Exploratory analysis tool
Linkspotter is a package of R software that allows to calculate and visualize using a graph all the bivariate links of a dataset.
Its main features are:¶
- calculation of several correlation matrices corresponding to different link coefficients.
- clustering of variables using an unsupervised learning.
- supervised discretization of one or a couple of variables.
It also offers a customizable user interface, allowing to:¶
- Visualize the links using a graph (the variables correspond to the nodes and the links correspond to the edges)
- View the distribution of each variable using its histogram or barplot
- Visualize a link between a couple of variables using scatter plots, box plots, etc.
Available link coefficients are:¶
- Pearson’s r
- Spearman’s rho
- Kendall’s tau
- the Maximal Information Coefficient (MIC)
- the distance correlation
- the Maximal Normalized Mutual Information (MaxNMI)
The MaxNMI is a bivariate link coefficient, which uses a supervised discretization called Best Equal-Frequency Discretization (BeEF). Both are introduced by Linkspotter. The interest of this new link coefficient is that it enables to assess and compare the link between a couple of variables, whatever their types (quantitative vs qualitative, quantitative vs quantitative, qualitative vs qualitative).
Linkspotter uses and combines some features coming from other R packages, namely infotheo, minerva, energy, mclust, Hmisc, shiny, visNetwork and rAmCharts.