Through the years, our mission at SigOpt has remained the same: to empower the world’s experts with software that enables them to design, explore, and optimize more impactful experiments. We are excited to share a new product development that will further advance this mission. SigOpt has released an open source offering. Learn more about the open source project.  

We will maintain the hosted software-as-a-service version of SigOpt and continue to enable the scalable customer experience while using this product. This open source version of SigOpt will enhance, rather than replace, the ways in which users can interact with our product. By offering SigOpt’s tooling in multiple formats, we hope a broader set of experts can use it to advance their research, experimentation, and modeling. We are grateful for the continued support across Intel that yielded this opportunity to  expand access to SigOpt. 

We are proud of what our customers have accomplished with the help of our product. Paul Leu from the University of Pittsburgh relied on SigOpt to optimize experiment design in the process of developing glass with better performance for solar panels and other use cases as documented in this Materials Horizons paper. Rafa Gomez-Bombarelli from MIT used SigOpt to develop generative neural networks capable of designing entirely new materials. And Subutai Ahmad from Numenta relied on SigOpt to custom-training neural networks capable of accelerating inference 100x when combined with Intel 4th Gen Xeon Processors. These examples are only scratching the surface. Starting today, we look forward to seeing the future advances that will be powered by SigOpt open source.  

We are also proud that SigOpt is part of Intel, which invests heavily in an open, portable and powerful software ecosystem that enables AI & HPC developers to maximize the potential of their hardware. SigOpt open source is only the most recent investment Intel has made in open software for AI & HPC, with significant investments in oneAPI, OpenVINO, Intel Neural Compressor, OpenFL, Modin and many other projects paving the way long before SigOpt. We are grateful to be part of this growing open ecosystem to enable software developers on whichever hardware they choose. 

Finally, we are grateful for the partnerships we have had through the years, each of which has inspired the version of SigOpt that we are open sourcing today.  Chief among these is our long-time advisor and collaborator Ruben Martinez-Cantin who we met at our first NIPS conference in 2016.  Peter Frazier,  Roman Garnett, Matthias Poloczek and Eytan Bakshy have been consistent supporters and friends of SigOpt, in some cases since even before our founding. Our collaborators have always been valued, with special thanks to Paul Leu, Jungtaek Kim and David Eriksson.  Additionally, we have been proud to sponsor the QMCPy open source project. 

Instructions for getting started with SigOpt open source are at https://sigopt.org. There are two versions of our product that we released as open source: 

• Self-Hosted: The full version of SigOpt that you host on your own machine, including our optimization library with all advanced features, AI module that includes experiment tracking, platform that enables easy parallelization of jobs and web dashboard for managing experiments. 
• SigOpt Lite: A streamlined version of the SigOpt optimization library that you can use to run sequential parameter optimization experiments without setting up a server. 

Whichever version of SigOpt you use, we hope it helps you achieve your modeling goals. 

Happy Optimizing, 

The SigOpt Team