Tailors are post-model so it didn't really fit.

We didn't include it in the blog post since we mentioned it in the tune post (tidyverse.org/blog/2025/11...) and it already had a post of its own (tidyverse.org/blog/2024/10...).

Let use know what you think of it!

I’m not aware of an Arrow or Parquet format, but there is the ONNX format (see onnx.ai). Depending on the model, you could try Posit’s orbital project, which translates your model to SQL (see here orbital.tidymodels.org).

This is probably the canonical answer to your question: varianceexplained.org/r/teach-tidy...

For some models that are known to tidymodels, you can trick it into just working:

gist.github.com/topepo/0d6f7...

You could create a small function/class/package that encapsulates the object and allows for predictions. The pain is making it work with tidymodels, especially if that involves some novel model type.

They also have simple vector interfaces so if you’re not interested in the tidy API’s that’s not a big deal

Yardstick can use case ways for all of its functions, including sensitivity, specificity, and ROC curves.

No. I generate the random seeds based on the wall clock (to better emulate true randomness).

We had a discussion in my group about this a while back. Use actual random numbers as seeds or common values/sequences, such as 1, 123, 999, 321, etc.

I think to their dismay, I always had 3-4 digit random values, leading to this slide annotation:

Emil!

... and after

I do the same thing over and over and over again.

The kangaroos are nice, but those handles are 💯