Visualising high-dimensional data

Tutorial notes for useR! 2019

To access the interactive tutorials online, it is split into two parts:

• https://ebsmonash.shinyapps.io/tutorial_highd_vis_part1/
• https://ebsmonash.shinyapps.io/tutorial_highd_vis_part2/

The workshop is interactive, bring your laptop set up with the latest versions of R (>3.5) and RStudio, and these R packages:

install.packages(c("knitr", "tidyverse", "here", "nullabor", "forecast", "readxl", "GGally", "broom", "plotly", "tourr", "spinifex", "geozoo", "mvtnorm", "randomForest", "RColorBrewer"))

Also install:

remotes::install_github("wmurphyrd/fiftystater")

If you have signed up for this tutorial it will be helpful to download a copy of these materials ahead of time, by downloading the tutorial_highd_vis.zip file above. Unzip the file into a project folder, to work through during the tutorial. It contains these files:

• data/TB_notifications_2019-07-01.csv
• data/tb_burden.rda
• data/TB_data_dictionary_2019-07-01.csv
• data/IncRate.xlsx
• tutorial_highd_vis.Rproj
• tutorial_highd_vis.R

Please have these on your computer by tutorial start (Tue Jul 9).

This is the planned structure for the tutorial:

A. Review basic visualisation and inference with graphics: This part covers making plots using the grammar of graphics and how this fits into statistical inference. We will use the packages ggplot2 and nullabor.

B. Plotting multiple dimensions in a single static plot, adding interaction: The building blocks to viewing high-dimensions are generalised pairs plots and parallel coordinate plots, available in the R package GGally. There are many variations and options that will be discussed, along with making these interactive with the plotly package.

C. Using dynamic plots (tours) to examine models in the data space, beyond 3D: This part will cover the use of tours to examine multivariate spaces, in relation to dimension reduction techniques like principal component analysis and t-SNE, supervised and unsupervised classification models. We will also examine high-dimension, low-sample size problems. The tourr and spinifex packages will be used.

Materials are designed for an intermediate audience, users who are familiar with R, basic visualisation and tidyverse tools, and who would like to improve their knowledge about data visualisation.

Looking forward to meeting you all! And excited to be at useR! 2019.