Gramian-Regularized Latent State Models / 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": 0.003,
 10      "weight_decay": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.0015,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.01389938301872462
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.0015,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 0.01393047539750114
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.006244247197173536
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 0.00622874335385859
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 0.007728174212388694
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 0.007723002112470567
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.006740233220625669,
 58      "std": 0.003932537515930856,
 59      "per_seed": [
 60        0.005818837322294712,
 61        0.00311187538318336,
 62        0.012360582128167152,
 63        0.003623678581789136,
 64        0.0028440626338124275,
 65        0.004673089366406202,
 66        0.007840032689273357,
 67        0.013649707660079002
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.006728245556587353,
 74    "std": 0.003915532713508233,
 75    "per_seed": [
 76      0.005818843841552734,
 77      0.003111880971118808,
 78      0.012264670804142952,
 79      0.003623704193159938,
 80      0.0028439818415790796,
 81      0.004673039074987173,
 82      0.00784000102430582,
 83      0.013649842701852322
 84    ],
 85    "n": 8,
 86    "best_cfg": {
 87      "lr": 0.003,
 88      "weight_decay": 0.0001,
 89      "reg": 0.1
 90    },
 91    "sweep": [
 92      {
 93        "cfg": {
 94          "lr": 0.003,
 95          "weight_decay": 0.0001,
 96          "reg": 0.01
 97        },
 98        "mean": 0.006728254316840321,
 99        "std": 0.003915514027606794,
100        "per_seed": [
101          0.005818838719278574,
102          0.003111876081675291,
103          0.012264743447303772,
104          0.0036236809100955725,
105          0.0028440479654818773,
106          0.004673085641115904,
107          0.007840028963983059,
108          0.013649732805788517
109        ],
110        "n": 8
111      },
112      {
113        "cfg": {
114          "lr": 0.003,
115          "weight_decay": 0.0001,
116          "reg": 0.03
117        },
118        "mean": 0.006728248088620603,
119        "std": 0.003915514311451397,
120        "per_seed": [
121          0.005818837787955999,
122          0.003111874917522073,
123          0.012264708988368511,
124          0.0036236881278455257,
125          0.002844026545062661,
126          0.0046730744652450085,
127          0.007840024307370186,
128          0.01364974956959486
129        ],
130        "n": 8
131      },
132      {
133        "cfg": {
134          "lr": 0.003,
135          "weight_decay": 0.0001,
136          "reg": 0.1
137        },
138        "mean": 0.006728245556587353,
139        "std": 0.003915532713508233,
140        "per_seed": [
141          0.005818843841552734,
142          0.003111880971118808,
143          0.012264670804142952,
144          0.003623704193159938,
145          0.0028439818415790796,
146          0.004673039074987173,
147          0.00784000102430582,
148          0.013649842701852322
149        ],
150        "n": 8
151      }
152    ]
153  },
154  "comparison": {
155    "delta_mean": -1.1987664038315415e-05,
156    "idea_wins": 4,
157    "n_pairs": 8,
158    "per_seed_diffs": [
159      6.51925802230835e-09,
160      5.587935447692871e-09,
161      -9.591132402420044e-05,
162      2.561137080192566e-08,
163      -8.079223334789276e-08,
164      -5.029141902923584e-08,
165      -3.166496753692627e-08,
166      1.3504177331924438e-07
167    ],
168    "p_value": 0.5437,
169    "mde": 2.8354259077042443e-05,
170    "mde_rel_pct": 0.42067178017336365,
171    "verdict": "no measurable effect",
172    "system_worked": false
173  },
174  "mechanism_signature": {
175    "baseline_normalized_observability": -1.568709631527554e-06,
176    "idea_normalized_observability": -1.7481720213652352e-06,
177    "baseline_normalized_reachability": -1.48555855794541e-06,
178    "idea_normalized_reachability": -1.1099894585697712e-06,
179    "predicted": "regularization increases both normalized minimum Gramian eigenvalues",
180    "confirmed": false
181  },
182  "protocol_notes": {
183    "epochs": 8,
184    "paired_seeds": 8,
185    "structural_match": "control/dynamics latent-state track",
186    "baseline_grid": [
187      {
188        "lr": 0.0015,
189        "weight_decay": 0.0
190      },
191      {
192        "lr": 0.0015,
193        "weight_decay": 0.0001
194      },
195      {
196        "lr": 0.003,
197        "weight_decay": 0.0
198      },
199      {
200        "lr": 0.003,
201        "weight_decay": 0.0001
202      },
203      {
204        "lr": 0.006,
205        "weight_decay": 0.0
206      },
207      {
208        "lr": 0.006,
209        "weight_decay": 0.0001
210      }
211    ],
212    "idea_grid": [
213      {
214        "lr": 0.003,
215        "weight_decay": 0.0001,
216        "reg": 0.01
217      },
218      {
219        "lr": 0.003,
220        "weight_decay": 0.0001,
221        "reg": 0.03
222      },
223      {
224        "lr": 0.003,
225        "weight_decay": 0.0001,
226        "reg": 0.1
227      }
228    ]
229  }
230}