{ "bench_version": 1, "track": "uncertainty_missingness_classification", "model": "mlp_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "full": { "per_seed": [ 0.25932323932647705, 0.32002052664756775, 0.2647111713886261, 0.27379462122917175, 0.334913432598114, 0.27944448590278625, 0.39740344882011414, 0.32050931453704834 ], "mean": 0.3062650300562382, "std": 0.04362192624557341, "n": 8 }, "best_config": { "lr": 0.01, "wd": 0.0 }, "sweep": [ { "config": { "lr": 0.003, "wd": 0.0 }, "mean_nll": 0.2802733927965164, "per_seed": [ { "nll": 0.2711596190929413, "err": 0.0949999988079071 }, { "nll": 0.31240639090538025, "err": 0.08999999612569809 }, { "nll": 0.2681650221347809, "err": 0.10499999672174454 }, { "nll": 0.26936253905296326, "err": 0.11249999701976776 } ] }, { "config": { "lr": 0.003, "wd": 0.0001 }, "mean_nll": 0.27974314987659454, "per_seed": [ { "nll": 0.27134039998054504, "err": 0.0949999988079071 }, { "nll": 0.3109743595123291, "err": 0.08999999612569809 }, { "nll": 0.2672112286090851, "err": 0.10249999910593033 }, { "nll": 0.26944661140441895, "err": 0.11249999701976776 } ] }, { "config": { "lr": 0.01, "wd": 0.0 }, "mean_nll": 0.27946238964796066, "per_seed": [ { "nll": 0.25932323932647705, "err": 0.10499999672174454 }, { "nll": 0.32002052664756775, "err": 0.0925000011920929 }, { "nll": 0.2647111713886261, "err": 0.10749999433755875 }, { "nll": 0.27379462122917175, "err": 0.10749999433755875 } ] }, { "config": { "lr": 0.01, "wd": 0.0001 }, "mean_nll": 0.283403217792511, "per_seed": [ { "nll": 0.27524203062057495, "err": 0.09999999403953552 }, { "nll": 0.31805169582366943, "err": 0.08999999612569809 }, { "nll": 0.26388776302337646, "err": 0.10249999910593033 }, { "nll": 0.2764313817024231, "err": 0.10749999433755875 } ] }, { "config": { "lr": 0.03, "wd": 0.0 }, "mean_nll": 0.30907802283763885, "per_seed": [ { "nll": 0.33665135502815247, "err": 0.10999999940395355 }, { "nll": 0.34054434299468994, "err": 0.08749999850988388 }, { "nll": 0.25733017921447754, "err": 0.11249999701976776 }, { "nll": 0.3017862141132355, "err": 0.10749999433755875 } ] }, { "config": { "lr": 0.03, "wd": 0.0001 }, "mean_nll": 0.31001827865839005, "per_seed": [ { "nll": 0.32909339666366577, "err": 0.10999999940395355 }, { "nll": 0.34493932127952576, "err": 0.08500000089406967 }, { "nll": 0.2608874440193176, "err": 0.11249999701976776 }, { "nll": 0.305152952671051, "err": 0.10999999940395355 } ] } ], "full_eval_config": { "lr": 0.01, "wd": 0.0 } }, "idea": { "per_seed": [ 0.2529575228691101, 0.3246569037437439, 0.25593283772468567, 0.2860279083251953, 0.3488270044326782, 0.29746517539024353, 0.3761008381843567, 0.3177599310874939 ], "mean": 0.3074660152196884, "std": 0.04032588351447927, "n": 8, "best_config": { "lr": 0.01, "wd": 0.0 }, "sweep": [ { "config": { "lr": 0.003, "wd": 0.0 }, "per_seed": [ { "nll": 0.27514806389808655, "err": 0.0925000011920929, "missing_corr": 0.9024436840399366, "b": 0.7666630148887634, "pred_missing": 0.44896629452705383, "observed_missing": 0.3125 }, { "nll": 0.33466601371765137, "err": 0.0925000011920929, "missing_corr": 0.9724430083762355, "b": 0.7647979855537415, "pred_missing": 0.45239901542663574, "observed_missing": 0.3425000011920929 }, { "nll": 0.26720207929611206, "err": 0.10249999910593033, "missing_corr": 0.9400309045157215, "b": 0.7660092115402222, "pred_missing": 0.4616314172744751, "observed_missing": 0.33500000834465027 }, { "nll": 0.285520076751709, "err": 0.11749999970197678, "missing_corr": 0.9499087957126232, "b": 0.7629536986351013, "pred_missing": 0.4592474699020386, "observed_missing": 0.3174999952316284 }, { "nll": 0.3349361717700958, "err": 0.1224999949336052, "missing_corr": 0.9228905173160256, "b": 0.7681577205657959, "pred_missing": 0.4607542157173157, "observed_missing": 0.3425000011920929 }, { "nll": 0.30076807737350464, "err": 0.11499999463558197, "missing_corr": 0.9735755481574243, "b": 0.7658361196517944, "pred_missing": 0.4581686854362488, "observed_missing": 0.33250001072883606 }, { "nll": 0.34624627232551575, "err": 0.14000000059604645, "missing_corr": 0.8828189879427482, "b": 0.7663794159889221, "pred_missing": 0.4717622399330139, "observed_missing": 0.3725000023841858 }, { "nll": 0.3216817080974579, "err": 0.14499999582767487, "missing_corr": 0.9267386520612199, "b": 0.7663792967796326, "pred_missing": 0.46759405732154846, "observed_missing": 0.3425000011920929 } ], "mean_nll": 0.30827105790376663 }, { "config": { "lr": 0.01, "wd": 0.0 }, "per_seed": [ { "nll": 0.2529575228691101, "err": 0.09749999642372131, "missing_corr": 0.8599101025078111, "b": 0.7472167015075684, "pred_missing": 0.4164726138114929, "observed_missing": 0.3125 }, { "nll": 0.3246569037437439, "err": 0.08999999612569809, "missing_corr": 0.9487513902968302, "b": 0.745702862739563, "pred_missing": 0.41046252846717834, "observed_missing": 0.3425000011920929 }, { "nll": 0.25593283772468567, "err": 0.11249999701976776, "missing_corr": 0.9430598899116818, "b": 0.7479268908500671, "pred_missing": 0.4195019602775574, "observed_missing": 0.33500000834465027 }, { "nll": 0.2860279083251953, "err": 0.11499999463558197, "missing_corr": 0.9281673470315555, "b": 0.7177038192749023, "pred_missing": 0.4063892066478729, "observed_missing": 0.3174999952316284 }, { "nll": 0.3488270044326782, "err": 0.125, "missing_corr": 0.8866245947416767, "b": 0.7491956949234009, "pred_missing": 0.4097135066986084, "observed_missing": 0.3425000011920929 }, { "nll": 0.29746517539024353, "err": 0.11749999970197678, "missing_corr": 0.9490765968164979, "b": 0.7454859614372253, "pred_missing": 0.41396957635879517, "observed_missing": 0.33250001072883606 }, { "nll": 0.3761008381843567, "err": 0.1525000035762787, "missing_corr": 0.8037557854015652, "b": 0.7294673919677734, "pred_missing": 0.4210052490234375, "observed_missing": 0.3725000023841858 }, { "nll": 0.3177599310874939, "err": 0.14000000059604645, "missing_corr": 0.9021314944562635, "b": 0.7316409945487976, "pred_missing": 0.41623786091804504, "observed_missing": 0.3425000011920929 } ], "mean_nll": 0.3074660152196884 }, { "config": { "lr": 0.03, "wd": 0.0 }, "per_seed": [ { "nll": 0.2912931442260742, "err": 0.10249999910593033, "missing_corr": 0.8220934083194001, "b": 0.790917158126831, "pred_missing": 0.3713911473751068, "observed_missing": 0.3125 }, { "nll": 0.35071563720703125, "err": 0.08249999582767487, "missing_corr": 0.9079289341663003, "b": 0.8076502680778503, "pred_missing": 0.33232399821281433, "observed_missing": 0.3425000011920929 }, { "nll": 0.24430160224437714, "err": 0.10999999940395355, "missing_corr": 0.9119544553233667, "b": 0.8473079800605774, "pred_missing": 0.3684086501598358, "observed_missing": 0.33500000834465027 }, { "nll": 0.29751330614089966, "err": 0.11249999701976776, "missing_corr": 0.8516922311869071, "b": 0.7000229358673096, "pred_missing": 0.3121287524700165, "observed_missing": 0.3174999952316284 }, { "nll": 0.35454967617988586, "err": 0.12999999523162842, "missing_corr": 0.9089462931313597, "b": 0.8252794742584229, "pred_missing": 0.32259058952331543, "observed_missing": 0.3425000011920929 }, { "nll": 0.28305864334106445, "err": 0.11999999731779099, "missing_corr": 0.9016399678001666, "b": 0.8164690732955933, "pred_missing": 0.3561153709888458, "observed_missing": 0.33250001072883606 }, { "nll": 0.47652503848075867, "err": 0.17749999463558197, "missing_corr": 0.7801492316430328, "b": 0.7549465298652649, "pred_missing": 0.3328292965888977, "observed_missing": 0.3725000023841858 }, { "nll": 0.32324108481407166, "err": 0.13499999046325684, "missing_corr": 0.878917930163105, "b": 0.7771673798561096, "pred_missing": 0.3450394868850708, "observed_missing": 0.3425000011920929 } ], "mean_nll": 0.32764976657927036 } ] }, "comparison": { "delta_mean": 0.001200985163450241, "idea_wins": 4, "n_pairs": 8, "per_seed_diffs": [ -0.006365716457366943, 0.0046363770961761475, -0.00877833366394043, 0.01223328709602356, 0.013913571834564209, 0.018020689487457275, -0.021302610635757446, -0.0027493834495544434 ], "p_value": 0.79055, "mde": 0.01120430419306732, "mde_rel_pct": 3.658368763488904, "verdict": "no measurable effect", "system_worked": false }, "custom_track": { "name": "uncertainty_missingness_classification", "file": "missingness_bench.py", "domain": "label_missingness" }, "math_check": { "a_grad": -0.0441457297454092, "a_expected": -0.0441457297454092, "b_grad": 0.034629446878288096, "b_expected": 0.03462944687828812, "entropy_grad_norm": 0.10446851953879388, "gradient_match": true }, "protocol": { "baseline_grid": [ { "lr": 0.003, "wd": 0.0 }, { "lr": 0.003, "wd": 0.0001 }, { "lr": 0.01, "wd": 0.0 }, { "lr": 0.01, "wd": 0.0001 }, { "lr": 0.03, "wd": 0.0 }, { "lr": 0.03, "wd": 0.0001 } ], "idea_grid": [ 0.003, 0.01, 0.03 ], "paired_deltas_nll": [ -0.006365716457366943, 0.0046363770961761475, -0.00877833366394043, 0.01223328709602356, 0.013913571834564209, 0.018020689487457275, -0.021302610635757446, -0.0027493834495544434 ], "permutation_p": 0.79055, "seeds": [ 0, 1, 2, 3, 4, 5, 6, 7 ] }, "mechanism_signature": { "predicted": "positive uncertainty-to-missingness slope and high predicted/observed missingness correlation", "observed_b_mean": 0.7392925396561623, "observed_missing_corr_mean": 0.9026846501454853, "confirmed": true } }