Data-driven invariant hidden-state ellipsoid / 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": 0.001864717691205442
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0010901302302954718
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.0006186682076076977
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.00054517226817552,
 33      "std": 0.00020361842727135583,
 34      "per_seed": [
 35        0.0004930912400595844,
 36        0.0004765945195686072,
 37        0.0004673648509196937,
 38        0.0010376222198829055,
 39        0.000511663849465549,
 40        0.00027233464061282575,
 41        0.0005571767687797546,
 42        0.0005455300561152399
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.24843271635472775,
 49    "std": 0.12890244786167743,
 50    "per_seed": [
 51      0.09243064373731613,
 52      0.48731154203414917,
 53      0.30256471037864685,
 54      0.0956249013543129,
 55      0.23182319104671478,
 56      0.24627986550331116,
 57      0.155330628156662,
 58      0.376096248626709
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.24788754408655223,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.09193755249725655,
 68      0.48683494751458056,
 69      0.30209734552772716,
 70      0.09458727913442999,
 71      0.23131152719724923,
 72      0.24600753086269833,
 73      0.15477345138788223,
 74      0.37555071857059374
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.1153001840292749,
 78    "mde_rel_pct": 21149.312017491255,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false,
 81    "permutation_pvalue": 0.0081
 82  },
 83  "idea_config": {
 84    "lr": 0.01,
 85    "lambda": 0.01
 86  },
 87  "idea_sweep": [
 88    {
 89      "cfg": {
 90        "lr": 0.001,
 91        "lambda": 0.01
 92      },
 93      "result": {
 94        "mean": 0.74566999822855,
 95        "std": 0.043829155721253926,
 96        "per_seed": [
 97          0.728725254535675,
 98          0.7290393114089966,
 99          0.78700190782547,
100          0.8437849879264832,
101          0.7193636894226074,
102          0.7325140237808228,
103          0.7265931367874146,
104          0.6983376741409302
105        ],
106        "n": 8
107      }
108    },
109    {
110      "cfg": {
111        "lr": 0.001,
112        "lambda": 0.1
113      },
114      "result": {
115        "mean": 0.7457223460078239,
116        "std": 0.04384320703957966,
117        "per_seed": [
118          0.7288874387741089,
119          0.729056179523468,
120          0.7870328426361084,
121          0.8438847064971924,
122          0.719398021697998,
123          0.7325379252433777,
124          0.7266111373901367,
125          0.6983705163002014
126        ],
127        "n": 8
128      }
129    },
130    {
131      "cfg": {
132        "lr": 0.001,
133        "lambda": 1.0
134      },
135      "result": {
136        "mean": 0.745727613568306,
137        "std": 0.043844610253164565,
138        "per_seed": [
139          0.7289036512374878,
140          0.7290578484535217,
141          0.7870359420776367,
142          0.8438947200775146,
143          0.7194015383720398,
144          0.7325403690338135,
145          0.7266129851341248,
146          0.6983738541603088
147        ],
148        "n": 8
149      }
150    },
151    {
152      "cfg": {
153        "lr": 0.003,
154        "lambda": 0.01
155      },
156      "result": {
157        "mean": 0.6492044553160667,
158        "std": 0.034374889139018386,
159        "per_seed": [
160          0.5766242742538452,
161          0.690006673336029,
162          0.6899384260177612,
163          0.6342498064041138,
164          0.6637277007102966,
165          0.655630886554718,
166          0.6534223556518555,
167          0.6300355195999146
168        ],
169        "n": 8
170      }
171    },
172    {
173      "cfg": {
174        "lr": 0.003,
175        "lambda": 0.1
176      },
177      "result": {
178        "mean": 0.6523455232381821,
179        "std": 0.03187828804668321,
180        "per_seed": [
181          0.5861462950706482,
182          0.6904615759849548,
183          0.692155659198761,
184          0.6431483030319214,
185          0.6647420525550842,
186          0.6567747592926025,
187          0.6541696190834045,
188          0.6311659216880798
189        ],
190        "n": 8
191      }
192    },
193    {
194      "cfg": {
195        "lr": 0.003,
196        "lambda": 1.0
197      },
198      "result": {
199        "mean": 0.6526677086949348,
200        "std": 0.03162840900701376,
201        "per_seed": [
202          0.5871396660804749,
203          0.6905071139335632,
204          0.6923784017562866,
205          0.6440579891204834,
206          0.6648440957069397,
207          0.6568903923034668,
208          0.6542445421218872,
209          0.631279468536377
210        ],
211        "n": 8
212      }
213    },
214    {
215      "cfg": {
216        "lr": 0.01,
217        "lambda": 0.01
218      },
219      "result": {
220        "mean": 0.24843271635472775,
221        "std": 0.12890244786167743,
222        "per_seed": [
223          0.09243064373731613,
224          0.48731154203414917,
225          0.30256471037864685,
226          0.0956249013543129,
227          0.23182319104671478,
228          0.24627986550331116,
229          0.155330628156662,
230          0.376096248626709
231        ],
232        "n": 8
233      }
234    },
235    {
236      "cfg": {
237        "lr": 0.01,
238        "lambda": 0.1
239      },
240      "result": {
241        "mean": 0.37393518909811974,
242        "std": 0.08752984377791141,
243        "per_seed": [
244          0.31651848554611206,
245          0.5420994162559509,
246          0.4366860091686249,
247          0.2581859529018402,
248          0.36661314964294434,
249          0.349292129278183,
250          0.28558772802352905,
251          0.43649864196777344
252        ],
253        "n": 8
254      }
255    },
256    {
257      "cfg": {
258        "lr": 0.01,
259        "lambda": 1.0
260      },
261      "result": {
262        "mean": 0.39147959649562836,
263        "std": 0.08067803171474866,
264        "per_seed": [
265          0.35244834423065186,
266          0.5486297011375427,
267          0.4513516426086426,
268          0.28766605257987976,
269          0.38437727093696594,
270          0.36130663752555847,
271          0.3028562068939209,
272          0.4432009160518646
273        ],
274        "n": 8
275      }
276    }
277  ],
278  "track_justification": "Dynamics is structurally matched: controlled actuated pendulum rollout with recurrent hidden state.",
279  "mechanism_signature": {
280    "prediction": "data-fitted ellipsoid barrier plus projection keeps normalized hidden energy bounded",
281    "baseline_mean_hidden_energy": 2.286437749862671,
282    "idea_mean_normalized_energy": 9.777409553527832,
283    "idea_max_normalized_energy": 22.335248947143555,
284    "idea_violation_rate": 1.0,
285    "predicted_max_energy": 1.0,
286    "confirmed": false
287  }
288}