refactor: Remove least squares cost function using Gauss-Newton approximation

This was a test implementation in the first place, and we are not
using it since -- as expected -- it only make sense when initial
values are close to the minimum.
Moreover, it is easy to achieve the same behavior by just providing
an all-zero model Hessian to the normal least squares cost function.
The performance difference was negligible for all test cases.

Author: AdrianDBS

minuit2.net/CostFunctions/CostFunction.cs

@@ -381,68 +381,6 @@ public static ICostFunction LeastSquares(
             errorDefinitionInSigma,
             false);
     }
-    
-    public static ICostFunction LeastSquaresWithGaussNewtonApproximation(
-        IReadOnlyList<double> x,
-        IReadOnlyList<double> y,
-        IReadOnlyList<string> parameters,
-        Func<double, IReadOnlyList<double>, double> model,
-        Func<double, IReadOnlyList<double>, IReadOnlyList<double>> modelGradient,
-        double errorDefinitionInSigma = 1)
-    {
-        DataValidation.EnsureMatchingSizesBetween((x, nameof(x)), (y, nameof(y)));
-        return new LeastSquaresWithGaussNewtonApproximation(
-            x,
-            y,
-            _ => 1,
-            parameters,
-            model,
-            modelGradient,
-            errorDefinitionInSigma,
-            true);
-    }
-    
-    public static ICostFunction LeastSquaresWithGaussNewtonApproximation(
-        IReadOnlyList<double> x,
-        IReadOnlyList<double> y,
-        double yError,
-        IReadOnlyList<string> parameters,
-        Func<double, IReadOnlyList<double>, double> model,
-        Func<double, IReadOnlyList<double>, IReadOnlyList<double>> modelGradient,
-        double errorDefinitionInSigma = 1)
-    {
-        DataValidation.EnsureMatchingSizesBetween((x, nameof(x)), (y, nameof(y)));
-        return new LeastSquaresWithGaussNewtonApproximation(
-            x,
-            y,
-            _ => yError,
-            parameters,
-            model,
-            modelGradient,
-            errorDefinitionInSigma,
-            false);
-    }
-    
-    public static ICostFunction LeastSquaresWithGaussNewtonApproximation(
-        IReadOnlyList<double> x,
-        IReadOnlyList<double> y,
-        IReadOnlyList<double> yError,
-        IReadOnlyList<string> parameters,
-        Func<double, IReadOnlyList<double>, double> model,
-        Func<double, IReadOnlyList<double>, IReadOnlyList<double>> modelGradient,
-        double errorDefinitionInSigma = 1)
-    {
-        DataValidation.EnsureMatchingSizesBetween((x, nameof(x)), (y, nameof(y)), (yError, nameof(yError)));
-        return new LeastSquaresWithGaussNewtonApproximation(
-            x,
-            y,
-            i => yError[i],
-            parameters,
-            model,
-            modelGradient,
-            errorDefinitionInSigma,
-            false);
-    }
 
     public static ICostFunction Sum(params ICostFunction[] components) => new CostFunctionSum(components);
 }

minuit2.net/CostFunctions/LeastSquaresWithGaussNewtonApproximation.cs

@@ -35,13 +35,10 @@ public class LeastSquaresCostBuilder(
         private bool _hasYErrors = true;
         private bool _hasGradient;
         private bool _hasHessian;
-        private bool _isUsingGaussNewtonApproximation;
         private double _errorDefinitionInSigma = 1;
 
         public ICostFunction Build() => _hasYErrors switch
         {
-            true when _isUsingGaussNewtonApproximation => CostFunction.LeastSquaresWithGaussNewtonApproximation(xValues, yValues, yError, _parameterNames, model, modelGradient, _errorDefinitionInSigma),
-            false when _isUsingGaussNewtonApproximation => CostFunction.LeastSquaresWithGaussNewtonApproximation(xValues, yValues, _parameterNames, model, modelGradient, _errorDefinitionInSigma),
             true when _hasHessian => CostFunction.LeastSquares(xValues, yValues, yError, _parameterNames, model, modelGradient, modelHessian, _errorDefinitionInSigma),
             false when _hasHessian => CostFunction.LeastSquares(xValues, yValues, _parameterNames, model, modelGradient, modelHessian, _errorDefinitionInSigma),
             true when _hasGradient => CostFunction.LeastSquares(xValues, yValues, yError, _parameterNames, model, modelGradient, _errorDefinitionInSigma),
@@ -68,12 +65,6 @@ public LeastSquaresCostBuilder WithHessian(bool hasHessian = true)
             return this;
         }
 
-        public LeastSquaresCostBuilder UsingGaussNewtonApproximation(bool isUsingGaussNewtonApproximation = true)
-        {
-            _isUsingGaussNewtonApproximation = isUsingGaussNewtonApproximation;
-            return this;
-        }
-        
         public LeastSquaresCostBuilder WithErrorDefinition(double sigma)
         {
             _errorDefinitionInSigma = sigma;