Simon Couch

Some news: I've just open-sourced the draft of a book I'm working on about how #rstats tidymodels users can make their code run faster without sacrificing predictive performance! simonpcouch.com/blog/2024-10-2…

Released a cauldron of R code today 🎃👻 Introducing Kolmogorov-Arnold Networks for Time Series in R! It's an R package called {kantime}, a super experimental port of Nixtla's {neuralforecast} from Python to {modeltime} in R

Monitoring Models in Production is essential to ensure accuracy, detect drift, and maintain fairness over time. For more, check out our recent blog by Myles Mitchell on "Vetiver: Monitoring Models in Production". #DataScience #MLOps #Rstats jumpingrivers.com/blog/vetiver-m…

This is tomorrow! #rstats folks, come through!

i like R

I'll be giving a free virtual talk on speeding up your #rstats tidymodels code at R/Pharma 2024 this Tuesday and would love to see you there! I'll also be announcing a new project.👀 Register here: rinpharma.com

@EatsleepfitJeff The fitted calibration results from the probably package will live inside of the fitted workflow! #calibration" target="_blank" rel="nofollow noopener">probably.tidymodels.org/reference/inde… With current dev packages, you can `extract_postprocessor(wflow)` and then subset into the result, but we should write some nicer extractors for that.

@simonpcouch Is/will there be a way to extract calibration-related metrics like calibration slope, calibration in the large, etc? Thanks!

The #rstats tidymodels team is hard at work implementing postprocessing! There will be changes in almost all of our core packages as well as an entirely new package included in this set of releases. We value your opinion—let us know what you think: tidyverse.org/blog/2024/10/p…

@markgburdon Via twitter reply is fine, you're also welcome to file issues on the relevant repository!

@simonpcouch How would you like to receive feedback?

@MartinHulin Thank you!

@simonpcouch especially for SOTA approach for classification probs github.com/ip200/venn-abe…

@predict_addict @MartinHulin Carrying out conformal inference on predictions from resampled models—i.e. only those predictions that are used to compute metrics—is what I'm saying is not on our roadmap. Should it be? References much appreciated, if so!

@predict_addict @MartinHulin Seems like I'm missing something! tidymodels supports conformal inference currently (to the extent shown in the linked article)... (1/2)

@MartinHulin Re: calibration, any calibration method supported in probably is supported in tailor (and thus postprocessors): #calibration" target="_blank" rel="nofollow noopener">probably.tidymodels.org/reference/inde… Do those methods cover what you're looking for? If not, could you provide a reference? (n/n).

@MartinHulin That said, conformal inference will work just fine on postprocessed predictions: tidymodels.org/learn/models/c… (2/n)

📢 Tomorrow is the official release of @bbiinnyyuu and my book, Veridical Data Science, with @mitpress. I'm so excited to for physical copies to be out in the world! Get your copy from mitpress.mit.edu/9780262049191/…

Updated the documentation for {maize} 🌽 String kernels are the newest addition which were a bit tricky to plant 👨🏻‍🌾 And I'm liking how this table turned out, thanks to @gt_package & gtUtils 🙌🏻

Announcing Shiny Assistant, an AI-powered tool that can help you build Shiny applications! shiny.posit.co/blog/posts/shi…

Postprocessing is coming to tidymodels! We've got a set of changes for updating model predictions coming to the next version of our #rstats packages. Let us know what you think: tidyverse.org/blog/2024/10/p…

@frankiethull 🥞🥞🥞

{maize} works with {stacks} too! 🌽🥞 {stacks} is part of the tidymodels ecosystem for combining many models into a new ensemble model Here's an ensemble example of three SVMs with different types of kernels 🌽🌽🌽