add dipeptides dataset

openqdc/datasets/potential/__init__.py

@@ -2,6 +2,7 @@
 from .ani import ANI1, ANI1CCX, ANI1CCX_V2, ANI1X, ANI2X
 from .bpa import BPA
 from .comp6 import COMP6
+from .dipeptides import Dipeptides
 from .dummy import Dummy, PredefinedDataset
 from .gdml import GDML
 from .geom import GEOM
@@ -37,6 +38,7 @@
     "ANI2X": ANI2X,
     "BPA": BPA,
     "COMP6": COMP6,
+    "Dipeptides": Dipeptides,
     "GDML": GDML,
     "GEOM": GEOM,
     "ISO17": ISO17,

openqdc/datasets/potential/dipeptides.py

@@ -0,0 +1,81 @@
+import numpy as np 
+from openqdc.datasets.base import BaseDataset
+from openqdc.methods import PotentialMethod
+
+def shape_atom_inputs(coords, atom_species):
+    reshaped_coords = coords.reshape(-1, 3)
+    frame, atoms, _ = coords.shape
+    z = np.tile(atom_species, frame)
+    xs = np.stack((z, np.zeros_like(z)), axis=-1)
+    return np.concatenate((xs, reshaped_coords), axis=-1, dtype=np.float32)
+
+
+def read_npz_entry(folder):
+    data, name = create_path(folder)
+    data = np.load(data)
+
+    nuclear_charges, coords, energies, forces = (
+        data["nuclear_charges"],
+        data["coords"],
+        data["energies"],
+        data["forces"],
+    )
+    frames = coords.shape[0]
+    res = dict(
+        name=np.array([name] * frames),
+        subset=np.array(["dipeptides"] * frames),
+        energies=energies[:, None].astype(np.float32),
+        forces=forces.reshape(-1, 3, 1).astype(np.float32),
+        atomic_inputs=shape_atom_inputs(coords, nuclear_charges),
+        n_atoms=np.array([len(nuclear_charges)] * frames, dtype=np.int32),
+    )
+    return res
+
+
+def create_path(folder):
+    name = folder.split("/")[-1]
+    return folder, name 
+
+folder="/network/scratch/s/semih.canturk/cache/openqdc/dipeptides/npz_files/mol_73.npz"
+
+trajectories={
+    "mol_73": folder
+}
+
+class Dipeptides(BaseDataset):
+    """
+    """
+
+    __name__ = "dipeptides"
+
+    __energy_methods__ = [PotentialMethod.WB97M_D3BJ_DEF2_TZVPPD]
+
+    energy_target_names = [
+        "",
+    ]
+
+    __energy_unit__ = "kj/mol"
+    __distance_unit__ = "ang"
+    __forces_unit__ = "kj/mol/ang"
+
+    __force_mask__ = [False]
+
+    @property
+    def data_types(self):
+        return {
+            "atomic_inputs": np.float32,
+            "position_idx_range": np.int32,
+            "energies": np.float32,
+            "forces": np.float32,
+        }
+
+    def read_raw_entries(self):
+        entries_list = []
+
+        for dummy_name, path_to_npz in trajectories.items():
+            entries_list.append(read_npz_entry(path_to_npz))
+        return entries_list
+
+
+# to store it in the cache and loading back (add the dataset in the __init__)
+# Dipeptides.no_init().preprocess(upload=False, overwrite=True)