pyproject.toml (1)

19-26:Upper bounds clash with newly-addedrequirements.txt

[project.dependencies] still caps NumPy and SciPy to<2.0.0, yet the freshly-addedrequirements.txt pins them to2.x. Decide which source of truth wins and adjust one of them. At the moment a user installing with Poetry and a user installing from the flat file will end up with divergent ABI versions.

requirements-dev.txt (1)

1-145:Huge fully-pinned dev snapshot will rot quickly

A 400-line frozen list is excellent for archival reproduction but painful to maintain manually. Recommend:

1. Generate it viapoetry export --with dev (orpip-compile).
2. Commit it asrequirements-dev.lockplus a short, readablerequirements-dev.in so humans review only meaningful changes.
3. Refresh automatically in CI (e.g. Renovate).

This keeps dev envs in sync without trapping the project on months-old packages.

requirements.txt (1)

1-6:Add environment markers to model Python-specific constraints

Thepyproject.toml correctly downgrades to NumPy 1.x / SciPy 1.x for Python < 3.13. Pinning unconditional 2.x here silently drops that safety net.

-numpy==2.3.1-scipy==1.16.0+numpy==2.3.1 ; python_version >= "3.13"+numpy==1.26.4 ; python_version <  "3.13"+scipy==1.16.0 ; python_version >= "3.13"+scipy==1.11.4 ; python_version <  "3.13"

pyproject.toml (3)

28-35:Pinningone package with== but leaving the rest floating is inconsistent

Indev you pinruff==0.6.6 but allowpytest,coverage, etc. to float. Either:

1. Adopt a freeze-everything philosophy (create/export a lock file), or
2. Keepall optional-dependency ranges flexible and rely on the lock produced by the installer (Poetry, PDM, etc.).

Mixing approaches invites non-reproducible CI failures.

---

36-47:Consider extracting the notebook / docs stack into its owndocs group

nbtools currently bundles runtime notebook dependencies with Sphinx extensions. Splitting those intonb anddocs (or re-using the growing “docs” convention) keeps environments lean: users who only want to run notebooks do not need to pull in the entire Sphinx tool-chain.

---

49-55:Redundant[tool.poetry] section can be dropped

Poetry ≥ 1.4 fully understands PEP-621 metadata. Keeping onlyrequires-poetry under[tool.poetry] is fine, butpackages can be moved to[tool.poetry.dynamic]or declared via the newpackages = [...] key in[tool.poetry.dependencies] for clarity. Removing the legacy table eliminates ambiguity about which metadata Poetry ultimately reads.

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

pyproject.toml (1)

39-40:Avoid hard-pinningruff to an exact patch version
Lockingruff to0.6.6 prevents bug-fix & security updates from reaching consumers. A conservative upper bound is usually sufficient (e.g.<0.7) while still allowing patch-level upgrades.

-    "ruff==0.6.6",+    "ruff>=0.6.6,<0.7",

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

.github/workflows/run_tests.yml (1)

21-25:Potential NumPy ≥ 2 incompatibility with Python 3.9

NumPy 2.0 is dropping support for some older Python versions; final wheels may require ≥3.10. You already exclude3.13+<2, but the reverse (3.9 +>=2) can still appear and break the matrix.

        exclude:+          - python-version: "3.9"+            numpy-version: ">=2"

.pre-commit-config.yaml (1)

8-9:Add a trailing newline to satisfy linters

yamllint is already flagging “no new line character at the end of file”.
Appending a single\n keeps CI happy and avoids needless diffs later.

-    rev: v0.12.3+    rev: v0.12.3+

scripts/format.sh (1)

4-5:Order of commands leads to double work

ruff formatwrites changes, thenruff check --fix may re-write the same files.
Runningcheck --fix first (or dropping--fix here) avoids two formatting passes and speeds CI.

-uv run ruff format bayes_opt tests-uv run ruff check bayes_opt --fix+uv run ruff check bayes_opt --fix+uv run ruff format bayes_opt tests

scripts/check_precommit.sh (1)

4-5:Consider skippingpre-commit install in CI runs

pre-commit install writes to.git/hooks, which is irrelevant in CI and costs ~1 s.
pre-commit run … works fine without it.

-uv run pre-commit install

If the script isonly for local dev keep it, otherwise you can drop it when invoking in pipelines.

scripts/check.sh (1)

4-5:Pass--config pyproject.toml explicitly for forward-compat

ruff auto-detects config today, but the project just switched to PEP-621 layout;
being explicit avoids surprises if the file is relocated.

-uv run ruff format --check bayes_opt tests-uv run ruff check bayes_opt tests+uv run ruff format --check --config pyproject.toml bayes_opt tests+uv run ruff check --config pyproject.toml bayes_opt tests

.github/workflows/build_docs.yml (3)

21-25:Pinsetup-uv major version to avoid breaking changes

astral-sh/setup-uv@v6 is fine today, but GitHub-Action major bumps can introduce breaking changes.
Sticking to the major tagand a minor pin (e.g.v6.0.1) keeps the workflow reproducible, aligning with the PR’s reproducibility goal.

-      - name: Install uv-        uses: astral-sh/setup-uv@v6+      - name: Install uv+        uses: astral-sh/setup-uv@v6.0.1

---

30-30:uv sync --extra dev fails if the extra is renamed

The extra is hard-coded; if the optional-dependency key changes (e.g.,dev ➜docs) docs will stop building.
Consider reusing a shared env var (defined inpyproject.toml or a matrix) to keep this DRY.

---

34-34:Parallel Sphinx build could speed up CI

make github defaults to a single process.
AddingSPHINXOPTS="-j auto" leverages parallel cores at no extra cost:

-          uv run make github+          uv run make github SPHINXOPTS="-j auto"

.github/workflows/run_tests.yml (1)

34-35:uv add afteruv sync defeats reproducibility

uv sync installs the locked set, thenuv add mutates the environment ad-hoc, bypassing the lock file and creating network churn for every CI run.
Prefer a fully-locked path:

1. Add a separateextra (e.g.numpy2) inpyproject.toml and lock it once.
2. In CI, calluv sync --extra dev --extra numpy2.

This keeps builds hermetic and cache-friendly.

.github/workflows/format_and_lint.yml (1)

16-22:Cache opportunity for faster lint jobs

Becauseuv sync will fetch all packages on every run, you can shave ~30-40 s off the job by adding the built-in cache:

-name:Install uvuses:astral-sh/setup-uv@v6with:python-version:"3.9"+    cache:true# enables automatic ~/.cache/uv caching

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

Learnt from: till-mPR: bayesian-optimization/BayesianOptimization#566File: bayes_opt/target_space.py:114-116Timestamp: 2025-05-30T08:08:06.867ZLearning: The `.lb` property of the `ConstraintModel` class in bayes_opt is always an array, so `.size` calls on `constraint.lb` are always valid and will not raise AttributeError.