/// <summary>
/// Initializes a new instance of the <see cref="NonlinearLeastSquares"/> class.
/// </summary>
///
/// <param name="regression">The regression model.</param>
/// <param name="algorithm">The <see cref="ILeastSquaresMethod">least squares</see>
/// algorithm to be used to estimate the regression parameters. Default is to
/// use a <see cref="LevenbergMarquardt">Levenberg-Marquardt</see> algorithm.</param>
///
public NonlinearLeastSquares(NonlinearRegression regression, ILeastSquaresMethod algorithm)
{
if (regression == null)
{
throw new ArgumentNullException("regression");
}
if (algorithm == null)
{
throw new ArgumentNullException("algorithm");
}
if (regression.Gradient == null)
{
throw new ArgumentException("The regression must have a gradient function defined.", "regression");
}
this.regression = regression;
this.NumberOfParameters = regression.Coefficients.Length;
this.solver = algorithm;
this.solver.Solution = regression.Coefficients;
this.solver.Function = new LeastSquaresFunction(regression.Function);
this.solver.Gradient = new LeastSquaresGradientFunction(regression.Gradient);
}
/// <summary>
/// Learns a model that can map the given inputs to the given outputs.
/// </summary>
/// <param name="x">The model inputs.</param>
/// <param name="y">The desired outputs associated with each <paramref name="x">inputs</paramref>.</param>
/// <param name="weights">The weight of importance for each input-output pair (if supported by the learning algorithm).</param>
/// <returns>
/// A model that has learned how to produce <paramref name="y" /> given <paramref name="x" />.
/// </returns>
public NonlinearRegression Learn(double[][] x, double[] y, double[] weights = null)
{
if (weights != null)
{
throw new ArgumentException(Accord.Properties.Resources.NotSupportedWeights, "weights");
}
if (NumberOfParameters == 0)
{
if (regression == null)
{
if (StartValues == null)
{
throw new InvalidOperationException("Please set the number of parameters, starting values, or the initial regression model.");
}
NumberOfParameters = StartValues.Length;
}
}
if (regression == null)
{
this.regression = new NonlinearRegression(numberOfParameters, function, gradient);
if (StartValues != null)
{
this.regression.Coefficients.SetTo(StartValues);
}
}
if (this.solver == null)
{
this.solver = new LevenbergMarquardt(numberOfParameters);
}
this.solver.NumberOfVariables = numberOfParameters;
this.solver.Solution = regression.Coefficients;
this.solver.Function = new LeastSquaresFunction(regression.Function);
this.solver.Gradient = new LeastSquaresGradientFunction(regression.Gradient);
this.solver.Token = Token;
double error = solver.Minimize(x, y);
if (Double.IsNaN(error) || Double.IsInfinity(error))
{
throw new Exception();
}
if (computeStandardErrors)
{
double[] errors = solver.StandardErrors;
for (int i = 0; i < errors.Length; i++)
{
regression.StandardErrors[i] = solver.StandardErrors[i];
}
}
return(regression);
}
/// <summary>
/// Initializes a new instance of the <see cref="NonlinearLeastSquares"/> class.
/// </summary>
///
/// <param name="regression">The regression model.</param>
/// <param name="algorithm">The <see cref="ILeastSquaresMethod">least squares</see>
/// algorithm to be used to estimate the regression parameters. Default is to
/// use a <see cref="LevenbergMarquardt">Levenberg-Marquardt</see> algorithm.</param>
///
public NonlinearLeastSquares(NonlinearRegression regression, ILeastSquaresMethod algorithm)
{
if (regression == null)
throw new ArgumentNullException("regression");
if (algorithm == null)
throw new ArgumentNullException("algorithm");
if (regression.Gradient == null)
throw new ArgumentException("The regression must have a gradient function defined.", "regression");
this.solver = algorithm;
this.solver.Solution = regression.Coefficients;
this.solver.Function = new LeastSquaresFunction(regression.Function);
this.solver.Gradient = new LeastSquaresGradientFunction(regression.Gradient);
this.regression = regression;
}