Skip to content

Vendored datasets

Small, public classic functional-data datasets used by the pages under docs/examples/. They are loaded at documentation build time through the helper scripts/docs_data.py (importable as docs_data because scripts/ is on PYTHONPATH during the build, alongside docs_fig).

Each loader returns (argvals, X, meta): a shared 1-D grid, an (n_obs, n_points) array of curves (one observation per row), and a pandas.DataFrame of per-observation labels aligned to the rows of X.

File Loader Shape (obs × pts) Source License
growth.csv load_growth() 93 × 31 growth dataset, R fda package GPL-2/3
canadian_weather.csv (+ _precip, _meta) load_canadian_weather() 35 × 365 CanadianWeather, R fda package GPL-2/3
tecator.csv load_tecator() 240 × 100 StatLib Tecator dataset Public domain (redistributable)
phoneme.csv load_phoneme() 400 × 256 ElemStatLearn phoneme data Public / redistributable (ESL)
wine.csv load_wine() 178 × 13 UCI Wine dataset CC BY 4.0 (UCI)
sonar.csv load_sonar() 208 × 60 UCI Connectionist Bench (Sonar) CC BY 4.0 (UCI)
(synthetic) load_penicillin() 46 × 200 Simulated in docs_data.py n/a (generated)

Details

growth.csv — Berkeley Growth Study

Heights (cm) of 39 boys and 54 girls measured at 31 ages (1–18 years, yearly to age 8 then biannual). Wide format: column age plus one column per child (M01…M39 boys, F01…F54 girls). load_growth() returns ages, X of shape (93, 31) (boys then girls), and meta with id, sex.

  • Source: data/growth.rda from the R fda package (https://github.com/cran/fda, originally Ramsay & Silverman, FDA).
  • License: GPL-2 | GPL-3 (the fda package license).

canadian_weather.csv — Canadian Weather

Daily mean temperature (°C) for 35 Canadian weather stations over a 365-day year (averaged over 1960–1994). Wide format: column day (1–365) plus one column per station. Companion files: canadian_weather_precip.csv (daily precipitation, mm) and canadian_weather_meta.csv (station, province, region, lat, lon). load_canadian_weather(variable="temperature"|"precipitation") returns the day grid, X of shape (35, 365), and the station metadata aligned to rows.

tecator.csv — Tecator NIR spectra

Near-infrared absorbance spectra (100 channels, 850–1050 nm) of 240 finely chopped meat samples, with lab-measured moisture, fat and protein content (percent). Columns: sample, ch001…ch100, then the three contents. load_tecator() returns the 100 wavelengths, X of shape (240, 100), and meta with the contents.

  • Source: StatLib, https://lib.stat.cmu.edu/datasets/tecator. Recorded on a Tecator Infratec Food and Feed Analyzer.
  • License: Public domain — the StatLib note explicitly permits redistribution ("The data can be redistributed as long as this permission note is attached"); if used in a publication, mention the instrument/company (Tecator). Only the 100 absorbances + contents are vendored (the 22 precomputed principal components in the original file are dropped).

phoneme.csv — Phoneme log-periodograms

Log-periodograms (256 frequencies) of spoken phonemes. This is a balanced, seeded subset: 80 curves from each of 5 classes (aa, ao, iy, sh, dcl) = 400 rows (the full ElemStatLearn file has 4509 rows across these classes; subset with numpy seed 0). Columns: phoneme (class) then f001…f256. load_phoneme() returns the frequency index, X of shape (400, 256), and meta with phoneme.

  • Source: phoneme.data from The Elements of Statistical Learning (https://hastie.su.domains/ElemStatLearn/), originally the TIMIT database.
  • License: distributed with ESL for research/teaching; redistributable subset.

wine.csv — Wine (UCI)

13 chemical measurements (alcohol, phenols, colour intensity, proline, …) for 178 wines from 3 cultivars. This is a multivariate table, used as the input to the Andrews transformation (feature vector → curve). Columns: class (1/2/3) then the 13 features. load_wine() returns the 13 feature names (not an argvals grid), X of shape (178, 13) (raw), and meta with cultivar.

sonar.csv — Sonar (UCI Connectionist Bench)

Sonar return energy in 60 frequency bands for 208 objects — 111 mines (metal cylinders) vs 97 rocks. The 60 values form a natural spectral curve. Columns: b00…b59 then label (Mine/Rock). load_sonar() returns the band index (1..60), X of shape (208, 60), and meta with label.

load_penicillin() — SYNTHETIC fermentation batches

Simulated penicillin-concentration trajectories (no CSV; generated deterministically in docs_data.py, seed 20260805). 40 normal + 6 faulty batches over a 0–400 h fermentation on a 200-point grid. Used purely to illustrate process monitoring; it is not measured data and is labelled as synthetic wherever it appears.

Subsetting / substitutions

  • Phoneme is subset from 4509 → 400 rows (80 per class, numpy seed 0) to keep the asset small and builds fast; all 256 frequency channels are kept.
  • Tecator drops the 22 precomputed principal-component columns present in the StatLib file, keeping the raw 100-channel spectra + contents.
  • No planned source was down; all four come from their canonical public sources (R fda on GitHub, StatLib, ESL).