t-eoSymreg.cpp

#ifdef _MSC_VER
#pragma warning(disable:4786)
#endif

#include <gp/eoParseTree.h>
#include <eo>

using namespace gp_parse_tree;
using namespace std;

//-----------------------------------------------------------------------------

class SymregNode
{
public :

        enum Operator {X = 'x', Plus = '+', Min = '-', Mult = '*', PDiv = '/'};

        SymregNode(void)                                { init(); }
        SymregNode(Operator _op)                { op = _op; }
        virtual ~SymregNode(void)                       {}

        // arity function, need this function!
        int arity(void) const      { return op == X? 0 : 2; }

    void randomize(void) {}
    
        // evaluation function, single case, using first argument to give value of variable
        template <class Children>
        void operator()(double& result, Children args, double var) const
        {
        double r1, r2;

        if (arity() == 2)
        {
            args[0].apply(r1, var);
            args[1].apply(r2, var);
        }

                switch (op)
                {
                case Plus : result = r1 + r2; break;
                case Min  : result = r1 - r2; break;
                case Mult : result = r1 * r2; break;
                case PDiv : 
                        {
                                if (r2 == 0.0)
                                        result = 1.0; // protection a la Koza, realistic implementations should maybe throw an exception
                else
                                    result = r1 / r2;
                break;
                        }

                case X    : result = var; break;
                }

        }

    template <class Children>
        void operator()(string& result, Children args) const
    {
        static const string lb = "(";
        static const string rb = ")";
        char opStr[4] = "   ";
        opStr[1] = op;
        
            if (arity() == 0)
                {
            result = "x";
            return;
        }
        // else
        string r1;
        args[0].apply(r1);
        result = lb + r1;
        result += opStr; 
        args[1].apply(r1);
        result += r1 + rb;
    }

    Operator getOp(void) const { return op; }    

protected :

        void init(void)                                 { op = X; }

private :

        Operator op; // the type of node
};

static SymregNode init_sequence[5] = {SymregNode::X, SymregNode::Plus, SymregNode::Min, SymregNode::Mult, SymregNode::PDiv}; // needed for intialization

// MSVC does not recognize the lt_arity<Node> in eoParseTreeDepthInit
// without this specialization ...
// 2 months later, it seems it does not accept this definition ...
// but dies accept the lt_arity<Node> in eoParseTreeDepthInit
// !!!
// #ifdef _MSC_VER 
// template <>
// bool lt_arity(const SymregNode &node1, const SymregNode &node2) 
// {
//              return (node1.arity() < node2.arity());
// }
// #endif

//-----------------------------------------------------------
// saving, loading

std::ostream& operator<<(std::ostream& os, const SymregNode& eot)
{
    os << static_cast<char>(eot.getOp());
    return os;
}

std::istream& operator>>(std::istream& is, SymregNode& eot)
{
    char type;
    type = (char) is.get();
    eot = SymregNode(static_cast<SymregNode::Operator>(type));
    return is;
}


//-----------------------------------------------------------------------------
double targetFunction(double x)
{ 
        return x * x * x * x - x * x * x + x * x * x - x * x + x - 10; 
}

// parameters controlling the sampling of points
const double xbegin = -10.0f;
const double xend   = 10.0f;
const double xstep  = 1.3f; 

template <class FType, class Node> struct RMS: public eoEvalFunc< eoParseTree<FType, Node> > 
{
public :

    typedef eoParseTree<FType, Node> EoType;

    typedef eoParseTree<FType, Node> argument_type;
    typedef double                   fitness_type;

        RMS(void) : eoEvalFunc<EoType>()
        {
                int n = int( (xend - xbegin) / xstep);
                
                inputs.resize(n);
                target.resize(n);

                int i = 0;

        for (double x = xbegin; x < xend && i < n; ++i, x+=xstep)
                {
                        target[i] = targetFunction(x);
                        inputs[i] = x;
                }
        }

    ~RMS() {}
    
        void operator()( EoType & _eo )  
        {
                vector<double> outputs;
                outputs.resize(inputs.size());
                
        double fitness = 0.0;
        
        for (unsigned i = 0; i < inputs.size(); ++i)
        {
                    _eo.apply(outputs[i], inputs[i]);
                fitness += (outputs[i] - target[i]) * (outputs[i] - target[i]);
        }
                
        fitness /= (double) target.size();
        fitness = sqrt(fitness);
                        
                if (fitness > 1e+20)
                        fitness = 1e+20;

                _eo.fitness(fitness);
        }

private :
        vector<double> inputs;
        vector<double> target;
};

template <class EOT, class FitnessType>
void print_best(eoPop<EOT>& pop)
{
    std::cout << std::endl;
    FitnessType best = pop[0].fitness();
    int index = 0;

    for (unsigned i = 1; i < pop.size(); ++i)
    {
        if (best < pop[i].fitness())
        {
            best = pop[i].fitness();
            index = i;
        }
    }
    
    std::cout << "\t";
        
    string str;
    pop[index].apply(str);
    
    std::cout << str.c_str();
    std::cout << std::endl << "RMS Error = " << pop[index].fitness() << std::endl;
}

int main()
{
    typedef eoMinimizingFitness FitnessType;
    typedef SymregNode GpNode;

    typedef eoParseTree<FitnessType, GpNode> EoType;
    typedef eoPop<EoType> Pop;  

    const int MaxSize = 100;
    const int nGenerations = 10; // only a test, so few generations

    // Initializor sequence, contains the allowable nodes
    vector<GpNode> init(init_sequence, init_sequence + 5);

    // Depth Initializor, defaults to grow method.
    eoGpDepthInitializer<FitnessType, GpNode> initializer(10, init);
    
    // Root Mean Squared Error Measure
    RMS<FitnessType, GpNode>              eval;

    Pop pop(50, initializer);

    apply<EoType>(eval, pop);

    eoSubtreeXOver<FitnessType, GpNode>   xover(MaxSize);
    eoBranchMutation<FitnessType, GpNode> mutation(initializer, MaxSize);

    // The operators are  encapsulated into an eoTRansform object, 
    // that performs sequentially crossover and mutation
    eoSGATransform<EoType> transform(xover, 0.75, mutation, 0.25);

    // The robust tournament selection
    eoDetTournamentSelect<EoType> selectOne(2);   // tSize in [2,POPSIZE]
    // is now encapsulated in a eoSelectMany: 2 at a time -> SteadyState
    eoSelectMany<EoType> select(selectOne,2, eo_is_an_integer);    
  
    // and the Steady-State replacement
    eoSSGAWorseReplacement<EoType> replace;
  
    // Terminators
    eoGenContinue<EoType> term(nGenerations);

    eoCheckPoint<EoType> checkPoint(term);

    eoAverageStat<EoType>     avg;
    eoBestFitnessStat<EoType> best;
    eoStdoutMonitor monitor;

    checkPoint.add(monitor);
    checkPoint.add(avg);
    checkPoint.add(best);

    monitor.add(avg);
    monitor.add(best);

    // GP generation
    eoEasyEA<EoType> gp(checkPoint, eval, select, transform, replace);

    std::cout << "Initialization done" << std::endl;

    print_best<EoType, FitnessType>(pop);

    try
    {
      gp(pop);
    }
    catch (std::exception& e)
    {
            std::cout << "exception: " << e.what() << std::endl;;
            exit(EXIT_FAILURE);
    }

    print_best<EoType, FitnessType>(pop);

}

---