As the number of variables grows, it becomes harder to fit an analysis into a single plot. Even after plotting the first two principal components, a question remains: are there clusters or outliers that appear only from another direction?

An online workshop titled “Visualising High-dimensional Data with R” will address that problem on October 1, 2026. It will be taught by Dianne Cook of Monash University, whose research includes interactive visualization of high-dimensional data.

The central idea is the tour: instead of fixing one low-dimensional projection, it moves through a sequence of projections. This article connects what a tour can reveal, how to try it in R, and who is likely to benefit from the workshop.

One two-dimensional view can miss high-dimensional structure

High-dimensional data records many variables for each observation. If six measurements are collected for every subject, each subject is a point in a six-dimensional space. Because six orthogonal axes cannot be displayed directly on a screen, analysts project the data into one or two dimensions.

PCA is a common way to do this, but a scatterplot of the first two principal components is still one projection. It does not automatically cycle through structures that appear only from other directions. The original tourr paper describes tours as smooth sequences of low-dimensional projections for finding features such as clusters, outliers, and nonlinear dependence.

A tour is therefore easier to understand as a complement to a fixed summary than as a replacement for PCA. Cook and Ursula Laa’s online book on high-dimensional visualization follows this broader path, using low-dimensional projections to examine linear and nonlinear dimension reduction, clustering, and classification diagnostics.

A tour moves the projection plane to search for clusters and outliers

Let X be a data matrix and F a two-dimensional projection basis. The displayed coordinates are X F. A tour changes F gradually, producing an animation in which the projected point cloud appears to rotate and change shape.

The book’s introductory chapter distinguishes three commonly used forms. A grand tour surveys many projections for a broad view; a guided tour searches for directions that emphasize a chosen structure, such as clusters or anomalies; and a radial tour removes and restores a variable’s contribution to test whether a visible pattern depends on it.

Historically, the grand tour builds on Asimov’s 1985 method for generating smooth sequences of low-dimensional projections. The 2011 paper introducing the R package tourr organized grand, guided, little, and other tours with several display options. The ideas behind today’s tools have been refined over decades.

tourr can animate a six-dimensional point cloud in a few lines

The current tourr website includes an example that projects six numeric variables from the bundled flea data into an animated two-dimensional scatterplot. This small example is enough to watch how overlap between groups changes across projections.

install.packages("tourr")

library(tourr)

flea_variables <- flea[, 1:6]
animate_xy(flea_variables, col = flea$species)

animate_xy() uses a two-dimensional grand tour by default. The official site recommends running animations from the IDE console rather than a Quarto document chunk, where evaluation may produce many static frames. A separate graphics device may also be needed in some environments.

Before substituting your own data, consider variable scale. The tourr changelog records that rescale has defaulted to FALSE since version 1.2.0. Whether to standardize variables or preserve their original units should be an explicit analytical choice, so that a large numeric scale does not unintentionally dominate the view.

The October 1 workshop is a next step for experienced R users

The workshop is aimed at scientists and data science practitioners who regularly work with high-dimensional data and models. Its stated scope includes recognizing clusters, outliers, and nonlinear relationships, then connecting those structures to supervised classification, cluster analysis, and nonlinear dimension reduction.

Participants are expected to have a good working knowledge of R and some background in multivariate statistics or data mining; this is not an introduction to R itself. An existing tutorial with the same title lists packages such as tourr, mulgar, geozoo, GGally, randomForest, and mclust, but it belongs to an earlier, separate workshop and has not been announced as the confirmed material for the October 2026 session. It is best treated as a preview of the concepts and environment, not as a definitive setup list.

The session runs from 10:00 to 12:00 CEST on October 1, 2026, corresponding to 17:00–19:00 JST on the same day. The minimum fee is EUR 20, USD 20, or UAH 800. Registration requires donating through one of the organizations linked by the announcement, saving the emailed donation receipt, and attaching a screenshot to the registration form. Confirmation is scheduled to arrive one day before the workshop rather than immediately.

Tours matter when the question is about relationships among variables

Having many variables does not automatically make a tour necessary. Cook and Laa’s discussion frames the purpose of high-dimensional visualization as learning about associations among variables; when no such association is present, univariate views may be sufficient.

When PCA or a clustering result does not show which directions separate groups, or which variable makes a structure appear or disappear, tours broaden the search. A fleeting pattern in an animation should not become a conclusion by itself, however. It should lead to a reproducible projection, model evaluation, or statistical investigation.

The answer to the opening question is not to make one supposedly best plot carry the whole high-dimensional space. For readers already comfortable with multivariate analysis in R, the workshop offers an entry point for adding tours to their analytical toolbox. Running the official minimal example first will also make it easier to decide how the method connects to a real dataset.

Source

• Title: Visualising High-dimensional Data with R workshop
• URL: https://www.r-bloggers.com/2026/09/visualising-high-dimensional-data-with-r-workshop/