Updating to match Numpy 2.0 requirements (#7857)

Fixes #7856. 

### Description

This introduces changes to meet Numpy 2.0 requirements. MONAI itself is
compatible with Numpy 2.0 however some dependencies are not such as
older versions of Pytorch. This PR adjusts the MAX_SEED value to be
compatible with Numpy 2.0 behaviour changes, uses the `ptp` function,
and some other minor tweaks. The versions for dependencies are also
fixed to exclude Numpy 2.0.

### Types of changes
<!--- Put an `x` in all the boxes that apply, and remove the not
applicable items -->
- [x] Non-breaking change (fix or new feature that would not break
existing functionality).
- [ ] Breaking change (fix or new feature that would cause existing
functionality to change).
- [ ] New tests added to cover the changes.
- [ ] Integration tests passed locally by running `./runtests.sh -f -u
--net --coverage`.
- [ ] Quick tests passed locally by running `./runtests.sh --quick
--unittests --disttests`.
- [ ] In-line docstrings updated.
- [ ] Documentation updated, tested `make html` command in the `docs/`
folder.

---------

Signed-off-by: Eric Kerfoot <[email protected]>
Co-authored-by: YunLiu <[email protected]>

Author: ericspod

environment-dev.yml

@@ -5,10 +5,10 @@ channels:
   - nvidia
   - conda-forge
 dependencies:
-  - numpy>=1.20
+  - numpy>=1.24,<2.0
   - pytorch>=1.9
   - torchvision
-  - pytorch-cuda=11.6
+  - pytorch-cuda>=11.6
   - pip
   - pip:
     - -r requirements-dev.txt

monai/data/utils.py

@@ -927,7 +927,7 @@ def compute_shape_offset(
     corners = in_affine_ @ corners
     all_dist = corners_out[:-1].copy()
     corners_out = corners_out[:-1] / corners_out[-1]
-    out_shape = np.round(corners_out.ptp(axis=1)) if scale_extent else np.round(corners_out.ptp(axis=1) + 1.0)
+    out_shape = np.round(np.ptp(corners_out, axis=1)) if scale_extent else np.round(np.ptp(corners_out, axis=1) + 1.0)
     offset = None
     for i in range(corners.shape[1]):
         min_corner = np.min(all_dist - all_dist[:, i : i + 1], 1)

monai/transforms/io/array.py

@@ -86,7 +86,7 @@ def switch_endianness(data, new="<"):
         if new not in ("<", ">"):
             raise NotImplementedError(f"Not implemented option new={new}.")
         if current_ != new:
-            data = data.byteswap().newbyteorder(new)
+            data = data.byteswap().view(data.dtype.newbyteorder(new))
     elif isinstance(data, tuple):
         data = tuple(switch_endianness(x, new) for x in data)
     elif isinstance(data, list):

monai/transforms/spatial/functional.py

@@ -373,7 +373,7 @@ def rotate(img, angle, output_shape, mode, padding_mode, align_corners, dtype, l
     if output_shape is None:
         corners = np.asarray(np.meshgrid(*[(0, dim) for dim in im_shape], indexing="ij")).reshape((len(im_shape), -1))
         corners = transform[:-1, :-1] @ corners  # type: ignore
-        output_shape = np.asarray(corners.ptp(axis=1) + 0.5, dtype=int)
+        output_shape = np.asarray(np.ptp(corners, axis=1) + 0.5, dtype=int)
     else:
         output_shape = np.asarray(output_shape, dtype=int)
     shift = create_translate(input_ndim, ((np.array(im_shape) - 1) / 2).tolist())

monai/transforms/transform.py

@@ -203,8 +203,8 @@ def set_random_state(self, seed: int | None = None, state: np.random.RandomState
 
         """
         if seed is not None:
-            _seed = id(seed) if not isinstance(seed, (int, np.integer)) else seed
-            _seed = _seed % MAX_SEED
+            _seed = np.int64(id(seed) if not isinstance(seed, (int, np.integer)) else seed)
+            _seed = _seed % MAX_SEED  # need to account for Numpy2.0 which doesn't silently convert to int64
             self.R = np.random.RandomState(_seed)
             return self
 

requirements.txt

@@ -1,2 +1,2 @@
 torch>=1.9
-numpy>=1.20,<=1.26.0
+numpy>=1.24,<2.0

setup.cfg

@@ -42,7 +42,7 @@ setup_requires =
     ninja
 install_requires =
     torch>=1.9
-    numpy>=1.20
+    numpy>=1.24,<2.0
 
 [options.extras_require]
 all =

tests/test_meta_tensor.py

@@ -448,7 +448,7 @@ def test_shape(self):
 
     def test_astype(self):
         t = MetaTensor([1.0], affine=torch.tensor(1), meta={"fname": "filename"})
-        for np_types in ("float32", "np.float32", "numpy.float32", np.float32, float, "int", np.compat.long, np.uint16):
+        for np_types in ("float32", "np.float32", "numpy.float32", np.float32, float, "int", np.uint16):
             self.assertIsInstance(t.astype(np_types), np.ndarray)
         for pt_types in ("torch.float", torch.float, "torch.float64"):
             self.assertIsInstance(t.astype(pt_types), torch.Tensor)

tests/test_nifti_endianness.py

@@ -82,7 +82,7 @@ def test_switch(self):  # verify data types
         after = switch_endianness(before)
         np.testing.assert_allclose(after.astype(float), expected_float)
 
-        before = np.array(["1.12", "-9.2", "42"], dtype=np.string_)
+        before = np.array(["1.12", "-9.2", "42"], dtype=np.bytes_)
         after = switch_endianness(before)
         np.testing.assert_array_equal(before, after)
 

tests/test_signal_fillempty.py

@@ -30,7 +30,7 @@ class TestSignalFillEmptyNumpy(unittest.TestCase):
     def test_correct_parameters_multi_channels(self):
         self.assertIsInstance(SignalFillEmpty(replacement=0.0), SignalFillEmpty)
         sig = np.load(TEST_SIGNAL)
-        sig[:, 123] = np.NAN
+        sig[:, 123] = np.nan
         fillempty = SignalFillEmpty(replacement=0.0)
         fillemptysignal = fillempty(sig)
         self.assertTrue(not np.isnan(fillemptysignal).any())
@@ -42,7 +42,7 @@ class TestSignalFillEmptyTorch(unittest.TestCase):
     def test_correct_parameters_multi_channels(self):
         self.assertIsInstance(SignalFillEmpty(replacement=0.0), SignalFillEmpty)
         sig = convert_to_tensor(np.load(TEST_SIGNAL))
-        sig[:, 123] = convert_to_tensor(np.NAN)
+        sig[:, 123] = convert_to_tensor(np.nan)
         fillempty = SignalFillEmpty(replacement=0.0)
         fillemptysignal = fillempty(sig)
         self.assertTrue(not torch.isnan(fillemptysignal).any())