add preprocess scripts

.gitignore

@@ -0,0 +1,4 @@
+**/.ipynb_checkpoints/
+**/__pycache__
+**.out
+**.err

jobs.py

@@ -0,0 +1,37 @@
+"""Module for submitting jobs"""
+
+import subprocess
+
+
+def submit_slurm(
+    cmd: str,
+    time: str,
+    partition: str,
+    nodes: int,
+    ntasks_per_node: int,
+    job_name: str,
+    account: str,
+    **kwargs,
+):
+    """Submit a job to the slurm scheduler."""
+    scmd = [
+        *f"""sbatch
+        --time={time}
+        --account={account}
+        --qos={partition}
+        --job-name={job_name}
+        --output=%x-%j.out
+        --error=%x-%j.err
+        --nodes={nodes}
+        --ntasks-per-node={ntasks_per_node}""".replace("'", "").split(),
+    ]
+
+    if kwargs.get("constraint", False):
+        scmd += [f"--constraint={kwargs['constraint']}"]
+
+    if kwargs.get("exclude", False):
+        scmd += ["--exclude"]
+
+    scmd += ["--wrap", cmd]
+
+    return subprocess.run(scmd, check=True)

preprocess.py

@@ -0,0 +1,84 @@
+# # %%
+# from datasets import load_dataset
+
+# dataset = load_dataset("atomind/alexandria", name='1D', streaming=True)
+
+# print(dataset)
+# # %%
+import argparse
+import bz2
+import json
+from pathlib import Path
+
+from tqdm.auto import tqdm
+
+from ase import Atoms
+from ase.calculators.singlepoint import SinglePointCalculator
+from ase.io import write, read
+from pymatgen.core import Structure
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--src_path", type=str, required=True)
+    parser.add_argument("--dst_dir", type=str, required=True)
+    return parser.parse_args()
+
+
+def main(src_path: Path | str, dst_dir: Path | str):
+    """Extracts the structures from a bz2 compressed json file and writes them to an extended xyz file."""
+
+    if isinstance(src_path, str):
+        src_path = Path(src_path)
+
+    if isinstance(dst_dir, str):
+        dst_dir = Path(dst_dir)
+
+    dst_dir.mkdir(exist_ok=True, parents=True)
+
+    with bz2.open(src_path, "rb") as f:
+        data = json.load(f)
+
+    assert isinstance(data, dict)
+
+    for alex_id, u in tqdm(data.items(), desc="Extracting structures"):
+        for calc_id, v in enumerate(u):
+            for ionic_step, w in enumerate(v["steps"]):
+                atoms = Structure.from_dict(w["structure"]).to_ase_atoms()
+
+                results = {
+                    "energy": w["energy"],
+                    "forces": w["forces"],
+                    "stress": w["stress"],
+                }
+
+                atoms.calc = SinglePointCalculator(atoms=atoms, **results)
+
+                atoms.info = {
+                    "alex_id": alex_id,
+                    "calc_id": calc_id,
+                    "ionic_step": ionic_step,
+                }
+
+                traj_file = dst_dir / f"{atoms.get_chemical_formula()}.traj"
+
+                exist = False
+                if traj_file.exists():
+                    traj = read(traj_file, index=":")
+                    for frame in traj:
+                        assert isinstance(frame, Atoms)
+                        if frame.info == atoms.info:
+                            exist = True
+                            break
+
+                if not exist:
+                    write(
+                        traj_file,
+                        atoms,
+                        append=True,
+                    )
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args.src_path, args.dst_dir)