ProCreations's picture
Publish validated ICML reproduction
6c2bb8e verified
Raw
History Blame Contribute Delete
6.49 kB
{
"provenance": "Re-run of the authors' released scripts (github.com/hlzhang109/prescriptive-scaling) on the authors' released tables, on this machine.",
"table1_authors_code": {
"BBH Raw": 0.6995598819257327,
"GPQA Raw": 0.4244877017896367,
"IFEval Raw": 0.827845230576459,
"MATH Lvl 5 Raw": 0.5395001305416097,
"MMLU-PRO Raw": 0.5626259737685013,
"MUSR Raw": 0.535245164843684
},
"table2_authors_code": {
"Constant": {
"pinball_ID": 0.005352683799141467,
"pinball_OOD": 0.006233912544681612,
"calib_ID": 0.04123274010277254,
"calib_OOD": 0.03597267501958549
},
"Binwise": {
"pinball_ID": 0.004014274443578322,
"pinball_OOD": 0.004995430505610828,
"calib_ID": 0.016646429776157354,
"calib_OOD": 0.028066352913576547
},
"I-spline": {
"pinball_ID": 0.004003277885706522,
"pinball_OOD": 0.004918276548778966,
"calib_ID": 0.018259079529548153,
"calib_OOD": 0.024059338259677387
},
"Sigmoid": {
"pinball_ID": 0.004082793246777277,
"pinball_OOD": 0.00492990898200165,
"calib_ID": 0.018373223589866268,
"calib_OOD": 0.02205737092377908
}
},
"n_task_period_pairs": 18,
"claim4_authors_code": {
"BBH Raw": {
"k1_MAE_OOS_macro": 0.0162621,
"k2_MAE_OOS_macro": 0.0157885,
"k3_MAE_OOS_macro": 0.0109922,
"k1_signed_coverage_micro": 0.013671086497890289,
"k2_signed_coverage_micro": 0.01554063991081382,
"k3_signed_coverage_micro": 0.007951856224899597,
"mean_MAE_OOS_macro": 0.0143476,
"mean_signed_coverage_micro": 0.012387860877867902
},
"GPQA Raw": {
"k1_MAE_OOS_macro": 0.02,
"k2_MAE_OOS_macro": 0.0189691,
"k3_MAE_OOS_macro": 0.0175445,
"k1_signed_coverage_micro": 0.020000000000000014,
"k2_signed_coverage_micro": 0.017770189520624313,
"k3_signed_coverage_micro": 0.013574371084337357,
"mean_MAE_OOS_macro": 0.018837866666666665,
"mean_signed_coverage_micro": 0.01711485353498723
},
"IFEval Raw": {
"k1_MAE_OOS_macro": 0.048011,
"k2_MAE_OOS_macro": 0.0201847,
"k3_MAE_OOS_macro": 0.019375,
"k1_signed_coverage_micro": -0.0664979345991561,
"k2_signed_coverage_micro": -0.007870622073578595,
"k3_signed_coverage_micro": 0.019196787148594394,
"mean_MAE_OOS_macro": 0.029190233333333333,
"mean_signed_coverage_micro": -0.0183905898413801
},
"MATH Lvl 5 Raw": {
"k1_MAE_OOS_macro": 0.032517,
"k2_MAE_OOS_macro": 0.0178855,
"k3_MAE_OOS_macro": 0.0379227,
"k1_signed_coverage_micro": -0.04962008227848103,
"k2_signed_coverage_micro": -0.012329929765886248,
"k3_signed_coverage_micro": -0.03140572771084337,
"mean_MAE_OOS_macro": 0.02944173333333333,
"mean_signed_coverage_micro": -0.03111857991840355
},
"MMLU-PRO Raw": {
"k1_MAE_OOS_macro": 0.0179721,
"k2_MAE_OOS_macro": 0.0151516,
"k3_MAE_OOS_macro": 0.0170103,
"k1_signed_coverage_micro": 0.0052320675105485765,
"k2_signed_coverage_micro": 0.014425835005574148,
"k3_signed_coverage_micro": 0.016787031325301224,
"mean_MAE_OOS_macro": 0.01671133333333333,
"mean_signed_coverage_micro": 0.01214831128047465
},
"MUSR Raw": {
"k1_MAE_OOS_macro": 0.0183673,
"k2_MAE_OOS_macro": 0.0194074,
"k3_MAE_OOS_macro": 0.0181618,
"k1_signed_coverage_micro": 0.015780727848101264,
"k2_signed_coverage_micro": 0.017770347826086975,
"k3_signed_coverage_micro": 0.01277109236947792,
"mean_MAE_OOS_macro": 0.0186455,
"mean_signed_coverage_micro": 0.015440722681222053
}
},
"claim5_authors_code": {
"MAE_OOS_macro_by_task_alpha": {
"BBH Raw": {
"5": 0.16658663333333334,
"10": 0.07861816666666667,
"20": 0.06196636666666666,
"50": 0.0471685,
"100": 0.0156571
},
"GPQA Raw": {
"5": 0.0236493,
"10": 0.0199271,
"20": 0.017966333333333334,
"50": 0.023808866666666668,
"100": 0.019059033333333333
},
"IFEval Raw": {
"5": 0.15543933333333335,
"10": 0.10541753333333333,
"20": 0.0680076,
"50": 0.04751193333333334,
"100": 0.038778
},
"MATH Lvl 5 Raw": {
"5": 0.12228146666666667,
"10": 0.09935396666666667,
"20": 0.09590846666666668,
"50": 0.08460853333333333,
"100": 0.0558949
},
"MMLU-PRO Raw": {
"5": 0.1612238,
"10": 0.08756603333333333,
"20": 0.0449784,
"50": 0.039775066666666664,
"100": 0.022609100000000003
},
"MUSR Raw": {
"5": 0.029723733333333335,
"10": 0.02311313333333333,
"20": 0.021194599999999997,
"50": 0.020050266666666667,
"100": 0.019651533333333332
}
},
"macro_mean_by_alpha": {
"5": 0.10981737777777778,
"10": 0.06899932222222223,
"20": 0.05167029444444444,
"50": 0.04382052777777778,
"100": 0.028608277777777774
}
},
"claim6_authors_code": {
"X_post_max": 5.109977737428537,
"X_post_min": -1.056389658245763,
"X_pre_max": 3.3468033095963556,
"X_pre_min": -1.8152899666382492,
"X_support_hi": 3.3468033095963556,
"X_support_lo": -1.056389658245763,
"coef_group_shift_gamma": 0.5756989528983756,
"coef_intercept_a": -4.278842982074016,
"coef_slope_b": 1.3557444173948832,
"eps_clip": 1e-06,
"gamma_perm_mean": 0.2918169125031601,
"gamma_perm_q05": -0.3614995547495127,
"gamma_perm_q50": 0.3697808951699048,
"gamma_perm_q95": 0.6725590405018136,
"gamma_perm_std": 0.31078256611596033,
"gamma_perm_zscore": 0.9134426166276405,
"n_bins": 10,
"n_dropped_outside_support": 25,
"n_missing_release_date": 0,
"n_perm": 2000,
"n_post": 62,
"n_post_release": 86,
"n_pre": 28,
"n_pre_release": 29,
"n_total_merged": 115,
"n_used_in_support": 90,
"n_used_total": 115,
"n_with_release_date": 115,
"odds_ratio_group_shift": 1.7783730916431397,
"pvalue_one_sided_gamma_gt_0": 0.14842578710644677,
"pvalue_two_sided": 0.1489255372313843,
"r2_logit_fit": 0.5288603684522073,
"release_cutoff": "2025-02-06",
"se_group_shift_gamma": 0.32613271712611586,
"seed": 0,
"t_group_shift_gamma": 1.7652290698444468,
"test_name": "logit_to_logit_group_shift"
}
}