[tune] Multi-objective support for Optuna #20489
Conversation
We may have fields in the result dict that are or None. Make sure our results are json serializable.
| # If only a mode was passed, use anonymous metric | ||
| self._metric = DEFAULT_METRIC | ||
|
|
||
| pruner = ot.pruners.NopPruner() | ||
| storage = ot.storages.InMemoryStorage() | ||
|
|
||
| if self._sampler: | ||
| sampler = self._sampler |
There was a problem hiding this comment.
[note] Following samplers don't support multi-objective optimization. So, when self._mode is a list and self._sampler is one of them, Optuna would raise an exception (You can choose other samplers listed in https://optuna.readthedocs.io/en/stable/reference/samplers.html).
- https://optuna.readthedocs.io/en/stable/reference/generated/optuna.samplers.CmaEsSampler.html#optuna.samplers.CmaEsSampler
- https://optuna.readthedocs.io/en/stable/reference/generated/optuna.integration.PyCmaSampler.html#optuna.integration.PyCmaSampler
- https://optuna.readthedocs.io/en/stable/reference/generated/optuna.integration.SkoptSampler.html#optuna.integration.SkoptSampler
There was a problem hiding this comment.
I think it's fine to let Optuna raise it here.
BTW. would it be possible for Optuna to add a property to its samplers that would inform whether the sampler can support multi-objective optimization?
There was a problem hiding this comment.
would it be possible for Optuna to add a property to its samplers that would inform whether the sampler can support multi-objective optimization?
Nice question. Such property is not available now but I'll discuss this with other developers. 👀
Why are these changes needed?
Adds support for multi-objective optimization to
OptunaSearch. Adds test, example, docs.Related issue number
Closes #18027
Checks
scripts/format.shto lint the changes in this PR.