File size: 8,730 Bytes
39ea985 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 | """Verify a Pass A Zarr store: structural checks on every shard, plus a full re-parse comparison
for a random sample of calculations. This is the gate before scaling to the full run.
"""
from __future__ import annotations
import argparse, glob, os, random, sys
import numpy as np
import zarr
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
from omol_parse import parse_archive
from omol_store import (read_calc, inflate_fock, frontier, SCALARS_F8, SCALARS_I, FLAGS,
ATOM_1D, ATOM_SHELL, PAIRS)
M5250 = "/global/cfs/projectdirs/m5250/OMol_elec"
def shard_paths(root, side):
out = []
for depth in ("*", "*/*/*"):
out += glob.glob(os.path.join(root, side, depth, "*.zarr"))
return sorted(set(out))
def close(a, b, tol=0.0):
if a is None or b is None:
return a is None and b is None
a = np.asarray(a, dtype=float)
b = np.asarray(b, dtype=float)
if a.shape != b.shape:
return False
return bool(np.all((np.isnan(a) & np.isnan(b)) | (np.abs(a - b) <= tol)))
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--store", required=True)
ap.add_argument("--sample", type=int, default=30)
ap.add_argument("--seed", type=int, default=7)
args = ap.parse_args()
p1 = shard_paths(args.store, "p1")
p2 = shard_paths(args.store, "p2")
print(f"shards: p1 {len(p1)}, p2 {len(p2)}")
if len(p1) != len(p2):
print(" ERROR: p1 and p2 shard counts differ")
total = 0
problems = []
nbas_all, natm_all, uhf_n, nofock = [], [], 0, 0
for sp in p2:
g = zarr.open_group(sp, mode="r")
n = int(g.attrs["n_calc"])
total += n
si = np.asarray(g["scalar_i"])
icol = {k: j for j, k in enumerate(g.attrs["scalar_i_cols"])}
nbas = si[:, icol["nbas"]]
natm = si[:, icol["n_atoms"]]
nbas_all.append(nbas)
natm_all.append(natm)
for nm, key in (("atom", "atom_offsets"), ("fock_a", "fock_a_offsets"),
("fock_b", "fock_b_offsets"), ("eps_a", "eps_a_offsets")):
o = np.asarray(g[key])
if len(o) != n + 1 or o[0] != 0 or np.any(np.diff(o) < 0):
problems.append(f"{os.path.basename(sp)}: bad {nm} offsets")
if not np.array_equal(np.diff(np.asarray(g["atom_offsets"])), natm):
problems.append(f"{os.path.basename(sp)}: atom offsets != n_atoms")
want = nbas.astype("i8") * (nbas.astype("i8") + 1) // 2
flags = np.asarray(g["flags"])
fcol = {k: j for j, k in enumerate(g.attrs["flag_cols"])}
is_uhf = flags[:, fcol["is_uhf"]].astype(bool)
has_f = flags[:, fcol["has_fock"]].astype(bool)
uhf_n += int(is_uhf.sum())
nofock += int((~has_f).sum())
fa = np.diff(np.asarray(g["fock_a_offsets"]))
if not np.array_equal(fa[has_f], want[has_f]):
problems.append(f"{os.path.basename(sp)}: fock_a length != nbas(nbas+1)/2")
if np.any(fa[~has_f] != 0):
problems.append(f"{os.path.basename(sp)}: fock_a stored where has_fock is false")
fb = np.diff(np.asarray(g["fock_b_offsets"]))
if np.any((fb > 0) != (is_uhf & has_f)):
problems.append(f"{os.path.basename(sp)}: beta Fock presence != is_uhf & has_fock")
if not np.array_equal(fb[is_uhf & has_f], want[is_uhf & has_f]):
problems.append(f"{os.path.basename(sp)}: beta Fock length wrong")
if len(g.attrs["calc_id"]) != n or len(g.attrs["rel_path"]) != n:
problems.append(f"{os.path.basename(sp)}: attrs length != n_calc")
if np.any(~np.asarray(g["flags"])[:, fcol["terminated_normally"]].astype(bool)):
problems.append(f"{os.path.basename(sp)}: some runs not terminated normally")
nbas_all = np.concatenate(nbas_all) if nbas_all else np.zeros(0)
natm_all = np.concatenate(natm_all) if natm_all else np.zeros(0)
print(f"calculations: {total:,} ({uhf_n:,} UHF, {total-uhf_n:,} RHF, "
f"{nofock:,} without a Fock matrix)")
if total:
print(f" nbas min {nbas_all.min()} median {int(np.median(nbas_all))} max {nbas_all.max()}")
print(f" atoms min {natm_all.min()} median {int(np.median(natm_all))} max {natm_all.max()}")
for p in problems[:12]:
print(" STRUCT ERROR:", p)
if not problems:
print(" structural checks passed on every shard")
rng = random.Random(args.seed)
picks = []
for sp in p2:
g = zarr.open_group(sp, mode="r")
n = int(g.attrs["n_calc"])
if n:
picks.append((sp, rng.randrange(n)))
rng.shuffle(picks)
picks = picks[:args.sample]
print(f"\nre-parse comparison on {len(picks)} calculations:")
nbad = 0
for sp, i in picks:
g = zarr.open_group(sp, mode="r")
stored = read_calc(g, i)
rel = stored["rel_path"]
fresh = parse_archive(os.path.join(M5250, rel, "orca.tar.zst"))
fresh["n_atoms"] = len(fresh["elements"])
fresh["homo_a"], fresh["lumo_a"], fresh["gap_a"] = frontier(fresh["eps_a"], fresh["occ_a"])
fresh["homo_b"], fresh["lumo_b"], fresh["gap_b"] = frontier(fresh["eps_b"], fresh["occ_b"])
bad = []
for k in SCALARS_F8:
a, b = stored[k], fresh.get(k)
if b is None:
continue
if not close(a, b, tol=abs(float(b)) * 1e-12 + 1e-12):
bad.append(f"{k}: {a} vs {b}")
for k in SCALARS_I:
b = fresh.get(k)
if b is not None and stored[k] != int(b):
bad.append(f"{k}: {stored[k]} vs {b}")
for k in ("scf_converged", "terminated_normally", "nbo_available", "npa_available"):
if stored[k] != bool(fresh.get(k)):
bad.append(f"{k}: {stored[k]} vs {fresh.get(k)}")
if stored["is_uhf"] != (fresh["hftyp"] == "UHF"):
bad.append("is_uhf mismatch")
if not close(stored["coords"], fresh["coords"], 1e-9):
bad.append("coords differ")
if fresh["forces"] is not None and not close(stored["forces"], fresh["forces"], 1e-12):
bad.append("forces differ")
if fresh["atomic_numbers"] is not None and not np.array_equal(
stored["atomic_numbers"], fresh["atomic_numbers"]):
bad.append("atomic numbers differ")
for k in ATOM_1D:
if fresh.get(k) is not None and not close(stored[k], fresh[k], 1e-9):
bad.append(f"{k} differs")
for k in ATOM_SHELL:
if fresh.get(k) is not None and not close(stored[k], fresh[k], 2e-5):
bad.append(f"{k} differs")
for k in PAIRS:
idx, val = stored[k]
ref = fresh.get(k) or []
if len(val) != len(ref):
bad.append(f"{k} count {len(val)} vs {len(ref)}")
elif ref:
if (not np.array_equal(idx, np.array([[a, b] for a, b, _ in ref]))
or not close(val, np.array([v for _, _, v in ref]), 1e-4)):
bad.append(f"{k} content differs")
if fresh["fock_a"] is None:
if stored["has_fock"] or len(stored["fock_a"]):
bad.append("fock_a stored but source has none")
elif not np.array_equal(stored["fock_a"], fresh["fock_a"]):
bad.append("fock_a differs")
fb = fresh.get("fock_b")
if fb is None:
if len(stored["fock_b"]):
bad.append("fock_b present but should be absent")
elif not np.array_equal(stored["fock_b"], fb):
bad.append("fock_b differs")
if not close(stored["eps_a"], fresh["eps_a"], 1e-12):
bad.append("eps_a differs")
if stored["has_fock"]:
F = inflate_fock(stored["fock_a"], stored["nbas"])
if np.abs(F - F.T).max() != 0:
bad.append("inflated Fock not symmetric")
if bad:
nbad += 1
print(f" MISMATCH {rel}")
for b in bad[:6]:
print(f" {b}")
else:
print(f" ok {rel[:76]}")
print(f"\n{len(picks)-nbad}/{len(picks)} exact, {len(problems)} structural problems")
def du(p):
return sum(os.path.getsize(os.path.join(r, f)) for r, _, fs in os.walk(p) for f in fs)
b1 = sum(du(p) for p in p1)
b2 = sum(du(p) for p in p2)
print(f"\nsize: p1 {b1/1e9:.3f} GB ({b1/max(total,1)/1e3:.1f} kB/calc), "
f"p2 {b2/1e9:.3f} GB ({b2/max(total,1)/1e6:.3f} MB/calc)")
if total:
print(f"extrapolated to 3.73 M: p1 {b1/total*3.73e6/1e12:.2f} TB, "
f"p2 {b2/total*3.73e6/1e12:.2f} TB")
if __name__ == "__main__":
main()
|