Census Income (Adult) dataset

Two charts from the 1994 Census 'Adult' extract: the share of people earning over 50K at each level of education, and a histogram of usual hours worked per week. — Data: Becker & Kohavi (1996), UCI Machine Learning Repository, CC BY 4.0

The standard 'Adult' benchmark extract of the 1994 US Census: 32,561 person records with demographic and employment attributes and an income label, '<=50K' or '>50K'. The recipe reads it from the UCI Machine Learning Repository, which publishes it under CC BY 4.0 with credit to Becker and Kohavi (1996). The recipe treats it as a shape of data, not as a source of fact about anyone; its columns use that census's own categories.

Load it

These commands install arc, Meridian’s open-source Protocol runner, clone the open-analytics repository and run this Protocol, which writes build/census_income.parquet in open-analytics/examples/census-income.

shell

curl -fsSL https://install.meridian.online/arc | bash
git clone https://github.com/meridian-online/open-analytics
cd open-analytics/examples/census-income && arc run

arc runs the Protocol’s SQL on DuckDB 2.x, which its installer puts in place when no DuckDB is on your PATH, saying what to add to PATH; the last step, which describes the Parquet, needs finetype 0.6.54 or later on your PATH.

Columns

Name	Type	Description
age	integer	Age in years.
workclass	string	Employment sector. The source's '?' placeholder is NULL here — 1,836 rows.
final_weight	number	Census sampling weight ('fnlwgt'): how many people in the population this record is taken to represent. Not a property of the person.
education	string	Highest education level reached, as a label.
education_num	integer	The same education level as an ordinal number, so it can be compared.
marital_status	string	Marital status as the 1994 Census coded it.
occupation	string	Occupation category. The source's '?' placeholder is NULL here — 1,843 rows.
relationship	string	Role within the household, as the 1994 Census coded it.
race	string	Race as recorded by the 1994 US Census, using that instrument's categories.
sex	string	Sex as recorded by the 1994 US Census, which offered two values.
capital_gain	integer	Capital gains in nominal 1994 US dollars; zero for most rows.
capital_loss	integer	Capital losses in nominal 1994 US dollars; zero for most rows.
hours_per_week	integer	Usual hours worked per week.
native_country	string	Country of origin. The source's '?' placeholder is NULL here — 583 rows.
income	string	The benchmark's label, kept as the source's own '<=50K' / '>50K' strings rather than folded to a boolean.

Source and licence

Licence

Creative Commons Attribution 4.0 International

Sources

Becker, Barry and Ronny Kohavi (1996). Adult [Dataset]. UCI Machine Learning Repository. https://doi.org/10.24432/C5XW20

Meridian hosts none of this data: the Protocol downloads it, from the address its fetch step names, when you run it.

The steps

4 stepsduckdb >=1.2

Every step runs in order. op: steps are typed, versioned operators; sql: steps are DuckDB models.

01fetch_adultop: http_fetch@1

Fetch adult.data from the UCI Machine Learning Repository, pinned by sha256. UCI has no revisions, so the hash is the only record of which bytes the recipe was written against.

producesbuild/adult.data

with

url: https://archive.ics.uci.edu/ml/machine-learning-databases/adult/adult.data
out: build/adult.data
sha256: 5b00264637dbfec36bdeaab5676b0b309ff9eb788d63554ca0a249491c86603d
headers:
  User-Agent: Meridian Protocol (open-analytics demo; research@meridian.online)

02loadsql: models/load.sql
Read the headerless, comma-separated file under UCI's column names, renamed to snake_case, and turn the source's '?' placeholder into NULL in workclass, occupation and native_country.
readsbuild/adult.dataproducescensus_income
model
```
models/load.sql
```
03export_census_incomeop: parquet_export@1
Write build/census_income.parquet, on your own disk. ORDER BY ALL makes the bytes reproducible: the only rows it leaves unordered are exact duplicates of each other, which cannot change the output.
readscensus_incomeproducesbuild/census_income.parquet
with
```
input: census_income
dest: build/census_income.parquet
compression: zstd
order_by: ALL
```
04describeop: datapackage_describe@1
Emit datapackage.json from the built Parquet: finetype types every column, and descriptor.overrides.json adds the curated title, licence, sources and per-field prose.
readsbuild/census_income.parquetdescriptor.overrides.jsonproducesdatapackage.json
with
```
parquet: build/census_income.parquet
overrides: descriptor.overrides.json
out: datapackage.json
```