NN.h

Go to the documentation of this file.

#ifndef NN_H
#define NN_H
 
#include "FunctionApproximator.h"       
#include <vector>
 
class NN : public FunctionApproximator {
public:
        NN(int inputs, int hiddens, int outputs, double learnrate);
        NN(NN *);
        NN(FunctionApproximator *);
        NN(char *);
        ~NN();
        void load(const char *);
        void load(FILE *);
        void load(const NN *);
        void load(const FunctionApproximator *fa) { load((NN*)fa); }
        static bool validSaveFile(char *fname);
        void save(const char *);
        void save(FILE *);
 
        double train(std::vector<double> &input, std::vector<double> &target);
        double *test(const std::vector<double> &input);
        double GetInput(std::vector<double> &input, const std::vector<double> &target);
        double train(std::vector<unsigned int> &input, std::vector<double> &target);
        double *test(const std::vector<unsigned int> &input);
 
        int getNumInputs() { return inputs; }
        double getInputWeight(int inp, int outp=0) { return weights[0][outp][inp]; }
 
        void Print();
private:
        void allocateMemory(const NN *nn = 0);
        void freeMemory();
        
        std::vector< std::vector< std::vector<double> > > weights;
        std::vector< std::vector< std::vector<double> > > updatedweights;
        std::vector< std::vector< std::vector<double> > > errors;
        std::vector<double> hidden;
        std::vector<double> output;
        int inputs, hiddens, outputs;
        
        double g(double a);
        double dg(double a);
        double outputerr(const std::vector<double> &output, const std::vector<double> &expected, int which);
        double internalerr(const std::vector<double> &output, const std::vector<double> &expected, int which);
        double internalinputerr(const std::vector<double> &output, const std::vector<double> &expected, int which);
        double error(const std::vector<double> &outputs);
};
 
 
#endif