Quack!

Next

Welcome to Quack 🦆!

Quack is a lightweight, clean, and extensible Python library dedicated to Quantification (also known as prevalence estimation).

Unlike standard classification tasks that aim to predict the labels of individual instances, quantification algorithms aim to estimate the class distribution (prevalences) of a target population. This is particularly useful in scenarios subject to dataset shift, such as prior probability shift and covariate shift.

Why quack?

  • Pure Python & SciPy Stack: Built on top of NumPy, SciPy, and Scikit-Learn.
  • Unified Interface: All quantifiers implement a clean, Scikit-Learn-compatible fit/predict API.
  • Shift Simulation: Built-in bag generators allow you to easily simulate and test algorithms against controlled dataset shifts.
  • Core Metrics: Standard error metrics specifically designed to measure quantification performance.

Next Steps

Feature Status
Implement bag generation and test pipelines
Implement ensembles strategies
Create the visualization module

And some other features that aren't listed here.

Features

  • Implementation of quantification baselines as Classify & Count, and its variants (Adjusted Classify & Count, Probabilistic Classify & Count, Probabilistic Adjusted Classify & Count).
  • Implementation of Threshold based quantification methods as T50, Max, X and MedianSweep.
  • Implementation of iteration based quantification methods as Expectation-Maximization Quantifier (EM).
  • Implementation of feature based quantification methods as HDx and ReadMe.
  • Implementation of distribution based quantification methods as HDy, DyS, FMM, etc.
  • Dataset loaders for almost 30 UCI datasets (used on Quantification Review papers).
  • Next steps will include add more loaders to classical papers as (Forman 2008), hierarchical and multilabel dataset.
  • Loaders to signal based datasets from PhysioNet.

Citing

If you use Quack, please consider citing the repository (at the moment we do not have any publication related to the original repository), in your paper.

@misc{quack,
  author = {de Medeiros Júnior, J.G.B.},
  title = {Quack: A Quantification Kit Library for Python},
  year = {2026},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/jose-gilberto/quack}},
}

Contributing

If you want to contribute in any way, look at the CONTRIBUTING.md file, and the instructions at the Contribution page. We appreciate that! :)

Getting Started