noislearn

noislearn is an scikit-learn compatible toolkit for label-noise generation, noise filtering, iterative cleaning, noise-score evaluation, and explainable inspection of noisy decisions.

Distance-based filters Classifier-based filters TabPFN explanations CNC-NOS cleaner Noise-score evaluation

Getting started API reference

• :material-filter: Filters

  ---

  Classical, distance-based, ensemble-based, TabPFN-based, and noise-score filters.

  Browse the filters

• :material-broom: Cleaners

  ---

  Higher-level cleaning pipelines built on top of the available filters.

  Open the cleaners API

• :material-book-open-page-variant: Concepts

  ---

  Short guides on noise models, filtering strategies, and local explanations.

  Read the guides

• :material-chart-box-outline: TabPFN explainability

  ---

  Local SHAP-based reports for noisy-instance inspection and auditability.

  Explore the explanation model

• :material-chart-line: Evaluation

  ---

  Fold-based experiments, persisted noise masks, and continuous noise-score ranking metrics.

  Open the evaluation guide

Note

The API pages are generated from the public docstrings in the source tree.