Symplectic Hamiltonian Optimizer / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 189.66867446899414
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 189.60668563842773
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 19.660808086395264
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 19.741307735443115
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 22.842825889587402
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 22.87500286102295
 54      }
 55    ],
 56    "full": {
 57      "mean": 20.874786138534546,
 58      "std": 4.428642414489589,
 59      "per_seed": [
 60        16.815889358520508,
 61        18.23108673095703,
 62        27.367088317871094,
 63        16.229167938232422,
 64        28.437650680541992,
 65        17.556045532226562,
 66        21.989797592163086,
 67        20.371562957763672
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 191.32705307006836,
 74    "std": 10.504082612747036,
 75    "per_seed": [
 76      198.2303466796875,
 77      194.18316650390625,
 78      201.94972229003906,
 79      189.76544189453125,
 80      178.002685546875,
 81      195.01913452148438,
 82      171.26991271972656,
 83      202.19601440429688
 84    ],
 85    "n": 8
 86  },
 87  "comparison": {
 88    "delta_mean": 170.4522669315338,
 89    "idea_wins": 0,
 90    "n_pairs": 8,
 91    "per_seed_diffs": [
 92      181.414457321167,
 93      175.95207977294922,
 94      174.58263397216797,
 95      173.53627395629883,
 96      149.565034866333,
 97      177.4630889892578,
 98      149.28011512756348,
 99      181.8244514465332
100    ],
101    "p_value": 0.0081,
102    "mde": 11.128805371835032,
103    "mde_rel_pct": 53.312188675750896,
104    "verdict": "idea worse (significant)",
105    "system_worked": false
106  },
107  "track_choice": "optimizer intervention structurally matches the tabular optimizer track",
108  "baseline_grid": [
109    {
110      "lr": 0.001,
111      "weight_decay": 0.0
112    },
113    {
114      "lr": 0.001,
115      "weight_decay": 0.0001
116    },
117    {
118      "lr": 0.003,
119      "weight_decay": 0.0
120    },
121    {
122      "lr": 0.003,
123      "weight_decay": 0.0001
124    },
125    {
126      "lr": 0.006,
127      "weight_decay": 0.0
128    },
129    {
130      "lr": 0.006,
131      "weight_decay": 0.0001
132    }
133  ],
134  "idea_grid": [
135    {
136      "h": 0.001,
137      "mass": 1.0
138    },
139    {
140      "h": 0.003,
141      "mass": 1.0
142    },
143    {
144      "h": 0.006,
145      "mass": 1.0
146    }
147  ],
148  "idea_best_cfg": {
149    "h": 0.006,
150    "mass": 1.0
151  },
152  "mechanism_signature": {
153    "prediction": "leapfrog trained-model energy oscillation scales approximately as h^2",
154    "step_sizes": [
155      0.001,
156      0.003,
157      0.006
158    ],
159    "observed_energy_ranges": [
160      0.011682417884003371,
161      0.10293819382786751,
162      0.5033693164587021
163    ],
164    "loglog_slope": 2.0894278143286225,
165    "confirmed": true
166  },
167  "idea_sweep": [
168    {
169      "cfg": {
170        "h": 0.001,
171        "mass": 1.0
172      },
173      "result": {
174        "mean": 235.28918075561523,
175        "std": 7.613234616794597,
176        "per_seed": [
177          243.13882446289062,
178          235.24093627929688,
179          240.7924346923828,
180          221.10401916503906,
181          244.659912109375,
182          230.86053466796875,
183          228.01934814453125,
184          238.4974365234375
185        ],
186        "n": 8
187      }
188    },
189    {
190      "cfg": {
191        "h": 0.003,
192        "mass": 1.0
193      },
194      "result": {
195        "mean": 226.3181381225586,
196        "std": 6.927247858722839,
197        "per_seed": [
198          233.64035034179688,
199          226.38275146484375,
200          233.14788818359375,
201          214.47064208984375,
202          231.64706420898438,
203          223.7176513671875,
204          216.8732147216797,
205          230.66554260253906
206        ],
207        "n": 8
208      }
209    },
210    {
211      "cfg": {
212        "h": 0.006,
213        "mass": 1.0
214      },
215      "result": {
216        "mean": 191.32705307006836,
217        "std": 10.504082612747036,
218        "per_seed": [
219          198.2303466796875,
220          194.18316650390625,
221          201.94972229003906,
222          189.76544189453125,
223          178.002685546875,
224          195.01913452148438,
225          171.26991271972656,
226          202.19601440429688
227        ],
228        "n": 8
229      }
230    }
231  ]
232}