Quotient-Fibre Mixing Network / 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": 0.01
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.004103701037820429
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0033861721749417484
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.0014241317112464458
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.001394080012687482,
 33      "std": 0.0003941947646392458,
 34      "per_seed": [
 35        0.0008339877240359783,
 36        0.0018332540057599545,
 37        0.001035665743984282,
 38        0.0019936193712055683,
 39        0.0009847625624388456,
 40        0.0013270708732306957,
 41        0.0016533465823158622,
 42        0.0014909332385286689
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.001602079973963555,
 49    "std": 0.0004381539879566494,
 50    "per_seed": [
 51      0.0021186538506299257,
 52      0.0017930945614352822,
 53      0.0016023782081902027,
 54      0.0021843218710273504,
 55      0.0014364364324137568,
 56      0.00164392264559865,
 57      0.0007154843187890947,
 58      0.001322347903624177
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.000207999961276073,
 64    "idea_wins": 3,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.0012846661265939474,
 68      -4.015944432467222e-05,
 69      0.0005667124642059207,
 70      0.0001907024998217821,
 71      0.0004516738699749112,
 72      0.00031685177236795425,
 73      -0.0009378622635267675,
 74      -0.0001685853349044919
 75    ],
 76    "p_value": 0.4041,
 77    "mde": 0.0005359412451298254,
 78    "mde_rel_pct": 38.44408070212897,
 79    "verdict": "no significant win",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "predicted_vs_observed": {
 84      "predicted_max_rate": 0.7171734331655341,
 85      "observed_full_rate": 0.6922671462080507,
 86      "observed_quotient_rate": 0.6404829150176061,
 87      "observed_fibre_rate": 0.7171734331655341,
 88      "relative_error": 0.03472840153538507,
 89      "confirmed": true,
 90      "measurement": "trained quotient-fibre GRU hidden-state perturbation decay"
 91    },
 92    "idea_lr_sweep": [
 93      {
 94        "lr": 0.01,
 95        "result": {
 96          "mean": 0.001602079973963555,
 97          "std": 0.0004381539879566494,
 98          "per_seed": [
 99            0.0021186538506299257,
100            0.0017930945614352822,
101            0.0016023782081902027,
102            0.0021843218710273504,
103            0.0014364364324137568,
104            0.00164392264559865,
105            0.0007154843187890947,
106            0.001322347903624177
107          ],
108          "n": 8
109        }
110      },
111      {
112        "lr": 0.001,
113        "result": {
114          "mean": 0.0051142975862603635,
115          "std": 0.0013857279567077699,
116          "per_seed": [
117            0.006965973414480686,
118            0.006302177906036377,
119            0.0032342439517378807,
120            0.006503377575427294,
121            0.005306449253112078,
122            0.003405582858249545,
123            0.003712605219334364,
124            0.005483970511704683
125          ],
126          "n": 8
127        }
128      },
129      {
130        "lr": 0.003,
131        "result": {
132          "mean": 0.002961420876090415,
133          "std": 0.00145164098988552,
134          "per_seed": [
135            0.0031063449569046497,
136            0.002748891245573759,
137            0.0011568840127438307,
138            0.006246649660170078,
139            0.003641354851424694,
140            0.0015332313487306237,
141            0.002529497491195798,
142            0.002728513441979885
143          ],
144          "n": 8
145        }
146      }
147    ],
148    "parameter_counts": {
149      "baseline": 13313,
150      "idea": 13581
151    },
152    "track_choice": "dynamics structurally matches stability/control and multi-step dynamical memory"
153  }
154}