Weak Koopman Latent 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": 0.01,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.0010981351661030203
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 0.0010876548330998048
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.0005773879893240519
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 0.0005898331510252319
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.01,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 0.00017333678806608077
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.01,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 0.00024068110360531136
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.0001381416768708732,
 58      "std": 8.561204737527774e-05,
 59      "per_seed": [
 60        0.00027561953174881637,
 61        0.00025423229089938104,
 62        6.920921441633254e-05,
 63        9.428611519979313e-05,
 64        0.00011270481627434492,
 65        4.6381595893763006e-05,
 66        5.4802185331936926e-05,
 67        0.00019789766520261765
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.00013874010073777754,
 74    "std": 8.310915831065427e-05,
 75    "per_seed": [
 76      0.0002849092415999621,
 77      0.00024259932979475707,
 78      6.42913073534146e-05,
 79      0.00012459092249628156,
 80      5.412771861301735e-05,
 81      7.187522714957595e-05,
 82      8.046071889111772e-05,
 83      0.00018706634000409395
 84    ],
 85    "n": 8
 86  },
 87  "comparison": {
 88    "delta_mean": 5.984238669043407e-07,
 89    "idea_wins": 4,
 90    "n_pairs": 8,
 91    "per_seed_diffs": [
 92      9.289709851145744e-06,
 93      -1.1632961104623973e-05,
 94      -4.917907062917948e-06,
 95      3.0304807296488434e-05,
 96      -5.857709766132757e-05,
 97      2.5493631255812943e-05,
 98      2.5658533559180796e-05,
 99      -1.08313251985237e-05
100    ],
101    "p_value": 0.969,
102    "mde": 2.4517188581434705e-05,
103    "mde_rel_pct": 17.74785795046628,
104    "verdict": "no measurable effect",
105    "system_worked": false
106  },
107  "mechanism_signature": {
108    "weak_koopman": {
109      "model_test_mse": 0.0006496671121567488,
110      "weak_residual_rms_observed": 0.009801944717764854,
111      "weak_scale_predicted_from_observed_B": 0.0648469477891922,
112      "relative_residual": 0.15115506668838702,
113      "confirmed": true
114    },
115    "selected_cfg": {
116      "lr": 0.01,
117      "weight_decay": 0.0,
118      "alpha": 0.01
119    },
120    "baseline_grid": [
121      {
122        "lr": 0.001,
123        "weight_decay": 0.0
124      },
125      {
126        "lr": 0.001,
127        "weight_decay": 0.0001
128      },
129      {
130        "lr": 0.003,
131        "weight_decay": 0.0
132      },
133      {
134        "lr": 0.003,
135        "weight_decay": 0.0001
136      },
137      {
138        "lr": 0.01,
139        "weight_decay": 0.0
140      },
141      {
142        "lr": 0.01,
143        "weight_decay": 0.0001
144      }
145    ],
146    "idea_grid": [
147      {
148        "lr": 0.01,
149        "weight_decay": 0.0,
150        "alpha": 0.01
151      },
152      {
153        "lr": 0.01,
154        "weight_decay": 0.0,
155        "alpha": 0.03
156      },
157      {
158        "lr": 0.01,
159        "weight_decay": 0.0,
160        "alpha": 0.1
161      }
162    ],
163    "track_justification": "Actuated pendulum rollout has explicit dynamical stability/control structure."
164  },
165  "idea_sweep": [
166    {
167      "cfg": {
168        "lr": 0.01,
169        "weight_decay": 0.0,
170        "alpha": 0.01
171      },
172      "result": {
173        "mean": 0.00013874010073777754,
174        "std": 8.310915831065427e-05,
175        "per_seed": [
176          0.0002849092415999621,
177          0.00024259932979475707,
178          6.42913073534146e-05,
179          0.00012459092249628156,
180          5.412771861301735e-05,
181          7.187522714957595e-05,
182          8.046071889111772e-05,
183          0.00018706634000409395
184        ],
185        "n": 8
186      }
187    },
188    {
189      "cfg": {
190        "lr": 0.01,
191        "weight_decay": 0.0,
192        "alpha": 0.03
193      },
194      "result": {
195        "mean": 0.00014541220389219234,
196        "std": 8.054320319092932e-05,
197        "per_seed": [
198          0.0002905830624513328,
199          0.0002372498856857419,
200          6.000923895044252e-05,
201          0.00013045869127381593,
202          8.125846943585202e-05,
203          6.058422877686098e-05,
204          0.00010983884567394853,
205          0.000193315208889544
206        ],
207        "n": 8
208      }
209    },
210    {
211      "cfg": {
212        "lr": 0.01,
213        "weight_decay": 0.0,
214        "alpha": 0.1
215      },
216      "result": {
217        "mean": 0.00019572334167605732,
218        "std": 7.176352745686476e-05,
219        "per_seed": [
220          0.00028579795616678894,
221          0.00025223862030543387,
222          0.00010874042345676571,
223          0.00025284208822995424,
224          0.00021868280600756407,
225          6.382106221280992e-05,
226          0.00016829039668664336,
227          0.00021537338034249842
228        ],
229        "n": 8
230      }
231    }
232  ]
233}