source: trunk/lib/statistics/Linear.h @ 385

// $Id: Linear.h 385 2005-08-13 09:24:18Z jari $

#ifndef _theplu_statistics_regression_linear_
#define _theplu_statistics_regression_linear_

#include <c++_tools/statistics/OneDimensional.h>
#include <c++_tools/gslapi/vector.h>

#include <cmath>

namespace theplu {
namespace statistics {  
namespace regression {

  ///
  /// Class for OneDimensional.   
  ///
  
  class Linear : public OneDimensional 
  {
  
  public:
    ///
    /// Default Constructor.
    ///
    inline Linear(void)
      : OneDimensional(), alpha_(0), alpha_var_(0), beta_(0), beta_var_(0),
        m_x_(0), s2_(0) {}

    ///
    /// Destructor 
    ///
    virtual ~Linear(void) {};
         
    ///
    /// @return the parameter \f$ \alpha \f$
    ///
    inline double alpha(void) const { return alpha_; }

    ///
    /// @return standard deviation of parameter \f$ \alpha \f$
    ///
    inline double alpha_err(void) const { return sqrt(alpha_var_); }

    ///
    /// @return the parameter \f$ \beta \f$
    ///
    inline double beta(void) const { return beta_; }

    ///
    /// @return standard deviation of parameter \f$ \beta \f$
    ///
    inline double beta_err(void) const { return sqrt(beta_var_); }
    
    ///
    /// This function computes the best-fit linear regression
    /// coefficients \f$ (\alpha, \beta)\f$ of the model \f$ y =
    /// \alpha + \beta (x-m_x) \f$ from vectors \a x and \a y, by
    /// minimizing \f$ \sum{(y_i - \alpha - \beta (x-m_x))^2} \f$. By
    /// construction \f$ \alpha \f$ and \f$ \beta \f$ are independent.
    ///
    void fit(const gslapi::vector& x, const gslapi::vector& y) ;
    
    ///
    /// This function computes the best-fit linear regression
    /// coefficients \f$ (\alpha, \beta)\f$ of the model \f$ y =
    /// \alpha + \beta (x-m_x) \f$ from vectors \a x and \a y, by
    /// minimizing \f$ \sum{w_i(y_i - \alpha - \beta (x-m_x))^2} \f$,
    /// where \f$ m_x \f$ is the weighted average. By construction \f$
    /// \alpha \f$ and \f$ \beta \f$ are independent.
    ///
    void fit(const gslapi::vector& x, const gslapi::vector& y,
             const gslapi::vector& w);
    
    ///
    /// Function predicting value using the linear model. \a y_err is
    /// the expected deviation from the line for a new data point. If
    /// weights are used a weight can be specified for the new point.
    ///
    inline void predict(const double x, double& y, double& y_err, 
                        const double w=1.0) 
    { 
      y = alpha_ + beta_ * (x-m_x_); 
      y_err = sqrt( alpha_var_+beta_var_*(x-m_x_)*(x-m_x_)+s2_/w ); 
      x_=x;
      y_=y;
      y_err_=y_err;
    }

    ///
    /// @return prediction value and parameters
    ///
    std::ostream& print(std::ostream&) const;
             
    ///
    /// @return header for print()
    ///
    std::ostream& print_header(std::ostream&) const;

    ///
    /// Function returning the coefficient of determination,
    /// i.e. share of variance explained by the linear model.
    ///
    inline double r2(void) const { return r2_; }

  private:
    ///
    /// Copy Constructor. (not implemented)
    ///
    Linear(const Linear&);

    double alpha_;
    double alpha_var_;
    double beta_;
    double beta_var_;
    double m_x_; // average of x values
    double s2_; // var(y|x)
    double r2_; // coefficient of determination
  };

}}} // of namespaces regression, statisitcs and thep

#endif