + * + * @param T type of the constants for the expression + * @param E type of the actual expression state */ -public interface ExpressionAlgebra : Algebra { +public interface ExpressionAlgebra : Algebra { + /** - * Introduce a variable into expression context + * Bind a given [Symbol] to this context variable and produce context-specific object. Return null if symbol could not be bound in current context. */ - public fun variable(name: String, default: T? = null): E + public fun bindOrNull(symbol: Symbol): E? + + /** + * Bind a string to a context using [StringSymbol] + */ + override fun symbol(value: String): E = bind(StringSymbol(value)) /** * A constant expression which does not depend on arguments */ public fun const(value: T): E } + +/** + * Bind a given [Symbol] to this context variable and produce context-specific object. + */ +public fun ExpressionAlgebra.bind(symbol: Symbol): E = + bindOrNull(symbol) ?: error("Symbol $symbol could not be bound to $this") + +/** + * A delegate to create a symbol with a string identity in this scope + */ +public val symbol: ReadOnlyProperty = ReadOnlyProperty { thisRef, property -> + StringSymbol(property.name) +} + +/** + * Bind a symbol by name inside the [ExpressionAlgebra] + */ +public fun ExpressionAlgebra.binding(): ReadOnlyProperty = ReadOnlyProperty { _, property -> + bind(StringSymbol(property.name)) ?: error("A variable with name ${property.name} does not exist") +} \ No newline at end of file diff --git a/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/FunctionalExpressionAlgebra.kt b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/FunctionalExpressionAlgebra.kt index 5b050dd36..0630e8e4b 100644 --- a/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/FunctionalExpressionAlgebra.kt +++ b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/FunctionalExpressionAlgebra.kt @@ -2,67 +2,43 @@ package kscience.kmath.expressions import kscience.kmath.operations.* -internal class FunctionalUnaryOperation(val context: Algebra, val name: String, private val expr: Expression) : - Expression { - override operator fun invoke(arguments: Map): T = - context.unaryOperation(name, expr.invoke(arguments)) -} - -internal class FunctionalBinaryOperation( - val context: Algebra, - val name: String, - val first: Expression, - val second: Expression -) : Expression { - override operator fun invoke(arguments: Map): T = - context.binaryOperation(name, first.invoke(arguments), second.invoke(arguments)) -} - -internal class FunctionalVariableExpression(val name: String, val default: T? = null) : Expression { - override operator fun invoke(arguments: Map): T = - arguments[name] ?: default ?: error("Parameter not found: $name") -} - -internal class FunctionalConstantExpression(val value: T) : Expression { - override operator fun invoke(arguments: Map): T = value -} - -internal class FunctionalConstProductExpression( - val context: Space, - private val expr: Expression, - val const: Number -) : Expression { - override operator fun invoke(arguments: Map): T = context.multiply(expr.invoke(arguments), const) -} - /** * A context class for [Expression] construction. * * @param algebra The algebra to provide for Expressions built. */ -public abstract class FunctionalExpressionAlgebra>(public val algebra: A) : - ExpressionAlgebra> { +public abstract class FunctionalExpressionAlgebra>( + public val algebra: A, +) : ExpressionAlgebra> { /** * Builds an Expression of constant expression which does not depend on arguments. */ - public override fun const(value: T): Expression = FunctionalConstantExpression(value) + public override fun const(value: T): Expression = Expression { value } /** * Builds an Expression to access a variable. */ - public override fun variable(name: String, default: T?): Expression = FunctionalVariableExpression(name, default) + public override fun bindOrNull(symbol: Symbol): Expression? = Expression { arguments -> + arguments[symbol] ?: error("Argument not found: $symbol") + } /** * Builds an Expression of dynamic call of binary operation [operation] on [left] and [right]. */ - public override fun binaryOperation(operation: String, left: Expression, right: Expression): Expression = - FunctionalBinaryOperation(algebra, operation, left, right) + public override fun binaryOperation( + operation: String, + left: Expression, + right: Expression, + ): Expression = Expression { arguments -> + algebra.binaryOperation(operation, left.invoke(arguments), right.invoke(arguments)) + } /** * Builds an Expression of dynamic call of unary operation with name [operation] on [arg]. */ - public override fun unaryOperation(operation: String, arg: Expression): Expression = - FunctionalUnaryOperation(algebra, operation, arg) + public override fun unaryOperation(operation: String, arg: Expression): Expression = Expression { arguments -> + algebra.unaryOperation(operation, arg.invoke(arguments)) + } } /** @@ -81,8 +57,9 @@ public open class FunctionalExpressionSpace>(algebra: A) : /** * Builds an Expression of multiplication of expression by number. */ - public override fun multiply(a: Expression, k: Number): Expression = - FunctionalConstProductExpression(algebra, a, k) + public override fun multiply(a: Expression, k: Number): Expression = Expression { arguments -> + algebra.multiply(a.invoke(arguments), k) + } public operator fun Expression.plus(arg: T): Expression = this + const(arg) public operator fun Expression.minus(arg: T): Expression = this - const(arg) @@ -118,8 +95,8 @@ public open class FunctionalExpressionRing(algebra: A) : FunctionalExpress } public open class FunctionalExpressionField(algebra: A) : - FunctionalExpressionRing(algebra), - Field> where A : Field, A : NumericAlgebra { + FunctionalExpressionRing(algebra), Field> + where A : Field, A : NumericAlgebra { /** * Builds an Expression of division an expression by another one. */ diff --git a/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/SimpleAutoDiff.kt b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/SimpleAutoDiff.kt new file mode 100644 index 000000000..5a9642690 --- /dev/null +++ b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/SimpleAutoDiff.kt @@ -0,0 +1,395 @@ +package kscience.kmath.expressions + +import kscience.kmath.linear.Point +import kscience.kmath.operations.* +import kscience.kmath.structures.asBuffer +import kotlin.contracts.InvocationKind +import kotlin.contracts.contract + +/* + * Implementation of backward-mode automatic differentiation. + * Initial gist by Roman Elizarov: https://gist.github.com/elizarov/1ad3a8583e88cb6ea7a0ad09bb591d3d + */ + + +public open class AutoDiffValue(public val value: T) + +/** + * Represents result of [simpleAutoDiff] call. + * + * @param T the non-nullable type of value. + * @param value the value of result. + * @property simpleAutoDiff The mapping of differentiated variables to their derivatives. + * @property context The field over [T]. + */ +public class DerivationResult( + public val value: T, + private val derivativeValues: Map, + public val context: Field, +) { + /** + * Returns derivative of [variable] or returns [Ring.zero] in [context]. + */ + public fun derivative(variable: Symbol): T = derivativeValues[variable.identity] ?: context.zero + + /** + * Computes the divergence. + */ + public fun div(): T = context { sum(derivativeValues.values) } +} + +/** + * Computes the gradient for variables in given order. + */ +public fun DerivationResult.grad(vararg variables: Symbol): Point { + check(variables.isNotEmpty()) { "Variable order is not provided for gradient construction" } + return variables.map(::derivative).asBuffer() +} + +/** + * Runs differentiation and establishes [SimpleAutoDiffField] context inside the block of code. + * + * The partial derivatives are placed in argument `d` variable + * + * Example: + * ``` + * val x by symbol // define variable(s) and their values + * val y = RealField.withAutoDiff() { sqr(x) + 5 * x + 3 } // write formulate in deriv context + * assertEquals(17.0, y.x) // the value of result (y) + * assertEquals(9.0, x.d) // dy/dx + * ``` + * + * @param body the action in [SimpleAutoDiffField] context returning [AutoDiffVariable] to differentiate with respect to. + * @return the result of differentiation. + */ +public fun > F.simpleAutoDiff( + bindings: Map, + body: SimpleAutoDiffField.() -> AutoDiffValue, +): DerivationResult { + contract { callsInPlace(body, InvocationKind.EXACTLY_ONCE) } + + return SimpleAutoDiffField(this, bindings).derivate(body) +} + +public fun > F.simpleAutoDiff( + vararg bindings: Pair, + body: SimpleAutoDiffField.() -> AutoDiffValue, +): DerivationResult = simpleAutoDiff(bindings.toMap(), body) + +/** + * Represents field in context of which functions can be derived. + */ +public open class SimpleAutoDiffField>( + public val context: F, + bindings: Map, +) : Field>, ExpressionAlgebra> { + + // this stack contains pairs of blocks and values to apply them to + private var stack: Array = arrayOfNulls(8) + private var sp: Int = 0 + private val derivatives: MutableMap, T> = hashMapOf() + + /** + * Differentiable variable with value and derivative of differentiation ([simpleAutoDiff]) result + * with respect to this variable. + * + * @param T the non-nullable type of value. + * @property value The value of this variable. + */ + private class AutoDiffVariableWithDerivative( + override val identity: String, + value: T, + var d: T, + ) : AutoDiffValue(value), Symbol { + override fun toString(): String = identity + override fun equals(other: Any?): Boolean = this.identity == (other as? Symbol)?.identity + override fun hashCode(): Int = identity.hashCode() + } + + private val bindings: Map> = bindings.entries.associate { + it.key.identity to AutoDiffVariableWithDerivative(it.key.identity, it.value, context.zero) + } + + override fun bindOrNull(symbol: Symbol): AutoDiffValue? = bindings[symbol.identity] + + private fun getDerivative(variable: AutoDiffValue): T = + (variable as? AutoDiffVariableWithDerivative)?.d ?: derivatives[variable] ?: context.zero + + private fun setDerivative(variable: AutoDiffValue, value: T) { + if (variable is AutoDiffVariableWithDerivative) variable.d = value else derivatives[variable] = value + } + + + @Suppress("UNCHECKED_CAST") + private fun runBackwardPass() { + while (sp > 0) { + val value = stack[--sp] + val block = stack[--sp] as F.(Any?) -> Unit + context.block(value) + } + } + + override val zero: AutoDiffValue get() = const(context.zero) + override val one: AutoDiffValue get() = const(context.one) + + override fun const(value: T): AutoDiffValue = AutoDiffValue(value) + + /** + * A variable accessing inner state of derivatives. + * Use this value in inner builders to avoid creating additional derivative bindings. + */ + public var AutoDiffValue.d: T + get() = getDerivative(this) + set(value) = setDerivative(this, value) + + public inline fun const(block: F.() -> T): AutoDiffValue = const(context.block()) + + /** + * Performs update of derivative after the rest of the formula in the back-pass. + * + * For example, implementation of `sin` function is: + * + * ``` + * fun AD.sin(x: Variable): Variable = derive(Variable(sin(x.x)) { z -> // call derive with function result + * x.d += z.d * cos(x.x) // update derivative using chain rule and derivative of the function + * } + * ``` + */ + @Suppress("UNCHECKED_CAST") + public fun derive(value: R, block: F.(R) -> Unit): R { + // save block to stack for backward pass + if (sp >= stack.size) stack = stack.copyOf(stack.size * 2) + stack[sp++] = block + stack[sp++] = value + return value + } + + + internal fun derivate(function: SimpleAutoDiffField.() -> AutoDiffValue): DerivationResult { + val result = function() + result.d = context.one // computing derivative w.r.t result + runBackwardPass() + return DerivationResult(result.value, bindings.mapValues { it.value.d }, context) + } + + // Overloads for Double constants + + override operator fun Number.plus(b: AutoDiffValue): AutoDiffValue = + derive(const { this@plus.toDouble() * one + b.value }) { z -> + b.d += z.d + } + + override operator fun AutoDiffValue.plus(b: Number): AutoDiffValue = b.plus(this) + + override operator fun Number.minus(b: AutoDiffValue): AutoDiffValue = + derive(const { this@minus.toDouble() * one - b.value }) { z -> b.d -= z.d } + + override operator fun AutoDiffValue.minus(b: Number): AutoDiffValue = + derive(const { this@minus.value - one * b.toDouble() }) { z -> this@minus.d += z.d } + + + // Basic math (+, -, *, /) + + override fun add(a: AutoDiffValue, b: AutoDiffValue): AutoDiffValue = + derive(const { a.value + b.value }) { z -> + a.d += z.d + b.d += z.d + } + + override fun multiply(a: AutoDiffValue, b: AutoDiffValue): AutoDiffValue = + derive(const { a.value * b.value }) { z -> + a.d += z.d * b.value + b.d += z.d * a.value + } + + override fun divide(a: AutoDiffValue, b: AutoDiffValue): AutoDiffValue = + derive(const { a.value / b.value }) { z -> + a.d += z.d / b.value + b.d -= z.d * a.value / (b.value * b.value) + } + + override fun multiply(a: AutoDiffValue, k: Number): AutoDiffValue = + derive(const { k.toDouble() * a.value }) { z -> + a.d += z.d * k.toDouble() + } +} + +/** + * A constructs that creates a derivative structure with required order on-demand + */ +public class SimpleAutoDiffExpression>( + public val field: F, + public val function: SimpleAutoDiffField.() -> AutoDiffValue, +) : FirstDerivativeExpression() { + public override operator fun invoke(arguments: Map): T { + //val bindings = arguments.entries.map { it.key.bind(it.value) } + return SimpleAutoDiffField(field, arguments).function().value + } + + override fun derivativeOrNull(symbol: Symbol): Expression = Expression { arguments -> + //val bindings = arguments.entries.map { it.key.bind(it.value) } + val derivationResult = SimpleAutoDiffField(field, arguments).derivate(function) + derivationResult.derivative(symbol) + } +} + +/** + * Generate [AutoDiffProcessor] for [SimpleAutoDiffExpression] + */ +public fun > simpleAutoDiff(field: F): AutoDiffProcessor, SimpleAutoDiffField> { + return object : AutoDiffProcessor, SimpleAutoDiffField> { + override fun process(function: SimpleAutoDiffField.() -> AutoDiffValue): DifferentiableExpression { + return SimpleAutoDiffExpression(field, function) + } + } +} + +// Extensions for differentiation of various basic mathematical functions + +// x ^ 2 +public fun > SimpleAutoDiffField.sqr(x: AutoDiffValue): AutoDiffValue = + derive(const { x.value * x.value }) { z -> x.d += z.d * 2 * x.value } + +// x ^ 1/2 +public fun > SimpleAutoDiffField.sqrt(x: AutoDiffValue): AutoDiffValue = + derive(const { sqrt(x.value) }) { z -> x.d += z.d * 0.5 / z.value } + +// x ^ y (const) +public fun > SimpleAutoDiffField.pow( + x: AutoDiffValue, + y: Double, +): AutoDiffValue = + derive(const { power(x.value, y) }) { z -> x.d += z.d * y * power(x.value, y - 1) } + +public fun > SimpleAutoDiffField.pow( + x: AutoDiffValue, + y: Int, +): AutoDiffValue = pow(x, y.toDouble()) + +// exp(x) +public fun > SimpleAutoDiffField.exp(x: AutoDiffValue): AutoDiffValue = + derive(const { exp(x.value) }) { z -> x.d += z.d * z.value } + +// ln(x) +public fun > SimpleAutoDiffField.ln(x: AutoDiffValue): AutoDiffValue = + derive(const { ln(x.value) }) { z -> x.d += z.d / x.value } + +// x ^ y (any) +public fun > SimpleAutoDiffField.pow( + x: AutoDiffValue, + y: AutoDiffValue, +): AutoDiffValue = + exp(y * ln(x)) + +// sin(x) +public fun > SimpleAutoDiffField.sin(x: AutoDiffValue): AutoDiffValue = + derive(const { sin(x.value) }) { z -> x.d += z.d * cos(x.value) } + +// cos(x) +public fun > SimpleAutoDiffField.cos(x: AutoDiffValue): AutoDiffValue = + derive(const { cos(x.value) }) { z -> x.d -= z.d * sin(x.value) } + +public fun > SimpleAutoDiffField.tan(x: AutoDiffValue): AutoDiffValue = + derive(const { tan(x.value) }) { z -> + val c = cos(x.value) + x.d += z.d / (c * c) + } + +public fun > SimpleAutoDiffField.asin(x: AutoDiffValue): AutoDiffValue = + derive(const { asin(x.value) }) { z -> x.d += z.d / sqrt(one - x.value * x.value) } + +public fun > SimpleAutoDiffField.acos(x: AutoDiffValue): AutoDiffValue = + derive(const { acos(x.value) }) { z -> x.d -= z.d / sqrt(one - x.value * x.value) } + +public fun > SimpleAutoDiffField.atan(x: AutoDiffValue): AutoDiffValue = + derive(const { atan(x.value) }) { z -> x.d += z.d / (one + x.value * x.value) } + +public fun > SimpleAutoDiffField.sinh(x: AutoDiffValue): AutoDiffValue = + derive(const { sinh(x.value) }) { z -> x.d += z.d * cosh(x.value) } + +public fun > SimpleAutoDiffField.cosh(x: AutoDiffValue): AutoDiffValue = + derive(const { cosh(x.value) }) { z -> x.d += z.d * sinh(x.value) } + +public fun > SimpleAutoDiffField.tanh(x: AutoDiffValue): AutoDiffValue = + derive(const { tanh(x.value) }) { z -> + val c = cosh(x.value) + x.d += z.d / (c * c) + } + +public fun > SimpleAutoDiffField.asinh(x: AutoDiffValue): AutoDiffValue = + derive(const { asinh(x.value) }) { z -> x.d += z.d / sqrt(one + x.value * x.value) } + +public fun > SimpleAutoDiffField.acosh(x: AutoDiffValue): AutoDiffValue = + derive(const { acosh(x.value) }) { z -> x.d += z.d / (sqrt((x.value - one) * (x.value + one))) } + +public fun > SimpleAutoDiffField.atanh(x: AutoDiffValue): AutoDiffValue = + derive(const { atanh(x.value) }) { z -> x.d += z.d / (one - x.value * x.value) } + +public class SimpleAutoDiffExtendedField>( + context: F, + bindings: Map, +) : ExtendedField>, SimpleAutoDiffField(context, bindings) { + // x ^ 2 + public fun sqr(x: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).sqr(x) + + // x ^ 1/2 + public override fun sqrt(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).sqrt(arg) + + // x ^ y (const) + public override fun power(arg: AutoDiffValue, pow: Number): AutoDiffValue = + (this as SimpleAutoDiffField).pow(arg, pow.toDouble()) + + // exp(x) + public override fun exp(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).exp(arg) + + // ln(x) + public override fun ln(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).ln(arg) + + // x ^ y (any) + public fun pow( + x: AutoDiffValue, + y: AutoDiffValue, + ): AutoDiffValue = exp(y * ln(x)) + + // sin(x) + public override fun sin(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).sin(arg) + + // cos(x) + public override fun cos(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).cos(arg) + + public override fun tan(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).tan(arg) + + public override fun asin(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).asin(arg) + + public override fun acos(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).acos(arg) + + public override fun atan(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).atan(arg) + + public override fun sinh(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).sinh(arg) + + public override fun cosh(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).cosh(arg) + + public override fun tanh(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).tanh(arg) + + public override fun asinh(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).asinh(arg) + + public override fun acosh(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).acosh(arg) + + public override fun atanh(arg: AutoDiffValue): AutoDiffValue = + (this as SimpleAutoDiffField).atanh(arg) +} \ No newline at end of file diff --git a/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/SymbolIndexer.kt b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/SymbolIndexer.kt new file mode 100644 index 000000000..6c61c7c7d --- /dev/null +++ b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/SymbolIndexer.kt @@ -0,0 +1,61 @@ +package kscience.kmath.expressions + +import kscience.kmath.linear.Point +import kscience.kmath.structures.BufferFactory +import kscience.kmath.structures.Structure2D + +/** + * An environment to easy transform indexed variables to symbols and back. + * TODO requires multi-receivers to be beutiful + */ +public interface SymbolIndexer { + public val symbols: List + public fun indexOf(symbol: Symbol): Int = symbols.indexOf(symbol) + + public operator fun List.get(symbol: Symbol): T { + require(size == symbols.size) { "The input list size for indexer should be ${symbols.size} but $size found" } + return get(this@SymbolIndexer.indexOf(symbol)) + } + + public operator fun Array.get(symbol: Symbol): T { + require(size == symbols.size) { "The input array size for indexer should be ${symbols.size} but $size found" } + return get(this@SymbolIndexer.indexOf(symbol)) + } + + public operator fun DoubleArray.get(symbol: Symbol): Double { + require(size == symbols.size) { "The input array size for indexer should be ${symbols.size} but $size found" } + return get(this@SymbolIndexer.indexOf(symbol)) + } + + public operator fun Point.get(symbol: Symbol): T { + require(size == symbols.size) { "The input buffer size for indexer should be ${symbols.size} but $size found" } + return get(this@SymbolIndexer.indexOf(symbol)) + } + + public fun DoubleArray.toMap(): Map { + require(size == symbols.size) { "The input array size for indexer should be ${symbols.size} but $size found" } + return symbols.indices.associate { symbols[it] to get(it) } + } + + public operator fun Structure2D.get(rowSymbol: Symbol, columnSymbol: Symbol): T = + get(indexOf(rowSymbol), indexOf(columnSymbol)) + + + public fun Map.toList(): List = symbols.map { getValue(it) } + + public fun Map.toPoint(bufferFactory: BufferFactory): Point = + bufferFactory(symbols.size) { getValue(symbols[it]) } + + public fun Map.toDoubleArray(): DoubleArray = DoubleArray(symbols.size) { getValue(symbols[it]) } +} + +public inline class SimpleSymbolIndexer(override val symbols: List) : SymbolIndexer + +/** + * Execute the block with symbol indexer based on given symbol order + */ +public inline fun withSymbols(vararg symbols: Symbol, block: SymbolIndexer.() -> R): R = + with(SimpleSymbolIndexer(symbols.toList()), block) + +public inline fun withSymbols(symbols: Collection, block: SymbolIndexer.() -> R): R = + with(SimpleSymbolIndexer(symbols.toList()), block) \ No newline at end of file diff --git a/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/expressionBuilders.kt b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/expressionBuilders.kt index 1702a5921..1603bc21d 100644 --- a/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/expressionBuilders.kt +++ b/kmath-core/src/commonMain/kotlin/kscience/kmath/expressions/expressionBuilders.kt @@ -7,6 +7,7 @@ import kscience.kmath.operations.Space import kotlin.contracts.InvocationKind import kotlin.contracts.contract + /** * Creates a functional expression with this [Space]. */ diff --git a/kmath-core/src/commonMain/kotlin/kscience/kmath/misc/AutoDiff.kt b/kmath-core/src/commonMain/kotlin/kscience/kmath/misc/AutoDiff.kt deleted file mode 100644 index bfcd5959f..000000000 --- a/kmath-core/src/commonMain/kotlin/kscience/kmath/misc/AutoDiff.kt +++ /dev/null @@ -1,266 +0,0 @@ -package kscience.kmath.misc - -import kscience.kmath.linear.Point -import kscience.kmath.operations.* -import kscience.kmath.structures.asBuffer -import kotlin.contracts.InvocationKind -import kotlin.contracts.contract - -/* - * Implementation of backward-mode automatic differentiation. - * Initial gist by Roman Elizarov: https://gist.github.com/elizarov/1ad3a8583e88cb6ea7a0ad09bb591d3d - */ - -/** - * Differentiable variable with value and derivative of differentiation ([deriv]) result - * with respect to this variable. - * - * @param T the non-nullable type of value. - * @property value The value of this variable. - */ -public open class Variable(public val value: T) - -/** - * Represents result of [deriv] call. - * - * @param T the non-nullable type of value. - * @param value the value of result. - * @property deriv The mapping of differentiated variables to their derivatives. - * @property context The field over [T]. - */ -public class DerivationResult( - value: T, - public val deriv: Map, T>, - public val context: Field -) : Variable(value) { - /** - * Returns derivative of [variable] or returns [Ring.zero] in [context]. - */ - public fun deriv(variable: Variable): T = deriv[variable] ?: context.zero - - /** - * Computes the divergence. - */ - public fun div(): T = context { sum(deriv.values) } - - /** - * Computes the gradient for variables in given order. - */ - public fun grad(vararg variables: Variable): Point { - check(variables.isNotEmpty()) { "Variable order is not provided for gradient construction" } - return variables.map(::deriv).asBuffer() - } -} - -/** - * Runs differentiation and establishes [AutoDiffField] context inside the block of code. - * - * The partial derivatives are placed in argument `d` variable - * - * Example: - * ``` - * val x = Variable(2) // define variable(s) and their values - * val y = deriv { sqr(x) + 5 * x + 3 } // write formulate in deriv context - * assertEquals(17.0, y.x) // the value of result (y) - * assertEquals(9.0, x.d) // dy/dx - * ``` - * - * @param body the action in [AutoDiffField] context returning [Variable] to differentiate with respect to. - * @return the result of differentiation. - */ -public inline fun > F.deriv(body: AutoDiffField.() -> Variable): DerivationResult { - contract { callsInPlace(body, InvocationKind.EXACTLY_ONCE) } - - return (AutoDiffContext(this)) { - val result = body() - result.d = context.one // computing derivative w.r.t result - runBackwardPass() - DerivationResult(result.value, derivatives, this@deriv) - } -} - -/** - * Represents field in context of which functions can be derived. - */ -public abstract class AutoDiffField> : Field> { - public abstract val context: F - - /** - * A variable accessing inner state of derivatives. - * Use this value in inner builders to avoid creating additional derivative bindings. - */ - public abstract var Variable.d: T - - /** - * Performs update of derivative after the rest of the formula in the back-pass. - * - * For example, implementation of `sin` function is: - * - * ``` - * fun AD.sin(x: Variable): Variable = derive(Variable(sin(x.x)) { z -> // call derive with function result - * x.d += z.d * cos(x.x) // update derivative using chain rule and derivative of the function - * } - * ``` - */ - public abstract fun derive(value: R, block: F.(R) -> Unit): R - - /** - * - */ - public abstract fun variable(value: T): Variable - - public inline fun variable(block: F.() -> T): Variable = variable(context.block()) - - // Overloads for Double constants - - override operator fun Number.plus(b: Variable): Variable = - derive(variable { this@plus.toDouble() * one + b.value }) { z -> - b.d += z.d - } - - override operator fun Variable.plus(b: Number): Variable = b.plus(this) - - override operator fun Number.minus(b: Variable): Variable = - derive(variable { this@minus.toDouble() * one - b.value }) { z -> b.d -= z.d } - - override operator fun Variable.minus(b: Number): Variable = - derive(variable { this@minus.value - one * b.toDouble() }) { z -> this@minus.d += z.d } -} - -/** - * Automatic Differentiation context class. - */ -@PublishedApi -internal class AutoDiffContext>(override val context: F) : AutoDiffField() { - // this stack contains pairs of blocks and values to apply them to - private var stack: Array = arrayOfNulls(8) - private var sp: Int = 0 - val derivatives: MutableMap, T> = hashMapOf() - override val zero: Variable get() = Variable(context.zero) - override val one: Variable get() = Variable(context.one) - - /** - * A variable coupled with its derivative. For internal use only - */ - private class VariableWithDeriv(x: T, var d: T) : Variable(x) - - override fun variable(value: T): Variable = - VariableWithDeriv(value, context.zero) - - override var Variable.d: T - get() = (this as? VariableWithDeriv)?.d ?: derivatives[this] ?: context.zero - set(value) = if (this is VariableWithDeriv) d = value else derivatives[this] = value - - @Suppress("UNCHECKED_CAST") - override fun derive(value: R, block: F.(R) -> Unit): R { - // save block to stack for backward pass - if (sp >= stack.size) stack = stack.copyOf(stack.size * 2) - stack[sp++] = block - stack[sp++] = value - return value - } - - @Suppress("UNCHECKED_CAST") - fun runBackwardPass() { - while (sp > 0) { - val value = stack[--sp] - val block = stack[--sp] as F.(Any?) -> Unit - context.block(value) - } - } - - // Basic math (+, -, *, /) - - override fun add(a: Variable, b: Variable): Variable = derive(variable { a.value + b.value }) { z -> - a.d += z.d - b.d += z.d - } - - override fun multiply(a: Variable, b: Variable): Variable = derive(variable { a.value * b.value }) { z -> - a.d += z.d * b.value - b.d += z.d * a.value - } - - override fun divide(a: Variable, b: Variable): Variable = derive(variable { a.value / b.value }) { z -> - a.d += z.d / b.value - b.d -= z.d * a.value / (b.value * b.value) - } - - override fun multiply(a: Variable, k: Number): Variable = derive(variable { k.toDouble() * a.value }) { z -> - a.d += z.d * k.toDouble() - } -} - -// Extensions for differentiation of various basic mathematical functions - -// x ^ 2 -public fun > AutoDiffField.sqr(x: Variable): Variable = - derive(variable { x.value * x.value }) { z -> x.d += z.d * 2 * x.value } - -// x ^ 1/2 -public fun > AutoDiffField.sqrt(x: Variable): Variable = - derive(variable { sqrt(x.value) }) { z -> x.d += z.d * 0.5 / z.value } - -// x ^ y (const) -public fun > AutoDiffField.pow(x: Variable, y: Double): Variable = - derive(variable { power(x.value, y) }) { z -> x.d += z.d * y * power(x.value, y - 1) } - -public fun > AutoDiffField.pow(x: Variable, y: Int): Variable = - pow(x, y.toDouble()) - -// exp(x) -public fun > AutoDiffField.exp(x: Variable): Variable = - derive(variable { exp(x.value) }) { z -> x.d += z.d * z.value } - -// ln(x) -public fun > AutoDiffField.ln(x: Variable): Variable = - derive(variable { ln(x.value) }) { z -> x.d += z.d / x.value } - -// x ^ y (any) -public fun > AutoDiffField.pow(x: Variable, y: Variable): Variable = - exp(y * ln(x)) - -// sin(x) -public fun > AutoDiffField.sin(x: Variable): Variable = - derive(variable { sin(x.value) }) { z -> x.d += z.d * cos(x.value) } - -// cos(x) -public fun > AutoDiffField.cos(x: Variable): Variable = - derive(variable { cos(x.value) }) { z -> x.d -= z.d * sin(x.value) } - -public fun > AutoDiffField.tan(x: Variable): Variable = - derive(variable { tan(x.value) }) { z -> - val c = cos(x.value) - x.d += z.d / (c * c) - } - -public fun > AutoDiffField.asin(x: Variable): Variable = - derive(variable { asin(x.value) }) { z -> x.d += z.d / sqrt(one - x.value * x.value) } - -public fun > AutoDiffField.acos(x: Variable): Variable = - derive(variable { acos(x.value) }) { z -> x.d -= z.d / sqrt(one - x.value * x.value) } - -public fun > AutoDiffField.atan(x: Variable): Variable = - derive(variable { atan(x.value) }) { z -> x.d += z.d / (one + x.value * x.value) } - -public fun > AutoDiffField.sinh(x: Variable): Variable = - derive(variable { sin(x.value) }) { z -> x.d += z.d * cosh(x.value) } - -public fun > AutoDiffField.cosh(x: Variable): Variable = - derive(variable { cos(x.value) }) { z -> x.d += z.d * sinh(x.value) } - -public fun > AutoDiffField.tanh(x: Variable): Variable = - derive(variable { tan(x.value) }) { z -> - val c = cosh(x.value) - x.d += z.d / (c * c) - } - -public fun > AutoDiffField.asinh(x: Variable): Variable = - derive(variable { asinh(x.value) }) { z -> x.d += z.d / sqrt(one + x.value * x.value) } - -public fun > AutoDiffField.acosh(x: Variable): Variable = - derive(variable { acosh(x.value) }) { z -> x.d += z.d / (sqrt((x.value - one) * (x.value + one))) } - -public fun > AutoDiffField.atanh(x: Variable): Variable = - derive(variable { atanh(x.value) }) { z -> x.d += z.d / (one - x.value * x.value) } - diff --git a/kmath-core/src/commonMain/kotlin/kscience/kmath/operations/Complex.kt b/kmath-core/src/commonMain/kotlin/kscience/kmath/operations/Complex.kt index 37055a5c8..703931c7c 100644 --- a/kmath-core/src/commonMain/kotlin/kscience/kmath/operations/Complex.kt +++ b/kmath-core/src/commonMain/kotlin/kscience/kmath/operations/Complex.kt @@ -195,6 +195,7 @@ public data class Complex(val re: Double, val im: Double) : FieldElement FunctionalExpressionField.expression(): Expression { - val x = variable("x") + val x by binding() return x * x + 2 * x + one } val expression = FunctionalExpressionField(RealField).expression() - assertEquals(expression("x" to 1.0), 4.0) + assertEquals(expression(x to 1.0), 4.0) } @Test fun valueExpression() { val expressionBuilder: FunctionalExpressionField.() -> Expression = { - val x = variable("x") + val x by binding() x * x + 2 * x + one } val expression = FunctionalExpressionField(RealField).expressionBuilder() - assertEquals(expression("x" to 1.0), 4.0) + assertEquals(expression(x to 1.0), 4.0) } } diff --git a/kmath-core/src/commonTest/kotlin/kscience/kmath/expressions/SimpleAutoDiffTest.kt b/kmath-core/src/commonTest/kotlin/kscience/kmath/expressions/SimpleAutoDiffTest.kt new file mode 100644 index 000000000..510ed23a9 --- /dev/null +++ b/kmath-core/src/commonTest/kotlin/kscience/kmath/expressions/SimpleAutoDiffTest.kt @@ -0,0 +1,285 @@ +package kscience.kmath.expressions + +import kscience.kmath.operations.RealField +import kscience.kmath.structures.asBuffer +import kotlin.math.E +import kotlin.math.PI +import kotlin.math.pow +import kotlin.math.sqrt +import kotlin.test.Test +import kotlin.test.assertEquals +import kotlin.test.assertTrue + +class SimpleAutoDiffTest { + + fun dx( + xBinding: Pair, + body: SimpleAutoDiffField.(x: AutoDiffValue) -> AutoDiffValue, + ): DerivationResult = RealField.simpleAutoDiff(xBinding) { body(bind(xBinding.first)) } + + fun dxy( + xBinding: Pair, + yBinding: Pair, + body: SimpleAutoDiffField.(x: AutoDiffValue, y: AutoDiffValue) -> AutoDiffValue, + ): DerivationResult = RealField.simpleAutoDiff(xBinding, yBinding) { + body(bind(xBinding.first), bind(yBinding.first)) + } + + fun diff(block: SimpleAutoDiffField.() -> AutoDiffValue): SimpleAutoDiffExpression { + return SimpleAutoDiffExpression(RealField, block) + } + + val x by symbol + val y by symbol + val z by symbol + + @Test + fun testPlusX2() { + val y = RealField.simpleAutoDiff(x to 3.0) { + // diff w.r.t this x at 3 + val x = bind(x) + x + x + } + assertEquals(6.0, y.value) // y = x + x = 6 + assertEquals(2.0, y.derivative(x)) // dy/dx = 2 + } + + @Test + fun testPlusX2Expr() { + val expr = diff { + val x = bind(x) + x + x + } + assertEquals(6.0, expr(x to 3.0)) // y = x + x = 6 + assertEquals(2.0, expr.derivative(x)(x to 3.0)) // dy/dx = 2 + } + + + @Test + fun testPlus() { + // two variables + val z = RealField.simpleAutoDiff(x to 2.0, y to 3.0) { + val x = bind(x) + val y = bind(y) + x + y + } + assertEquals(5.0, z.value) // z = x + y = 5 + assertEquals(1.0, z.derivative(x)) // dz/dx = 1 + assertEquals(1.0, z.derivative(y)) // dz/dy = 1 + } + + @Test + fun testMinus() { + // two variables + val z = RealField.simpleAutoDiff(x to 7.0, y to 3.0) { + val x = bind(x) + val y = bind(y) + + x - y + } + assertEquals(4.0, z.value) // z = x - y = 4 + assertEquals(1.0, z.derivative(x)) // dz/dx = 1 + assertEquals(-1.0, z.derivative(y)) // dz/dy = -1 + } + + @Test + fun testMulX2() { + val y = dx(x to 3.0) { x -> + // diff w.r.t this x at 3 + x * x + } + assertEquals(9.0, y.value) // y = x * x = 9 + assertEquals(6.0, y.derivative(x)) // dy/dx = 2 * x = 7 + } + + @Test + fun testSqr() { + val y = dx(x to 3.0) { x -> sqr(x) } + assertEquals(9.0, y.value) // y = x ^ 2 = 9 + assertEquals(6.0, y.derivative(x)) // dy/dx = 2 * x = 7 + } + + @Test + fun testSqrSqr() { + val y = dx(x to 2.0) { x -> sqr(sqr(x)) } + assertEquals(16.0, y.value) // y = x ^ 4 = 16 + assertEquals(32.0, y.derivative(x)) // dy/dx = 4 * x^3 = 32 + } + + @Test + fun testX3() { + val y = dx(x to 2.0) { x -> + // diff w.r.t this x at 2 + x * x * x + } + assertEquals(8.0, y.value) // y = x * x * x = 8 + assertEquals(12.0, y.derivative(x)) // dy/dx = 3 * x * x = 12 + } + + @Test + fun testDiv() { + val z = dxy(x to 5.0, y to 2.0) { x, y -> + x / y + } + assertEquals(2.5, z.value) // z = x / y = 2.5 + assertEquals(0.5, z.derivative(x)) // dz/dx = 1 / y = 0.5 + assertEquals(-1.25, z.derivative(y)) // dz/dy = -x / y^2 = -1.25 + } + + @Test + fun testPow3() { + val y = dx(x to 2.0) { x -> + // diff w.r.t this x at 2 + pow(x, 3) + } + assertEquals(8.0, y.value) // y = x ^ 3 = 8 + assertEquals(12.0, y.derivative(x)) // dy/dx = 3 * x ^ 2 = 12 + } + + @Test + fun testPowFull() { + val z = dxy(x to 2.0, y to 3.0) { x, y -> + pow(x, y) + } + assertApprox(8.0, z.value) // z = x ^ y = 8 + assertApprox(12.0, z.derivative(x)) // dz/dx = y * x ^ (y - 1) = 12 + assertApprox(8.0 * kotlin.math.ln(2.0), z.derivative(y)) // dz/dy = x ^ y * ln(x) + } + + @Test + fun testFromPaper() { + val y = dx(x to 3.0) { x -> 2 * x + x * x * x } + assertEquals(33.0, y.value) // y = 2 * x + x * x * x = 33 + assertEquals(29.0, y.derivative(x)) // dy/dx = 2 + 3 * x * x = 29 + } + + @Test + fun testInnerVariable() { + val y = dx(x to 1.0) { x -> + const(1.0) * x + } + assertEquals(1.0, y.value) // y = x ^ n = 1 + assertEquals(1.0, y.derivative(x)) // dy/dx = n * x ^ (n - 1) = n - 1 + } + + @Test + fun testLongChain() { + val n = 10_000 + val y = dx(x to 1.0) { x -> + var res = const(1.0) + for (i in 1..n) res *= x + res + } + assertEquals(1.0, y.value) // y = x ^ n = 1 + assertEquals(n.toDouble(), y.derivative(x)) // dy/dx = n * x ^ (n - 1) = n - 1 + } + + @Test + fun testExample() { + val y = dx(x to 2.0) { x -> sqr(x) + 5 * x + 3 } + assertEquals(17.0, y.value) // the value of result (y) + assertEquals(9.0, y.derivative(x)) // dy/dx + } + + @Test + fun testSqrt() { + val y = dx(x to 16.0) { x -> sqrt(x) } + assertEquals(4.0, y.value) // y = x ^ 1/2 = 4 + assertEquals(1.0 / 8, y.derivative(x)) // dy/dx = 1/2 / x ^ 1/4 = 1/8 + } + + @Test + fun testSin() { + val y = dx(x to PI / 6.0) { x -> sin(x) } + assertApprox(0.5, y.value) // y = sin(PI/6) = 0.5 + assertApprox(sqrt(3.0) / 2, y.derivative(x)) // dy/dx = cos(pi/6) = sqrt(3)/2 + } + + @Test + fun testCos() { + val y = dx(x to PI / 6) { x -> cos(x) } + assertApprox(sqrt(3.0) / 2, y.value) //y = cos(pi/6) = sqrt(3)/2 + assertApprox(-0.5, y.derivative(x)) // dy/dx = -sin(pi/6) = -0.5 + } + + @Test + fun testTan() { + val y = dx(x to PI / 6) { x -> tan(x) } + assertApprox(1.0 / sqrt(3.0), y.value) // y = tan(pi/6) = 1/sqrt(3) + assertApprox(4.0 / 3.0, y.derivative(x)) // dy/dx = sec(pi/6)^2 = 4/3 + } + + @Test + fun testAsin() { + val y = dx(x to PI / 6) { x -> asin(x) } + assertApprox(kotlin.math.asin(PI / 6.0), y.value) // y = asin(pi/6) + assertApprox(6.0 / sqrt(36 - PI * PI), y.derivative(x)) // dy/dx = 6/sqrt(36-pi^2) + } + + @Test + fun testAcos() { + val y = dx(x to PI / 6) { x -> acos(x) } + assertApprox(kotlin.math.acos(PI / 6.0), y.value) // y = acos(pi/6) + assertApprox(-6.0 / sqrt(36.0 - PI * PI), y.derivative(x)) // dy/dx = -6/sqrt(36-pi^2) + } + + @Test + fun testAtan() { + val y = dx(x to PI / 6) { x -> atan(x) } + assertApprox(kotlin.math.atan(PI / 6.0), y.value) // y = atan(pi/6) + assertApprox(36.0 / (36.0 + PI * PI), y.derivative(x)) // dy/dx = 36/(36+pi^2) + } + + @Test + fun testSinh() { + val y = dx(x to 0.0) { x -> sinh(x) } + assertApprox(kotlin.math.sinh(0.0), y.value) // y = sinh(0) + assertApprox(kotlin.math.cosh(0.0), y.derivative(x)) // dy/dx = cosh(0) + } + + @Test + fun testCosh() { + val y = dx(x to 0.0) { x -> cosh(x) } + assertApprox(1.0, y.value) //y = cosh(0) + assertApprox(0.0, y.derivative(x)) // dy/dx = sinh(0) + } + + @Test + fun testTanh() { + val y = dx(x to 1.0) { x -> tanh(x) } + assertApprox((E * E - 1) / (E * E + 1), y.value) // y = tanh(pi/6) + assertApprox(1.0 / kotlin.math.cosh(1.0).pow(2), y.derivative(x)) // dy/dx = sech(pi/6)^2 + } + + @Test + fun testAsinh() { + val y = dx(x to PI / 6) { x -> asinh(x) } + assertApprox(kotlin.math.asinh(PI / 6.0), y.value) // y = asinh(pi/6) + assertApprox(6.0 / sqrt(36 + PI * PI), y.derivative(x)) // dy/dx = 6/sqrt(pi^2+36) + } + + @Test + fun testAcosh() { + val y = dx(x to PI / 6) { x -> acosh(x) } + assertApprox(kotlin.math.acosh(PI / 6.0), y.value) // y = acosh(pi/6) + assertApprox(-6.0 / sqrt(36.0 - PI * PI), y.derivative(x)) // dy/dx = -6/sqrt(36-pi^2) + } + + @Test + fun testAtanh() { + val y = dx(x to PI / 6) { x -> atanh(x) } + assertApprox(kotlin.math.atanh(PI / 6.0), y.value) // y = atanh(pi/6) + assertApprox(-36.0 / (PI * PI - 36.0), y.derivative(x)) // dy/dx = -36/(pi^2-36) + } + + @Test + fun testDivGrad() { + val res = dxy(x to 1.0, y to 2.0) { x, y -> x * x + y * y } + assertEquals(6.0, res.div()) + assertTrue(res.grad(x, y).contentEquals(doubleArrayOf(2.0, 4.0).asBuffer())) + } + + private fun assertApprox(a: Double, b: Double) { + if ((a - b) > 1e-10) assertEquals(a, b) + } +} diff --git a/kmath-core/src/commonTest/kotlin/kscience/kmath/misc/AutoDiffTest.kt b/kmath-core/src/commonTest/kotlin/kscience/kmath/misc/AutoDiffTest.kt deleted file mode 100644 index 3b1813185..000000000 --- a/kmath-core/src/commonTest/kotlin/kscience/kmath/misc/AutoDiffTest.kt +++ /dev/null @@ -1,261 +0,0 @@ -package kscience.kmath.misc - -import kscience.kmath.operations.RealField -import kscience.kmath.structures.asBuffer -import kotlin.math.PI -import kotlin.math.pow -import kotlin.math.sqrt -import kotlin.test.Test -import kotlin.test.assertEquals -import kotlin.test.assertTrue - -class AutoDiffTest { - inline fun deriv(body: AutoDiffField.() -> Variable): DerivationResult = - RealField.deriv(body) - - @Test - fun testPlusX2() { - val x = Variable(3.0) // diff w.r.t this x at 3 - val y = deriv { x + x } - assertEquals(6.0, y.value) // y = x + x = 6 - assertEquals(2.0, y.deriv(x)) // dy/dx = 2 - } - - @Test - fun testPlus() { - // two variables - val x = Variable(2.0) - val y = Variable(3.0) - val z = deriv { x + y } - assertEquals(5.0, z.value) // z = x + y = 5 - assertEquals(1.0, z.deriv(x)) // dz/dx = 1 - assertEquals(1.0, z.deriv(y)) // dz/dy = 1 - } - - @Test - fun testMinus() { - // two variables - val x = Variable(7.0) - val y = Variable(3.0) - val z = deriv { x - y } - assertEquals(4.0, z.value) // z = x - y = 4 - assertEquals(1.0, z.deriv(x)) // dz/dx = 1 - assertEquals(-1.0, z.deriv(y)) // dz/dy = -1 - } - - @Test - fun testMulX2() { - val x = Variable(3.0) // diff w.r.t this x at 3 - val y = deriv { x * x } - assertEquals(9.0, y.value) // y = x * x = 9 - assertEquals(6.0, y.deriv(x)) // dy/dx = 2 * x = 7 - } - - @Test - fun testSqr() { - val x = Variable(3.0) - val y = deriv { sqr(x) } - assertEquals(9.0, y.value) // y = x ^ 2 = 9 - assertEquals(6.0, y.deriv(x)) // dy/dx = 2 * x = 7 - } - - @Test - fun testSqrSqr() { - val x = Variable(2.0) - val y = deriv { sqr(sqr(x)) } - assertEquals(16.0, y.value) // y = x ^ 4 = 16 - assertEquals(32.0, y.deriv(x)) // dy/dx = 4 * x^3 = 32 - } - - @Test - fun testX3() { - val x = Variable(2.0) // diff w.r.t this x at 2 - val y = deriv { x * x * x } - assertEquals(8.0, y.value) // y = x * x * x = 8 - assertEquals(12.0, y.deriv(x)) // dy/dx = 3 * x * x = 12 - } - - @Test - fun testDiv() { - val x = Variable(5.0) - val y = Variable(2.0) - val z = deriv { x / y } - assertEquals(2.5, z.value) // z = x / y = 2.5 - assertEquals(0.5, z.deriv(x)) // dz/dx = 1 / y = 0.5 - assertEquals(-1.25, z.deriv(y)) // dz/dy = -x / y^2 = -1.25 - } - - @Test - fun testPow3() { - val x = Variable(2.0) // diff w.r.t this x at 2 - val y = deriv { pow(x, 3) } - assertEquals(8.0, y.value) // y = x ^ 3 = 8 - assertEquals(12.0, y.deriv(x)) // dy/dx = 3 * x ^ 2 = 12 - } - - @Test - fun testPowFull() { - val x = Variable(2.0) - val y = Variable(3.0) - val z = deriv { pow(x, y) } - assertApprox(8.0, z.value) // z = x ^ y = 8 - assertApprox(12.0, z.deriv(x)) // dz/dx = y * x ^ (y - 1) = 12 - assertApprox(8.0 * kotlin.math.ln(2.0), z.deriv(y)) // dz/dy = x ^ y * ln(x) - } - - @Test - fun testFromPaper() { - val x = Variable(3.0) - val y = deriv { 2 * x + x * x * x } - assertEquals(33.0, y.value) // y = 2 * x + x * x * x = 33 - assertEquals(29.0, y.deriv(x)) // dy/dx = 2 + 3 * x * x = 29 - } - - @Test - fun testInnerVariable() { - val x = Variable(1.0) - val y = deriv { - Variable(1.0) * x - } - assertEquals(1.0, y.value) // y = x ^ n = 1 - assertEquals(1.0, y.deriv(x)) // dy/dx = n * x ^ (n - 1) = n - 1 - } - - @Test - fun testLongChain() { - val n = 10_000 - val x = Variable(1.0) - val y = deriv { - var res = Variable(1.0) - for (i in 1..n) res *= x - res - } - assertEquals(1.0, y.value) // y = x ^ n = 1 - assertEquals(n.toDouble(), y.deriv(x)) // dy/dx = n * x ^ (n - 1) = n - 1 - } - - @Test - fun testExample() { - val x = Variable(2.0) - val y = deriv { sqr(x) + 5 * x + 3 } - assertEquals(17.0, y.value) // the value of result (y) - assertEquals(9.0, y.deriv(x)) // dy/dx - } - - @Test - fun testSqrt() { - val x = Variable(16.0) - val y = deriv { sqrt(x) } - assertEquals(4.0, y.value) // y = x ^ 1/2 = 4 - assertEquals(1.0 / 8, y.deriv(x)) // dy/dx = 1/2 / x ^ 1/4 = 1/8 - } - - @Test - fun testSin() { - val x = Variable(PI / 6.0) - val y = deriv { sin(x) } - assertApprox(0.5, y.value) // y = sin(PI/6) = 0.5 - assertApprox(sqrt(3.0) / 2, y.deriv(x)) // dy/dx = cos(pi/6) = sqrt(3)/2 - } - - @Test - fun testCos() { - val x = Variable(PI / 6) - val y = deriv { cos(x) } - assertApprox(sqrt(3.0) / 2, y.value) //y = cos(pi/6) = sqrt(3)/2 - assertApprox(-0.5, y.deriv(x)) // dy/dx = -sin(pi/6) = -0.5 - } - - @Test - fun testTan() { - val x = Variable(PI / 6) - val y = deriv { tan(x) } - assertApprox(1.0 / sqrt(3.0), y.value) // y = tan(pi/6) = 1/sqrt(3) - assertApprox(4.0 / 3.0, y.deriv(x)) // dy/dx = sec(pi/6)^2 = 4/3 - } - - @Test - fun testAsin() { - val x = Variable(PI / 6) - val y = deriv { asin(x) } - assertApprox(kotlin.math.asin(PI / 6.0), y.value) // y = asin(pi/6) - assertApprox(6.0 / sqrt(36 - PI * PI), y.deriv(x)) // dy/dx = 6/sqrt(36-pi^2) - } - - @Test - fun testAcos() { - val x = Variable(PI / 6) - val y = deriv { acos(x) } - assertApprox(kotlin.math.acos(PI / 6.0), y.value) // y = acos(pi/6) - assertApprox(-6.0 / sqrt(36.0 - PI * PI), y.deriv(x)) // dy/dx = -6/sqrt(36-pi^2) - } - - @Test - fun testAtan() { - val x = Variable(PI / 6) - val y = deriv { atan(x) } - assertApprox(kotlin.math.atan(PI / 6.0), y.value) // y = atan(pi/6) - assertApprox(36.0 / (36.0 + PI * PI), y.deriv(x)) // dy/dx = 36/(36+pi^2) - } - - @Test - fun testSinh() { - val x = Variable(0.0) - val y = deriv { sinh(x) } - assertApprox(kotlin.math.sinh(0.0), y.value) // y = sinh(0) - assertApprox(kotlin.math.cosh(0.0), y.deriv(x)) // dy/dx = cosh(0) - } - - @Test - fun testCosh() { - val x = Variable(0.0) - val y = deriv { cosh(x) } - assertApprox(1.0, y.value) //y = cosh(0) - assertApprox(0.0, y.deriv(x)) // dy/dx = sinh(0) - } - - @Test - fun testTanh() { - val x = Variable(PI / 6) - val y = deriv { tanh(x) } - assertApprox(1.0 / sqrt(3.0), y.value) // y = tanh(pi/6) - assertApprox(1.0 / kotlin.math.cosh(PI / 6.0).pow(2), y.deriv(x)) // dy/dx = sech(pi/6)^2 - } - - @Test - fun testAsinh() { - val x = Variable(PI / 6) - val y = deriv { asinh(x) } - assertApprox(kotlin.math.asinh(PI / 6.0), y.value) // y = asinh(pi/6) - assertApprox(6.0 / sqrt(36 + PI * PI), y.deriv(x)) // dy/dx = 6/sqrt(pi^2+36) - } - - @Test - fun testAcosh() { - val x = Variable(PI / 6) - val y = deriv { acosh(x) } - assertApprox(kotlin.math.acosh(PI / 6.0), y.value) // y = acosh(pi/6) - assertApprox(-6.0 / sqrt(36.0 - PI * PI), y.deriv(x)) // dy/dx = -6/sqrt(36-pi^2) - } - - @Test - fun testAtanh() { - val x = Variable(PI / 6.0) - val y = deriv { atanh(x) } - assertApprox(kotlin.math.atanh(PI / 6.0), y.value) // y = atanh(pi/6) - assertApprox(-36.0 / (PI * PI - 36.0), y.deriv(x)) // dy/dx = -36/(pi^2-36) - } - - @Test - fun testDivGrad() { - val x = Variable(1.0) - val y = Variable(2.0) - val res = deriv { x * x + y * y } - assertEquals(6.0, res.div()) - assertTrue(res.grad(x, y).contentEquals(doubleArrayOf(2.0, 4.0).asBuffer())) - } - - private fun assertApprox(a: Double, b: Double) { - if ((a - b) > 1e-10) assertEquals(a, b) - } -} diff --git a/kmath-functions/src/commonMain/kotlin/kscience/kmath/functions/Polynomial.kt b/kmath-functions/src/commonMain/kotlin/kscience/kmath/functions/Polynomial.kt index c513a6889..820076c4c 100644 --- a/kmath-functions/src/commonMain/kotlin/kscience/kmath/functions/Polynomial.kt +++ b/kmath-functions/src/commonMain/kotlin/kscience/kmath/functions/Polynomial.kt @@ -8,13 +8,13 @@ import kotlin.contracts.contract import kotlin.math.max import kotlin.math.pow -// TODO make `inline`, when KT-41771 gets fixed /** * Polynomial coefficients without fixation on specific context they are applied to * @param coefficients constant is the leftmost coefficient */ public inline class Polynomial(public val coefficients: List) +@Suppress("FunctionName") public fun Polynomial(vararg coefficients: T): Polynomial = Polynomial(coefficients.toList()) public fun Polynomial.value(): Double = coefficients.reduceIndexed { index, acc, d -> acc + d.pow(index) } @@ -33,14 +33,6 @@ public fun > Polynomial.value(ring: C, arg: T): T = ring res } -/** - * Represent a polynomial as a context-dependent function - */ -public fun > Polynomial.asMathFunction(): MathFunction = - object : MathFunction { - override fun C.invoke(arg: T): T = value(this, arg) - } - /** * Represent the polynomial as a regular context-less function */ @@ -49,7 +41,7 @@ public fun > Polynomial.asFunction(ring: C): (T) -> T = /** * An algebra for polynomials */ -public class PolynomialSpace>(public val ring: C) : Space> { +public class PolynomialSpace>(private val ring: C) : Space> { public override val zero: Polynomial = Polynomial(emptyList()) public override fun add(a: Polynomial, b: Polynomial): Polynomial { diff --git a/kmath-functions/src/commonMain/kotlin/kscience/kmath/functions/functions.kt b/kmath-functions/src/commonMain/kotlin/kscience/kmath/functions/functions.kt deleted file mode 100644 index d780c16f3..000000000 --- a/kmath-functions/src/commonMain/kotlin/kscience/kmath/functions/functions.kt +++ /dev/null @@ -1,34 +0,0 @@ -package kscience.kmath.functions - -import kscience.kmath.operations.Algebra -import kscience.kmath.operations.RealField - -// TODO make fun interface when KT-41770 is fixed -/** - * A regular function that could be called only inside specific algebra context - * @param T source type - * @param C source algebra constraint - * @param R result type - */ -public /*fun*/ interface MathFunction, R> { - public operator fun C.invoke(arg: T): R -} - -public fun MathFunction.invoke(arg: Double): R = RealField.invoke(arg) - -/** - * A suspendable function defined in algebraic context - */ -// TODO make fun interface, when the new JVM IR is enabled -public interface SuspendableMathFunction, R> { - public suspend operator fun C.invoke(arg: T): R -} - -public suspend fun SuspendableMathFunction.invoke(arg: Double): R = RealField.invoke(arg) - -/** - * A parametric function with parameter - */ -public fun interface ParametricFunction> { - public operator fun C.invoke(arg: T, parameter: P): T -} diff --git a/kmath-prob/build.gradle.kts b/kmath-stat/build.gradle.kts similarity index 100% rename from kmath-prob/build.gradle.kts rename to kmath-stat/build.gradle.kts diff --git a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/Distribution.kt b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/Distribution.kt similarity index 98% rename from kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/Distribution.kt rename to kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/Distribution.kt index 72660e20d..c4ceb29eb 100644 --- a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/Distribution.kt +++ b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/Distribution.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kscience.kmath.chains.Chain import kscience.kmath.chains.collect diff --git a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/FactorizedDistribution.kt b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/FactorizedDistribution.kt similarity index 98% rename from kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/FactorizedDistribution.kt rename to kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/FactorizedDistribution.kt index 4d713fc4e..1ed9deba9 100644 --- a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/FactorizedDistribution.kt +++ b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/FactorizedDistribution.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kscience.kmath.chains.Chain import kscience.kmath.chains.SimpleChain diff --git a/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/Fitting.kt b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/Fitting.kt new file mode 100644 index 000000000..01fdf4c5e --- /dev/null +++ b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/Fitting.kt @@ -0,0 +1,59 @@ +package kscience.kmath.stat + +import kscience.kmath.expressions.* +import kscience.kmath.operations.ExtendedField +import kscience.kmath.structures.Buffer +import kscience.kmath.structures.indices +import kotlin.math.pow + +public object Fitting { + + /** + * Generate a chi squared expression from given x-y-sigma data and inline model. Provides automatic differentiation + */ + public fun chiSquared( + autoDiff: AutoDiffProcessor, + x: Buffer, + y: Buffer, + yErr: Buffer, + model: A.(I) -> I, + ): DifferentiableExpression where A : ExtendedField, A : ExpressionAlgebra { + require(x.size == y.size) { "X and y buffers should be of the same size" } + require(y.size == yErr.size) { "Y and yErr buffer should of the same size" } + return autoDiff.process { + var sum = zero + x.indices.forEach { + val xValue = const(x[it]) + val yValue = const(y[it]) + val yErrValue = const(yErr[it]) + val modelValue = model(xValue) + sum += ((yValue - modelValue) / yErrValue).pow(2) + } + sum + } + } + + /** + * Generate a chi squared expression from given x-y-sigma model represented by an expression. Does not provide derivatives + */ + public fun chiSquared( + x: Buffer, + y: Buffer, + yErr: Buffer, + model: Expression, + xSymbol: Symbol = StringSymbol("x"), + ): Expression { + require(x.size == y.size) { "X and y buffers should be of the same size" } + require(y.size == yErr.size) { "Y and yErr buffer should of the same size" } + return Expression { arguments -> + x.indices.sumByDouble { + val xValue = x[it] + val yValue = y[it] + val yErrValue = yErr[it] + val modifiedArgs = arguments + (xSymbol to xValue) + val modelValue = model(modifiedArgs) + ((yValue - modelValue) / yErrValue).pow(2) + } + } + } +} \ No newline at end of file diff --git a/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/OptimizationProblem.kt b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/OptimizationProblem.kt new file mode 100644 index 000000000..ea522bff9 --- /dev/null +++ b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/OptimizationProblem.kt @@ -0,0 +1,91 @@ +package kscience.kmath.stat + +import kscience.kmath.expressions.DifferentiableExpression +import kscience.kmath.expressions.Expression +import kscience.kmath.expressions.Symbol + +public interface OptimizationFeature + +public class OptimizationResult( + public val point: Map, + public val value: T, + public val features: Set = emptySet(), +) { + override fun toString(): String { + return "OptimizationResult(point=$point, value=$value)" + } +} + +public operator fun OptimizationResult.plus( + feature: OptimizationFeature, +): OptimizationResult = OptimizationResult(point, value, features + feature) + +/** + * A configuration builder for optimization problem + */ +public interface OptimizationProblem { + /** + * Define the initial guess for the optimization problem + */ + public fun initialGuess(map: Map): Unit + + /** + * Set an objective function expression + */ + public fun expression(expression: Expression): Unit + + /** + * Set a differentiable expression as objective function as function and gradient provider + */ + public fun diffExpression(expression: DifferentiableExpression): Unit + + /** + * Update the problem from previous optimization run + */ + public fun update(result: OptimizationResult) + + /** + * Make an optimization run + */ + public fun optimize(): OptimizationResult +} + +public interface OptimizationProblemFactory> { + public fun build(symbols: List): P + +} + +public operator fun > OptimizationProblemFactory.invoke( + symbols: List, + block: P.() -> Unit, +): P = build(symbols).apply(block) + + +/** + * Optimize expression without derivatives using specific [OptimizationProblemFactory] + */ +public fun > Expression.optimizeWith( + factory: OptimizationProblemFactory, + vararg symbols: Symbol, + configuration: F.() -> Unit, +): OptimizationResult { + require(symbols.isNotEmpty()) { "Must provide a list of symbols for optimization" } + val problem = factory(symbols.toList(),configuration) + problem.expression(this) + return problem.optimize() +} + +/** + * Optimize differentiable expression using specific [OptimizationProblemFactory] + */ +public fun > DifferentiableExpression.optimizeWith( + factory: OptimizationProblemFactory, + vararg symbols: Symbol, + configuration: F.() -> Unit, +): OptimizationResult { + require(symbols.isNotEmpty()) { "Must provide a list of symbols for optimization" } + val problem = factory(symbols.toList(), configuration) + problem.diffExpression(this) + return problem.optimize() +} + diff --git a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/RandomChain.kt b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/RandomChain.kt similarity index 94% rename from kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/RandomChain.kt rename to kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/RandomChain.kt index b4a80f6c5..0f10851b9 100644 --- a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/RandomChain.kt +++ b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/RandomChain.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kscience.kmath.chains.Chain diff --git a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/RandomGenerator.kt b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/RandomGenerator.kt similarity index 99% rename from kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/RandomGenerator.kt rename to kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/RandomGenerator.kt index 2dd4ce51e..4486ae016 100644 --- a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/RandomGenerator.kt +++ b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/RandomGenerator.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kotlin.random.Random diff --git a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/SamplerAlgebra.kt b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/SamplerAlgebra.kt similarity index 97% rename from kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/SamplerAlgebra.kt rename to kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/SamplerAlgebra.kt index e363ba30b..f416028a5 100644 --- a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/SamplerAlgebra.kt +++ b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/SamplerAlgebra.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kscience.kmath.chains.Chain import kscience.kmath.chains.ConstantChain diff --git a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/Statistic.kt b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/Statistic.kt similarity index 99% rename from kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/Statistic.kt rename to kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/Statistic.kt index 6720a3d7f..a4624fc21 100644 --- a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/Statistic.kt +++ b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/Statistic.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kotlinx.coroutines.CoroutineDispatcher import kotlinx.coroutines.Dispatchers diff --git a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/UniformDistribution.kt b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/UniformDistribution.kt similarity index 96% rename from kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/UniformDistribution.kt rename to kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/UniformDistribution.kt index 8df2c01e1..1ba5c96f1 100644 --- a/kmath-prob/src/commonMain/kotlin/kscience/kmath/prob/UniformDistribution.kt +++ b/kmath-stat/src/commonMain/kotlin/kscience/kmath/stat/UniformDistribution.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kscience.kmath.chains.Chain import kscience.kmath.chains.SimpleChain diff --git a/kmath-prob/src/jvmMain/kotlin/kscience/kmath/prob/RandomSourceGenerator.kt b/kmath-stat/src/jvmMain/kotlin/kscience/kmath/stat/RandomSourceGenerator.kt similarity index 98% rename from kmath-prob/src/jvmMain/kotlin/kscience/kmath/prob/RandomSourceGenerator.kt rename to kmath-stat/src/jvmMain/kotlin/kscience/kmath/stat/RandomSourceGenerator.kt index 18be6f019..5cba28a95 100644 --- a/kmath-prob/src/jvmMain/kotlin/kscience/kmath/prob/RandomSourceGenerator.kt +++ b/kmath-stat/src/jvmMain/kotlin/kscience/kmath/stat/RandomSourceGenerator.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import org.apache.commons.rng.UniformRandomProvider import org.apache.commons.rng.simple.RandomSource diff --git a/kmath-prob/src/jvmMain/kotlin/kscience/kmath/prob/distributions.kt b/kmath-stat/src/jvmMain/kotlin/kscience/kmath/stat/distributions.kt similarity index 99% rename from kmath-prob/src/jvmMain/kotlin/kscience/kmath/prob/distributions.kt rename to kmath-stat/src/jvmMain/kotlin/kscience/kmath/stat/distributions.kt index ff20572cc..9a77b0bd2 100644 --- a/kmath-prob/src/jvmMain/kotlin/kscience/kmath/prob/distributions.kt +++ b/kmath-stat/src/jvmMain/kotlin/kscience/kmath/stat/distributions.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kscience.kmath.chains.BlockingIntChain import kscience.kmath.chains.BlockingRealChain diff --git a/kmath-prob/src/jvmTest/kotlin/kscience/kmath/prob/CommonsDistributionsTest.kt b/kmath-stat/src/jvmTest/kotlin/kscience/kmath/stat/CommonsDistributionsTest.kt similarity index 96% rename from kmath-prob/src/jvmTest/kotlin/kscience/kmath/prob/CommonsDistributionsTest.kt rename to kmath-stat/src/jvmTest/kotlin/kscience/kmath/stat/CommonsDistributionsTest.kt index 12a00684b..fe58fac08 100644 --- a/kmath-prob/src/jvmTest/kotlin/kscience/kmath/prob/CommonsDistributionsTest.kt +++ b/kmath-stat/src/jvmTest/kotlin/kscience/kmath/stat/CommonsDistributionsTest.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kotlinx.coroutines.flow.take import kotlinx.coroutines.flow.toList diff --git a/kmath-prob/src/jvmTest/kotlin/kscience/kmath/prob/SamplerTest.kt b/kmath-stat/src/jvmTest/kotlin/kscience/kmath/stat/SamplerTest.kt similarity index 92% rename from kmath-prob/src/jvmTest/kotlin/kscience/kmath/prob/SamplerTest.kt rename to kmath-stat/src/jvmTest/kotlin/kscience/kmath/stat/SamplerTest.kt index 75db5c402..afed4c5d0 100644 --- a/kmath-prob/src/jvmTest/kotlin/kscience/kmath/prob/SamplerTest.kt +++ b/kmath-stat/src/jvmTest/kotlin/kscience/kmath/stat/SamplerTest.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kotlinx.coroutines.runBlocking import kotlin.test.Test diff --git a/kmath-prob/src/jvmTest/kotlin/kscience/kmath/prob/StatisticTest.kt b/kmath-stat/src/jvmTest/kotlin/kscience/kmath/stat/StatisticTest.kt similarity index 96% rename from kmath-prob/src/jvmTest/kotlin/kscience/kmath/prob/StatisticTest.kt rename to kmath-stat/src/jvmTest/kotlin/kscience/kmath/stat/StatisticTest.kt index 22ca472a8..5cee4d172 100644 --- a/kmath-prob/src/jvmTest/kotlin/kscience/kmath/prob/StatisticTest.kt +++ b/kmath-stat/src/jvmTest/kotlin/kscience/kmath/stat/StatisticTest.kt @@ -1,4 +1,4 @@ -package kscience.kmath.prob +package kscience.kmath.stat import kotlinx.coroutines.flow.drop import kotlinx.coroutines.flow.first diff --git a/settings.gradle.kts b/settings.gradle.kts index 7ece3f25c..fa9edcf22 100644 --- a/settings.gradle.kts +++ b/settings.gradle.kts @@ -10,8 +10,8 @@ pluginManagement { maven("https://dl.bintray.com/kotlin/kotlin-dev/") } - val toolsVersion = "0.6.1-dev-1.4.20-M1" - val kotlinVersion = "1.4.20-M1" + val toolsVersion = "0.6.4-dev-1.4.20-M2" + val kotlinVersion = "1.4.20-M2" plugins { id("kotlinx.benchmark") version "0.2.0-dev-20" @@ -34,11 +34,11 @@ include( ":kmath-histograms", ":kmath-commons", ":kmath-viktor", - ":kmath-prob", + ":kmath-stat", ":kmath-dimensions", ":kmath-for-real", ":kmath-geometry", ":kmath-ast", - ":examples", - ":kmath-ejml" + ":kmath-ejml", + ":examples" )