使用gsl做最小二乘法的非线性拟合
由 hubdog 于 星期一, 2022-01-10 17:20 发表
拟合的函数方程是
y=alpha*exp(-kappa*x);
拟合后能正确算出alpha和kappa的值,用gsl写这个玩意好啰嗦,用python的curve_fit10行代码就能搞定。
type PFitData=^TFitData; TFitData=record t:Pdouble; y:Pdouble; n:Integer; end; function market(alpha,kappa, t:double):double; begin result:=alpha*exp(-kappa*t); end; function func_f(x: Pgsl_vector; params: Pointer; f: Pgsl_vector): Integer;cdecl; var data:PFitData; alpha:double; kappa:double; i:Integer; t,yi,y:double; begin data:= PFitData(params); alpha := gsl_vector_get(x, 0); kappa := gsl_vector_get(x, 1); for I := 0 to data^.n-1 do begin t := data^.t[i]; yi := data^.y[i]; y := market(alpha,kappa, t); gsl_vector_set(f, i, yi - y); end; Result:= Ord(GSL_SUCCESS); end; procedure TFormMain.btnASClick(Sender: TObject); var fdf:gsl_multifit_nlinear_fdf; n:Integer; p:Integer; x:pgsl_vector; fdf_params:gsl_multifit_nlinear_parameters; r:pgsl_rng; i:Integer; t:double; fit_data:TFitData; alpha,kappa:double; tp:pgsl_multifit_nlinear_type; gsl_multifit_nlinear_trs_lmaccel2:function():Pgsl_multifit_nlinear_trs;cdecl; gsl_vector_set2:procedure (v: Pgsl_vector; i: NativeUInt; x: Double); cdecl; begin n:=300; //number of data points to fit p:=2; //parameters count x := gsl_vector_alloc(p); fdf_params :=gsl_multifit_nlinear_default_parameters(); GetMem(fit_data.t, n * sizeof(double)); GetMem(fit_data.y, n * sizeof(double)); fit_data.n := n; //generate simulate data for I := 0 to n-1 do begin t:=i/n; fit_data.t[i]:=t; fit_data.y[i]:=market(3,4, t); end; // /* define function to be minimized */ FillChar(fdf, sizeof(gsl_multifit_nlinear_fdf),0); fdf.f := @func_f; fdf.df := nil;//func_df; fdf.fvv := nil;//func_fvv; fdf.n := n; fdf.p := p; fdf.params := @fit_data; // /* starting point */ gsl_vector_set(x, 0, 1.0); gsl_vector_set(x, 1, 1.0); fdf_params.trs := gsl_multifit_nlinear_trs_lmaccel; // solve_system(x, &fdf, &fdf_params); tp:=gsl_multifit_nlinear_trust; var max_iter:Integer := 200; var xtol:double := 1.0e-8; var gtol:double := 1.0e-8; var ftol:double := 1.0e-8; //fdf_params.trs^.alloc(@fdf,n,p); var work:pgsl_multifit_nlinear_workspace := gsl_multifit_nlinear_alloc(tp, @fdf_params, n, p); var f2:pgsl_vector := gsl_multifit_nlinear_residual(work); var y:pgsl_vector := gsl_multifit_nlinear_position(work); var info:integer; var chisq0, chisq, rcond:double; ///* initialize solver */ gsl_multifit_nlinear_init(x, @fdf, work); //* store initial cost */ gsl_blas_ddot(f2, f2, @chisq0); //* iterate until convergence */ gsl_multifit_nlinear_driver(max_iter, xtol, gtol, ftol, nil, nil, @info, work); //* store final cost */ gsl_blas_ddot(f2, f2, @chisq); //* store cond(J(x)) */ gsl_multifit_nlinear_rcond(@rcond, work); gsl_vector_memcpy(x, y); gsl_multifit_nlinear_free(work); alpha := gsl_vector_get(x, 0); kappa := gsl_vector_get(x, 1); gsl_vector_free(x); end;