Libra is an ergonomic machine learning library that lets non-technical users build, tune and explain models with a handful of high-level function calls. Where typical AutoML tools still expect data-science fluency, Libra hides preprocessing, model selection and hyperparameter search behind plain-English style APIs such as training a model to predict a named column. It is an open-source project on GitHub with around 1,900 stars, updated in 2025.
Written in Python on top of scikit-learn, Keras and related libraries, Libra automates the repetitive middle of a machine learning project: cleaning and encoding data, splitting it correctly, trying suitable model families, tuning them, and producing plots and metrics a beginner can interpret. Queries are written as short function calls with sensible defaults, and the library returns fitted models that can be saved, reloaded and used for prediction. Its documentation and tutorials target analysts, students and founders who need working models without first becoming ML engineers.
Free. Libra is open source under the MIT license and installs from PyPI at no cost.
Libra also generates explanatory plots โ feature relationships, target distributions and model performance charts โ so users can sanity-check what the model learned instead of trusting a single score. Trained models expose straightforward predict calls and can be persisted to disk, and the library’s dataset helpers cover common loading and cleaning steps. For many tabular problems, a Libra script of a dozen lines replaces what would otherwise be a long scikit-learn pipeline.
Analysts, students, product managers and small teams who need credible machine learning models fast, and educators teaching applied ML without heavy framework overhead.