Support MOCMAES

[tenzen-y]

/kind feature

Describe the solution you'd like [A clear and concise description of what you want to happen.] Previously, we supported MOCMAES with the chocolate suggestion service. Although, we removed one suggestion service in #2058 since it is no longer maintained.

So it would be good to support MOCMAES with other suggestion services.

Anything else you would like to add: [Miscellaneous information that will assist in solving the issue.] Follow up on https://github.com/kubeflow/katib/issues/2058#issuecomment-1347749443.

---

Love this feature? Give it a 👍 We prioritize the features with the most 👍

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[tenzen-y]

/lifecycle frozen /help

[google-oss-prow[bot]]

@tenzen-y: This request has been marked as needing help from a contributor.

Please ensure the request meets the requirements listed here.

If this request no longer meets these requirements, the label can be removed by commenting with the /remove-help command.

In response to this:

/lifecycle frozen /help

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[hnanacc]

Hello,

I was trying to work on this issue, but from the current discussions and some research, I found out that there are currently no HPO libraries that support multi-objective CMA-ES, besides chocolate.

The only decent implementation of the algorithm is pycomocma but is no longer maintained (last commit 4 years ago).

For now the following two options seem doable,

1. Fork either chocolate or pycomocma and maintain it.
2. If it's alright with @c-bata, I can add an implementation of MOCMAES with the recent research findings to optuna or add pycomocma as optuna-integration.

NOTE: pycomocma is tested only for bi-objective, although it supports multi-objective.

Please let me know if one of the above works or if there is a better alternative.

Thanks, Harshal

[tenzen-y]

Hello,

I was trying to work on this issue, but from the current discussions and some research, I found out that there are currently no HPO libraries that support multi-objective CMA-ES, besides chocolate.

The only decent implementation of the algorithm is pycomocma but is no longer maintained (last commit 4 years ago).

For now the following two options seem doable,

1. Fork either chocolate or pycomocma and maintain it.
2. If it's alright with @c-bata, I can add an implementation of MOCMAES with the recent research findings to optuna or add pycomocma as optuna-integration.

NOTE: pycomocma is tested only for bi-objective, although it supports multi-objective.

Please let me know if one of the above works or if there is a better alternative.

Thanks, Harshal

I prefer to implement MOCMAES on the optuna side. But I'm not sure if the Optuna community can be acceptable for the support. @hnanacc Could you open a dedicated issue in the optuna side?