source: trunk/lib/statistics/Local.h @ 429

// $Id: Local.h 429 2005-12-08 19:50:11Z peter $

#ifndef _theplu_statistics_regression_local_
#define _theplu_statistics_regression_local_

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


namespace theplu {
namespace statistics {
namespace regression {

  ///
  /// Class for Locally weighted regression. 
  ///
  /// Locally weighted regression is an algorithm for learning
  /// continuous non-linear mappings in a non-parametric manner.  In
  /// locally weighted regression, points are weighted by proximity to
  /// the current x in question using a Kernel. A weighted regression
  /// is then computed using the weighted points and a specific
  /// Regression method. This procedure is repeated, which results in
  /// a pointwise approximation of the underlying (unknown) function.
  ///
  class Local
  {
  
  public:
    ///
    /// Constructor taking type of \a regressor,
    /// type of \a kernel.
    ///
    inline Local(OneDimensionalWeighted& r, Kernel& k) 
      : kernel_(&k), regressor_(&r) {}

    ///
    /// Destructor 
    ///
    virtual ~Local(void) {};


    ///
    /// adding a data point
    ///
    inline void add(const double x, const double y) 
    { data_.push_back(std::make_pair(x,y)); }

    ///
    /// Function returning the points where to predict
    ///
    inline const std::vector<double>& x(void) const { return x_; }

    ///
    /// Function returning predicted values
    ///
    inline const std::vector<double>& y(void) const { return y_; }

    ///
    /// Function returning error of predictions
    ///
    inline const std::vector<double>& y_err(void) const { return y_err_; }
  
    ///
    /// Performs the fit in data defined by add using a
    /// RegressionKernel and a Regression method defined in the
    /// constructor. The function starts by regressing over the lowest
    /// x value, followed by stepping forward \a step_size number of
    /// points, et cetera. The kernel is centralized over the x-value
    /// in question and the width is set so \a fraction of all points
    /// are used. The result is sent to ostream \a s, using print
    /// function in used Regression.
    ///
    void fit(std::ostream& s, const double fraction, const u_int step_size=1);


  private:
    ///
    /// Copy Constructor. (Not implemented)
    ///
    Local(const Local&);

    std::vector<std::pair<double, double> > data_;
    Kernel* kernel_;
    OneDimensionalWeighted* regressor_;
    std::vector<double> x_; 
    std::vector<double> y_;
    std::vector<double> y_err_;

  };

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

#endif