666ofmine/nutrition-food-composition

营养成分权威数据 / Authoritative food composition data

Collected under assistant 营养成分 into this dataset. Goal: official, citable food composition sources with reproducible scripts and full provenance.

Hard rules used during collection

1. Keep all fetch/processing scripts (scripts/, team/*/scripts/).
2. Provenance in sources.md / team/*/sources.md / manifests (agency, dataset, version, URL, access date).
3. Cross-validate when multiple sources exist; never fabricate values.
4. Prefer government / UN official bulk downloads.

Layout

Path	Contents
team/0/	USDA FoodData Central — Foundation Foods (2026-04-30) + SR Legacy (2018-04) official CSV zips, extracted tables, download script + SHA256 manifest
team/3/	UK CoFID / McCance & Widdowson 2021 official Excel + user guide + download script
raw/cofid/	CoFID Excel provenance mirror
parquet/	Hub-mainstream Parquet mirrors of CSV / XLSX sheets (for DuckDB hf://, Polars, datasets)
team/1,2,4,5/	Reserved for China FCT, FAO/INFOODS, Japan STFC, ANSES CIQUAL
sources.md	Master provenance index
ASSIGNMENT_BOARD.md / TEAM_INDEX.md	Team board

Original CSV/XLSX are kept as provenance. PDFs and source ZIP archives are not converted.

Parquet naming

• CSV → parquet/<same relative path>.parquet
• XLSX sheet → parquet/<dir>/{workbook_stem}__{sanitized_sheet}.parquet

Configs (selected tables via load_dataset)

config_name	Source
usda_foundation_food	FDC Foundation food
usda_foundation_food_nutrient	FDC Foundation food_nutrient
usda_foundation_nutrient	FDC Foundation nutrient
usda_sr_legacy_food	FDC SR Legacy food
usda_sr_legacy_food_nutrient	FDC SR Legacy food_nutrient
usda_sr_legacy_nutrient	FDC SR Legacy nutrient
cofid_proximates	CoFID 2021 sheet 1.3 Proximates
cofid_inorganics	CoFID 2021 sheet 1.4 Inorganics
cofid_vitamins	CoFID 2021 sheet 1.5 Vitamins

Additional tables live under parquet/ (Foundation/SR Legacy companions, CoFID fatty-acid sheets, field-description workbooks). Use path-based DuckDB/Polars below for any file.

Usage

Hugging Face datasets

from datasets import load_dataset

# Private repo: ensure HF_TOKEN / huggingface-cli login
ds = load_dataset("666ofmine/nutrition-food-composition", "usda_sr_legacy_food")
print(ds["train"][0])

cofid = load_dataset("666ofmine/nutrition-food-composition", "cofid_proximates")

DuckDB (hf://)

import duckdb

con = duckdb.connect()
# Private: SET hf_token='hf_...';  or rely on huggingface-cli credential helper
df = con.sql("""
  SELECT f.description, n.name AS nutrient, fn.amount
  FROM read_parquet(
    'hf://datasets/666ofmine/nutrition-food-composition/parquet/team/0/extracted/sr_legacy/FoodData_Central_sr_legacy_food_csv_2018-04/food.parquet'
  ) f
  JOIN read_parquet(
    'hf://datasets/666ofmine/nutrition-food-composition/parquet/team/0/extracted/sr_legacy/FoodData_Central_sr_legacy_food_csv_2018-04/food_nutrient.parquet'
  ) fn ON f.fdc_id = fn.fdc_id
  JOIN read_parquet(
    'hf://datasets/666ofmine/nutrition-food-composition/parquet/team/0/extracted/sr_legacy/FoodData_Central_sr_legacy_food_csv_2018-04/nutrient.parquet'
  ) n ON fn.nutrient_id = n.id
  WHERE lower(f.description) LIKE '%apple%'
    AND n.name = 'Energy'
  LIMIT 20
""").df()
print(df)

CoFID proximates example:

SELECT "Food Code", "Food Name", "Protein (g)", "Fat (g)", "Carbohydrate (g)", "Energy (kcal) (kcal)"
FROM read_parquet(
  'hf://datasets/666ofmine/nutrition-food-composition/parquet/team/3/raw/McCance_Widdowsons_Composition_of_Foods_Integrated_Dataset_2021__1.3_Proximates.parquet'
)
WHERE "Food Name" IS NOT NULL
LIMIT 20;

Polars

import polars as pl

food = pl.read_parquet(
    "hf://datasets/666ofmine/nutrition-food-composition/parquet/team/0/extracted/foundation/FoodData_Central_foundation_food_csv_2026-04-30/food.parquet"
)
print(food.head())

Licensing / redistribution

• USDA FoodData Central: U.S. government work; see USDA / FDC terms. Redistribution should preserve attribution to USDA ARS.
• UK CoFID: Crown / UK government publication; follow gov.uk / Open Government Licence terms as applicable.
• This repo’s scripts and documentation are provided for reproducibility of the collection pipeline.

Access date

Primary downloads: 2026-09-10 (Asia/Shanghai). Dataset packaged for Hub upload: 2026-09-12. Parquet conversion: 2026-09-12.