Actionable-Information 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": 0.006,
 10      "weight_decay": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 7.018276214599609
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 6.867959260940552
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 6.590747117996216
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 6.44768226146698
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 5.89583158493042
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 5.276575446128845
 54      }
 55    ],
 56    "full": {
 57      "mean": 5.42892998456955,
 58      "std": 0.41047416569067036,
 59      "per_seed": [
 60        5.578059196472168,
 61        5.473832130432129,
 62        5.200394153594971,
 63        4.854016304016113,
 64        6.361762046813965,
 65        5.232422828674316,
 66        5.255797863006592,
 67        5.475155353546143
 68      ],
 69      "n": 8
 70    },
 71    "grid_union": [
 72      {
 73        "lr": 0.001,
 74        "weight_decay": 0.0
 75      },
 76      {
 77        "lr": 0.001,
 78        "weight_decay": 0.0001
 79      },
 80      {
 81        "lr": 0.003,
 82        "weight_decay": 0.0
 83      },
 84      {
 85        "lr": 0.003,
 86        "weight_decay": 0.0001
 87      },
 88      {
 89        "lr": 0.006,
 90        "weight_decay": 0.0
 91      },
 92      {
 93        "lr": 0.006,
 94        "weight_decay": 0.0001
 95      }
 96    ]
 97  },
 98  "idea": {
 99    "mean": 6.97602242231369,
100    "std": 0.26847889510235706,
101    "per_seed": [
102      6.515726566314697,
103      6.685352802276611,
104      7.203780651092529,
105      7.1638712882995605,
106      6.83967924118042,
107      7.354576110839844,
108      6.9066338539123535,
109      7.138558864593506
110    ],
111    "n": 8,
112    "best_cfg": {
113      "lr": 0.003,
114      "delta": 0.5,
115      "lambda": 0.03
116    }
117  },
118  "comparison": {
119    "delta_mean": 1.5470924377441406,
120    "idea_wins": 0,
121    "n_pairs": 8,
122    "per_seed_diffs": [
123      0.9376673698425293,
124      1.2115206718444824,
125      2.0033864974975586,
126      2.3098549842834473,
127      0.4779171943664551,
128      2.1221532821655273,
129      1.6508359909057617,
130      1.6634035110473633
131    ],
132    "p_value": 0.0081,
133    "mde": 0.5261925661705102,
134    "mde_rel_pct": 9.692380776066154,
135    "verdict": "idea worse (significant)",
136    "system_worked": false
137  },
138  "mechanism_signature": {
139    "math_sanity": {
140      "rows": [
141        {
142          "delta": 2.0,
143          "I_acq": 0.9397995625962782,
144          "I_use": 0.05860868669192204
145        },
146        {
147          "delta": 1.0,
148          "I_acq": 1.5434491151928815,
149          "I_use": 0.08289914666261401
150        },
151        {
152          "delta": 0.5,
153          "I_acq": 2.2116158746500787,
154          "I_use": 0.09305665042940188
155        },
156        {
157          "delta": 0.25,
158          "I_acq": 2.899102061566953,
159          "I_use": 0.09884338786391983
160        },
161        {
162          "delta": 0.125,
163          "I_acq": 3.589851585897336,
164          "I_use": 0.10102097970960289
165        },
166        {
167          "delta": 0.0625,
168          "I_acq": 4.281741767709677,
169          "I_use": 0.10177035755518295
170        },
171        {
172          "delta": 0.03125,
173          "I_acq": 4.97355053968472,
174          "I_use": 0.10251417051178979
175        }
176      ],
177      "acq_log_slope": 0.9963986336397874,
178      "acq_log_r2": 0.9999983290351192,
179      "fine_acq_increment": 0.691808771975043,
180      "fine_use_increment": 0.0007438129566068458
181    },
182    "idea_sweep": [
183      {
184        "cfg": {
185          "lr": 0.001,
186          "delta": 0.5,
187          "lambda": 0.03
188        },
189        "mean": 9.066823542118073,
190        "per_seed": [
191          11.198817253112793,
192          8.665545463562012,
193          9.605782508850098,
194          8.392980575561523,
195          7.9980645179748535,
196          9.001076698303223,
197          8.63749885559082,
198          9.034822463989258
199        ]
200      },
201      {
202        "cfg": {
203          "lr": 0.001,
204          "delta": 1.0,
205          "lambda": 0.03
206        },
207        "mean": 56.89865565299988,
208        "per_seed": [
209          13.655369758605957,
210          14.693795204162598,
211          112.16630554199219,
212          9.177323341369629,
213          139.04576110839844,
214          13.599967002868652,
215          138.39633178710938,
216          14.454391479492188
217        ]
218      },
219      {
220        "cfg": {
221          "lr": 0.001,
222          "delta": 2.0,
223          "lambda": 0.03
224        },
225        "mean": 27.349403142929077,
226        "per_seed": [
227          14.268783569335938,
228          11.69338607788086,
229          12.973438262939453,
230          11.941506385803223,
231          12.154528617858887,
232          11.955282211303711,
233          93.44461059570312,
234          50.36368942260742
235        ]
236      },
237      {
238        "cfg": {
239          "lr": 0.003,
240          "delta": 0.5,
241          "lambda": 0.03
242        },
243        "mean": 6.97602242231369,
244        "per_seed": [
245          6.515726566314697,
246          6.685352802276611,
247          7.203780651092529,
248          7.1638712882995605,
249          6.83967924118042,
250          7.354576110839844,
251          6.9066338539123535,
252          7.138558864593506
253        ]
254      },
255      {
256        "cfg": {
257          "lr": 0.006,
258          "delta": 0.5,
259          "lambda": 0.03
260        },
261        "mean": 7.816548943519592,
262        "per_seed": [
263          8.138494491577148,
264          8.348912239074707,
265          7.035199165344238,
266          6.89385461807251,
267          6.767551422119141,
268          7.8906731605529785,
269          7.620196342468262,
270          9.837510108947754
271        ]
272      }
273    ],
274    "mechanism_signature": {
275      "predicted": "finite-resolution channel suppresses noisy observations and rejects anti-aligned updates",
276      "observed": {
277        "accepted_fraction": 0.7734375,
278        "mean_raw_grad_norm": 27.050026551771833,
279        "mean_channel_update_norm": 27.05706130868445,
280        "delta": 0.5
281      },
282      "confirmed": true
283    }
284  }
285}