/*
----
This file is NOT part of SECONDO.
Authors: Greg Hamerly and Jonathan Drake
Feedback: hamerly@cs.baylor.edu
See: http://cs.baylor.edu/~hamerly/software/kmeans.php
Copyright 2014
The Annulus K-means algorithm is based on Hamerly's algorithm, but also 
sortsthe centers by their norms (distances from the origin). Doing this 
allows searching the centers using the norm of the point to exclude
centers that cannot be close.
----

//paragraph [1] Title: [{\Large \bf \begin{center}] [\end{center}}]
//[TOC] [\tableofcontents]

[1] Declarations for the annulus algorithm


1 Declarations for the annulus algorithm

*/

#ifndef ANNULUS_KMEANS_H
#define ANNULUS_KMEANS_H



#include "hamerly_kmeans.h"
#include <utility>

class AnnulusKmeans;

class AnnulusKmeans : public HamerlyKmeans {
    public:
        AnnulusKmeans() : xNorm(NULL), cOrder(NULL), guard(NULL) {}
        virtual ~AnnulusKmeans() { free(); }
        virtual void free();
        virtual void initialize(Dataset const *aX, unsigned short aK, 
        unsigned short *initialAssignment, int aNumThreads);
        virtual std::string getName() const { return "annulus"; }

    protected:
        virtual int runThread(int threadId, int maxIterations);
        void sort_means_by_norm();

        // The norm of each point.
        double *xNorm;

        // The order of the centers (first is the norm, second is the 
        //center
        // index).
        std::pair<double, int> *cOrder;

        // Guard is an index, for each point, of what is (or might be) the
        // second-closest center. When we tighten the bound, it is the
        // second-closest; however, this is not guaranteed to hold as the bounds
        // change. It is still useful as an index of a close center that is not
        // the closest.
        unsigned short *guard;
};

#endif