Entropy-Symmetrized Neural Flux / bench_report.json

✓✓ Beats tuned baseline

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": 0.003,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.004509380669333041
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 0.0036578410072252154
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.0037718922831118107
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.0034481814655009657,
 37      "std": 0.0004415458604303511,
 38      "per_seed": [
 39        0.003405338153243065,
 40        0.003420044668018818,
 41        0.003606032580137253,
 42        0.004199948627501726,
 43        0.003293032059445977,
 44        0.002657515462487936,
 45        0.003106778720393777,
 46        0.003896761452779174
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.0011822001833934337,
 53    "std": 0.00033321397960779696,
 54    "per_seed": [
 55      0.0010503326775506139,
 56      0.0014236074639484286,
 57      0.0007342637982219458,
 58      0.0010695835808292031,
 59      0.0010554390028119087,
 60      0.00107160909101367,
 61      0.0019350601360201836,
 62      0.0011177057167515159
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": -0.002265981282107532,
 68    "idea_wins": 8,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      -0.002355005475692451,
 72      -0.0019964372040703893,
 73      -0.002871768781915307,
 74      -0.003130365046672523,
 75      -0.0022375930566340685,
 76      -0.001585906371474266,
 77      -0.0011717185843735933,
 78      -0.002779055736027658
 79    ],
 80    "p_value": 0.0081,
 81    "mde": 0.000558113615628125,
 82    "mde_rel_pct": 16.185737937868073,
 83    "verdict": "idea better (significant)",
 84    "system_worked": true
 85  },
 86  "idea_sweep": [
 87    {
 88      "cfg": {
 89        "lr": 0.003,
 90        "weight_decay": 0.0
 91      },
 92      "mean": 0.0011822001833934337,
 93      "per_seed": [
 94        0.0010503326775506139,
 95        0.0014236074639484286,
 96        0.0007342637982219458,
 97        0.0010695835808292031,
 98        0.0010554390028119087,
 99        0.00107160909101367,
100        0.0019350601360201836,
101        0.0011177057167515159
102      ]
103    },
104    {
105      "cfg": {
106        "lr": 0.001,
107        "weight_decay": 0.0
108      },
109      "mean": 0.002837543041096069,
110      "per_seed": [
111        0.0023743249475955963,
112        0.002447130624204874,
113        0.0012939035659655929,
114        0.0055610244162380695,
115        0.0027316580526530743,
116        0.0028867414221167564,
117        0.0028890997637063265,
118        0.0025164615362882614
119      ]
120    },
121    {
122      "cfg": {
123        "lr": 0.01,
124        "weight_decay": 0.0
125      },
126      "mean": 0.0019820211964542978,
127      "per_seed": [
128        0.0011709327809512615,
129        0.0052402568981051445,
130        0.0006008912459947169,
131        0.0012470174115151167,
132        0.0014493301277980208,
133        0.0010334531543776393,
134        0.0033070051576942205,
135        0.0018072827951982617
136      ]
137    }
138  ],
139  "custom_track": null,
140  "mechanism_signature": {
141    "baseline_imag_max_mean": 0.0,
142    "idea_imag_max_mean": 0.0,
143    "prediction": "symmetric S and SPD H yield real A eigenvalues",
144    "confirmed": true
145  }
146}