add multi objective BOHB optimizer

[14e6]

Reuses most of single objective BOHB, but replaces the single-objective based decisions with ranking configurations by nondominated (Pareto) front and breaks ties within the boundary front by crowding distance (NSGA-II style).
I moved a couple of tests around and the shared KDE sampler bit around to get a clean base for the single and multi objective variants, apologies for that - hope it doesn't make this PR that unnecessarily complex to read.
Also, I broke some of the public import paths, thus the bump of the major version.

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
blackboxopt
__init__.py
blackboxopt/optimizers
bohb.py					28-29
hyperband.py					29-30
mobohb.py					21-22
blackboxopt/optimizers/staged
bohb.py
hyperband.py
iteration.py
kde_sampler.py					71-72, 76-82, 273, 321, 375-380, 392-396, 475
mobohb.py					34
optimizer.py
utils.py
Project Total

This report was generated by python-coverage-comment-action

[14e6]

Most of this just moved to kde_sampler.py, I like a side by side diff to make this easier to read.

[14e6]

These also just moved from the staged/bohb_test.py

[14e6]

This also just moved from staged/bohb_test.py

[Copilot]

🟡 Changes recommended

Supported dependency versions, invalid-loss ranking, and KDE fallback handling can currently produce import or optimization failures.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Adds MO-BOHB with Pareto-front ranking and NSGA-II crowding distance, while extracting BOHB’s shared KDE and staged scheduling infrastructure.

Changes:

• Adds the MO-BOHB optimizer, dependency extra, documentation, and tests.
• Generalizes staged optimization for single and multiple objectives.
• Moves shared KDE sampling code and bumps the package to 6.0.0.

File summaries

File	Description
blackboxopt/__init__.py	Bumps version to 6.0.0.
blackboxopt/optimizers/bohb.py	Adopts generalized staged scheduling.
blackboxopt/optimizers/hyperband.py	Adopts generalized staged scheduling.
blackboxopt/optimizers/mobohb.py	Adds the public MO-BOHB optimizer.
blackboxopt/optimizers/staged/bohb.py	Retains single-objective KDE specialization.
blackboxopt/optimizers/staged/hyperband.py	Generalizes iteration construction and promotion.
blackboxopt/optimizers/staged/iteration.py	Stores and processes objective vectors.
blackboxopt/optimizers/staged/kde_sampler.py	Extracts shared KDE sampling machinery.
blackboxopt/optimizers/staged/mobohb.py	Adds Pareto ranking and MO-KDE splitting.
blackboxopt/optimizers/staged/optimizer.py	Splits staged schedulers by objective type.
blackboxopt/optimizers/staged/utils.py	Updates scalar promotion for loss vectors.
docs/optimizers/mobohb.md	Documents MO-BOHB behavior.
mkdocs.yml	Adds MO-BOHB documentation navigation.
pyproject.toml	Adds the MO-BOHB extra and version bump.
tests/optimizers/bohb_test.py	Relocates BOHB sampling coverage.
tests/optimizers/mobohb_test.py	Tests public MO-BOHB behavior.
tests/optimizers/staged/bohb_test.py	Retains BOHB sampler-specific tests.
tests/optimizers/staged/iteration_test.py	Updates tests for vector losses.
tests/optimizers/staged/kde_sampler_test.py	Tests shared KDE utilities.
tests/optimizers/staged/mobohb_test.py	Tests Pareto ranking and MO sampling.
uv.lock	Locks version and MO-BOHB dependencies.

Review details

• Files reviewed: 20/21 changed files
• Comments generated: 3
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[14e6]

That fallback was missing the convert_from_statsmodels_kde_representation which I added.

[14e6]

Seems to have moved to:

from pymoo.functions.standard.non_dominated_sorting import (
    efficient_non_dominated_sort,
)