vallydia-register 1.1.0

Thin loader for the Vallydia Ingredient-Evidence Register: evidence-graded cosmetic & research ingredients as pandas DataFrames.

vallydia-register

A thin Python loader for the Vallydia Ingredient-Evidence Register — an evidence-graded open dataset of cosmetic and research ingredients. Every table comes back as a pandas DataFrame.

pip install vallydia-register

from vallydia_register import load_compounds, load_grades, load_citations

compounds = load_compounds()          # 85 rows
grades    = load_grades()             # 456 per-outcome rows

# every A-graded outcome, with the compound it belongs to
best = grades[grades.grade == "A"].merge(compounds[["slug", "name"]], on="slug")

# citations with scholarly cross-links (OpenAlex / Semantic Scholar)
cites = load_citations(enriched=True)

Loaders

Function	Rows	Notes
load_compounds()	85	flat core + pipe-joined lists
load_grades()	456	per outcome; grade may be null (safety/penetration rows)
load_legal_status()	132	per region
load_citations(enriched=False)	117	DOI provenance; enriched=True adds OpenAlex/S2 cross-links
load_cosmetic_claims()	246	permitted / forbidden wordings
load_identifiers()	85	CAS/CID/InChIKey/SMILES with confidence; blanks are honest

Where the data comes from

By default the loader fetches the published parquet from GitHub raw and caches it under your user cache dir. Options:

load_compounds(source="hf")                       # Hugging Face mirror instead of GitHub
load_compounds(source="/path/to/checkout/data")   # a local checkout, fully offline
load_compounds(use_cache=False)                    # always re-fetch

or set VALLYDIA_REGISTER_DATA=/path/to/data in the environment.

Attribution

Data: Vallydia Ingredient-Evidence Register (https://vallydia.com), CC-BY-4.0. Any use — including in derived datasets or model training — must attribute Vallydia. Appearance/evidence reference only; not medical advice.