elbow.py

"""!

@brief Elbow method to determine the optimal number of clusters for k-means clustering.
@details Implementation based on paper @cite article::cluster::elbow::1.

@authors Andrei Novikov (pyclustering@yandex.ru)
@date 2014-2018
@copyright GNU Public License

@cond GNU_PUBLIC_LICENSE
    PyClustering 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.

    PyClustering is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with this program.  If not, see <http://www.gnu.org/licenses/>.
@endcond

"""


import math

from pyclustering.cluster.kmeans import kmeans
from pyclustering.cluster.center_initializer import kmeans_plusplus_initializer, random_center_initializer
from pyclustering.core.wrapper import ccore_library

import pyclustering.core.elbow_wrapper as wrapper


class elbow:
    """!
    @brief Class represents Elbow method that is used to find out appropriate amount of clusters in a dataset.
    @details Elbow method performs clustering using K-Means algorithm for each K and estimate clustering results using
              sum of square erros. By default K-Means++ algorithm is used to calculate initial centers that are used by
              K-Means algorithm.

    The Elbow is determined by max distance from each point (x, y) to segment from kmin-point (x0, y0) to kmax-point (x1, y1),
    where 'x' is K (amount of clusters), and 'y' is within-cluster error. Following expression is used to calculate Elbow
    length:
    \f[Elbow_{k} = \frac{\left ( y_{0} - y_{1} \right )x_{k} + \left ( x_{1} - x_{0} \right )y_{k} + \left ( x_{0}y_{1} - x_{1}y_{0} \right )}{\sqrt{\left ( x_{1} - x_{0} \right )^{2} + \left ( y_{1} - y_{0} \right )^{2}}}\f]

    Usage example of Elbow method for cluster analysis:
    @code
        # read sample 'Simple3' from file (sample contains four clusters)
        sample = read_sample(SIMPLE_SAMPLES.SAMPLE_SIMPLE3)

        # create instance of Elbow method using K value from 1 to 10.
        kmin, kmax = 1, 10
        elbow_instance = elbow(sample, kmin, kmax)

        # process input data and obtain results of analysis
        elbow_instance.process()
        amount_clusters = elbow_instance.get_amount()   # most probable amount of clusters
        wce = elbow_instance.get_wce()                  # total within-cluster errors for each K

        # perform cluster analysis using K-Means algorithm
        centers = kmeans_plusplus_initializer(sample, amount_clusters).initialize()
        kmeans_instance = kmeans(sample, centers)
        kmeans_instance.process()

        # obtain clustering results and visualize them
        clusters = kmeans_instance.get_clusters()
        centers = kmeans_instance.get_centers()
        kmeans_visualizer.show_clusters(sample, clusters, centers)
    @endcode

    By default Elbow uses K-Means++ initializer to calculate initial centers for K-Means algorithm, it can be changed
    using argument 'initializer':
    @code
        # perform analysis using Elbow method with random center initializer for K-Means algorithm inside of the method.
        kmin, kmax = 1, 10
        elbow_instance = elbow(sample, kmin, kmax, initializer=random_center_initializer)
        elbow_instance.process()
    @endcode

    @image html elbow_example_simple_03.png "Elbows analysis with further K-Means clustering."

    """

    def __init__(self, data, kmin, kmax, **kwargs):
        """!
        @brief Construct Elbow method.

        @param[in] data (array_like): Input data that is presented as array of points (objects), each point should be represented by array_like data structure.
        @param[in] kmin (int): Minimum amount of clusters that should be considered.
        @param[in] kmax (int): Maximum amount of clusters that should be considered.
        @param[in] **kwargs: Arbitrary keyword arguments (available arguments: 'ccore', 'initializer').

        <b>Keyword Args:</b><br>
            - ccore (bool): If True then CCORE (C++ implementation of pyclustering library) is used (be default True).
            - initializer (callable): Center initializer that is used by K-Means algorithm (by default K-Means++).

        """
        if kmax - kmin < 3:
            raise ValueError("Amount of K (" + str(kmax - kmin) + ") is too small for analysis. "
                             "It is require to have at least three K to build elbow.")

        if len(data) < kmax:
            raise ValueError("K max value '%d' is greater than amount of points in data '%d'." % (kmax, len(data)))

        self.__initializer = kwargs.get('initializer', kmeans_plusplus_initializer)

        self.__ccore = kwargs.get('ccore', True) or \
                       isinstance(self.__initializer, kmeans_plusplus_initializer) or \
                       isinstance(self.__initializer, random_center_initializer)

        if self.__ccore:
            self.__ccore = ccore_library.workable()

        self.__data = data
        self.__kmin = kmin
        self.__kmax = kmax

        self.__wce = []
        self.__elbows = []
        self.__kvalue = -1


    def process(self):
        """!
        @brief Performs analysis to find out appropriate amount of clusters.

        """
        if self.__ccore:
            self.__process_by_ccore()
        else:
            self.__process_by_python()


    def __process_by_ccore(self):
        """!
        @brief Performs processing using C++ implementation.

        """
        if isinstance(self.__initializer, kmeans_plusplus_initializer):
            initializer = wrapper.elbow_center_initializer.KMEANS_PLUS_PLUS
        else:
            initializer = wrapper.elbow_center_initializer.RANDOM

        result = wrapper.elbow(self.__data, self.__kmin, self.__kmax, initializer)

        self.__kvalue = result[0]
        self.__wce = result[1]


    def __process_by_python(self):
        """!
        @brief Performs processing using python implementation.

        """
        for amount in range(self.__kmin, self.__kmax):
            centers = self.__initializer(self.__data, amount).initialize()
            instance = kmeans(self.__data, centers, ccore=True)
            instance.process()

            self.__wce.append(instance.get_total_wce())

        self.__calculate_elbows()
        self.__find_optimal_kvalue()


    def get_amount(self):
        """!
        @brief Returns appropriate amount of clusters.

        """
        return self.__kvalue


    def get_wce(self):
        """!
        @brief Returns list of total within cluster errors for each K-value (kmin, kmin + 1, ..., kmax - 1).

        """

        return self.__wce


    def __calculate_elbows(self):
        """!
        @brief Calculates potential elbows.
        @details Elbow is calculated as a distance from each point (x, y) to segment from kmin-point (x0, y0) to kmax-point (x1, y1).

        """

        x0, y0 = 0.0, self.__wce[0]
        x1, y1 = float(len(self.__wce)), self.__wce[-1]

        for index_elbow in range(1, len(self.__wce) - 1):
            x, y = float(index_elbow), self.__wce[index_elbow]

            segment = abs((y0 - y1) * x + (x1 - x0) * y + (x0 * y1 - x1 * y0))
            norm = math.sqrt((x1 - x0) ** 2 + (y1 - y0) ** 2)
            distance = segment / norm

            self.__elbows.append(distance)


    def __find_optimal_kvalue(self):
        """!
        @brief Finds elbow and returns corresponding K-value.

        """
        optimal_elbow_value = max(self.__elbows)
        self.__kvalue = self.__elbows.index(optimal_elbow_value) + 1 + self.__kmin