ENH: Collapse linear and nonlinear transforms chains

Very undertested, but currently there is a test that uses a "collapsed"
transform on an ITK's .h5 file with one affine and one nonlinear.

BSpline transforms not currently supported.

Resolves #89.

nitransforms/linear.py

@@ -123,19 +123,17 @@ def __matmul__(self, b):
         True
 
         >>> xfm1 = Affine([[1, 0, 0, 4], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]])
-        >>> xfm1 @ np.eye(4) == xfm1
+        >>> xfm1 @ Affine() == xfm1
         True
 
         """
-        if not isinstance(b, self.__class__):
-            _b = self.__class__(b)
-        else:
-            _b = b
+        if isinstance(b, self.__class__):
+            return self.__class__(
+                b.matrix @ self.matrix,
+                reference=b.reference,
+            )
 
-        retval = self.__class__(self.matrix.dot(_b.matrix))
-        if _b.reference:
-            retval.reference = _b.reference
-        return retval
+        return b @ self
 
     @property
     def matrix(self):

nitransforms/manip.py

@@ -140,17 +140,17 @@ def map(self, x, inverse=False):
 
         return x
 
-    def asaffine(self, indices=None):
+    def collapse(self):
         """
-        Combine a succession of linear transforms into one.
+        Combine a succession of transforms into one.
 
         Example
         ------
         >>> chain = TransformChain(transforms=[
         ...     Affine.from_matvec(vec=(2, -10, 3)),
         ...     Affine.from_matvec(vec=(-2, 10, -3)),
         ... ])
-        >>> chain.asaffine()
+        >>> chain.collapse()
         array([[1., 0., 0., 0.],
                [0., 1., 0., 0.],
                [0., 0., 1., 0.],
@@ -160,15 +160,15 @@ def asaffine(self, indices=None):
         ...     Affine.from_matvec(vec=(1, 2, 3)),
         ...     Affine.from_matvec(mat=[[0, 1, 0], [0, 0, 1], [1, 0, 0]]),
         ... ])
-        >>> chain.asaffine()
+        >>> chain.collapse()
         array([[0., 1., 0., 2.],
                [0., 0., 1., 3.],
                [1., 0., 0., 1.],
                [0., 0., 0., 1.]])
 
         >>> np.allclose(
         ...     chain.map((4, -2, 1)),
-        ...     chain.asaffine().map((4, -2, 1)),
+        ...     chain.collapse().map((4, -2, 1)),
         ... )
         True
 
@@ -178,9 +178,8 @@ def asaffine(self, indices=None):
             The indices of the values to extract.
 
         """
-        affines = self.transforms if indices is None else np.take(self.transforms, indices)
-        retval = affines[0]
-        for xfm in affines[1:]:
+        retval = self.transforms[-1]
+        for xfm in reversed(self.transforms[:-1]):
             retval = xfm @ retval
         return retval
 

nitransforms/tests/test_linear.py

@@ -328,10 +328,10 @@ def test_mulmat_operator(testdata_path):
     mat2 = from_matvec(np.eye(3), (4, 2, -1))
     aff = nitl.Affine(mat1, reference=ref)
 
-    composed = aff @ mat2
+    composed = aff @ nitl.Affine(mat2)
     assert composed.reference is None
-    assert composed == nitl.Affine(mat1.dot(mat2))
+    assert composed == nitl.Affine(mat2 @ mat1)
 
     composed = nitl.Affine(mat2) @ aff
     assert composed.reference == aff.reference
-    assert composed == nitl.Affine(mat2.dot(mat1), reference=ref)
+    assert composed == nitl.Affine(mat1 @ mat2, reference=ref)

nitransforms/tests/test_manip.py

@@ -60,6 +60,12 @@ def test_itk_h5(tmp_path, testdata_path):
     # A certain tolerance is necessary because of resampling at borders
     assert (np.abs(diff) > 1e-3).sum() / diff.size < RMSE_TOL
 
+    col_moved = xfm.collapse().apply(img_fname, order=0)
+    col_moved.to_filename("nt_collapse_resampled.nii.gz")
+    diff = sw_moved.get_fdata() - col_moved.get_fdata()
+    # A certain tolerance is necessary because of resampling at borders
+    assert (np.abs(diff) > 1e-3).sum() / diff.size < RMSE_TOL
+
 
 @pytest.mark.parametrize("ext0", ["lta", "tfm"])
 @pytest.mark.parametrize("ext1", ["lta", "tfm"])
@@ -81,7 +87,7 @@ def test_collapse_affines(tmp_path, data_path, ext0, ext1, ext2):
         ]
     )
     assert np.allclose(
-        chain.asaffine().matrix,
+        chain.collapse().matrix,
         Affine.from_filename(
             data_path / "regressions" / f"from-fsnative_to-bold_mode-image.{ext2}",
             fmt=f"{FMT[ext2]}",