Package status and compatibility

This site documents Model Auditor release v0.1.16 at commit 9408be104264ef04115974a6ae4dc2ec357f1f95.

Metadata

Property	Value
PyPI distribution	model-auditor
Python import package	model_auditor
Python requirement	>=3.10
Python classifiers	3.10, 3.11, 3.12
Release date	September 5, 2026
Development-status classifier	Alpha
License metadata	MIT
Build backend	setuptools.build_meta
Versioning	setuptools_scm
Typed package marker	py.typed
Core runtime style	synchronous Python library

The source tree does not expose a command-line entry point.

Supported problem shape

The current public API is designed for:

• binary outcomes represented by 0 and 1;
• continuous scores where larger values indicate the positive class;
• scalar or discrete-feature conditional thresholds;
• pandas DataFrames;
• independent marginal subgroup features;
• optional registered intersection features;
• paired score and reference-group contrasts; and
• probability calibration and decision-curve summaries.

It does not provide a native multiclass result model.

Dependencies

Core:

pandas >= 2.2
numpy >= 2.1
scikit-learn >= 1.5
scipy >= 1.14
tqdm >= 4.0

Optional:

plotting: matplotlib >= 3.7, plotly >= 5
styling: jinja2 >= 3.1

Stability implications

The Alpha classifier means users should pin a version for reproducible work and review release-specific behavior before upgrading.

Important stable-release constraints include:

• hierarchy thresholds are accepted but unused;
• custom metric inputs are limited to the built-in input registry; and
• ordinary DataFrame exports are display-formatted strings.

See Exceptions and edge cases.

Testing in the stable tree

The repository includes pytest coverage for:

• known metric values;
• confusion input calculations;
• scalar and conditional thresholds;
• threshold optimization;
• analytic and resampled confidence intervals, diagnostics, and local RNG state;
• score and group comparisons;
• calibration and decision evaluation;
• categorical ordering and placeholders;
• error odds ratios and exports;
• result styling;
• interval plots; and
• score-distribution plots.

The dev extra declares the verification and optional rendering dependencies. The release workflow calls the complete CI workflow before publishing.

Publishing

A GitHub Actions workflow runs when a GitHub Release is created:

1. run tests on Python 3.10, 3.11, and 3.12;
2. run Ruff lint and formatting checks;
3. build distributions; and
4. publish distributions to PyPI with OIDC trusted publishing.

Package users should rely on the selected stable tag rather than unreleased default-branch state.

License note

The package metadata declares MIT. Consult the source repository and distribution metadata for package licensing details. The documentation site itself has separate CC BY 4.0 and MIT licensing boundaries described on Licensing and attribution.