| """Last-ionic-step forces/stress for replacement tasks whose output.forces/stress are empty. |
| |
| MP's task documents often leave output.forces/stress empty for relaxations; the values are |
| in calcs_reversed[0].output.ionic_steps[-1]. Kept only if that step is the output: same |
| species order, lattice and positions to 1e-4 A, and an energy term equal to the output |
| energy to 1e-5 eV (as in scripts/recover_mp_forces.py).""" |
| import sys, json, re, gzip, time, numpy as np, pandas as pd, requests |
| from pathlib import Path |
| from concurrent.futures import ProcessPoolExecutor, as_completed |
| W=Path(sys.argv[1]); OUT=W/"s3_steps"; OUT.mkdir(exist_ok=True) |
| MAN=Path("/home/kna/.local/share/wyformer/cache/mp_forces_recovery") |
| S3="https://materialsproject-parsed.s3.amazonaws.com/" |
| TASK_ID_RE=re.compile(rb'"task_id":\s*"([a-z]+-\d+)"') |
| TOL_X, TOL_E = 1e-4, 1e-5 |
|
|
| def xyz(s): |
| lat=np.asarray(s["lattice"]["matrix"],float) |
| return lat,[x["species"][0]["element"] for x in s["sites"]],np.array([x["xyz"] if "xyz" in x else np.dot(x["abc"],lat) for x in s["sites"]],float) |
|
|
| def extract(doc): |
| o=doc.get("output") or {}; c=(doc.get("calcs_reversed") or [{}])[0] |
| steps=(c.get("output") or {}).get("ionic_steps") or [] |
| rec={"task_id":doc["task_id"],"ok":False} |
| if not steps or not o.get("structure"): rec["why"]="no ionic steps"; return rec |
| st=steps[-1] |
| if not st.get("forces") or not st.get("stress") or not st.get("structure"): rec["why"]="step lacks forces/stress"; return rec |
| la,sa,xa=xyz(o["structure"]); lb,sb,xb=xyz(st["structure"]) |
| if sa!=sb: rec["why"]="species order differs"; return rec |
| fr=(xb-xa)@np.linalg.inv(la); fr-=np.round(fr) |
| dpos=float(np.linalg.norm(fr@la,axis=1).max()); dlat=float(np.abs(lb-la).max()) |
| dE=min(abs(st[k]-o["energy"]) for k in ("e_fr_energy","e_wo_entrp","e_0_energy") if isinstance(st.get(k),(int,float))) |
| rec.update(dpos=dpos,dlat=dlat,dE=dE) |
| if dpos>TOL_X or dlat>TOL_X or dE>TOL_E: rec["why"]="last step is not the output"; return rec |
| rec.update(ok=True,forces=st["forces"],stress=st["stress"]) |
| return rec |
|
|
| def scan(key, needed): |
| wanted=set(needed); got=[] |
| for a in range(4): |
| try: |
| with requests.get(S3+key,stream=True,timeout=900) as resp: |
| resp.raise_for_status() |
| with gzip.GzipFile(fileobj=resp.raw) as fh: |
| for line in fh: |
| if not {m.decode() for m in TASK_ID_RE.findall(line)} & wanted: continue |
| d=json.loads(line) |
| if d.get("task_id") in wanted: |
| got.append(extract(d)); wanted.discard(d["task_id"]) |
| if not wanted: break |
| return got, sorted(wanted) |
| except Exception: time.sleep(5*(a+1)) |
| return got, sorted(wanted) |
|
|
| if __name__=="__main__": |
| need=set(open(W/"need_steps.txt").read().split()) |
| a2=pd.read_parquet(MAN/"manifest_atomate2.parquet",columns=["task_id","sourced_from_path"]); a2=a2[a2.task_id.isin(need)] |
| leg=pd.read_parquet(MAN/"manifest_legacy.parquet",columns=["task_id","nelements","spacegroup","dt"]); leg=leg[leg.task_id.isin(need)] |
| keys={} |
| |
| for t,k in zip(a2.task_id,a2.sourced_from_path.str.replace("s3://materialsproject-parsed/","",regex=False)): keys.setdefault(k,[]).append(t) |
| for r in leg[~leg.task_id.isin(set(a2.task_id))].itertuples(): |
| keys.setdefault(f"tasks-legacy/tasks/nelements={r.nelements}/output_spacegroup_number={r.spacegroup}/dt={r.dt}.jsonl.gz",[]).append(r.task_id) |
| print("tasks",len(need),"files",len(keys),flush=True) |
| out=open(OUT/"steps.jsonl","w"); missing=[] |
| with ProcessPoolExecutor(12) as ex: |
| futs=[ex.submit(scan,k,v) for k,v in keys.items()] |
| for i,fu in enumerate(as_completed(futs),1): |
| got,left=fu.result(); missing+=left |
| for g in got: out.write(json.dumps(g)+"\n") |
| if i%100==0: print("files",i,len(futs),flush=True) |
| out.close(); print("DONE missing",len(missing),flush=True) |
|
|
