@sebffischer

79 Followers
74 Following
310 Posts

PhD Student in Statistics / Machine Learning.
Working on mlr3, which is a machine learning framework in R.

Also working on a code transformation framework for R (jit compilation, autodiff): https://r-xla.github.io/anvl/

sebffischerMay 7
shikokuchuoMay 7

OpenTelemetry has come to R 🎉

We've instrumented Shiny, plumber2, mirai, httr2, ellmer, knitr, testthat, and DBI — so you can get production-grade traces from your R apps with no code changes. Just set a few env vars and telemetry flows to the backend of your choice.

Full write-up: https://opensource.posit.co/blog/2026-05-07_opentelemetry/

#RStats #OpenTelemetry #Observability

Bringing OpenTelemetry to R in production

Posit has instrumented Shiny, plumber2, mirai, httr2, ellmer, knitr, testthat, and DBI with OpenTelemetry, and created tools for you to instrument your own packages, bringing production-grade observability to R.

Posit Open Source
sebffischerMay 7
In an R package, I am linking against R's LAPACK but ideally would like to have access to the the single precision float routines, which are not bundled there (on windows) because R does not use them. Does anyone have a recommendation how to also get access to f32 routines on windows? #rstats
Show threadsebffischerMay 6
@defuneste Use man-roxygen templates to deduplicate. They are not recommended anymore (for reasons that are not clear to me) but I love them and they work mostly. I use them heavily, e.g. here: https://github.com/r-xla/anvl/blob/main/man-roxygen/section_nv_cum_relation.R or here: https://github.com/mlr-org/mlr3torch/tree/main/man-roxygen
anvl/man-roxygen/section_nv_cum_relation.R at main · r-xla/anvl

Accelerated array computing and code transformations for R - r-xla/anvl

GitHub
Show threadsebffischerMay 1
One thing which might be nice is to have a ..(n) function that allows to skip n dimensions, so slicing the fourth dimension would become x[..(3), i, ..]
Show threadsebffischerMay 1
as more and more code is produced by LLMs I think it becomes even more important for code to be intuitive and readable.
Show threadsebffischerMay 1
Generally one pain is the:" select i-th axis, where i is a variable", because you can't do x[,,,2] because you don't know how many of those ","s you need.
sebffischerMay 1

What's the best "data.table"-like DSL for n-dimensional arrays? I am already aware of DumPy: https://dynomight.net/dumpy/

I am exploring ideas how to make array computations more readable. #rstats

DumPy: NumPy except it’s OK if you’re dum

reject clever

DYNOMIGHT
Show threadsebffischerMay 1

@dodecadron the capacity of a vector is the true length of a vector. Like I might know that I have to allocate up until n elements, so I could do>

x <- with_capacity("integer", n)
length(x)
#> 0
x[n] <- 100 # no reallocation because internal length is already n
length(x)
#> n

sebffischerMay 1
O Iago  Apr 30

In Memoriam: Tomáš Kalibera

by Jan Vitek

https://prl-prg.github.io/tomas-kalibera.html

#TomášKalibera #RStats #RCore #R

In Memoriam: Tomáš Kalibera - Programming Research Laboratory - CVUT

sebffischerMay 1
Why do I have to write C Code in R to pre-allocate a vector with a certain capacity? #rstats