Exemplary usage of our functions is presented through a script performing our algorithm
on one of our samples. On a computer with MATLAB (version newer than/or 2013b) please right
click file 'example.m' and select option 'Run' in order to start computations.

During the calculations there will appear some figures which will be further saved.

Here is brief description of what this script does. It is also commented, almost line by
line to make it easy to analyze it.

File 'example.m' contains code for reading the data from file 'sample_1.mat' and calling
recursive partitioning procedure. Folder 'libs' contains all libraries necessary for this
purpose. Result will be saved to the newly created folder 'sample_1_results' in the same
directory as extracted data.

In data file there must be present following variables:
- 'data', which is matrix of n by p size, where n is the number of features and p is the
  number of observations, containing features in rows and observations in columns;
- 'group_xy_region', where 2-nd row contains coordinates on abscissa and 3-rd on ordinate
  axes and there are p columns, where p is the number of observations.

For considered sample, noise filtration is applied. Clusterization is conducted with usage
of Pearson correlative distance. Maximal number of 10 clusters is set while total level of
nesting reaches three.

Folder with results will be organised in the same manner as our algorithm worked: as a tree
with root at top level and for every obtained cluster a branch (directory) is created
recursively.
In such a folder there are following elements:
- 'levelX.mat' file where X denotes number of levels considered further (so for 3 levels
  of nesting, 'level3.mat' will be created at top level, 'level2.mat' at lower level and
  'level1.mat' at the lowest level)
- 'primary__A_features__k_B__dunn_C__mono.png' where A denotes the number of features used
  in clusterization process, B denotes the number of clusters found and C the Dunn's index
  for obtained partition. This file contains grayscale image showing clusters assignment in
  space. The darker the cluster is marked, the higher is the number of cluster used.
- 'downmerged_mono.png' also shows clusters assignment in space, however taking into
  account all clusters generated in recursion process, so the total number of clusters
  shown is higher than for the specific step.
- 'clusterD' where D denotes the number of cluster for which results of analysis is stored
  inside. D ranges from 1 to B, however it may happen, that some of the folders are not
  present. This can be caused by insufficient number of objects in specific cluster to be
  partitioned (analysis is omitted when cluster contains less than 0.1% of whole sample) or
  no partition made sense (there were empty clusters and Dunn index was infinite).
- 'kmeans_cache' is usually large directory with plenty of files inside, however it is used
  by our implementation of k-means algorithms to cache results (this directory is useful
  when disrupting calculations, however in some cases is unnecessary - next version will
  provide possibility to disable this feature)
- 'cache' is much smaller directory with less files, however allows to quickly recover
  results when all calculations were once completed (also useful during disruptions).

Declared RAM usage is optimal right now, as this option is deprecated and different setting
may influence negatively time needed for calculations with absolutely no gain in any other
parameter.

Two calls of recursive partitioning for the same data are done:
- first one shows time of calculations in total
- second one shows time of calculations when results have been cached on hard disk
This approach of caching results allows user to disrupt calculation process and resume it
with significantly lower loss as compared to usual disruption. It may be advantage for
anyone using notebook PC for calculations but has to stop them in many cases. Also, in case
of messing result files (as for now there are many of them generated), they can be just
deleted (keeping just cache files) and function called with the same parameters as
previously will reconstruct proper structure of files in very short time.

For further informations of features implemented, one may check help provided in
documentation of k_means function implemented in our library.