Lyapunov Fading-Memory Optimizer / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "momentum": 0.9
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "momentum": 0.0
 17        },
 18        "mean": 0.6811865866184235
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "momentum": 0.9
 24        },
 25        "mean": 0.2693689875304699
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "momentum": 0.0
 31        },
 32        "mean": 0.5496408864855766
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "momentum": 0.9
 38        },
 39        "mean": 0.005768774193711579
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "momentum": 0.0
 45        },
 46        "mean": 0.3778964877128601
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "momentum": 0.9
 52        },
 53        "mean": 0.0021511380036827177
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.0024354744236916304,
 58      "std": 0.0014542907868172839,
 59      "per_seed": [
 60        0.004145679529756308,
 61        0.002980618504807353,
 62        0.0008099595434032381,
 63        0.000668294436763972,
 64        0.0030060403514653444,
 65        0.003159687388688326,
 66        0.004236925859004259,
 67        0.00047658977564424276
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.008837972913170233,
 74    "std": 0.006216926394780772,
 75    "per_seed": [
 76      0.0022076719906181097,
 77      0.006650018040090799,
 78      0.011529006995260715,
 79      0.02273235283792019,
 80      0.0036975364200770855,
 81      0.006488712504506111,
 82      0.004945824388414621,
 83      0.012452660128474236
 84    ],
 85    "n": 8,
 86    "best_cfg": {
 87      "lr": 0.003,
 88      "momentum": 0.9,
 89      "kappa": 0.5,
 90      "beta": 2.0
 91    }
 92  },
 93  "comparison": {
 94    "delta_mean": 0.006402498489478603,
 95    "idea_wins": 1,
 96    "n_pairs": 8,
 97    "per_seed_diffs": [
 98      -0.001938007539138198,
 99      0.0036693995352834463,
100      0.010719047451857477,
101      0.022064058401156217,
102      0.0006914960686117411,
103      0.0033290251158177853,
104      0.0007088985294103622,
105      0.011976070352829993
106    ],
107    "p_value": 0.037,
108    "mde": 0.006677981109294924,
109    "mde_rel_pct": 274.19631445657353,
110    "verdict": "idea worse (significant)",
111    "system_worked": false
112  },
113  "track_choice": "Lyapunov/stability optimizer structurally matches actuated pendulum dynamics track",
114  "mechanism_signature": {
115    "prediction": "exponential memory state decay rate beta*lr in discrete small-step response",
116    "predicted_decay_rate": 0.006,
117    "observed_decay_rate_mean": -0.10988588097760517,
118    "observed_force_ratio_final_mean": 3.349307292855885,
119    "observed_force_ratio_final_per_seed": [
120      3.3285459078945556,
121      3.357568656130888,
122      3.345152320579811,
123      3.365962286818286
124    ],
125    "confirmed": false
126  }
127}