/**
   \file logchisq.hpp
   \brief chi-square statistic
   \author Junhua Gu
 */

#ifndef LOG_CHI_SQ_HPP
#define LOG_CHI_SQ_HPP
#define OPT_HEADER
#include <core/fitter.hpp>
#include <iostream>
#include <vector>
#include <misc/optvec.hpp>
#include <cmath>
using std::cerr;using std::endl;

namespace opt_utilities
{
  class negative_data_value
    :public opt_exception
  {
  public:
    negative_data_value()
      :opt_exception("log chisq statistics cannot be used when has negative data!")
    {}
  };



  /**
     \brief chi-square statistic
     \tparam Ty the return type of model
     \tparam Tx the type of the self-var
     \tparam Tp the type of model parameter
     \tparam Ts the type of the statistic
     \tparam Tstr the type of the string used
   */
  template<typename Ty,typename Tx,typename Tp,typename Ts,typename Tstr>
  class logchisq
    :public statistic<Ty,Tx,Tp,Ts,Tstr>
  {
  private:
    bool verb;
    int n;


    statistic<Ty,Tx,Tp,Ts,Tstr>* do_clone()const
    {
      // return const_cast<statistic<Ty,Tx,Tp>*>(this);
      return new logchisq<Ty,Tx,Tp,Ts,Tstr>(*this);
    }

    const char* do_get_type_name()const
    {
      return "chi^2 statistic";
    }

  public:
    void verbose(bool v)
    {
      verb=v;
    }
  public:
    logchisq()
      :verb(false)
    {}



    Ts do_eval(const Tp& p)
    {
      Ts result(0);
      for(int i=(this->get_data_set()).size()-1;i>=0;--i)
	{
	  Ty y=std::log(this->get_data_set().get_data(i).get_y());
	  Ty ym=std::log(this->eval_model(this->get_data_set().get_data(i).get_x(),p));
	  Ty ye1=std::log(1+this->get_data_set().get_data(i).get_y_upper_err()/this->get_data_set().get_data(i).get_y());
	  Ty chi=(y-ym)/ye1;
	  result+=chi*chi;
	}
      if(verb)
	{
	  n++;
	  if(n%10==0)
	    {

	      cerr<<result<<"\t";
	      for(size_t i=0;i<get_size(p);++i)
		{
		  cerr<<get_element(p,i)<<",";
		}
	      cerr<<endl;
	    }

	}

      return result;
    }
  };

#if 1

  template<>
  class logchisq<double,double,std::vector<double>,double,std::string>
    :public statistic<double,double,std::vector<double> ,double,std::string>
  {
  public:
    typedef double Ty;
    typedef double Tx;
    typedef std::vector<double> Tp;
    typedef double Ts;
    typedef std::string Tstr;
  private:
    bool verb;
    int n;

    statistic<Ty,Tx,Tp,Ts,Tstr>* do_clone()const
    {
      // return const_cast<statistic<Ty,Tx,Tp>*>(this);
      return new logchisq<Ty,Tx,Tp,Ts,Tstr>(*this);
    }

    const char* do_get_type_name()const
    {
      return "chi^2 statistics (specialized for double)";
    }
  public:
    void verbose(bool v)
    {
      verb=v;
    }
  public:
    logchisq()
      :verb(false)
    {}



    Ty do_eval(const Tp& p)
    {
      Ty result(0);
      for(size_t i=0;i!=(this->get_data_set()).size();++i)
	{

	  Ty y_model=this->eval_model(this->get_data_set().get_data(i).get_x(),p);
	  Ty y_obs=this->get_data_set().get_data(i).get_y();
	  Ty y_err;

	  if(y_model>y_obs)
	    {
	      y_err=std::abs(this->get_data_set().get_data(i).get_y_upper_err());
	    }
	  else
	    {
	      y_err=-std::abs(this->get_data_set().get_data(i).get_y_lower_err());
	    }
	  if(y_obs+y_err<0)
	    {
	      throw negative_data_value();
	    }
	  Ty logy=std::log(y_obs);
	  Ty logym=std::log(y_model);
	  Ty logerr=std::log(y_obs+y_err)-log(y_obs);


	  Ty chi=(logy-logym)/logerr;

	  //	  Ty chi=(this->get_data_set().get_data(i).get_y()-this->eval_model(this->get_data_set().get_data(i).get_x(),p));
	  //	  cerr<<chi<<"\n";
	  result+=chi*chi;
	  //std::cerr<<chi<<std::endl;
	  //cerr<<this->eval_model(this->get_data_set()[i].x,p)<<endl;
	  //cerr<<this->get_data_set()[i].y_upper_err<<endl;
	  //	  cerr<<this->get_data_set()[i].x<<"\t"<<this->get_data_set()[i].y<<"\t"<<this->eval_model(this->get_data_set()[i].x,p)<<endl;
	}
      if(verb)
	{
	  n++;
	  if(n%10==0)
	    {
	      cerr<<result<<"\t";
	      for(int i=0;i<(int)get_size(p);++i)
		{
		  cerr<<get_element(p,i)<<",";
		}
	      cerr<<endl;
	    }

	}

      return result;
    }
  };
#endif

}

#endif
//EOF