Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

R Weekly Highlights

R Weekly Highlights artwork

Issue 2026-W03 Highlights

Jan 14, 202643 minAudio
ShareXLinkedInFacebookWhatsAppEmail

It's a new year of R Weekly Highlights! In this episode we learn of the positive intellectual diversity impacting R and Python communities, a great starting point to create your first RAG-powered knowledge store, and a new package enters the Shiny ecosystem that is certainly not tiny in its ambitions. Episode Links This week's curator: Eric Nantz: @rpodcast@podcastindex.social (Mastodon) & @rpodcast.bsky.social (BlueSky) & @theRcast (X/Twitter) R + Python: From polyglot to pluralism: A very targeted recap on posit::conf(2025) and a broader reflection on how the best design and devtools ideas from R and python have influenced each others' stacks for the better Retrieval-Augmented Generation: Setting up a Knowledge Store in R {tinyshinyserver} 0.1.0: Tiny 'shiny' Server - Lightweight Multi-App 'shiny' Proxy Entire issue available at rweekly.org/2026-W03 Supplement Resources tinyshinyserver GitHub repository: https://github.com/lab1702/tinyshinyserver Yann Tourman's R Weekly episode length analysis https://bsky.app/profile/yannco.bsky.social/post/3mbarm75kik2j Supporting the show Use the contact page at https://serve.podhome.fm/custompage/r-weekly-highlights/contact to send us your feedback R-Weekly Highlights on the Podcastindex.org - You can send a boost into the show directly in the Podcast Index. First, top-up with Alby, and then head over to the R-Weekly Highlights podcast entry on the index. A new way to think about value: https://value4value.info Get in touch with us on social media Eric Nantz: @rpodcast@podcastindex.social (Mastodon), @rpodcast.bsky.social (BlueSky) and @theRcast (X/Twitter) Mike Thomas: @mike_thomas@fosstodon.org (Mastodon), @mike-thomas.bsky.social (BlueSky), and @mike_ketchbrook (X/Twitter) Music credits powered by OCRemix We're the Fugawe Forest - Legend of Zelda: A Link to the Past - XPRTNovice feat. Weisty - https://ocremix.org/remix/OCR03597 Zelda's Graceful Nightmare - 25YEARLEGEND: A Legend of Zelda Indie Game Composer Tribute - Matheus Manente - https://zelda25.ocremix.org/index.html

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Hello, friends. Happy New Year twenty twenty six. The R Weekly Highlights Podcast is back for the New Year. We definitely had a little bit of a break to recharge our respective data, science, batteries, however you want to call it, time with family, but I hope you all are having a great start to 2026. But yeah, it feels great to finally be back behind the microphone and talking to you all about the latest highlights that have been shared in this week's Our Weekly Issue. My name is Eric Nance, and I'm delighted that you're joining us wherever you are around the world.

And, yes, thankfully, I'm not alone in 2026 as my awesome co host Mike Thomas is here virtually joining me as always. Mike, yeah, hard to believe 2026 already. It's amazing how Yep. It Another year, more Mike and Eric, our weekly highlights. Let's keep going strong. We'll keep going as long as this train doesn't stop moving. So we'll see how far we get. But speaking of how far we get, there's a little inside baseball here.

Around the end of the year, usually there's a certain person on the highlight our weekly team that's in charge of making the schedules for the curation. Guess who forgot to do that until last weekend? Yes. That was me. So that meant that with a short notice of notifying everyone else about the schedule, I guess I better do the issue. So it was me curating this week, but we'll have the full rotation already programmed up, ready to go for the rest of our team. But, yeah, had to dust off the holiday cobwebs a little bit, but as always, we had some terrific poll requests to our GitHub repository for R Weekly,

which was a massive help. So thanks to all of you around the world and my fellow curators for helping with this week's issue. And we're going to lead off with kind of a recap of a very important and very highly talked about theme, stemming from the recent POSIT conference in 2025, and that is seeing the influence that goes kind of bidirectional, but definitely in ways that us as R users can appreciate in the world of not just the

verses, so to speak, of R and Python, but really working together and benefiting from each other in terms of dev practices and community practices, and that is very succinctly talked about in our first highlight here coming to us from Emily Reederer, who is a senior manager of data science at Capital One, and she has been one of the top thought leaders in this space of going across different data science languages and using them to your best advantage across both ways,

and her post here is entitled R in Python, from Polyglot to Pluralism, and her main thesis here of the post is about, you know what, the days of the verses of the language wars, yeah, they're kind of done. This is now more about how can we get better data science accomplished, leveraging the best of each of these ecosystems and how can they learn from each other. So in her mind, you know, having the diversity

from the intellectual dev standpoint and the community standpoint, it just literally helps make the tooling better. And this was highlighted quite a lot in many of the POSIT conference talks, and she talks about a few of them here, and of course my co host looking at me is no stranger to this topic, I'll touch on that in a little bit here, but she shouts out two of the talks that leads off this post here regarding how best practices with our package development in recent years

have really begun to influence Python development. And these talks are given by Rich Iona and Michael Chow over at POSIT, talking about their various Python packages and what they have learned from their days of building the R package equivalents. So on Rich's viewpoint, he has made the great tables package in Python, which is the analog to GT from the R side of things. So he has some really nice points in the post here that Emily highlights in this,

in that it's about not just the code itself, but really understanding that different users are going to have different needs for their package and having that user experience, that onboarding about how to use the package being heavily featured and addressed in the package documentation, as well as built in data sets to help try these things out. So Rich's talk was excellent, I didn't see a live, but I've seen the recording of it, talking about these

attributes and how they've informed the GreatTables development. And I've been on record many times on this very podcast saying Rich authored some of the best documentation I've ever seen for packages, and he's carrying that same mindset to the Python side as well with gray tables. And then Michael Chow's talk about, called the cursive documentation, this was an interesting take on this as well, but it's also his experience

with making sure that there isn't just all terse documentation from a quote unquote API like standpoint, which a lot of Python modules and packages do get very terse with their documentation, but having more of the long form vignette style documentation is very helpful for those that are getting to know the package for the first time. And that is something that the Python community, I think, is slowly adopting.

But R had it here first, as they say, thinking all the way back to Fritz Leisch, building what became Sweve that was built into R since the very beginning. Of course, we've had R Markdown since then, but being able to build literate programming literally in the language itself, and we're seeing a lot of tooling in Python and elsewhere adopt those kind of mindsets. And then she also mentions that it kind of goes both ways as well, so to speak, the engineering side of it.

We're seeing some advancements in the R ecosystem to take advantage of newer libraries to help with co development, such as a highly regarded air utility based in Rust to help lint our code, and it is blazing fast. I love using it. I use it every day in my positron editor. It just cleans up so much for me in that little control s keyboard shortcut. And then also a shout out to the rig utility,

built by posit as well as a way to manage multiple r installations on a host system no matter Windows, Mac, or Linux. That is something that Python has been trying to address, sometimes a mixed success depending on the framework you use, for managing Python libraries and installations. And we're seeing also how, say, the TidyModels team was heavily informed by what scikit learn has done in their machine learning exploits on the Python side. So there's a lot of

interesting ways that both language communities are benefiting from each other in these practices, and it's really, really interesting to see how far this takes. And Mike, you're no stranger to this, you deal with both languages and you've been a a highly, regarded voice in this matter too. What did you think of Emily's post here? Yeah. Absolutely. And I've been very inspired by Emily in the past really on this topic. I know she gave a talk at posit comp. I think it was two years ago. It was called, cleverly titled Python

Ergonomics. And I took a lot of inspiration from that in developing the talk that I gave at this this past year's positconf on building multilingual data science teams, which I believe was in the sort of the same session, as as Rich and Michael's talk, which was fantastic. And obviously, developers like Rich and Michael are are really doing a great job of bringing ours great focus on user experience to Python through their their table, packages like great tables and point blank. And we also have other

packages like plot nine and Python that mirrors g g plot two and R. It's more great work from from posit to essentially make these capabilities language agnostic. It's it's just pick whichever language you want, and the same tooling essentially will be there for you. Some of the key traits, you know, that I think are being adopted by Python, as Emily mentions, is making packages nicer, by including, you know, sample datasets. We take for granted that forever. You know, if you install a package,

there is almost always a package dataset that enables you to get up to speed quicker and and run examples, that you see, I of dplyr. We always have the Star Wars dataset. Right? It's just built in for us to test out all of the different, you know, crazy things that dplyr can do for us without needing to load some CSV that we need to have on our laptop first. So it allows users to get up and running a lot quicker. And, you know, I think that really bleeds into documentation in terms of, vignettes

for Python and them trying to move beyond, you know, what they may have had in the past around package documentation. My guess is that package down, the the R package that allows us to build these these HTML websites that we can host with, you know, GitHub pages and things like that, probably really distance R from Python in terms of robustness of the documentation in the last few years. I'm assuming that the Python community caught wind of what package down could do and I I think created a fairly obvious gap.

And obviously, you know, think Quarto has also been a game changer here. And the more that we can enable developers to leverage Quarto in package documentation and wire those two things together, I think the more that we can provide clarity to end users, and LLMs, unfortunately, nowadays, on how to best use our R and Python packages. And as you mentioned, you know, Python like robustness in R, R is gaining, you know, high performance developer tooling that's very much inspired by Python. We think of Air, the the code formatter you mentioned. Rig is another one that Emily mentions that helps to mirror the the functionality and feel of, PY N for helping users manage multiple r versions.

If you are a docker user, there is a dev container feature, in the rocker project that, allows you to essentially leverage rig to install the exact version of r that you want to install in that container. So if you were trying to do reproducible data science as as we really have a focus on on our team, that may be helpful for you. It's something that we use in just about every project that we hand off to clients in particular because we know that it will work on their machine exactly as intended.

So I I, you know, really think that this blog post is another fantastic, one from from Emily on this particular topic. Anything that she puts out is essentially required reading, in my view. Yeah. I think sort of this concept of intellectual diversity, you know, makes tools better across both R and Python. And and when developers from these different backgrounds collide and come together, we get better engineering in R and we get better design in Python. Yeah. Yeah. And I I definitely agree with all those points. Another utility that we've been seeing slowly slowly get some attention

on the Python side back to package management side of it, UV has been highly regarded as another way to manage your project specific dependencies, and we're now seeing an R sort of equivalent called RV that we talked about, I think, late last year. So, yeah, reproducible data science on the tooling side, as well as being able to leverage a package effectively. We're in a good time here and with the advent of having to, you know, use the best tool for the job. The gaps have definitely tightened a bit, but there will be cases where you need to hop back and forth and that's where having the choices, but also, you know, what to look for is highly important. So I dare say this is a great start for stuffing my data science polygout wallet, if you will, with some with some great content. Yes. I've been watching too much ESPN lately. You're welcome.

I will second second the UV comment. If you are starting a new Python project or just getting into Python, start with UV instead of any of any of those other Python environment managers. Is so easy to use and it just works fast. And I did just see RV come across my desk as you mentioned, Eric. I have not tried it yet. I don't know if you have, but I'm excited to take a look. It is really promising. So I look forward to playing with this even more, even as the

veteran of PACRAD and RM, which have, we have a, mostly beneficial relationship. Sometimes things go haywire though. So I'm always looking to see what else is out there. So yeah, there's a couple mic drop moments right away in this podcast. Choices are good.

And yes, 2026. You can't escape the use of LLMs across much of our dev workflow, and boy, when they work well, they really work well. But there's always that age old trade off, right, of if what I'm building now requires something that has some knowledge, so to speak, or resource that wasn't necessarily used to train these models, just what do you do? What if you have this very intricate source of information domain specific wise

that you want to have available to you, but you know that off the shelf Gemini or quad model just doesn't have any exposure to it? That's where a really important technique that I've heard about for a long time, but admittedly was kind of intimidated by until recently, called Retrieval Augmented Generation, RAG for short, is a very important set of tooling, so to speak, that you can have in your next project that has a leverage, the use of AI.

And how the heck do you get started from this as an R user? Well, guess what? The tooling has definitely caught up speaking of, you know, tooling and R ecosystem getting up to speed much quicker these days. There is some great packages out there that you can use to build your own knowledge storage in R to be tapped into by various LLMs. And this tutorial blog post comes to us from

Miles Mitchell over at Jumping Rivers, and this is a great practical post on how you can build a Rag knowledge store using just off the shelf packages in the R ecosystem. So as we talked about, this is where RAG is important for a lot of reasons here, one of which, like I said, is if you have another additional context source, if you will, typically documentation,

maybe reports or whatnot, and you want to make that available to your LLM session or model provider to be able to tap into as needed. So there are lots of different phases of this, but luckily in R, there is a particular package that can help you get started really quickly, and that comes to us from the POSIT team called Ragnar, and this is a very important package in this space, I'll link to it in show notes as always,

but this is helping you implement Rag all in the confines itself. And it's got a few different steps in this process, but we'll walk through a little bit here verbally what that entails. First of which, you got to have that knowledge source available, hopefully in a web kind of friendly text format. A lot of times say markdown documentation sites are a great source of this. So in this example, they're actually looking at the web version of a textbook called Efficient R Programming,

co authored by one of the, you know, the founders of Jumping Rivers, Colin Jawispie, and that URL is available. You can find that in the show notes as well. So what Ragnar lets you do is read the contents of that website as Markdown with the appropriate name, read as Markdown function. Once you have that, this is now split up into what are called kind of chunks in a nice tidy tibble

with each chunk being a different row, and then the text actually containing the the actual, like, verbatim text in that particular chunk as defined by the starting and ending index of characters. Now that's just step one. You gotta you know, you can actually visualize this in your R session with a Ragnar chunks view function, which is neat to kinda see what's behind the scenes there, but this is not ready for the models models yet. There is something you need to do to create under the hood what's been known as like a vector database,

a way to translate this text into tokens that then the LLMs can use, and that's where there is a Ragnar store create function that will help you take into this, or define, I should say, what this kind of internal vector database will look like, but the embeddings are model specific. So depending on which model you're using, you might have to use a different embed function that comes from the package,

such as one for Gemini, maybe one for OpenAI, whatever else, they have separate functions for each of these. Once you have that kind of store set up, then you feed in those chunks that you got from the website into that store, and then from there, you got it in the store, but imagine like a grocery store where all the products are not organized yet, you got to actually start organizing this and create an index, and that's where the Ragnar store build index function comes in to get it actually ready for the LLM usage. So all this is stepped up through in the blog post, but once you have that, then you can actually start to

use this using the Ragnar retrieve function, such as in the example here, they take that store as the first input and then ask the question, who are the authors of EfficientR programming? And sure enough, you get a tipple back and it will give you text to actually tell you in the well, actually, not so much text yet, but the actual well, not just the text, it's a context, it's a distance calculations, all in the confines of this vector database investigation,

and you will actually see then the text showing the authors there. So there is a lot of interesting things you can do with this because if you want to feed in more than one source, you can feed multiple, you know, sets of knowledge chunks into these stores and just kind of rinse and repeat along the way, and then lo and behold, you can set up via the Elmer package a chat function, or chat object I should say, and then register this store

as another means for that particular model in Elmer to get this knowledge, and sure enough, they do just that with asking us some questions about, say, good practice for parallel computing in R, and then it can even tell you where it got the answer from. The one some things to watch out for is this, is as you're building these stores, these different chunks into these stores, this is going to add to your token count when you go back and forth between these model providers, so cost

may or may not be an issue, depends on your organization, but this is a great way to get started with getting that internal resource ready for your models to run. And I think this is a great way to jumpstart your Rag journey into using packages like Elmer and the like to supercharge your next chat sessions. So, Mike, have you done much of RAGs lately? This sounds like a pretty interesting technique here. Yeah. We've done, you know, quite a bit with RAG and I think there's always been a trade off for AI engineers.

You know, I I don't know if it's starting to change or if it it has changed, but the concept of leveraging Rag versus actually trying to, like, fine tune your own model, which I believe would be, you know, actually potentially training an LM or or doing some form of that would be the the fine tuning, which is, I think, much more difficult than leveraging rag. And if you can get good results out of your rag setup, it's probably a lot, easier to sort of stand that solution up. And right? The the real problem that everybody's dealing with is that LLMs, you know, tend to hallucinate a lot because they lack, you know, real real time information or

they will hallucinate because they don't have access to the data that you are asking it about because it wasn't trained on that. And they have gotten better at this, I think, in, you know, more recent times where maybe they'll actually admit that they don't know the answer to your question. But I remember when LLMs were first coming out, if you ask them, you know, a a question, like, I had asked them who who Catchbrooke Analytics was, I think, at one point in time, and it it just made up a completely fake company, that said they were in pharmaceutical sales or something like that. You you couldn't find it on the Internet. I don't know. Maybe there is another catch, Brooke, analytics.

You know, it was it was interesting. They hallucinated, you know, all the time, unless you, you know, had that rag set up and you gave them specific information about the question that you were asking. And that rag workflow, you know, really helps to provide that model with sort of this external knowledge store, as we call it, which, you know, are often our our documentation, to look up the information before answering. And the first step of that process is parsing that documentation into something that the LLM can read. And as you mentioned, that's where the Ragnar package comes in to build that knowledge store and extract

text, you know, from these, you know, PDF documents or other sources, chunking it into contextual pieces, and storing that in a vector database so that the model, can search on it efficiently. And then Elmer is sort of the second piece of that equation that allows our users to interact with various LLM APIs, you know, such as OpenAI, Google Gemini, Claude, and leverage that connection to the rag store that they were able to create. And Miles does a great job,

providing us with an example, you know, building that knowledge store using the efficient r programming textbook, registering it with Elmer, and then the you know, clearly accurately answers some of those technical questions. Think about parallel computing and even sites exactly where it found the information, which I know is an important sort of development and very important for, you know, production level workflows as well and auditability and and traceability in a lot of domain settings. And I think sort of at a higher level, you know, Rag sort of makes LLMs

production ready, for a lot of organizational use cases where they need specialized access to to knowledge that's really specific to your organization and might be outside of the corpus of data that the LLM was trained on. And or you you may potentially just wanna have the LLM focused on the information that you wanted to know about, and really not care about, you know, all of the other potential information that it was was trained on. Last year, you know, we had built out a simple radio app to demonstrate rag and gave it a single PDF. You know, this wasn't publicly available information

in the PDF for its knowledge store. And we had a really cool component to that app that was a button that could toggle rag on or off. And if rag was on, I would ask it some random question like, you know, give me a recipe for banana bread, which was totally different, you know, than the PDF context we were giving. And it would actually come back to me and say, hey. You know, I don't think that's appropriate right now because I'm sort of set up to answer questions about this particular domain. And I thought thought that that was awesome. Yeah. So that might be an interesting thing if you need to sort of demonstrate the capabilities

of Rag, in your Shiny app or some sort of a web interface like Gradeo to consider doing something like that because it was really eye opening, for a lot of our audience. Yeah. I'm jotting that down as I start to build more interfaces on top of these. But, yeah, it's it's a very interesting interesting technique and it does make me it makes important another thing to underscore is if you want to use this, you know, having that knowledge store or that source of that knowledge store

hopefully easy to digest via these mechanisms. I know in a lot of organizations, lot of knowledge is locked in proprietary files and whatnot, so it can be a Herculean effort to get that ready for consumption into RAG. But boy, I think the benefits will definitely pay off if you can get support to translate some of this internal domain knowledge into ways that we can leverage more effectively. I mean, basics like PDFs

of these certain, in the pharma side of things, something called the schedule of events that determine when patients get certain, you know, treatment drug versus like lab assessments versus other assessments. A lot of that's locked away in some very bad formats. And we're trying to make more machine readable formats of all this so that in the future, you know, things like RAD could be built to help interrogate some past, you know, study information or whatnot. Still a very long ways to go in this space, but I'm excited to see just where you can take these techniques and building interfaces on top of it because

these models can't do everything folks. You got to help it out along the way. And rounding out our highlights today, you know, Mike and I are always fans of Shiny. So as I'm curating this issue and I get this list of the new packages that have come since our last issue as well as the updated packages,

and there was one that particularly caught my eye as far as like, you know, what we might talk about or at least might be featured in the highlights section, and I saw this package called Tiny Shiny Server, and I thought, that's interesting. What is this all about? So I started reading up the it's just landed on GRAND recently, so I started looking up the GitHub repo and I saw things, Mike, that boy sure sounded kind of familiar, but this is all done in R itself. So let's let's talk about what this has been built for, at least according to the README.

This is appropriate for hosting multiple shiny apps behind a single server. Sounds familiar. Development environments of multiple projects. Yeah. Maybe. Small scale production deployments. Oh, you and I deal with production all the time. And hosting R Markdown and Cordial dashboards with the Shiny runtime. Okay. This is this is intriguing, so I actually gave it a spin on my, Nixify dev environment set up at my other computer,

and for the most part, it did basically what was advertised, and what's interesting is once you define a directory of your app dashboard, if you're doing like portal dashboard or flex dashboard in the R Markdown side, you basically make a subdirectory and they have an example directory you can copy straight from the package itself, and within that you could have any number of Shiny apps as subdirectories. So they have simple inventory one, simple reporting one, the sales one, and whatnot.

And then there is a configuration file in JSON called config dot JSON, where you spell out all these different apps, each get their own, you know, sub entry, the name of it, where it's located on the file system, and then another interesting flag called resident. If resident is set to true, that means that the app is always running. This would be very similar to the classic, you know, Shiny server days, the app is just running no matter what, or I pause the connect internally.

Whereas if it's set to false, it'll only spin up when it's accessed. That part sounds really familiar to me. You probably will comment on that in a little bit, but I'm wondering just how does that work just from the R side of things? That usually is kind of like a container kind of thing to make sure that is only spun up on demand, much like services like shinyapps.io would do, or you'll probably mention another one a little, but I won't steal your thunder there.

Now I will say I tried that feature. It did not work on my system, so there may be a little bit of gotchas in how that was all implemented, but it was intriguing nonetheless to see them trying to pull it off. So this readme is quite comprehensive. I kept looking, it's like there's a lot of options here, a lot of configuration, even about if you're going to host this behind a proxy or SSL, some instructions for that. If you want to do like an NGINX kind of reverse proxy or a caddy file.

So all this is very DevOps y kind of features here too, with a management interface as well if you want to look at a high level of all these apps. So I was, you know, admittedly very blown away just how much could be accomplished literally with just an R package to do all this. And then I did a little, you know, I like to, you know, shout out whoever authors these utilities on our weekly, so I took a look at the root of the repository,

lab seventeen oh two, I believe is the alias, and then I read the description, just some random hobby projects I made throughout the years, The more recent stuff is mostly AI created as a way to learn how to work with AI. And sure enough, if you go to the TinyServer GitHub repo, on GitHub you always see like a contributors link or set of contributors on the right side. Sure enough,

right under lab seventeen oh two is Claude. So I'm guessing this was a mostly AI generated package. I will say though, how it got there, I have no idea. I mean, it sounds like Claude somehow, but I must say pretty impressed with how far it went. I'm not going to pretend to say I'm going to use this in production anytime soon, but it's a great way to see just using some of the dependency packages that they reference here in the description file,

such as like the web sockets kind of packages, but really it's HTTP UAV. It's not a huge dependency footprint. Might be an interesting learning opportunity nonetheless. So yeah, impressed in a few different ways, but definitely caught my eye and it just shows you. If I hadn't started looking around that GitHub repo, I was almost completely fooled. Maybe it was because of lack of sleep too. I don't know. But this is this is probably going to be more common than we realize, Mike. So

very interested in your takes on what Chinese Shiny server is trying to accomplish here. Yeah. I mean, I don't want to make any assumptions on how much of it was AI generated versus how much of it was sort of human reviewed. I assume all of us, you know, are doing that to some extent these but I think, you know, as you mentioned, Eric, probably there is a a real thing going on in the ecosystem and data science in general, where fully AI generated stuff is is getting out there for better or for worse. This this concept is, I think,

excellent. Having a sort of more lightweight, you know, WebSocket enabled, you know, proxy server designed to host maybe a couple shiny apps in our markdown documents and and Cordo dashboards behind a single server. I think, you know, obviously, there's there's great need for making productionizing of your data science content easier, right, if possible, especially in smaller settings. I think WebR has done some of that for us, but I think for a lot of folks, it's probably still a little scary territory because there are some gotchas,

from my experience on hosting, you know, Shiny apps, for example, with WebR that are very very niche to, that that WebR framework as opposed to if you were just, you know, running it, obviously, on your own laptop. But, I think this is really, really interesting. As you mentioned, one of that that standout features that they market in the the repositories readme is the the resident versus on demand apps where, you know, resident apps are always running.

I would I think this gets into R sessions. Right? You would have to have different apps run on different ports from the the same R session. I imagine that we're only spawning a single R session in this process or maybe not. It seemed like one to me when I ran it last night. It was pretty bizarre. Obviously, you couldn't get that on demand apps, the other side of it, up and running. And I would imagine that that would have to mean that you would have multiple R sessions running, in order to have one spin down. But, obviously, I had two apps, one was a resident and one was on demand,

I would assume that I would need two hour sessions. I'm not I'm not sure. You know? And then sort of the questions then become, you know, what's the separation between users? Are they sharing sessions, not sharing sessions? You know, a project that we talk a lot about and implement quite a bit for a lot of our clients and even ourselves is the ShinyProxy project, which this feels kind of like a lightweight ShinyProxy, environment. And that is very much Docker based such that every

app that gets spun up for a particular user is in a separate container, sort of by default. You can do some things to to mess with that. But by default, every user that logs in and and wants to start an app essentially launches a docker container that creates a lot of separation between users. But, you know, the other interesting thing here, you know, that I think a lot of these types of frameworks have, like, you know, shiny proxy and obviously, posit connect, is that tiny shiny server boasts a web based dashboard

on port thirty eight thirty nine that shows you sort of admin level statistics on, you know, real time connection counts, IP addresses of users, app health status. Supposedly, again, I haven't played around with this. I don't know if you peaked at port thirty eight thirty nine while you were trying it out yourself, Eric. Yeah. I did. It was, yeah, some pretty interesting metrics. I'm, again, all impressed just from the R side of things. It's crazy to me. Very interesting. And then, the the last sort of feature around management and monitoring is automatic memory management. It boasts, you know, such as, you know, cleaning up stale connections after thirty

minutes, assuming that's the default, and then clearing, you know, dead process objects that obviously, may remain after, you know, users end their session. But very, very interesting, you know, particularly I think for those that find or don't need, you know, crazy full scale production deployments for, what they're trying to host. This might be sort of a quick start version, you know, simple sort of JSON configuration here, that may get you up and running in a smaller

production ish setting or POC setting. Yeah. I mean, again, impressive technically how all this was done, and I can definitely see value in exploring kind of under the hood from the friendly confines of R how all these different concepts work. So yep, this got my attention and yeah, I may play with it a bit more just to see just what are the internals of this package really doing. And my quick glance last night, it looked like some use of reference classes and whatnot. So it's definitely using some kind of object oriented programming,

which again, interesting to see what, you know, what drives behind the choices of these different frameworks and also seeing how this compares with some of the great packages that Thomas Lynn Peterson at POSIT has created with his take on web technology and R with like his Fiery package and other packages that the Plumber two reinvention of a plumber is basing off of as its dependency footprint. So yeah, definitely something I'm going to ponder and like I always say every time we do this show, I learn something new and this was by far

a new take on Shiny session management. Again, straight from R itself. So who knows what you find out on these issues and you can find out a lot more. Guess what? You just go to rlk.org. This is the first issue of 2026 so we hope to have obviously many more after this and we have our curator team ready to help out with the rest of the issues. Before we go, I do want to acknowledge some interesting feedback not so much feedback from our contact page but we were recently,

I recently got a nice note from a, listener, I believe, Yan Terman over on Blue Sky, who did an interesting analysis of scraping the transcripts of our very podcast and doing the episode length across time of what we've talked about. And, you know, I put a link in our show notes, Mike, if you want to check it out, but there's an interesting spike apparently in 2022 somewhere where we had a long episode and it's been a general trend back up.

You and I doing long episodes, who would have thought that, right? But it's an interesting graph and apparently he's been doing listening to us while biking across across America. So, hey. If we can make your bike rides more fun, we're we're here for you, buddy. That is awesome. Thank you for that analysis. I can't wait to check it out. And we may ramble once in a while, but only only once in in a while. It's sporadic. Mean, the data shows it. Perfect. Who knows? Yeah. Yeah. No. Anyway, you tune in for the R content. Maybe you stay for the rants, depending on how that flipped around. I have no idea. Nonetheless,

if you want to find out more, obviously everything's at rok.org. This is a community project, obviously we value your input and how you can help out is sending those great resources via a poll request to the github repository of R Weekly. Everything's available at rweekly.org. You know, it's not hard to find and also we love hearing from you wherever you shout us out on social media or you get into our contact page on the episode show notes. We're all right there. This podcast has great transcripts as we just heard, as well as chapter markers, so you can always skip to whatever highlight you're interested in. Hopefully, you listen to the whole episode, but we won't judge.

Nonetheless, we're just happy you're listening at all. So with that, you can get in touch with me on social media in various ways. I am on Blue Sky with r podcast@bsky.social. I'm also on Mastodon or I'm rpodcast@podcastindex.social, and I'm on LinkedIn. Just search my name and you will find me causing trouble there. Mike, where can listeners find you? You can find me on blue sky at mike dash thomas dot b s k y dot social,

or you can find me on LinkedIn by searching Ketchbrooke Analytics, k e t c h b r o o k, and see what I'm up to lately. Very nice stuff, and it's great to get back in 2026 with you. And, yeah, I'll try to keep my ranting to a a minimum sometimes, but I can't guarantee it. But nonetheless, I think we got relatively unscathed here. Any feedback's welcome no matter how far I take it. Alright. Well, that'll do it for episode 217 of highlights, and we will be back with another edition of RWKY highlights

next week.

Open the transcript file

Hosts and Guests

Hosts

  • Eric Nantz Host Feed-stated
  • Mike Thomas Host Feed-stated

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Issue 2026-W47 Highlights

    Nov 20, 202648 minAudio

    Our takes on the increasing removal of friction in good and not-so-good ways thanks to AI, a visual tour of hot air balloon accident data, and a case study of novel image recognition with the rplatipus package. Episode...

    Credit the people on this episode

  2. Issue 2026-W46 Highlights

    Nov 13, 202633 minAudio

    The flow of both data and analytical jobs just got smoother with mizu, opening the door for multiple agent harnesses in Positron, and an optimal statistical model tailored to complex experimental designs with a friendly...

    Credit the people on this episode

  3. Issue 2026-W45 Highlights

    Nov 6, 202644 minAudio

    What happens when that online data set isn't so easy to download anymore? We learn of two packages to ease the pain of obtaining these files and then some. Plus a visit to the visualization corner with a handy way to...

    Credit the people on this episode

  4. Issue 2026-W44 Highlights

    Oct 30, 202631 minAudio

    The celebration of rOpenSci's 15-year anniversary continues with the community sharing their perspectives, and a fun Shiny app that could be quite useful during celebrations on a weekend or the holidays. If you ever...

    Credit the people on this episode

  5. Issue 2026-W43 Highlights

    Oct 23, 202659 minAudio

    We learn about the variety of ways you can contribute back to the core of R, a roundup of the excellent choices available for package and environment reproducibility, and an amazing collection of Shiny applications and...

    Credit the people on this episode