//
//Please find details of the method from
//
//	Q. Zhang, A. Zhou and Y. Jin, "RM-MEDA: A Regularity Model Based Multiobjective Estimation of Distribution Algorithm", IEEE Trans. Evolutionary Computation, Vol. 12, no. 1, pp41-63, 2008.
//
//The source codes are free for reserach work. If you have any problem with the source codes, please contact with 
//	Qingfu Zhang, 
//		Department of Computing and Electronic Systems,
//		University of Essex,
//		Colchester, CO4 3SQ, UK
//		http://cswww.essex.ac.uk/staff/zhang
//		Email: qzhang@essex.ac.uk
//    Aimin Zhou
//		Department of Computing and Electronic Systems,
//		University of Essex,
//		Colchester, CO4 3SQ, UK
//		http://privatewww.essex.ac.uk/~azhou/
//		Email: azhou@essex.ac.uk or amzhou@gmail.com 
//Programmer:		
//		Aimin Zhou
//Last Update:
//		Feb. 21, 2008
//
//main.cpp is the RM-MEDA main program
//
//Please run command 'Front('F')' or 'Front('X')' in Matlab to view the final results 
//

#include <cmath>
#include <ctime>
#include <vector>
#include "Random.h"
#include "Parameter.h"
#include "Initialization.h"
#include "Alg.h"
#include "GenMod.h"
#include "Sel.h"
#include "Problem.h"

// test problem
// min (f1(X),...,fm(X))
// g1(X)  = 0; ... gk(X)  = 0;
// h1(X) <= 0; ... hs(X) <= 0;
// F is objective vector (return)
// E is equality constraint vector (return)
// I is inequality constraint vector (return)
// X is decision vector (input)

// main function
int main()
{
	// NOTE: the following parameters MUST be set correctly and the other parameters may use default values
	// population size, maximum generation, number of clusters,
	// number of objectives, number of equality constraints, number of inequality constraints, fitness function,
	// dimension of decision variables, and their boundaries

	unsigned int i, 
		popsize,	// population size
		maxgen,		// maximum generation
		cluster;	// number of clusters

	//------------------------------------------------------------------------
	// set random seed
	az::rnd::seed((long) time(NULL));

	//------------------------------------------------------------------------
	// algorithm parameters
	popsize = 200;	// population size
	maxgen	= 200;	// maximum generation
	cluster = 5;	// number of clusters

	az::mea::CParameter	mPar;	// parameter object
	
	mPar.TolF()	= 1.0E-5;		// objective tolerance
	mPar.TolX()	= 1.0E-5;		// decision variable tolerance
	mPar.TolC()	= 1.0E-5;		// constraint tolerance

	mPar.FSize( 3 );			// set number of objectives
	mPar.ESize( 0 );			// set number of equality constraints
	mPar.ISize( 0 );			// set number of inequality constraints
	mPar.Evaluator( PROBLEM::ZZJ08_F8 );		// set fitness function
	mPar.XSize(30);				// set dimension of decision variables
	mPar.XCoding() = false;		// do not normalize the decision varialbes
	for(i=0; i<mPar.XSize(); i++)				// set the boundaries of decision variables
	{ 
		mPar.XRealLow(i) = 0.0;					//	[XRealLow, XRealUpp] is the real boundary (the boundary of search space)
		mPar.XRealUpp(i) = 1.0;					//
		mPar.XLow(i)	 = mPar.XRealLow(i);	//	[XLow, XUpp] is the working boundary
		mPar.XUpp(i)	 = mPar.XRealUpp(i);	//  since the decision variables are not normalized, [XLow, XUpp] = [XRealLow, XRealUpp]
	}
	//mPar.XCoding() = true;		//  normalize the decision varialbes if the range of search space is quite different, for exammple, -1 <= x1 <= 0, -1000 <= x2 <= 1000
	//for(i=0; i<mPar.XSize(); i++)				// set the boundaries of decision variables
	//{ 
	//	mPar.XRealLow(i) = low[i];				//	[XRealLow, XRealUpp] is the real boundary (the boundary of search space)
	//	mPar.XRealUpp(i) = upp[i];				//
	//	mPar.XLow(i)	 = 0.0;					//	[XLow, XUpp] is the working boundary
	//	mPar.XUpp(i)	 = 1.0;					//  it is normalized to [XLow, XUpp] = [0.0, 1.0]
	//}

	//------------------------------------------------------------------------
	// initialize population and algorithm
	az::mea::CPopulationMO inipop(mPar);
	az::mea::ini::Random init;
	init.Initialize(inipop, popsize);					// set populatin size here

	typedef az::mea::CEA< az::mea::gen::mod::ModelLocalPCAU, az::mea::sel::SCrowd2 > EA;
	EA* pEA			 = new EA(maxgen, mPar, inipop);	// set maximum generation here
	pEA->EvaTimes() += init.EvaTimes();
	// set parameters of Local PCA method
	pEA->Set( mPar.FSize()-1,							// dimension of manifold of Pareto set = number of objectives - 1				 
			  cluster,									// number of clusters partitioned by Local PCA 
			  50,										// training steps in Local PCA, 50 (default, usually it is more than enough)
			  0.25);									// extension ratio, 0.25 (default)

	//------------------------------------------------------------------------
	// main loop
	while(!pEA->IsTerminate())
	{
		pEA->Step();
		std::cout<<pEA->CurStep()<<std::endl;
	}

	//------------------------------------------------------------------------
	// save final population (nondominated solutions)
	// format is: 
	// number of point (n)
	// number of objectives (m) number of decision variables (k)
	// obj11 ... obj1m x11 ... x1k
	// ...
	// objn1 ... objnm xn1 ... xnk
	// in Matlab, use command 'Front('F')' to observe the Pareto front
	// and command 'Front('X')' to observe the Pareto set
	pEA->Population().Write("pop.dat");

	std::cout<<"Over."<<std::endl;

   return 0;
}