architecture - c++ loose coupling function dependency -

as title might suggest having hard time describing architectural problem , same reason searching has been without result. writing command line application aims take in two-dimensional data physics model can fitted means of least squares fitting. have implemented nelder-mead , levenberg-marquardt routine work properly, i.e. given set of data, model , cost function, optimize array of fitting parameters.

for purpose want write class facilitates abstract functionality of optimization, independent of actual model , cost function used.

class fit { public:     fit();     ~fit();      run();  private:     std::vector<std::vector<double>> m_data;     std::vector<double> m_parameters;     std::vector<double> m_evaluated; } 

the actual optimization method run() following:

void run() {     // set initial guess parameters     m_parameters = {...};      while(condition) {         double cost_value;         // evaluate model function current parameters         (*model)(&m_evaluated, m_data, m_parameters);         (*cost)(&cost_value, m_data, m_evaluated);          if (cost_value < threshold_value)             break;         else             update_parameters(&m_parameters);     } } 

this questions comes in. fit::run() method needs call model , cost function there no knowing front these or function arguments require. arguments take in example bare minimum required arguments seen fitting procedure there might additional required arguments.

specific cost functions , model defined elsewhere have no clue how let fit class know functions needs call. coming c thinking function pointers have no clue how preferably done in c++. should fit have function pointer member m_model has set after instantiated? spitballing here. how best approach , problem called maybe additional approaches?

edit: give better idea of trying accomplish pseudo code.

// imagine multiple available functions work differently take array of parameters  // used evaluate function set of data values void model_a(std::vector<double>* y, std::vector<std::vector<double>> data, std::vector<double> p) {     (int i; < data[0].size(); ++i)         (*y)[i] = parameters[0] + parameters[1]*data[0][i]; };  void model_b(std::vector<double>* y, std::vector<std::vector<double>> data, std::vector<double> p) {     (int i; < data[0].size(); ++i)         (*y)[i] = parameters[0] - pow(data[0][i], parameters[1]); };  // assume have data std::vector<std::vector<double>> data = {};  fit fit = fit(); fit->set_data(data) fit->set_model(model_a); // question fit->run();  // can extract , display optimized parameter array std::vector<double> p = fit.parameters(); (int i; < p.size(); ++i)     std::cout << p[i] << std::endl; 

to me, sounds either want interface class provides interface, or making class hieararchy, data stored in base-class, , derived class knows how calculate cost, etc.

there several ways can implement - if want data persist within class, pass in either @ construction or later, in instance of cost model.

unfortunately, have rush now, later check in comments, etc.