Padé-Hermite Neural ODE Integrator / bench_report.json

Mechanism confirmed, baseline not beaten

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      "h": 0.025
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "h": 0.025
 17        },
 18        "mean": 1.4525697110912006e-05
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "h": 0.05
 24        },
 25        "mean": 1.4545178487423982e-05
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.001,
 30          "h": 0.1
 31        },
 32        "mean": 1.4584162499886588e-05
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "h": 0.025
 38        },
 39        "mean": 8.580010785408376e-06
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "h": 0.05
 45        },
 46        "mean": 8.575699666835135e-06
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "h": 0.1
 52        },
 53        "mean": 8.567132681491785e-06
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.01,
 58          "h": 0.025
 59        },
 60        "mean": 5.892069680157874e-06
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.01,
 65          "h": 0.05
 66        },
 67        "mean": 5.897995151826763e-06
 68      },
 69      {
 70        "cfg": {
 71          "lr": 0.01,
 72          "h": 0.1
 73        },
 74        "mean": 5.909915387292131e-06
 75      }
 76    ],
 77    "full": {
 78      "mean": 5.681601408014103e-06,
 79      "std": 2.730715161483723e-06,
 80      "per_seed": [
 81        2.3133106878958642e-06,
 82        4.037152848468395e-06,
 83        1.1316467862343416e-05,
 84        5.901347321923822e-06,
 85        5.3298954298952594e-06,
 86        3.4796667023329064e-06,
 87        8.45680187921971e-06,
 88        4.618168532033451e-06
 89      ],
 90      "n": 8
 91    }
 92  },
 93  "idea": {
 94    "mean": 5.6799913750182895e-06,
 95    "std": 2.7279983370928816e-06,
 96    "per_seed": [
 97      2.3146424155129353e-06,
 98      4.040079147671349e-06,
 99      1.1296857337583788e-05,
100      5.893076377105899e-06,
101      5.312819212122122e-06,
102      3.484348098936607e-06,
103      8.480390533804893e-06,
104      4.617717877408722e-06
105    ],
106    "n": 8
107  },
108  "comparison": {
109    "delta_mean": -1.6100329958135262e-09,
110    "idea_wins": 4,
111    "n_pairs": 8,
112    "per_seed_diffs": [
113      1.3317276170710102e-09,
114      2.9262992029543966e-09,
115      -1.961052475962788e-08,
116      -8.27094481792301e-09,
117      -1.7076217773137614e-08,
118      4.681396603700705e-09,
119      2.3588654585182667e-08,
120      -4.5065462472848594e-10
121    ],
122    "p_value": 0.78145,
123    "mde": 1.1450469552506787e-08,
124    "mde_rel_pct": 0.20153595316199915,
125    "verdict": "no measurable effect",
126    "system_worked": false
127  },
128  "mechanism_signature": {
129    "signature": {
130      "prediction": "two-derivative correction should improve finite-step rollout",
131      "predicted": {
132        "order": 4,
133        "correction_coefficient": 0.5
134      },
135      "observed": {
136        "mean_abs_field": 0.08425881783477962,
137        "mean_abs_ph_correction": 7.20391401642928e-07
138      },
139      "confirmed": true
140    },
141    "idea_sweep": [
142      {
143        "cfg": {
144          "lr": 0.001,
145          "h": 0.025
146        },
147        "mean": 1.4506283719128987e-05
148      },
149      {
150        "cfg": {
151          "lr": 0.001,
152          "h": 0.05
153        },
154        "mean": 1.4506301454275672e-05
155      },
156      {
157        "cfg": {
158          "lr": 0.001,
159          "h": 0.1
160        },
161        "mean": 1.4506304410133453e-05
162      },
163      {
164        "cfg": {
165          "lr": 0.003,
166          "h": 0.025
167        },
168        "mean": 8.584338502259925e-06
169      },
170      {
171        "cfg": {
172          "lr": 0.003,
173          "h": 0.05
174        },
175        "mean": 8.584342822359758e-06
176      },
177      {
178        "cfg": {
179          "lr": 0.003,
180          "h": 0.1
181        },
182        "mean": 8.58435225836729e-06
183      },
184      {
185        "cfg": {
186          "lr": 0.01,
187          "h": 0.025
188        },
189        "mean": 5.886169105906447e-06
190      },
191      {
192        "cfg": {
193          "lr": 0.01,
194          "h": 0.05
195        },
196        "mean": 5.886163819468493e-06
197      },
198      {
199        "cfg": {
200          "lr": 0.01,
201          "h": 0.1
202        },
203        "mean": 5.886164046842168e-06
204      }
205    ]
206  }
207}