NEWS

DatabionicSwarm 2.0.0 (2024-06-20)

Major update to efficient parallelized CPU version of the central algorithm Pswarm.

Major bugfix of Quirin Stier in calcStressC.cpp and PswarmCurrentRadiusC2botsPositive: swarm allows every databot to change position by checking a new index. Priorly, databot with index 0 was frozen in the projection.
RelativeDifference can now be called silent and as a bugfix for vectors x and y optional parameter are now used

Moved RobustNormalization and RobustNorm_BackTrafo to DataVisualizations package available on CRAN
Corrected and extended UniquePoints
ClusteringAccuracy deleted. An improved version called ClusterAccuracy can be found in the FCPS package available on CRAN
DelaunayClassifiationError deleted. An new function with a better principle for evalaution the quality of DR methods called GabrielClassificationError can be found in the DRquality package available on CRAN.

GeneratePswarmVisualization reimplemented in a parallelized Cpp function, Pswarm will follow soon.

Bugfix in UniquePoints by Quirin Stier
DBSclustering allows now to set title and xlab if dendrogramm is plottet
Delaunay4Points changed argument Grid to LC and within the function LC is converted to Grid

Corrected notes of CRAN derived from documentation
Bugfixes in internal procedure of computing unique points for Delaunay4Points. Procedue is now available to the user as the function UniquePoints()
Print() replaced with message() in Pswarm()

Improved documentation significantly.
Pswarm now falls back to the dist function if the package parallelDist is not installed.
sESOM agorithm for the generalized Umatrix improved and is described now in the MethodsX publication of [Thrun/Ultsch, 2020]. It differs from the first publication of [Thrun, 2018].

Bugfix: checkInputDistancesOrData, DBSclustering, Pswarm, setGridSize and ShortestGraphPathsC: using unname in isSymmetric to check correctly of symmetric distance matrix
checkInputDistancesOrData is now used in RobustNormalization and Pswarm correctly and indicates in which function the errors lies.
Improvement: DBSclustering now provides the same colored branches (depending on frequency) as the topographicMap visualization with default colors. This makes a direct comparison of cluster structres more evident as it is clearly presented which projected point belongs to which high-dimensional point in the dendrogram.

Improvement: DBSclustering has now a colored dendrogram using package dendextend.
RobustNorm_BackTrafo added: transforms the Robust Normalization back.
bugfix: Delaunay4Points uses inherits instead of class to check for matrix (see https://developer.r-project.org/Blog/public/2019/11/09/when-you-think-class.-think-again/index.html)

bugfix: RelativeDifference now stops if non finite values in either x and y expaining the error.
bugfix: RobustNormalization fixed in some special cases like non finite values.
NoLevels of GeneralizedUmatrix integrated for fast plotting.

Delaunay Classification Error (DCE) added. DCE evaluates projection methods unbiased.

ClusteringAccuracy added. Given a prior Classification this function evaluates a clustering algorithm unbiased.
Now on GitHub.

RelativeDifference added which calculates the difference of a positive x and y value in the range [-2,2].