T-DominO is a simple method of optimizing multiple objective in MAP-Elites, targeting design exploration problems. It does not find a Pareto front of solutions, but a single balanced solution in each bin.
Published at PPSN 2020. Full paper w/supplemental material on arxiv.
T-DominO is implemented as a type of archive in the PyRibs framework, overriding the add
method to base tournaments on T-DominO rather than fitness. PyRibs and other dependencies are included in the environment, so getting it to run should be as simple as this:
Install the virtual environment with conda and activate it:
$ conda env create -f environment.yml
$ conda activate tdomino
Install alpha version of bbq
(PyRibs boilerplate) in the virtual environment:
$ cd bbq; pip install -e .; cd ../
Install tdomino
in the virtual environment:
$ pip install -e .
Test that a basic example runs
$ python test_run.py