Change format

src/openqdc/__init__.py

@@ -9,7 +9,7 @@
 
 _lazy_imports_obj = {}
 
-_lazy_imports_mod = {"datasets": "openqdc.datamodule", "utils": "openqdc.utils"}
+_lazy_imports_mod = {"datasets": "openqdc.datasets", "utils": "openqdc.utils"}
 
 
 def __getattr__(name):

src/openqdc/datasets/ani.py

@@ -39,6 +39,12 @@ class ANI1(BaseDataset):
     def root(self):
         return p_join(get_local_cache(), "ani")
 
+    def __smiles_converter__(self, x):
+        """util function to convert string to smiles: useful if the smiles is
+        encoded in a different format than its display format
+        """
+        return "-".join(x.decode("ascii").split("-")[:-1])
+
     @property
     def preprocess_path(self):
         path = p_join(self.root, "preprocessed", self.__name__)
@@ -89,6 +95,12 @@ class ANI1CCX(ANI1):
     __force_methods__ = []
     force_target_names = []
 
+    def __smiles_converter__(self, x):
+        """util function to convert string to smiles: useful if the smiles is
+        encoded in a different format than its display format
+        """
+        return x
+
 
 class ANI1X(ANI1):
     """
@@ -145,3 +157,9 @@ class ANI1X(ANI1):
 
     def convert_forces(self, x):
         return super().convert_forces(x) * 0.529177249  # correct the Dataset error
+
+    def __smiles_converter__(self, x):
+        """util function to convert string to smiles: useful if the smiles is
+        encoded in a different format than its display format
+        """
+        return x

src/openqdc/datasets/base.py

@@ -1,4 +1,5 @@
 import os
+import pickle as pkl
 from os.path import join as p_join
 from typing import Dict, List, Optional, Union
 
@@ -24,7 +25,7 @@
     push_remote,
     set_cache_dir,
 )
-from openqdc.utils.molecule import atom_table
+from openqdc.utils.molecule import atom_table, z_to_formula
 from openqdc.utils.package_utils import requires_package
 from openqdc.utils.units import get_conversion
 
@@ -43,7 +44,7 @@ def extract_entry(
 
     res = dict(
         name=np.array([df["name"][i]]),
-        subset=np.array([subset]),
+        subset=np.array([subset if subset is not None else z_to_formula(x)]),
         energies=energies.reshape((1, -1)).astype(np.float32),
         atomic_inputs=np.concatenate((xs, positions), axis=-1, dtype=np.float32),
         n_atoms=np.array([x.shape[0]], dtype=np.int32),
@@ -64,8 +65,8 @@ def read_qc_archive_h5(
 ) -> List[Dict[str, np.ndarray]]:
     data = load_hdf5_file(raw_path)
     data_t = {k2: data[k1][k2][:] for k1 in data.keys() for k2 in data[k1].keys()}
-    n = len(data_t["molecule_id"])
 
+    n = len(data_t["molecule_id"])
     samples = [extract_entry(data_t, i, subset, energy_target_names, force_target_names) for i in tqdm(range(n))]
     return samples
 
@@ -96,9 +97,6 @@ def __init__(
         self._set_units(energy_unit, distance_unit)
         if not self.is_preprocessed():
             logger.info("This dataset not available. Please open an issue on Github for the team to look into it.")
-            # entries = self.read_raw_entries()
-            # res = self.collate_list(entries)
-            # self.save_preprocess(res)
         else:
             self.read_preprocess(overwrite_local_cache=overwrite_local_cache)
             self._set_isolated_atom_energies()
@@ -107,12 +105,12 @@ def __init__(
     def numbers(self):
         if hasattr(self, "_numbers"):
             return self._numbers
-        self._numbers = np.array(list(set(self.data["atomic_inputs"][..., 0])), dtype=np.int32)
+        self._numbers = np.unique(self.data["atomic_inputs"][..., 0]).astype(np.int32)
         return self._numbers
 
     @property
     def chemical_species(self):
-        return [chemical_symbols[z] for z in self.numbers]
+        return np.array(chemical_symbols)[self.numbers]
 
     @property
     def energy_unit(self):
@@ -211,10 +209,11 @@ def collate_list(self, list_entries):
         # concatenate entries
         res = {key: np.concatenate([r[key] for r in list_entries if r is not None], axis=0) for key in list_entries[0]}
 
-        csum = np.cumsum(res.pop("n_atoms"))
+        csum = np.cumsum(res.get("n_atoms"))
         x = np.zeros((csum.shape[0], 2), dtype=np.int32)
         x[1:, 0], x[:, 1] = csum[:-1], csum
         res["position_idx_range"] = x
+
         return res
 
     def save_preprocess(self, data_dict):
@@ -228,12 +227,13 @@ def save_preprocess(self, data_dict):
             push_remote(local_path, overwrite=True)
 
         # save smiles and subset
+        local_path = p_join(self.preprocess_path, "props.pkl")
         for key in ["name", "subset"]:
-            local_path = p_join(self.preprocess_path, f"{key}.npz")
-            uniques, inv_indices = np.unique(data_dict[key], return_inverse=True)
-            with open(local_path, "wb") as f:
-                np.savez_compressed(f, uniques=uniques, inv_indices=inv_indices)
-            push_remote(local_path)
+            data_dict[key] = np.unique(data_dict[key], return_inverse=True)
+
+        with open(local_path, "wb") as f:
+            pkl.dump(data_dict, f)
+        push_remote(local_path, overwrite=True)
 
     def read_preprocess(self, overwrite_local_cache=False):
         logger.info("Reading preprocessed data")
@@ -247,32 +247,29 @@ def read_preprocess(self, overwrite_local_cache=False):
         for key in self.data_keys:
             filename = p_join(self.preprocess_path, f"{key}.mmap")
             pull_locally(filename, overwrite=overwrite_local_cache)
-            self.data[key] = np.memmap(
-                filename,
-                mode="r",
-                dtype=self.data_types[key],
-            ).reshape(self.data_shapes[key])
+            self.data[key] = np.memmap(filename, mode="r", dtype=self.data_types[key]).reshape(self.data_shapes[key])
+
+        filename = p_join(self.preprocess_path, "props.pkl")
+        pull_locally(filename, overwrite=overwrite_local_cache)
+        with open(filename, "rb") as f:
+            tmp = pkl.load(f)
+            for key in ["name", "subset", "n_atoms"]:
+                x = tmp.pop(key)
+                if len(x) == 2:
+                    self.data[key] = x[0][x[1]]
+                else:
+                    self.data[key] = x
 
         for key in self.data:
             print(f"Loaded {key} with shape {self.data[key].shape}, dtype {self.data[key].dtype}")
 
-        for key in ["name", "subset"]:
-            filename = p_join(self.preprocess_path, f"{key}.npz")
-            pull_locally(filename)
-            self.data[key] = dict()
-            with open(filename, "rb") as f:
-                tmp = np.load(f)
-                for k in tmp:
-                    self.data[key][k] = tmp[k]
-                    print(f"Loaded {key}_{k} with shape {self.data[key][k].shape}, dtype {self.data[key][k].dtype}")
-
     def is_preprocessed(self):
         predicats = [copy_exists(p_join(self.preprocess_path, f"{key}.mmap")) for key in self.data_keys]
-        predicats += [copy_exists(p_join(self.preprocess_path, f"{x}.npz")) for x in ["name", "subset"]]
+        predicats += [copy_exists(p_join(self.preprocess_path, "props.pkl"))]
         return all(predicats)
 
-    def preprocess(self):
-        if not self.is_preprocessed():
+    def preprocess(self, overwrite=False):
+        if overwrite or not self.is_preprocessed():
             entries = self.read_raw_entries()
             res = self.collate_list(entries)
             self.save_preprocess(res)
@@ -305,7 +302,7 @@ def get_ase_atoms(self, idx: int, ext=True):
 
     @requires_package("dscribe")
     @requires_package("datamol")
-    def chemical_space(
+    def soap_descriptors(
         self,
         n_samples: Optional[Union[List[int], int]] = None,
         return_idxs: bool = True,
@@ -350,7 +347,7 @@ def chemical_space(
             idxs = list(range(len(self)))
         elif isinstance(n_samples, int):
             idxs = np.random.choice(len(self), size=n_samples, replace=False)
-        elif isinstance(n_samples, list):
+        else:  # list, set, np.ndarray
             idxs = n_samples
         datum = {}
         r_cut = soap_kwargs.pop("r_cut", 5.0)
@@ -383,7 +380,7 @@ def wrapper(idx):
             entry = self.get_ase_atoms(idx, ext=False)
             return soap.create(entry, centers=entry.positions)
 
-        descr = dm.parallelized(wrapper, idxs, progress=progress, scheduler="threads")
+        descr = dm.parallelized(wrapper, idxs, progress=progress, scheduler="threads", n_jobs=-1)
         datum["soap"] = np.vstack(descr)
         if return_idxs:
             datum["idxs"] = idxs
@@ -392,6 +389,12 @@ def wrapper(idx):
     def __len__(self):
         return self.data["energies"].shape[0]
 
+    def __smiles_converter__(self, x):
+        """util function to convert string to smiles: useful if the smiles is
+        encoded in a different format than its display format
+        """
+        return x
+
     def __getitem__(self, idx: int):
         shift = IsolatedAtomEnergyFactory.max_charge
         p_start, p_end = self.data["position_idx_range"][idx]
@@ -402,8 +405,8 @@ def __getitem__(self, idx: int):
             self.convert_distance(np.array(input[:, -3:], dtype=np.float32)),
             self.convert_energy(np.array(self.data["energies"][idx], dtype=np.float32)),
         )
-        name = self.data["name"]["uniques"][self.data["name"]["inv_indices"][idx]]
-        subset = self.data["subset"]["uniques"][self.data["subset"]["inv_indices"][idx]]
+        name = self.__smiles_converter__(self.data["name"][idx])
+        subset = self.data["subset"][idx]
 
         if "forces" in self.data:
             forces = self.convert_forces(np.array(self.data["forces"][p_start:p_end], dtype=np.float32))

src/openqdc/datasets/comp6.py

@@ -35,8 +35,8 @@ class COMP6(BaseDataset):
         "pbe-d3bj/def2-tzvp",
         "pbe/def2-tzvp",
         "svwn/def2-tzvp",
-        "wb97m-d3bj/def2-tzvp",
-        "wb97m/def2-tzvp",
+        # "wb97m-d3bj/def2-tzvp",
+        # "wb97m/def2-tzvp",
     ]
 
     energy_target_names = [
@@ -47,8 +47,8 @@ class COMP6(BaseDataset):
         "PBE-D3M(BJ):def2-tzvp",
         "PBE:def2-tzvp",
         "SVWN:def2-tzvp",
-        "WB97M-D3(BJ):def2-tzvp",
-        "WB97M:def2-tzvp",
+        # "WB97M-D3(BJ):def2-tzvp",
+        # "WB97M:def2-tzvp",
     ]
 
     __force_methods__ = [
@@ -59,6 +59,12 @@ class COMP6(BaseDataset):
         "Gradient",
     ]
 
+    def __smiles_converter__(self, x):
+        """util function to convert string to smiles: useful if the smiles is
+        encoded in a different format than its display format
+        """
+        return "-".join(x.decode("ascii").split("_")[:-1])
+
     def read_raw_entries(self):
         samples = []
         for subset in ["ani_md", "drugbank", "gdb7_9", "gdb10_13", "s66x8", "tripeptides"]:

src/openqdc/datasets/iso_17.py

@@ -43,6 +43,12 @@ class ISO17(BaseDataset):
     __distance_unit__ = "bohr"  # bohr
     __forces_unit__ = "ev/bohr"
 
+    def __smiles_converter__(self, x):
+        """util function to convert string to smiles: useful if the smiles is
+        encoded in a different format than its display format
+        """
+        return "-".join(x.decode("ascii").split("_")[:-1])
+
     def read_raw_entries(self):
         raw_path = p_join(self.root, "iso_17.h5")
         samples = read_qc_archive_h5(raw_path, "iso_17", self.energy_target_names, self.force_target_names)

src/openqdc/datasets/nabladft.py

@@ -4,38 +4,46 @@
 
 import datamol as dm
 import numpy as np
-from tqdm import tqdm
+import pandas as pd
 
 from openqdc.datasets.base import BaseDataset
+from openqdc.utils.molecule import z_to_formula
 from openqdc.utils.package_utils import requires_package
 
 
-def to_mol(entry) -> Dict[str, np.ndarray]:
+def to_mol(entry, metadata) -> Dict[str, np.ndarray]:
     Z, R, E, F = entry[:4]
     C = np.zeros_like(Z)
+    E[0] = metadata["DFT TOTAL ENERGY"]
 
     res = dict(
         atomic_inputs=np.concatenate((Z[:, None], C[:, None], R), axis=-1).astype(np.float32),
-        name=np.array([""]),
+        name=np.array([metadata["SMILES"]]),
         energies=E[:, None].astype(np.float32),
         forces=F[:, :, None].astype(np.float32),
         n_atoms=np.array([Z.shape[0]], dtype=np.int32),
-        subset=np.array(["nabla"]),
+        subset=np.array([z_to_formula(Z)]),
     )
 
     return res
 
 
 @requires_package("nablaDFT")
-def read_chunk_from_db(raw_path, start_idx, stop_idx, step_size=1000):
+def read_chunk_from_db(raw_path, start_idx, stop_idx, labels, step_size=1000):
     from nablaDFT.dataset import HamiltonianDatabase
 
     print(f"Loading from {start_idx} to {stop_idx}")
     db = HamiltonianDatabase(raw_path)
     idxs = list(np.arange(start_idx, stop_idx))
     n, s = len(idxs), step_size
 
-    samples = [to_mol(entry) for i in tqdm(range(0, n, s)) for entry in db[idxs[i : i + s]]]
+    cursor = db._get_connection().cursor()
+    data_idxs = cursor.execute("""SELECT * FROM dataset_ids WHERE id IN (""" + str(idxs)[1:-1] + ")").fetchall()
+    c_idxs = [tuple(x[1:]) for x in data_idxs]
+
+    samples = [
+        to_mol(entry, labels[c_idxs[i + j]]) for i in range(0, n, s) for j, entry in enumerate(db[idxs[i : i + s]])
+    ]
     return samples
 
 
@@ -68,12 +76,16 @@ class NablaDFT(BaseDataset):
     def read_raw_entries(self):
         from nablaDFT.dataset import HamiltonianDatabase
 
+        label_path = p_join(self.root, "summary.csv")
+        df = pd.read_csv(label_path, usecols=["MOSES id", "CONFORMER id", "SMILES", "DFT TOTAL ENERGY"])
+        labels = df.set_index(keys=["MOSES id", "CONFORMER id"]).to_dict("index")
+
         raw_path = p_join(self.root, "dataset_full.db")
         train = HamiltonianDatabase(raw_path)
         n, c = len(train), 20
         step_size = int(np.ceil(n / os.cpu_count()))
 
-        fn = lambda i: read_chunk_from_db(raw_path, i * step_size, min((i + 1) * step_size, n))
+        fn = lambda i: read_chunk_from_db(raw_path, i * step_size, min((i + 1) * step_size, n), labels=labels)
         samples = dm.parallelized(
             fn, list(range(c)), n_jobs=c, progress=False, scheduler="threads"
         )  # don't use more than 1 job