A library for Bayesian optimization. This software library provides the core technology for Bayesian optimization of machine learning systems. In particular, calls to the Bayesian optimization utility with Thor are used to generate parameter configurations that are recommended to the consumer of the technology. Thor supports a variety of acquisition functions, support for distributed Bayesian optimization, and persistence across multiple client sessions.
Thor was developed by James Brofos. James is a senior data scientist at The MITRE Corporation where he works to support the data analysis needs of the United States Government. If you encounter problems while using Thor, or have ideas or suggestions for improving the underlying technology, please feel free to contact James by visiting his website.