moledit-benchmark / code /validate_with_executor.py
MOLYHECI's picture
Upload folder using huggingface_hub
ae9909c verified
"""
Validate pubchem_dataset records using ActionExecutor.
Translates old-format operation dicts (raw atom_idx params) to the new
ActionExecutor JSON protocol, then checks that executing the action on
wrong_smiles reproduces correct_smiles.
Usage:
python validate_with_executor.py [--dataset ./data/pubchem_dataset] [--verbose]
"""
import os, sys, json, argparse
from collections import Counter, defaultdict
sys.path.insert(0, os.path.dirname(__file__))
from rdkit import Chem, RDLogger
RDLogger.DisableLog("rdApp.*")
from action_executor import ActionExecutor, ActionError
# ---------------------------------------------------------------------------
# Old-format → ActionExecutor JSON translation
# ---------------------------------------------------------------------------
def translate_operation(op: dict) -> dict:
"""
Convert a mol_corrupt.py correction dict to an ActionExecutor action dict.
New format (atom_map / anchor_map / atom1_map / atom2_map) uses map_num refs.
Old format (atom_idx / anchor_idx / atom_idx1 / atom_idx2) uses idx refs.
Both are handled transparently.
"""
old_type = op["type"]
params = op.get("params", {})
# Determine ref style: map-num (new) or raw idx (old)
has_maps = any(k in params for k in ("atom_map", "anchor_map", "atom1_map", "atom2_map"))
def atom_ref(map_key, idx_key):
if has_maps and map_key in params:
return {"map_num": params[map_key]}
return {"idx": params[idx_key]}
def bond_ref(map1, idx1, map2, idx2):
return {"atom1": atom_ref(map1, idx1), "atom2": atom_ref(map2, idx2)}
if old_type == "change_atom_element":
return {
"type": "ChangeAtom",
"atom": atom_ref("atom_map", "atom_idx"),
"element": params["to_elem"],
}
if old_type == "change_charge":
return {
"type": "ChangeAtom",
"atom": atom_ref("atom_map", "atom_idx"),
"charge": params["to_charge"],
}
if old_type == "add_atom":
return {
"type": "AddAtom",
"anchor": atom_ref("anchor_map", "anchor_idx"),
"element": params["new_elem"],
"bond_order": params.get("bond_type", "SINGLE"),
}
if old_type == "remove_atom":
return {
"type": "RemoveAtom",
"atom": atom_ref("atom_map", "atom_idx"),
}
if old_type == "change_bond_order":
return {
"type": "ChangeBond",
"bond": bond_ref("atom1_map", "atom_idx1", "atom2_map", "atom_idx2"),
"order": params["to_order"],
}
if old_type == "add_bond":
return {
"type": "AddBond",
"bond": bond_ref("atom1_map", "atom_idx1", "atom2_map", "atom_idx2"),
"order": params.get("bond_type", "SINGLE"),
}
if old_type == "remove_bond":
return {
"type": "RemoveBond",
"bond": bond_ref("atom1_map", "atom_idx1", "atom2_map", "atom_idx2"),
}
if old_type == "add_functional_group":
return {
"type": "AddGroup",
"anchor": atom_ref("anchor_map", "anchor_idx"),
"group": params["group_name"],
}
if old_type == "remove_functional_group":
return {
"type": "RemoveGroup",
"anchor": atom_ref("anchor_map", "anchor_idx"),
"group": params["group_name"],
}
if old_type == "flip_chirality":
return {
"type": "FlipChirality",
"atom": atom_ref("atom_map", "atom_idx"),
}
if old_type == "flip_ez":
# SMILES-level op — find the double bond automatically
return {"type": "FlipEZ"}
if old_type == "move_substituent":
return {
"type": "MoveSubstituent",
"substituent": {"map_num": params["substituent_map"]},
"from_atom": {"map_num": params["from_map"]},
"to_atom": {"map_num": params["to_map"]},
}
if old_type == "swap_substituents":
return {
"type": "SwapSubstituents",
"atom1": {"map_num": params["ring1_map"]},
"atom2": {"map_num": params["ring2_map"]},
}
raise ValueError(f"Unhandled old type: {old_type!r}")
# ---------------------------------------------------------------------------
# Validation
# ---------------------------------------------------------------------------
def validate(records, verbose=False):
ex = ActionExecutor()
results = defaultdict(lambda: {"pass": 0, "fail": 0, "error": 0})
failures = []
for r in records:
op_type = r["operation"]["type"]
# Prefer mapped SMILES (new format) so map_num refs work correctly
wrong = r.get("wrong_smiles_mapped") or r.get("wrong_smiles", "")
correct = r["correct_smiles"]
try:
action = translate_operation(r["operation"])
except Exception as e:
results[op_type]["error"] += 1
failures.append({"id": r.get("id"), "op": op_type, "kind": "translate_error", "msg": str(e)})
continue
try:
restored = ex.execute(wrong, action)
except ActionError as e:
results[op_type]["error"] += 1
failures.append({
"id": r.get("id"), "op": op_type, "kind": "executor_error",
"msg": str(e), "wrong": wrong, "action": action,
})
continue
except Exception as e:
results[op_type]["error"] += 1
failures.append({
"id": r.get("id"), "op": op_type, "kind": "unexpected_error",
"msg": str(e), "wrong": wrong,
})
continue
# Canonicalize both for fair comparison
mol_r = Chem.MolFromSmiles(restored) if restored else None
mol_c = Chem.MolFromSmiles(correct)
canon_r = Chem.MolToSmiles(mol_r) if mol_r else None
canon_c = Chem.MolToSmiles(mol_c) if mol_c else None
if canon_r == canon_c:
results[op_type]["pass"] += 1
else:
results[op_type]["fail"] += 1
failures.append({
"id": r.get("id"), "op": op_type, "kind": "mismatch",
"wrong": wrong, "expected": correct, "got": restored,
"action": action,
})
return results, failures
# ---------------------------------------------------------------------------
# Report
# ---------------------------------------------------------------------------
def report(results, failures, verbose=False):
print("=" * 76)
print(f"{'Op Type':<35} {'Total':>6} {'Pass':>5} {'Fail':>5} {'Error':>6} {'Rate':>6}")
print("-" * 76)
total_p = total_f = total_e = 0
for op in sorted(results):
r = results[op]
p, f, e = r["pass"], r["fail"], r["error"]
n = p + f + e
rate = p / n * 100 if n else 0
mark = "" if (f + e) == 0 else " ✗"
print(f"{op:<35} {n:>6} {p:>5} {f:>5} {e:>6} {rate:>5.0f}%{mark}")
total_p += p; total_f += f; total_e += e
total = total_p + total_f + total_e
print("-" * 76)
rate = total_p / total * 100 if total else 0
print(f"{'TOTAL':<35} {total:>6} {total_p:>5} {total_f:>5} {total_e:>6} {rate:>5.0f}%")
print("=" * 76)
if failures:
show = failures if verbose else failures[:5]
print(f"\n--- {len(failures)} failure(s){' (showing first 5)' if not verbose and len(failures) > 5 else ''} ---")
for fl in show:
print(f"\n [{fl['op']}] id={fl.get('id')} kind={fl['kind']}")
if fl["kind"] == "mismatch":
print(f" wrong : {fl['wrong']}")
print(f" action : {fl['action']}")
print(f" expected : {fl['expected']}")
print(f" got : {fl['got']}")
elif fl["kind"] in ("executor_error", "translate_error", "unexpected_error"):
print(f" wrong : {fl.get('wrong', '')}")
print(f" action : {fl.get('action', '')}")
print(f" error : {fl['msg']}")
else:
print("\n✓ All records pass validation.")
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--dataset", default="./data/pubchem_dataset")
parser.add_argument("--verbose", action="store_true", help="Show all failures")
args = parser.parse_args()
for name in ("dataset.json", "roundtrip_test.json"):
p = os.path.join(args.dataset, name)
if os.path.exists(p):
json_path = p
break
else:
print(f"No dataset JSON found in {args.dataset}")
sys.exit(1)
with open(json_path) as f:
records = json.load(f)
print(f"Loaded {len(records)} records from {json_path}\n")
results, failures = validate(records, verbose=args.verbose)
report(results, failures, verbose=args.verbose)
if __name__ == "__main__":
main()