codeformat fixes

Signed-off-by: Wenqi Li <wenqil@nvidia.com>

Author: wyli

monai/transforms/spatial/array.py

@@ -462,7 +462,7 @@ def __init__(
         self.padding_mode: GridSamplePadMode = look_up_option(padding_mode, GridSamplePadMode)
         self.align_corners = align_corners
         self.dtype = dtype
-        self._rotation_matrix: Optional[np.ndarray] = None
+        self._rotation_matrix: Optional[NdarrayOrTensor] = None
 
     def __call__(
         self,
@@ -511,7 +511,7 @@ def __call__(
             corners = np.asarray(np.meshgrid(*[(0, dim) for dim in im_shape], indexing="ij")).reshape(
                 (len(im_shape), -1)
             )
-            corners = transform[:-1, :-1] @ corners
+            corners = transform[:-1, :-1] @ corners  # type: ignore
             output_shape = np.asarray(corners.ptp(axis=1) + 0.5, dtype=int)
         shift_1 = create_translate(input_ndim, (-(output_shape - 1) / 2).tolist())
         transform = shift @ transform @ shift_1
@@ -532,7 +532,7 @@ def __call__(
         out, *_ = convert_to_dst_type(output, dst=img, dtype=output.dtype)
         return out
 
-    def get_rotation_matrix(self) -> Optional[np.ndarray]:
+    def get_rotation_matrix(self) -> Optional[NdarrayOrTensor]:
         """
         Get the most recently applied rotation matrix
         This is not thread-safe.

monai/transforms/utils.py

@@ -602,8 +602,8 @@ def create_rotate(
         return _create_rotate(
             spatial_dims=spatial_dims,
             radians=radians,
-            sin_func=lambda th: torch.sin(torch.as_tensor(th)),
-            cos_func=lambda th: torch.cos(torch.as_tensor(th)),
+            sin_func=lambda th: torch.sin(torch.as_tensor(th, dtype=torch.float32)),
+            cos_func=lambda th: torch.cos(torch.as_tensor(th, dtype=torch.float32)),
             array_func=torch.as_tensor,
         )
     raise ValueError("backend {} is not supported".format(backend))
@@ -676,7 +676,6 @@ def create_shear(
 
     """
     if look_up_option(backend, TransformBackends) == TransformBackends.NUMPY:
-        array_func = np.array
         return _create_shear(spatial_dims=spatial_dims, coefs=coefs, array_func=np.array)
     if look_up_option(backend, TransformBackends) == TransformBackends.TORCH:
         return _create_shear(spatial_dims=spatial_dims, coefs=coefs, array_func=torch.as_tensor)
@@ -759,7 +758,7 @@ def _create_translate(
     affine = eye_func(spatial_dims + 1)
     for i, a in enumerate(shift[:spatial_dims]):
         affine[i, spatial_dims] = a
-    return array_func(affine)
+    return array_func(affine)  # type: ignore
 
 
 def generate_spatial_bounding_box(