Parallel Quadratic Tree Layer / report_bench_2026-08-31T195753.md
Beats tuned baseline
Стенд-проверка (stage-2) · промт оператора:
(универсальный)
Ответ агента:
{ "worked": true, "confidence": 8, "verdict": "Implemented an end-to-end quadratic Schur-contraction temporal model and compared it with a matched sequential GRU system on the structurally appropriate dynamics track. Across 8 paired seeds, the idea reduced test MSE from 0.004203 to 0.002017, with all 8 paired wins and permutation p=0.0081, so the benchmark reports a significant win. The trained-model signature observed finite damped latent contractions with ratios 0.429–0.515 and was marked confirmed.", "metrics": { "baseline": "Sequential dynamics model, best lr=0.006, 8 epochs: mean test MSE 0.0042030427, std 0.0010498074.", "idea": "QuadraticTree Schur elimination, best lr=0.006, eps=0.1, 8 epochs: mean test MSE 0.0020170198, std 0.0005818964; paired delta -0.0021860230, p=0.0081, 8/8 wins." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_bench.py", "files": [ "stage2_bench.py", "bench_report.json" ], "limitations": "Only the built-in 8-step pendulum dynamics track was tested, using 800 training and 300 test samples per seed and 8 epochs. The forward contraction is sequential over this short chain and does not realize a genuinely parallel GPU rake-compress schedule; latency, FLOPs, peak memory, long-tree, and 100k-node scaling were not measured.", "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "epochs": 8, "eps": 0.1 }, "sweep": [ { "cfg": { "lr": 0.001, "epochs": 8, "eps": 0.02 }, "mean": 0.0411750716 }, { "cfg": { "lr": 0.003, "epochs": 8, "eps": 0.05 }, "mean": 0.0074238189 }, { "cfg": { "lr": 0.006, "epochs": 8, "eps": 0.1 }, "mean": 0.0050647462 } ], "full": { "mean": 0.0042030427, "std": 0.0010498074, "per_seed": [ 0.0038119142, 0.0051749824, 0.0055093565, 0.0057627317, 0.0028818215, 0.003411748, 0.0039346945, 0.0031370933 ], "n": 8 } }, "idea": { "best_cfg": { "lr": 0.006, "epochs": 8, "eps": 0.1 }, "mean": 0.0020170198, "std": 0.0005818964, "per_seed": [ 0.0029471358, 0.0018542189, 0.002415082, 0.0026398895, 0.0011361651, 0.0013343744, 0.0019472047, 0.0018620879 ], "n": 8 }, "comparison": { "delta_mean": -0.002186023, "idea_wins": 8, "n_pairs": 8, "per_seed_diffs": [ -0.0008647784, -0.0033207635, -0.0030942746, -0.0031228422, -0.0017456564, -0.0020773736, -0.0019874898, -0.0012750054 ], "p_value": 0.0081, "verdict": "idea better (significant)", "system_worked": true }, "mechanism_signature": { "predicted": "Schur elimination yields a finite, damped latent contraction.", "observed_ratio_range": [ 0.4285934, 0.5154902 ], "confirmed": true }, "custom_track": null }, "system_verdict": "worked", "practical_verdict": "helps", "mechanism_ok": 1, "system_judged": true }