Adjusted R^2 missing from yardstick

[Andrew_Mcmahon]

I see that yardstick is missing adjusted R^2. I'm not troubled by this, but I am in a working group that is approaching tidymodels for the first time and I keep getting asked why it isn't available.

I'm trying to understand the justification for leaving it out. Is it that:

1. Using adjusted R^2 to evaluate a trained model something you want to discourage; and
2. The advantages of using adjusted R^2 are slight when fitting a complex model?

Thank you.

[Max]

We won't be adding it; tidymodels uses a holdout set to measure performance.

Adjusted R^2 is only used when the data set used for modeling is re-predicted. It basically penalizes via the df.

Using a different set of data to evaluate the model gives valid R^2 estimates.

[Andrew_Mcmahon]

Thank you That makes sense now that I understand.

I actually didn't know that that was the purpose of adjusted R^2. There are so many online "learn data science" resources that just say "always use adjusted R^2, never R^2". There are so many data analysts out there who don't always know the theory as well as we should (myself included).