Merge branch 'v1' of github.com:openkim/kliff into v1

Author: mjwen

kliff/dataset/dataset.py

@@ -430,7 +430,7 @@ def fingerprint(self, fingerprint):
         """
         Set the fingerprint of the configuration.
         Args:
-         fingerprint: Numpy array which is the fingerprint of the configuration.
+         fingerprint: Object which is the fingerprint of the configuration.
         """
         self._fingerprint = fingerprint
 

kliff/transforms/configuration_transforms/configuration_transform.py

@@ -32,7 +32,7 @@ def forward(self, configuration: Configuration) -> Any:
     def __call__(self, configuration: Configuration) -> Any:
         fingerprint = self.forward(configuration)
         if self.copy_to_config:
-            configuration.fingerprint(fingerprint)
+            configuration.fingerprint = fingerprint
         return fingerprint
 
     def inverse(self, *args, **kargs) -> Configuration:

kliff/transforms/property_transforms.py

@@ -96,7 +96,7 @@ def _get_property_values(
     def _set_property_values(
         self,
         dataset: Union[List[Configuration], Dataset],
-        property_values: Union[np.ndarray, List[float, int]],
+        property_values: Union[np.ndarray, List[Union[float, int]]],
     ):
         """
         Set the property values of all the configurations in a dataset. This method

kliff/utils.py

@@ -194,9 +194,9 @@ def stress_to_voigt(input_stress: np.ndarray) -> list:
         stress[0] = input_stress[0, 0]
         stress[1] = input_stress[1, 1]
         stress[2] = input_stress[2, 2]
-        stress[3] = input_stress[0, 1]
+        stress[3] = input_stress[1, 2]
         stress[4] = input_stress[0, 2]
-        stress[5] = input_stress[1, 2]
+        stress[5] = input_stress[0, 1]
     else:
         raise ValueError("input_stress must be a 2D array")
 
@@ -218,7 +218,7 @@ def stress_to_tensor(input_stress: list) -> np.ndarray:
     stress[1, 1] = input_stress[1]
     stress[2, 2] = input_stress[2]
     stress[1, 2] = stress[2, 1] = input_stress[3]
-    stress[0, 2] = stress[0, 2] = input_stress[4]
+    stress[0, 2] = stress[2, 0] = input_stress[4]
     stress[0, 1] = stress[1, 0] = input_stress[5]
 
     return stress

setup.py

@@ -124,6 +124,8 @@ def get_readme():
             "torch",
             "numpy",
             "ase",
+            "libdescriptor",
+            "torch_geometric",
         ],
         "docs": [
             "sphinx",

tests/dataset/test_ase_parser.py

@@ -2,12 +2,13 @@
 
 import numpy as np
 import pytest
+from ase import io
 
 from kliff.dataset import Dataset
 
 
-def test_ase_parser():
-    """Test ASE parser."""
+def test_dataset_from_ase():
+    """Test ASE parser reading from file using ASE parser"""
 
     # training set
     filename = Path(__file__).parents[1].joinpath("test_data/configs/Si_4.xyz")
@@ -34,6 +35,54 @@ def test_ase_parser():
     assert np.allclose(configs[1].stress, np.array([9.9, 5.5, 1.1, 8.8, 7.7, 4.4]))
 
 
+def test_dataset_add_from_ase():
+    """Test adding configurations to dataset using ASE parser."""
+    filename = Path(__file__).parents[1].joinpath("test_data/configs/Si_4.xyz")
+    data = Dataset()
+    data.add_from_ase(filename, energy_key="Energy", forces_key="force")
+    configs = data.get_configs()
+
+    assert len(configs) == 4
+    assert configs[0].species == ["Si" for _ in range(4)]
+    assert configs[0].coords.shape == (4, 3)
+    assert configs[0].energy == 123.45
+    assert np.allclose(configs[0].stress, np.array([1.1, 5.5, 9.9, 8.8, 7.7, 4.4]))
+    assert configs[1].forces.shape == (8, 3)
+
+
+def test_dataset_from_ase_atoms_list():
+    """Test ASE parser reading from file using ASE atoms list."""
+    filename = Path(__file__).parents[1].joinpath("test_data/configs/Si_4.xyz")
+    atoms = io.read(filename, index=":")
+    data = Dataset.from_ase(
+        ase_atoms_list=atoms, energy_key="Energy", forces_key="force"
+    )
+    configs = data.get_configs()
+
+    assert len(configs) == 4
+    assert configs[0].species == ["Si" for _ in range(4)]
+    assert configs[0].coords.shape == (4, 3)
+    assert configs[0].energy == 123.45
+    assert np.allclose(configs[0].stress, np.array([1.1, 5.5, 9.9, 8.8, 7.7, 4.4]))
+    assert configs[1].forces.shape == (8, 3)
+
+
+def test_dataset_add_from_ase_atoms_list():
+    """Test adding configurations to dataset using ASE atoms list."""
+    filename = Path(__file__).parents[1].joinpath("test_data/configs/Si_4.xyz")
+    atoms = io.read(filename, index=":")
+    data = Dataset()
+    data.add_from_ase(ase_atoms_list=atoms, energy_key="Energy", forces_key="force")
+    configs = data.get_configs()
+
+    assert len(configs) == 4
+    assert configs[0].species == ["Si" for _ in range(4)]
+    assert configs[0].coords.shape == (4, 3)
+    assert configs[0].energy == 123.45
+    assert np.allclose(configs[0].stress, np.array([1.1, 5.5, 9.9, 8.8, 7.7, 4.4]))
+    assert configs[1].forces.shape == (8, 3)
+
+
 # def test_colabfit_parser():
 # TODO: figure minimal colabfit example to run tests on
 #    pass

tests/dataset/test_configurations.py

@@ -0,0 +1,62 @@
+from pathlib import Path
+
+import numpy as np
+import pytest
+from ase import io
+
+from kliff.dataset import Configuration
+from kliff.dataset.dataset import ConfigurationError
+from kliff.utils import stress_to_voigt
+
+
+def test_configuration_from_ase():
+    """Test initializing Configuration from ASE atoms object"""
+
+    filename = Path(__file__).parents[1].joinpath("test_data/configs/Si_4.xyz")
+    atoms = io.read(filename, index=":")
+    config = Configuration.from_ase_atoms(
+        atoms[0], energy_key="Energy", forces_key="force"
+    )
+
+    assert config.species == ["Si" for _ in range(4)]
+    assert config.coords.shape == (4, 3)
+    assert config.energy == 123.45
+    assert np.allclose(config.stress, np.array([1.1, 5.5, 9.9, 8.8, 7.7, 4.4]))
+    assert config.forces.shape == (4, 3)
+    assert config.stress.shape == (6,)
+
+
+def test_configuration_to_ase():
+    """Test converting Configuration to ASE atoms object"""
+
+    filename = Path(__file__).parents[1].joinpath("test_data/configs/Si_4.xyz")
+    atoms = io.read(filename, index=":")
+    config = Configuration.from_ase_atoms(
+        atoms[0], energy_key="Energy", forces_key="force"
+    )
+
+    atoms = config.to_ase_atoms()
+    assert np.allclose(atoms.get_positions(), config.coords)
+    assert atoms.info["energy"] == config.energy
+    assert np.allclose(atoms.arrays["forces"], config.forces)
+    assert np.allclose(stress_to_voigt(atoms.info["stress"]), config.stress)
+    # TODO: As per Marcos' suggestion, we should use the dict
+    # method to get the ASE atoms properties. It solves the issue
+    # of associated calculator.
+
+
+def test_configuration_from_file():
+    """Test initializing Configuration from file"""
+
+    filename = (
+        Path(__file__).parents[1].joinpath("test_data/configs/Si_4/Si_T300_step_0.xyz")
+    )
+    config = Configuration.from_file(filename)
+
+    assert config.species == ["Si" for _ in range(64)]
+    assert config.coords.shape == (64, 3)
+    assert config.energy == 0.0
+    assert config.forces.shape == (64, 3)
+    # stress should raise exception
+    with pytest.raises(ConfigurationError):
+        stress = config.stress

tests/misc/test_utils.py

@@ -0,0 +1,25 @@
+import numpy as np
+import pytest
+
+from kliff.utils import stress_to_tensor, stress_to_voigt
+
+
+def test_stress_conversion():
+    stress_tensor = np.random.rand(3, 3)
+    stress_tensor = stress_tensor + stress_tensor.T  # make it symmetric
+    stress_voigt = stress_to_voigt(stress_tensor)
+    assert np.allclose(
+        stress_voigt,
+        np.array(
+            [
+                stress_tensor[0, 0],
+                stress_tensor[1, 1],
+                stress_tensor[2, 2],
+                stress_tensor[1, 2],
+                stress_tensor[0, 2],
+                stress_tensor[0, 1],
+            ]
+        ),
+    )
+    print(stress_tensor, stress_to_tensor(stress_voigt))
+    assert np.allclose(stress_tensor, stress_to_tensor(stress_voigt))

tests/test_data/precomputed_numpy_arrays/cm_si.npy

@@ -0,0 +1,74 @@
+import numpy as np
+from ase.data import atomic_numbers
+
+from kliff.dataset import Configuration
+from kliff.transforms.configuration_transforms import ConfigurationTransform
+
+
+class GlobalCoulombMatrix(ConfigurationTransform):
+    """
+    Coulomb matrix representation of the configuration.
+    """
+
+    def __init__(self, max_atoms: int = 5, copy_to_config: bool = False):
+        super().__init__(copy_to_config)
+        self.max_atoms = max_atoms
+
+    def forward(self, configuration: Configuration):
+        """
+        Generate the Coulomb matrix for the configuration.
+
+        Args:
+            configuration: Instance of ~:class:`kliff.dataset.Configuration`. For which the
+                Coulomb matrix is to be generated.
+
+        Returns:
+            Coulomb matrix of the configuration.
+        """
+        coords = configuration.coords
+        n_atoms = configuration.get_num_atoms()
+        coulomb_mat = np.zeros((self.max_atoms, self.max_atoms))
+        species = [atomic_numbers[elem] for elem in configuration.species]
+        for i in range(n_atoms):
+            for j in range(i + 1):
+                if i == j:
+                    coulomb_mat[i, j] = 0.5 * (species[i] ** 2.4)
+                else:
+                    r = np.linalg.norm(coords[i] - coords[j])
+                    coulomb_mat[i, j] = species[i] * species[j] / r
+                    coulomb_mat[j, i] = coulomb_mat[i, j]
+        return coulomb_mat
+
+    def backward(self, fingerprint, configuration):
+        """
+        Inverse mapping of the transform. This is not implemented for any of the transforms,
+        but is there for future use.
+        """
+        NotImplementedError(
+            "Any of the implemented transforms do not support inverse mapping.\n"
+            "For computing jacobian-vector product use `backward` function."
+        )
+
+    def collate_fn(self, config_list):
+        """
+        Collate function for the Coulomb matrix transform.
+        """
+        return [self.forward(config) for config in config_list]
+
+
+def test_configuration_transform(test_data_dir):
+    config = Configuration.from_file(test_data_dir / "configs/Si.xyz")
+    transform = GlobalCoulombMatrix(max_atoms=8)
+    fingerprint = transform(config)
+    assert fingerprint.shape == (8, 8)
+    assert np.allclose(fingerprint, fingerprint.T)
+    assert np.allclose(
+        fingerprint, np.load(test_data_dir / "precomputed_numpy_arrays/cm_si.npy")
+    )
+    assert config.fingerprint is None
+    dataset = [config, config]
+    fingerprints = transform.collate_fn(dataset)
+    assert len(fingerprints) == 2
+    assert fingerprints[0].shape == (8, 8)
+    assert fingerprints[1].shape == (8, 8)
+    assert np.array(fingerprints).shape == (2, 8, 8)