BOP-Elites

[szhaovas]

Description

Implement BOP-Elites.

TODO

• Implement an emitter that maintains GP surrogrates on the objective and measure functions, and emits solutions maximizing the Expected Joint Improvement of Elites (EJIE) acquisition function.
• Implement a scheduler to accommodate archive upscaling.
• Write a simple script that demos how to run a BOP-Elites experiment.

Questions

• None

Status

• I have read the guidelines in
  CONTRIBUTING.md
• I have formatted my code using yapf
• I have tested my code by running pytest
• I have linted my code with pylint
• I have added a one-line description of my change to the changelog in
  HISTORY.md
• This PR is ready to go

[btjanaka]

Thanks @szhaovas! Left a bunch of comments. I haven't really looked at the emitter and scheduler themselves yet; I'll wait for you to fix the big things like how the scheduler works before taking a closer look.

[btjanaka]

We should create another extra for pymoo if we are adding it here. Also, do we actually use pymoo in BOP-Elites?

[szhaovas]

do we actually use pymoo in BOP-Elites?

BOP-Elites uses PatternSearch optimizer to optimize EJIE acquisition. I initially considered making our own pattern search implementation to avoid adding extra dependencies, but we might eventually need pymoo anyways if we decide to add multi-objective optimization to pyribs.

[btjanaka]

Since this is a demo script, I think it's better to just put the params in here directly. Users can then figure out how to configure their parameters on their own later.

[btjanaka]

The general idea of an example is to make something that's fairly clear and easy to understand.

[btjanaka]

extra_fields would need to be included. As stated on my ArchiveBase comment, I think there may be a better way to do this method, but let's come back to it later.