cataScience

A Journey of Data Science — an interactive training app for people who are new to data: import → clean → visualize → understand, with AI as your assistant and you as the judge.

Developed for WHO data trainings (Global Health Learning Center and country-office sessions), but the content is general: anyone learning data science or starting to use AI assistants at work can use it for self-study.

Documentation: https://shanlong-who.github.io/cataScience/

Guides

The website contains documentation. Launch the Shiny app from R for the interactive lessons and exercises.

Installation

install.packages("cataScience")

or

# install.packages("remotes")
remotes::install_github("shanlong-who/cataScience")

Usage

library(cataScience)
run_cata()

Version 2.2 adds an extended workshop app while keeping the original entry point above:

run_catas()

The extended app starts with Why data matters and adds one connected Workflow: understand the data, review problems, join and summarise, visualise, then communicate. Actions preserve an input snapshot and a preparation log, with undo and reset. Dataset A (a fictional care cascade) and B/C (synthetic ART records and facilities) support the Lao workshop. They are training examples, not country estimates; A is independent of B/C.

New lessons include John Snow’s map, explicit numerators and denominators, multiple imputation with synthetic cats, and the workshop’s chi-square and logistic-regression examples. AI chart comparisons remain a separate classroom exercise using the same exported synthetic data.

Workflow > Visualise offers 11 chart types: bar, sorted dot, histogram, density, line, scatter, boxplot, heatmap, dumbbell, forest and cascade. Plots include editable titles, units, grouping, facets, missing-row counts, and PNG/PDF exports with explicit dimensions.

Workflow > Communicate creates a self-contained HTML report or dashboard that opens offline. Supply the question, finding, denominator and limitations; the app does not invent conclusions. The project ZIP includes input and selected data, reviewed preparation code, chart settings, HTML, Quarto sources, an RStudio project and a local Shiny dashboard. Open the project in RStudio and run source("analysis.R") to reproduce it. Quarto is needed only to render the editable .qmd files. The ZIP contains the input data as well as selected records; review it before sharing.

Reusable functions

The package can also be used without launching Shiny:

library(cataScience)

patients <- cata_example("art")
facilities <- cata_example("facilities")
cata_profile(patients)
cata_missing(patients)
cata_check_keys(patients, "patient_id")

joined <- cata_join(patients, facilities, by = "facility_id")
joined$audit

lesson <- cata_lao_analysis()
lesson$flow
summary <- lesson$summary
figure <- cata_plot(summary, "dot", "group", "percent",
  percent = TRUE, sort = TRUE,
  title = "Reported viral-load testing by facility type",
  ylab = "Tested among records with known status (%)")
figure

# Choose a new output file; existing files are protected.
cata_report(lesson$selected, "training-report.html", plot = figure,
  tables = list("Reported testing" = summary),
  source = "Synthetic Lao workshop B/C example",
  unit = "Selected annual ART record",
  denominator = "Records with known reported VL testing status in each group",
  limitations = "Synthetic data; unresolved source and selection issues remain.")

Other helpers include cata_read(), cata_flag_outliers(), cata_proportion(), cata_count_status(), cata_mice_demo(), theme_cata() and cata_export(). Use ?function_name for arguments and worked examples. Flagging a value does not delete or replace it.

Dependencies

• mice remains required: the complete-data cat Easter egg is preserved, and the extended lesson demonstrates predictive mean matching, completed datasets, chain diagnostics and pooled model estimates.
• VIM is optional and used only for its existing kNN imputation. Install it with install.packages("VIM") to enable that choice. Missingness patterns use mice::md.pattern().
• rio is removed. Excel uses readxl; CSV uses readr.
• naniar and skimr remain required for missingness demonstrations and compact data summaries in the original app.
• DSIR is optional. Install it to choose style = "dsi" or the DSIR theme in the chart builder. The app also has a built-in theme.

The app opens in your browser. Everything runs locally — no internet connection is needed after installation.

Start with Import → Or use the example data, then follow Cleaning → Missing data, Cleaning → Outliers, Visualize and Quiz. Use Reset to original data to compare cleaning choices.

Use RStudio’s Stop button or Escape in the console to stop the app. Live demonstrations with an external AI assistant need internet access.

What is inside

For trainers

The Training → Training guide page inside the app contains a full playbook: a module menu with suggested timings, ready-made agendas (a 3-hour data-science session, a 2-hour AI session, and a full day), and a pre-course checklist for participants.

Participants who do not have R installed can use the portable Windows edition instead — contact the maintainer.

Development

Reusable functions live under R/; Shiny modules and teaching materials live under inst/app/. From this source checkout, use pkgload::load_all() and then run_catas() to try changes. Run devtools::test() and devtools::check() before a release. The existing installed package is unchanged until you install the new version.

Contact

Shanlong Ding — dings@who.int