Persistent Spectral Noise for Recurrent GNNs / 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      "epochs": 10,
 11      "sigma": 0.0
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "epochs": 10,
 18          "sigma": 0.0
 19        },
 20        "mean": 0.009392368956469
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.003,
 25          "epochs": 10,
 26          "sigma": 0.0
 27        },
 28        "mean": 0.005014403606764972
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.01,
 33          "epochs": 10,
 34          "sigma": 0.0
 35        },
 36        "mean": 0.004590500087942928
 37      }
 38    ],
 39    "full": {
 40      "mean": 0.003632939886301756,
 41      "std": 0.001261873819762896,
 42      "per_seed": [
 43        0.005352384876459837,
 44        0.005787275265902281,
 45        0.0038195690140128136,
 46        0.003402771195396781,
 47        0.0029491649474948645,
 48        0.00237746792845428,
 49        0.003468934213742614,
 50        0.0019059516489505768
 51      ],
 52      "n": 8
 53    }
 54  },
 55  "idea": {
 56    "mean": 0.004008013929706067,
 57    "std": 0.0023638783550923134,
 58    "per_seed": [
 59      0.005741244647651911,
 60      0.003205370157957077,
 61      0.0026500194799154997,
 62      0.0029542287811636925,
 63      0.0027808116283267736,
 64      0.0022908966057002544,
 65      0.0027672694995999336,
 66      0.009674270637333393
 67    ],
 68    "n": 8
 69  },
 70  "comparison": {
 71    "delta_mean": 0.00037507404340431094,
 72    "idea_wins": 6,
 73    "n_pairs": 8,
 74    "per_seed_diffs": [
 75      0.0003888597711920738,
 76      -0.0025819051079452038,
 77      -0.0011695495340973139,
 78      -0.0004485424142330885,
 79      -0.00016835331916809082,
 80      -8.657132275402546e-05,
 81      -0.0007016647141426802,
 82      0.007768318988382816
 83    ],
 84    "p_value": 0.9688,
 85    "mde": 0.002608707656005132,
 86    "mde_rel_pct": 71.80706914093018,
 87    "verdict": "no significant win",
 88    "system_worked": false
 89  },
 90  "mechanism_signature": {
 91    "mechanism_signature": {
 92      "prediction": "trained recurrent hidden temporal energy is nonzero and increases with sigma^2",
 93      "trained_model_observed": [
 94        {
 95          "sigma": 0.01,
 96          "sigma2": 0.0001,
 97          "observed_temporal_energy": 0.018960949033498764
 98        },
 99        {
100          "sigma": 0.03,
101          "sigma2": 0.0009,
102          "observed_temporal_energy": 0.018547870218753815
103        },
104        {
105          "sigma": 0.1,
106          "sigma2": 0.010000000000000002,
107          "observed_temporal_energy": 0.018667757511138916
108        }
109      ],
110      "baseline_sigma0": {
111        "metric": 0.005352384876459837,
112        "hidden_variance": 0.19951528310775757,
113        "temporal_energy": 0.020083513110876083,
114        "sigma": 0.0
115      },
116      "best_sigma": {
117        "metric": 0.005741244647651911,
118        "hidden_variance": 0.19854561984539032,
119        "temporal_energy": 0.018960949033498764,
120        "sigma": 0.01
121      },
122      "through_origin_slope": 2.035972557147564,
123      "through_origin_r2": -7019.474509912291,
124      "confirmed": false,
125      "note": "measured from trained benchmark model hidden trajectories"
126    },
127    "idea_sweep": [
128      {
129        "cfg": {
130          "lr": 0.01,
131          "epochs": 10,
132          "sigma": 0.01
133        },
134        "result": {
135          "mean": 0.004008013929706067,
136          "std": 0.0023638783550923134,
137          "per_seed": [
138            0.005741244647651911,
139            0.003205370157957077,
140            0.0026500194799154997,
141            0.0029542287811636925,
142            0.0027808116283267736,
143            0.0022908966057002544,
144            0.0027672694995999336,
145            0.009674270637333393
146          ],
147          "n": 8
148        }
149      },
150      {
151        "cfg": {
152          "lr": 0.01,
153          "epochs": 10,
154          "sigma": 0.03
155        },
156        "result": {
157          "mean": 0.004176608053967357,
158          "std": 0.002662518875846031,
159          "per_seed": [
160            0.005716613493859768,
161            0.0032172678038477898,
162            0.0026425563264638186,
163            0.0028939577750861645,
164            0.002898444654420018,
165            0.0025497509632259607,
166            0.0027517646085470915,
167            0.010742508806288242
168          ],
169          "n": 8
170        }
171      },
172      {
173        "cfg": {
174          "lr": 0.01,
175          "epochs": 10,
176          "sigma": 0.1
177        },
178        "result": {
179          "mean": 0.0049640603247098625,
180          "std": 0.0029402421909946234,
181          "per_seed": [
182            0.005744725465774536,
183            0.0041050356812775135,
184            0.0026244467590004206,
185            0.0034257906954735518,
186            0.004549694713205099,
187            0.0031880675815045834,
188            0.0036924462765455246,
189            0.01238227542489767
190          ],
191          "n": 8
192        }
193      }
194    ],
195    "track_match": "stability/control and recurrent state evolution -> dynamics"
196  },
197  "custom_track": null
198}