Alignment-Section Floquet Training for Recurrent Dynamics / 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": 0.006,
 10      "weight_decay": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.005222459672950208
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 0.0032230458600679412
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.0022474244324257597
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.001,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 0.005207901762332767
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "weight_decay": 0.0001
 45        },
 46        "mean": 0.003198997670551762
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 0.0022117759071988985
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.0022117759071988985,
 58      "std": 0.0004126881098400608,
 59      "per_seed": [
 60        0.002800536109134555,
 61        0.001347046927548945,
 62        0.001825663261115551,
 63        0.0023813715670257807,
 64        0.0023169058840721846,
 65        0.0024658460170030594,
 66        0.0022247303277254105,
 67        0.002332107163965702
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "cfg": {
 74      "lr": 0.006,
 75      "weight_decay": 0.0
 76    },
 77    "mean": 0.0018650132114998996,
 78    "std": 0.0008113546405834626,
 79    "per_seed": [
 80      0.0019402806647121906,
 81      0.000917025376111269,
 82      0.00216244300827384,
 83      0.003614878747612238,
 84      0.0023528579622507095,
 85      0.0011243090266361833,
 86      0.0014117673272266984,
 87      0.0013965435791760683
 88    ],
 89    "n": 8
 90  },
 91  "comparison": {
 92    "delta_mean": -0.00034676269569899887,
 93    "idea_wins": 5,
 94    "n_pairs": 8,
 95    "per_seed_diffs": [
 96      -0.0008602554444223642,
 97      -0.00043002155143767595,
 98      0.00033677974715828896,
 99      0.0012335071805864573,
100      3.595207817852497e-05,
101      -0.0013415369903668761,
102      -0.0008129630004987121,
103      -0.0009355635847896338
104    ],
105    "p_value": 0.25545,
106    "mde": 0.0007020822623155458,
107    "mde_rel_pct": 31.742920249307595,
108    "verdict": "no significant win",
109    "system_worked": false
110  },
111  "idea_sweep": [
112    {
113      "cfg": {
114        "lr": 0.001,
115        "weight_decay": 0.0
116      },
117      "mean": 0.005775320241809823,
118      "std": 0.0032488792459768865,
119      "per_seed": [
120        0.013064713217318058,
121        0.001774493488483131,
122        0.005221209488809109,
123        0.0029860660433769226,
124        0.005788332317024469,
125        0.0035972988698631525,
126        0.006834390107542276,
127        0.006936058402061462
128      ],
129      "n": 8
130    },
131    {
132      "cfg": {
133        "lr": 0.003,
134        "weight_decay": 0.0
135      },
136      "mean": 0.003850284934742376,
137      "std": 0.0013980431321628994,
138      "per_seed": [
139        0.005784670822322369,
140        0.0026722021866589785,
141        0.003800347214564681,
142        0.005372739862650633,
143        0.004920608829706907,
144        0.0028360977303236723,
145        0.0014036931097507477,
146        0.004011919721961021
147      ],
148      "n": 8
149    },
150    {
151      "cfg": {
152        "lr": 0.006,
153        "weight_decay": 0.0
154      },
155      "mean": 0.0018650132114998996,
156      "std": 0.0008113546405834626,
157      "per_seed": [
158        0.0019402806647121906,
159        0.000917025376111269,
160        0.00216244300827384,
161        0.003614878747612238,
162        0.0023528579622507095,
163        0.0011243090266361833,
164        0.0014117673272266984,
165        0.0013965435791760683
166      ],
167      "n": 8
168    },
169    {
170      "cfg": {
171        "lr": 0.001,
172        "weight_decay": 0.0001
173      },
174      "mean": 0.005752541357651353,
175      "std": 0.003251318835613531,
176      "per_seed": [
177        0.013069597072899342,
178        0.0017632902599871159,
179        0.005176736507564783,
180        0.00296420487575233,
181        0.005783510394394398,
182        0.003592958441004157,
183        0.006851297803223133,
184        0.006818735506385565
185      ],
186      "n": 8
187    },
188    {
189      "cfg": {
190        "lr": 0.003,
191        "weight_decay": 0.0001
192      },
193      "mean": 0.0038360641920007765,
194      "std": 0.0013847205833962984,
195      "per_seed": [
196        0.005729560274630785,
197        0.0026723749469965696,
198        0.003779245540499687,
199        0.005330996587872505,
200        0.004909616895020008,
201        0.002826679963618517,
202        0.0013972243759781122,
203        0.004042814951390028
204      ],
205      "n": 8
206    },
207    {
208      "cfg": {
209        "lr": 0.006,
210        "weight_decay": 0.0001
211      },
212      "mean": 0.0018727781061897986,
213      "std": 0.000814390952787726,
214      "per_seed": [
215        0.0019529082346707582,
216        0.0009154137806035578,
217        0.002177728107199073,
218        0.0036332407034933567,
219        0.0023326349910348654,
220        0.0011026168940588832,
221        0.0014148338232189417,
222        0.0014528483152389526
223      ],
224      "n": 8
225    }
226  ],
227  "idea_best_signature_metrics": {
228    "test_metric": 0.0019402806647121906,
229    "return_residual_mean": 0.6867897510528564,
230    "floquet_rho": 0.0386572889983654
231  },
232  "mechanism_signature": {
233    "baseline_predicted": "rho<1 implies reduced return residual",
234    "predicted_rho": "<1 should be stable",
235    "baseline_rho": 0.03690243512392044,
236    "idea_rho": 0.0386572889983654,
237    "baseline_return_residual": 0.9603143334388733,
238    "idea_return_residual": 0.6867897510528564,
239    "observed_direction": "not lower",
240    "confirmed": false
241  }
242}