v0.2.0

This release updates optimas to use the latest libEnsemble version, introduces a new Generator, and greatly expands the documentation.

The main changes can be summarized as follow:

• Update to libEnsemble 0.10.2. This implies:
  • Many changes under the hood that should improve resource detection and assignment in HPC environments.
  • The output of an optimas exploration, including the log files, is now fully contained in a single folder (see the documentation for details).
  • The n_procs argument of the TemplateEvaluator is now functional. Therefore, you can now specify the number of MPI processes that your simulation needs (previously, the number of processes was always equal to the number of GPUs given in n_gpus).
  • The default number of GPUs assigned to a TemplateEvaluator is now 0 (i.e., n_gpus=0).
  • An exploration will now fail to start if the workers require more resources than available in the system (e.g., require more workers than CPU cores, or more GPUs than available).
• A new AxClientGenerator has been implemented. This generator allows the user to provide fully-customized AxClient, thus allowing for a higher control of the optimization. Basically, you can now use optimas as a platform to launch an Ax optimization in parallel.
• The documentation has been greatly expanded and now includes a user guide covering the basic usage of optimas as well as many detailed examples.

For all details, see the pull requests below.

What's Changed

• Implement generator with user-defined AxClient by @AngelFP in #100
• Update generator after running exploration by @AngelFP in #101
• Updating for new libE interface by @shuds13 in #88
• Improve documentation and examples by @AngelFP in #104
• Documentation: add submission script on Perlmutter by @RemiLehe in #106
• Add installation instructions for JUWELS Booster by @AngelFP in #102

New Contributors

• @shuds13 made their first contribution in #88

Full Changelog: v0.1.1...v0.2.0