ASE neighborlist computer (#120)

docs/api.rst

@@ -45,6 +45,13 @@ NeuroChem
 .. automodule:: torchani.neurochem.trainer
 
 
+ASE Interface
+=============
+
+.. automodule:: torchani.ase
+.. autoclass:: torchani.ase.NeighborList
+    :members:
+
 Ignite Helpers
 ==============
 

docs/conf.py

@@ -38,6 +38,7 @@
     'numpy': ('http://docs.scipy.org/doc/numpy/', None),
     'torch': ('https://pytorch.org/docs/master/', None),
     'ignite': ('https://pytorch.org/ignite/', None),
+    'ase': ('https://wiki.fysik.dtu.dk/ase/', None),
 }
 
 latex_documents = [

setup.py

@@ -21,6 +21,7 @@
         'nose',
         'tensorboardX',
         'tqdm',
+        'ase',
     ],
 }
 

tests/test_aev.py

@@ -56,5 +56,12 @@ def testPadding(self):
             self._assertAEVEqual(expected_radial, expected_angular, aev_)
 
 
+class TestAEVASENeighborList(TestAEV):
+
+    def setUp(self):
+        super(TestAEVASENeighborList, self).setUp()
+        self.aev_computer.neighborlist = torchani.ase.NeighborList()
+
+
 if __name__ == '__main__':
     unittest.main()

torchani/__init__.py

@@ -34,3 +34,9 @@
 
 __all__ = ['AEVComputer', 'EnergyShifter', 'ANIModel', 'Ensemble',
            'ignite', 'utils', 'neurochem', 'data']
+
+try:
+    from . import ase  # noqa: F401
+    __all__.append('ase')
+except ImportError:
+    pass

torchani/ase.py

@@ -0,0 +1,82 @@
+# -*- coding: utf-8 -*-
+"""Tools for interfacing with `ASE`_.
+
+.. _ASE:
+    https://wiki.fysik.dtu.dk/ase
+"""
+
+import math
+import torch
+import ase.neighborlist
+from . import utils
+
+
+class NeighborList:
+    """ASE neighborlist computer
+
+    Arguments:
+        cell: same as in :class:`ase.Atoms`
+        pbc: same as in :class:`ase.Atoms`
+    """
+
+    def __init__(self, cell=None, pbc=None):
+        self.pbc = pbc
+        self.cell = cell
+
+    def __call__(self, species, coordinates, cutoff):
+        conformations = species.shape[0]
+        max_atoms = species.shape[1]
+        neighbor_species = []
+        neighbor_distances = []
+        neighbor_vecs = []
+        for i in range(conformations):
+            s = species[i].unsqueeze(0)
+            c = coordinates[i].unsqueeze(0)
+            s, c = utils.strip_redundant_padding(s, c)
+            s = s.squeeze()
+            c = c.squeeze()
+            atoms = s.shape[0]
+            atoms_object = ase.Atoms(
+                'C'*atoms,  # chemical symbols are not important here
+                positions=c.numpy(),
+                pbc=self.pbc,
+                cell=self.cell)
+            idx1, idx2, d, D = ase.neighborlist.neighbor_list(
+                'ijdD', atoms_object, cutoff)
+            idx1 = torch.from_numpy(idx1).to(coordinates.device)
+            idx2 = torch.from_numpy(idx2).to(coordinates.device)
+            d = torch.from_numpy(d).to(coordinates.device) \
+                .to(coordinates.dtype)
+            D = torch.from_numpy(D).to(coordinates.device) \
+                .to(coordinates.dtype)
+            neighbor_species1 = []
+            neighbor_distances1 = []
+            neighbor_vecs1 = []
+            for i in range(atoms):
+                this_atom_indices = (idx1 == i).nonzero().flatten()
+                neighbor_indices = idx2[this_atom_indices]
+                neighbor_species1.append(s[neighbor_indices])
+                neighbor_distances1.append(d[this_atom_indices])
+                neighbor_vecs1.append(D.index_select(0, this_atom_indices))
+            for i in range(max_atoms - atoms):
+                neighbor_species1.append(torch.full((1,), -1))
+                neighbor_distances1.append(torch.full((1,), math.inf))
+                neighbor_vecs1.append(torch.full((1, 3), 0))
+            neighbor_species1 = torch.nn.utils.rnn.pad_sequence(
+                neighbor_species1, padding_value=-1)
+            neighbor_distances1 = torch.nn.utils.rnn.pad_sequence(
+                neighbor_distances1, padding_value=math.inf)
+            neighbor_vecs1 = torch.nn.utils.rnn.pad_sequence(
+                neighbor_vecs1, padding_value=0)
+            neighbor_species.append(neighbor_species1)
+            neighbor_distances.append(neighbor_distances1)
+            neighbor_vecs.append(neighbor_vecs1)
+        neighbor_species = torch.nn.utils.rnn.pad_sequence(
+            neighbor_species, batch_first=True, padding_value=-1)
+        neighbor_distances = torch.nn.utils.rnn.pad_sequence(
+            neighbor_distances, batch_first=True, padding_value=math.inf)
+        neighbor_vecs = torch.nn.utils.rnn.pad_sequence(
+            neighbor_vecs, batch_first=True, padding_value=0)
+        return neighbor_species.permute(0, 2, 1), \
+            neighbor_distances.permute(0, 2, 1), \
+            neighbor_vecs.permute(0, 2, 1, 3)

torchani/neurochem/__init__.py

@@ -32,7 +32,7 @@ def __init__(self, filename):
                     name = line[0]
                     value = line[1]
                     if name == 'Rcr' or name == 'Rca':
-                        setattr(self, name, torch.tensor(float(value)))
+                        setattr(self, name, float(value))
                     elif name in ['EtaR', 'ShfR', 'Zeta',
                                   'ShfZ', 'EtaA', 'ShfA']:
                         value = [float(x.strip()) for x in value.replace(