Make unit tests compatible with NumPy 2.x

.github/workflows/ReceivePR.yml

@@ -22,7 +22,7 @@ jobs:
       - name: Use Python
         uses: actions/setup-python@42375524e23c412d93fb67b49958b491fce71c38 # v5.4.0
         with:
-          python-version: 3.9
+          python-version: '3.10'
           architecture: x64
 
       - name: Setup MPI

.github/workflows/ci.yaml

@@ -12,7 +12,6 @@ jobs:
       fail-fast: false
       matrix:
         py-version:
-          - 3.9
           - '3.10'
           - 3.11
           - 3.12

heat/core/linalg/tests/test_basics.py

@@ -61,19 +61,20 @@ def test_cross(self):
         self.assert_array_equal(cross_axisc, np_cross_axisc)
 
         # test vector axes with 2 elements
-        b_2d = ht.array(np_b[:-1, :, :], split=1)
-        cross_3d_2d = ht.cross(a, b_2d, axisa=1, axisb=0)
-        np_cross_3d_2d = np.cross(np_a, np_b[:-1, :, :], axisa=1, axisb=0)
-        self.assert_array_equal(cross_3d_2d, np_cross_3d_2d)
-
-        a_2d = ht.array(np_a[:, :-1, :], split=0)
-        cross_2d_3d = ht.cross(a_2d, b, axisa=1, axisb=0)
-        np_cross_2d_3d = np.cross(np_a[:, :-1, :], np_b, axisa=1, axisb=0)
-        self.assert_array_equal(cross_2d_3d, np_cross_2d_3d)
-
-        cross_z_comp = ht.cross(a_2d, b_2d, axisa=1, axisb=0)
-        np_cross_z_comp = np.cross(np_a[:, :-1, :], np_b[:-1, :, :], axisa=1, axisb=0)
-        self.assert_array_equal(cross_z_comp, np_cross_z_comp)
+        if np.lib.NumpyVersion(np.__version__) < "2.0.0b1":
+            b_2d = ht.array(np_b[:-1, :, :], split=1)
+            cross_3d_2d = ht.cross(a, b_2d, axisa=1, axisb=0)
+            np_cross_3d_2d = np.cross(np_a, np_b[:-1, :, :], axisa=1, axisb=0)
+            self.assert_array_equal(cross_3d_2d, np_cross_3d_2d)
+
+            a_2d = ht.array(np_a[:, :-1, :], split=0)
+            cross_2d_3d = ht.cross(a_2d, b, axisa=1, axisb=0)
+            np_cross_2d_3d = np.cross(np_a[:, :-1, :], np_b, axisa=1, axisb=0)
+            self.assert_array_equal(cross_2d_3d, np_cross_2d_3d)
+
+            cross_z_comp = ht.cross(a_2d, b_2d, axisa=1, axisb=0)
+            np_cross_z_comp = np.cross(np_a[:, :-1, :], np_b[:-1, :, :], axisa=1, axisb=0)
+            self.assert_array_equal(cross_z_comp, np_cross_z_comp)
 
         a_wrong_split = ht.array(np_a[:, :-1, :], split=2)
         with self.assertRaises(ValueError):

heat/core/tests/test_exponential.py

@@ -96,7 +96,8 @@ def test_expm1(self):
 
     def test_exp2(self):
         elements = 10
-        tmp = np.exp2(torch.arange(elements, dtype=torch.float64))
+        tmp = np.exp2(torch.arange(elements, dtype=torch.float64).numpy())
+        tmp = torch.tensor(tmp)
         tmp = tmp.to(self.device.torch_device)
         comparison = ht.array(tmp, device=self.device)
 

heat/core/tests/test_manipulations.py

@@ -2687,22 +2687,31 @@ def test_row_stack(self):
         # test local row_stack, 2-D arrays
         a = np.arange(10, dtype=np.float32).reshape(2, 5)
         b = np.arange(15, dtype=np.float32).reshape(3, 5)
-        np_rstack = np.row_stack((a, b))
+        if np.lib.NumpyVersion(np.__version__) >= "2.0.0b1":
+            np_rstack = np.vstack((a, b))
+        else:
+            np_rstack = np.row_stack((a, b))
         ht_a = ht.array(a)
         ht_b = ht.array(b)
         ht_rstack = ht.row_stack((ht_a, ht_b))
         self.assertTrue((np_rstack == ht_rstack.numpy()).all())
 
         # 2-D and 1-D arrays
         c = np.arange(5, dtype=np.float32)
-        np_rstack = np.row_stack((a, b, c))
+        if np.lib.NumpyVersion(np.__version__) >= "2.0.0b1":
+            np_rstack = np.vstack((a, b, c))
+        else:
+            np_rstack = np.row_stack((a, b, c))
         ht_c = ht.array(c)
         ht_rstack = ht.row_stack((ht_a, ht_b, ht_c))
         self.assertTrue((np_rstack == ht_rstack.numpy()).all())
 
         # 2-D and 1-D arrays, distributed
         c = np.arange(5, dtype=np.float32)
-        np_rstack = np.row_stack((a, b, c))
+        if np.lib.NumpyVersion(np.__version__) >= "2.0.0b1":
+            np_rstack = np.vstack((a, b, c))
+        else:
+            np_rstack = np.row_stack((a, b, c))
         ht_a = ht.array(a, split=0)
         ht_b = ht.array(b, split=0)
         ht_c = ht.array(c, split=0)
@@ -2713,7 +2722,10 @@ def test_row_stack(self):
         # 1-D arrays, distributed, different dtypes
         d = np.arange(10).astype(np.float32)
         e = np.arange(10)
-        np_rstack = np.row_stack((d, e))
+        if np.lib.NumpyVersion(np.__version__) >= "2.0.0b1":
+            np_rstack = np.vstack((d, e))
+        else:
+            np_rstack = np.row_stack((d, e))
         ht_d = ht.array(d, split=0)
         ht_e = ht.array(e, split=0)
         ht_rstack = ht.row_stack((ht_d, ht_e))

heat/fft/tests/test_fft.py

@@ -123,7 +123,7 @@ def test_fftn_ifftn(self):
         x = ht.random.randn(10, 8, 6, dtype=ht.float64, split=0)
         # FFT along last 2 axes
         y = ht.fft.fftn(x, s=(6, 6))
-        np_y = np.fft.fftn(x.numpy(), s=(6, 6))
+        np_y = np.fft.fftn(x.numpy(), s=(6, 6), axes=(1, 2))
         self.assertIsInstance(y, ht.DNDarray)
         self.assertEqual(y.shape, np_y.shape)
         self.assertTrue(y.split == 0)

setup.py

@@ -21,10 +21,9 @@
     author_email="[email protected]",
     url="https://github.com/helmholtz-analytics/heat",
     keywords=["data", "analytics", "tensors", "distributed", "gpu"],
-    python_requires=">=3.9",
+    python_requires=">=3.10",
     classifiers=[
         "Development Status :: 4 - Beta",
-        "Programming Language :: Python :: 3.9",
         "Programming Language :: Python :: 3.10",
         "Programming Language :: Python :: 3.11",
         "Programming Language :: Python :: 3.12",
@@ -34,9 +33,9 @@
     ],
     install_requires=[
         "mpi4py>=3.0.0",
-        "numpy>=1.22.0, <2",
+        "numpy>=1.23.5",
         "torch>=2.0.0, <2.6.1",
-        "scipy>=1.10.0",
+        "scipy>=1.14.0",
         "pillow>=6.0.0",
         "torchvision>=0.15.2, <0.21.1",
     ],