James Balamuta

@coatless

omnipresent explorer of the unknown, fashionably caffeinated informatics phd, amorous cavalier. design, build, analyze. #rstats on #macos + #webr and #python with #pyodide

16 Siguiendo1187 Seguidores468 Publicaciones

YouTube Bluesky

¿Es tu perfil?Reclámalo y coatless.dirty.social es tuyo.Reclamar perfil

James Balamuta@coatless· 1d

Nothing says your niche R Observatory project has made it like a hosting provider gently asking if you are aware you are being attacked. Happy Monday!

Screenshot of an email from a hosting provider that reads: "Hello, This email is being sent to inform you that your site r-observatory.thecoatlessprofessor.com is currently receiving an unusually high volume of traffic from a large number of IP addresses simultaneously. This pattern is consistent with a distributed denial-of-service (DDoS) attack or aggressive bot/crawler activity, and it is generating" (text cuts off).

James Balamuta@coatless· 4d

Every painter has a gpaint_*() twin that returns a real ggplot, so it works with ggsave() and patchwork() like other ggplot2 extension packages. Thus, you get the same drawing either way, in four palettes.

A four-up comparison showing one data frame drawn by paintr in each of its four palettes. Three modern cards, mint, slate, and warm, share the refined look: a tinted header band, bold column names over lighter type tags, hairline borders, and softly rounded corners, each in its own colour family of cool green, cool grey, and warm amber. The fourth, classic, shows the original flat look with a hard black grid and square corners. All four draw the identical structure and the same labels, so only the styling differs. Both the base graphics and ggplot2 backends produce these identical cards because the style travels as data on the cell table.

James Balamuta@coatless· 4d

Formatting further aims to emphasize magnitude by showing significant figures, but the digits it drops go grey rather than vanishing, so 123456.789 still reads as a bold 123 with a faded 457 with decimals rounded if its a large number. NA is red. Inf is blue.

The vector 123456.789, one third, NA, Inf, and e drawn by paintr's paint_vector() as a horizontal row of five boxes in the mint palette, chosen to show how the package handles values that are awkward to print. The first box reads 123457 with the leading 123 in solid black and the trailing 457 in light grey, so the three significant figures stand out while the dropped digits stay visible and the magnitude is never hidden. One third shows as 0.333 and e as 2.72. NA is drawn in red and Inf in blue, so missing and infinite values are recognisable at a glance rather than blending into the numbers around them.

James Balamuta@coatless· 4d

show_indices = 'cell' stamps every matrix cell with the [row, column] that reaches it. Useful when the thing you are explaining is not the values but the addressing.

A matrix of the numbers 1 through 20, laid out in four rows and five columns and drawn by paintr's paint_matrix() with show_indices set to cell, in the mint palette. Every cell carries two things, the value itself and the bracketed row and column subscript that reaches it, running from bracket 1 comma 1 in the top left corner to bracket 4 comma 5 in the bottom right. Reading the picture and reading the code are the same act, so m[2, 4] stops being an incantation and becomes a box you can point at.

James Balamuta@coatless· 4d

You already know a data frame is a list whose elements happen to share a length. Your students do not. paint_list() and paint_data_frame() draw the same card until one element changes length and the bottom edge goes ragged. That is the lesson, drawn.

A list of three numeric elements named mpg, cyl, and gear, drawn by paintr's paint_list() in the mint palette. Each element becomes a column of boxed values under its own bold name and a grey dbl type tag, and every cell carries the double-bracket accessor that reaches it, such as bracket bracket 1 bracket bracket bracket 1. The mpg and gear columns run four rows deep, but cyl holds only three values, so its fourth cell is missing and the bottom edge of the card is ragged rather than square. That ragged edge is the whole difference between a list and a data frame, since a data frame is a list whose elements happen to share a length.

James Balamuta@coatless· 4d

Every R course has the whiteboard moment where you draw a vector as a row of boxes. {paintr} draws it and writes the accessor in each cell, so the diagram and the code are one object. Now on CRAN. #rstats blog.thecoatlessprofessor.com/software-rel...

The numeric vector 10, 20, 30, 40, 50 drawn by paintr's paint_vector() as a single horizontal row of five boxes in the mint palette. Each box holds its value in large type with the bracketed position it answers to underneath, running from bracket 1 through bracket 5. The third box, holding 30, is filled with a soft mint highlight, so the picture shows what x[3] selects rather than describing it. A caption above reads that the vector has 5 elements and is numeric.

James Balamuta@coatless· 18d

Now on CRAN: {livelink}. Hand it a string, a file path, or your clipboard and get back a URL that opens a live R session in the browser, code already loaded. CRAN: cran.r-project.org/web/packages... Release notes: blog.thecoatlessprofessor.com/software-rel... #rstats

Three-panel diagram showing how livelink works. Panel 1, "You write R": a code block wrapping a short script in webr_repl_link(), where the script loads the mtcars dataset and plots mpg against weight with steelblue points. A caption notes that a string, a file path, or the clipboard work too. Panel 2, "You get a URL": a long webr.r-wasm.org link with the code encoded in the fragment. Panel 3, "They get a live R session": a browser window with four panes, an editor holding script.R, a file browser showing home/web_user/script.R, a terminal where source('script.R') and mean(mtcars$mpg) return 20.09062, and a pane displaying the resulting scatter plot.

James Balamuta@coatless· 22d

{livelink} will also soon be expanded over to Python just after Pyodidarian launches as the later comes with a Pyodide powered Python enhanced 4-panel editor based off of webR's. #pyodide #livelink #python

Screenshot of the Pyodide REPL, a browser-based Python environment. The left pane shows example code using pandas and matplotlib; the right pane shows an Environment panel of variables and a Plots tab with a line chart titled "n squared."

James Balamuta@coatless· 22d

#webR and #Shinylive already share browser-runnable links from their editors. livelink is a small #Rstats wrapper around that idea: build those links from R (scripts & projects) and decode any of them back into files. GH github.com/coatless-rpk... Post blog.thecoatlessprofessor.com/programming/...

A split comparison of two ways to share an R project. The top path, 'send the instructions,' runs through install steps to a laptop marked 'runs, if it installs.' The bottom path, 'send a link,' goes straight to a browser already running the plot.

A loop showing encode and decode as inverse steps. Three project files flow along webr_repl_project() into a single URL, then back along decode_webr_link() into the same three files.

James Balamuta@coatless· Jul 12

I know I'm sleeping, 'cause this dream's too amazin'

Screenshot showing Apptrope Data Science fixed with a Shiny for Python app open along with 10 other app frameworks.

James Balamuta@coatless· Jul 10

It was part of the underlying rule set in addition to the CRAN cookbook. blog.thecoatlessprofessor.com/programming/...

Screenshot showing part of the ruleset from different R sources regarding CRAN checks that needed a home inside of checktor.

James Balamuta@coatless· Jul 10

After 2 years collecting dust on GH, peeky has finally peeked out onto CRAN and kept its own release quiet for a week. It downloads & rebuilds the source behind any published #Shinylive app, in #Rstats | #Python CRAN: cran.r-project.org/package=peeky blog.thecoatlessprofessor.com/programming/...

James Balamuta@coatless· Jul 10

The best part of getting on CRAN wasn't the acceptance email. It was the PS. This made my whole week.

Screenshot of a Gmail message from a CRAN maintainer, dated Thursday, July 2. The sender's name and address are blacked out. The message reads: "Thanks, on its way to CRAN. Best, [name redacted]. PS: I hope your package will be used a lot!" Below it, the quoted original email shows it was submitted via the CRAN Package Submission Form on 26/06/2026.

James Balamuta@coatless· Jul 10

R CMD check comes back clean. Then CRAN emails you about title case and a missing @ return tag. checktor is in. It catches those before you submit. On CRAN as of last week 7/2! CRAN: cran.r-project.org/package=chec... Blog: blog.thecoatlessprofessor.com/programming/... #rstats

Dark terminal screenshot of a checktor() run on a demo package: the Package Doctor diagnosis summary listing five check categories with an overall verdict of "NEEDS ATTENTION (10 issues)."

James Balamuta@coatless· Jul 9

Yes, yes it was a #quarto playground I spied back in the day. Confirmed by @shikokuchuo.net 's #user2026 talk. Quarto levels up via: quarto-hub.com

Screenshot of Charlie's talk posted on Linkedin showing the playground and it's home over at quarto-hub.com

James Balamuta@coatless· Jul 8

Thanks, we were overprotective regarding periods. data.table and other packages with periods should now show up on search now.

Screenshot of the R Observatory package search page showing results for the query "data.table." The interface lists 133 matching CRAN packages, each shown as a card with the package name, version, download rank, short description, maintainer, and 30-day download count.

James Balamuta@coatless· Jul 8

That's a slice of it. It's free and works for decently well for many of the 25K packages on CRAN as well as having some Bioconductor support (still improving!), so go look up one you maintain or rely on. r-observatory.thecoatlessprofessor.com

The Compare downloads view on R Observatory, showing daily downloads for dplyr, ggplot2, tidyr, and tidyverse, each in its own color. An "Add a package" box with suggestions sits above a "Split Stream" row of source pills (CRAN, r2u, autoCRAN, conda-forge, bioconda, c2d4u) that breaks the comparison out by download source. Two stacked line charts plot the series over the past year, one for CRAN and one for conda-forge, with a hover tooltip listing each package's downloads on a given day and dashed markers for releases and R versions.

James Balamuta@coatless· Jul 8

R Observatory has pages for authors, not just packages: your published packages, their check results, your overall impact, and a collaborator network graph. Go look up your own. r-observatory.thecoatlessprofessor.com/authors/

An author page on R Observatory for Kurt Hornik. A header notes 85 packages, time on CRAN and Bioconductor, and an ORCID link. Cards show downloads (4M in 30 days, 13.4M in 90 days, 53.9M in a year), a Roles breakdown (maintains, co-authors, contributes), and an Impact panel: 4.9K packages depend on their work, a dependency h-index of 27, and the most-depended-on packages. Further panels list top packages by downloads, an aggregate Portfolio Check Status across maintained packages (error, note, and OK counts, plus packages needing attention), frequent collaborators with a link to the collaborator network, and recent package activity.

James Balamuta@coatless· Jul 8

ggplot2's license quietly moved from GPL-2 to MIT. R Observatory's version diff catches that, plus minimum R, download size, contributors added or dropped, and how code, tests, and docs grew. r-observatory.thecoatlessprofessor.com/packages/ggp...

A version comparison on R Observatory for ggplot2, set to 3.0.0 versus 4.0.0. From and To version selectors sit at the top. A Metadata panel summarizes what changed between the releases: the dates (flagged a major bump), a license change (from GPL-2 to MIT), a higher minimum R version, a larger download size, and contributors added and removed (four new, one removed, with roles shown). A Structure table then compares counts side by side with a change column, for exported and internal functions, total and R lines, compiled lines, and test lines (for example, exported functions 451 to 643, test lines 5,221 to 17,565). A Documentation table does the same for doc lines, roxygen coverage, and references coverage.

James Balamuta@coatless· Jul 8

Then there's the license side. You add one small package, it brings a few of its own, and now a GPL dependency is sitting under your MIT project. RObs walks the whole install tree so you can see the license picture instead of guessing at it. r-observatory.thecoatlessprofessor.com/packages/ggp...

The Dependency licenses page for ggplot2 on R Observatory. A green verdict reads that all 16 installed dependencies are permissively licensed, with no copyleft obligations in the install tree. A make-up bar shows 15 MIT and 1 Apache, plus 6 base or recommended packages that ship with R. Below, an "Explore the dependency graph" panel walks the tree with toggles for view (by dependency or by license family), which kinds to include (Depends, Imports, LinkingTo, Suggests), and depth (one level through all levels), followed by an indented, license-tagged listing of the imports and their own dependencies.

James Balamuta@coatless· Jul 8

On each package page, the call graph is tinted by test coverage: functions green where tests reach them, red where they don't. That needs the package to have tests. covr measures your own coverage; here you can see it for other packages too. r-observatory.thecoatlessprofessor.com/packages/Rcp...

The Rcpp call graph on R Observatory. An interactive network plots the package's 411 functions as nodes tinted by test coverage on a red-to-green scale, with node shape marking language and rings marking exported functions; edges show calls, including dashed edges from R into compiled code. A side panel shows per-function detail on hover, here the exported sourceCpp function with its line count, cyclomatic complexity, 100% line coverage, and call list. Below the graph, summaries break coverage down by type and by file, tally the structure (411 functions, 18 exported, 291 compiled, 497 call edges), and rank functions by lowest coverage.

James Balamuta@coatless· Jul 8

The download number everyone cites is a single CRAN mirror. R Observatory also counts conda-forge, Bioconductor, bioconda, r2u, & more, so you see much more of where a package actually gets installed. Still not every source, but well past one mirror. r-observatory.thecoatlessprofessor.com/downloads

The Downloads overview on R Observatory, titled "Package download activity across every data source we track." A row of source filters includes CRAN, r2u, Bioconductor, autoCRAN, conda-forge, bioconda, and c2d4u, with Posit P3M and Fedora COPR marked "soon." A headline reads 277.1M downloads in the last 30 days, combined across 7 active sources, broken down by a stacked bar (CRAN 84.7%, conda-forge 10.3%, Bioconductor 2.7%, and smaller shares from the rest). An ecosystem-trend chart runs from 2012 to 2026, and a "Top packages, combined" table lists the most-downloaded packages with per-source columns.

James Balamuta@coatless· Jul 8

Checking on an R package usually takes five open tabs: CRAN, GitHub, download counts, reverse deps, check results. R Observatory collects all of that on one page, for every CRAN and Bioconductor package. r-observatory.thecoatlessprofessor.com #rstats

The ggplot2 package page on R Observatory. A header gives the name, version 4.0.3, and tagline, with four headline numbers: 6,665 packages depend on it, 26.4M downloads a year, 81.1% test coverage, and 13 of 13 CRAN checks passing. Sections below cover an Overview (maintainer, first-published date, description), install commands in tabs (CRAN, pak, Ubuntu r2u, Fedora COPR, openSUSE OBS), Health (CRAN check result, code-health facts, and a dated check history), and the start of Downloads. A Vital Signs sidebar summarizes health, coverage, docs, downloads, reverse dependencies (6,665 direct and 1,909 indirect), release, license, and links to reference docs.

James Balamuta@coatless· Jul 2

Signing desktop apps is per developer, per platform: macOS needs Apple Developer credentials ($99/yr), Windows a certificate ($100+/yr). We cover Apple but do not develop on Windows, so the demos there trip SmartScreen. Click More info → Run anyway to launch.

Windows Defender SmartScreen dialog for demo-r-app-suite-shinylive-win-x64.exe, "Unknown publisher," showing the "More info" link you click first.

The same SmartScreen dialog after expanding, with Run anyway and Don't run buttons.

James Balamuta@coatless· Jul 2

Five runtime strategies to pick from: offline shinylive (WebR/Pyodide), a bundled interpreter, the system install, auto-download on first launch, or a container. One export() call converts the app and builds the installer for each platform.

A three-stage pipeline diagram. A Shiny app (app.R or app.py, R or Python) flows through shinyelectron, which converts and packages it with electron-builder via an export() call, into a desktop app shipped as .app, .exe, or AppImage.

James Balamuta@coatless· Jul 2

Try the demos, read the write-up, or browse the code: Demos + docs: r-pkg.thecoatlessprofessor.com/shinyelectron Write-up: blog.thecoatlessprofessor.com/posts/reintr... Source: github.com/coatless-rpk...

James Balamuta@coatless· Jul 2

Turn a Shiny app into a desktop app your users just double-click. shinyelectron wraps #rstats or #Python Shiny in Electron and builds a native installer for macOS, Windows, and Linux, with a runtime of your choice (including a fully offline build). Build locally or in CI. #rshiny #pyshiny

macOS drag-to-install window for demo-r-app-suite: the app icon with an arrow pointing to the Applications folder.

The app's About panel showing the r-shinylive (WebR) backend, R 4.6.0, 37 packages, platform wasm32-unknown-emscripten, inside the desktop app.

shinyelectron launcher listing three R Shiny apps as cards (Dashboard, Data Explorer, About), with an "Apps → Back to Launcher" menu open.

GitHub Actions run "Build demo installers" for shinyelectron 0.2.0, matrix building 73 of 76 jobs across macOS, Windows, and Linux.

James Balamuta@coatless· Jun 30

Apparently I shipped Mac versions of my iPhone/iPad apps by leaving a checkbox on. Pyodios #python and webRios #rstats run on macOS now. #WebAssembly really does go anywhere. TIL.

A Mac screen split between two windows. On the left, the Mac App Store shows an account page with Pyodios (displaying a download button) and webRios (with an Open button), both marked "Designed for iPad. Not verified for macOS." On the right, the webRios app is open to a "Welcome to webRios" onboarding screen reading "Run R on your iOS device," explaining that webRios brings R statistical computing to iPhone and iPad through webR, a WebAssembly distribution of R. A sidebar lists Console, Editor, Plots, Environment, File Browser, Projects, Packages, History, R Help, and Settings.

A Mac screen split between two windows. On the left, the Mac App Store shows an account page with an "iPhone & iPad Apps" section listing Pyodios and webRios, both marked "Designed for iPad. Not verified for macOS" with Open buttons. On the right, the Pyodios app runs a Jupyter-style Python notebook called ExampleNotebook.ipynb locally through Pyodide. Code cells demonstrate basic arithmetic, f-strings, and list comprehensions with their printed output below. A right-hand panel shows environment variables, loaded modules, and two small plot thumbnails.

James Balamuta@coatless· Jun 27

@posit.co any chance #P3M download stats could be published like cran-logs.rstudio.com? cran-logs only logs CRAN-mirror source pulls, so the Linux binaries P3M serves (CI, Docker, etc.) are invisible to authors. The API has a downloads field but returns count: 0. #rstats

$Terminal session running a curl POST request to https://packagemanager.posit.co/__api__/filter/packages with header Content-Type: application/json and body {"repo":"cran","names":["ggplot2"],"omit_downloads":false}. The server returns a JSON error: {"code":6,"error":"downloads are not available because usage data is disabled; set omit_downloads to true or enable Server.UsageDataEnabled","payload":null}.$

$Browser showing the JSON response from https://packagemanager.posit.co/__api__/repos/cran/packages/ggplot2. It lists metadata for the ggplot2 package: source "cran", version 4.0.3, repository "RSPM", license "MIT + file LICENSE", maintainer Thomas Lin Pedersen, publication date 2026-04-22, plus depends, imports, and suggests fields. The "downloads" field is highlighted and reads { count: 0 }, showing no download counts are exposed on the public instance.$

James Balamuta@coatless· Jun 26

A well-earned Rousseeuw Prize for the R Core Team, the quietly heroic group behind #rstats, with Brian Ripley speaking for everyone. The stray photo of Excel is an accidental but perfect tribute to where most of us started before we found their work. www.bbc.com/news/article...