File size: 4,836 Bytes
b47994e | 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 | #!/usr/bin/env python3
"""Regenerate approved-questions/POOL_STATS.txt.
Reports, for the accepted archive (bank/), the frozen benchmark (benchmark_600/)
and the reserve (leftover_pool/): counts, GT distributions, object diversity,
and the per-type signature reuse that drove the selection.
Run from repo root: python approved-questions/scripts/gen_pool_stats.py
"""
import json, glob, os
from collections import Counter, defaultdict
ROOT=os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
AQ=os.path.join(ROOT,"approved-questions")
def qtype(q):
t=q.get("question_type")
return {"marked_comparison":"compare","counting":"count",
"existence":"find","under_object":"beneath"}.get(t,t)
def scene_folders(ep_dir):
sp=os.path.join(ep_dir,"scene_snapshot","scene_snapshot.json"); s=set()
if os.path.isfile(sp):
for o in (json.load(open(sp)).get("objects") or []):
if o.get("folder"): s.add(o["folder"])
return s
def name_folder(ep_dir):
sp=os.path.join(ep_dir,"scene_snapshot","scene_snapshot.json"); m={}
if os.path.isfile(sp):
for o in (json.load(open(sp)).get("objects") or []):
if o.get("name") and o.get("folder"): m[o["name"]]=o["folder"]
return m
def analyze(bank_dir, label, out):
rows=defaultdict(list)
for rj in sorted(glob.glob(os.path.join(bank_dir,"ep_*/review.json"))):
ep=os.path.dirname(rj); d=json.load(open(rj)); q=d.get("question") or {}
ct=qtype(q); fmap=name_folder(ep); scene=scene_folders(ep)
r={"scene":scene}
if ct=="count":
r["gt"]=str(q.get("gt_answer")); r["sig"]=q.get("target_ref_folder")
r["pol"]="no" if r["gt"]=="0" else "yes"
elif ct=="find":
yes=str(q.get("gt_answer_text") or "").lower().startswith("y")
r["gt"]="yes" if yes else "no"; r["sig"]=q.get("queried_ref_folder")
r["pol"]=r["gt"]
elif ct=="beneath":
r["gt"]=str(q.get("gt_answer"))
r["sig"]=(fmap.get(q.get("cover_object")), fmap.get(q.get("target_name")))
elif ct=="compare":
r["gt"]=str(q.get("gt_answer")); r["ext"]=q.get("comparison_extremum")
ml=q.get("marked_labels") or {}
r["sig"]=tuple(sorted(v.get("obj_folder") for v in ml.values() if v.get("obj_folder")))
rows[ct].append(r)
out.append("="*72); out.append(f"{label}"); out.append("="*72)
allf=set(); total=0
for ct in ["count","find","beneath","compare"]:
rs=rows[ct]; total+=len(rs)
for r in rs: allf|=r["scene"]
sigs=Counter(r["sig"] for r in rs if r.get("sig"))
reuse=Counter(sigs.values()); maxr=max(sigs.values()) if sigs else 0
line=f" {ct:8s} n={len(rs):<4d} distinct-signatures={len(sigs):<4d} "\
f"max-reuse={maxr} reuse-hist={dict(sorted(reuse.items()))}"
out.append(line)
if ct=="count":
out.append(f" GT: {dict(sorted(Counter(r['gt'] for r in rs).items()))}"
f" (yes={sum(1 for r in rs if r['pol']=='yes')}"
f" no={sum(1 for r in rs if r['pol']=='no')})")
elif ct=="find":
out.append(f" yes/no: {dict(Counter(r['gt'] for r in rs))}")
elif ct=="compare":
out.append(f" letter: {dict(sorted(Counter(r['gt'] for r in rs).items()))}"
f" extremum: {dict(Counter(r['ext'] for r in rs))}")
elif ct=="beneath":
gt=Counter(r["gt"] for r in rs)
out.append(f" distinct-GT={len(gt)} NOTA={gt.get('none of the above',0)}")
out.append(f" TOTAL n={total} UNIQUE object folders (scene union)={len(allf)}")
out.append("")
return allf
def main():
out=[]
out.append("PROBE-Bench V2 — approved-questions pool statistics")
out.append("(regenerate: python approved-questions/scripts/gen_pool_stats.py)")
out.append("")
f1=analyze(os.path.join(AQ,"benchmark_600"), "FROZEN BENCHMARK — benchmark_600/ (the deliverable)", out)
f2=analyze(os.path.join(AQ,"leftover_pool"), "RESERVE — leftover_pool/ (swap-in if questions removed)", out)
f3=analyze(os.path.join(AQ,"bank"), "FULL ACCEPTED ARCHIVE — bank/ (783, everything accepted)", out)
out.append("="*72)
out.append("CROSS-BANK OBJECT COVERAGE")
out.append("="*72)
out.append(f" benchmark_600 unique objects : {len(f1)}")
out.append(f" leftover_pool unique objects : {len(f2)}")
out.append(f" full archive unique objects : {len(f3)}")
out.append(f" benchmark captures {100*len(f1)//max(len(f3),1)}% of the archive's object variety")
txt="\n".join(out)+"\n"
open(os.path.join(AQ,"POOL_STATS.txt"),"w").write(txt)
print(txt)
if __name__=="__main__":
main()
|