Trusted Polytopic Optimizer Steps / report_bench_2026-09-01T194315.md
Mechanism confirmed, baseline not beaten
Стенд-проверка (stage-2) · промт оператора:
(универсальный)
Ответ агента:
{ "worked": false, "confidence": 9, "verdict": "The idea was implemented and evaluated on the registered tabular Friedman regression track with the shared mlp_tiny architecture and eight paired seeds. Tuned Adam achieved test MSE 7.1906, while the best trusted-polytopic configuration achieved 14.9916; paired delta was +7.8010 with permutation p=0.0081, so the idea was significantly worse. The trained-model mechanism signature nevertheless confirmed the predicted radius-to-nonlinearity relationship.", "metrics": { "baseline": "Adam; best configuration lr=0.01, weight_decay=0.0001; mean test MSE 7.1905966401", "idea": "Trusted polytopic optimizer; best configuration lr=0.01, radius=0.1; mean test MSE 14.9916086197", "paired_delta": 7.8010119795799255, "permutation_p_value": 0.0081, "mechanism_loglog_slope": 1.9431904882117401, "mechanism_confirmed": true }, "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_bench.py", "files": [ "stage2_bench.py", "bench_report.json", "bench_run.txt" ], "limitations": "Only the tabular track, mlp_tiny model, 10 epochs, 1200 training samples, and two optimizer directions were tested. The implementation uses one-step finite-difference probes rather than multi-step rollouts or exact JVPs, and no speed or FLOP advantage was demonstrated.", "bench_report": { "bench_version": 1, "track": "tabular", "model": "mlp_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.01, "weight_decay": 0.0001 }, "sweep": [ { "cfg": { "lr": 0.001, "weight_decay": 0.0 }, "mean": 197.47264099121094 }, { "cfg": { "lr": 0.001, "weight_decay": 0.0001 }, "mean": 197.46313858032227 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 31.18530511856079 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001 }, "mean": 31.172401428222656 }, { "cfg": { "lr": 0.01, "weight_decay": 0.0 }, "mean": 7.314293384552002 }, { "cfg": { "lr": 0.01, "weight_decay": 0.0001 }, "mean": 7.230101585388184 } ], "full": { "mean": 7.190596640110016, "std": 0.44163564514728276, "per_seed": [ 7.42201042175293, 7.114084720611572, 7.479846477508545, 6.873255252838135, 6.192978382110596, 7.314598083496094, 7.447637557983398, 7.680362224578857 ], "n": 8 } }, "idea": { "mean": 14.991608619689941, "std": 1.4869862368564055, "per_seed": [ 15.754435539245605, 14.699921607971191, 14.950590133666992, 13.512487411499023, 12.468204498291016, 15.603882789611816, 15.10988712310791, 17.833459854125977 ], "n": 8 }, "comparison": { "delta_mean": 7.8010119795799255, "idea_wins": 0, "n_pairs": 8, "per_seed_diffs": [ 8.332425117492676, 7.585836887359619, 7.470743656158447, 6.639232158660889, 6.27522611618042, 8.289284706115723, 7.662249565124512, 10.15309762954712 ], "p_value": 0.0081, "mde": 0.9946617786244648, "mde_rel_pct": 13.832812886153583, "verdict": "idea worse (significant)", "system_worked": false }, "mechanism_signature": { "prediction": "local nonlinear affine-model violation increases superlinearly with trusted radius", "loglog_slope_radius_violation": 1.9431904882117401, "spearman_like_rank_correlation": 0.9486956521739129, "confirmed": true } }, "system_verdict": "partial", "practical_verdict": "harms", "mechanism_ok": 1, "system_judged": true }