Interactive Shiny App

The dppca Shiny app provides an interactive interface for estimating and visualizing differentially private principal component analysis (PCA) results. You can explore the example datasets, choose privacy budgets and estimation methods, and compare available visualizations without writing R code.

Launch the online Shiny app

The online application is hosted on Posit Connect Cloud. It runs separately from this documentation website, so you do not need to install R or dppca to try the example analyses.

1. Available analyses

Use the tabs at the top of the results panel to switch between four analyses.

Tab What you can explore
PCA An integrated analysis with scree/PVE, score, loading, variable, and biplot views based on a shared private PCA fit.
Loading Private PC directions displayed as vector, point, or heatmap plots.
Scree Private scree values and proportions of variance explained (PVE), using the clipped, PMWM, or Huber estimators.
Score Private two-dimensional score histograms and synthetic sampled points, using additive and/or sparse histogram methods. Grouped displays are available when group labels are selected.

For more detail on the methods, see the DP Loading, DP Scree, and DP Score articles.

2. Getting started

  1. Open the online app. Under Data, select Example data and choose Gaussian, Gaussian with groups, or Vowel.
  2. Choose an analysis. The PCA tab gives an overview of five complementary visualizations; the other tabs let you focus on one part of the analysis.
  3. Set the privacy parameters. Enter the desired epsilon and delta values. For integrated analyses, configure the privacy budget allocation using either total-budget proportions or individual component budgets.
  4. Adjust the analysis settings. Select the PC directions, scree or histogram method, and other parameters relevant to the active tab. Group labels can be used for PCA and Score analyses.
  5. Run the analysis. Click Run PCA, Run Loading, Run Scree, or Run Score, depending on the selected tab, to view the results.

After an analysis has run, you can adjust supported plot settings to change the display without re-estimating the underlying private quantities. For PCA, the View menu lets you inspect individual plots or the integrated overview.

3. Data and comparison options

The app includes three example datasets: gau (synthetic Gaussian data), gau_g (Gaussian data with group labels), and vowel (vowel measurements). It also supports CSV upload through the Data panel.

Use Include non-private comparison only when a non-private reference is appropriate. Ordinary PCA results are not differentially private and should not be released from sensitive data without authorization.

For the R interfaces behind the app, see the Getting Started tutorial and DP PCA Visualization tutorial.

4. Run the app locally

If you prefer to run the application on your own computer, install dppca and shiny, then launch the bundled app:

install.packages("shiny")
# Install the current development version of dppca if needed:
# remotes::install_github("yejinjo0220/dppca")

library(dppca)
dppca_app()

You can also supply your own data using dppca_app(X) or dppca_app(X, group = "group"); see dppca_app() for details.

Online application: yejinjo0220-dppca.share.connect.posit.cloud