File size: 3,912 Bytes
515fb62 | 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
"""Answer audit: re-derive question answers with 2x integration resolution.
The shipped answers come from the generation sim (8 substeps/frame = 96 Hz).
This script re-simulates sampled videos at 16 substeps/frame (192 Hz) and checks
that every question's answer is unchanged. Any mismatch indicates numerical or
conceptual fragility.
Usage: python audit_answers.py [dataset_root] [--n 100]
"""
import argparse
import csv
import json
import sys
from collections import Counter
from pathlib import Path
import numpy as np
sys.path.insert(0, str(Path(__file__).parent))
import pendulum_gen as G
from questions import build_questions
def resim(cfg):
rows = G.simulate(cfg, frames=G.EXT_FRAMES, substeps=16)
return G.rows_to_np(rows)
def cfg_from_row(r):
cfg = G.Configuration(name=r["name"])
cfg.theta0_deg = float(r["theta0_deg"])
cfg.omega0_rad_s = float(r["omega0_rad_s"])
cfg.damping_s_inv = float(r["damping_s_inv"])
cfg.light_mean_deg = float(r["light_mean_deg"])
cfg.light_amplitude_deg = float(r["light_amplitude_deg"])
cfg.light_frequency_hz = float(r["light_frequency_hz"])
cfg.light_phase_rad = float(r["light_phase_rad"])
cfg.gravity_m_s2 = float(r["gravity_m_s2"])
cfg.length_m = float(r["length_m"])
cfg.damping_model = r["damping_model"]
cfg.quad_drag_beta = float(r["quad_drag_beta"])
cfg.drive_amp_rad_s2 = float(r["drive_amp_rad_s2"])
cfg.drive_freq_hz = float(r["drive_freq_hz"])
cfg.drive_phase_rad = float(r["drive_phase_rad"])
return cfg
def main():
ap = argparse.ArgumentParser()
ap.add_argument("root", nargs="?", default="dataset")
ap.add_argument("--n", type=int, default=100)
ap.add_argument("--seed", type=int, default=20261014)
args = ap.parse_args()
root = Path(args.root)
idx = list(csv.DictReader(open(root / "index.csv")))
qs = {}
for split in ("train", "val"):
for e in json.load(open(root / "questions" / f"questions_{split}.json"))["scene_annotation"]:
qs[e["video_id"]] = e["questions"]
rng = np.random.default_rng(0)
n_train = args.n // 2
n_val = args.n - n_train
pool = ([r for r in idx if r["split"] == "train"], [r for r in idx if r["split"] == "val"])
sample = [pool[0][i] for i in rng.permutation(len(pool[0]))[:n_train]] + \
[pool[1][i] for i in rng.permutation(len(pool[1]))[:n_val]]
mismatches = Counter()
n_checked = 0
for r in sample:
vid = r["video_id"]
index_num = int(vid.split("_")[1]) if r["split"] == "train" else \
100000 + (int(vid.split("_")[1]) - 10000)
cfg = cfg_from_row(r)
A_fine = resim(cfg)
q_rng = np.random.default_rng(np.random.SeedSequence([args.seed, index_num, 777]))
rebuilt = build_questions(cfg, A_fine, q_rng)
orig = {q["question_id"]: q for q in qs[vid]}
for q2 in rebuilt:
q1 = orig.get(q2["question_id"])
if q1 is None or q1["question"] != q2["question"]:
continue
n_checked += 1
a1, a2 = q1["answer"], q2["answer"]
if "choices" in q1:
ok = q1["choices"][a1] == q2["choices"][a2]
else:
ok = str(a1) == str(a2)
if not ok:
mismatches[q2["concept"] or q2["question_type"]] += 1
print(f"MISMATCH {vid} [{q2['question_type']}] {q2['question']}\n"
f" shipped: {a1 if 'choices' not in q1 else q1['choices'][a1]}\n"
f" 192Hz : {a2 if 'choices' not in q2 else q2['choices'][a2]}")
print(f"\nchecked {n_checked} questions on {len(sample)} videos "
f"({n_train} train / {n_val} val)")
print(f"mismatches: {sum(mismatches.values())} {dict(mismatches)}")
sys.exit(0 if not mismatches else 1)
if __name__ == "__main__":
main()
|