* other features (easy)
** pkgdown documentation
** add plot.legend in plotSC for 'lines' 'barplot' 'boxplot' and 'dendro3d'
** MC : ficelle -> check all features (especially 3D plots)
** MC : for lines, display all observations and not the mean
** MC : improved colors in prototypes/MDS because SC are very light
** improve predict for Heskes: maybe possible to do it entirely with a matrix
operation
** calculateProtoDist: use other type of neighbors (but add a parameter for the
radius because it should not affect the poly.dist plot)
** superclusters : make projGraph and output super-clusters for observations
** superclusters : allows to cut the super-classes afterward (method)
** add weights to observations
** kaski lagus
** implement weighted graphs projection
** change information about maxit (default = 5*n.obs, not 500 as claimed)
** add additional variables for superclasses plots

* other features (hard)
** grid.topology: hexagonal
** repeat SOM and combine
** multiple dissimilarities
** kacm, kdij
** batch version

* implementation enhancement
** should maybe improve the distance handling and propagate it everywhere
** colors handling in "names" seems to sometimes bugs
** make a function for initializing colors init.color(clustering, def.color)
** dissimilarity should maybe algorithm requires a "dist" type input

* package changes
** make unit test
** roxygenize documentation
** C++ for training
** improved modern graphics (ggplot2)

* shiny
** JB : remove variable names for somplot and scplot for Relational?
** JB : what is standard value for h in superClass?
** MC : add an option in input to indicate which variable contains the
observation name
** MC : possibility to choose additional variables within the original table
** fix bug in download file names
