pyproject.toml

[build-system]
requires = ["setuptools"]
build-backend = "setuptools.build_meta"

[project]
name = "vanguard-gp"
dynamic = ["version"]
description = "Various easy-to-use extensions for Gaussian process models and a framework for composition of extensions."
readme = "README.md"
requires-python = ">=3.9"
keywords = [
    "gaussian-process",
    "pytorch",
]
classifiers = [
    "Development Status :: 4 - Beta",
    "License :: OSI Approved :: GNU General Public License v3 (GPLv3)",
    "Programming Language :: Python",
    "Programming Language :: Python :: 3.9",
    "Programming Language :: Python :: 3.10",
    "Programming Language :: Python :: 3.11",
    "Programming Language :: Python :: 3.12",
    "Programming Language :: Python :: 3.13",
    "Typing :: Typed",
]
license = {file = "LICENSE.md"}
authors = [
    {name = "GCHQ", email = "oss@gchq.gov.uk"},
]
dependencies = [
    "bibtexparser>=1.1",  # 1.0 fails on import (?)
    "gpytorch>=1.11",  # LinearOperator support added in 1.9;
                       #   our SoftmaxLMC performance is improved in 1.9.1 (a unit test fails due to bad predictions
                       #   on 1.9.0);
                       #   one of our multitask tests fails on 1.10.
    "kmedoids>=0.3.3",  # first version with prebuilt wheels; earlier versions require a rust toolchain to build
    "kmedoids>=0.4; python_version>='3.11'",  # earlier versions don't have prebuilt wheels for python 3.11
    "kmedoids>=0.4.4; python_version>='3.12'",  # earlier versions don't have prebuilt wheels for python 3.12
    "kmedoids>=0.5.3; python_version>='3.13'",  # earlier versions don't have prebuilt wheels for python 3.13
    "matplotlib>=3.8.4",  # first version to support numpy 2
    "numpy>=2",  # modern versions of packages are incompatible with numpy<2
    "numpy>=2.1.0; python_version>='3.13'",  # first version to support python 3.13
    "pandas>=2",
    "pandas>=2.2.3; python_version>='3.13'",  # earlier versions don't provide prebuilt wheels for python 3.13
    "scikit-learn>=1.5.0",  # CVE-2024-5206 affects versions earlier than 1.5.0
    "scikit-learn>=1.6.0; python_version>='3.13'",  # earlier versions don't provide prebuilt wheels for python 3.13
    "scipy>=1.13",  # first version to support numpy 2
    "scipy>=1.14.1; python_version>='3.13'",  # earlier versions don't provide prebuilt wheels for python 3.13
    "torch>=2.6",  # earlier versions have issues in CI in lower-bound tests
    "typing_extensions>=4.12",  # ParamSpec behaviour improved in 4.12
    "urllib3>=2.0.4",  # BaseHttpResponse added to urllib3.__all__ in 2.0.4
]

[project.optional-dependencies]
# These match the suffixes on the various requirements-*.txt files
# Run tests
test = [
    "ipykernel>=6",  # for notebook tests
    "nbconvert>=7",  # for notebook tests
    "nbformat>=5",  # for notebook tests
    "tqdm>=4",  # for notebook tests
    "pytest-beartype>=0.1",
    "pytest-cov>=5",
    "pytest>=8.3.3",
    "beartype>=0.19",
]
# Compile documentation
doc = [
    "furo>=2024",
    "ipykernel>=6",
    "ipython>=8",
    "nbconvert>=7",
    "nbconvert>=7.1; python_version>='3.11'",
    "nbformat>=5",
    "nbsphinx>=0.9",
    "pillow>=10",
    "pybtex>=0.24",
    "setuptools>=75; python_version>='3.12'",  # TODO: Remove once pybtex releases > 0.24.0
    # https://github.com/gchq/Vanguard/issues/246
    "sphinx>=8, !=8.2.*; python_version>='3.10'",
    "sphinx>=7, !=8.2.*; python_version<'3.10'",
    "sphinx-autodoc-typehints>=2",
    "sphinxcontrib-bibtex>=2",
    "tqdm>=4",
]
# Run notebooks in the jupyter interface
notebook = [
    "jupyterlab>=4",
    "tqdm>=4",
]

[tool.uv]
dev-dependencies = [
    "vanguard-gp[test, doc, notebook]",
    "isort>=5",
    "jupyterlab>=4",
    "pre-commit>=3",
    "pylint>=3",
    "pyright>=1",
    "pyroma>=4",
    "ruff>=0.6",
]

# Constraints on dependencies. These do *not* select a package to be installed, but if *another* package requires
# one of these packages, then uv will respect these constraints. Note that pip does *not* respect these constraints,
# however - they are only for ensuring that `--resolution lowest` only picks packages compatible with modern Python.
# Only include *transitive* dependencies in this list - to apply a constraint on a direct dependency, do so at the
# actual dependency specification!
constraint-dependencies = [
    "networkx>=2",
    "sympy>=1.5",
    "appnope>=0.1",
    "wcwidth>=0.2",
    "cmdln>=2",
    "distribute>=1",
    "python-json-logger>=1",
    "py>=1",
    "fsspec>=2024",
    "six>=1.16; python_version>='3.10'",  # earlier versions don't work on 3.10+
    "pyzmq>=26",
    "contourpy>=1.1; python_version>='3.11'",
    "lxml>=4.7.1",  # earlier versions fail to build to missing libxml2
    "argon2-cffi>=19", # earlier versions have issues building on Linux
    "bleach>=1",  # some pre-1.0 versions fail to build
    "webencodings>=0.5",  # older versions fail to build
    "kiwisolver>=1.3.0",
    "defusedxml>=0.5",  # older versions are incompatible with modern Python versions
    "pywin32>=302; python_version>='3.10'",  # first version to support 3.10
    "pywin32>=303; python_version>='3.11'",  # first version to support 3.11
    "pywin32>=306; python_version>='3.12'",  # first version to support 3.12
    "pywin32>=307; python_version>='3.13'",  # first version to support 3.13
    "wheel>=0.30",  # picking a too-low version of this was causing CI failures that couldn't be replicated locally,
                    # so this is a guessed lower bound
    "cffi>=1.17; python_version>='3.13'",
    "pywinpty>=2.0.14; python_version>='3.13'",
    "pillow>=11; python_version>='3.13'",
    "html5lib>=1",
]

# force the lockfile to be split by Python version, ensuring we get the latest dependencies for each version
environments = [
    "python_version>='3.13'",
    "python_version=='3.12.*'",
    "python_version=='3.11.*'",
    "python_version=='3.10.*'",
    "python_version=='3.9.*'",
]

[project.urls]
Documentation = "https://vanguard.readthedocs.io/en/latest/"
Repository = "https://github.com/gchq/Vanguard"
Issues = "https://github.com/gchq/Vanguard/issues"
Changelog = "https://github.com/gchq/Vanguard/blob/main/CHANGELOG.md"

[tool.setuptools]
packages = [
    "vanguard",
    "vanguard.base",
    "vanguard.base.posteriors",
    "vanguard.classification",
    "vanguard.datasets",
    "vanguard.datasets.data",
    "vanguard.decoratorutils",
    "vanguard.distribute",
    "vanguard.hierarchical",
    "vanguard.multitask",
    "vanguard.optimise",
    "vanguard.variational",
    "vanguard.warps",
]

[tool.setuptools.dynamic]
version = {attr = "vanguard.__version__"}

[tool.setuptools.package-data]
vanguard = ["references.bib"]
"vanguard.datasets.data" = ["*.csv", "*.rst"]

[tool.ruff]
line-length = 120
extend-include = ["*.ipynb"]  # Handle Jupyter Notebook files as well.

[tool.ruff.lint]
extend-select = [
    "D102",  # Public methods must have docstrings
    "D201",  # No blank lines allowed before function docstring
    "D202",  # No blank lines allowed after function docstring
    "D211",  # No blank lines allowed before class docstring
    "E501",  # Docstring line lengths
    # TODO: We should enable more of the pydocstyle lints!
    # https://github.com/gchq/Vanguard/issues/256
    "S102",  # use of `exec` builtin
    "S307",  # Suspicious `eval` usage
    "I",     # isort rules
    "UP",    # pyupgrade rules
    "NPY",   # numpy linting rules
    "TID252",# ban relative imports
]

[tool.ruff.lint.per-file-ignores]
"tests/**.py" = [ "TID252" ]  # allow relative imports within tests

[tool.ruff.lint.isort]
known-local-folder = ["confutils","refstyle"]

[tool.ruff.lint.flake8-tidy-imports]
ban-relative-imports = "all"

[tool.coverage.run]
command_line = "-m pytest tests/units --cov"
branch = true
relative_files = true
source = ["vanguard"]

[tool.coverage.report]
show_missing = true
sort = "Cover"
exclude_also = [
    "pass",
    "raise NotImplementedError",
    "if TYPE_CHECKING:",
    "raise$",
]

[tool.pytest.ini_options]
markers = [
    "no_beartype: for tests incompatible with beartype (e.g. checking for TypeErrors)",
]

# TODO: fix as many of these as possible, and for those we can't fix, suppress as many as possible at the point that
#  they're emitted using the catch_warnings context manager. Suppress globally here only as a last resort.
# https://github.com/gchq/Vanguard/issues/281
filterwarnings = [
    # Numerical error warnings
    "ignore::linear_operator.utils.warnings.NumericalWarning",
    "ignore:Hit a numerical error:UserWarning",
    "ignore:A regression problem with no warping",

    # Other warnings
    "ignore::pytest.PytestCollectionWarning",
    "ignore::vanguard.warnings.ExperimentalFeatureWarning",  # we already know what parts of the code are experimental!
    "ignore:The input matches the stored training data. Did you forget to call model.train\\(\\)\\?:gpytorch.utils.warnings.GPInputWarning",
    "ignore:To copy construct from a tensor, it is recommended to use sourceTensor.clone\\(\\).detach\\(\\):UserWarning",
    "ignore:Pass auto_restart=True to the controller to automatically restart training up to the last stable iterations:UserWarning",
    "ignore:std\\(\\). degrees of freedom is <= 0:UserWarning",
    "ignore:You have passed data through a FixedNoiseGaussianLikelihood that did not match the size of the fixed noise:gpytorch.utils.warnings.GPInputWarning",
    "ignore:To get the last learning rate computed by the scheduler, please use `get_last_lr\\(\\)`:UserWarning",

    # -- Deprecations --
    # TODO: replace with sparse_coo_tensor
    # https://github.com/gchq/Vanguard/issues/278
    "ignore:torch.sparse.SparseTensor\\(indices, values, shape, \\*, device=\\) is deprecated:UserWarning",
    # TODO: replace with some alternative (e.g. beartype.typing.*)
    "ignore::beartype.roar.BeartypeDecorHintPep585DeprecationWarning"
]