/*!
  \file
  \brief Probability distributions for Mixtures of pdfs
  \author Vaclav Smidl.

  -----------------------------------
  BDM++ - C++ library for Bayesian Decision Making under Uncertainty

  Using IT++ for numerical operations
  -----------------------------------
*/

#ifndef MX_H
#define MX_H

#include "libBM.h"
#include "libEF.h"
//#include <std>

using namespace itpp;

/*!
* \brief Mixture of epdfs

Density function:
\f[
f(x) = \sum_{i=1}^{n} w_{i} f_i(x), \quad \sum_{i=1}^n w_i = 1.
\f]
where \f$f_i(x)\f$ is any density on random variable \f$x\f$, called \a component,

*/
class emix : public epdf {
protected:
	//! weights of the components
	vec w;
	//! Component (epdfs)
	Array<epdf*> Coms;
public:
	//!Default constructor
	emix ( RV &rv ) : epdf ( rv ) {};
	//! Set weights \c w and components \c R
	void set_parameters ( const vec &w, const Array<epdf*> &Coms );

	vec sample() const;
	vec mean() const {
		int i; vec mu = zeros ( rv.count() );
		for ( i = 0;i < w.length();i++ ) {mu += w ( i ) * Coms ( i )->mean(); }
		return mu;
	}
	double evalpdflog ( const vec &val ) const {
		int i;
		double sum = 0.0;
		for ( i = 0;i < w.length();i++ ) {sum += w ( i ) * Coms ( i )->evalpdflog ( val );}
		return log ( sum );
	};

//Access methods
	//! returns a pointer to the internal mean value. Use with Care!
	vec& _w() {return w;}
};

/*! \brief Chain rule decomposition of epdf

Probability density in the form of Chain-rule decomposition:
\[
f(x_1,x_2,x_3) = f(x_1|x_2,x_3)f(x_2,x_3)f(x_3)
\]
Note that
*/
class mprod: public compositepdf, public mpdf {
protected:
	// pointers to epdfs
	Array<epdf*> epdfs;
	//! Indeces of rvc in common rvc
	Array<ivec> rvcinds;
public:
	/*!\brief Constructor from list of mFacs,
	Additional parameter overlap is left for future use. Do not set to true for mprod.
	*/
	mprod ( Array<mpdf*> mFacs): compositepdf(mFacs), mpdf(getrv(true),RV()), epdfs(n), rvcinds(n)
	{ 	setrvc(rv,rvc);
		setrvcinrv(rvc,rvcinds);
		setindices(rv); 
	for(int i=0;i<n;i++){epdfs(i)=&(mpdfs(i)->_epdf());}
	};

	double evalpdflog ( const vec &val ) const {
		int i;
		double res = 0.0;
		for ( i = n - 1;i > 0;i++ ) {
			if ( rvcinds ( i ).length() > 0 )
				{mpdfs ( i )->condition ( val ( rvcinds ( i ) ) );}
			// add logarithms
			res += epdfs ( i )->evalpdflog ( val ( rvsinrv ( i ) ) );
		}
		return res;
	}
	vec samplecond ( const vec &cond, double &ll ) {
		vec smp=zeros ( rv.count() );
		vec condi;
		vec smpi;
		ll = 0;
		for ( int i = ( n - 1 );i >= 0;i-- ) {
			if ( rvcinds ( i ).length() > 0 ) {
				condi = zeros ( rvcsinrv.length() + rvcinds.length() );
				// copy data in condition
				set_subvector ( condi,rvcinds ( i ), cond );
				// copy data in already generated sample
				set_subvector ( condi,rvcsinrv ( i ), smp );

				mpdfs ( i )->condition ( condi );
			}
			smpi = epdfs ( i )->sample();
			// copy contribution of this pdf into smp
			set_subvector ( smp,rvsinrv ( i ), smpi );
			// add ith likelihood contribution
			ll+=epdfs ( i )->evalpdflog ( smpi );
		}
		return smp;
	}
	mat samplecond ( const vec &cond, vec &ll, int N ) {
		mat Smp ( rv.count(),N );
		for ( int i=0;i<N;i++ ) {Smp.set_col ( i,samplecond ( cond,ll ( i ) ) );}
		return Smp;
	}

	~mprod() {};
};

//! Product of independent epdfs. For dependent pdfs, use mprod.
class eprod: public epdf {
protected:
	//! Components (epdfs)
	Array<const epdf*> epdfs;
	//! Array of indeces
	Array<ivec> rvinds;
public:
	eprod ( const Array<const epdf*> epdfs0 ) : epdf ( RV() ),epdfs ( epdfs0 ),rvinds ( epdfs.length() ) {
		bool independent=true;
		for ( int i=0;i<epdfs.length();i++ ) {
			independent=rv.add ( epdfs ( i )->_rv() );
			it_assert_debug ( independent==true, "eprod:: given components are not independent ." );
			rvinds ( i ) = ( epdfs ( i )->_rv() ).dataind ( rv );
		}
	}

	vec mean() const {
		vec tmp ( rv.count() );
		for ( int i=0;i<epdfs.length();i++ ) {
			vec pom = epdfs ( i )->mean();
			set_subvector ( tmp,rvinds ( i ), pom );
		}
		return tmp;
	}
	vec sample() const {
		vec tmp ( rv.count() );
		for ( int i=0;i<epdfs.length();i++ ) {
			vec pom = epdfs ( i )->sample();
			set_subvector ( tmp,rvinds ( i ), pom );
		}
		return tmp;
	}
	double evalpdflog ( const vec &val ) const {
		double tmp=0;
		for ( int i=0;i<epdfs.length();i++ ) {
			tmp+=epdfs(i)->evalpdflog(val(rvinds(i)));
		}
		return tmp;
	}
	//!access function
	const epdf* operator () (int i) const {it_assert_debug(i<epdfs.length(),"wrong index");return epdfs(i);}
};


/*! \brief Mixture of mpdfs with constant weights, all mpdfs are of equal type

*/
class mmix : public mpdf {
protected:
	//! Component (epdfs)
	Array<mpdf*> Coms;
	//!Internal epdf
	emix Epdf;
public:
	//!Default constructor
	mmix ( RV &rv, RV &rvc ) : mpdf ( rv, rvc ), Epdf ( rv ) {ep = &Epdf;};
	//! Set weights \c w and components \c R
	void set_parameters ( const vec &w, const Array<mpdf*> &Coms ) {
		Array<epdf*> Eps ( Coms.length() );

		for ( int i = 0;i < Coms.length();i++ ) {
			Eps ( i ) = & ( Coms ( i )->_epdf() );
		}
		Epdf.set_parameters ( w, Eps );
	};

	void condition ( const vec &cond ) {
		for ( int i = 0;i < Coms.length();i++ ) {Coms ( i )->condition ( cond );}
	};
};
#endif //MX_H