Skip to content

build: migrate from Poetry to uv, update dependencies to latest#113

Merged
kimusaku merged 8 commits intomainfrom
build/update-dependencies
Mar 19, 2026
Merged

build: migrate from Poetry to uv, update dependencies to latest#113
kimusaku merged 8 commits intomainfrom
build/update-dependencies

Conversation

@kimusaku
Copy link
Copy Markdown
Contributor

@kimusaku kimusaku commented Mar 18, 2026 •
edited
Loading

Summary

Migrate the package manager from Poetry to uv, and update all dependencies to their latest compatible versions.

Changes

pyproject.toml — Poetry → PEP 621 format

Before After
[tool.poetry] + [tool.poetry.dependencies] [project] (PEP 621)
[tool.poetry.group.dev.dependencies] [dependency-groups] dev (PEP 735)
[tool.poetry.plugins.*] [project.entry-points.*]
build-backend = "poetry.core.masonry.api" build-backend = "hatchling.build"
Poetry ^ shorthand Explicit >=X,<Y PEP 508 ranges

pyproject.toml — dependency version updates

Package Before After
imbalanced-learn >=0.11,<0.13 >=0.11,<0.15
ipykernel ^6.25.1 >=6.25.1,<8.0.0
numba >=0.57.1,<0.61.0 >=0.57.1,<0.65.0
scikit-learn 1.6.1 (pinned) >=1.6.1,<2.0
shap >=0.43,<0.47 >=0.43,<0.52
xgboost >=1.7.6,<3.0.0 >=1.7.6,<4.0.0
black (dev) >=23.7,<26.0 >=23.7,<27.0
isort (dev) >=5.12,<7.0 >=5.12,<9.0
pysen (dev) >=0.10.5,<0.12.0 >=0.10.5,<0.13.0
pytest (dev) >=7.4,<9.0 >=7.4,<10.0
pytest-cov (dev) >=4.1,<7.0 >=4.1,<8.0

requirements-training.txt

Package Before After
category-encoders 2.8.1 2.9.0
statsmodels 0.14.4 0.14.6
tensorflow 2.18.0 2.21.0
wordcloud 1.9.4 1.9.6

uv.lock (new file)

Generated by uv lock — 192 packages resolved.

GitHub Actions workflows

File Change
test.yml Install Poetry + setup-python → astral-sh/setup-uv@v5; poetry install → uv sync --group dev; poetry run → uv run
release.yml Same as above for test/additional_test jobs; release job: poetry build → uv build; sed version replace → uv version $SEMVER; poetry publish → uv publish (env: UV_PUBLISH_TOKEN)
lint.yml pip install pysen flake8 black isort → uv sync --group dev; pysen run lint → uv run pysen run lint

Developer workflow after this PR

# Install deps (creates .venv automatically)
uv sync --group dev

# Run tests
uv run pytest

# Run lint
uv run pysen run lint

# Update lock file after changing pyproject.toml
uv lock

@kimusaku kimusaku requested a review from a team as a code owner March 18, 2026 06:11
@kimusaku kimusaku requested review from AkiraUra and fukuta-flab and removed request for a team March 18, 2026 06:11
@kimusaku kimusaku changed the title build(deps): update dependencies to latest compatible versions build: migrate from Poetry to uv, update dependencies to latest Mar 18, 2026
@openhands-agent
build(deps): update dependencies to latest compatible versions
354a05c 
pyproject.toml:
- imbalanced-learn: >=0.11,<0.13 → >=0.11,<0.15 (latest 0.14.1)
- ipykernel: ^6.25.1 → >=6.25.1,<8.0.0 (latest 7.2.0)
- numba: >=0.57.1,<0.61.0 → >=0.57.1,<0.65.0 (latest 0.64.0)
- scikit-learn: 1.6.1 (pinned) → >=1.6.1,<2.0 (latest 1.8.0)
- shap: >=0.43,<0.47 → >=0.43,<0.52 (latest 0.51.0)
- xgboost: >=1.7.6,<3.0.0 → >=1.7.6,<4.0.0 (latest 3.2.0)
- black: >=23.7,<26.0 → >=23.7,<27.0 (latest 26.3.1)
- isort: >=5.12,<7.0 → >=5.12,<9.0 (latest 8.0.1)
- pysen: >=0.10.5,<0.12.0 → >=0.10.5,<0.13.0 (latest 0.12.0)
- pytest: >=7.4,<9.0 → >=7.4,<10.0 (latest 9.0.2)
- pytest-cov: >=4.1,<7.0 → >=4.1,<8.0 (latest 7.0.0)

requirements-training.txt:
- category-encoders: 2.8.1 → 2.9.0
- statsmodels: 0.14.4 → 0.14.6
- tensorflow: 2.18.0 → 2.21.0
- wordcloud: 1.9.4 → 1.9.6

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>
@openhands-agent
build: migrate package manager from Poetry to uv
ab78524 
- pyproject.toml: rewrite from [tool.poetry] to PEP 621 [project] format
  - [tool.poetry.dependencies] -> [project] dependencies list (PEP 508)
  - [tool.poetry.group.dev.dependencies] -> [dependency-groups] dev (PEP 735)
  - [tool.poetry.plugins.*] -> [project.entry-points.*]
  - build-system: poetry-core -> hatchling
  - [tool.hatch.build.targets.wheel] packages = ['sapientml_core'] added
  - Poetry '^' constraints expanded to explicit >=X,<Y ranges

- uv.lock: generated (192 packages resolved)

- .github/workflows/test.yml:
  - Remove POETRY_VERSION/POETRY_URL env vars
  - Install Poetry + setup-python -> astral-sh/setup-uv@v5
  - poetry install -> uv sync --group dev
  - poetry run -> uv run
  - pip install coverage -> uv tool install coverage

- .github/workflows/release.yml: same test job changes as test.yml, plus
  - release job: poetry build -> uv build
  - Set Version: sed -> uv version $SEMVER
  - Check Version: poetry version --short -> tag regex check directly
  - Publish: POETRY_PYPI_TOKEN_PYPI/poetry publish -> UV_PUBLISH_TOKEN/uv publish

- .github/workflows/lint.yml:
  - setup-python + pip install pysen/flake8/black/isort
    -> astral-sh/setup-uv@v5 + uv sync --group dev
  - pysen run lint -> uv run pysen run lint

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>
@openhands-agent
fix: constrain isort<8.0 and bump [tool.pysen] to 0.12.0
0d73e88 
pysen does not support isort v8.x (supports v4-v7 only).
isort was resolved to 8.0.1 under the previous >=5.12,<9.0 range,
causing the lint workflow to abort at config load time.

- isort: >=5.12,<9.0 -> >=5.12,<8.0 (resolves to 6.1.0)
- [tool.pysen] version: 0.11.0 -> 0.12.0 (match installed pysen)
- uv.lock: regenerated (191 packages)

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>
@openhands-agent
fix: replace removed sklearn _validate_data with public validate_data
d8812fe 
scikit-learn >= 1.6 removed BaseEstimator._validate_data in favour of
the free function sklearn.utils.validation.validate_data(estimator, X).
The force_all_finite kwarg was also renamed to ensure_all_finite.

- Import validate_data from sklearn.utils.validation
- Replace scaler._validate_data(..., estimator=scaler, force_all_finite=...)
  with    validate_data(scaler, ..., ensure_all_finite=...)

Fixes: AttributeError: 'StandardScaler' object has no attribute '_validate_data'

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>
@openhands-agent
fix: remove pandas FutureWarnings and deprecated API usage
1c21ffc 
- Replace df.loc[:, col] = ... with df[col] = ... in real_feature_preprocess
  to avoid FutureWarning under pandas 2.x copy-on-write semantics
- Replace meta_features[col].fillna(0, inplace=True) with assignment form
  in _predict_preprocessors; ensures NaN values are actually filled
  (inplace= is a no-op on a copy under CoW)
- Add .infer_objects(copy=False) after fillna in _predict_models to suppress
  FutureWarning about object-dtype inference
- Replace deprecated pd.api.types.is_categorical_dtype with
  isinstance(column.dtype, pd.CategoricalDtype)

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>
@openhands-agent
fix: re-serialize model pkl files to match current sklearn versions p…
c4b7f8a 
…er Python

Re-pickle all model files (pp_models, mp_model_1, mp_model_2) for each
supported Python version using the sklearn release locked in uv.lock:

  PY39/     → sklearn 1.6.1 (Python 3.9)
  PY310/    → sklearn 1.7.2 (Python 3.10)
  PY311/    → sklearn 1.8.0 (Python 3.11)
  models/   → sklearn 1.8.0 (Python >=3.12 fallback)

This eliminates InconsistentVersionWarning on load, which was causing
incorrect model predictions due to NaN not being propagated through the
copy-on-write path in fillna (fixed separately).

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>
@openhands-agent
fix: pin xgboost<3.0.0 to fix SHAP 0.49.x incompatibility
c5a6b2c 
XGBoost 3.x stores base_score in UBJ bracket-string format (e.g.
'[5.0482047E-1]'), which SHAP 0.49.1's XGBTreeModelLoader cannot parse
via float(), raising ValueError.  The fix in SHAP (0.50+) requires
numpy>=2, but sapientml constrains numpy<2.0.0, making the SHAP upgrade
path blocked.

Tighten the xgboost upper bound to <3.0.0 so uv resolves xgboost 2.1.4
across all Python versions (previously 3.2.0 on Python>=3.10).  SHAP
0.49.1 + XGBoost 2.1.4 is a verified-working combination.

Also simplify shap.py.jinja: replace the Explainer/LGBMClassifier
conditional with a direct shap.TreeExplainer() call, which is correct
for all tree-based models in models_for_shap.

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>
@openhands-agent
docs: add AGENTS.md with repo-specific agent memory
d0d9f88 
Records key dependency constraints (xgboost<3.0.0, shap, numpy),
SHAP/XGBoost compatibility root-cause analysis, common uv commands,
and commit history for build/update-dependencies.

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>
@kimusaku kimusaku force-pushed the build/update-dependencies branch from 89e610d to d0d9f88 Compare March 19, 2026 01:21
@kimusaku kimusaku merged commit efd22d6 into main Mar 19, 2026
31 checks passed
@kimusaku kimusaku deleted the build/update-dependencies branch March 19, 2026 01:30
kimusaku pushed a commit that referenced this pull request Mar 19, 2026
@openhands-agent
fix: use PY311 models for Python 3.12+ to fix monotonic_cst Attribute…
eb49a56 
…Error

predictor.py hardcoded version checks for [9, 10, 11], causing Python 3.12
and 3.13 to fall through to the root-level pkl files which were serialized
with an old sklearn (<1.4) that lacks the monotonic_cst attribute.

Fix: clamp to the newest versioned directory (PY311) for any unknown minor
version. sklearn is pinned at >=1.6.1,<2.0 for all Python versions, so the
PY311 models (re-serialized in the uv migration, #113) are pickle-compatible
with Python 3.12 and 3.13.

Fixes: AttributeError: 'DecisionTreeClassifier' object has no attribute 'monotonic_cst'
kimusaku pushed a commit that referenced this pull request Mar 19, 2026
@openhands-agent
fix: use PY311 models for Python 3.12+ to fix monotonic_cst Attribute…
3628830 
…Error

predictor.py hardcoded version checks for [9, 10, 11], causing Python 3.12
and 3.13 to fall through to the root-level pkl files which were serialized
with an old sklearn (<1.4) that lacks the monotonic_cst attribute.

Fix: clamp to the newest versioned directory (PY311) for any unknown minor
version. sklearn is pinned at >=1.6.1,<2.0 for all Python versions, so the
PY311 models (re-serialized in the uv migration, #113) are pickle-compatible
with Python 3.12 and 3.13.

Fixes: AttributeError: 'DecisionTreeClassifier' object has no attribute 'monotonic_cst'
Signed-off-by: openhands <openhands@all-hands.dev>
kimusaku added a commit that referenced this pull request Mar 20, 2026
@kimusaku @openhands-agent
feat: add Python 3.13 support (#116)
b7aea02 
* feat: add Python 3.13 support

- Bump requires-python from <3.13 to <3.14
- Replace fasttext-wheel with langdetect (pure Python, supports 3.13)
  - fasttext-wheel has no cp313 wheel; langdetect 1.0.9 is pure Python
  - Rewrite check_column_language() to use langdetect.detect()
  - DetectorFactory.seed=0 ensures deterministic detection
- Remove requests dependency (was only used to download fasttext model)
- Add Python 3.12 and 3.13 to CI test matrix
- Regenerate uv.lock

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>

* fix(deps): resolve numpy 2.x for cp313 wheel support on Python 3.13

numpy 1.26.4 has no cp313 wheel, causing source builds that time out in
CI. sapientml (PyPI <=0.4.16) required numpy<2.0.0 due to fasttext-wheel
incompatibility; that dependency has been replaced by langdetect.

Add [tool.uv] override-dependencies to bypass sapientml's numpy<2.0.0
constraint so uv can resolve:
  - numpy 2.0.2  for Python <3.10  (numba 0.60.0 ceiling <2.1)
  - numpy 2.2.6  for Python 3.10   (last numpy supporting 3.10)
  - numpy 2.4.3  for Python >=3.11 (includes cp313 wheels)

CI fixes:
  - Add fail-fast: false so independent matrix jobs aren't canceled
  - Prefix coverage artifact names with Python version (py${{ver}}-*)
    to avoid collisions across the version × test matrix

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>

* docs: update AGENTS.md — Python 3.13 numpy 2.x support and CI fixes

Co-authored-by: openhands <openhands@all-hands.dev>
Signed-off-by: openhands <openhands@all-hands.dev>

* fix: use PY311 models for Python 3.12+ to fix monotonic_cst AttributeError

predictor.py hardcoded version checks for [9, 10, 11], causing Python 3.12
and 3.13 to fall through to the root-level pkl files which were serialized
with an old sklearn (<1.4) that lacks the monotonic_cst attribute.

Fix: clamp to the newest versioned directory (PY311) for any unknown minor
version. sklearn is pinned at >=1.6.1,<2.0 for all Python versions, so the
PY311 models (re-serialized in the uv migration, #113) are pickle-compatible
with Python 3.12 and 3.13.

Fixes: AttributeError: 'DecisionTreeClassifier' object has no attribute 'monotonic_cst'
Signed-off-by: openhands <openhands@all-hands.dev>

---------

Signed-off-by: openhands <openhands@all-hands.dev>
Co-authored-by: openhands <openhands@all-hands.dev>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@AkiraUra AkiraUra Awaiting requested review from AkiraUra AkiraUra is a code owner automatically assigned from sapientml/maintainers

@fukuta-flab fukuta-flab Awaiting requested review from fukuta-flab fukuta-flab is a code owner automatically assigned from sapientml/maintainers

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@kimusaku @openhands-agent