Install - onceover

onceover is a Python package with a Rust core. Binary wheels are the normal path; building from source needs a Rust toolchain.

pip install onceover

The package requires CPython 3.10+ and Polars 1.28–2.x. Wheels cover Linux x86_64 and aarch64, macOS Intel and Apple silicon, and Windows x86_64. The source distribution additionally needs Rust and a C compiler — the same setup used for development, below.

Verify the installation¶

Verify the installation by importing onceover and profiling a small DataFrame. This loads the compiled extension and runs the pipeline.

onceover can profile in-memory DataFrames with millions of rows and hundreds of columns. Column summaries and integrity checks examine every row. Relationship discovery uses bounded samples and searches; see Coverage, budgets, and deadlines for the limits and coverage fields.

import polars as pl
import onceover

print(onceover.__version__)

df = pl.DataFrame({"id": [1, 2, 3], "value": [0.5, 1.5, None]})
report = onceover.profile(df, budget="quick")
print(report.overview["headline"])

0.1.2
3 rows × 2 columns (quick budget); key candidate: id; 0 functional dependencies, 0 predictive dependencies, 0 conditional rules

Optional dependencies¶

Two extras install what the surrounding tooling needs:

Extra	Installs	Used by
`onceover[notebook]`	altair, great-tables, vl-convert, numpy, jupyterlab	the analysis notebook, charts, tables
`onceover[benchmark]`	numpy, psutil	the scale benchmark

pip install 'onceover[notebook]'

Development installation¶

Development builds the Rust core and binds it to a virtual environment:

python -m venv .venv
source .venv/bin/activate
pip install 'maturin>=1.15,<2' pytest scipy
maturin develop --release
cargo test --workspace --locked
pytest tests/

maturin develop --release compiles the workspace and installs the extension into the active environment. Repeat it after changing Rust sources; the Python side is plain source and needs no build step.

The profiler releases the GIL during computation. Multiple threads can therefore call profile concurrently without serializing on the interpreter.