liminal is an R package for constructing interactive visualisations designed for exploratory high-dimensional data analysis. It’s main purpose is to combine tours with (non-linear) dimension reduction algorithms to provide a more holistic view of the geometry and topology of a dataset. These are designed for data analysts first, so they render either inside the RStudio Viewer pane or from a web-browser using shiny.

There are two main functions for generating tour interfaces:

The goal of liminal is to provide complementary visualisations for use with understanding embedding algorithms such as tSNE. It has been shown that in order to produce an ‘effective’ embedding one may have to play with hyperparamters and various settings for these algorithms. liminal allows you to see how different parameterisations warps the underlying high-dimensional space.

See the liminal vignette for details of package usage and our arXiv preprint for a complete discussion on how to apply liminal to real data analysis workflows like clustering.

Quick Start

The release version of liminal is available on CRAN:

install.packages("liminal")

The development version of liminal can be installed as follows:

# install.packages("remotes")
remotes::install_github("sa-lee/liminal")

You can generate a tour view that will load in the Rstudio Viewer pane:

library(liminal)
limn_tour(fake_trees, dim1:dim10)

The interface provides instructions on how to use it, click on the help button to get started!