Missingness-as-a-Label Signal / bench_report.json
Mechanism confirmed, baseline not beaten
1{
2 "bench_version": 1,
3 "track": "uncertainty_missingness_classification",
4 "model": "mlp_tiny",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "full": {
9 "per_seed": [
10 0.25932323932647705,
11 0.32002052664756775,
12 0.2647111713886261,
13 0.27379462122917175,
14 0.334913432598114,
15 0.27944448590278625,
16 0.39740344882011414,
17 0.32050931453704834
18 ],
19 "mean": 0.3062650300562382,
20 "std": 0.04362192624557341,
21 "n": 8
22 },
23 "best_config": {
24 "lr": 0.01,
25 "wd": 0.0
26 },
27 "sweep": [
28 {
29 "config": {
30 "lr": 0.003,
31 "wd": 0.0
32 },
33 "mean_nll": 0.2802733927965164,
34 "per_seed": [
35 {
36 "nll": 0.2711596190929413,
37 "err": 0.0949999988079071
38 },
39 {
40 "nll": 0.31240639090538025,
41 "err": 0.08999999612569809
42 },
43 {
44 "nll": 0.2681650221347809,
45 "err": 0.10499999672174454
46 },
47 {
48 "nll": 0.26936253905296326,
49 "err": 0.11249999701976776
50 }
51 ]
52 },
53 {
54 "config": {
55 "lr": 0.003,
56 "wd": 0.0001
57 },
58 "mean_nll": 0.27974314987659454,
59 "per_seed": [
60 {
61 "nll": 0.27134039998054504,
62 "err": 0.0949999988079071
63 },
64 {
65 "nll": 0.3109743595123291,
66 "err": 0.08999999612569809
67 },
68 {
69 "nll": 0.2672112286090851,
70 "err": 0.10249999910593033
71 },
72 {
73 "nll": 0.26944661140441895,
74 "err": 0.11249999701976776
75 }
76 ]
77 },
78 {
79 "config": {
80 "lr": 0.01,
81 "wd": 0.0
82 },
83 "mean_nll": 0.27946238964796066,
84 "per_seed": [
85 {
86 "nll": 0.25932323932647705,
87 "err": 0.10499999672174454
88 },
89 {
90 "nll": 0.32002052664756775,
91 "err": 0.0925000011920929
92 },
93 {
94 "nll": 0.2647111713886261,
95 "err": 0.10749999433755875
96 },
97 {
98 "nll": 0.27379462122917175,
99 "err": 0.10749999433755875
100 }
101 ]
102 },
103 {
104 "config": {
105 "lr": 0.01,
106 "wd": 0.0001
107 },
108 "mean_nll": 0.283403217792511,
109 "per_seed": [
110 {
111 "nll": 0.27524203062057495,
112 "err": 0.09999999403953552
113 },
114 {
115 "nll": 0.31805169582366943,
116 "err": 0.08999999612569809
117 },
118 {
119 "nll": 0.26388776302337646,
120 "err": 0.10249999910593033
121 },
122 {
123 "nll": 0.2764313817024231,
124 "err": 0.10749999433755875
125 }
126 ]
127 },
128 {
129 "config": {
130 "lr": 0.03,
131 "wd": 0.0
132 },
133 "mean_nll": 0.30907802283763885,
134 "per_seed": [
135 {
136 "nll": 0.33665135502815247,
137 "err": 0.10999999940395355
138 },
139 {
140 "nll": 0.34054434299468994,
141 "err": 0.08749999850988388
142 },
143 {
144 "nll": 0.25733017921447754,
145 "err": 0.11249999701976776
146 },
147 {
148 "nll": 0.3017862141132355,
149 "err": 0.10749999433755875
150 }
151 ]
152 },
153 {
154 "config": {
155 "lr": 0.03,
156 "wd": 0.0001
157 },
158 "mean_nll": 0.31001827865839005,
159 "per_seed": [
160 {
161 "nll": 0.32909339666366577,
162 "err": 0.10999999940395355
163 },
164 {
165 "nll": 0.34493932127952576,
166 "err": 0.08500000089406967
167 },
168 {
169 "nll": 0.2608874440193176,
170 "err": 0.11249999701976776
171 },
172 {
173 "nll": 0.305152952671051,
174 "err": 0.10999999940395355
175 }
176 ]
177 }
178 ],
179 "full_eval_config": {
180 "lr": 0.01,
181 "wd": 0.0
182 }
183 },
184 "idea": {
185 "per_seed": [
186 0.2529575228691101,
187 0.3246569037437439,
188 0.25593283772468567,
189 0.2860279083251953,
190 0.3488270044326782,
191 0.29746517539024353,
192 0.3761008381843567,
193 0.3177599310874939
194 ],
195 "mean": 0.3074660152196884,
196 "std": 0.04032588351447927,
197 "n": 8,
198 "best_config": {
199 "lr": 0.01,
200 "wd": 0.0
201 },
202 "sweep": [
203 {
204 "config": {
205 "lr": 0.003,
206 "wd": 0.0
207 },
208 "per_seed": [
209 {
210 "nll": 0.27514806389808655,
211 "err": 0.0925000011920929,
212 "missing_corr": 0.9024436840399366,
213 "b": 0.7666630148887634,
214 "pred_missing": 0.44896629452705383,
215 "observed_missing": 0.3125
216 },
217 {
218 "nll": 0.33466601371765137,
219 "err": 0.0925000011920929,
220 "missing_corr": 0.9724430083762355,
221 "b": 0.7647979855537415,
222 "pred_missing": 0.45239901542663574,
223 "observed_missing": 0.3425000011920929
224 },
225 {
226 "nll": 0.26720207929611206,
227 "err": 0.10249999910593033,
228 "missing_corr": 0.9400309045157215,
229 "b": 0.7660092115402222,
230 "pred_missing": 0.4616314172744751,
231 "observed_missing": 0.33500000834465027
232 },
233 {
234 "nll": 0.285520076751709,
235 "err": 0.11749999970197678,
236 "missing_corr": 0.9499087957126232,
237 "b": 0.7629536986351013,
238 "pred_missing": 0.4592474699020386,
239 "observed_missing": 0.3174999952316284
240 },
241 {
242 "nll": 0.3349361717700958,
243 "err": 0.1224999949336052,
244 "missing_corr": 0.9228905173160256,
245 "b": 0.7681577205657959,
246 "pred_missing": 0.4607542157173157,
247 "observed_missing": 0.3425000011920929
248 },
249 {
250 "nll": 0.30076807737350464,
251 "err": 0.11499999463558197,
252 "missing_corr": 0.9735755481574243,
253 "b": 0.7658361196517944,
254 "pred_missing": 0.4581686854362488,
255 "observed_missing": 0.33250001072883606
256 },
257 {
258 "nll": 0.34624627232551575,
259 "err": 0.14000000059604645,
260 "missing_corr": 0.8828189879427482,
261 "b": 0.7663794159889221,
262 "pred_missing": 0.4717622399330139,
263 "observed_missing": 0.3725000023841858
264 },
265 {
266 "nll": 0.3216817080974579,
267 "err": 0.14499999582767487,
268 "missing_corr": 0.9267386520612199,
269 "b": 0.7663792967796326,
270 "pred_missing": 0.46759405732154846,
271 "observed_missing": 0.3425000011920929
272 }
273 ],
274 "mean_nll": 0.30827105790376663
275 },
276 {
277 "config": {
278 "lr": 0.01,
279 "wd": 0.0
280 },
281 "per_seed": [
282 {
283 "nll": 0.2529575228691101,
284 "err": 0.09749999642372131,
285 "missing_corr": 0.8599101025078111,
286 "b": 0.7472167015075684,
287 "pred_missing": 0.4164726138114929,
288 "observed_missing": 0.3125
289 },
290 {
291 "nll": 0.3246569037437439,
292 "err": 0.08999999612569809,
293 "missing_corr": 0.9487513902968302,
294 "b": 0.745702862739563,
295 "pred_missing": 0.41046252846717834,
296 "observed_missing": 0.3425000011920929
297 },
298 {
299 "nll": 0.25593283772468567,
300 "err": 0.11249999701976776,
301 "missing_corr": 0.9430598899116818,
302 "b": 0.7479268908500671,
303 "pred_missing": 0.4195019602775574,
304 "observed_missing": 0.33500000834465027
305 },
306 {
307 "nll": 0.2860279083251953,
308 "err": 0.11499999463558197,
309 "missing_corr": 0.9281673470315555,
310 "b": 0.7177038192749023,
311 "pred_missing": 0.4063892066478729,
312 "observed_missing": 0.3174999952316284
313 },
314 {
315 "nll": 0.3488270044326782,
316 "err": 0.125,
317 "missing_corr": 0.8866245947416767,
318 "b": 0.7491956949234009,
319 "pred_missing": 0.4097135066986084,
320 "observed_missing": 0.3425000011920929
321 },
322 {
323 "nll": 0.29746517539024353,
324 "err": 0.11749999970197678,
325 "missing_corr": 0.9490765968164979,
326 "b": 0.7454859614372253,
327 "pred_missing": 0.41396957635879517,
328 "observed_missing": 0.33250001072883606
329 },
330 {
331 "nll": 0.3761008381843567,
332 "err": 0.1525000035762787,
333 "missing_corr": 0.8037557854015652,
334 "b": 0.7294673919677734,
335 "pred_missing": 0.4210052490234375,
336 "observed_missing": 0.3725000023841858
337 },
338 {
339 "nll": 0.3177599310874939,
340 "err": 0.14000000059604645,
341 "missing_corr": 0.9021314944562635,
342 "b": 0.7316409945487976,
343 "pred_missing": 0.41623786091804504,
344 "observed_missing": 0.3425000011920929
345 }
346 ],
347 "mean_nll": 0.3074660152196884
348 },
349 {
350 "config": {
351 "lr": 0.03,
352 "wd": 0.0
353 },
354 "per_seed": [
355 {
356 "nll": 0.2912931442260742,
357 "err": 0.10249999910593033,
358 "missing_corr": 0.8220934083194001,
359 "b": 0.790917158126831,
360 "pred_missing": 0.3713911473751068,
361 "observed_missing": 0.3125
362 },
363 {
364 "nll": 0.35071563720703125,
365 "err": 0.08249999582767487,
366 "missing_corr": 0.9079289341663003,
367 "b": 0.8076502680778503,
368 "pred_missing": 0.33232399821281433,
369 "observed_missing": 0.3425000011920929
370 },
371 {
372 "nll": 0.24430160224437714,
373 "err": 0.10999999940395355,
374 "missing_corr": 0.9119544553233667,
375 "b": 0.8473079800605774,
376 "pred_missing": 0.3684086501598358,
377 "observed_missing": 0.33500000834465027
378 },
379 {
380 "nll": 0.29751330614089966,
381 "err": 0.11249999701976776,
382 "missing_corr": 0.8516922311869071,
383 "b": 0.7000229358673096,
384 "pred_missing": 0.3121287524700165,
385 "observed_missing": 0.3174999952316284
386 },
387 {
388 "nll": 0.35454967617988586,
389 "err": 0.12999999523162842,
390 "missing_corr": 0.9089462931313597,
391 "b": 0.8252794742584229,
392 "pred_missing": 0.32259058952331543,
393 "observed_missing": 0.3425000011920929
394 },
395 {
396 "nll": 0.28305864334106445,
397 "err": 0.11999999731779099,
398 "missing_corr": 0.9016399678001666,
399 "b": 0.8164690732955933,
400 "pred_missing": 0.3561153709888458,
401 "observed_missing": 0.33250001072883606
402 },
403 {
404 "nll": 0.47652503848075867,
405 "err": 0.17749999463558197,
406 "missing_corr": 0.7801492316430328,
407 "b": 0.7549465298652649,
408 "pred_missing": 0.3328292965888977,
409 "observed_missing": 0.3725000023841858
410 },
411 {
412 "nll": 0.32324108481407166,
413 "err": 0.13499999046325684,
414 "missing_corr": 0.878917930163105,
415 "b": 0.7771673798561096,
416 "pred_missing": 0.3450394868850708,
417 "observed_missing": 0.3425000011920929
418 }
419 ],
420 "mean_nll": 0.32764976657927036
421 }
422 ]
423 },
424 "comparison": {
425 "delta_mean": 0.001200985163450241,
426 "idea_wins": 4,
427 "n_pairs": 8,
428 "per_seed_diffs": [
429 -0.006365716457366943,
430 0.0046363770961761475,
431 -0.00877833366394043,
432 0.01223328709602356,
433 0.013913571834564209,
434 0.018020689487457275,
435 -0.021302610635757446,
436 -0.0027493834495544434
437 ],
438 "p_value": 0.79055,
439 "mde": 0.01120430419306732,
440 "mde_rel_pct": 3.658368763488904,
441 "verdict": "no measurable effect",
442 "system_worked": false
443 },
444 "custom_track": {
445 "name": "uncertainty_missingness_classification",
446 "file": "missingness_bench.py",
447 "domain": "label_missingness"
448 },
449 "math_check": {
450 "a_grad": -0.0441457297454092,
451 "a_expected": -0.0441457297454092,
452 "b_grad": 0.034629446878288096,
453 "b_expected": 0.03462944687828812,
454 "entropy_grad_norm": 0.10446851953879388,
455 "gradient_match": true
456 },
457 "protocol": {
458 "baseline_grid": [
459 {
460 "lr": 0.003,
461 "wd": 0.0
462 },
463 {
464 "lr": 0.003,
465 "wd": 0.0001
466 },
467 {
468 "lr": 0.01,
469 "wd": 0.0
470 },
471 {
472 "lr": 0.01,
473 "wd": 0.0001
474 },
475 {
476 "lr": 0.03,
477 "wd": 0.0
478 },
479 {
480 "lr": 0.03,
481 "wd": 0.0001
482 }
483 ],
484 "idea_grid": [
485 0.003,
486 0.01,
487 0.03
488 ],
489 "paired_deltas_nll": [
490 -0.006365716457366943,
491 0.0046363770961761475,
492 -0.00877833366394043,
493 0.01223328709602356,
494 0.013913571834564209,
495 0.018020689487457275,
496 -0.021302610635757446,
497 -0.0027493834495544434
498 ],
499 "permutation_p": 0.79055,
500 "seeds": [
501 0,
502 1,
503 2,
504 3,
505 4,
506 5,
507 6,
508 7
509 ]
510 },
511 "mechanism_signature": {
512 "predicted": "positive uncertainty-to-missingness slope and high predicted/observed missingness correlation",
513 "observed_b_mean": 0.7392925396561623,
514 "observed_missing_corr_mean": 0.9026846501454853,
515 "confirmed": true
516 }
517}