Osprey is a tool for practical hyperparameter optimization of machine learning algorithms. It’s designed to provide a practical, easy to use way for application scientists to find parameters that maximize the cross-validation score of a model on their dataset. Osprey is being developed by researchers at Stanford University with primary application areas in computational protein dynamics and drug design.


osprey is a command line tool. It runs using a simple config file which sets up the problem by describing the estimator, search space, strategy, dataset, cross validation, and storage for the results.

Related tools include and spearmint, hyperopt, and MOE. Both hyperopt and MOE can serve as backend search strategies for osprey.

To get started, run osprey skeleton to create an example config file, and then boot up one or more parallel instances of osprey worker.