Cleanlab

Cleanlab is a Python library for data scientists and ML engineers checking training data for errors. It uses model predictions or feature embeddings to identify problems in text, image, audio and tabular datasets.

• Dataset auditing: Datalab flags likely label errors, outliers, near duplicates, missing values, class imbalance and groups where a model performs poorly, with scores to prioritize review.
• Learning with noisy labels: CleanLearning assigns label quality scores and trains scikit-learn-compatible classifiers on an automatically filtered dataset that excludes low-quality examples.
• Dataset curation: Identify overlapping classes, compare label quality across classes and measure overall dataset health as you revise the data.
• Annotation review: Estimate consensus labels and annotator quality when several people label the same examples, and suggest which examples to label or review next.

Dedicated functionality also covers regression, multi-label classification, token classification, image segmentation and object detection. Cleanlab accepts outputs from models built with PyTorch, TensorFlow, Hugging Face and XGBoost, among other frameworks. Its label checks perform better with predictions on examples the model wasn't trained on.

The open-source library runs on your own machine on Linux, macOS or Windows and uses the Apache 2.0 license.