Volume-Mass Diffusion GNN / 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      "epochs": 30,
 12      "batch": 64
 13    },
 14    "sweep": [
 15      {
 16        "cfg": {
 17          "lr": 0.001,
 18          "weight_decay": 0.0,
 19          "epochs": 30,
 20          "batch": 64
 21        },
 22        "mean": 0.0008596329862484708
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.001,
 27          "weight_decay": 0.0001,
 28          "epochs": 30,
 29          "batch": 64
 30        },
 31        "mean": 0.0008667934744153172
 32      },
 33      {
 34        "cfg": {
 35          "lr": 0.003,
 36          "weight_decay": 0.0,
 37          "epochs": 30,
 38          "batch": 64
 39        },
 40        "mean": 0.0003459557265159674
 41      },
 42      {
 43        "cfg": {
 44          "lr": 0.003,
 45          "weight_decay": 0.0001,
 46          "epochs": 30,
 47          "batch": 64
 48        },
 49        "mean": 0.0004122954633203335
 50      },
 51      {
 52        "cfg": {
 53          "lr": 0.01,
 54          "weight_decay": 0.0,
 55          "epochs": 30,
 56          "batch": 64
 57        },
 58        "mean": 3.945273101635394e-05
 59      },
 60      {
 61        "cfg": {
 62          "lr": 0.01,
 63          "weight_decay": 0.0001,
 64          "epochs": 30,
 65          "batch": 64
 66        },
 67        "mean": 0.00015876943143666722
 68      }
 69    ],
 70    "full": {
 71      "mean": 4.022148073090648e-05,
 72      "std": 1.0539636376553464e-05,
 73      "per_seed": [
 74        2.7519910872797482e-05,
 75        5.704866998712532e-05,
 76        2.5148139684461057e-05,
 77        4.8094203521031886e-05,
 78        3.4156204492319375e-05,
 79        3.871386797982268e-05,
 80        4.010241536889225e-05,
 81        5.098843394080177e-05
 82      ],
 83      "n": 8
 84    }
 85  },
 86  "idea": {
 87    "mean": 8.502966829837533e-05,
 88    "std": 3.866288220789e-05,
 89    "per_seed": [
 90      3.906466736225411e-05,
 91      0.00010664308501873165,
 92      6.796212983317673e-05,
 93      6.528497760882601e-05,
 94      6.44689062028192e-05,
 95      0.0001517514610895887,
 96      0.00013560302613768727,
 97      4.945909313391894e-05
 98    ],
 99    "n": 8
100  },
101  "comparison": {
102    "delta_mean": 4.480818756746885e-05,
103    "idea_wins": 1,
104    "n_pairs": 8,
105    "per_seed_diffs": [
106      1.1544756489456631e-05,
107      4.959441503160633e-05,
108      4.2813990148715675e-05,
109      1.7190774087794125e-05,
110      3.0312701710499823e-05,
111      0.00011303759310976602,
112      9.550061076879501e-05,
113      -1.5293408068828285e-06
114    ],
115    "p_value": 0.0156,
116    "mde": 3.385917105994864e-05,
117    "mde_rel_pct": 84.18181142180329,
118    "verdict": "idea worse (significant)",
119    "system_worked": false
120  },
121  "mechanism_signature": {
122    "mass_diffusion": {
123      "predicted_max_mode_amplitude": 0.19999999999999996,
124      "observed_mean_test_input_norm_ratio": 0.6925333812832832,
125      "observed_per_seed": [
126        0.6863239407539368,
127        0.7053196430206299,
128        0.6866878867149353,
129        0.6967299580574036,
130        0.6953513622283936,
131        0.6847705245018005,
132        0.6918107271194458,
133        0.6932730078697205
134      ],
135      "lambda_max": 2.0,
136      "confirmed": false
137    },
138    "idea_sweep": [
139      {
140        "cfg": {
141          "lr": 0.01,
142          "weight_decay": 0.0,
143          "epochs": 30,
144          "batch": 64,
145          "alpha": 1.0,
146          "gamma": 0.6
147        },
148        "result": {
149          "mean": 8.502966829837533e-05,
150          "std": 3.866288220789e-05,
151          "per_seed": [
152            3.906466736225411e-05,
153            0.00010664308501873165,
154            6.796212983317673e-05,
155            6.528497760882601e-05,
156            6.44689062028192e-05,
157            0.0001517514610895887,
158            0.00013560302613768727,
159            4.945909313391894e-05
160          ],
161          "n": 8
162        }
163      },
164      {
165        "cfg": {
166          "lr": 0.01,
167          "weight_decay": 0.0,
168          "epochs": 30,
169          "batch": 64,
170          "alpha": 1.0,
171          "gamma": 0.8
172        },
173        "result": {
174          "mean": 0.000431537780968938,
175          "std": 0.0003206459743519331,
176          "per_seed": [
177            0.00012159255857113749,
178            0.0005679481546394527,
179            0.0004245344316586852,
180            0.0001826822553994134,
181            0.00037008809158578515,
182            0.00020022669923491776,
183            0.0011974757071584463,
184            0.00038775434950366616
185          ],
186          "n": 8
187        }
188      },
189      {
190        "cfg": {
191          "lr": 0.01,
192          "weight_decay": 0.0,
193          "epochs": 30,
194          "batch": 64,
195          "alpha": 1.0,
196          "gamma": 0.95
197        },
198        "result": {
199          "mean": 0.008008880191482604,
200          "std": 0.0025785752995045367,
201          "per_seed": [
202            0.007743558846414089,
203            0.00685552041977644,
204            0.008397368714213371,
205            0.007993439212441444,
206            0.003967979457229376,
207            0.008484840393066406,
208            0.01381105836480856,
209            0.006817276123911142
210          ],
211          "n": 8
212        }
213      }
214    ]
215  },
216  "notes": "Dynamics is structurally matched: controlled pendulum rollout and stability-sensitive recurrent prediction; diffusion acts on the three physical channels at every timestep."
217}