LinearRegression.h

Go to the documentation of this file.

/*
 *  LinearRegression.h
 *  hog2
 *
 *  Created by Nathan Sturtevant on 6/1/06.
 *  Copyright 2006 Nathan Sturtevant. All rights reserved.
 *
 */
 
#ifndef LinearRegression_H
#define LinearRegression_H
 
#include "FPUtil.h"
#include "FunctionApproximator.h"
#include <cassert>
 
class updateData {
public:
        updateData() :mean(0), S(0), totErr(0), n(0) {}
        void reset() { mean = 0.0; S = 0.0; totErr = 0.0; n = 0; }
        double mean;
        double S;
        double totErr;
        int n;
};
 
class LinearRegression : public FunctionApproximator  {
public:
        LinearRegression(int _inputs, int _outputs, double learnrate);
        LinearRegression(LinearRegression *);
        LinearRegression(char *);
        LinearRegression(FunctionApproximator *);
        ~LinearRegression();
        
        void setUseBinary(bool binary) { useBinary = binary; }
        
        void resizeInputs(int newSize, double newVal);
        void resizeInputs(int newSize);
 
        void save(const char *);
        void save(FILE *);
        void load(const char *);
        void load(FILE *);
        void load(const FunctionApproximator *fa) { load((LinearRegression*)fa); }
        void load(const LinearRegression *);
        static bool validSaveFile(char *fname);
        
        double train(std::vector<double> &input, std::vector<double> &output2);
        double *test(const std::vector<double> &input);
        double train(std::vector<unsigned int> &input, std::vector<double> &output2);
        double *test(const std::vector<unsigned int> &input);
        
        void Print();
        
        int getNumInputs() { return inputs; }
        void setInputWeight(double value, unsigned int weightNum, unsigned int whichOutput=0);
        double getInputWeight(int inp, int outp=0) { return weight[outp][inp]; }
        void getWeightUpdateVariance(std::vector<double> &var, unsigned int which=0);
        double getWeightUpdateVariance(unsigned int weightNum, unsigned int whichOutput=0);
 
        void getWeightUpdateAverage(std::vector<double> &var, unsigned int which=0);
        double getWeightUpdateAverage(unsigned int weightNum, unsigned int whichOutput=0);
 
        void getWeightUpdateSum(std::vector<double> &var, unsigned int which=0);
        double getWeightUpdateSum(unsigned int weightNum, unsigned int whichOutput=0);
        void resetWeightVariance(unsigned int weightNum, unsigned int whichOutput=0);
        void resetWeightVariance();
 
        int getWeightFrequency(unsigned int weightNum, unsigned int whichOutput=0);
 
        double GetInput(std::vector<double> &input, const std::vector<double> &target)
        { assert(!"Unimplemented"); }
 
private:
        void allocateMemory();
        void freeMemory();
        
        double g(double a);
        double dg(double a);
        double outputerr(std::vector<double> &output, std::vector<double> &expected, int which);
        
        double error(double* output);
        std::vector<std::vector<double> > weight;
        std::vector<std::vector<updateData> > updates;
        //std::vector<double > weight;
        std::vector<double> output;
        //double** weight;
        //double* output;
        int inputs, outputs;
        bool useBinary;
};
 
#endif