| """Shortest interatomic distance of each replacement structure vs LeMat's original.""" |
| import sys, json, numpy as np, pandas as pd |
| from multiprocessing import Pool |
| from pymatgen.core import Structure, Lattice |
| W=sys.argv[1] |
| L=pd.read_parquet(W+"/lemat_review_rows.parquet",columns=["immutable_id","lattice_vectors","cartesian_site_positions","species_at_sites"]).drop_duplicates("immutable_id").set_index("immutable_id") |
| reps=[json.loads(l) for l in open(W+"/replacements.jsonl")] |
| reps=[p for p in reps if p["action"] in ("replace_structure","replace_energy")] |
| def mind(s): |
| dm=s.distance_matrix+np.eye(len(s))*99 |
| return float(dm.min()) if len(s)>1 else float(min(s.lattice.abc)) |
| def job(p): |
| l=L.loc[p["immutable_id"]] |
| a=Structure(Lattice(np.stack(l.lattice_vectors)),list(l.species_at_sites),np.stack(l.cartesian_site_positions),coords_are_cartesian=True) |
| b=Structure(Lattice(p["lattice"]),p["species"],p["xyz"],coords_are_cartesian=True) |
| return p["immutable_id"],p["action"],mind(a),mind(b) |
| with Pool(16) as pool: r=pd.DataFrame(pool.map(job,reps,chunksize=50),columns=["immutable_id","action","dmin_old","dmin_new"]) |
| r["ratio"]=r.dmin_new/r.dmin_old |
| r.to_parquet(W+"/contacts.parquet") |
| for a,g in r.groupby("action"): |
| print(a,len(g),"ratio<0.9:",(g.ratio<0.9).sum(),"<0.8:",(g.ratio<0.8).sum(),"new dmin<1.0A (non-H):",(g.dmin_new<1.0).sum(), "ratio q01:",round(g.ratio.quantile(0.01),3)) |
| print(r[r.ratio<0.9].sort_values("ratio").head(15).to_string()) |
|
|
