Matching-Controllable Recurrent State Space / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "explicit_tanh_rnn_shared",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.32027048990130424
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 0.01070007259841077
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.0029869703284930438
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.0031130630668485537,
 37      "std": 0.0015298698474074236,
 38      "per_seed": [
 39        0.002583631779998541,
 40        0.001341367606073618,
 41        0.0019145020050927997,
 42        0.006108379922807217,
 43        0.005108166951686144,
 44        0.002453721361234784,
 45        0.0029577165842056274,
 46        0.002437018323689699
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.004186775186099112,
 53    "std": 0.0021045838143091996,
 54    "per_seed": [
 55      0.0009666543919593096,
 56      0.005662341136485338,
 57      0.0049017248675227165,
 58      0.0013687764294445515,
 59      0.007184116635471582,
 60      0.0038662750739604235,
 61      0.003245178610086441,
 62      0.006299134343862534
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": 0.0010737121192505583,
 68    "idea_wins": 2,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      -0.0016169773880392313,
 72      0.00432097353041172,
 73      0.002987222862429917,
 74      -0.004739603493362665,
 75      0.0020759496837854385,
 76      0.0014125537127256393,
 77      0.0002874620258808136,
 78      0.0038621160201728344
 79    ],
 80    "p_value": 0.3437,
 81    "mde": 0.002543128344355023,
 82    "mde_rel_pct": 81.69215623792383,
 83    "verdict": "no significant win",
 84    "system_worked": false
 85  },
 86  "mechanism_signature": {
 87    "prediction": "row-saturating matching gives full-rank trained controllability and improves long-horizon input access",
 88    "trained_model_measurements": [
 89      {
 90        "seed": 0,
 91        "test_mse": 0.002090232679620385,
 92        "controllability_rank": 16,
 93        "min_singular": 6.442173137209217e-16,
 94        "max_singular": 1.0159012079238892
 95      },
 96      {
 97        "seed": 1,
 98        "test_mse": 0.08538427203893661,
 99        "controllability_rank": 17,
100        "min_singular": 4.1185198621923365e-18,
101        "max_singular": 0.7894802689552307
102      },
103      {
104        "seed": 2,
105        "test_mse": 0.04124434292316437,
106        "controllability_rank": 17,
107        "min_singular": 4.013088390668608e-20,
108        "max_singular": 1.0363144874572754
109      },
110      {
111        "seed": 3,
112        "test_mse": 0.020918531343340874,
113        "controllability_rank": 15,
114        "min_singular": 2.736597840696797e-13,
115        "max_singular": 0.9130778908729553
116      },
117      {
118        "seed": 4,
119        "test_mse": 0.1084025576710701,
120        "controllability_rank": 15,
121        "min_singular": 2.54601735032937e-15,
122        "max_singular": 0.9427530765533447
123      },
124      {
125        "seed": 5,
126        "test_mse": 0.07313309609889984,
127        "controllability_rank": 18,
128        "min_singular": 1.5868783161841502e-24,
129        "max_singular": 0.8598010540008545
130      },
131      {
132        "seed": 6,
133        "test_mse": 0.05488329380750656,
134        "controllability_rank": 17,
135        "min_singular": 6.466245724631725e-13,
136        "max_singular": 0.9993555545806885
137      },
138      {
139        "seed": 7,
140        "test_mse": 0.07927680015563965,
141        "controllability_rank": 16,
142        "min_singular": 2.0967682614041888e-13,
143        "max_singular": 0.8383661508560181
144      }
145    ],
146    "idea_grid": [
147      {
148        "cfg": {
149          "lr": 0.001,
150          "weight_decay": 0.0
151        },
152        "full": {
153          "mean": 0.4885829910635948,
154          "std": 0.06858866720808456,
155          "per_seed": [
156            0.33161619305610657,
157            0.5656974911689758,
158            0.48498693108558655,
159            0.4976513087749481,
160            0.5585150122642517,
161            0.52698814868927,
162            0.47414085268974304,
163            0.4690679907798767
164          ],
165          "n": 8
166        }
167      },
168      {
169        "cfg": {
170          "lr": 0.003,
171          "weight_decay": 0.0
172        },
173        "full": {
174          "mean": 0.0581666408397723,
175          "std": 0.033096846800440026,
176          "per_seed": [
177            0.002090232679620385,
178            0.08538427203893661,
179            0.04124434292316437,
180            0.020918531343340874,
181            0.1084025576710701,
182            0.07313309609889984,
183            0.05488329380750656,
184            0.07927680015563965
185          ],
186          "n": 8
187        }
188      },
189      {
190        "cfg": {
191          "lr": 0.01,
192          "weight_decay": 0.0
193        },
194        "full": {
195          "mean": 0.004186775186099112,
196          "std": 0.0021045838143091996,
197          "per_seed": [
198            0.0009666543919593096,
199            0.005662341136485338,
200            0.0049017248675227165,
201            0.0013687764294445515,
202            0.007184116635471582,
203            0.0038662750739604235,
204            0.003245178610086441,
205            0.006299134343862534
206          ],
207          "n": 8
208        }
209      }
210    ],
211    "confirmed": false
212  }
213}