Lazy eval refactor

docs/source/transforms.rst

@@ -958,6 +958,17 @@ MRI Transforms
     :special-members: __call__
 
 
+Lazy
+^^^^
+
+`ApplyPending`
+""""""""""""""
+
+.. autoclass:: ApplyPending
+    :members:
+    :special-members: __call__
+
+
 Utility
 ^^^^^^^
 
@@ -1912,6 +1923,17 @@ Smooth Field (Dict)
     :special-members: __call__
 
 
+Lazy (Dict)
+^^^^^^^^^^^
+
+`ApplyPendingd`
+"""""""""""""""
+
+.. autoclass:: ApplyPendingd
+    :members:
+    :special-members: __call__
+
+
 Utility (Dict)
 ^^^^^^^^^^^^^^
 
@@ -2211,9 +2233,3 @@ Utilities
 
 .. automodule:: monai.transforms.utils_pytorch_numpy_unification
     :members:
-
-Lazy
-----
-.. automodule:: monai.transforms.lazy
-    :members:
-    :imported-members:

monai/apps/auto3dseg/data_analyzer.py

@@ -212,7 +212,7 @@ def get_all_case_stats(self, key="training", transform_list=None):
                 manager_list = manager.list()
                 processes = []
                 for rank in range(nprocs):
-                    p = tmp_ctx.Process(
+                    p = tmp_ctx.Process(  # type: ignore[attr-defined]
                         target=self._get_all_case_stats, args=(rank, nprocs, manager_list, key, transform_list)
                     )
                     processes.append(p)

monai/apps/detection/transforms/dictionary.py

@@ -17,6 +17,7 @@
 
 from __future__ import annotations
 
+import warnings
 from collections.abc import Hashable, Mapping, Sequence
 from copy import deepcopy
 from typing import Any
@@ -1326,7 +1327,11 @@ def __init__(
         super().__init__(self.image_keys + self.box_keys, prob, max_k, spatial_axes, allow_missing_keys)
         self.box_ref_image_keys = ensure_tuple_rep(box_ref_image_keys, len(self.box_keys))
 
-    def __call__(self, data: Mapping[Hashable, torch.Tensor]) -> Mapping[Hashable, torch.Tensor]:
+    def __call__(
+        self, data: Mapping[Hashable, torch.Tensor], lazy: bool | None = None
+    ) -> Mapping[Hashable, torch.Tensor]:
+        if lazy is True:
+            warnings.warn("RandRotateBox90d cannot currently execute lazily; ignoring lazy=True")
         self.randomize()
         d = dict(data)
 

monai/apps/reconstruction/transforms/dictionary.py

@@ -11,6 +11,7 @@
 
 from __future__ import annotations
 
+import warnings
 from collections.abc import Hashable, Mapping, Sequence
 
 import numpy as np
@@ -213,10 +214,10 @@ class ReferenceBasedSpatialCropd(Cropd):
     """
 
     def __init__(self, keys: KeysCollection, ref_key: str, allow_missing_keys: bool = False) -> None:
-        super().__init__(keys, cropper=None, allow_missing_keys=allow_missing_keys)  # type: ignore
+        super().__init__(keys, cropper=None, allow_missing_keys=allow_missing_keys, lazy=False)  # type: ignore
         self.ref_key = ref_key
 
-    def __call__(self, data: Mapping[Hashable, Tensor]) -> dict[Hashable, Tensor]:
+    def __call__(self, data: Mapping[Hashable, Tensor], lazy: bool | None = None) -> dict[Hashable, Tensor]:
         """
         This transform can support to crop ND spatial (channel-first) data.
         It also supports pseudo ND spatial data (e.g., (C,H,W) is a pseudo-3D
@@ -229,6 +230,9 @@ def __call__(self, data: Mapping[Hashable, Tensor]) -> dict[Hashable, Tensor]:
         Returns:
             the new data dictionary
         """
+        if lazy is True:
+            warnings.warn("ReferenceBasedSpatialCropd cannot currently execute lazily; ignoring lazy=True")
+
         d = dict(data)
 
         # compute roi_size according to self.ref_key

monai/transforms/__init__.py

@@ -230,7 +230,9 @@
 from .inverse_batch_transform import BatchInverseTransform, Decollated, DecollateD, DecollateDict
 from .io.array import SUPPORTED_READERS, LoadImage, SaveImage
 from .io.dictionary import LoadImaged, LoadImageD, LoadImageDict, SaveImaged, SaveImageD, SaveImageDict
-from .lazy.functional import apply_transforms
+from .lazy.array import ApplyPending
+from .lazy.dictionary import ApplyPendingd, ApplyPendingD, ApplyPendingDict
+from .lazy.functional import apply_pending
 from .lazy.utils import combine_transforms, resample
 from .meta_utility.dictionary import (
     FromMetaTensord,