Solving the model representation problem with broom
summary
Broom standardizes messy R model objects by exposing three consistent generics tidy, glance, augment that always return tidy tibbles, so coefficients, one-row fit summaries, and per-observation predictions are easy to access. It pairs with list-columns and purrr::map to fit and summarize many models at once, demonstrated with bootstrap workflows (example uses 100 resamples), nls fits, and ggplot visualizations of coefficient and prediction uncertainty. Broom 0.5 ships these conventions and works reliably to cut cognitive load when comparing or reporting models.
AI-written summary · about
chapters (15)
- 0:00Intro
- 0:40These slides are available online!
- 1:38How does R represent models, estimators and fits?
- 2:44The model representation problem
- 2:54broon provides a standard way to represent fits
- 4:38The normal model: an example
- 5:01What is normal_fit?
- 5:30What is the tidy representation of normal_fit?
- 6:28Another example: the linear model
- 6:59The tidy representation of ln objects
- 9:04Report model coefficients with tidy
- 9:29Comparing models by goodness of fit measures
- 10:23Inspecting residuals from multiple linear regression
- 11:08Bootstrapping
- 16:25Thank you! Questions?
related talks
22mModelling, Tidyverse Way
57mGood Practices for Applied Machine Learning
18mtidymodels: Adventures in Rewriting a Modeling Pipeline
15mInterpretable Machine Learning with R
17mMaking sense of marginal effects
51mProgramming in the Tidyverse
20mgtsummary: Streamlining Summary Tables for Research and Regulatory Submissions
02h 16mTech Tutorial: Create Twitter and APIs