Datasets:
File size: 4,914 Bytes
84a242a | 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 | #!/usr/bin/env python3
"""Cross-check final acquisition artifacts with existing inputs and fixed folds.
Checks existing data only: no docking, acquisition loop, completion, or fitting.
The original CSV exporter used %.8g; comparisons respect that serialization.
"""
import argparse
from concurrent.futures import ThreadPoolExecutor
import csv
import io
import json
import math
from pathlib import Path
import tarfile
ALGORITHMS = ("surfdock", "unimol", "diffdock", "diffdock_l", "gnina", "autodock", "smina")
def rows(raw):
return list(csv.DictReader(io.StringIO(raw.decode())))
def main():
p = argparse.ArgumentParser(description=__doc__)
p.add_argument("--root", type=Path, default=Path(__file__).resolve().parents[1])
p.add_argument("--report", type=Path)
args = p.parse_args()
root = args.root
truth_rows = rows((root / "data/targets/full_strict_score_matrix.csv").read_bytes())
cost_rows = rows((root / "data/raw/moad_runtime_trimmed.csv").read_bytes())
truth = {r["PDB_CCD_ID"]: {a: float(r[a + "_strict_score"]) for a in ALGORITHMS} for r in truth_rows}
costs = {r["PDB_CCD_ID"]: {a: float(r[a + "_runtime"]) for a in ALGORITHMS} for r in cost_rows}
split = rows((root / "data/splits/moad_5fold_split_seed42.csv").read_bytes())
folds = {r["instance_id"]: int(r["test_fold"]) for r in split}
archives = sorted((root / "results/acquisition").glob("*/fold*_seed*.tar.xz"))
def compatible(actual, expected):
return math.isclose(float(actual), float("%.8g" % expected), rel_tol=1e-12, abs_tol=1e-14)
def check(path):
fold = int(path.name.split("_")[0][4:])
seed = int(path.name.split("_")[1].split(".")[0][4:])
expected = {x for x, f in folds.items() if f != fold}
wanted = {"strict_score_matrices/strict_score_100pct.csv", "queried_cells/cumulative/queried_100pct.csv"}
payload = {}
with tarfile.open(path, "r:xz") as tf:
for item in tf:
if item.name in wanted:
payload[item.name] = tf.extractfile(item).read()
matrix = rows(payload["strict_score_matrices/strict_score_100pct.csv"])
ledger = rows(payload["queried_cells/cumulative/queried_100pct.csv"])
ids = [r["PDB_CCD_ID"] for r in matrix]
if len(ids) != len(expected) or set(ids) != expected:
raise ValueError("Matrix training-fold membership mismatch: " + str(path))
for r in matrix:
for a in ALGORITHMS:
if not compatible(r[a], truth[r["PDB_CCD_ID"]][a]):
raise ValueError("Full matrix differs from observed strict targets: " + str(path))
n = len(expected) * len(ALGORITHMS)
cells = [(r["row_label"], r["algorithm"]) for r in ledger]
if len(cells) != n or len(set(cells)) != n or [int(r["query_order"]) for r in ledger] != list(range(1, n + 1)):
raise ValueError("Final ledger coverage/order mismatch: " + str(path))
warm = []
for r in ledger:
identity, a = r["row_label"], r["algorithm"]
if identity not in expected or a not in ALGORITHMS:
raise ValueError("Held-out or invalid cell in acquisition ledger")
if not compatible(r["score"], truth[identity][a]) or not compatible(r["runtime"], costs[identity][a]):
raise ValueError("Recorded outcome/runtime differs from raw inputs: " + str(path))
if int(r["round"]) == 0:
warm.append((identity, a))
return {"archive": str(path.relative_to(root)), "fold": fold, "seed": seed,
"training_complexes": len(ids), "observed_cells": n, "warm_start": warm}
with ThreadPoolExecutor(max_workers=4) as ex:
checked = list(ex.map(check, archives))
signatures = {}
for r in checked:
key = (r["fold"], r["seed"])
warm = r.pop("warm_start")
r["warm_start_cells"] = len(warm)
if key in signatures and signatures[key] != warm:
raise ValueError("Warm starts differ between policies for fold/seed " + str(key))
signatures[key] = warm
if len(checked) != 90:
raise ValueError("Expected 90 runs")
report = {"status": "pass", "runs_checked": len(checked),
"full_matrices_match_existing_strict_targets": True,
"final_ledgers_cover_training_cells_exactly_once": True,
"held_out_cells_in_acquisition_ledgers": 0,
"recorded_scores_and_runtimes_match_inputs_at_export_precision": True,
"warm_starts_match_across_six_policies_for_every_fold_seed": True,
"new_experiments_executed": False, "details": checked}
if args.report:
args.report.write_text(json.dumps(report, indent=2) + "\n")
print(json.dumps({k: v for k, v in report.items() if k != "details"}, indent=2))
if __name__ == "__main__":
main()
|