KMeans.h

Go to the documentation of this file.

/*
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 3 of the License, or
 * (at your option) any later version.
 *
 * Written (W) 1999-2008 Gunnar Raetsch
 * Written (W) 2007-2009 Soeren Sonnenburg
 * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society
 */

#ifndef _KMEANS_H__
#define _KMEANS_H__

#include <shogun/lib/config.h>

#include <shogun/lib/common.h>
#include <shogun/io/SGIO.h>
#include <shogun/features/DenseFeatures.h>
#include <shogun/distance/Distance.h>
#include <shogun/machine/DistanceMachine.h>

namespace shogun
{
class CDistanceMachine;

enum EKMeansMethod
{
    KMM_MINI_BATCH,

    /* Standard KMeans with Lloyds algorithm */
    KMM_LLOYD
};

class CKMeans : public CDistanceMachine
{
    public:
        CKMeans();

        CKMeans(int32_t k, CDistance* d, EKMeansMethod f);

        CKMeans(int32_t k, CDistance* d, bool kmeanspp=false, EKMeansMethod f=KMM_LLOYD);

        CKMeans(int32_t k_i, CDistance* d_i, SGMatrix<float64_t> centers_i, EKMeansMethod f=KMM_LLOYD);
        virtual ~CKMeans();


        MACHINE_PROBLEM_TYPE(PT_MULTICLASS)

        
        virtual EMachineType get_classifier_type() { return CT_KMEANS; }

        virtual bool load(FILE* srcfile);

        virtual bool save(FILE* dstfile);

        void set_k(int32_t p_k);

        int32_t get_k();

        void set_use_kmeanspp(bool kmpp);

        bool get_use_kmeanspp() const;

        void set_fixed_centers(bool fixed);

        bool get_fixed_centers();

        void set_max_iter(int32_t iter);

        float64_t get_max_iter();

        SGVector<float64_t> get_radiuses();

        SGMatrix<float64_t> get_cluster_centers();

        int32_t get_dimensions();

        virtual const char* get_name() const { return "KMeans"; }

        virtual void set_initial_centers(SGMatrix<float64_t> centers);

        void set_train_method(EKMeansMethod f);

        EKMeansMethod get_train_method() const;

        void set_mbKMeans_batch_size(int32_t b);

        int32_t get_mbKMeans_batch_size() const;

        void set_mbKMeans_iter(int32_t t);

        int32_t get_mbKMeans_iter() const;

        void set_mbKMeans_params(int32_t b, int32_t t);

    private:
        virtual bool train_machine(CFeatures* data=NULL);

        virtual void store_model_features();

        virtual bool train_require_labels() const { return false; }

        SGMatrix<float64_t> kmeanspp();
        void init();

        void set_random_centers(SGVector<float64_t> weights_set, SGVector<int32_t> ClList, int32_t XSize);
        void set_initial_centers(SGVector<float64_t> weights_set,
                    SGVector<int32_t> ClList, int32_t XSize);
        void compute_cluster_variances();

    private:
        int32_t max_iter;

        bool fixed_centers;

        int32_t k;

        int32_t dimensions;

        SGVector<float64_t> R;

        SGMatrix<float64_t> mus_initial;

        bool use_kmeanspp;

        int32_t batch_size;

        int32_t minib_iter;

        SGMatrix<float64_t> mus;

        EKMeansMethod train_method;
};
}
#endif