In Albluca/rpgraph: r interface to VDAOEngine to build elastic principal graphs

# library("knitr")
# library("rgl")
# knit_hooks$set(webgl = hook_webgl)

This vignette is intended to provide a very quick tutorial to use the rpgraph package.

An essential overview of Pircipal elastic circles

rpgraph prerequisite and installation

The package is currently under development and oly available on GitHub. It relises on the Java code written by Andrei Zynoviev available in the java library VDAOEngine (https://github.com/auranic/VDAOEngine). Therefore, a functional Java Virtual machine for your system is necessary. Before installing the package, it is advisable to install rJava from sources by using

install.packages(pkgs = "rJava", repos="http://rforge.net", type = 'source')

Compiling from source requires the appropriate development tools, e.g., C/C++ compiler. The installation of the package requires the devtools package, which is available from CRAN. The rpgraph package can be installed using

library(devtools)
install_github("Albluca/rpgraph")

To take advantage of all the feature of the package, it is advisable to also install (not necessary from source in this case) bigpca, flashpcaR, irlba, nsprcomp, and plotly. flashpcaR is available from GitHub

library(devtools)
install_github("gabraham/flashpca")

The other packages can be found on CRAN.

Workaround for common problems

The installation and loading of rJava is known to be problematics under certain circumstances on MacOS. A number of workaround can be found on the internet, possible solutions include recompiling the package, setting certain environment variables manually, ~ and changing the operating system~. Most of the times the problem is connected with the package struggling to find the appropriate information in the global environment.

If rJava fails to load try typing

options("java.home"="/Library/Java/JavaVirtualMachines/jdk1.7.0_79.jdk/Contents/Home/jre")
dyn.load('/Library/Java/JavaVirtualMachines/jdk1.7.0_79.jdk/Contents/Home/jre/lib/server/libjvm.dylib')

before loading the library. These lines may need to get adjusted depending on the version of the virtual machine that the user intendt to use.

A Few examples

The package contains a few ezample datasets thqt can be udes to test its functionalities.

Example 1 - Circle

This section describes how to produce a circular principal graph.

library(rpgraph)

Data <- simple_circle

Results <- computeElasticPrincipalGraph(Data = Data, NumNodes = 40, Method = 'CircleConfiguration')

Now Results contain the processed principal graph. Diagnostic information can be obtained using

plotMSDEnergyPlot(Results, Main = "Pincipal Circle", Cex.Main = 1)

and

accuracyComplexityPlot(Results, Main = "Pincipal Circle", Cex.Main = 1, Mode = 5)

It it also possible to zoom into a specific area of the accuracy/complexity plot by using the Xlims parameter.

accuracyComplexityPlot(Results, Main = "Pincipal Circle", Cex.Main = 1, Xlims = c(.97, .98))

Data can be plotted in 2D using the R built-in functions

plotData2D(Data = simple_circle, PrintGraph = Results,
           GroupsLab = rep(1, nrow(simple_circle)), Xlab = "Dimension 1", Ylab = "Dimension 2")

or plotly, which produces an interactive plot. Using plotly interactivelly requires running the code in RStudio (does it?)

plotData2D(Data = simple_circle, PrintGraph = Results, Plot.ly = TRUE,
           GroupsLab = rep(1, nrow(simple_circle)), Xlab = "Dimension 1", Ylab = "Dimension 2")

This commands will produce a list of warnings. Unfortunately there is not an easy way to fix it ...

The plotly graph can be exported on the web, for example on plot.ly using the instruction provided here.

Data can also be plotted in 3D using the functionalities provided by the rgl package

plotData3D(Data = simple_circle, PrintGraph = Results, Plot.ly = FALSE,
           GroupsLab = rep(1, nrow(simple_circle)), NodeSizeMult = 0.05,
           Xlab = "Dimension 1", Ylab = "Dimension 2", Zlab = "Dimension 3")

The code is not executed in the html version of the vignette. This can be fixed (in priciple) at a later stage of development.

For rgl to work correctly on MacOS, a proper X11 environment need to be installed. The most common solution is the installation of XQuartz. If the code crashes R, try using

library("rgl")
open3d()

before invoking the plotData3D function.

It is also possible to produce 3D plots using plot.ly.

plotData3D(Data = simple_circle, PrintGraph = Results, Plot.ly = TRUE,
           GroupsLab = rep(1, nrow(simple_circle)),
           Xlab = "Dimension 1", Ylab = "Dimension 2", Zlab = "Dimension 3")

Similarly to 2D, it is possible to export the plot to a web resource.

To see how different populations distribute among the nodes of the graph by providing a population identifier to each point in the original data (The Categories vector). In the following example, three populations are randomly assigned.

Net <- ConstructGraph(Results = Results, DirectionMat = NULL, Thr = 0.05)
TaxonList <- getTaxonMap(Results = Results, Data = Data)

InfoData <- plotPieNet(Results = Results, Data = simple_circle, NodeSizeMult = 4,
                       Categories = sample(1:3, nrow(simple_circle), replace = TRUE),
           Graph = Net, TaxonList = TaxonList, LayOut = 'circle', Main = "Pincipal Circle")

Example 2 - Curve

Data <- simple_circle

Results <- computeElasticPrincipalGraph(Data = Data, NumNodes = 20, Method = 'CurveConfiguration')

plotData2D(Data = simple_circle, PrintGraph = Results,
           GroupsLab = rep(1, nrow(simple_circle)), Xlab = "Dimension 1", Ylab = "Dimension 2")

Example 3 - tree

Data <- simple_tree

Results <- computeElasticPrincipalGraph(Data = Data, NumNodes = 40, Method = 'DefaultPrincipalTreeConfiguration')

plotData2D(Data = simple_tree, PrintGraph = Results,
           GroupsLab = rep(1, nrow(simple_circle)), Xlab = "Dimension 1", Ylab = "Dimension 2")

References

Albluca/rpgraph documentation built on May 5, 2019, 1:35 p.m.