Tyler H Chang
Tyler H Chang
When libEnsemble runs are started on MacOS and Windows (although Windows is not currently supported) using Python's multiprocessing library, the default is to use ``spawn`` instead of ``fork``. This means...
Should support design variable types inputs that are not embedded/rescaled at all, and passed directly to surrogates. This is necessary to support certain custom surrogate models.
Should implement design variables as a class, so that additional design variable types can be added in the future. - must define a way for design variables to be embedded...
Simply exporting solution data to a ``numpy structured array``, does not help with decision making. We should add tools for visualizing Pareto front data, such as - 2D/3D scatter plots...
Not all users are comfortable working in Python, and ParMOO's interface is necessarily complicated. We should build a GUI interface (for example, using ``tkinter`` or a web-interface), which allows users...
When you solve a MOOP using ParMOO, a surrogate model is trained for each of the underlying simulation outputs. In certain use cases, this surrogate may be the desired product...
We should be able to support ``funcx`` in order to run simulations remotely, by leveraging the interface that already exists in ``libEnsemble`` release 0.9.0: https://github.com/Libensemble/libensemble/releases/tag/v0.9.0
We should allow users to toggle whether they will use named variables, instead of choosing for them based on whether they name all design variables. ParMOO is already set-up to...
MDML provides an interface to support problems involving real-world experiments, while automatically tagging, logging, and timestamping data from real-world experiments in the cloud: https://github.com/anl-mdml/MDML_Client ParMOO has already been integrated with...
It would be nice to track code coverage in ``README.rst``. The project: https://github.com/dbrgn/coverage-badge Seems promising, but we will need a place to host the badge after it has been generated....