All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.apache.commons.math3.ode.JacobianMatrices Maven / Gradle / Ivy

Go to download

A Java's Collaborative Filtering library to carry out experiments in research of Collaborative Filtering based Recommender Systems. The library has been designed from researchers to researchers.

The newest version!
/*
 * Licensed to the Apache Software Foundation (ASF) under one or more
 * contributor license agreements.  See the NOTICE file distributed with
 * this work for additional information regarding copyright ownership.
 * The ASF licenses this file to You under the Apache License, Version 2.0
 * (the "License"); you may not use this file except in compliance with
 * the License.  You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
package org.apache.commons.math3.ode;

import java.lang.reflect.Array;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;

import org.apache.commons.math3.exception.DimensionMismatchException;
import org.apache.commons.math3.exception.MathIllegalArgumentException;
import org.apache.commons.math3.exception.MaxCountExceededException;
import org.apache.commons.math3.exception.util.LocalizedFormats;

/**
 * This class defines a set of {@link SecondaryEquations secondary equations} to
 * compute the Jacobian matrices with respect to the initial state vector and, if
 * any, to some parameters of the primary ODE set.
 * 

* It is intended to be packed into an {@link ExpandableStatefulODE} * in conjunction with a primary set of ODE, which may be: *

    *
  • a {@link FirstOrderDifferentialEquations}
  • *
  • a {@link MainStateJacobianProvider}
  • *
* In order to compute Jacobian matrices with respect to some parameters of the * primary ODE set, the following parameter Jacobian providers may be set: *
    *
  • a {@link ParameterJacobianProvider}
  • *
  • a {@link ParameterizedODE}
  • *
*

* * @see ExpandableStatefulODE * @see FirstOrderDifferentialEquations * @see MainStateJacobianProvider * @see ParameterJacobianProvider * @see ParameterizedODE * * @since 3.0 */ public class JacobianMatrices { /** Expandable first order differential equation. */ private ExpandableStatefulODE efode; /** Index of the instance in the expandable set. */ private int index; /** FODE with exact primary Jacobian computation skill. */ private MainStateJacobianProvider jode; /** FODE without exact parameter Jacobian computation skill. */ private ParameterizedODE pode; /** Main state vector dimension. */ private int stateDim; /** Selected parameters for parameter Jacobian computation. */ private ParameterConfiguration[] selectedParameters; /** FODE with exact parameter Jacobian computation skill. */ private List jacobianProviders; /** Parameters dimension. */ private int paramDim; /** Boolean for selected parameters consistency. */ private boolean dirtyParameter; /** State and parameters Jacobian matrices in a row. */ private double[] matricesData; /** Simple constructor for a secondary equations set computing Jacobian matrices. *

* Parameters must belong to the supported ones given by {@link * Parameterizable#getParametersNames()}, so the primary set of differential * equations must be {@link Parameterizable}. *

*

Note that each selection clears the previous selected parameters.

* * @param fode the primary first order differential equations set to extend * @param hY step used for finite difference computation with respect to state vector * @param parameters parameters to consider for Jacobian matrices processing * (may be null if parameters Jacobians is not desired) * @exception DimensionMismatchException if there is a dimension mismatch between * the steps array {@code hY} and the equation dimension */ public JacobianMatrices(final FirstOrderDifferentialEquations fode, final double[] hY, final String... parameters) throws DimensionMismatchException { this(new MainStateJacobianWrapper(fode, hY), parameters); } /** Simple constructor for a secondary equations set computing Jacobian matrices. *

* Parameters must belong to the supported ones given by {@link * Parameterizable#getParametersNames()}, so the primary set of differential * equations must be {@link Parameterizable}. *

*

Note that each selection clears the previous selected parameters.

* * @param jode the primary first order differential equations set to extend * @param parameters parameters to consider for Jacobian matrices processing * (may be null if parameters Jacobians is not desired) */ public JacobianMatrices(final MainStateJacobianProvider jode, final String... parameters) { this.efode = null; this.index = -1; this.jode = jode; this.pode = null; this.stateDim = jode.getDimension(); if (parameters == null) { selectedParameters = null; paramDim = 0; } else { this.selectedParameters = new ParameterConfiguration[parameters.length]; for (int i = 0; i < parameters.length; ++i) { selectedParameters[i] = new ParameterConfiguration(parameters[i], Double.NaN); } paramDim = parameters.length; } this.dirtyParameter = false; this.jacobianProviders = new ArrayList(); // set the default initial state Jacobian to the identity // and the default initial parameters Jacobian to the null matrix matricesData = new double[(stateDim + paramDim) * stateDim]; for (int i = 0; i < stateDim; ++i) { matricesData[i * (stateDim + 1)] = 1.0; } } /** Register the variational equations for the Jacobians matrices to the expandable set. * @param expandable expandable set into which variational equations should be registered * @throws DimensionMismatchException if the dimension of the partial state does not * match the selected equations set dimension * @exception MismatchedEquations if the primary set of the expandable set does * not match the one used to build the instance * @see ExpandableStatefulODE#addSecondaryEquations(SecondaryEquations) */ public void registerVariationalEquations(final ExpandableStatefulODE expandable) throws DimensionMismatchException, MismatchedEquations { // safety checks final FirstOrderDifferentialEquations ode = (jode instanceof MainStateJacobianWrapper) ? ((MainStateJacobianWrapper) jode).ode : jode; if (expandable.getPrimary() != ode) { throw new MismatchedEquations(); } efode = expandable; index = efode.addSecondaryEquations(new JacobiansSecondaryEquations()); efode.setSecondaryState(index, matricesData); } /** Add a parameter Jacobian provider. * @param provider the parameter Jacobian provider to compute exactly the parameter Jacobian matrix */ public void addParameterJacobianProvider(final ParameterJacobianProvider provider) { jacobianProviders.add(provider); } /** Set a parameter Jacobian provider. * @param parameterizedOde the parameterized ODE to compute the parameter Jacobian matrix using finite differences */ public void setParameterizedODE(final ParameterizedODE parameterizedOde) { this.pode = parameterizedOde; dirtyParameter = true; } /** Set the step associated to a parameter in order to compute by finite * difference the Jacobian matrix. *

* Needed if and only if the primary ODE set is a {@link ParameterizedODE}. *

*

* Given a non zero parameter value pval for the parameter, a reasonable value * for such a step is {@code pval * FastMath.sqrt(Precision.EPSILON)}. *

*

* A zero value for such a step doesn't enable to compute the parameter Jacobian matrix. *

* @param parameter parameter to consider for Jacobian processing * @param hP step for Jacobian finite difference computation w.r.t. the specified parameter * @see ParameterizedODE * @exception UnknownParameterException if the parameter is not supported */ public void setParameterStep(final String parameter, final double hP) throws UnknownParameterException { for (ParameterConfiguration param: selectedParameters) { if (parameter.equals(param.getParameterName())) { param.setHP(hP); dirtyParameter = true; return; } } throw new UnknownParameterException(parameter); } /** Set the initial value of the Jacobian matrix with respect to state. *

* If this method is not called, the initial value of the Jacobian * matrix with respect to state is set to identity. *

* @param dYdY0 initial Jacobian matrix w.r.t. state * @exception DimensionMismatchException if matrix dimensions are incorrect */ public void setInitialMainStateJacobian(final double[][] dYdY0) throws DimensionMismatchException { // Check dimensions checkDimension(stateDim, dYdY0); checkDimension(stateDim, dYdY0[0]); // store the matrix in row major order as a single dimension array int i = 0; for (final double[] row : dYdY0) { System.arraycopy(row, 0, matricesData, i, stateDim); i += stateDim; } if (efode != null) { efode.setSecondaryState(index, matricesData); } } /** Set the initial value of a column of the Jacobian matrix with respect to one parameter. *

* If this method is not called for some parameter, the initial value of * the column of the Jacobian matrix with respect to this parameter is set to zero. *

* @param pName parameter name * @param dYdP initial Jacobian column vector with respect to the parameter * @exception UnknownParameterException if a parameter is not supported * @throws DimensionMismatchException if the column vector does not match state dimension */ public void setInitialParameterJacobian(final String pName, final double[] dYdP) throws UnknownParameterException, DimensionMismatchException { // Check dimensions checkDimension(stateDim, dYdP); // store the column in a global single dimension array int i = stateDim * stateDim; for (ParameterConfiguration param: selectedParameters) { if (pName.equals(param.getParameterName())) { System.arraycopy(dYdP, 0, matricesData, i, stateDim); if (efode != null) { efode.setSecondaryState(index, matricesData); } return; } i += stateDim; } throw new UnknownParameterException(pName); } /** Get the current value of the Jacobian matrix with respect to state. * @param dYdY0 current Jacobian matrix with respect to state. */ public void getCurrentMainSetJacobian(final double[][] dYdY0) { // get current state for this set of equations from the expandable fode double[] p = efode.getSecondaryState(index); int j = 0; for (int i = 0; i < stateDim; i++) { System.arraycopy(p, j, dYdY0[i], 0, stateDim); j += stateDim; } } /** Get the current value of the Jacobian matrix with respect to one parameter. * @param pName name of the parameter for the computed Jacobian matrix * @param dYdP current Jacobian matrix with respect to the named parameter */ public void getCurrentParameterJacobian(String pName, final double[] dYdP) { // get current state for this set of equations from the expandable fode double[] p = efode.getSecondaryState(index); int i = stateDim * stateDim; for (ParameterConfiguration param: selectedParameters) { if (param.getParameterName().equals(pName)) { System.arraycopy(p, i, dYdP, 0, stateDim); return; } i += stateDim; } } /** Check array dimensions. * @param expected expected dimension * @param array (may be null if expected is 0) * @throws DimensionMismatchException if the array dimension does not match the expected one */ private void checkDimension(final int expected, final Object array) throws DimensionMismatchException { int arrayDimension = (array == null) ? 0 : Array.getLength(array); if (arrayDimension != expected) { throw new DimensionMismatchException(arrayDimension, expected); } } /** Local implementation of secondary equations. *

* This class is an inner class to ensure proper scheduling of calls * by forcing the use of {@link JacobianMatrices#registerVariationalEquations(ExpandableStatefulODE)}. *

*/ private class JacobiansSecondaryEquations implements SecondaryEquations { /** {@inheritDoc} */ public int getDimension() { return stateDim * (stateDim + paramDim); } /** {@inheritDoc} */ public void computeDerivatives(final double t, final double[] y, final double[] yDot, final double[] z, final double[] zDot) throws MaxCountExceededException, DimensionMismatchException { // Lazy initialization if (dirtyParameter && (paramDim != 0)) { jacobianProviders.add(new ParameterJacobianWrapper(jode, pode, selectedParameters)); dirtyParameter = false; } // variational equations: // from d[dy/dt]/dy0 and d[dy/dt]/dp to d[dy/dy0]/dt and d[dy/dp]/dt // compute Jacobian matrix with respect to primary state double[][] dFdY = new double[stateDim][stateDim]; jode.computeMainStateJacobian(t, y, yDot, dFdY); // Dispatch Jacobian matrix in the compound secondary state vector for (int i = 0; i < stateDim; ++i) { final double[] dFdYi = dFdY[i]; for (int j = 0; j < stateDim; ++j) { double s = 0; final int startIndex = j; int zIndex = startIndex; for (int l = 0; l < stateDim; ++l) { s += dFdYi[l] * z[zIndex]; zIndex += stateDim; } zDot[startIndex + i * stateDim] = s; } } if (paramDim != 0) { // compute Jacobian matrices with respect to parameters double[] dFdP = new double[stateDim]; int startIndex = stateDim * stateDim; for (ParameterConfiguration param: selectedParameters) { boolean found = false; for (int k = 0 ; (!found) && (k < jacobianProviders.size()); ++k) { final ParameterJacobianProvider provider = jacobianProviders.get(k); if (provider.isSupported(param.getParameterName())) { provider.computeParameterJacobian(t, y, yDot, param.getParameterName(), dFdP); for (int i = 0; i < stateDim; ++i) { final double[] dFdYi = dFdY[i]; int zIndex = startIndex; double s = dFdP[i]; for (int l = 0; l < stateDim; ++l) { s += dFdYi[l] * z[zIndex]; zIndex++; } zDot[startIndex + i] = s; } found = true; } } if (! found) { Arrays.fill(zDot, startIndex, startIndex + stateDim, 0.0); } startIndex += stateDim; } } } } /** Wrapper class to compute jacobian matrices by finite differences for ODE * which do not compute them by themselves. */ private static class MainStateJacobianWrapper implements MainStateJacobianProvider { /** Raw ODE without jacobians computation skill to be wrapped into a MainStateJacobianProvider. */ private final FirstOrderDifferentialEquations ode; /** Steps for finite difference computation of the jacobian df/dy w.r.t. state. */ private final double[] hY; /** Wrap a {@link FirstOrderDifferentialEquations} into a {@link MainStateJacobianProvider}. * @param ode original ODE problem, without jacobians computation skill * @param hY step sizes to compute the jacobian df/dy * @exception DimensionMismatchException if there is a dimension mismatch between * the steps array {@code hY} and the equation dimension */ MainStateJacobianWrapper(final FirstOrderDifferentialEquations ode, final double[] hY) throws DimensionMismatchException { this.ode = ode; this.hY = hY.clone(); if (hY.length != ode.getDimension()) { throw new DimensionMismatchException(ode.getDimension(), hY.length); } } /** {@inheritDoc} */ public int getDimension() { return ode.getDimension(); } /** {@inheritDoc} */ public void computeDerivatives(double t, double[] y, double[] yDot) throws MaxCountExceededException, DimensionMismatchException { ode.computeDerivatives(t, y, yDot); } /** {@inheritDoc} */ public void computeMainStateJacobian(double t, double[] y, double[] yDot, double[][] dFdY) throws MaxCountExceededException, DimensionMismatchException { final int n = ode.getDimension(); final double[] tmpDot = new double[n]; for (int j = 0; j < n; ++j) { final double savedYj = y[j]; y[j] += hY[j]; ode.computeDerivatives(t, y, tmpDot); for (int i = 0; i < n; ++i) { dFdY[i][j] = (tmpDot[i] - yDot[i]) / hY[j]; } y[j] = savedYj; } } } /** * Special exception for equations mismatch. * @since 3.1 */ public static class MismatchedEquations extends MathIllegalArgumentException { /** Serializable UID. */ private static final long serialVersionUID = 20120902L; /** Simple constructor. */ public MismatchedEquations() { super(LocalizedFormats.UNMATCHED_ODE_IN_EXPANDED_SET); } } }




© 2015 - 2024 Weber Informatics LLC | Privacy Policy