contramundum53

Refactor `test_pareto_front`

8

## Motivation This is a followup of #3389 and #3754. We refactor the tests of `plot_pareto_front`. ## Description of the changes Refactor the tests of `plot_pareto_front`. We first test whether...

Further refactor `ImportanceEvaluator`s

13

## Motivation This is a follow-up PR to #3552. By slightly breaking backward compatibility of the interface of `ImportanceEvaluator`s, we could make further refactoring. ## Description of the changes We...

Add batch-evaluate mode

2

Kurobako has `evaluate` subcommand that evaluates a problem at a given set of params. However, when we want to repeatedly evaluate different sets of params and the problem needs to...

Currently, debugging subprocesses run by kurobako is rather painful. There can be multiple possible improvements. * We could show that "During study <study setting>, subprocess <id> finished with an error...

Introducing sampler-dependent visualization

18

### Motivation Black-box optimization is inherently heuristic, and in many problems the algorithms don't work well out of the box but needs some hyper-hyper-parameter tuning. However, Optuna currently don't provide...

Enhance performance of GPSampler

5

## Motivation This PR enhances the performance of `GPSampler` to the same level of (if not above) BoTorch. ## Description of the changes This PR introduces the following changes: *...

Migrate from `threading.local` to `contextvars`

5

### Motivation https://github.com/optuna/optuna/blob/3fcbf8839c760916cc0cf38fb545f526d5efe85d/optuna/study/study.py#L56 uses `threading.local`, but `contextvars` can make the code more sophisticated (https://peps.python.org/pep-0567/#converting-code-that-uses-threading-local). ### Suggestion Use `contextvars` instead of `threading.local`. ### Additional context (optional) _No response_

Bring `_CachedStorage` into `RDBStorage`

5

### Motivation `_CachedStorage` is only used from `RDBStorage`. It can be moved into `RDBStorage`. ### Suggestion Move `_CachedStorage` (https://github.com/optuna/optuna/blob/3fcbf8839c760916cc0cf38fb545f526d5efe85d/optuna/storages/_cached_storage.py#L38) into `RDBStorage`. ### Additional context (optional) _No response_

Make `logei_candidate_func` in `BoTorchSampler` support constraints

1

### Motivation Straightforward extension of #4546 for constrained objective functions. Currently `logei_candidate_func` does not support constraints, because `botorch.acquisition.analytic.LogExpectedImprovement` does not take constraint argument. However, BoTorch has a separate class `botorch.acquisition.analytic.LogConstrainedExpectedImprovement`...

## Reference Issues/PRs ## What does this implement/fix? Explain your changes. The initial direction number of Sobol engine is not rectangular. This is likely unintentional.