HyperparameterHunter

HyperparameterHunter is an open-source Python library that makes hyperparameter optimization painless and keeps a permanent record of every experiment you run. It wraps the estimators you already use — from Keras and scikit-learn to XGBoost, LightGBM and CatBoost — so tuning, cross-validation and result saving happen through one consistent interface. The project is hosted on GitHub, where it has earned roughly 705 stars.

Overview

Machine learning experimentation has a bookkeeping problem: dozens of model variants get trained, and a week later nobody can say which hyperparameters produced the best score or reproduce the run. HyperparameterHunter treats experiment management as a first-class concern. As you optimize — with random search, Bayesian-style optimization and other strategies — it automatically saves each experiment’s parameters, cross-validation scores and predictions to disk in a queryable form, so results accumulate into a personal knowledge base instead of vanishing into terminal scrollback. Over months of work, that accumulated record becomes genuinely valuable — you can see at a glance which ideas you have already tried and what they achieved. It also reuses past results intelligently, skipping retraining when an identical experiment already exists. The library is documented on ReadTheDocs with extensive examples.

Key Features

• Unified hyperparameter optimization across Keras, scikit-learn, XGBoost, LightGBM, CatBoost and more
• Automatic saving of every experiment’s parameters, scores and predictions
• Result reuse — identical past experiments are recalled instead of retrained
• Multiple optimization strategies in one consistent interface
• Cross-validation and feature-engineering workflows built in
• Thorough documentation and examples on ReadTheDocs

💳 Pricing

HyperparameterHunter is free and open source under the MIT license. There is no paid tier, cloud dashboard or enterprise edition.

Who Is It For

HyperparameterHunter is for data scientists who run lots of model experiments and want tuning and meticulous record-keeping in the same tool, without adopting a heavyweight experiment-tracking platform.