Improve dataset card

#6
by davanstrien HF Staff - opened
BigLAM: BigScience Libraries, Archives and Museums org

Replaces the boilerplate template card (ToC + 14 [More Information Needed] stubs) with a short written card: sampling design, label counts, field descriptions, the greyscale/score/OCR traps, licence and citation.

Sources: Zenodo deposit 10.5281/zenodo.5838410; Programming Historian lesson 10.46430/phen0101; Newspaper Navigator (news-navigator.labs.loc.gov); the nnanno repo. Counts and the sampling pattern (61 adverts per year, every fifth year 1860-1900; 301 titles; 21 empty OCR fields; score range 0.900-0.998; all images PIL mode L) were computed from the parquet files.

YAML frontmatter is unchanged byte-for-byte.

davanstrien changed pull request status to merged

Sign up or log in to comment