Colmena is a library that supports applications which steer large campaigns of simulations on supercomputers. Such "high-throughput" searches are commonly deployed on HPC and are, historically, guided by humans designating a search space manually — a time-consuming process. Colmena was created to explore building applications high-throughput sweeps that replace human steering with Artificial Intelligence (AI) experimental design methods.
We use Anaconda to define an environments:
conda env create --file environment.yml --force
will install all packages needed for the colmena library and demo applications.
Consult our Installation Guide.
We are gradually building demo_apps which illustrate different approaches to using the prototype.
This project was supported in part by the Exascale Computing Project (17-SC-20-SC) of the U.S. Department of Energy (DOE) and by DOE’s Advanced Scientific Research Office (ASCR) under contract DE-AC02-06CH11357.