File size: 4,553 Bytes
15d1f1d | 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 | """The paper's fixed-population ten-column aggregation for offline DXF scores."""
import math
COMPONENTS = {'SV': 'structure', 'VC': 'view', 'GF': 'geometry',
'DF': 'physical', 'AQ': 'annotation', 'LC': 'layout'}
THRESHOLDS = {'structure': 1., 'view': 1., 'geometry': .9,
'physical': .9, 'annotation': .9, 'layout': .85}
def parse_atoms(result, gt):
"""Frozen CAMR atomic checks from unify_annotation_columns_20260924_v1.py."""
if result is None:
return None, "missing_result"
ann = result.get("components", {}).get("annotation", {})
d = ann.get("details", {})
items = d.get("items")
if d.get("scope") != "full_gt" or not isinstance(items, list):
return None, "no_full_gt_items:" + str(ann.get("reason"))
want = sorted((x["id"], x["gt_dimension_index"]) for x in gt["specs"])
got = sorted((x.get("id"), x.get("gt_dimension_index")) for x in items)
if want != got or d.get("n_selected_target") != len(want):
return None, "target_inventory_mismatch:" + str(ann.get("reason"))
if "n_matched_predicted_native" not in d:
return None, "missing_native_matching_counters"
n = len(want)
tp = int(d["n_matched_predicted_native"])
assert n > 0 and 0 <= tp <= n
assert tp == sum(float(x["presence"]) for x in items)
assert n == int(d["n_total_gt_dimension_targets"])
assert len({x["pred_dimension_index"] for x in items if x["pred_dimension_index"] is not None}) == tp
return {"native_target_count": n, "native_matched_count": tp, "all_targets_matched": tp == n,
"matching_policy": d.get("matching_policy")}, "observed"
def make_record(sid, contract, result=None, status='scored', error=None):
eligible = bool(contract['full_gt']['annotation_eligible'])
row = {'id': sid, 'status': status, 'annotation_eligible': eligible,
'components': {k: 0.0 for k in COMPONENTS},
'view_pass': False, 'layout_pass': False, 'camr_complete': False,
'annotation_unknown': False, 'error': error}
if status != 'scored':
if result is not None:
raise ValueError('A failed output must not contribute a scored result')
return row
if result is None:
raise ValueError('A scored record requires a result')
assert result['metric_version'] == 'v5.2-rc5k-final'
benchmark = result['benchmark']
assert benchmark['thresholds'] == THRESHOLDS
for column, key in COMPONENTS.items():
component = result['components'][key]
value = component['score']
row['components'][column] = float(value) if value is not None else 0.0
assert math.isfinite(row['components'][column]) and 0 <= row['components'][column] <= 1
if key in {'view', 'layout'}:
inferred = bool(value >= THRESHOLDS[key]) if component['eligible'] and value is not None else None
assert benchmark[key + '_pass'] is inferred
row[key + '_pass'] = inferred
if eligible:
atoms, reason = parse_atoms(result, contract)
row['annotation_unknown'] = atoms is None
row['annotation_evidence_status'] = reason
row['annotation_atoms'] = atoms
row['camr_complete'] = atoms is not None and atoms['all_targets_matched']
return row
def aggregate(rows):
if not rows:
raise ValueError('Cannot aggregate an empty population')
if len({r['id'] for r in rows}) != len(rows):
raise ValueError('Duplicate sample IDs')
n = len(rows)
eligible = [r for r in rows if r['annotation_eligible']]
scored = sum(r['status'] == 'scored' for r in rows)
camr_count = sum(r['camr_complete'] is True for r in eligible)
metrics = {'SDR': 100 * scored / n}
metrics.update({k: 100 * math.fsum(r['components'][k] for r in rows) / n for k in COMPONENTS})
metrics.update(CAMR=(100 * camr_count / len(eligible) if eligible else None),
CVR=100 * sum(r['view_pass'] is True for r in rows) / n,
LAR=100 * sum(r['layout_pass'] is True for r in rows) / n)
return {'sample_count': n, 'scorable_count': scored,
'camr_denominator': len(eligible), 'camr_confirmed_count': camr_count,
'annotation_unknown_count': sum(r['annotation_unknown'] for r in eligible),
'metrics': metrics, 'scale': '0--100; rate columns are percentages',
'failed_outputs': 'retained in fixed denominators with zero component scores',
'optional_inapplicable_components': 'fixed-column zero filling'}
|