Jul 31, 2026
|
53 min
Building Modeling Tools in R | Max Kuhn | Data Science Hangout
ADD THE DATA SCIENCE HANGOUT TO YOUR CALENDAR HERE: https://pos.it/dsh - All are welcome! We’d love to see you!
This week’s guest was Max Kuhn, Principal Software Engineer for tidymodels at Posit!
Some topics covered in this week’s Hangout were the origins of the caret package and the move to tidymodels, deep learning for tabular data (TabPFN, TabICL, and TabFM), Max’s background in pharma and computational chemistry (how cool is that), and career advice for breaking into data science.
One community member asked: “What do you think about the recent improvements in deep learning for tabular data, such as TabPFN, TabICL, and TabFM, compared to LightGBM or XGBoost?”
Max’s paraphrased answer: They’re really impressive. These tabular models take things to the next level. The idea is that they train a massive neural network, but not on any real data. They simulate essentially every possible tabular data scenario, with different numbers of rows, columns, and correlations, and train on hundreds of millions of synthetic datasets at once. When you give the model your training data, much like a prompt to an LLM, it uses that as context to funnel down and concentrate on the parts of the network most relevant to your data, so you get an immediate answer without any actual estimation. Because there’s no real fitting, they don’t tend to over- or underfit, and they’re remarkably resistant to things like noise columns that would tank a regular neural network, in ways that are honestly hard to fully explain.
Resources mentioned in the video and chat: caret package on GitHub → https://github.com/topepo/caret Tidy Modeling with R (TMWR) book → https://www.tmwr.org/ Applied Machine Learning for Tabular Data (Max’s book in progress) → https://aml4td.org/ Tidymodels code companion for the book → https://tidymodels.aml4td.org/ tabpfn tidymodels package → https://tabpfn.tidymodels.org/ Max’s R/Pharma talk slides on tabular deep learning → https://topepo.github.io/2025-r-pharma/#/title-slide Julia Silge’s tidymodels blog posts → https://juliasilge.com/categories/tidymodels/ Emil Hvitfeldt’s Quarto slide-crafting workshop → https://pos.it/slide-wrkshp Posit’s Data Science Lab → https://pos.it/dslab Double descent (Wikipedia) → https://en.wikipedia.org/wiki/Double_descent Ab initio quantum chemistry methods (Wikipedia) → https://en.wikipedia.org/wiki/Ab_initio_quantum_chemistry_methods Max Kuhn’s Applied Predictive Modeling talk on YouTube → https://www.youtube.com/watch?v=dB-JHhEJvQA Max Kuhn’s talk on TabPFN: A Deep-Learning Solution for Tabular Data → https://youtu.be/vJVB5N8-21c Western PA Regional Data Center (open data portal example) → https://www.wprdc.org/en Building Engaging Internal Data Communities blog post → https://posit.co/blog/building-engaging-internal-communities Building Engaging Internal Data Communities live discussion (August 5th 2026) → https://pos.it/build-comm
► Subscribe to Our Channel Here: https://bit.ly/2TzgcOu
Follow Us Here: Website: https://www.posit.co Hangout: https://pos.it/dsh The Lab: https://pos.it/dslab LinkedIn: https://www.linkedin.com/company/posit-software Bluesky: https://bsky.app/profile/posit.co
Thanks for hanging out with us!
Timestamps: 00:00 Introduction 03:31 “Can you tell us about caret and how it came about, and where it falls in the timeline of you joining Posit?” 11:14 “Are you working on anything right now?” 16:14 “What do you think about the recent improvements in deep learning for tabular data, such as TabPFN, TabICL, TabFM, etcetera, compared to LightGBM or XGBoost?” 24:32 “What does a normal day or week look like for you, Max?” 28:28 “What type of computational chemistry did you work on?” 34:02 “Can you tell us about the JoCo cruise?” 37:22 “If you could analyze any dataset in the world, what would it be, and what would you be looking for?” 39:01 “Do you have any book recommendations for getting into machine learning using tidymodels?” 41:09 “What do you find most challenging or rewarding about writing packages that do both inference and prediction?” 45:24 “Do you have a piece of career advice that you find meaningful, was helpful for you, or something you wish you knew before you got into the space you’re in?”
Emil Hvitfeldt, Julia Silge, Max Kuhn
Quarto
tidymodels
TMwR
