Data card — Palmer Penguins

The Palmer Archipelago penguins dataset (Horst et al., 2020) is our real-data running example all week: plots, regression, classification, tidymodels pipelines, and Day 4 metrics. Day 2 (Tuesday) slides walk through a full prediction workflow on a course slice of this table.

NoteSynthetic gene notebooks (unchanged)

For LASSO, ridge, and AIC, we keep simulated gene notebooks where we wrote the rules ourselves:

See gene trait and gene disease data cards.

How to load

library(palmerpenguins)
data(penguins, package = "palmerpenguins")

Variables (344 rows in package)

Variable Meaning
species Adelie, Chinstrap, or Gentoo
island Torgersen, Biscoe, or Dream
bill_length_mm, bill_depth_mm Bill size (mm)
flipper_length_mm Flipper length (mm)
body_mass_g Body mass (g)
sex female / male (some NA)
year Study year (2007, 2008, 2009)

Why we use it in this course

  • Plots read well in class (colour by species, pairs of measurements).
  • Same table everywhere — Day 1 EDA, Day 2 fit-check-tune, Day 4 metrics and importance.
  • Practice regression (body mass), binary (species pair or sex), and multiclass (three species).

Student notebooks vs tidymodels modules

Student notebooks (Palmer Penguins)

Caveats (say these out loud)

  • Small sample (~344 rows; fewer after drop_na()) — pedagogy, not big data.
  • Observational — predictions do not prove causal biology.
  • Missing sex — many analyses drop incomplete rows; Day 4 practices imputation inside recipes on an artificial NA slice (see below).
  • Leakage awareness — near-perfect separation is a chance to discuss honest splits, not to oversell accuracy.

Citation

Horst AM, Hill AP, Gorman KB (2020). palmerpenguins: Palmer Archipelago (Antarctica) penguin data. R package version 0.1.1.
DOI: 10.5281/zenodo.3960218.

Data card (raw penguins table)

1. Source

Palmer Archipelago (Antarctica) penguin data via R package palmerpenguins (Horst et al., 2020). Field measurements compiled for teaching and research; not simulated.

2. Outcome

Depends on notebook or slide block:

Task Outcome
Body mass regression body_mass_g (numeric)
Adelie vs Gentoo species (two levels)
Sex classification sex (female / male)
Three species species (three levels)

3. Predictors

Morphometrics (bill_*, flipper_length_mm, body_mass_g), island, sex, year. Which columns enter the model is task-specific (e.g. sex task may omit species).

4. Sample size

344 rows in penguins; 333 after drop_na() on all variables. sex is the main source of missingness in the raw table.

5. Leakage risks

  • Imputing or scaling on the full table before splitting leaks test information into training (Day 2 recipe + CV fixes this).
  • Dropping NA on the full table before split can bias which penguins remain (Day 4 imputation block).
  • Tuning on the same rows you report as final test performance.

6. Subgroups

species, island, sex, year — use for fairness-style checks (e.g. accuracy by island) when discussing limits of a model.

Course teaching slices (pedagogical derivatives)

These are not separate field datasets. Instructors build them from penguins in slides (_includes/day04-penguins-setup.qmd) to stress imbalance, missing data, and PCA. Document them on your data card when you use them.

Object What it is Typical n / notes
peng_cls / peng_bal Adelie + Gentoo only; outcome y; predictors bill length/depth, island, sex, year; no flipper or body mass; drop_na() ~219 rows
peng_imb3 Three-species slice with downsampled Chinstrap (deterministic retention of 15 rows closest to Adelie cloud) for imbalance / upsampling demos ~146 Adelie, ~119 Gentoo, ~15 Chinstrap
peng_na Copy of balanced slice with injected NA on bill_length_mm (~22 rows) and sex (~5 rows), set.seed(4) For imputation demos only
peng_pca Adelie + Gentoo with flipper and body mass kept for PCA recipe step ~219 after drop_na()

Leakage note for slices: artificial NA and imbalance are for teaching. In a model card, state that prevalence and missingness on peng_imb3 / peng_na do not match the wild population.

Back to all datasets · Lab exercises