Maslov Phase Budget for Symplectic Recurrence / 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      "wd": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "wd": 0.0
 17        },
 18        "mean": 0.004538182052783668
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "wd": 0.0001
 24        },
 25        "mean": 0.004533302329946309
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "wd": 0.0
 31        },
 32        "mean": 0.0035770683316513896
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "wd": 0.0001
 38        },
 39        "mean": 0.0035493404720909894
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "wd": 0.0
 45        },
 46        "mean": 0.0014125758316367865
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "wd": 0.0001
 52        },
 53        "mean": 0.0013684789882972836
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.0015742218820378184,
 58      "std": 0.0005383299843650786,
 59      "per_seed": [
 60        0.0017803608207032084,
 61        0.001374608720652759,
 62        0.0006994276773184538,
 63        0.0016195187345147133,
 64        0.0015369732864201069,
 65        0.0012765798019245267,
 66        0.0027483156882226467,
 67        0.0015579903265461326
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.0015715822883066721,
 74    "std": 0.0005392184858503133,
 75    "per_seed": [
 76      0.001779490732587874,
 77      0.0013879159232601523,
 78      0.0007163107511587441,
 79      0.0014804680831730366,
 80      0.0015849629417061806,
 81      0.0012773972703143954,
 82      0.002764256903901696,
 83      0.0015818557003512979
 84    ],
 85    "n": 8
 86  },
 87  "comparison": {
 88    "delta_mean": -2.639593731146306e-06,
 89    "idea_wins": 2,
 90    "n_pairs": 8,
 91    "per_seed_diffs": [
 92      -8.700881153345108e-07,
 93      1.3307202607393265e-05,
 94      1.6883073840290308e-05,
 95      -0.0001390506513416767,
 96      4.7989655286073685e-05,
 97      8.174683898687363e-07,
 98      1.5941215679049492e-05,
 99      2.386537380516529e-05
100    ],
101    "p_value": 0.985,
102    "mde": 4.7793207118518294e-05,
103    "mde_rel_pct": 3.0359892505528094,
104    "verdict": "no measurable effect",
105    "system_worked": false
106  },
107  "track_justification": "dynamics is the matched actuated-pendulum rollout track for stability/control ideas",
108  "idea_sweep": [
109    {
110      "cfg": {
111        "lr": 0.006,
112        "wd": 0.0001,
113        "phase": 0.0,
114        "tv": 0.0
115      },
116      "mean": 0.0013684789882972836
117    },
118    {
119      "cfg": {
120        "lr": 0.006,
121        "wd": 0.0001,
122        "phase": 0.002,
123        "tv": 0.01
124      },
125      "mean": 0.0013676858798135072
126    },
127    {
128      "cfg": {
129        "lr": 0.006,
130        "wd": 0.0001,
131        "phase": 0.008,
132        "tv": 0.04
133      },
134      "mean": 0.0013410463725449517
135    }
136  ],
137  "math_check": {
138    "expected_sum_theta_dot": -0.15,
139    "finite_difference_q": -0.15000000002777925,
140    "absolute_error": 2.7779251121629045e-11,
141    "passed": true
142  },
143  "custom_track": null,
144  "mechanism_signature": {
145    "quantity": "test hidden angular-velocity variance q",
146    "prediction": "phase budget lowers q variance",
147    "baseline_mean": 0.06730567710474133,
148    "idea_mean": 0.06802853476256132,
149    "ratio_idea_over_baseline": 1.0107399210425456,
150    "relative_reduction": -0.01073992104254562,
151    "confirmed": false
152  }
153}