Kac-rotated fast projection / bench_report.json

Failed on benchmark

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": 0.01,
 10      "T": 40
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "T": 40
 17        },
 18        "mean": 178.87363052368164
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "T": 80
 24        },
 25        "mean": 178.87363052368164
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.001,
 30          "T": 160
 31        },
 32        "mean": 178.87363052368164
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "T": 40
 38        },
 39        "mean": 19.137221336364746
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "T": 80
 45        },
 46        "mean": 19.137221336364746
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "T": 160
 52        },
 53        "mean": 19.137221336364746
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.01,
 58          "T": 40
 59        },
 60        "mean": 15.155841588973999
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.01,
 65          "T": 80
 66        },
 67        "mean": 15.155841588973999
 68      },
 69      {
 70        "cfg": {
 71          "lr": 0.01,
 72          "T": 160
 73        },
 74        "mean": 15.155841588973999
 75      }
 76    ],
 77    "full": {
 78      "mean": 15.751612901687622,
 79      "std": 1.864652435728868,
 80      "per_seed": [
 81        16.355323791503906,
 82        12.874679565429688,
 83        16.660974502563477,
 84        14.732388496398926,
 85        16.593015670776367,
 86        13.093815803527832,
 87        17.25185203552246,
 88        18.45085334777832
 89      ],
 90      "n": 8
 91    }
 92  },
 93  "idea": {
 94    "mean": 17.193141102790833,
 95    "std": 3.2702713922060123,
 96    "per_seed": [
 97      18.909334182739258,
 98      14.264732360839844,
 99      21.892744064331055,
100      20.79402732849121,
101      11.515158653259277,
102      14.863554000854492,
103      18.321683883666992,
104      16.98389434814453
105    ],
106    "n": 8
107  },
108  "comparison": {
109    "delta_mean": 1.4415282011032104,
110    "idea_wins": 2,
111    "n_pairs": 8,
112    "per_seed_diffs": [
113      2.5540103912353516,
114      1.3900527954101562,
115      5.231769561767578,
116      6.061638832092285,
117      -5.07785701751709,
118      1.7697381973266602,
119      1.0698318481445312,
120      -1.466958999633789
121    ],
122    "p_value": 0.26985,
123    "mde": 2.9673645326602944,
124    "mde_rel_pct": 18.838480549140286,
125    "verdict": "no significant win",
126    "system_worked": false
127  },
128  "mechanism_signature": {
129    "prediction": {
130      "statistic": "scaled projected input norm ratio",
131      "prediction": 1.0,
132      "observed_mean": 0.9315284490585327,
133      "observed_abs_error": 0.06847155094146729,
134      "n_observations": 3200,
135      "tolerance": 0.08,
136      "confirmed": true
137    },
138    "idea_cfg": {
139      "lr": 0.01,
140      "T": 80
141    },
142    "baseline_cfg": {
143      "lr": 0.01,
144      "T": 40
145    },
146    "method": "fixed projection preprocessing plus identically trained MLP"
147  },
148  "idea_sweep": [
149    {
150      "cfg": {
151        "lr": 0.001,
152        "T": 40
153      },
154      "mean": 186.76729202270508,
155      "std": 11.439657972898413,
156      "per_seed": [
157        197.9423370361328,
158        198.1376190185547,
159        196.3520050048828,
160        165.76368713378906,
161        176.966796875,
162        177.15956115722656,
163        186.6017608642578,
164        195.21456909179688
165      ],
166      "n": 8
167    },
168    {
169      "cfg": {
170        "lr": 0.003,
171        "T": 40
172      },
173      "mean": 25.9119371175766,
174      "std": 8.593153852711152,
175      "per_seed": [
176        29.929969787597656,
177        43.78696823120117,
178        20.91279411315918,
179        15.755831718444824,
180        21.09282112121582,
181        22.17743682861328,
182        20.18206024169922,
183        33.45761489868164
184      ],
185      "n": 8
186    },
187    {
188      "cfg": {
189        "lr": 0.01,
190        "T": 40
191      },
192      "mean": 19.208696007728577,
193      "std": 4.840492557796767,
194      "per_seed": [
195        21.061805725097656,
196        28.855024337768555,
197        17.191299438476562,
198        14.6019287109375,
199        17.883209228515625,
200        17.179380416870117,
201        13.033936500549316,
202        23.86298370361328
203      ],
204      "n": 8
205    },
206    {
207      "cfg": {
208        "lr": 0.001,
209        "T": 80
210      },
211      "mean": 186.9507598876953,
212      "std": 10.651138931586377,
213      "per_seed": [
214        191.85867309570312,
215        200.1656036376953,
216        192.61083984375,
217        163.03077697753906,
218        188.59230041503906,
219        179.8059844970703,
220        185.5874786376953,
221        193.9544219970703
222      ],
223      "n": 8
224    },
225    {
226      "cfg": {
227        "lr": 0.003,
228        "T": 80
229      },
230      "mean": 25.6825692653656,
231      "std": 7.43283927020537,
232      "per_seed": [
233        20.19032096862793,
234        40.638572692871094,
235        28.931119918823242,
236        26.276248931884766,
237        17.19772720336914,
238        16.450904846191406,
239        25.62046241760254,
240        30.155197143554688
241      ],
242      "n": 8
243    },
244    {
245      "cfg": {
246        "lr": 0.01,
247        "T": 80
248      },
249      "mean": 17.193141102790833,
250      "std": 3.2702713922060123,
251      "per_seed": [
252        18.909334182739258,
253        14.264732360839844,
254        21.892744064331055,
255        20.79402732849121,
256        11.515158653259277,
257        14.863554000854492,
258        18.321683883666992,
259        16.98389434814453
260      ],
261      "n": 8
262    },
263    {
264      "cfg": {
265        "lr": 0.001,
266        "T": 160
267      },
268      "mean": 185.08572959899902,
269      "std": 10.167016861250847,
270      "per_seed": [
271        191.33917236328125,
272        197.4365234375,
273        181.3788604736328,
274        161.34991455078125,
275        192.1107940673828,
276        187.85206604003906,
277        185.2051544189453,
278        184.0133514404297
279      ],
280      "n": 8
281    },
282    {
283      "cfg": {
284        "lr": 0.003,
285        "T": 160
286      },
287      "mean": 25.298375844955444,
288      "std": 4.551763171760175,
289      "per_seed": [
290        18.683156967163086,
291        33.22486877441406,
292        29.26276206970215,
293        22.208969116210938,
294        21.04352569580078,
295        24.836837768554688,
296        29.051376342773438,
297        24.075510025024414
298      ],
299      "n": 8
300    },
301    {
302      "cfg": {
303        "lr": 0.01,
304        "T": 160
305      },
306      "mean": 18.80278468132019,
307      "std": 2.1794834679139643,
308      "per_seed": [
309        15.243776321411133,
310        17.74627113342285,
311        17.490156173706055,
312        19.760400772094727,
313        17.134641647338867,
314        20.429834365844727,
315        19.96878433227539,
316        22.648412704467773
317      ],
318      "n": 8
319    }
320  ],
321  "protocol_notes": {
322    "paired_seeds": [
323      0,
324      1,
325      2,
326      3,
327      4,
328      5,
329      6,
330      7
331    ],
332    "epochs": 8,
333    "n_train": 400,
334    "n_test": 400,
335    "structural_match": "tabular MLP projection bottleneck",
336    "baseline_lr_union": [
337      0.001,
338      0.003,
339      0.01
340    ],
341    "baseline_method_knob": "projection width M=6",
342    "projection_note": "Kac stream is precomputed once because it is fixed; this is algebraically identical to inserting it before the MLP."
343  }
344}