pyproject.toml

[tool.poetry]
name = "verpex"
# Placeholder: the real version is derived from the latest git tag at build time by
# poetry-dynamic-versioning (see [tool.poetry-dynamic-versioning] below).
version = "0.0.0"
description = "Deep-learning prediction of anatomical points-of-interest (POIs) on vertebrae from CT/MRI spine segmentations."
authors = ["Hendrik Möller <hendrik.moeller@tum.de>"]
license = "Apache-2.0"
readme = "README.md"
packages = [{ include = "verpex" }]
include = [{ path = "verpex/configs/*.json" }, { path = "verpex/py.typed" }]

[tool.poetry.dependencies]
python = "^3.10 || ^3.11 || ^3.12"
TPTBox = ">=0.8"
TypeSaveArgParse = "*"
torch = ">=2.1"
pytorch-lightning = ">=2.1"
monai = ">=1.3"
numpy = ">=1.26"
pandas = ">=2.0"
scipy = ">=1.11"
scikit-image = ">=0.21"
scikit-learn = ">=1.4"
nibabel = ">=5.1"
PyYAML = ">=6.0"
tqdm = "*"
pqdm = "*"
openpyxl = "*"
einops = "*"
tensorboard = "*"

[tool.poetry.group.dev.dependencies]
pytest = ">=8.0"
poetry-dynamic-versioning = ">=1.0.0,<2.0.0"
ruff = ">=0.11"
pre-commit = "*"
coverage = ">=7.2"
mypy = ">=1.8"
pandas-stubs = "*"
types-PyYAML = "*"

[tool.poetry.group.extras.dependencies]
# Sparse-convolution backbones (SubmDenseNet). CUDA-specific; not required for
# the dense pipeline, so kept out of the default install.
spconv-cu120 = "*"

[tool.poetry.scripts]
verpex-prepare-data = "verpex.cli.prepare_data:main"
verpex-train = "verpex.cli.train:main"
verpex-train-cv = "verpex.cli.train_cv:main"
verpex-eval = "verpex.evaluation.metrics:main"
verpex-infer = "verpex.cli.infer:main"

[build-system]
requires = ["poetry-core>=1.0.0", "poetry-dynamic-versioning>=1.0.0,<2.0.0"]
build-backend = "poetry_dynamic_versioning.backend"

[tool.poetry-dynamic-versioning]
enable = true

[tool.ruff]
exclude = [".git", ".ruff_cache", ".venv", "venv", "build", "dist", "__pypackages__", "*.ipynb"]
line-length = 140
indent-width = 4
target-version = "py310"

[tool.ruff.lint]
select = [
    "E", "F", "W", "I", "N", "UP", "B", "A", "C4", "ICN", "G", "PIE", "PYI",
    "SIM", "TID", "ARG", "D", "ANN201", "PD", "PGH", "PL", "TRY", "FLY", "NPY",
    "PERF", "FURB", "RUF",
]
ignore = [
    "E501",    # line length handled by the formatter
    "BLE001",
    "N801", "N802", "N803", "N806", "N999",
    "NPY002",
    "PLR0911", "PLR0912", "PLR0913", "PLR0915", "PLR2004",
    "TRY003", "TRY301",
    "SIM105", "SIM118",
    "UP007",
    "PLC0415",
    # Rebinding a loop variable is this codebase's idiom for staged transforms
    # (batch = to_device(batch); batch = model(batch)); renaming each stage hurts
    # readability more than the shadowing costs.
    "PLW2901",
    "PLW0108",
    "B905",
    "D100",    # module docstring - covered per-package
    "D104",    # package docstring
    "D105",    # magic method
    "D107",    # __init__ (covered by class docstring)
    "D417",
    "ANN202",
    "RUF002",
]
fixable = ["ALL"]
unfixable = []
dummy-variable-rgx = "^(_+|(_+[a-zA-Z0-9_]*[a-zA-Z0-9]+?))$"

[tool.ruff.lint.pydocstyle]
convention = "google"

[tool.ruff.lint.per-file-ignores]
# Re-exports in package __init__ files are intentional, not dead imports.
"**/__init__.py" = ["F401"]
"unit_tests/**" = ["D101", "D102", "D103", "D205", "D415", "ANN201", "ARG001"]
# Largely a vendored copy of MONAI's affine transforms, kept close to upstream so it
# stays diffable against it. The project's own transforms are at the top of the file
# and are documented; MONAI's docstring and typing style is left as-is.
"verpex/data/transforms.py" = [
    "D101", "D102", "D103", "D205", "D415", "ANN201", "ARG002", "PGH003", "RUF012",
    "N803", "N806", "UP035", "UP045", "RUF005", "B028", "C408",
]

[tool.ruff.lint.mccabe]
max-complexity = 20

[tool.pytest.ini_options]
testpaths = ["unit_tests"]

[tool.coverage.run]
source = ["verpex"]
branch = true
# Vendored MONAI transforms - exempted here the same way they are for ruff and mypy.
omit = ["verpex/data/transforms.py"]

[tool.coverage.report]
exclude_also = [
    "if TYPE_CHECKING:",
    "raise NotImplementedError",
    "if __name__ == .__main__.:",
    "@(abc\\.)?abstractmethod",
]

[tool.mypy]
python_version = "3.10"
files = ["verpex"]
warn_redundant_casts = true
warn_unused_ignores = true
no_implicit_optional = true
show_error_codes = true

[[tool.mypy.overrides]]
# Third-party packages that ship no stubs and no py.typed marker.
module = [
    "TPTBox.*",
    "TypeSaveArgParse.*",
    "sklearn.*",
    "spconv.*",
    "monai.*",
    "nibabel.*",
    "skimage.*",
    "scipy.*",
    "pqdm.*",
]
ignore_missing_imports = true

[[tool.mypy.overrides]]
# Largely a vendored copy of MONAI's affine transforms, kept close to upstream so
# it stays diffable against it - exempted here the same way it is for ruff.
module = "verpex.data.transforms"
ignore_errors = true