Installation¶

Requirements¶

Python 3.12 or newer
A supported platform for NumPy, SciPy, pandas, and scikit-learn

The core package includes batch conformal detection, FDR selection, sequential conformal monitoring, and conformal martingales.

Core installation¶

Use the core installation with supported scikit-learn estimators or a custom detector.

pipuv

pip install nonconform

uv add nonconform

Optional extras¶

Install only the capabilities your application needs.

Extra	Adds	Needed for
`[pyod]`	PyOD	PyOD's detector collection
`[data]`	oddball and PyArrow	Packaged benchmark-dataset workflows
`[fdr]`	online-fdr	Online multiple-testing procedures such as GAI and LORD
`[probabilistic]`	KDEpy and Optuna	`Probabilistic()` KDE estimation and tuning
`[all]`	Every optional dependency above	Development or environments that need every feature

For PyOD models and the example datasets used throughout this site:

pipuv

pip install "nonconform[pyod,data]"

uv add "nonconform[pyod,data]"

For every optional capability:

pipuv

pip install "nonconform[all]"

uv add "nonconform[all]"

Sequential monitoring needs no extra

nonconform.monitoring and nonconform.martingales belong to the core installation. The [fdr] extra is for controlling false discoveries across hypotheses tested online. It is not required for conformal martingale change monitoring, and the two guarantee types are not interchangeable.

Verify the installation¶

import nonconform
from nonconform.martingales import SimpleJumperMartingale
from nonconform.monitoring import ExchangeabilityMonitor

print(f"nonconform {nonconform.__version__}")
print(SimpleJumperMartingale.__name__)
print(ExchangeabilityMonitor.__name__)

If an optional import fails, verify that its matching extra was installed into the same Python environment that runs your code.

Next steps¶

Run both core workflows in the Quick Start.
Check supported score interfaces in Detector Compatibility.
Review the API Stability Contract before building a reusable integration.