Merge remote-tracking branch 'upstream/main' into enh/estimator-refactor

Author: oesteban

.github/workflows/test.yml

@@ -95,8 +95,7 @@ jobs:
     continue-on-error: true
     strategy:
       matrix:
-        check: ['spellcheck']
-
+        check: ['spellcheck', 'typecheck']
     steps:
       - uses: actions/checkout@v4
       - name: Install the latest version of uv

docs/conf.py

@@ -54,7 +54,6 @@
     "nipype",
     "nitime",
     "nitransforms",
-    "numpy",
     "pandas",
     "scipy",
     "seaborn",
@@ -154,20 +153,20 @@
 
 # -- Options for LaTeX output ------------------------------------------------
 
-latex_elements = {
-    # The paper size ('letterpaper' or 'a4paper').
-    #
-    # 'papersize': 'letterpaper',
-    # The font size ('10pt', '11pt' or '12pt').
-    #
-    # 'pointsize': '10pt',
-    # Additional stuff for the LaTeX preamble.
-    #
-    # 'preamble': '',
-    # Latex figure (float) alignment
-    #
-    # 'figure_align': 'htbp',
-}
+# latex_elements = {
+#     # The paper size ('letterpaper' or 'a4paper').
+#     #
+#     # 'papersize': 'letterpaper',
+#     # The font size ('10pt', '11pt' or '12pt').
+#     #
+#     # 'pointsize': '10pt',
+#     # Additional stuff for the LaTeX preamble.
+#     #
+#     # 'preamble': '',
+#     # Latex figure (float) alignment
+#     #
+#     # 'figure_align': 'htbp',
+# }
 
 # Grouping the document tree into LaTeX files. List of tuples
 # (source start file, target name, title,

pyproject.toml

@@ -75,6 +75,15 @@ test = [
     "pytest-env",
     "pytest-xdist >= 1.28"
 ]
+types = [
+  "pandas-stubs",
+  "types-setuptools",
+  "scipy-stubs",
+  "types-PyYAML",
+  "types-tqdm",
+  "pytest",
+  "microsoft-python-type-stubs @ git+https://github.com/microsoft/python-type-stubs.git",
+]
 
 notebooks = [
     "jupyter",
@@ -138,6 +147,21 @@ version-file = "src/nifreeze/_version.py"
 # Developer tool configurations
 #
 
+[[tool.mypy.overrides]]
+module = [
+  "nipype.*",
+  "nilearn.*",
+  "nireports.*",
+  "nitransforms.*",
+  "seaborn",
+  "dipy.*",
+  "smac.*",
+  "joblib",
+  "h5py",
+  "ConfigSpace",
+]
+ignore_missing_imports = true
+
 [tool.ruff]
 line-length = 99
 target-version = "py310"

scripts/dwi_gp_estimation_error_analysis.py

@@ -49,7 +49,7 @@ def cross_validate(
     cv: int,
     n_repeats: int,
     gpr: DiffusionGPR,
-) -> dict[int, list[tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray]]]:
+) -> np.ndarray:
     """
     Perform the experiment by estimating the dMRI signal using a Gaussian process model.
 
@@ -74,7 +74,14 @@ def cross_validate(
     """
 
     rkf = RepeatedKFold(n_splits=cv, n_repeats=n_repeats)
-    scores = cross_val_score(gpr, X, y, scoring="neg_root_mean_squared_error", cv=rkf)
+    # scikit-learn stubs do not recognize rkf as a BaseCrossValidator
+    scores = cross_val_score(
+        gpr,
+        X,
+        y,
+        scoring="neg_root_mean_squared_error",
+        cv=rkf,  # type: ignore[arg-type]
+    )
     return scores
 
 
@@ -204,10 +211,10 @@ def main() -> None:
 
     if args.kfold:
         # Use Scikit-learn cross validation
-        scores = defaultdict(list, {})
+        scores: dict[str, list] = defaultdict(list, {})
         for n in args.kfold:
             for i in range(args.repeats):
-                cv_scores = -1.0 * cross_validate(X, y.T, n, gpr)
+                cv_scores = -1.0 * cross_validate(X, y.T, n, i, gpr)
                 scores["rmse"] += cv_scores.tolist()
                 scores["repeat"] += [i] * len(cv_scores)
                 scores["n_folds"] += [n] * len(cv_scores)
@@ -217,7 +224,7 @@ def main() -> None:
             print(f"Finished {n}-fold cross-validation")
 
         scores_df = pd.DataFrame(scores)
-        scores_df.to_csv(args.output_scores, sep="\t", index=None, na_rep="n/a")
+        scores_df.to_csv(args.output_scores, sep="\t", index=False, na_rep="n/a")
 
         grouped = scores_df.groupby(["n_folds"])
         print(grouped[["rmse"]].mean())

scripts/dwi_gp_estimation_error_analysis_plot.py

@@ -89,10 +89,18 @@ def main() -> None:
     df = pd.read_csv(args.error_data_fname, sep="\t", keep_default_na=False, na_values="n/a")
 
     # Plot the prediction error
-    kfolds = sorted(np.unique(df["n_folds"].values))
-    snr = np.unique(df["snr"].values).item()
-    bval = np.unique(df["bval"].values).item()
-    rmse_data = [df.groupby("n_folds").get_group(k)["rmse"].values for k in kfolds]
+    kfolds = sorted(pd.unique(df["n_folds"]))
+    snr = pd.unique(df["snr"])
+    if len(snr) == 1:
+        snr = snr[0]
+    else:
+        raise ValueError(f"More than one unique SNR value: {snr}")
+    bval = pd.unique(df["bval"])
+    if len(bval) == 1:
+        bval = bval[0]
+    else:
+        raise ValueError(f"More than one unique bval value: {bval}")
+    rmse_data = np.asarray([df.groupby("n_folds").get_group(k)["rmse"].values for k in kfolds])
     axis = 1
     mean = np.mean(rmse_data, axis=axis)
     std_dev = np.std(rmse_data, axis=axis)

scripts/dwi_gp_estimation_simulated_signal.py

@@ -132,11 +132,11 @@ def main() -> None:
 
     # Fit the Gaussian Process regressor and predict on an arbitrary number of
     # directions
-    a = 1.15
-    lambda_s = 120
+    beta_a = 1.15
+    beta_l = 120
     alpha = 100
     gpr = DiffusionGPR(
-        kernel=SphericalKriging(a=a, lambda_s=lambda_s),
+        kernel=SphericalKriging(beta_a=beta_a, beta_l=beta_l),
         alpha=alpha,
         optimizer=None,
     )
@@ -154,6 +154,8 @@ def main() -> None:
     X_test = np.vstack([gtab[~gtab.b0s_mask].bvecs, sph.vertices])
 
     predictions = gpr_fit.predict(X_test)
+    if isinstance(predictions, tuple):
+        predictions = predictions[0]
 
     # Save the predicted data
     testsims.serialize_dwi(predictions.T, args.dwi_pred_data_fname)

scripts/optimize_registration.py

@@ -127,12 +127,13 @@ async def train_coro(
             moving_path = tmp_folder / f"test-{index:04d}.nii.gz"
             (~xfm).apply(refnii, reference=refnii).to_filename(moving_path)
 
+            _kwargs = {"output_transform_prefix": f"conversion-{index:04d}", **align_kwargs}
+
             cmdline = erants.generate_command(
                 fixed_path,
                 moving_path,
                 fixedmask_path=brainmask_path,
-                output_transform_prefix=f"conversion-{index:04d}",
-                **align_kwargs,
+                **_kwargs,
             ).cmdline
 
             tasks.append(

src/nifreeze/cli/parser.py

@@ -29,13 +29,13 @@
 import yaml
 
 
-def _parse_yaml_config(file_path: Path) -> dict:
+def _parse_yaml_config(file_path: str) -> dict:
     """
     Parse YAML configuration file.
 
     Parameters
     ----------
-    file_path : Path
+    file_path : str
         Path to the YAML configuration file.
 
     Returns

src/nifreeze/data/base.py

@@ -27,17 +27,23 @@
 from collections import namedtuple
 from pathlib import Path
 from tempfile import mkdtemp
-from typing import Any
+from typing import Any, Generic, TypeVarTuple
 
 import attr
 import h5py
 import nibabel as nb
 import numpy as np
+from nibabel.spatialimages import SpatialHeader, SpatialImage
 from nitransforms.linear import Affine
 
+from nifreeze.utils.ndimage import load_api
+
 NFDH5_EXT = ".h5"
 
 
+Ts = TypeVarTuple("Ts")
+
+
 def _data_repr(value: np.ndarray | None) -> str:
     if value is None:
         return "None"
@@ -52,7 +58,7 @@ def _cmp(lh: Any, rh: Any) -> bool:
 
 
 @attr.s(slots=True)
-class BaseDataset:
+class BaseDataset(Generic[*Ts]):
     """
     Base dataset representation structure.
 
@@ -68,15 +74,15 @@ class BaseDataset:
 
     """
 
-    dataobj = attr.ib(default=None, repr=_data_repr, eq=attr.cmp_using(eq=_cmp))
+    dataobj: np.ndarray = attr.ib(default=None, repr=_data_repr, eq=attr.cmp_using(eq=_cmp))
     """A :obj:`~numpy.ndarray` object for the data array."""
-    affine = attr.ib(default=None, repr=_data_repr, eq=attr.cmp_using(eq=_cmp))
+    affine: np.ndarray = attr.ib(default=None, repr=_data_repr, eq=attr.cmp_using(eq=_cmp))
     """Best affine for RAS-to-voxel conversion of coordinates (NIfTI header)."""
-    brainmask = attr.ib(default=None, repr=_data_repr, eq=attr.cmp_using(eq=_cmp))
+    brainmask: np.ndarray = attr.ib(default=None, repr=_data_repr, eq=attr.cmp_using(eq=_cmp))
     """A boolean ndarray object containing a corresponding brainmask."""
-    motion_affines = attr.ib(default=None, eq=attr.cmp_using(eq=_cmp))
+    motion_affines: np.ndarray = attr.ib(default=None, eq=attr.cmp_using(eq=_cmp))
     """List of :obj:`~nitransforms.linear.Affine` realigning the dataset."""
-    datahdr = attr.ib(default=None)
+    datahdr: SpatialHeader = attr.ib(default=None)
     """A :obj:`~nibabel.spatialimages.SpatialHeader` header corresponding to the data."""
 
     _filepath = attr.ib(
@@ -93,9 +99,13 @@ def __len__(self) -> int:
 
         return self.dataobj.shape[-1]
 
+    def _getextra(self, idx: int | slice | tuple | np.ndarray) -> tuple[*Ts]:
+        # PY312: Default values for TypeVarTuples are not yet supported
+        return ()  # type: ignore[return-value]
+
     def __getitem__(
         self, idx: int | slice | tuple | np.ndarray
-    ) -> tuple[np.ndarray, np.ndarray | None]:
+    ) -> tuple[np.ndarray, np.ndarray | None, *Ts]:
         """
         Returns volume(s) and corresponding affine(s) through fancy indexing.
 
@@ -118,7 +128,7 @@ def __getitem__(
             raise ValueError("No data available (dataobj is None).")
 
         affine = self.motion_affines[idx] if self.motion_affines is not None else None
-        return self.dataobj[..., idx], affine
+        return self.dataobj[..., idx], affine, *self._getextra(idx)
 
     @classmethod
     def from_filename(cls, filename: Path | str) -> BaseDataset:
@@ -159,9 +169,8 @@ def set_transform(self, index: int, affine: np.ndarray, order: int = 3) -> None:
             The order of the spline interpolation.
 
         """
-        reference = namedtuple("ImageGrid", ("shape", "affine"))(
-            shape=self.dataobj.shape[:3], affine=self.affine
-        )
+        ImageGrid = namedtuple("ImageGrid", ("shape", "affine"))
+        reference = ImageGrid(shape=self.dataobj.shape[:3], affine=self.affine)
 
         xform = Affine(matrix=affine, reference=reference)
 
@@ -227,7 +236,7 @@ def to_filename(
                         compression_opts=compression_opts,
                     )
 
-    def to_nifti(self, filename: Path) -> None:
+    def to_nifti(self, filename: Path | str) -> None:
         """
         Write a NIfTI file to disk.
 
@@ -247,7 +256,7 @@ def load(
     filename: Path | str,
     brainmask_file: Path | str | None = None,
     motion_file: Path | str | None = None,
-) -> BaseDataset:
+) -> BaseDataset[()]:
     """
     Load 4D data from a filename or an HDF5 file.
 
@@ -279,11 +288,11 @@ def load(
     if filename.name.endswith(NFDH5_EXT):
         return BaseDataset.from_filename(filename)
 
-    img = nb.load(filename)
-    retval = BaseDataset(dataobj=img.dataobj, affine=img.affine)
+    img = load_api(filename, SpatialImage)
+    retval: BaseDataset[()] = BaseDataset(dataobj=np.asanyarray(img.dataobj), affine=img.affine)
 
     if brainmask_file:
-        mask = nb.load(brainmask_file)
+        mask = load_api(brainmask_file, SpatialImage)
         retval.brainmask = np.asanyarray(mask.dataobj)
 
     return retval