* ci: try Python 3.15 release candidate (DRAFT)
Point the repo's .python-version at 3.15 so every workflow that reads
python-version-file runs on the 3.15 release candidate (allow-prereleases
is already enabled across all workflows). Add the 3.15 trove classifier
and bump scipy>=1.18.1, which is the first scipy with cp315 manylinux
wheels, so uv installs it as a wheel rather than building from source.
Rebuilt on current master: the earlier version of this PR edited the
per-workflow `python-version: 3.14` lines, but master has since moved
the interpreter to a single `.python-version` file, so the change is now
a one-line switch there.
Purpose is to document which dependencies/tests are not yet 3.15-ready.
Must stay DRAFT until Python 3.15 GA (early October). Requested in #15081.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Upgrade to Python 3.15t
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Christian Clauss <cclauss@me.com>
Per discussion in #15418, these four files are not algorithms (they are
how-to-use scripts wrapping a deep-learning framework) and dragged in the
heavy keras/tensorflow dependency stack:
- computer_vision/cnn_classification.py
- dynamic_programming/k_means_clustering_tensorflow.py
- machine_learning/lstm/lstm_prediction.py
- neural_network/input_data.py (TF MNIST data loader; nothing imports it)
Also removes the now-orphaned machine_learning/lstm/ package (only
__init__.py + sample_data.csv, which served lstm_prediction.py).
Cleanups:
- Drop keras from pyproject.toml dependencies; regenerate uv.lock
(removes absl-py, h5py, keras, ml-dtypes, namex, optree).
- Remove the pre-release libhdf5-dev install step from build.yml and
sphinx.yml (it existed only because keras needs hdf5).
- Drop the four stale pytest --ignore entries in build.yml.
- Remove the four DIRECTORY.md entries and the empty Lstm heading.
Polars only ships cp310-abi3 wheels, which are not loadable under the
free-threaded CPython the repo targets (.python-version = 3.14t). uv
therefore builds the ~3 MB Rust sdist from source on every uv.lock
change, making CI builds very slow. Upstream (pola-rs/polars#27955)
is not yet ready to publish free-threaded wheels.
Remove polars from pyproject.toml and relock (uv sync --upgrade && uv lock).
No source files import polars, so nothing else changes.
Drop machine_learning/xgboost_classifier.py and
machine_learning/xgboost_regressor.py. Both were thin "how-to-use"
wrappers around sklearn's XGBClassifier/XGBRegressor rather than
from-scratch implementations, and the classifier's only doctest was
already disabled (# THIS TEST IS BROKEN!!), so it was never exercised
in CI.
xgboost is one of the heaviest compiled dependencies in the tree (large
wheel, needs OpenMP/libgomp at runtime, no free-threaded wheel yet), and
gradient boosting is already implemented from scratch in
machine_learning/gradient_boosting_classifier.py and
gradient_boosting_regressor.py, so no algorithm coverage is lost.
Removes the xgboost dependency from pyproject.toml, its (and its
xgboost-only transitive dep nvidia-nccl-cu13) entries from uv.lock, and
the two DIRECTORY.md links.
Refs #15081
* Use Astral uv
* uvx vs uv run
* uv sync --group=euler-validate,test
* uv sync --group=euler-validate --group=test
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* --group=test
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
