Support 0 molecule subsets when loading dataset (#257)

* Support 0 molecule subsets when loading dataset

* fix

* fix

torchani/utils.py

@@ -48,14 +48,17 @@ def pad_atomic_properties(atomic_properties, padding_values=defaultdict(lambda:
     max_atoms = max(x[anykey].shape[1] for x in atomic_properties)
     padded = {k: [] for k in keys}
     for p in atomic_properties:
-        num_molecules = max(v.shape[0] for v in p.values())
+        num_molecules = 1
+        for v in p.values():
+            assert num_molecules in {1, v.shape[0]}, 'Number of molecules in different atomic properties mismatch'
+            if v.shape[0] != 1:
+                num_molecules = v.shape[0]
         for k, v in p.items():
             shape = list(v.shape)
             padatoms = max_atoms - shape[1]
             shape[1] = padatoms
             padding = v.new_full(shape, padding_values[k])
             v = torch.cat([v, padding], dim=1)
-            if v.shape[0] < num_molecules:
             shape = list(v.shape)
             shape[0] = num_molecules
             v = v.expand(*shape)