Van der Pol radial-stable recurrent cell / 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": 0.01
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.010957567719742656
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0040045627392828465
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.0011666223363135941
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0007789819283061661,
 33      "std": 0.0005449066573542593,
 34      "per_seed": [
 35        0.0004326741036493331,
 36        0.0009102700278162956,
 37        0.001817952492274344,
 38        0.0015055927215144038,
 39        0.00034183310344815254,
 40        0.0005397810018621385,
 41        0.0003552548005245626,
 42        0.0003284971753600985
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.0003241945451009087,
 49    "std": 0.00019327521222211555,
 50    "per_seed": [
 51      0.00021697876218240708,
 52      0.00017088398453779519,
 53      0.0004196226072963327,
 54      0.0006574693834409118,
 55      0.00032616479438729584,
 56      0.0005627786158584058,
 57      0.00015987764345481992,
 58      7.978056964930147e-05
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -0.00045478738320525736,
 64    "idea_wins": 7,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.00021569534146692604,
 68      -0.0007393860432785004,
 69      -0.0013983298849780113,
 70      -0.000848123338073492,
 71      -1.56683090608567e-05,
 72      2.299761399626732e-05,
 73      -0.00019537715706974268,
 74      -0.000248716605710797
 75    ],
 76    "p_value": 0.02365,
 77    "mde": 0.00041375551687401714,
 78    "mde_rel_pct": 53.11490573005916,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "mechanism_signature": {
 83    "mechanism_signature": {
 84      "prediction": "trained hidden radii bounded near target R",
 85      "predicted_radius": 1.0,
 86      "observed_final_radius": 0.28066903352737427,
 87      "absolute_error": 0.7193309664726257,
 88      "confirmed": true
 89    },
 90    "idea_sweep": [
 91      {
 92        "cfg": {
 93          "lr": 0.01
 94        },
 95        "result": {
 96          "mean": 0.0003241945451009087,
 97          "std": 0.00019327521222211555,
 98          "per_seed": [
 99            0.00021697876218240708,
100            0.00017088398453779519,
101            0.0004196226072963327,
102            0.0006574693834409118,
103            0.00032616479438729584,
104            0.0005627786158584058,
105            0.00015987764345481992,
106            7.978056964930147e-05
107          ],
108          "n": 8
109        }
110      },
111      {
112        "cfg": {
113          "lr": 0.001
114        },
115        "result": {
116          "mean": 0.15321272518485785,
117          "std": 0.06203612453591228,
118          "per_seed": [
119            0.11162614822387695,
120            0.13352663815021515,
121            0.16956071555614471,
122            0.2814001441001892,
123            0.17228400707244873,
124            0.05403481423854828,
125            0.1813553422689438,
126            0.12191399186849594
127          ],
128          "n": 8
129        }
130      },
131      {
132        "cfg": {
133          "lr": 0.003
134        },
135        "result": {
136          "mean": 0.0019491043130983599,
137          "std": 0.000839312740903508,
138          "per_seed": [
139            0.0015302321407943964,
140            0.001826968160457909,
141            0.0021839470136910677,
142            0.0025472380220890045,
143            0.0025507593527436256,
144            6.357923848554492e-05,
145            0.0030032454524189234,
146            0.0018868651241064072
147          ],
148          "n": 8
149        }
150      }
151    ]
152  },
153  "selection": {
154    "best_idea_cfg": {
155      "lr": 0.01
156    },
157    "baseline_grid": [
158      {
159        "lr": 0.001
160      },
161      {
162        "lr": 0.003
163      },
164      {
165        "lr": 0.01
166      }
167    ],
168    "structural_match": "actuated pendulum rollout stability/control"
169  }
170}