simonpcouch

simonpcouch, 6 days ago to random

A new release of #rstats broom is on CRAN! v1.0.6 includes several changes to well-used tidiers from the package, e.g. for lm(), gam(), and survfit() output.

https://github.com/tidymodels/broom/releases/tag/v1.0.6

simonpcouch, 10 days ago to random

tidymodels has long supported parallelizing model fits across CPU cores. A couple of the modeling engines that #rstats #tidymodels supports for gradient boosting—#XGBoost and #LightGBM—have their own tools to parallelize model fits. A new blog post explores whether tidymodels users should use tidymodels' implementation, the engines', or both.

https://www.simonpcouch.com/blog/2024-05-13-parallel/

simonpcouch, 1 month ago to random

Since the #rstats tidymodels team sent out a bunch of package releases a few weeks ago, we've mostly shared about the two major features that those releases implemented: fairness assessment and survival analysis. There are all sorts of other goodies in those releases, though; this post on the tidyverse blog highlights a few from tune 1.2.0.

https://www.tidyverse.org/blog/2024/04/tune-1-2-0/

simonpcouch, 1 month ago to random

The development versions of #rstats tidymodels packages now include tools to benchmark training time. In this example, we identify an ML model that trains >10x faster than the most performant option with very little decrease in predictive performance: https://www.simonpcouch.com/blog/2024-04-08-fit-time/

simonpcouch, 1 month ago to random

Many hospital systems use machine learning models to help allot limited care resources. A new article on the #rstats #tidymodels website explores claims that these models may be discriminatory:

https://www.tidymodels.org/learn/work/fairness-readmission/

simonpcouch, 1 month ago to random

Many companies have popped up in the last couple years claiming that they can detect whether text was AI-generated or not using a machine learning model. A new article on the tidymodels website investigates claims that these detectors treat non-native English writers unfairly using #rstats:

https://www.tidymodels.org/learn/work/fairness-detectors/

simonpcouch, 2 months ago to random

New on the tidyverse blog: a whole bunch of tidymodels releases just made it to #rstats CRAN.🚀

One major addition is a set of tools for assessing model fairness, designed to help practitioners understand potential harms that could arise from a machine learning workflow.

Read more: https://www.tidyverse.org/blog/2024/03/tidymodels-fairness/

simonpcouch, 2 months ago to random

The newest tidymodels user survey is still live! Please fill it out while it's still up; responses to our previous surveys led to many of the new packages we've released in recent years.

Read more on the #rstats tidyverse blog: https://www.tidyverse.org/blog/2024/02/tidymodels-2024-survey/

simonpcouch, 2 months ago to random

Every couple years, the #rstats tidymodels team puts out a user survey to help us better prioritize what we'll work on next. The results of this survey led to the {agua}, {stacks}, and {spatialsample} packages, among others. Our newest survey is up--please take it!

https://www.tidyverse.org/blog/2024/02/tidymodels-2024-survey/

simonpcouch, 3 months ago to random

A new infer release is on CRAN! infer is an #rstats package for tidyverse-aligned statistical inference using randomization. infer v1.0.6 improves documentation, makes error messages more informative, and fixes a few bugs.

Release notes: https://infer.tidymodels.org/news/index.html#infer-v106

simonpcouch, 5 months ago to random

I (virtually) dropped by University of Wisconsin-Madison last week to present a webinar on tidymodels. Heading into the holidays, I thought a fun example problem might be to try and predict flight delays using flights data from Madison's airport. That example notebook is adapted as a post on my blog!🛩️

https://www.simonpcouch.com/blog/2023-11-28-flights/

simonpcouch, 6 months ago to random

The #rstats tidymodels team has continued refining our error messages to help get folks back on track in their modeling more quickly. I wrote about some upcoming changes on the tidyverse blog: https://www.tidyverse.org/blog/2023/11/tidymodels-errors-q4/

simonpcouch, 7 months ago to random

I wrote a bit about a trick that tidymodels uses to tune models more quickly and the things we think about when designing user interfaces.

A story about a longstanding FIXME in our source code: https://www.simonpcouch.com/blog/submodels-rabbit-hole/

simonpcouch, 8 months ago to random

The tidymodels session at #positconf2023 started out with a great talk from @ryantimpe on The LEGO Group's transition to #rstats tidymodels. The tidymodels team will be hanging out at the Lounge if you want to drop by after the session!

simonpcouch, 8 months ago to random

Feeling so, so stoked to see #rstats friends at posit::conf(2023) next week.🙂

Find me for a limited edition raspberry variant of the stacks hex.👀

The usual sticker, except that this one is composed of shades of pink with raspberries on the pancakes.

simonpcouch, 8 months ago to random

infer v1.0.5 is now on #rstats CRAN!

Among other things, this release introduces support for randomization-based inference on paired data.

More on that functionality in a new vignette: https://infer.tidymodels.org/articles/paired.html

juliasilge, 9 months ago to random

I know I've said this before in various venues, but one thing I am so jealous of Python :python: developers for is that there is A LOT of cute snake-themed jewelry. 🐍

simonpcouch, 9 months ago to random

This post is now a few months overdue, but I think it's still worth sharing. A short update for all those who've supported me in my work (and personal) life—I appreciate yall.🙂 https://www.simonpcouch.com/blog/moving-on-s23/

simonpcouch, 10 months ago to random

I came across a paper "GPT Detectors Are Biased Against Non-Native English Writers" a few weeks ago, where the authors argue that models predicting whether a text sample was written by AI or a human will disproportionately classify samples written by non-native english writers as AI-written.

I took some time to tidy the source data + drop it into an #rstats data package, and have really enjoyed poking around with it. You might too: https://github.com/simonpcouch/detectors

simonpcouch, 11 months ago to random

The #rstats tidymodels team recently converted our website to use #QuartoPub, allowing us to more easily maintain and contribute content.

There are a few bonus features that came along with this conversion too, though! One that's super exciting to me is the searchable broom method table—if you have a model object to summarize but can't find tidier methods for it, you can search across all CRAN packages here🤯

https://www.tidymodels.org/find/broom/

simonpcouch, 1 year ago to random

New post on the #rstats tidyverse blog: 10-ish tips and tricks for speeding up tidy code🏎

When computational efficiency is the priority, switching from functions we love in dplyr and tidyr to the backend tools underlying them can result in substantial speedups.

https://www.tidyverse.org/blog/2023/04/performant-packages/