Merge TorchScript tests with regular tests (#32)

Co-authored-by: Luthaf <[email protected]>
Co-authored-by: frostedoyster <[email protected]>

setup.py

@@ -1,31 +1,32 @@
-from setuptools import setup, find_packages
-import sys
 import subprocess
 
+from setuptools import find_packages, setup
 
 # Detecting if pytorch with or without CUDA support should be installed
 try:
-    subprocess.check_output('nvidia-smi')
+    subprocess.check_output("nvidia-smi")
     HAS_NVIDIA = True
 except:
     HAS_NVIDIA = False
 
 if HAS_NVIDIA:
     dependency_links = []
 else:
-    dependency_links = ['https://download.pytorch.org/whl/cpu']
-    print("torch_spex setup info: Did not find NVIDIA card defaulting to CPU-only installation")
+    dependency_links = ["https://download.pytorch.org/whl/cpu"]
+    print(
+        "torch_spex setup info: Did not find NVIDIA card, defaulting to CPU-only installation"
+    )
 
 setup(
-    name='torch_spex',
-    packages = find_packages(),
+    name="torch_spex",
+    packages=find_packages(),
     install_requires=[
-        'sphericart[torch] @ git+https://github.com/lab-cosmo/sphericart.git@99761b0', # pre-built wheels don't work
-        'numpy',
-        'ase',
-        'torch',
-        'scipy',
-        'metatensor[torch] @ https://github.com/lab-cosmo/metatensor/archive/0436e27.zip',
+        "sphericart[torch] @ git+https://github.com/lab-cosmo/sphericart.git@ecf4145",  # pre-built wheels don't work
+        "numpy",
+        "ase",
+        "torch",
+        "scipy",
+        "metatensor[torch]",
     ],
-    dependency_links = dependency_links
+    dependency_links=dependency_links,
 )

tests/test_spherical_expansions.py

@@ -4,7 +4,6 @@
 
 import metatensor.torch
 from metatensor.torch import Labels, TensorBlock, TensorMap
-import numpy as np
 import ase.io
 
 from torch_spex.spherical_expansions import VectorExpansion, SphericalExpansion
@@ -18,7 +17,9 @@ class TestEthanol1SphericalExpansion:
     device = "cpu"
     dtype = torch.float32
     frames = ase.io.read('datasets/rmd17/ethanol1.extxyz', ':1')
-    all_species = list(np.unique([frame.numbers for frame in frames]))
+    all_species = torch.unique(torch.concatenate([torch.tensor(frame.numbers)
+                    for frame in frames]))
+    all_species = [int(species) for species in all_species]
     with open("tests/data/expansion_coeffs-ethanol1_0-hypers.json", "r") as f:
         hypers = json.load(f)
 
@@ -34,16 +35,22 @@ def test_vector_expansion_coeffs(self):
         tm_ref = metatensor.torch.to(tm_ref, device=self.device, dtype=self.dtype)
         # we need to sort both computed and reference pair expansion coeffs,
         # because ase.neighborlist can get different neighborlist order for some reasons
-        tm_ref = sort_tm(tm_ref)
+        tm_ref = metatensor.torch.sort(tm_ref)
         vector_expansion = VectorExpansion(self.hypers, self.all_species,
                 device=self.device, dtype=self.dtype)
         with torch.no_grad():
-            tm = sort_tm(vector_expansion.forward(**self.batch))
+            tm = metatensor.torch.sort(vector_expansion.forward(**self.batch))
         # Default types are float32 so we cannot get higher accuracy than 1e-7.
         # Because the reference value have been cacluated using float32 and
         # now we using float64 computation the accuracy had to be decreased again
         assert metatensor.torch.allclose(tm_ref, tm, atol=1e-5, rtol=1e-5)
 
+        vector_expansion_script = torch.jit.script(vector_expansion)
+        with torch.no_grad():
+            tm_script = metatensor.torch.sort(vector_expansion_script.forward(**self.batch))
+        assert metatensor.torch.allclose(tm, tm_script, atol=1e-5,
+                rtol=torch.finfo(self.dtype).eps*10)
+
     def test_spherical_expansion_coeffs(self):
         tm_ref = metatensor.torch.load("tests/data/spherical_expansion_coeffs-ethanol1_0-data.npz")
         tm_ref = metatensor.torch.to(tm_ref, device=self.device, dtype=self.dtype)
@@ -56,6 +63,12 @@ def test_spherical_expansion_coeffs(self):
         # now we using float64 computation the accuracy had to be decreased again
         assert metatensor.torch.allclose(tm_ref, tm, atol=1e-5, rtol=1e-5)
 
+        spherical_expansion_script = torch.jit.script(spherical_expansion_calculator)
+        with torch.no_grad():
+            tm_script = metatensor.torch.sort(spherical_expansion_script.forward(**self.batch))
+        assert metatensor.torch.allclose(tm, tm_script, atol=1e-5,
+                rtol=torch.finfo(self.dtype).eps*10)
+
     def test_spherical_expansion_coeffs_alchemical(self):
         with open("tests/data/expansion_coeffs-ethanol1_0-alchemical-hypers.json", "r") as f:
             hypers = json.load(f)
@@ -88,7 +101,9 @@ class TestArtificialSphericalExpansion:
     device = "cpu"
     dtype = torch.float32
     frames = ase.io.read('tests/datasets/artificial.extxyz', ':')
-    all_species = list(np.unique(np.hstack([frame.numbers for frame in frames])))
+    all_species = torch.unique(torch.concatenate([torch.tensor(frame.numbers)
+                    for frame in frames]))
+    all_species = [int(species) for species in all_species]
     with open("tests/data/expansion_coeffs-artificial-hypers.json", "r") as f:
         hypers = json.load(f)
 
@@ -101,11 +116,11 @@ class TestArtificialSphericalExpansion:
     def test_vector_expansion_coeffs(self):
         tm_ref = metatensor.torch.load("tests/data/vector_expansion_coeffs-artificial-data.npz")
         tm_ref = metatensor.torch.to(tm_ref, device=self.device, dtype=self.dtype)
-        tm_ref = sort_tm(tm_ref)
+        tm_ref = metatensor.torch.sort(tm_ref)
         vector_expansion = VectorExpansion(self.hypers, self.all_species,
                 device=self.device, dtype=self.dtype)
         with torch.no_grad():
-            tm = sort_tm(vector_expansion.forward(**self.batch))
+            tm = metatensor.torch.sort(vector_expansion.forward(**self.batch))
         assert metatensor.torch.allclose(tm_ref, tm, atol=1e-5, rtol=1e-5)
 
     def test_spherical_expansion_coeffs(self):
@@ -137,39 +152,3 @@ def test_spherical_expansion_coeffs_artificial(self):
         with torch.no_grad():
             tm = spherical_expansion_calculator.forward(**self.batch)
         assert metatensor.torch.allclose(tm_ref, tm, atol=1e-5, rtol=1e-5)
-
-### these util functions will be removed once lab-cosmo/metatensor/pull/281 is merged
-def native_list_argsort(native_list):
-    return sorted(range(len(native_list)), key=native_list.__getitem__)
-
-def sort_tm(tm):
-    blocks = []
-    for _, block in tm.items():
-        values = block.values
-
-        samples_values = block.samples.values
-        sorted_idx = native_list_argsort([tuple(row.tolist()) for row in block.samples.values])
-        samples_values = samples_values[sorted_idx]
-        values = values[sorted_idx]
-
-        components_values = []
-        for i, component in enumerate(block.components):
-            component_values = component.values
-            sorted_idx = native_list_argsort([tuple(row.tolist()) for row in component.values])
-            components_values.append( component_values[sorted_idx] )
-            values = np.take(values, sorted_idx, axis=i+1)
-
-        properties_values = block.properties.values
-        sorted_idx = native_list_argsort([tuple(row.tolist()) for row in block.properties.values])
-        properties_values = properties_values[sorted_idx]
-        values = values[..., sorted_idx]
-
-        blocks.append(
-            TensorBlock(
-                values=values,
-                samples=Labels(values=samples_values, names=block.samples.names),
-                components=[Labels(values=components_values[i], names=component.names) for i, component in enumerate(block.components)],
-                properties=Labels(values=properties_values, names=block.properties.names)
-            )
-        )
-    return TensorMap(keys=tm.keys, blocks=blocks)