SecJev-4B / evaluation /results.json
macmacmacmac's picture
Complete runnable SecJev checkpoint for jev-stageA JEV defense
caf14ad verified
Raw History Blame Contribute Delete
294 kB
{
"selected_epoch": 3,
"epochs": [
{
"epoch": 1,
"development": {
"raw": {
"questions": 14912,
"accuracy": 0.8959898068669528,
"mean_class_recall": 0.8086847041598828,
"nll": 0.2676199905412643,
"brier": 0.14581168647550916,
"ece15": 0.024724176262894793,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.773806330472103,
"error_rate_at_confidence_0_9": 0.0240055464078343,
"ordinal_expected_level_mae": 0.00949861612167232,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 222,
"accuracy": 0.9819819819819819,
"mean_class_recall": 0.9670930232558139,
"nll": 0.06952124509948641,
"brier": 0.033360936223710204,
"ece15": 0.01963763286720348,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9774774774774775,
"error_rate_at_confidence_0_9": 0.013824884792626729,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 222,
"accuracy": 0.9864864864864865,
"mean_class_recall": 0.97,
"nll": 0.06796588080667874,
"brier": 0.02442578405456454,
"ece15": 0.013534851962772593,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.972972972972973,
"error_rate_at_confidence_0_9": 0.009259259259259259,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 1400,
"accuracy": 0.9742857142857143,
"mean_class_recall": 0.9504464285714286,
"nll": 0.07440929350732509,
"brier": 0.03861493679124129,
"ece15": 0.01061565077999396,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.925,
"error_rate_at_confidence_0_9": 0.008494208494208495,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 5000,
"accuracy": 0.8334,
"mean_class_recall": 0.7313333333333334,
"nll": 0.4357438451410003,
"brier": 0.2420765591030422,
"ece15": 0.05574236954650583,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.644,
"error_rate_at_confidence_0_9": 0.06677018633540373,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.004390672226438891,
"brier": 0.0015490359962901127,
"ece15": 0.003915493134373699,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9919354838709677,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.004780661486616687,
"brier": 0.0026501860965206722,
"ece15": 0.0038321931288341792,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9908536585365854,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 2.8436047490948972e-05,
"brier": 2.213459113159646e-07,
"ece15": 2.8380504783511674e-05,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 2500,
"accuracy": 0.9832,
"mean_class_recall": 0.9856,
"nll": 0.0716936772308886,
"brier": 0.032718644101254024,
"ece15": 0.01640596644430267,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9948,
"error_rate_at_confidence_0_9": 0.01648572577402493,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 797,
"accuracy": 0.9937264742785445,
"mean_class_recall": 0.9934913409721132,
"nll": 0.0153080496936535,
"brier": 0.009627399097334195,
"ece15": 0.004599932218083785,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9912170639899623,
"error_rate_at_confidence_0_9": 0.0037974683544303796,
"ordinal_expected_level_mae": 0.00949861612167232
},
"twins_streamlet:twins_policy_action": {
"questions": 797,
"accuracy": 0.9924717691342535,
"mean_class_recall": 0.9923419156847569,
"nll": 0.017883068917921716,
"brier": 0.011294880553062277,
"ece15": 0.0038184765498114437,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9811794228356336,
"error_rate_at_confidence_0_9": 0.0012787723785166241,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 797,
"accuracy": 0.9962358845671268,
"mean_class_recall": 0.995834315034425,
"nll": 0.011957269820791244,
"brier": 0.007393224542639707,
"ece15": 0.00670405372457154,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9836888331242158,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 797,
"accuracy": 0.9974905897114178,
"mean_class_recall": 0.9969442527413541,
"nll": 0.010437285979484219,
"brier": 0.0038211519865233618,
"ece15": 0.0009184477173700703,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9962358845671268,
"error_rate_at_confidence_0_9": 0.0012594458438287153,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 800,
"accuracy": 0.63,
"mean_class_recall": 0.6354549163606691,
"nll": 0.5876157448193948,
"brier": 0.4142276501377162,
"ece15": 0.031917999077751745,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.10375,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 800,
"accuracy": 0.59875,
"mean_class_recall": 0.4115791153670771,
"nll": 1.2269806519983115,
"brier": 0.571523062943736,
"ece15": 0.08060690424281568,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.11875,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9262877786032521,
"macro_task_balanced_accuracy": 0.9021513315229265
},
"recalibrated": {
"questions": 14912,
"accuracy": 0.8959898068669528,
"mean_class_recall": 0.8086847041598828,
"nll": 0.2562444405938866,
"brier": 0.1441282781781607,
"ece15": 0.016168446150980925,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.7631437768240343,
"error_rate_at_confidence_0_9": 0.020913884007029877,
"ordinal_expected_level_mae": 0.010086743500138846,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 222,
"accuracy": 0.9819819819819819,
"mean_class_recall": 0.9670930232558139,
"nll": 0.062064251368360995,
"brier": 0.03274641091592082,
"ece15": 0.018974706155035728,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9774774774774775,
"error_rate_at_confidence_0_9": 0.013824884792626729,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 222,
"accuracy": 0.9864864864864865,
"mean_class_recall": 0.97,
"nll": 0.06063186595056717,
"brier": 0.02487828836064738,
"ece15": 0.013603842355616328,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.972972972972973,
"error_rate_at_confidence_0_9": 0.009259259259259259,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 1400,
"accuracy": 0.9742857142857143,
"mean_class_recall": 0.9504464285714286,
"nll": 0.08099031193469938,
"brier": 0.03952213251707266,
"ece15": 0.021719508126191233,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9007142857142857,
"error_rate_at_confidence_0_9": 0.007137192704203013,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 5000,
"accuracy": 0.8334,
"mean_class_recall": 0.7313333333333334,
"nll": 0.4055190039113824,
"brier": 0.23828084117194667,
"ece15": 0.047374338949895636,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.6276,
"error_rate_at_confidence_0_9": 0.058954748247291265,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0058641097080354134,
"brier": 0.002007000102514143,
"ece15": 0.0052393541355612895,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9838709677419355,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.005681917468602561,
"brier": 0.002881636759121932,
"ece15": 0.004649618743350316,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9908536585365854,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 7.967277164193898e-05,
"brier": 1.0603152088487418e-06,
"ece15": 7.940563040823767e-05,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 2500,
"accuracy": 0.9832,
"mean_class_recall": 0.9856,
"nll": 0.06283986398882055,
"brier": 0.031969827800453,
"ece15": 0.015381335792818867,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.992,
"error_rate_at_confidence_0_9": 0.014516129032258065,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 797,
"accuracy": 0.9937264742785445,
"mean_class_recall": 0.9934913409721132,
"nll": 0.014979010411112437,
"brier": 0.009232024594375448,
"ece15": 0.0059751389669031076,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.986198243412798,
"error_rate_at_confidence_0_9": 0.001272264631043257,
"ordinal_expected_level_mae": 0.010086743500138846
},
"twins_streamlet:twins_policy_action": {
"questions": 797,
"accuracy": 0.9924717691342535,
"mean_class_recall": 0.9923419156847569,
"nll": 0.017946444333002753,
"brier": 0.011016825969457151,
"ece15": 0.005758410373299677,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9799247176913425,
"error_rate_at_confidence_0_9": 0.0012804097311139564,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 797,
"accuracy": 0.9962358845671268,
"mean_class_recall": 0.995834315034425,
"nll": 0.012786451464898273,
"brier": 0.00742639513463676,
"ece15": 0.007511975455200797,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9824341279799247,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 797,
"accuracy": 0.9974905897114178,
"mean_class_recall": 0.9969442527413541,
"nll": 0.010010236373401123,
"brier": 0.003879830332877274,
"ece15": 0.0006442096172233073,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9962358845671268,
"error_rate_at_confidence_0_9": 0.0012594458438287153,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 800,
"accuracy": 0.63,
"mean_class_recall": 0.6354549163606691,
"nll": 0.5899608657261348,
"brier": 0.41478182050444184,
"ece15": 0.02797725458606776,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.095,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 800,
"accuracy": 0.59875,
"mean_class_recall": 0.4115791153670771,
"nll": 1.22100156554001,
"brier": 0.5645239942686594,
"ece15": 0.03015245360912201,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.09125,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9262877786032521,
"macro_task_balanced_accuracy": 0.9021513315229265
}
},
"test": {
"raw": {
"questions": 21634,
"accuracy": 0.8587870943884626,
"mean_class_recall": 0.8070634668409765,
"nll": 0.38882363258516517,
"brier": 0.19821716542849313,
"ece15": 0.010114739193927335,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.6294259036701488,
"error_rate_at_confidence_0_9": 0.016596900932657706,
"ordinal_expected_level_mae": 0.010228305213524986,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 426,
"accuracy": 0.9953051643192489,
"mean_class_recall": 0.9895833333333333,
"nll": 0.025162438436180455,
"brier": 0.012711490628916015,
"ece15": 0.009157764610888798,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9694835680751174,
"error_rate_at_confidence_0_9": 0.002421307506053269,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 426,
"accuracy": 0.9835680751173709,
"mean_class_recall": 0.9635416666666667,
"nll": 0.039729487219800165,
"brier": 0.024078530771277282,
"ece15": 0.008718119493403046,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9765258215962441,
"error_rate_at_confidence_0_9": 0.007211538461538462,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 800,
"accuracy": 0.93875,
"mean_class_recall": 0.8984375,
"nll": 0.19453941595728141,
"brier": 0.09689065974959639,
"ece15": 0.027271759841056658,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.87875,
"error_rate_at_confidence_0_9": 0.034139402560455195,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 6000,
"accuracy": 0.9085,
"mean_class_recall": 0.8498888888888889,
"nll": 0.30026072378768554,
"brier": 0.1436264364242417,
"ece15": 0.037464827151113794,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.6986666666666667,
"error_rate_at_confidence_0_9": 0.029818702290076337,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0009360610136724553,
"brier": 5.531806653087891e-05,
"ece15": 0.0009217275317422124,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0029619998489780613,
"brier": 0.0010650062308243538,
"ece15": 0.0026318040347967494,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9925925925925926,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 4.009888988278539e-05,
"brier": 8.791167963328194e-07,
"ece15": 3.98768557623178e-05,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 3500,
"accuracy": 0.9882857142857143,
"mean_class_recall": 0.9807619047619047,
"nll": 0.0658259406128984,
"brier": 0.02104448067811733,
"ece15": 0.0069241929022703305,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9845714285714285,
"error_rate_at_confidence_0_9": 0.00783517121300058,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 659,
"accuracy": 0.9908952959028832,
"mean_class_recall": 0.9899995024628091,
"nll": 0.022081750206618136,
"brier": 0.01158884274796682,
"ece15": 0.006288646106945627,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9954476479514416,
"error_rate_at_confidence_0_9": 0.004573170731707317,
"ordinal_expected_level_mae": 0.010228305213524986
},
"twins_streamlet:twins_policy_action": {
"questions": 659,
"accuracy": 0.9939301972685888,
"mean_class_recall": 0.9932998324958123,
"nll": 0.017255345447197507,
"brier": 0.008938281239516815,
"ece15": 0.004558085120676935,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9893778452200304,
"error_rate_at_confidence_0_9": 0.003067484662576687,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 659,
"accuracy": 0.9969650986342944,
"mean_class_recall": 0.9966867772750125,
"nll": 0.005807395652175951,
"brier": 0.0034083401253427324,
"ece15": 0.0031822580740001574,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9893778452200304,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 659,
"accuracy": 0.9969650986342944,
"mean_class_recall": 0.9949748743718593,
"nll": 0.009270762829542105,
"brier": 0.00610585714141352,
"ece15": 0.0051900114831071945,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9878603945371776,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 3321,
"accuracy": 0.6546221017765733,
"mean_class_recall": 0.6604367829458773,
"nll": 0.6040832398786956,
"brier": 0.4230681455327693,
"ece15": 0.012144611565586284,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.08220415537488708,
"error_rate_at_confidence_0_9": 0.014652014652014652,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 3321,
"accuracy": 0.6248118036735922,
"mean_class_recall": 0.3856953745435135,
"nll": 1.250473514719575,
"brier": 0.5523161094538271,
"ece15": 0.04328702573140643,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.1093044263775971,
"error_rate_at_confidence_0_9": 0.10192837465564739,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9337570392580401,
"macro_task_balanced_accuracy": 0.9073790312675484
},
"recalibrated": {
"questions": 21634,
"accuracy": 0.8587870943884626,
"mean_class_recall": 0.8070634668409765,
"nll": 0.3819588765605003,
"brier": 0.1988268469084605,
"ece15": 0.019530087212558746,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.6154663954885828,
"error_rate_at_confidence_0_9": 0.014344723995493804,
"ordinal_expected_level_mae": 0.01072047331602824,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 426,
"accuracy": 0.9953051643192489,
"mean_class_recall": 0.9895833333333333,
"nll": 0.028102839471687808,
"brier": 0.013352375860260531,
"ece15": 0.011924724016462706,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9624413145539906,
"error_rate_at_confidence_0_9": 0.0024390243902439024,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 426,
"accuracy": 0.9835680751173709,
"mean_class_recall": 0.9635416666666667,
"nll": 0.03830754803404818,
"brier": 0.02344796092548543,
"ece15": 0.009297528928361507,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9694835680751174,
"error_rate_at_confidence_0_9": 0.004842615012106538,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 800,
"accuracy": 0.93875,
"mean_class_recall": 0.8984375,
"nll": 0.18853618602333774,
"brier": 0.09702707273078204,
"ece15": 0.01936307032837748,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.84125,
"error_rate_at_confidence_0_9": 0.03417533432392273,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 6000,
"accuracy": 0.9085,
"mean_class_recall": 0.8498888888888889,
"nll": 0.2843389317505148,
"brier": 0.14586901269531594,
"ece15": 0.04356789248344108,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.6861666666666667,
"error_rate_at_confidence_0_9": 0.028175856205975226,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0018322563352568268,
"brier": 0.0001319762188887517,
"ece15": 0.0017975613028342768,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.004015391153194959,
"brier": 0.0013677195263880856,
"ece15": 0.0035859188244034786,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9888888888888889,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 9.895528751872936e-05,
"brier": 3.024028962806258e-06,
"ece15": 9.818509110354407e-05,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 3500,
"accuracy": 0.9882857142857143,
"mean_class_recall": 0.9807619047619047,
"nll": 0.059841713174448447,
"brier": 0.021127495151853498,
"ece15": 0.005242220550012023,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.976,
"error_rate_at_confidence_0_9": 0.007318501170960188,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 659,
"accuracy": 0.9908952959028832,
"mean_class_recall": 0.9899995024628091,
"nll": 0.020306967381801818,
"brier": 0.011159005956953514,
"ece15": 0.005337269869592108,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9908952959028832,
"error_rate_at_confidence_0_9": 0.0030627871362940277,
"ordinal_expected_level_mae": 0.01072047331602824
},
"twins_streamlet:twins_policy_action": {
"questions": 659,
"accuracy": 0.9939301972685888,
"mean_class_recall": 0.9932998324958123,
"nll": 0.016710459187929856,
"brier": 0.008886576198762068,
"ece15": 0.0032207054419096,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9893778452200304,
"error_rate_at_confidence_0_9": 0.003067484662576687,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 659,
"accuracy": 0.9969650986342944,
"mean_class_recall": 0.9966867772750125,
"nll": 0.006802759775612477,
"brier": 0.003478846491952278,
"ece15": 0.004168012801242965,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9893778452200304,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 659,
"accuracy": 0.9969650986342944,
"mean_class_recall": 0.9949748743718593,
"nll": 0.009405080434222151,
"brier": 0.006029858296029803,
"ece15": 0.004873038962556261,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.9863429438543247,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 3321,
"accuracy": 0.6546221017765733,
"mean_class_recall": 0.6604367829458773,
"nll": 0.6050451348525548,
"brier": 0.4236393366516397,
"ece15": 0.022216352122328073,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.0671484492622704,
"error_rate_at_confidence_0_9": 0.008968609865470852,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 3321,
"accuracy": 0.6248118036735922,
"mean_class_recall": 0.3856953745435135,
"nll": 1.2411381163756228,
"brier": 0.5515874786124301,
"ece15": 0.032556668330635724,
"temperature": 1.1759291426711453,
"coverage_at_confidence_0_9": 0.07768744354110207,
"error_rate_at_confidence_0_9": 0.06976744186046512,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9337570392580401,
"macro_task_balanced_accuracy": 0.9073790312675484
}
},
"general": {
"decision-v7": {
"raw_clean": {
"n": 1200,
"nll": 0.5075688394241022,
"acc": 0.8666666666666667,
"ece": 0.07962904500464703,
"brier": 0.2102660845468813,
"mean_conf": 0.9440260985257297,
"confident_error_rate": 0.0575,
"coverage_at_0_9": 0.8266666666666667,
"accuracy_at_0_9": 0.9304435483870968,
"coverage_at_5pct_error": 0.7891666666666667,
"coverage_at_1pct_error": 0.6058333333333333,
"aurc": 0.028250531410146766,
"error_rate_at_0_9": 0.06955645161290322,
"confidence_bias": 0.07735943185906302,
"top_bins": {
"0.9": {
"n": 992,
"errors": 69,
"error_rate": 0.06955645161290322
},
"0.95": {
"n": 931,
"errors": 44,
"error_rate": 0.047261009667024706
},
"0.99": {
"n": 853,
"errors": 25,
"error_rate": 0.029308323563892145
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9916666666666667,
"confidence_cutoff": 0.9999431212544798
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9458333333333333,
"confidence_cutoff": 0.9297460863402224
}
},
"score_mae": 0.4607497712522395,
"ranked_probability_score": 0.08554771406167339
},
"as_shipped_clean": {
"n": 1200,
"nll": 0.454544651852342,
"acc": 0.8666666666666667,
"ece": 0.06910317787787193,
"brier": 0.20468221280246773,
"mean_conf": 0.9344856098346994,
"confident_error_rate": 0.041666666666666664,
"coverage_at_0_9": 0.795,
"accuracy_at_0_9": 0.9475890985324947,
"coverage_at_5pct_error": 0.7866666666666666,
"coverage_at_1pct_error": 0.6041666666666666,
"aurc": 0.02824430039449196,
"error_rate_at_0_9": 0.05241090146750524,
"confidence_bias": 0.06781894316803272,
"top_bins": {
"0.9": {
"n": 954,
"errors": 50,
"error_rate": 0.05241090146750524
},
"0.95": {
"n": 901,
"errors": 37,
"error_rate": 0.041065482796892344
},
"0.99": {
"n": 820,
"errors": 16,
"error_rate": 0.01951219512195122
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9916666666666667,
"confidence_cutoff": 0.9997511018421836
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9458333333333333,
"confidence_cutoff": 0.894833765474474
}
},
"score_mae": 0.46716677242673305,
"ranked_probability_score": 0.08353281985680432
},
"sources": {
"agnews": {
"n": 240,
"nll": 0.2969571491022943,
"acc": 0.9166666666666666,
"ece": 0.05669702215129781,
"brier": 0.13610080496920665,
"mean_conf": 0.9733636888179642,
"confident_error_rate": 0.04583333333333333,
"coverage_at_0_9": 0.9291666666666667,
"accuracy_at_0_9": 0.9506726457399103,
"coverage_at_5pct_error": 0.9291666666666667,
"coverage_at_1pct_error": 0.7333333333333333,
"aurc": 0.011459961623908478,
"error_rate_at_0_9": 0.04932735426008968,
"confidence_bias": 0.05669702215129757,
"top_bins": {
"0.9": {
"n": 223,
"errors": 11,
"error_rate": 0.04932735426008968
},
"0.95": {
"n": 216,
"errors": 10,
"error_rate": 0.046296296296296294
},
"0.99": {
"n": 197,
"errors": 4,
"error_rate": 0.02030456852791878
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.999603972973965
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.984375,
"confidence_cutoff": 0.9927218344770156
}
}
},
"amazon": {
"n": 80,
"nll": 1.1950196702821447,
"acc": 0.6125,
"ece": 0.13013344623625947,
"brier": 0.5588490719296325,
"mean_conf": 0.7426334462362595,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.2625,
"accuracy_at_0_9": 0.7619047619047619,
"coverage_at_5pct_error": 0.075,
"coverage_at_1pct_error": 0.075,
"aurc": 0.2609215557981968,
"error_rate_at_0_9": 0.23809523809523808,
"confidence_bias": 0.13013344623625944,
"top_bins": {
"0.9": {
"n": 21,
"errors": 5,
"error_rate": 0.23809523809523808
},
"0.95": {
"n": 13,
"errors": 2,
"error_rate": 0.15384615384615385
},
"0.99": {
"n": 4,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.775,
"confidence_cutoff": 0.7941759769280257
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.65625,
"confidence_cutoff": 0.5659829933652708
}
},
"score_mae": 0.4892758257511954,
"ranked_probability_score": 0.08858888411388968
},
"banking77": {
"n": 80,
"nll": 0.5758902612345038,
"acc": 0.85,
"ece": 0.08719569564847485,
"brier": 0.2407008008106888,
"mean_conf": 0.898073693405655,
"confident_error_rate": 0.0375,
"coverage_at_0_9": 0.75,
"accuracy_at_0_9": 0.95,
"coverage_at_5pct_error": 0.7625,
"coverage_at_1pct_error": 0.625,
"aurc": 0.03238808988486375,
"error_rate_at_0_9": 0.05,
"confidence_bias": 0.04807369340565504,
"top_bins": {
"0.9": {
"n": 60,
"errors": 3,
"error_rate": 0.05
},
"0.95": {
"n": 55,
"errors": 2,
"error_rate": 0.03636363636363636
},
"0.99": {
"n": 47,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9967902498867545
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.90625,
"confidence_cutoff": 0.8441430830009494
}
}
},
"boolq": {
"n": 80,
"nll": 0.5929837671372773,
"acc": 0.875,
"ece": 0.12193170422191033,
"brier": 0.2235795473979305,
"mean_conf": 0.9525887301568842,
"confident_error_rate": 0.0875,
"coverage_at_0_9": 0.85,
"accuracy_at_0_9": 0.8970588235294118,
"coverage_at_5pct_error": 0.6875,
"coverage_at_1pct_error": 0.1,
"aurc": 0.05282386578594207,
"error_rate_at_0_9": 0.10294117647058823,
"confidence_bias": 0.07758873015688417,
"top_bins": {
"0.9": {
"n": 68,
"errors": 7,
"error_rate": 0.10294117647058823
},
"0.95": {
"n": 67,
"errors": 7,
"error_rate": 0.1044776119402985
},
"0.99": {
"n": 61,
"errors": 4,
"error_rate": 0.06557377049180328
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.975,
"confidence_cutoff": 0.999911295185737
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.921875,
"confidence_cutoff": 0.9794008835333886
}
}
},
"contrastive": {
"n": 160,
"nll": 0.05996336403593442,
"acc": 0.96875,
"ece": 0.02417385494863131,
"brier": 0.04256566388357948,
"mean_conf": 0.9929238549486312,
"confident_error_rate": 0.0125,
"coverage_at_0_9": 0.98125,
"accuracy_at_0_9": 0.9872611464968153,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.975,
"aurc": 0.0005908980274374722,
"error_rate_at_0_9": 0.012738853503184714,
"confidence_bias": 0.024173854948631224,
"top_bins": {
"0.9": {
"n": 157,
"errors": 2,
"error_rate": 0.012738853503184714
},
"0.95": {
"n": 155,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 148,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.999992580361535
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9997815066400773
}
}
},
"dbpedia14": {
"n": 80,
"nll": 0.018296334655618373,
"acc": 0.9875,
"ece": 0.010929643397143786,
"brier": 0.013547082606835407,
"mean_conf": 0.9948529231687339,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.9875,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.9875,
"aurc": 0.00015625,
"error_rate_at_0_9": 0.0,
"confidence_bias": 0.007352923168733816,
"top_bins": {
"0.9": {
"n": 79,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 77,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 77,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999968083760991
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.999980508233233
}
}
},
"imdb": {
"n": 80,
"nll": 0.06650112598274324,
"acc": 0.9625,
"ece": 0.021127312041626652,
"brier": 0.047311917135916516,
"mean_conf": 0.9836273120416268,
"confident_error_rate": 0.0125,
"coverage_at_0_9": 0.9625,
"accuracy_at_0_9": 0.987012987012987,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.95,
"aurc": 0.0012680561790530144,
"error_rate_at_0_9": 0.012987012987012988,
"confidence_bias": 0.021127312041626767,
"top_bins": {
"0.9": {
"n": 77,
"errors": 1,
"error_rate": 0.012987012987012988
},
"0.95": {
"n": 75,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 71,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999868991306584
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.999740450053584
}
}
},
"mnli": {
"n": 80,
"nll": 0.4800709042620615,
"acc": 0.9125,
"ece": 0.07753013500835038,
"brier": 0.16332522643400862,
"mean_conf": 0.9812306735501618,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.925,
"accuracy_at_0_9": 0.9324324324324325,
"coverage_at_5pct_error": 0.8875,
"coverage_at_1pct_error": 0.6375,
"aurc": 0.016122981848020253,
"error_rate_at_0_9": 0.06756756756756757,
"confidence_bias": 0.06873067355016182,
"top_bins": {
"0.9": {
"n": 74,
"errors": 5,
"error_rate": 0.06756756756756757
},
"0.95": {
"n": 74,
"errors": 5,
"error_rate": 0.06756756756756757
},
"0.99": {
"n": 71,
"errors": 3,
"error_rate": 0.04225352112676056
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.999991586558766
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.96875,
"confidence_cutoff": 0.9998147211094783
}
}
},
"sst5": {
"n": 80,
"nll": 1.2448974820413372,
"acc": 0.6125,
"ece": 0.22505814089878196,
"brier": 0.5919475867934173,
"mean_conf": 0.8141449665678422,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.2875,
"accuracy_at_0_9": 0.782608695652174,
"coverage_at_5pct_error": 0.0875,
"coverage_at_1pct_error": 0.0875,
"aurc": 0.30028061666521133,
"error_rate_at_0_9": 0.21739130434782608,
"confidence_bias": 0.20164496656784214,
"top_bins": {
"0.9": {
"n": 23,
"errors": 5,
"error_rate": 0.21739130434782608
},
"0.95": {
"n": 8,
"errors": 1,
"error_rate": 0.125
},
"0.99": {
"n": 0,
"errors": 0,
"error_rate": null
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.6,
"confidence_cutoff": 0.855339285968297
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.625,
"confidence_cutoff": 0.7133087914054365
}
},
"score_mae": 0.4884492482076522,
"ranked_probability_score": 0.08971730097335581
},
"trec": {
"n": 80,
"nll": 0.28480951015184036,
"acc": 0.9625,
"ece": 0.030856363792743223,
"brier": 0.058704890234289875,
"mean_conf": 0.9903723975747134,
"confident_error_rate": 0.025,
"coverage_at_0_9": 0.975,
"accuracy_at_0_9": 0.9743589743589743,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.325,
"aurc": 0.015881097479045377,
"error_rate_at_0_9": 0.02564102564102564,
"confidence_bias": 0.02787239757471338,
"top_bins": {
"0.9": {
"n": 78,
"errors": 2,
"error_rate": 0.02564102564102564
},
"0.95": {
"n": 78,
"errors": 2,
"error_rate": 0.02564102564102564
},
"0.99": {
"n": 77,
"errors": 2,
"error_rate": 0.025974025974025976
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.975,
"confidence_cutoff": 0.9999910962505757
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.984375,
"confidence_cutoff": 0.9999026419356177
}
}
},
"yelp": {
"n": 160,
"nll": 0.6744512733294261,
"acc": 0.76875,
"ece": 0.11010641434949675,
"brier": 0.33941666300975865,
"mean_conf": 0.8769106142337293,
"confident_error_rate": 0.05625,
"coverage_at_0_9": 0.5875,
"accuracy_at_0_9": 0.9042553191489362,
"coverage_at_5pct_error": 0.425,
"coverage_at_1pct_error": 0.16875,
"aurc": 0.08692888800847512,
"error_rate_at_0_9": 0.09574468085106383,
"confidence_bias": 0.10816061423372925,
"top_bins": {
"0.9": {
"n": 94,
"errors": 9,
"error_rate": 0.09574468085106383
},
"0.95": {
"n": 83,
"errors": 8,
"error_rate": 0.0963855421686747
},
"0.99": {
"n": 67,
"errors": 3,
"error_rate": 0.04477611940298507
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.925,
"confidence_cutoff": 0.9744081429606993
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.84375,
"confidence_cutoff": 0.7335309795017884
}
},
"score_mae": 0.4237752433213515,
"ranked_probability_score": 0.07229227448316741
}
},
"types": {
"choice": {
"n": 480,
"nll": 0.3107149818460764,
"acc": 0.93125,
"ece": 0.04989938310601072,
"brier": 0.11702630095906619,
"mean_conf": 0.9725961663793784,
"confident_error_rate": 0.03333333333333333,
"coverage_at_0_9": 0.925,
"accuracy_at_0_9": 0.963963963963964,
"coverage_at_5pct_error": 0.95,
"coverage_at_1pct_error": 0.8333333333333334,
"aurc": 0.011065724252754237,
"error_rate_at_0_9": 0.036036036036036036,
"confidence_bias": 0.04134616637937838,
"top_bins": {
"0.9": {
"n": 444,
"errors": 16,
"error_rate": 0.036036036036036036
},
"0.95": {
"n": 436,
"errors": 15,
"error_rate": 0.034403669724770644
},
"0.99": {
"n": 415,
"errors": 7,
"error_rate": 0.016867469879518072
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9958333333333333,
"confidence_cutoff": 0.9999830700690301
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9895833333333334,
"confidence_cutoff": 0.9987196152255894
}
}
},
"noul": {
"n": 480,
"nll": 0.2523830232779272,
"acc": 0.925,
"ece": 0.04746414517655202,
"brier": 0.115343937114107,
"mean_conf": 0.9724641451765519,
"confident_error_rate": 0.0375,
"coverage_at_0_9": 0.91875,
"accuracy_at_0_9": 0.9591836734693877,
"coverage_at_5pct_error": 0.94375,
"coverage_at_1pct_error": 0.7041666666666667,
"aurc": 0.015342193433417054,
"error_rate_at_0_9": 0.04081632653061224,
"confidence_bias": 0.04746414517655184,
"top_bins": {
"0.9": {
"n": 441,
"errors": 18,
"error_rate": 0.04081632653061224
},
"0.95": {
"n": 427,
"errors": 14,
"error_rate": 0.03278688524590164
},
"0.99": {
"n": 395,
"errors": 8,
"error_rate": 0.020253164556962026
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9958333333333333,
"confidence_cutoff": 0.9998511569646503
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9791666666666666,
"confidence_cutoff": 0.9945634588215609
}
}
},
"score": {
"n": 240,
"nll": 1.146527249013703,
"acc": 0.6208333333333333,
"ece": 0.16277510247546034,
"brier": 0.5586705878659922,
"mean_conf": 0.7823074260616364,
"confident_error_rate": 0.06666666666666667,
"coverage_at_0_9": 0.2875,
"accuracy_at_0_9": 0.7681159420289855,
"coverage_at_5pct_error": 0.016666666666666666,
"coverage_at_1pct_error": 0.016666666666666666,
"aurc": 0.2783513871168549,
"error_rate_at_0_9": 0.2318840579710145,
"confidence_bias": 0.16147409272830304,
"top_bins": {
"0.9": {
"n": 69,
"errors": 16,
"error_rate": 0.2318840579710145
},
"0.95": {
"n": 38,
"errors": 8,
"error_rate": 0.21052631578947367
},
"0.99": {
"n": 10,
"errors": 1,
"error_rate": 0.1
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.7166666666666667,
"confidence_cutoff": 0.8132439678266241
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.6614583333333334,
"confidence_cutoff": 0.6191827053278941
}
},
"score_mae": 0.46716677242673305,
"ranked_probability_score": 0.08353281985680432
}
}
},
"transfer-v4": {
"raw_clean": {
"n": 656,
"nll": 0.7060605455091327,
"acc": 0.8384146341463414,
"ece": 0.10225794465378485,
"brier": 0.26913580782629415,
"mean_conf": 0.936023112192006,
"confident_error_rate": 0.08689024390243902,
"coverage_at_0_9": 0.8246951219512195,
"accuracy_at_0_9": 0.8946395563770795,
"coverage_at_5pct_error": 0.5975609756097561,
"coverage_at_1pct_error": 0.38109756097560976,
"aurc": 0.04670337348873893,
"error_rate_at_0_9": 0.10536044362292052,
"confidence_bias": 0.09760847804566453,
"top_bins": {
"0.9": {
"n": 541,
"errors": 57,
"error_rate": 0.10536044362292052
},
"0.95": {
"n": 508,
"errors": 46,
"error_rate": 0.09055118110236221
},
"0.99": {
"n": 449,
"errors": 28,
"error_rate": 0.062360801781737196
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9695121951219512,
"confidence_cutoff": 0.9996513990871982
},
"0.8": {
"coverage": 0.8003048780487805,
"accuracy": 0.900952380952381,
"confidence_cutoff": 0.9260639080978317
}
},
"score_mae": 0.28732928759135146,
"ranked_probability_score": 0.12507059546399885
},
"as_shipped_clean": {
"n": 656,
"nll": 0.6270891070988179,
"acc": 0.8384146341463414,
"ece": 0.09035981505437185,
"brier": 0.26355975613405186,
"mean_conf": 0.9253405484079532,
"confident_error_rate": 0.07469512195121951,
"coverage_at_0_9": 0.7896341463414634,
"accuracy_at_0_9": 0.9054054054054054,
"coverage_at_5pct_error": 0.5960365853658537,
"coverage_at_1pct_error": 0.3795731707317073,
"aurc": 0.046713600417941994,
"error_rate_at_0_9": 0.0945945945945946,
"confidence_bias": 0.08692591426161178,
"top_bins": {
"0.9": {
"n": 518,
"errors": 49,
"error_rate": 0.0945945945945946
},
"0.95": {
"n": 494,
"errors": 44,
"error_rate": 0.08906882591093117
},
"0.99": {
"n": 413,
"errors": 23,
"error_rate": 0.05569007263922518
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9695121951219512,
"confidence_cutoff": 0.9988538098904463
},
"0.8": {
"coverage": 0.8003048780487805,
"accuracy": 0.900952380952381,
"confidence_cutoff": 0.8926384237014877
}
},
"score_mae": 0.28661481608517747,
"ranked_probability_score": 0.1219924564403267
},
"sources": {
"composition_holdout": {
"n": 96,
"nll": 0.7716759343811087,
"acc": 0.8854166666666666,
"ece": 0.13161705342878371,
"brier": 0.2418445379901801,
"mean_conf": 0.9798888855966136,
"confident_error_rate": 0.11458333333333333,
"coverage_at_0_9": 0.9479166666666666,
"accuracy_at_0_9": 0.8791208791208791,
"coverage_at_5pct_error": 0.6875,
"coverage_at_1pct_error": 0.6666666666666666,
"aurc": 0.02787241712770423,
"error_rate_at_0_9": 0.12087912087912088,
"confidence_bias": 0.09447221892994695,
"top_bins": {
"0.9": {
"n": 91,
"errors": 11,
"error_rate": 0.12087912087912088
},
"0.95": {
"n": 90,
"errors": 11,
"error_rate": 0.12222222222222222
},
"0.99": {
"n": 86,
"errors": 11,
"error_rate": 0.12790697674418605
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.999944022537299
},
"0.8": {
"coverage": 0.8020833333333334,
"accuracy": 0.948051948051948,
"confidence_cutoff": 0.9994106055377326
}
}
},
"contrastive": {
"n": 80,
"nll": 0.5084316515078089,
"acc": 0.85,
"ece": 0.11914860470823861,
"brier": 0.24157018604381048,
"mean_conf": 0.9691486047082385,
"confident_error_rate": 0.0875,
"coverage_at_0_9": 0.9,
"accuracy_at_0_9": 0.9027777777777778,
"coverage_at_5pct_error": 0.85,
"coverage_at_1pct_error": 0.8,
"aurc": 0.018225744933534453,
"error_rate_at_0_9": 0.09722222222222222,
"confidence_bias": 0.11914860470823851,
"top_bins": {
"0.9": {
"n": 72,
"errors": 7,
"error_rate": 0.09722222222222222
},
"0.95": {
"n": 72,
"errors": 7,
"error_rate": 0.09722222222222222
},
"0.99": {
"n": 68,
"errors": 3,
"error_rate": 0.04411764705882353
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999840553223158
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9992294841723544
}
},
"score_mae": 0.28661481608517747,
"ranked_probability_score": 0.1219924564403267
},
"emotion": {
"n": 80,
"nll": 1.1842140096136198,
"acc": 0.6625,
"ece": 0.16344433654532525,
"brier": 0.49695454986128207,
"mean_conf": 0.7770470897943043,
"confident_error_rate": 0.075,
"coverage_at_0_9": 0.4375,
"accuracy_at_0_9": 0.8285714285714286,
"coverage_at_5pct_error": 0.075,
"coverage_at_1pct_error": 0.075,
"aurc": 0.18204891186299516,
"error_rate_at_0_9": 0.17142857142857143,
"confidence_bias": 0.11454708979430428,
"top_bins": {
"0.9": {
"n": 35,
"errors": 6,
"error_rate": 0.17142857142857143
},
"0.95": {
"n": 30,
"errors": 4,
"error_rate": 0.13333333333333333
},
"0.99": {
"n": 12,
"errors": 1,
"error_rate": 0.08333333333333333
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.8,
"confidence_cutoff": 0.8455177052439261
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.734375,
"confidence_cutoff": 0.5895563704278984
}
}
},
"mmlu": {
"n": 80,
"nll": 0.8811102729379066,
"acc": 0.7125,
"ece": 0.1128354245761854,
"brier": 0.4333401015309152,
"mean_conf": 0.8253354245761854,
"confident_error_rate": 0.05,
"coverage_at_0_9": 0.475,
"accuracy_at_0_9": 0.8947368421052632,
"coverage_at_5pct_error": 0.3625,
"coverage_at_1pct_error": 0.3125,
"aurc": 0.11284549915448887,
"error_rate_at_0_9": 0.10526315789473684,
"confidence_bias": 0.11283542457618534,
"top_bins": {
"0.9": {
"n": 38,
"errors": 4,
"error_rate": 0.10526315789473684
},
"0.95": {
"n": 32,
"errors": 2,
"error_rate": 0.0625
},
"0.99": {
"n": 19,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9,
"confidence_cutoff": 0.8934639334502172
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.796875,
"confidence_cutoff": 0.6529787827885164
}
}
},
"paws": {
"n": 80,
"nll": 0.9149710000217235,
"acc": 0.8,
"ece": 0.15742064884121132,
"brier": 0.3163214505434944,
"mean_conf": 0.9481862302428528,
"confident_error_rate": 0.125,
"coverage_at_0_9": 0.8625,
"accuracy_at_0_9": 0.855072463768116,
"coverage_at_5pct_error": 0.275,
"coverage_at_1pct_error": 0.175,
"aurc": 0.09027125896066751,
"error_rate_at_0_9": 0.14492753623188406,
"confidence_bias": 0.1481862302428527,
"top_bins": {
"0.9": {
"n": 69,
"errors": 10,
"error_rate": 0.14492753623188406
},
"0.95": {
"n": 69,
"errors": 10,
"error_rate": 0.14492753623188406
},
"0.99": {
"n": 61,
"errors": 6,
"error_rate": 0.09836065573770492
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9,
"confidence_cutoff": 0.9997018073599228
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.890625,
"confidence_cutoff": 0.9816976299326602
}
}
},
"qnli": {
"n": 80,
"nll": 0.3010750024235448,
"acc": 0.925,
"ece": 0.08174310999100654,
"brier": 0.13992055357407857,
"mean_conf": 0.9736296740922775,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.925,
"accuracy_at_0_9": 0.9324324324324325,
"coverage_at_5pct_error": 0.8625,
"coverage_at_1pct_error": 0.625,
"aurc": 0.014993418347854443,
"error_rate_at_0_9": 0.06756756756756757,
"confidence_bias": 0.048629674092277475,
"top_bins": {
"0.9": {
"n": 74,
"errors": 5,
"error_rate": 0.06756756756756757
},
"0.95": {
"n": 71,
"errors": 4,
"error_rate": 0.056338028169014086
},
"0.99": {
"n": 64,
"errors": 2,
"error_rate": 0.03125
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9996265539820621
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.96875,
"confidence_cutoff": 0.9930251162586794
}
}
},
"sciq": {
"n": 80,
"nll": 0.007314621146129305,
"acc": 1.0,
"ece": 0.006816864281393114,
"brier": 0.0015733066903292278,
"mean_conf": 0.993183135718607,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.975,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 1.0,
"aurc": 0.0,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.0068168642813930536,
"top_bins": {
"0.9": {
"n": 78,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 77,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 72,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999608059905641
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9995954309046333
}
}
},
"tweet_offensive": {
"n": 80,
"nll": 0.4190029993022442,
"acc": 0.8625,
"ece": 0.08458362708432104,
"brier": 0.2412964064670991,
"mean_conf": 0.9253956750968154,
"confident_error_rate": 0.075,
"coverage_at_0_9": 0.7625,
"accuracy_at_0_9": 0.9016393442622951,
"coverage_at_5pct_error": 0.575,
"coverage_at_1pct_error": 0.5,
"aurc": 0.04870527959080802,
"error_rate_at_0_9": 0.09836065573770492,
"confidence_bias": 0.06289567509681537,
"top_bins": {
"0.9": {
"n": 61,
"errors": 6,
"error_rate": 0.09836065573770492
},
"0.95": {
"n": 53,
"errors": 6,
"error_rate": 0.11320754716981132
},
"0.99": {
"n": 31,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9833703279193756
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.890625,
"confidence_cutoff": 0.891081054907394
}
}
}
},
"types": {
"choice": {
"n": 336,
"nll": 0.7139642916559491,
"acc": 0.8184523809523809,
"ece": 0.08621197602515834,
"brier": 0.29097176277874826,
"mean_conf": 0.8979600744773888,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.7202380952380952,
"accuracy_at_0_9": 0.9132231404958677,
"coverage_at_5pct_error": 0.5208333333333334,
"coverage_at_1pct_error": 0.3958333333333333,
"aurc": 0.05304090653277752,
"error_rate_at_0_9": 0.08677685950413223,
"confidence_bias": 0.07950769352500786,
"top_bins": {
"0.9": {
"n": 242,
"errors": 21,
"error_rate": 0.08677685950413223
},
"0.95": {
"n": 229,
"errors": 17,
"error_rate": 0.07423580786026202
},
"0.99": {
"n": 189,
"errors": 12,
"error_rate": 0.06349206349206349
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9702380952380952,
"confidence_cutoff": 0.9970506958588666
},
"0.8": {
"coverage": 0.8005952380952381,
"accuracy": 0.895910780669145,
"confidence_cutoff": 0.7834280359430882
}
}
},
"noul": {
"n": 280,
"nll": 0.4671583132891337,
"acc": 0.8821428571428571,
"ece": 0.07465765160771948,
"brier": 0.1992966887782935,
"mean_conf": 0.9563447099149558,
"confident_error_rate": 0.075,
"coverage_at_0_9": 0.8714285714285714,
"accuracy_at_0_9": 0.9139344262295082,
"coverage_at_5pct_error": 0.7642857142857142,
"coverage_at_1pct_error": 0.26785714285714285,
"aurc": 0.03774979623265823,
"error_rate_at_0_9": 0.0860655737704918,
"confidence_bias": 0.07420185277209868,
"top_bins": {
"0.9": {
"n": 244,
"errors": 21,
"error_rate": 0.0860655737704918
},
"0.95": {
"n": 233,
"errors": 20,
"error_rate": 0.08583690987124463
},
"0.99": {
"n": 196,
"errors": 8,
"error_rate": 0.04081632653061224
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9642857142857143,
"confidence_cutoff": 0.9991586798493963
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9375,
"confidence_cutoff": 0.964939495122677
}
}
},
"score": {
"n": 40,
"nll": 1.0168531134867067,
"acc": 0.7,
"ece": 0.23830739887567784,
"brier": 0.48314037180891034,
"mean_conf": 0.9383073988756777,
"confident_error_rate": 0.175,
"coverage_at_0_9": 0.8,
"accuracy_at_0_9": 0.78125,
"coverage_at_5pct_error": 0.65,
"coverage_at_1pct_error": 0.6,
"aurc": 0.07895709505044536,
"error_rate_at_0_9": 0.21875,
"confidence_bias": 0.23830739887567776,
"top_bins": {
"0.9": {
"n": 32,
"errors": 7,
"error_rate": 0.21875
},
"0.95": {
"n": 32,
"errors": 7,
"error_rate": 0.21875
},
"0.99": {
"n": 28,
"errors": 3,
"error_rate": 0.10714285714285714
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9997651676938517
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.78125,
"confidence_cutoff": 0.9590969197346936
}
},
"score_mae": 0.28661481608517747,
"ranked_probability_score": 0.1219924564403267
}
}
}
},
"temperature": 1.1759291426711453
},
{
"epoch": 2,
"development": {
"raw": {
"questions": 14912,
"accuracy": 0.9046405579399142,
"mean_class_recall": 0.8351706914084861,
"nll": 0.2609479185046837,
"brier": 0.13871271598030968,
"ece15": 0.02912533581539033,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.811762339055794,
"error_rate_at_confidence_0_9": 0.030318050392399834,
"ordinal_expected_level_mae": 0.007142014228908907,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 222,
"accuracy": 0.9774774774774775,
"mean_class_recall": 0.9570930232558139,
"nll": 0.0857490162426134,
"brier": 0.03780111867446018,
"ece15": 0.020326742033581136,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9864864864864865,
"error_rate_at_confidence_0_9": 0.0136986301369863,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 222,
"accuracy": 0.9819819819819819,
"mean_class_recall": 0.96,
"nll": 0.08583505640070498,
"brier": 0.03170617674969949,
"ece15": 0.018778574788420183,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9819819819819819,
"error_rate_at_confidence_0_9": 0.009174311926605505,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 1400,
"accuracy": 0.9828571428571429,
"mean_class_recall": 0.9678571428571429,
"nll": 0.05038447430212686,
"brier": 0.026644462415551518,
"ece15": 0.006643072262975827,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9757142857142858,
"error_rate_at_confidence_0_9": 0.008784773060029283,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 5000,
"accuracy": 0.839,
"mean_class_recall": 0.7132000000000001,
"nll": 0.4567278707638309,
"brier": 0.24289094068797595,
"ece15": 0.07089882316707959,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.7078,
"error_rate_at_confidence_0_9": 0.08194405199208817,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0030323687345377927,
"brier": 0.0010147751519150774,
"ece15": 0.0027304101493378944,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9919354838709677,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0013590718776517526,
"brier": 0.0005853758373684321,
"ece15": 0.001174090997847941,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9969512195121951,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 2.706087239227311e-05,
"brier": 4.4631636950527135e-07,
"ece15": 2.6948652833984242e-05,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 2500,
"accuracy": 0.9788,
"mean_class_recall": 0.98,
"nll": 0.08755100200107269,
"brier": 0.038158588944742826,
"ece15": 0.017829224144769444,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9896,
"error_rate_at_confidence_0_9": 0.016976556184316895,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 797,
"accuracy": 0.9949811794228356,
"mean_class_recall": 0.9949406163344321,
"nll": 0.020100883811983803,
"brier": 0.009532705405460478,
"ece15": 0.004664870689987746,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9962358845671268,
"error_rate_at_confidence_0_9": 0.003778337531486146,
"ordinal_expected_level_mae": 0.007142014228908907
},
"twins_streamlet:twins_policy_action": {
"questions": 797,
"accuracy": 0.9887076537013801,
"mean_class_recall": 0.9889237691623501,
"nll": 0.029531627570397425,
"brier": 0.01682083664455817,
"ece15": 0.008194444119065863,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9887076537013801,
"error_rate_at_confidence_0_9": 0.0038071065989847717,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 797,
"accuracy": 0.9937264742785445,
"mean_class_recall": 0.9922118380062306,
"nll": 0.019942020828876864,
"brier": 0.011733672902765547,
"ece15": 0.006519660642330038,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9912170639899623,
"error_rate_at_confidence_0_9": 0.0037974683544303796,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 797,
"accuracy": 0.9974905897114178,
"mean_class_recall": 0.998236331569665,
"nll": 0.016769983022509982,
"brier": 0.0049992664597086935,
"ece15": 0.0021278163552494573,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.002509410288582183,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 800,
"accuracy": 0.6875,
"mean_class_recall": 0.6966034271725826,
"nll": 0.5142628800857739,
"brier": 0.36155138388149327,
"ece15": 0.02205690591199928,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.22125,
"error_rate_at_confidence_0_9": 0.011299435028248588,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 800,
"accuracy": 0.67375,
"mean_class_recall": 0.5089660282717738,
"nll": 0.998813782244008,
"brier": 0.4775012041931279,
"ece15": 0.03873448735113525,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.20625,
"error_rate_at_confidence_0_9": 0.030303030303030304,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9354480356736271,
"macro_task_balanced_accuracy": 0.9112880126164279
},
"recalibrated": {
"questions": 14912,
"accuracy": 0.9046405579399142,
"mean_class_recall": 0.8351706914084861,
"nll": 0.24022147566124724,
"brier": 0.13631204830326463,
"ece15": 0.023735839802210217,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.790973712446352,
"error_rate_at_confidence_0_9": 0.02467147096227215,
"ordinal_expected_level_mae": 0.007897347093607133,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 222,
"accuracy": 0.9774774774774775,
"mean_class_recall": 0.9570930232558139,
"nll": 0.06980345492279078,
"brier": 0.036072899559119326,
"ece15": 0.020348852979949178,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9819819819819819,
"error_rate_at_confidence_0_9": 0.013761467889908258,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 222,
"accuracy": 0.9819819819819819,
"mean_class_recall": 0.96,
"nll": 0.0697534508038578,
"brier": 0.030632585867415095,
"ece15": 0.018874774396367578,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9819819819819819,
"error_rate_at_confidence_0_9": 0.009174311926605505,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 1400,
"accuracy": 0.9828571428571429,
"mean_class_recall": 0.9678571428571429,
"nll": 0.05022700385733144,
"brier": 0.026160310719648357,
"ece15": 0.005133274295293359,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9642857142857143,
"error_rate_at_confidence_0_9": 0.007407407407407408,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 5000,
"accuracy": 0.839,
"mean_class_recall": 0.7132000000000001,
"nll": 0.4007512736870225,
"brier": 0.23542102872463042,
"ece15": 0.05680924935578224,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.6622,
"error_rate_at_confidence_0_9": 0.0664451827242525,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.00473204279698464,
"brier": 0.0015928649780981342,
"ece15": 0.004243332776407401,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9919354838709677,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0022032517878559357,
"brier": 0.0007796053221569698,
"ece15": 0.0019494525692642349,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9969512195121951,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 9.041263403421138e-05,
"brier": 3.609861838766288e-06,
"ece15": 8.949527479169929e-05,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 2500,
"accuracy": 0.9788,
"mean_class_recall": 0.98,
"nll": 0.07409278691965634,
"brier": 0.03695008653192076,
"ece15": 0.013363996653905837,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.984,
"error_rate_at_confidence_0_9": 0.016666666666666666,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 797,
"accuracy": 0.9949811794228356,
"mean_class_recall": 0.9949406163344321,
"nll": 0.017338830453877865,
"brier": 0.00923947768144349,
"ece15": 0.0037098432118081266,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9899623588456713,
"error_rate_at_confidence_0_9": 0.0038022813688212928,
"ordinal_expected_level_mae": 0.007897347093607133
},
"twins_streamlet:twins_policy_action": {
"questions": 797,
"accuracy": 0.9887076537013801,
"mean_class_recall": 0.9889237691623501,
"nll": 0.025810322060464137,
"brier": 0.015590356305548642,
"ece15": 0.0063347054801968505,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9824341279799247,
"error_rate_at_confidence_0_9": 0.002554278416347382,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 797,
"accuracy": 0.9937264742785445,
"mean_class_recall": 0.9922118380062306,
"nll": 0.017984967092368,
"brier": 0.011131458808574956,
"ece15": 0.005638113127633191,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9912170639899623,
"error_rate_at_confidence_0_9": 0.0037974683544303796,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 797,
"accuracy": 0.9974905897114178,
"mean_class_recall": 0.998236331569665,
"nll": 0.013810181529074162,
"brier": 0.004981534821676031,
"ece15": 0.0021482520070447817,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.998745294855709,
"error_rate_at_confidence_0_9": 0.002512562814070352,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 800,
"accuracy": 0.6875,
"mean_class_recall": 0.6966034271725826,
"nll": 0.5181534061242398,
"brier": 0.36342749885940506,
"ece15": 0.019537157326473344,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.205,
"error_rate_at_confidence_0_9": 0.012195121951219513,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 800,
"accuracy": 0.67375,
"mean_class_recall": 0.5089660282717738,
"nll": 1.020378053391093,
"brier": 0.48493001129627705,
"ece15": 0.07743873861233927,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.1725,
"error_rate_at_confidence_0_9": 0.021739130434782608,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9354480356736271,
"macro_task_balanced_accuracy": 0.9112880126164279
}
},
"test": {
"raw": {
"questions": 21634,
"accuracy": 0.8712212258482019,
"mean_class_recall": 0.8343950403877007,
"nll": 0.36437325393469283,
"brier": 0.1826183728339598,
"ece15": 0.016790325586147077,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.6702412868632708,
"error_rate_at_confidence_0_9": 0.01903448275862069,
"ordinal_expected_level_mae": 0.005114573364023923,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 426,
"accuracy": 0.9953051643192489,
"mean_class_recall": 0.9895833333333333,
"nll": 0.020419671491798054,
"brier": 0.008500766418781964,
"ece15": 0.004676929181256299,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9765258215962441,
"error_rate_at_confidence_0_9": 0.002403846153846154,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 426,
"accuracy": 0.9835680751173709,
"mean_class_recall": 0.9635416666666667,
"nll": 0.03064858581307839,
"brier": 0.020119931286614087,
"ece15": 0.009967851623295965,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9788732394366197,
"error_rate_at_confidence_0_9": 0.004796163069544364,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 800,
"accuracy": 0.9525,
"mean_class_recall": 0.91640625,
"nll": 0.17731219866650544,
"brier": 0.08433190754386494,
"ece15": 0.03454275208230118,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.95,
"error_rate_at_confidence_0_9": 0.03289473684210526,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 6000,
"accuracy": 0.8995,
"mean_class_recall": 0.8221111111111111,
"nll": 0.30893788807797395,
"brier": 0.15122747877892262,
"ece15": 0.028859158451860423,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.737,
"error_rate_at_confidence_0_9": 0.03640886476707372,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.00019733561912440662,
"brier": 1.9375590740059706e-06,
"ece15": 0.0001968485399507404,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0017246613457991484,
"brier": 0.0007621008009601763,
"ece15": 0.001458779457079616,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9981481481481481,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 1.1611401866187557e-06,
"brier": 5.747739403769056e-11,
"ece15": 1.1611258165194832e-06,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 3500,
"accuracy": 0.9894285714285714,
"mean_class_recall": 0.9826666666666667,
"nll": 0.059799522199125266,
"brier": 0.02016997152334037,
"ece15": 0.00639039716386258,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9728571428571429,
"error_rate_at_confidence_0_9": 0.0076358296622613805,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 659,
"accuracy": 0.9984825493171472,
"mean_class_recall": 0.9983249581239532,
"nll": 0.0075671027444564045,
"brier": 0.0038579171936140126,
"ece15": 0.0029747801235328448,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9939301972685888,
"error_rate_at_confidence_0_9": 0.0015267175572519084,
"ordinal_expected_level_mae": 0.005114573364023923
},
"twins_streamlet:twins_policy_action": {
"questions": 659,
"accuracy": 0.9969650986342944,
"mean_class_recall": 0.9966499162479062,
"nll": 0.007401075027874373,
"brier": 0.004494834346902429,
"ece15": 0.0034418284412443194,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9939301972685888,
"error_rate_at_confidence_0_9": 0.0015267175572519084,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 659,
"accuracy": 0.9984825493171472,
"mean_class_recall": 0.9978632478632479,
"nll": 0.0026045927984607647,
"brier": 0.00133772985585407,
"ece15": 0.002039732462895639,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9954476479514416,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 659,
"accuracy": 0.9969650986342944,
"mean_class_recall": 0.9978260869565218,
"nll": 0.007434124878991116,
"brier": 0.004449797406116125,
"ece15": 0.002637941445288803,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9969650986342944,
"error_rate_at_confidence_0_9": 0.0015220700152207,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 3321,
"accuracy": 0.6838301716350497,
"mean_class_recall": 0.6880928586224566,
"nll": 0.5412193961251557,
"brier": 0.3797960822714433,
"ece15": 0.028068617701296383,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.19451972297500753,
"error_rate_at_confidence_0_9": 0.01393188854489164,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 3321,
"accuracy": 0.6859379704908161,
"mean_class_recall": 0.4840299509564737,
"nll": 1.1567283681183516,
"brier": 0.4884417807744329,
"ece15": 0.038297669894539775,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.18307738632941886,
"error_rate_at_confidence_0_9": 0.0805921052631579,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9414975177781386,
"macro_task_balanced_accuracy": 0.9169354318963097
},
"recalibrated": {
"questions": 21634,
"accuracy": 0.8712212258482019,
"mean_class_recall": 0.8343950403877007,
"nll": 0.35395590855273523,
"brier": 0.18501029425464544,
"ece15": 0.030017031519889922,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.6506887307016733,
"error_rate_at_confidence_0_9": 0.016054557078923067,
"ordinal_expected_level_mae": 0.006011043764813851,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 426,
"accuracy": 0.9953051643192489,
"mean_class_recall": 0.9895833333333333,
"nll": 0.023835787531761698,
"brier": 0.009460218069430947,
"ece15": 0.00990292779030044,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9671361502347418,
"error_rate_at_confidence_0_9": 0.0024271844660194173,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 426,
"accuracy": 0.9835680751173709,
"mean_class_recall": 0.9635416666666667,
"nll": 0.02937486871225543,
"brier": 0.01905460565156044,
"ece15": 0.011296185892511,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9694835680751174,
"error_rate_at_confidence_0_9": 0.002421307506053269,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 800,
"accuracy": 0.9525,
"mean_class_recall": 0.91640625,
"nll": 0.15527362489298477,
"brier": 0.08190930430000586,
"ece15": 0.02574101560783749,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.92625,
"error_rate_at_confidence_0_9": 0.029689608636977057,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 6000,
"accuracy": 0.8995,
"mean_class_recall": 0.8221111111111111,
"nll": 0.2798275436837799,
"brier": 0.15134523246061166,
"ece15": 0.029238705666976703,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.7075,
"error_rate_at_confidence_0_9": 0.0303886925795053,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0008001907223348217,
"brier": 1.793746254505973e-05,
"ece15": 0.0007956440534692383,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0027723209709698475,
"brier": 0.0009529345145823382,
"ece15": 0.0024453488801810066,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9925925925925926,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 1.8070166383123626e-05,
"brier": 5.357685907682576e-09,
"ece15": 1.806882660338892e-05,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 3500,
"accuracy": 0.9894285714285714,
"mean_class_recall": 0.9826666666666667,
"nll": 0.05655522961511102,
"brier": 0.021374549663926688,
"ece15": 0.009425764157233412,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.95,
"error_rate_at_confidence_0_9": 0.0069172932330827065,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 659,
"accuracy": 0.9984825493171472,
"mean_class_recall": 0.9983249581239532,
"nll": 0.007520014659349455,
"brier": 0.003991311156523451,
"ece15": 0.00377357794140357,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.992412746585736,
"error_rate_at_confidence_0_9": 0.0015290519877675841,
"ordinal_expected_level_mae": 0.006011043764813851
},
"twins_streamlet:twins_policy_action": {
"questions": 659,
"accuracy": 0.9969650986342944,
"mean_class_recall": 0.9966499162479062,
"nll": 0.008102634890372476,
"brier": 0.004356473026911708,
"ece15": 0.004867738330078791,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9908952959028832,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 659,
"accuracy": 0.9984825493171472,
"mean_class_recall": 0.9978632478632479,
"nll": 0.0038576227479580737,
"brier": 0.0014748958385226618,
"ece15": 0.003282715492202558,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9939301972685888,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 659,
"accuracy": 0.9969650986342944,
"mean_class_recall": 0.9978260869565218,
"nll": 0.006788751578147029,
"brier": 0.004171759480970426,
"ece15": 0.003177496987382077,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.9939301972685888,
"error_rate_at_confidence_0_9": 0.0015267175572519084,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 3321,
"accuracy": 0.6838301716350497,
"mean_class_recall": 0.6880928586224566,
"nll": 0.5429719234394278,
"brier": 0.3809937714731844,
"ece15": 0.02889956013524373,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.17765733212887685,
"error_rate_at_confidence_0_9": 0.010169491525423728,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 3321,
"accuracy": 0.6859379704908161,
"mean_class_recall": 0.4840299509564737,
"nll": 1.1477144959529286,
"brier": 0.5019380206266735,
"ece15": 0.11401240417222197,
"temperature": 1.2873604198505975,
"coverage_at_confidence_0_9": 0.16079494128274616,
"error_rate_at_confidence_0_9": 0.07865168539325842,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9414975177781386,
"macro_task_balanced_accuracy": 0.9169354318963097
}
},
"general": {
"decision-v7": {
"raw_clean": {
"n": 1200,
"nll": 0.5237394417093835,
"acc": 0.865,
"ece": 0.08203125833030098,
"brier": 0.21310411533941315,
"mean_conf": 0.9455011120677191,
"confident_error_rate": 0.06,
"coverage_at_0_9": 0.8308333333333333,
"accuracy_at_0_9": 0.9277833500501504,
"coverage_at_5pct_error": 0.7791666666666667,
"coverage_at_1pct_error": 0.6158333333333333,
"aurc": 0.028219153802824883,
"error_rate_at_0_9": 0.07221664994984955,
"confidence_bias": 0.08050111206771915,
"top_bins": {
"0.9": {
"n": 997,
"errors": 72,
"error_rate": 0.07221664994984955
},
"0.95": {
"n": 937,
"errors": 47,
"error_rate": 0.05016008537886873
},
"0.99": {
"n": 861,
"errors": 27,
"error_rate": 0.0313588850174216
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9933333333333333,
"confidence_cutoff": 0.9999626357556778
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9458333333333333,
"confidence_cutoff": 0.9342652506942892
}
},
"score_mae": 0.4647882331825802,
"ranked_probability_score": 0.08637141204655961
},
"as_shipped_clean": {
"n": 1200,
"nll": 0.44174931033716425,
"acc": 0.865,
"ece": 0.0702397073006123,
"brier": 0.20431799318642196,
"mean_conf": 0.930435938959295,
"confident_error_rate": 0.04,
"coverage_at_0_9": 0.7858333333333334,
"accuracy_at_0_9": 0.9490986214209968,
"coverage_at_5pct_error": 0.7783333333333333,
"coverage_at_1pct_error": 0.6125,
"aurc": 0.028152549316370398,
"error_rate_at_0_9": 0.05090137857900318,
"confidence_bias": 0.065435938959295,
"top_bins": {
"0.9": {
"n": 943,
"errors": 48,
"error_rate": 0.05090137857900318
},
"0.95": {
"n": 893,
"errors": 35,
"error_rate": 0.03919372900335946
},
"0.99": {
"n": 803,
"errors": 15,
"error_rate": 0.0186799501867995
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9916666666666667,
"confidence_cutoff": 0.9996192297027195
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9447916666666667,
"confidence_cutoff": 0.8818964099133315
}
},
"score_mae": 0.47431418086986715,
"ranked_probability_score": 0.08319761613591657
},
"sources": {
"agnews": {
"n": 240,
"nll": 0.2796999395688286,
"acc": 0.9125,
"ece": 0.05892681778016237,
"brier": 0.1350754514620047,
"mean_conf": 0.9714268177801624,
"confident_error_rate": 0.04583333333333333,
"coverage_at_0_9": 0.9208333333333333,
"accuracy_at_0_9": 0.9502262443438914,
"coverage_at_5pct_error": 0.9333333333333333,
"coverage_at_1pct_error": 0.75,
"aurc": 0.011330835078134182,
"error_rate_at_0_9": 0.049773755656108594,
"confidence_bias": 0.0589268177801624,
"top_bins": {
"0.9": {
"n": 221,
"errors": 11,
"error_rate": 0.049773755656108594
},
"0.95": {
"n": 214,
"errors": 9,
"error_rate": 0.04205607476635514
},
"0.99": {
"n": 189,
"errors": 3,
"error_rate": 0.015873015873015872
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9994019817408221
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.984375,
"confidence_cutoff": 0.9890606274392002
}
}
},
"amazon": {
"n": 80,
"nll": 1.1725112640711228,
"acc": 0.6,
"ece": 0.14900986362097515,
"brier": 0.5611978409181134,
"mean_conf": 0.7279331238481536,
"confident_error_rate": 0.0375,
"coverage_at_0_9": 0.2125,
"accuracy_at_0_9": 0.8235294117647058,
"coverage_at_5pct_error": 0.075,
"coverage_at_1pct_error": 0.075,
"aurc": 0.268674359040746,
"error_rate_at_0_9": 0.17647058823529413,
"confidence_bias": 0.12793312384815358,
"top_bins": {
"0.9": {
"n": 17,
"errors": 3,
"error_rate": 0.17647058823529413
},
"0.95": {
"n": 12,
"errors": 2,
"error_rate": 0.16666666666666666
},
"0.99": {
"n": 2,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.725,
"confidence_cutoff": 0.7565342654626949
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.640625,
"confidence_cutoff": 0.5366925117472928
}
},
"score_mae": 0.5012771753189752,
"ranked_probability_score": 0.0890286871906647
},
"banking77": {
"n": 80,
"nll": 0.5995652320511431,
"acc": 0.85,
"ece": 0.08817712605098262,
"brier": 0.24932641415691287,
"mean_conf": 0.8916570456057876,
"confident_error_rate": 0.05,
"coverage_at_0_9": 0.75,
"accuracy_at_0_9": 0.9333333333333333,
"coverage_at_5pct_error": 0.6875,
"coverage_at_1pct_error": 0.6125,
"aurc": 0.03525784971074972,
"error_rate_at_0_9": 0.06666666666666667,
"confidence_bias": 0.041657045605787646,
"top_bins": {
"0.9": {
"n": 60,
"errors": 4,
"error_rate": 0.06666666666666667
},
"0.95": {
"n": 55,
"errors": 2,
"error_rate": 0.03636363636363636
},
"0.99": {
"n": 46,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9933430562867247
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.90625,
"confidence_cutoff": 0.8362446757691954
}
}
},
"boolq": {
"n": 80,
"nll": 0.5712250447804231,
"acc": 0.875,
"ece": 0.11692367054964475,
"brier": 0.22310665871912266,
"mean_conf": 0.948749077509772,
"confident_error_rate": 0.0875,
"coverage_at_0_9": 0.85,
"accuracy_at_0_9": 0.8970588235294118,
"coverage_at_5pct_error": 0.6875,
"coverage_at_1pct_error": 0.1,
"aurc": 0.05278244760094704,
"error_rate_at_0_9": 0.10294117647058823,
"confidence_bias": 0.073749077509772,
"top_bins": {
"0.9": {
"n": 68,
"errors": 7,
"error_rate": 0.10294117647058823
},
"0.95": {
"n": 66,
"errors": 6,
"error_rate": 0.09090909090909091
},
"0.99": {
"n": 60,
"errors": 4,
"error_rate": 0.06666666666666667
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.975,
"confidence_cutoff": 0.9998615919766226
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.921875,
"confidence_cutoff": 0.979302462267494
}
}
},
"contrastive": {
"n": 160,
"nll": 0.06554428006012027,
"acc": 0.96875,
"ece": 0.024615155693287222,
"brier": 0.04485687620328682,
"mean_conf": 0.9933651556932872,
"confident_error_rate": 0.0125,
"coverage_at_0_9": 0.98125,
"accuracy_at_0_9": 0.9872611464968153,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.975,
"aurc": 0.000671805023667049,
"error_rate_at_0_9": 0.012738853503184714,
"confidence_bias": 0.024615155693287205,
"top_bins": {
"0.9": {
"n": 157,
"errors": 2,
"error_rate": 0.012738853503184714
},
"0.95": {
"n": 154,
"errors": 1,
"error_rate": 0.006493506493506494
},
"0.99": {
"n": 148,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999882930499685
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9997675703852573
}
}
},
"dbpedia14": {
"n": 80,
"nll": 0.01926796153789536,
"acc": 0.9875,
"ece": 0.011278233290686156,
"brier": 0.014217088541729535,
"mean_conf": 0.9949340220579689,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.9875,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.9875,
"aurc": 0.00015625,
"error_rate_at_0_9": 0.0,
"confidence_bias": 0.0074340220579688365,
"top_bins": {
"0.9": {
"n": 79,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 77,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 76,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999950740905034
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9999745354541041
}
}
},
"imdb": {
"n": 80,
"nll": 0.06446545489140978,
"acc": 0.9625,
"ece": 0.025460341159037583,
"brier": 0.045829990154110004,
"mean_conf": 0.9836424817249945,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.9375,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.95,
"aurc": 0.0012680561790530144,
"error_rate_at_0_9": 0.0,
"confidence_bias": 0.021142481724994444,
"top_bins": {
"0.9": {
"n": 75,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 75,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 71,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999788743849367
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9996364092230372
}
}
},
"mnli": {
"n": 80,
"nll": 0.46179093448986874,
"acc": 0.9125,
"ece": 0.07967308878997346,
"brier": 0.16221252791471635,
"mean_conf": 0.985103225410972,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.925,
"accuracy_at_0_9": 0.9324324324324325,
"coverage_at_5pct_error": 0.8875,
"coverage_at_1pct_error": 0.65,
"aurc": 0.015873702018260973,
"error_rate_at_0_9": 0.06756756756756757,
"confidence_bias": 0.07260322541097197,
"top_bins": {
"0.9": {
"n": 74,
"errors": 5,
"error_rate": 0.06756756756756757
},
"0.95": {
"n": 74,
"errors": 5,
"error_rate": 0.06756756756756757
},
"0.99": {
"n": 71,
"errors": 3,
"error_rate": 0.04225352112676056
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.999987939363811
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.96875,
"confidence_cutoff": 0.9997555402093751
}
}
},
"sst5": {
"n": 80,
"nll": 1.1930808545317393,
"acc": 0.6125,
"ece": 0.23848723097529123,
"brier": 0.5810003072599018,
"mean_conf": 0.7987986774228062,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.2625,
"accuracy_at_0_9": 0.7619047619047619,
"coverage_at_5pct_error": 0.0875,
"coverage_at_1pct_error": 0.0875,
"aurc": 0.2998766100005863,
"error_rate_at_0_9": 0.23809523809523808,
"confidence_bias": 0.18629867742280615,
"top_bins": {
"0.9": {
"n": 21,
"errors": 5,
"error_rate": 0.23809523809523808
},
"0.95": {
"n": 5,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 0,
"errors": 0,
"error_rate": null
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.625,
"confidence_cutoff": 0.8363874325149153
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.625,
"confidence_cutoff": 0.6881083881339022
}
},
"score_mae": 0.4905379704011906,
"ranked_probability_score": 0.08831634129641223
},
"trec": {
"n": 80,
"nll": 0.268396899685072,
"acc": 0.9625,
"ece": 0.02818577556782262,
"brier": 0.058300946365123306,
"mean_conf": 0.9906857755678227,
"confident_error_rate": 0.025,
"coverage_at_0_9": 0.9875,
"accuracy_at_0_9": 0.9746835443037974,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.325,
"aurc": 0.015881097479045377,
"error_rate_at_0_9": 0.02531645569620253,
"confidence_bias": 0.028185775567822646,
"top_bins": {
"0.9": {
"n": 79,
"errors": 2,
"error_rate": 0.02531645569620253
},
"0.95": {
"n": 78,
"errors": 2,
"error_rate": 0.02564102564102564
},
"0.99": {
"n": 77,
"errors": 2,
"error_rate": 0.025974025974025976
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.975,
"confidence_cutoff": 0.9999834449407883
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.984375,
"confidence_cutoff": 0.9998382341930544
}
}
},
"yelp": {
"n": 160,
"nll": 0.6528738150960317,
"acc": 0.76875,
"ece": 0.10032078827768093,
"brier": 0.3373190084870059,
"mean_conf": 0.867012445257043,
"confident_error_rate": 0.05625,
"coverage_at_0_9": 0.575,
"accuracy_at_0_9": 0.9021739130434783,
"coverage_at_5pct_error": 0.41875,
"coverage_at_1pct_error": 0.18125,
"aurc": 0.08667681796341256,
"error_rate_at_0_9": 0.09782608695652174,
"confidence_bias": 0.09826244525704297,
"top_bins": {
"0.9": {
"n": 92,
"errors": 9,
"error_rate": 0.09782608695652174
},
"0.95": {
"n": 83,
"errors": 8,
"error_rate": 0.0963855421686747
},
"0.99": {
"n": 63,
"errors": 3,
"error_rate": 0.047619047619047616
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.925,
"confidence_cutoff": 0.9656164925493048
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.84375,
"confidence_cutoff": 0.7039733738891296
}
},
"score_mae": 0.4311273968894357,
"ranked_probability_score": 0.07224781992067277
}
},
"types": {
"choice": {
"n": 480,
"nll": 0.3037822176153947,
"acc": 0.93125,
"ece": 0.050099480347752155,
"brier": 0.11775456449644288,
"mean_conf": 0.9718507476012318,
"confident_error_rate": 0.035416666666666666,
"coverage_at_0_9": 0.9270833333333334,
"accuracy_at_0_9": 0.9617977528089887,
"coverage_at_5pct_error": 0.94375,
"coverage_at_1pct_error": 0.725,
"aurc": 0.010673395731944459,
"error_rate_at_0_9": 0.038202247191011236,
"confidence_bias": 0.04060074760123178,
"top_bins": {
"0.9": {
"n": 445,
"errors": 17,
"error_rate": 0.038202247191011236
},
"0.95": {
"n": 435,
"errors": 15,
"error_rate": 0.034482758620689655
},
"0.99": {
"n": 408,
"errors": 6,
"error_rate": 0.014705882352941176
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9958333333333333,
"confidence_cutoff": 0.9999737433376136
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9895833333333334,
"confidence_cutoff": 0.9984835061671959
}
}
},
"noul": {
"n": 480,
"nll": 0.24523830247044484,
"acc": 0.9229166666666667,
"ece": 0.04796953512867931,
"brier": 0.11605074761778626,
"mean_conf": 0.970886201795346,
"confident_error_rate": 0.035416666666666666,
"coverage_at_0_9": 0.9104166666666667,
"accuracy_at_0_9": 0.9610983981693364,
"coverage_at_5pct_error": 0.94375,
"coverage_at_1pct_error": 0.71875,
"aurc": 0.01472209854520942,
"error_rate_at_0_9": 0.038901601830663615,
"confidence_bias": 0.0479695351286793,
"top_bins": {
"0.9": {
"n": 437,
"errors": 17,
"error_rate": 0.038901601830663615
},
"0.95": {
"n": 424,
"errors": 13,
"error_rate": 0.030660377358490566
},
"0.99": {
"n": 389,
"errors": 8,
"error_rate": 0.02056555269922879
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9958333333333333,
"confidence_cutoff": 0.9997753303513292
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9791666666666666,
"confidence_cutoff": 0.9930519257681523
}
}
},
"score": {
"n": 240,
"nll": 1.110705511514142,
"acc": 0.6166666666666667,
"ece": 0.16184498803254416,
"brier": 0.5539793417036516,
"mean_conf": 0.7667057960033195,
"confident_error_rate": 0.058333333333333334,
"coverage_at_0_9": 0.25416666666666665,
"accuracy_at_0_9": 0.7704918032786885,
"coverage_at_5pct_error": 0.016666666666666666,
"coverage_at_1pct_error": 0.016666666666666666,
"aurc": 0.28057303182426463,
"error_rate_at_0_9": 0.22950819672131148,
"confidence_bias": 0.15003912933665275,
"top_bins": {
"0.9": {
"n": 61,
"errors": 14,
"error_rate": 0.22950819672131148
},
"0.95": {
"n": 34,
"errors": 7,
"error_rate": 0.20588235294117646
},
"0.99": {
"n": 6,
"errors": 1,
"error_rate": 0.16666666666666666
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.6916666666666667,
"confidence_cutoff": 0.8026983558729256
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.6666666666666666,
"confidence_cutoff": 0.6054379586272015
}
},
"score_mae": 0.47431418086986715,
"ranked_probability_score": 0.08319761613591657
}
}
},
"transfer-v4": {
"raw_clean": {
"n": 656,
"nll": 0.7491303449341485,
"acc": 0.836890243902439,
"ece": 0.10484780065311013,
"brier": 0.2706697225616099,
"mean_conf": 0.9369188798045006,
"confident_error_rate": 0.08689024390243902,
"coverage_at_0_9": 0.8277439024390244,
"accuracy_at_0_9": 0.8950276243093923,
"coverage_at_5pct_error": 0.5701219512195121,
"coverage_at_1pct_error": 0.3567073170731707,
"aurc": 0.048474736987579994,
"error_rate_at_0_9": 0.10497237569060773,
"confidence_bias": 0.10002863590206157,
"top_bins": {
"0.9": {
"n": 543,
"errors": 57,
"error_rate": 0.10497237569060773
},
"0.95": {
"n": 512,
"errors": 48,
"error_rate": 0.09375
},
"0.99": {
"n": 455,
"errors": 30,
"error_rate": 0.06593406593406594
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9634146341463414,
"confidence_cutoff": 0.9997756567114997
},
"0.8": {
"coverage": 0.8003048780487805,
"accuracy": 0.8990476190476191,
"confidence_cutoff": 0.9312810361300458
}
},
"score_mae": 0.2771201607516616,
"ranked_probability_score": 0.12002374674647091
},
"as_shipped_clean": {
"n": 656,
"nll": 0.6222520301084513,
"acc": 0.836890243902439,
"ece": 0.09048309685573616,
"brier": 0.2622492394102412,
"mean_conf": 0.9203846154028757,
"confident_error_rate": 0.07317073170731707,
"coverage_at_0_9": 0.7820121951219512,
"accuracy_at_0_9": 0.9064327485380117,
"coverage_at_5pct_error": 0.5670731707317073,
"coverage_at_1pct_error": 0.3551829268292683,
"aurc": 0.0484684761006514,
"error_rate_at_0_9": 0.0935672514619883,
"confidence_bias": 0.0834943715004367,
"top_bins": {
"0.9": {
"n": 513,
"errors": 48,
"error_rate": 0.0935672514619883
},
"0.95": {
"n": 477,
"errors": 37,
"error_rate": 0.07756813417190776
},
"0.99": {
"n": 405,
"errors": 23,
"error_rate": 0.056790123456790124
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9634146341463414,
"confidence_cutoff": 0.9985297137731126
},
"0.8": {
"coverage": 0.8003048780487805,
"accuracy": 0.900952380952381,
"confidence_cutoff": 0.8749540187432164
}
},
"score_mae": 0.2768980570793622,
"ranked_probability_score": 0.11577841556824224
},
"sources": {
"composition_holdout": {
"n": 96,
"nll": 0.8314301766731748,
"acc": 0.8645833333333334,
"ece": 0.12259156340261834,
"brier": 0.25260603356221595,
"mean_conf": 0.9783762896019588,
"confident_error_rate": 0.11458333333333333,
"coverage_at_0_9": 0.9479166666666666,
"accuracy_at_0_9": 0.8791208791208791,
"coverage_at_5pct_error": 0.7291666666666666,
"coverage_at_1pct_error": 0.6666666666666666,
"aurc": 0.029361211962366768,
"error_rate_at_0_9": 0.12087912087912088,
"confidence_bias": 0.1137929562686254,
"top_bins": {
"0.9": {
"n": 91,
"errors": 11,
"error_rate": 0.12087912087912088
},
"0.95": {
"n": 90,
"errors": 11,
"error_rate": 0.12222222222222222
},
"0.99": {
"n": 86,
"errors": 11,
"error_rate": 0.12790697674418605
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.99991410687893
},
"0.8": {
"coverage": 0.8020833333333334,
"accuracy": 0.948051948051948,
"confidence_cutoff": 0.9986476483509216
}
}
},
"contrastive": {
"n": 80,
"nll": 0.46595606789147936,
"acc": 0.85,
"ece": 0.11588755121828759,
"brier": 0.22911401994889413,
"mean_conf": 0.9658875512182877,
"confident_error_rate": 0.0875,
"coverage_at_0_9": 0.9,
"accuracy_at_0_9": 0.9027777777777778,
"coverage_at_5pct_error": 0.85,
"coverage_at_1pct_error": 0.8,
"aurc": 0.017400442212878862,
"error_rate_at_0_9": 0.09722222222222222,
"confidence_bias": 0.11588755121828775,
"top_bins": {
"0.9": {
"n": 72,
"errors": 7,
"error_rate": 0.09722222222222222
},
"0.95": {
"n": 70,
"errors": 5,
"error_rate": 0.07142857142857142
},
"0.99": {
"n": 68,
"errors": 3,
"error_rate": 0.04411764705882353
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999823513518187
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9993279314433398
}
},
"score_mae": 0.2768980570793622,
"ranked_probability_score": 0.11577841556824224
},
"emotion": {
"n": 80,
"nll": 1.1680421729742996,
"acc": 0.6625,
"ece": 0.15215129647550107,
"brier": 0.49482559482198313,
"mean_conf": 0.767505586427657,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.425,
"accuracy_at_0_9": 0.8529411764705882,
"coverage_at_5pct_error": 0.075,
"coverage_at_1pct_error": 0.075,
"aurc": 0.18480906195062094,
"error_rate_at_0_9": 0.14705882352941177,
"confidence_bias": 0.105005586427657,
"top_bins": {
"0.9": {
"n": 34,
"errors": 5,
"error_rate": 0.14705882352941177
},
"0.95": {
"n": 26,
"errors": 3,
"error_rate": 0.11538461538461539
},
"0.99": {
"n": 8,
"errors": 1,
"error_rate": 0.125
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.8,
"confidence_cutoff": 0.8315422519453946
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.734375,
"confidence_cutoff": 0.5608119629701582
}
}
},
"mmlu": {
"n": 80,
"nll": 0.8706892848634225,
"acc": 0.725,
"ece": 0.14396914879993838,
"brier": 0.42715412327538493,
"mean_conf": 0.8115029447218166,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.475,
"accuracy_at_0_9": 0.868421052631579,
"coverage_at_5pct_error": 0.375,
"coverage_at_1pct_error": 0.3375,
"aurc": 0.11356864595903862,
"error_rate_at_0_9": 0.13157894736842105,
"confidence_bias": 0.08650294472181663,
"top_bins": {
"0.9": {
"n": 38,
"errors": 5,
"error_rate": 0.13157894736842105
},
"0.95": {
"n": 28,
"errors": 1,
"error_rate": 0.03571428571428571
},
"0.99": {
"n": 18,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.85,
"confidence_cutoff": 0.8706443942318668
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.796875,
"confidence_cutoff": 0.5685435872401945
}
}
},
"paws": {
"n": 80,
"nll": 0.8953763773747229,
"acc": 0.8,
"ece": 0.16414062885679848,
"brier": 0.31529713556607303,
"mean_conf": 0.947798906827078,
"confident_error_rate": 0.125,
"coverage_at_0_9": 0.8625,
"accuracy_at_0_9": 0.855072463768116,
"coverage_at_5pct_error": 0.3,
"coverage_at_1pct_error": 0.15,
"aurc": 0.09544667963478717,
"error_rate_at_0_9": 0.14492753623188406,
"confidence_bias": 0.14779890682707797,
"top_bins": {
"0.9": {
"n": 69,
"errors": 10,
"error_rate": 0.14492753623188406
},
"0.95": {
"n": 69,
"errors": 10,
"error_rate": 0.14492753623188406
},
"0.99": {
"n": 60,
"errors": 6,
"error_rate": 0.1
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.875,
"confidence_cutoff": 0.9995333943902521
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.890625,
"confidence_cutoff": 0.9804900927788639
}
}
},
"qnli": {
"n": 80,
"nll": 0.2866833743812309,
"acc": 0.925,
"ece": 0.07720126969420893,
"brier": 0.1353782597318704,
"mean_conf": 0.969044048844119,
"confident_error_rate": 0.05,
"coverage_at_0_9": 0.9,
"accuracy_at_0_9": 0.9444444444444444,
"coverage_at_5pct_error": 0.8625,
"coverage_at_1pct_error": 0.6125,
"aurc": 0.01474092888988553,
"error_rate_at_0_9": 0.05555555555555555,
"confidence_bias": 0.044044048844118944,
"top_bins": {
"0.9": {
"n": 72,
"errors": 4,
"error_rate": 0.05555555555555555
},
"0.95": {
"n": 69,
"errors": 3,
"error_rate": 0.043478260869565216
},
"0.99": {
"n": 64,
"errors": 2,
"error_rate": 0.03125
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9994500997969659
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.96875,
"confidence_cutoff": 0.9906893719466282
}
}
},
"sciq": {
"n": 80,
"nll": 0.007797247790721174,
"acc": 1.0,
"ece": 0.007304169605922202,
"brier": 0.00159066044034253,
"mean_conf": 0.9926958303940776,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.975,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 1.0,
"aurc": 0.0,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.007304169605922395,
"top_bins": {
"0.9": {
"n": 78,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 76,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 71,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999411463988364
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.999496345294104
}
}
},
"tweet_offensive": {
"n": 80,
"nll": 0.4102059096056144,
"acc": 0.8625,
"ece": 0.082947906119192,
"brier": 0.24395672910477048,
"mean_conf": 0.9186674303481943,
"confident_error_rate": 0.075,
"coverage_at_0_9": 0.7375,
"accuracy_at_0_9": 0.8983050847457628,
"coverage_at_5pct_error": 0.575,
"coverage_at_1pct_error": 0.5125,
"aurc": 0.05023960594359948,
"error_rate_at_0_9": 0.1016949152542373,
"confidence_bias": 0.05616743034819427,
"top_bins": {
"0.9": {
"n": 59,
"errors": 6,
"error_rate": 0.1016949152542373
},
"0.95": {
"n": 49,
"errors": 4,
"error_rate": 0.08163265306122448
},
"0.99": {
"n": 30,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9765148863831499
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.875,
"confidence_cutoff": 0.8834322098576286
}
}
}
},
"types": {
"choice": {
"n": 336,
"nll": 0.724820218484822,
"acc": 0.8154761904761905,
"ece": 0.09541834159449251,
"brier": 0.29207086162199275,
"mean_conf": 0.8918466450156909,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.7172619047619048,
"accuracy_at_0_9": 0.9128630705394191,
"coverage_at_5pct_error": 0.48214285714285715,
"coverage_at_1pct_error": 0.36904761904761907,
"aurc": 0.05557065380684749,
"error_rate_at_0_9": 0.08713692946058091,
"confidence_bias": 0.07637045453950042,
"top_bins": {
"0.9": {
"n": 241,
"errors": 21,
"error_rate": 0.08713692946058091
},
"0.95": {
"n": 220,
"errors": 15,
"error_rate": 0.06818181818181818
},
"0.99": {
"n": 183,
"errors": 12,
"error_rate": 0.06557377049180328
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9464285714285714,
"confidence_cutoff": 0.9967131547386102
},
"0.8": {
"coverage": 0.8005952380952381,
"accuracy": 0.8921933085501859,
"confidence_cutoff": 0.7705104187347486
}
}
},
"noul": {
"n": 280,
"nll": 0.4549349162962509,
"acc": 0.8821428571428571,
"ece": 0.08452238489264785,
"brier": 0.19846632133861988,
"mean_conf": 0.9530010972612158,
"confident_error_rate": 0.07142857142857142,
"coverage_at_0_9": 0.8571428571428571,
"accuracy_at_0_9": 0.9166666666666666,
"coverage_at_5pct_error": 0.7607142857142857,
"coverage_at_1pct_error": 0.2571428571428571,
"aurc": 0.03904590143532812,
"error_rate_at_0_9": 0.08333333333333333,
"confidence_bias": 0.07085824011835873,
"top_bins": {
"0.9": {
"n": 240,
"errors": 20,
"error_rate": 0.08333333333333333
},
"0.95": {
"n": 227,
"errors": 17,
"error_rate": 0.07488986784140969
},
"0.99": {
"n": 194,
"errors": 8,
"error_rate": 0.041237113402061855
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9571428571428572,
"confidence_cutoff": 0.9989211416259292
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9330357142857143,
"confidence_cutoff": 0.95900292163327
}
}
},
"score": {
"n": 40,
"nll": 0.9318990444323394,
"acc": 0.7,
"ece": 0.2317881936468471,
"brier": 0.45822803933287715,
"mean_conf": 0.9317881936468471,
"confident_error_rate": 0.175,
"coverage_at_0_9": 0.8,
"accuracy_at_0_9": 0.78125,
"coverage_at_5pct_error": 0.65,
"coverage_at_1pct_error": 0.6,
"aurc": 0.07545101908306909,
"error_rate_at_0_9": 0.21875,
"confidence_bias": 0.2317881936468471,
"top_bins": {
"0.9": {
"n": 32,
"errors": 7,
"error_rate": 0.21875
},
"0.95": {
"n": 30,
"errors": 5,
"error_rate": 0.16666666666666666
},
"0.99": {
"n": 28,
"errors": 3,
"error_rate": 0.10714285714285714
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9997762256879271
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.78125,
"confidence_cutoff": 0.9360940664483365
}
},
"score_mae": 0.2768980570793622,
"ranked_probability_score": 0.11577841556824224
}
}
}
},
"temperature": 1.2873604198505975
},
{
"epoch": 3,
"development": {
"raw": {
"questions": 14912,
"accuracy": 0.9075912017167382,
"mean_class_recall": 0.8447857522159936,
"nll": 0.24807737680415345,
"brier": 0.12998082994334342,
"ece15": 0.027028147820359012,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.8103540772532188,
"error_rate_at_confidence_0_9": 0.023502151605428664,
"ordinal_expected_level_mae": 0.00481932257817722,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 222,
"accuracy": 0.9774774774774775,
"mean_class_recall": 0.9570930232558139,
"nll": 0.10407641066616782,
"brier": 0.039347876403718875,
"ece15": 0.021432863016197557,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9864864864864865,
"error_rate_at_confidence_0_9": 0.0136986301369863,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 222,
"accuracy": 0.9819819819819819,
"mean_class_recall": 0.96,
"nll": 0.10023555924012383,
"brier": 0.031563944714978254,
"ece15": 0.015911253108198964,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9864864864864865,
"error_rate_at_confidence_0_9": 0.0136986301369863,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 1400,
"accuracy": 0.9885714285714285,
"mean_class_recall": 0.9821428571428572,
"nll": 0.0442702602822601,
"brier": 0.02139513973607548,
"ece15": 0.007710400104216533,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9842857142857143,
"error_rate_at_confidence_0_9": 0.008708272859216255,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 5000,
"accuracy": 0.8418,
"mean_class_recall": 0.7537333333333334,
"nll": 0.4173917752941542,
"brier": 0.22451704505631584,
"ece15": 0.06284084146027137,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.6928,
"error_rate_at_confidence_0_9": 0.06293302540415704,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0014205610073774013,
"brier": 0.0002211290260091384,
"ece15": 0.0013613576932818498,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9919354838709677,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.000610657881540054,
"brier": 8.099314527924365e-05,
"ece15": 0.0005891478461821947,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 1.5480302068279906e-07,
"brier": 4.4217021643060835e-12,
"ece15": 1.548019148098009e-07,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 2500,
"accuracy": 0.9792,
"mean_class_recall": 0.9818666666666667,
"nll": 0.11218220058914397,
"brier": 0.03724045269615108,
"ece15": 0.01842697861877182,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9928,
"error_rate_at_confidence_0_9": 0.015310233682514102,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 797,
"accuracy": 0.9962358845671268,
"mean_class_recall": 0.9963898916967509,
"nll": 0.012217254972570101,
"brier": 0.006818829719414822,
"ece15": 0.004133398699556409,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9949811794228356,
"error_rate_at_confidence_0_9": 0.0012610340479192938,
"ordinal_expected_level_mae": 0.00481932257817722
},
"twins_streamlet:twins_policy_action": {
"questions": 797,
"accuracy": 0.9949811794228356,
"mean_class_recall": 0.9951865222623345,
"nll": 0.013078119892699127,
"brier": 0.007721112955228842,
"ece15": 0.003423998250058524,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9899623588456713,
"error_rate_at_confidence_0_9": 0.0012674271229404308,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 797,
"accuracy": 0.9949811794228356,
"mean_class_recall": 0.994276682635671,
"nll": 0.01027058602627885,
"brier": 0.007209131739335666,
"ece15": 0.004201163796495765,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9912170639899623,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 797,
"accuracy": 0.9974905897114178,
"mean_class_recall": 0.998236331569665,
"nll": 0.017287182530912663,
"brier": 0.004953169208636321,
"ece15": 0.0025266183132433223,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.998745294855709,
"error_rate_at_confidence_0_9": 0.002512562814070352,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 800,
"accuracy": 0.69,
"mean_class_recall": 0.7116483068135455,
"nll": 0.5092641106435474,
"brier": 0.3569714849542106,
"ece15": 0.03650030376694629,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.2425,
"error_rate_at_confidence_0_9": 0.010309278350515464,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 800,
"accuracy": 0.68875,
"mean_class_recall": 0.529392069171139,
"nll": 0.96833512912194,
"brier": 0.46247412161427887,
"ece15": 0.02731423542286039,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.22625,
"error_rate_at_confidence_0_9": 0.022099447513812154,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9379621229396502,
"macro_task_balanced_accuracy": 0.9185689774676985
},
"recalibrated": {
"questions": 14912,
"accuracy": 0.9075912017167382,
"mean_class_recall": 0.8447857522159936,
"nll": 0.2269545932235271,
"brier": 0.12792407717450321,
"ece15": 0.01858051585491712,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.7862124463519313,
"error_rate_at_confidence_0_9": 0.018679631525076765,
"ordinal_expected_level_mae": 0.005329911013904147,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 222,
"accuracy": 0.9774774774774775,
"mean_class_recall": 0.9570930232558139,
"nll": 0.08092683597443519,
"brier": 0.037546344393533104,
"ece15": 0.021320738677048364,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9819819819819819,
"error_rate_at_confidence_0_9": 0.013761467889908258,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 222,
"accuracy": 0.9819819819819819,
"mean_class_recall": 0.96,
"nll": 0.07809919245507242,
"brier": 0.030623326048867618,
"ece15": 0.015019163882298042,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9819819819819819,
"error_rate_at_confidence_0_9": 0.009174311926605505,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 1400,
"accuracy": 0.9885714285714285,
"mean_class_recall": 0.9821428571428572,
"nll": 0.04094699270351949,
"brier": 0.021018406686723334,
"ece15": 0.0038434318338614677,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9678571428571429,
"error_rate_at_confidence_0_9": 0.006642066420664207,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 5000,
"accuracy": 0.8418,
"mean_class_recall": 0.7537333333333334,
"nll": 0.36679030702727994,
"brier": 0.21757064336558796,
"ece15": 0.043357678681868704,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.6394,
"error_rate_at_confidence_0_9": 0.05004691898654989,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0028701157894541487,
"brier": 0.0006216354903482641,
"ece15": 0.0026971564840139535,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9919354838709677,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.001405432942745028,
"brier": 0.00025083300955486444,
"ece15": 0.0013358908781109478,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9939024390243902,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 328,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 4.6272880953518815e-06,
"brier": 9.796066038293659e-10,
"ece15": 4.627043141135246e-06,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 2500,
"accuracy": 0.9792,
"mean_class_recall": 0.9818666666666667,
"nll": 0.0877031817842779,
"brier": 0.03599036114227489,
"ece15": 0.01538869285344686,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.988,
"error_rate_at_confidence_0_9": 0.014574898785425101,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 797,
"accuracy": 0.9962358845671268,
"mean_class_recall": 0.9963898916967509,
"nll": 0.010603240403009088,
"brier": 0.006373682001028304,
"ece15": 0.004218405852177047,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9937264742785445,
"error_rate_at_confidence_0_9": 0.0012626262626262627,
"ordinal_expected_level_mae": 0.005329911013904147
},
"twins_streamlet:twins_policy_action": {
"questions": 797,
"accuracy": 0.9949811794228356,
"mean_class_recall": 0.9951865222623345,
"nll": 0.01230072195659841,
"brier": 0.007337903591834537,
"ece15": 0.0031241457310850957,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.986198243412798,
"error_rate_at_confidence_0_9": 0.001272264631043257,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 797,
"accuracy": 0.9949811794228356,
"mean_class_recall": 0.994276682635671,
"nll": 0.009584146250019401,
"brier": 0.00659904677807738,
"ece15": 0.005028933100804552,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9899623588456713,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 797,
"accuracy": 0.9974905897114178,
"mean_class_recall": 0.998236331569665,
"nll": 0.013437428285294059,
"brier": 0.004858731838764845,
"ece15": 0.0022785888551786117,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.998745294855709,
"error_rate_at_confidence_0_9": 0.002512562814070352,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 800,
"accuracy": 0.69,
"mean_class_recall": 0.7116483068135455,
"nll": 0.5102124840713312,
"brier": 0.35753050242002155,
"ece15": 0.03811996421246104,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.22125,
"error_rate_at_confidence_0_9": 0.011299435028248588,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 800,
"accuracy": 0.68875,
"mean_class_recall": 0.529392069171139,
"nll": 0.9911460027249259,
"brier": 0.4737144267281807,
"ece15": 0.09647043512079573,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.18625,
"error_rate_at_confidence_0_9": 0.020134228187919462,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9379621229396502,
"macro_task_balanced_accuracy": 0.9185689774676985
}
},
"test": {
"raw": {
"questions": 21634,
"accuracy": 0.8801423684940372,
"mean_class_recall": 0.8422636415920437,
"nll": 0.35756736839135106,
"brier": 0.17329397786588344,
"ece15": 0.018370944185424988,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.681750947582509,
"error_rate_at_confidence_0_9": 0.017289307749677945,
"ordinal_expected_level_mae": 0.004830545729071066,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 426,
"accuracy": 0.9953051643192489,
"mean_class_recall": 0.9895833333333333,
"nll": 0.019578322598972707,
"brier": 0.008470972915844083,
"ece15": 0.0032192125141472924,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9882629107981221,
"error_rate_at_confidence_0_9": 0.0023752969121140144,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 426,
"accuracy": 0.9835680751173709,
"mean_class_recall": 0.9635416666666667,
"nll": 0.03323309985013501,
"brier": 0.020700086530374747,
"ece15": 0.012858446963474107,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9835680751173709,
"error_rate_at_confidence_0_9": 0.007159904534606206,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 800,
"accuracy": 0.95875,
"mean_class_recall": 0.9296875,
"nll": 0.14798192410675134,
"brier": 0.07208296294789789,
"ece15": 0.031237402444916955,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9575,
"error_rate_at_confidence_0_9": 0.02741514360313316,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 6000,
"accuracy": 0.9171666666666667,
"mean_class_recall": 0.8741111111111111,
"nll": 0.2872065840367128,
"brier": 0.1291518236859529,
"ece15": 0.02997403903668212,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.7491666666666666,
"error_rate_at_confidence_0_9": 0.03181312569521691,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 6.416181236130915e-05,
"brier": 3.664856836177276e-07,
"ece15": 6.406994017019252e-05,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0003909349271697467,
"brier": 5.411140895974333e-05,
"ece15": 0.000376397356322764,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9981481481481481,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 1.9301873747749096e-07,
"brier": 4.461551704214049e-12,
"ece15": 1.9301762319156735e-07,
"temperature": 1.0,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 3500,
"accuracy": 0.988,
"mean_class_recall": 0.9820952380952381,
"nll": 0.052315917826989855,
"brier": 0.021002166306533596,
"ece15": 0.005742982405630235,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9757142857142858,
"error_rate_at_confidence_0_9": 0.007320644216691069,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 659,
"accuracy": 0.9954476479514416,
"mean_class_recall": 0.9949748743718594,
"nll": 0.006489090778116727,
"brier": 0.004985928434486637,
"ece15": 0.0035010905751582675,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9954476479514416,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": 0.004830545729071066
},
"twins_streamlet:twins_policy_action": {
"questions": 659,
"accuracy": 0.9984825493171472,
"mean_class_recall": 0.9983249581239532,
"nll": 0.007759166929265847,
"brier": 0.0031279766576844436,
"ece15": 0.0017302646891685911,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9969650986342944,
"error_rate_at_confidence_0_9": 0.0015220700152207,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 659,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0012916521858483854,
"brier": 0.0006719742504699832,
"ece15": 0.0010543087689309648,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9969650986342944,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 659,
"accuracy": 0.9984825493171472,
"mean_class_recall": 0.9989130434782609,
"nll": 0.0030258654215359524,
"brier": 0.0020473738981638956,
"ece15": 0.0002839890142556098,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9954476479514416,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 3321,
"accuracy": 0.7034025895814514,
"mean_class_recall": 0.7106693655944849,
"nll": 0.5271449745997652,
"brier": 0.3679898300956757,
"ece15": 0.03254066478455375,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.21499548328816623,
"error_rate_at_confidence_0_9": 0.011204481792717087,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 3321,
"accuracy": 0.692261367058115,
"mean_class_recall": 0.49711800080440777,
"nll": 1.181957709780574,
"brier": 0.48216427161657194,
"ece15": 0.04703079874679226,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.20776874435411022,
"error_rate_at_confidence_0_9": 0.07681159420289856,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9450619006663279,
"macro_task_balanced_accuracy": 0.9242156493985226
},
"recalibrated": {
"questions": 21634,
"accuracy": 0.8801423684940372,
"mean_class_recall": 0.8422636415920437,
"nll": 0.3441350952603644,
"brier": 0.1767857045083663,
"ece15": 0.035627089616313175,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.6567440140519553,
"error_rate_at_confidence_0_9": 0.014358108108108109,
"ordinal_expected_level_mae": 0.004973955616734103,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 426,
"accuracy": 0.9953051643192489,
"mean_class_recall": 0.9895833333333333,
"nll": 0.02085470758455793,
"brier": 0.009033562612123906,
"ece15": 0.005062121140742335,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9741784037558685,
"error_rate_at_confidence_0_9": 0.0024096385542168677,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 426,
"accuracy": 0.9835680751173709,
"mean_class_recall": 0.9635416666666667,
"nll": 0.029576905474440387,
"brier": 0.019371229609764937,
"ece15": 0.0128149515297424,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9741784037558685,
"error_rate_at_confidence_0_9": 0.004819277108433735,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 800,
"accuracy": 0.95875,
"mean_class_recall": 0.9296875,
"nll": 0.1252495255860491,
"brier": 0.06891977027707019,
"ece15": 0.02206119521855819,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.93375,
"error_rate_at_confidence_0_9": 0.0214190093708166,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 6000,
"accuracy": 0.9171666666666667,
"mean_class_recall": 0.8741111111111111,
"nll": 0.258923302286342,
"brier": 0.1323925339560934,
"ece15": 0.040786104751424794,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.6995,
"error_rate_at_confidence_0_9": 0.027162258756254467,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.00036120828507376734,
"brier": 6.118819643947097e-06,
"ece15": 0.0003596634947089362,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0010259839180316675,
"brier": 0.00015355821904600842,
"ece15": 0.0009831313211751498,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9981481481481481,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 540,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 5.4752266593721125e-06,
"brier": 1.1349090161477055e-09,
"ece15": 5.474942881804694e-06,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 1.0,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 3500,
"accuracy": 0.988,
"mean_class_recall": 0.9820952380952381,
"nll": 0.0471904212042291,
"brier": 0.02136284177729358,
"ece15": 0.004930332206950378,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9597142857142857,
"error_rate_at_confidence_0_9": 0.006251860672819291,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 659,
"accuracy": 0.9954476479514416,
"mean_class_recall": 0.9949748743718594,
"nll": 0.005989405513587896,
"brier": 0.004363630530423605,
"ece15": 0.003686325313576514,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9954476479514416,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": 0.004973955616734103
},
"twins_streamlet:twins_policy_action": {
"questions": 659,
"accuracy": 0.9984825493171472,
"mean_class_recall": 0.9983249581239532,
"nll": 0.0069631222649601515,
"brier": 0.003204452557959292,
"ece15": 0.001531985329986691,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9969650986342944,
"error_rate_at_confidence_0_9": 0.0015220700152207,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 659,
"accuracy": 1.0,
"mean_class_recall": 1.0,
"nll": 0.0018952623564185236,
"brier": 0.0008114028234666237,
"ece15": 0.0016108880631044219,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9954476479514416,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 659,
"accuracy": 0.9984825493171472,
"mean_class_recall": 0.9989130434782609,
"nll": 0.0033168872089549845,
"brier": 0.0020980871703588605,
"ece15": 0.0012518453771527926,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.9954476479514416,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 3321,
"accuracy": 0.7034025895814514,
"mean_class_recall": 0.7106693655944849,
"nll": 0.5278136514287293,
"brier": 0.3689683794121852,
"ece15": 0.03240031037986953,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.1969286359530262,
"error_rate_at_confidence_0_9": 0.009174311926605505,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 3321,
"accuracy": 0.692261367058115,
"mean_class_recall": 0.49711800080440777,
"nll": 1.1560335557868109,
"brier": 0.498611302076152,
"ece15": 0.12617627958797675,
"temperature": 1.334968103387446,
"coverage_at_confidence_0_9": 0.17856067449563384,
"error_rate_at_confidence_0_9": 0.07251264755480608,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.9450619006663279,
"macro_task_balanced_accuracy": 0.9242156493985226
}
},
"general": {
"decision-v7": {
"raw_clean": {
"n": 1200,
"nll": 0.5734957671127382,
"acc": 0.8658333333333333,
"ece": 0.08617509900940432,
"brier": 0.21792288610472857,
"mean_conf": 0.9509841477415141,
"confident_error_rate": 0.0675,
"coverage_at_0_9": 0.8525,
"accuracy_at_0_9": 0.9208211143695014,
"coverage_at_5pct_error": 0.7833333333333333,
"coverage_at_1pct_error": 0.6166666666666667,
"aurc": 0.0281999905676559,
"error_rate_at_0_9": 0.07917888563049853,
"confidence_bias": 0.08515081440818073,
"top_bins": {
"0.9": {
"n": 1023,
"errors": 81,
"error_rate": 0.07917888563049853
},
"0.95": {
"n": 962,
"errors": 53,
"error_rate": 0.0550935550935551
},
"0.99": {
"n": 885,
"errors": 34,
"error_rate": 0.0384180790960452
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9933333333333333,
"confidence_cutoff": 0.9999894749108045
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9447916666666667,
"confidence_cutoff": 0.9524106736120255
}
},
"score_mae": 0.4625639513458072,
"ranked_probability_score": 0.08810095245487483
},
"as_shipped_clean": {
"n": 1200,
"nll": 0.46540017253913407,
"acc": 0.8658333333333333,
"ece": 0.07412868458003202,
"brier": 0.20783721757683873,
"mean_conf": 0.9351787338923708,
"confident_error_rate": 0.041666666666666664,
"coverage_at_0_9": 0.7983333333333333,
"accuracy_at_0_9": 0.9478079331941545,
"coverage_at_5pct_error": 0.7783333333333333,
"coverage_at_1pct_error": 0.6141666666666666,
"aurc": 0.02814862771926311,
"error_rate_at_0_9": 0.05219206680584551,
"confidence_bias": 0.06934540055903748,
"top_bins": {
"0.9": {
"n": 958,
"errors": 50,
"error_rate": 0.05219206680584551
},
"0.95": {
"n": 912,
"errors": 40,
"error_rate": 0.043859649122807015
},
"0.99": {
"n": 822,
"errors": 18,
"error_rate": 0.021897810218978103
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9933333333333333,
"confidence_cutoff": 0.9998006494793819
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9458333333333333,
"confidence_cutoff": 0.8959586216727804
}
},
"score_mae": 0.47318044838075235,
"ranked_probability_score": 0.08431774694216215
},
"sources": {
"agnews": {
"n": 240,
"nll": 0.2925312694754093,
"acc": 0.9166666666666666,
"ece": 0.05790914903059574,
"brier": 0.1357836869201196,
"mean_conf": 0.9733704974572437,
"confident_error_rate": 0.04583333333333333,
"coverage_at_0_9": 0.9291666666666667,
"accuracy_at_0_9": 0.9506726457399103,
"coverage_at_5pct_error": 0.9333333333333333,
"coverage_at_1pct_error": 0.7583333333333333,
"aurc": 0.011635752911899793,
"error_rate_at_0_9": 0.04932735426008968,
"confidence_bias": 0.05670383079057706,
"top_bins": {
"0.9": {
"n": 223,
"errors": 11,
"error_rate": 0.04932735426008968
},
"0.95": {
"n": 217,
"errors": 9,
"error_rate": 0.041474654377880185
},
"0.99": {
"n": 197,
"errors": 4,
"error_rate": 0.02030456852791878
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9996851444568575
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.984375,
"confidence_cutoff": 0.9921275226980688
}
}
},
"amazon": {
"n": 80,
"nll": 1.2207591140196519,
"acc": 0.6125,
"ece": 0.1615943676102411,
"brier": 0.5707631816876242,
"mean_conf": 0.7431377036571911,
"confident_error_rate": 0.0375,
"coverage_at_0_9": 0.2375,
"accuracy_at_0_9": 0.8421052631578947,
"coverage_at_5pct_error": 0.075,
"coverage_at_1pct_error": 0.075,
"aurc": 0.272315413996461,
"error_rate_at_0_9": 0.15789473684210525,
"confidence_bias": 0.13063770365719107,
"top_bins": {
"0.9": {
"n": 19,
"errors": 3,
"error_rate": 0.15789473684210525
},
"0.95": {
"n": 14,
"errors": 2,
"error_rate": 0.14285714285714285
},
"0.99": {
"n": 3,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.7,
"confidence_cutoff": 0.7847517191083221
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.640625,
"confidence_cutoff": 0.5534749021853737
}
},
"score_mae": 0.5042438581009662,
"ranked_probability_score": 0.09073105215000309
},
"banking77": {
"n": 80,
"nll": 0.6278787803563556,
"acc": 0.8375,
"ece": 0.1237375681167378,
"brier": 0.25438988606734264,
"mean_conf": 0.8981608539632144,
"confident_error_rate": 0.05,
"coverage_at_0_9": 0.7625,
"accuracy_at_0_9": 0.9344262295081968,
"coverage_at_5pct_error": 0.6875,
"coverage_at_1pct_error": 0.6125,
"aurc": 0.035783745285874746,
"error_rate_at_0_9": 0.06557377049180328,
"confidence_bias": 0.06066085396321441,
"top_bins": {
"0.9": {
"n": 61,
"errors": 4,
"error_rate": 0.06557377049180328
},
"0.95": {
"n": 57,
"errors": 4,
"error_rate": 0.07017543859649122
},
"0.99": {
"n": 48,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9966194161989207
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.90625,
"confidence_cutoff": 0.8503200856915479
}
}
},
"boolq": {
"n": 80,
"nll": 0.6125951836430432,
"acc": 0.875,
"ece": 0.11756712414134643,
"brier": 0.22505069986723222,
"mean_conf": 0.9514740522819943,
"confident_error_rate": 0.0875,
"coverage_at_0_9": 0.85,
"accuracy_at_0_9": 0.8970588235294118,
"coverage_at_5pct_error": 0.6875,
"coverage_at_1pct_error": 0.1125,
"aurc": 0.0511510509173725,
"error_rate_at_0_9": 0.10294117647058823,
"confidence_bias": 0.07647405228199433,
"top_bins": {
"0.9": {
"n": 68,
"errors": 7,
"error_rate": 0.10294117647058823
},
"0.95": {
"n": 67,
"errors": 7,
"error_rate": 0.1044776119402985
},
"0.99": {
"n": 61,
"errors": 4,
"error_rate": 0.06557377049180328
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.975,
"confidence_cutoff": 0.9999426387743003
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.921875,
"confidence_cutoff": 0.9819850746275051
}
}
},
"contrastive": {
"n": 160,
"nll": 0.07962468888836484,
"acc": 0.96875,
"ece": 0.026633113578874344,
"brier": 0.05007247893674288,
"mean_conf": 0.9953831135788744,
"confident_error_rate": 0.0125,
"coverage_at_0_9": 0.98125,
"accuracy_at_0_9": 0.9872611464968153,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.96875,
"aurc": 0.000752718799433766,
"error_rate_at_0_9": 0.012738853503184714,
"confidence_bias": 0.026633113578874434,
"top_bins": {
"0.9": {
"n": 157,
"errors": 2,
"error_rate": 0.012738853503184714
},
"0.95": {
"n": 156,
"errors": 2,
"error_rate": 0.01282051282051282
},
"0.99": {
"n": 148,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999957907736378
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9999070137203582
}
}
},
"dbpedia14": {
"n": 80,
"nll": 0.02165084849307894,
"acc": 0.9875,
"ece": 0.01152608625473029,
"brier": 0.016104080808397182,
"mean_conf": 0.9959462330871315,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.9875,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.9875,
"aurc": 0.00015625,
"error_rate_at_0_9": 0.0,
"confidence_bias": 0.008446233087131438,
"top_bins": {
"0.9": {
"n": 79,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 78,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 77,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999981627832084
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9999896312065648
}
}
},
"imdb": {
"n": 80,
"nll": 0.06521915089368877,
"acc": 0.9625,
"ece": 0.02850319056358286,
"brier": 0.04639252879503235,
"mean_conf": 0.985401588169408,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.95,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.95,
"aurc": 0.0012680561790530144,
"error_rate_at_0_9": 0.0,
"confidence_bias": 0.022901588169408016,
"top_bins": {
"0.9": {
"n": 76,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 75,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 71,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999903501374642
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9998069111372814
}
}
},
"mnli": {
"n": 80,
"nll": 0.5051164719750952,
"acc": 0.9125,
"ece": 0.08269743141349266,
"brier": 0.16426794867102618,
"mean_conf": 0.9883215126078937,
"confident_error_rate": 0.075,
"coverage_at_0_9": 0.95,
"accuracy_at_0_9": 0.9210526315789473,
"coverage_at_5pct_error": 0.8875,
"coverage_at_1pct_error": 0.65,
"aurc": 0.015551107945666897,
"error_rate_at_0_9": 0.07894736842105263,
"confidence_bias": 0.07582151260789372,
"top_bins": {
"0.9": {
"n": 76,
"errors": 6,
"error_rate": 0.07894736842105263
},
"0.95": {
"n": 74,
"errors": 5,
"error_rate": 0.06756756756756757
},
"0.99": {
"n": 72,
"errors": 4,
"error_rate": 0.05555555555555555
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999963968506167
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.96875,
"confidence_cutoff": 0.9998914656545862
}
}
},
"sst5": {
"n": 80,
"nll": 1.2493438944839201,
"acc": 0.6125,
"ece": 0.24936946322063214,
"brier": 0.5917259991138234,
"mean_conf": 0.8126564640723977,
"confident_error_rate": 0.075,
"coverage_at_0_9": 0.3,
"accuracy_at_0_9": 0.75,
"coverage_at_5pct_error": 0.075,
"coverage_at_1pct_error": 0.075,
"aurc": 0.3025049973121495,
"error_rate_at_0_9": 0.25,
"confidence_bias": 0.20015646407239762,
"top_bins": {
"0.9": {
"n": 24,
"errors": 6,
"error_rate": 0.25
},
"0.95": {
"n": 10,
"errors": 1,
"error_rate": 0.1
},
"0.99": {
"n": 0,
"errors": 0,
"error_rate": null
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.625,
"confidence_cutoff": 0.8548749588336553
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.625,
"confidence_cutoff": 0.7024572592103087
}
},
"score_mae": 0.48716165684142193,
"ranked_probability_score": 0.08942157579848162
},
"trec": {
"n": 80,
"nll": 0.2837433451436081,
"acc": 0.9625,
"ece": 0.028515907037196753,
"brier": 0.057680311047061936,
"mean_conf": 0.9910159070371967,
"confident_error_rate": 0.025,
"coverage_at_0_9": 0.9875,
"accuracy_at_0_9": 0.9746835443037974,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.275,
"aurc": 0.01757155053741791,
"error_rate_at_0_9": 0.02531645569620253,
"confidence_bias": 0.02851590703719664,
"top_bins": {
"0.9": {
"n": 79,
"errors": 2,
"error_rate": 0.02531645569620253
},
"0.95": {
"n": 78,
"errors": 2,
"error_rate": 0.02564102564102564
},
"0.99": {
"n": 77,
"errors": 2,
"error_rate": 0.025974025974025976
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.975,
"confidence_cutoff": 0.9999924765263039
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.984375,
"confidence_cutoff": 0.9999100942077365
}
}
},
"yelp": {
"n": 160,
"nll": 0.6789263064378058,
"acc": 0.76875,
"ece": 0.10810206862028397,
"brier": 0.34184380448059815,
"mean_conf": 0.8753444869898281,
"confident_error_rate": 0.05625,
"coverage_at_0_9": 0.6,
"accuracy_at_0_9": 0.90625,
"coverage_at_5pct_error": 0.41875,
"coverage_at_1pct_error": 0.1875,
"aurc": 0.08672761133002724,
"error_rate_at_0_9": 0.09375,
"confidence_bias": 0.1065944869898281,
"top_bins": {
"0.9": {
"n": 96,
"errors": 9,
"error_rate": 0.09375
},
"0.95": {
"n": 86,
"errors": 8,
"error_rate": 0.09302325581395349
},
"0.99": {
"n": 68,
"errors": 4,
"error_rate": 0.058823529411764705
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.925,
"confidence_cutoff": 0.971684991366676
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.84375,
"confidence_cutoff": 0.7267623543739319
}
},
"score_mae": 0.428135830199869,
"ranked_probability_score": 0.07280061287800174
}
},
"types": {
"choice": {
"n": 480,
"nll": 0.3225147096064765,
"acc": 0.9291666666666667,
"ece": 0.05269727298309242,
"brier": 0.11939209773780447,
"mean_conf": 0.9740635473132214,
"confident_error_rate": 0.0375,
"coverage_at_0_9": 0.9333333333333333,
"accuracy_at_0_9": 0.9598214285714286,
"coverage_at_5pct_error": 0.9479166666666666,
"coverage_at_1pct_error": 0.7354166666666667,
"aurc": 0.010523805656224608,
"error_rate_at_0_9": 0.04017857142857143,
"confidence_bias": 0.04489688064655473,
"top_bins": {
"0.9": {
"n": 448,
"errors": 18,
"error_rate": 0.04017857142857143
},
"0.95": {
"n": 439,
"errors": 17,
"error_rate": 0.0387243735763098
},
"0.99": {
"n": 416,
"errors": 8,
"error_rate": 0.019230769230769232
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9958333333333333,
"confidence_cutoff": 0.999989087291256
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9895833333333334,
"confidence_cutoff": 0.9991522595300467
}
}
},
"noul": {
"n": 480,
"nll": 0.2630675588624098,
"acc": 0.925,
"ece": 0.048426932319500904,
"brier": 0.11894188754077414,
"mean_conf": 0.9734269323195011,
"confident_error_rate": 0.035416666666666666,
"coverage_at_0_9": 0.91875,
"accuracy_at_0_9": 0.9614512471655329,
"coverage_at_5pct_error": 0.9416666666666667,
"coverage_at_1pct_error": 0.71875,
"aurc": 0.014292966145900442,
"error_rate_at_0_9": 0.03854875283446712,
"confidence_bias": 0.048426932319501015,
"top_bins": {
"0.9": {
"n": 441,
"errors": 17,
"error_rate": 0.03854875283446712
},
"0.95": {
"n": 431,
"errors": 15,
"error_rate": 0.03480278422273782
},
"0.99": {
"n": 397,
"errors": 9,
"error_rate": 0.022670025188916875
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9958333333333333,
"confidence_cutoff": 0.9999067506240331
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9791666666666666,
"confidence_cutoff": 0.9950087776054062
}
}
},
"score": {
"n": 240,
"nll": 1.1558363257578976,
"acc": 0.6208333333333333,
"ece": 0.1746159580142597,
"brier": 0.5625181173270365,
"mean_conf": 0.7809127101964097,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.2875,
"accuracy_at_0_9": 0.782608695652174,
"coverage_at_5pct_error": 0.016666666666666666,
"coverage_at_1pct_error": 0.016666666666666666,
"aurc": 0.28163895271092954,
"error_rate_at_0_9": 0.21739130434782608,
"confidence_bias": 0.16007937686307638,
"top_bins": {
"0.9": {
"n": 69,
"errors": 15,
"error_rate": 0.21739130434782608
},
"0.95": {
"n": 42,
"errors": 8,
"error_rate": 0.19047619047619047
},
"0.99": {
"n": 9,
"errors": 1,
"error_rate": 0.1111111111111111
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.6916666666666667,
"confidence_cutoff": 0.8195696673851967
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.6614583333333334,
"confidence_cutoff": 0.6215591950077792
}
},
"score_mae": 0.47318044838075235,
"ranked_probability_score": 0.08431774694216215
}
}
},
"transfer-v4": {
"raw_clean": {
"n": 656,
"nll": 0.840653344789253,
"acc": 0.8384146341463414,
"ece": 0.11570395215040587,
"brier": 0.2772648298322645,
"mean_conf": 0.9433821481191502,
"confident_error_rate": 0.0899390243902439,
"coverage_at_0_9": 0.8429878048780488,
"accuracy_at_0_9": 0.8933092224231465,
"coverage_at_5pct_error": 0.5777439024390244,
"coverage_at_1pct_error": 0.3551829268292683,
"aurc": 0.0494438083179723,
"error_rate_at_0_9": 0.10669077757685352,
"confidence_bias": 0.10496751397280879,
"top_bins": {
"0.9": {
"n": 553,
"errors": 59,
"error_rate": 0.10669077757685352
},
"0.95": {
"n": 527,
"errors": 54,
"error_rate": 0.10246679316888045
},
"0.99": {
"n": 472,
"errors": 34,
"error_rate": 0.07203389830508475
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9634146341463414,
"confidence_cutoff": 0.9999382193883597
},
"0.8": {
"coverage": 0.8003048780487805,
"accuracy": 0.900952380952381,
"confidence_cutoff": 0.951274711387887
}
},
"score_mae": 0.2762329771748361,
"ranked_probability_score": 0.12228321967451691
},
"as_shipped_clean": {
"n": 656,
"nll": 0.6710659810048166,
"acc": 0.8384146341463414,
"ece": 0.09231046175305074,
"brier": 0.26721979560494846,
"mean_conf": 0.9260829686930101,
"confident_error_rate": 0.07926829268292683,
"coverage_at_0_9": 0.7987804878048781,
"accuracy_at_0_9": 0.9007633587786259,
"coverage_at_5pct_error": 0.5746951219512195,
"coverage_at_1pct_error": 0.3551829268292683,
"aurc": 0.049407167334717506,
"error_rate_at_0_9": 0.09923664122137404,
"confidence_bias": 0.08766833454666867,
"top_bins": {
"0.9": {
"n": 524,
"errors": 52,
"error_rate": 0.09923664122137404
},
"0.95": {
"n": 493,
"errors": 42,
"error_rate": 0.08519269776876268
},
"0.99": {
"n": 422,
"errors": 23,
"error_rate": 0.054502369668246446
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9573170731707317,
"confidence_cutoff": 0.9992490894147191
},
"0.8": {
"coverage": 0.8003048780487805,
"accuracy": 0.8990476190476191,
"confidence_cutoff": 0.8992331757610149
}
},
"score_mae": 0.27589619803682164,
"ranked_probability_score": 0.11777542530670457
},
"sources": {
"composition_holdout": {
"n": 96,
"nll": 0.946778069631199,
"acc": 0.8645833333333334,
"ece": 0.14288844371789414,
"brier": 0.2646184933261653,
"mean_conf": 0.9787006948112301,
"confident_error_rate": 0.11458333333333333,
"coverage_at_0_9": 0.9479166666666666,
"accuracy_at_0_9": 0.8791208791208791,
"coverage_at_5pct_error": 0.6875,
"coverage_at_1pct_error": 0.65625,
"aurc": 0.032783504799663264,
"error_rate_at_0_9": 0.12087912087912088,
"confidence_bias": 0.11411736147789675,
"top_bins": {
"0.9": {
"n": 91,
"errors": 11,
"error_rate": 0.12087912087912088
},
"0.95": {
"n": 90,
"errors": 11,
"error_rate": 0.12222222222222222
},
"0.99": {
"n": 89,
"errors": 11,
"error_rate": 0.12359550561797752
}
},
"selective": {
"0.5": {
"coverage": 0.5104166666666666,
"accuracy": 1.0,
"confidence_cutoff": 0.9999696869828117
},
"0.8": {
"coverage": 0.8125,
"accuracy": 0.8974358974358975,
"confidence_cutoff": 0.9992490894147191
}
}
},
"contrastive": {
"n": 80,
"nll": 0.5115728584657595,
"acc": 0.875,
"ece": 0.13006292045346943,
"brier": 0.23406192561698297,
"mean_conf": 0.9705133073135578,
"confident_error_rate": 0.0875,
"coverage_at_0_9": 0.9125,
"accuracy_at_0_9": 0.9041095890410958,
"coverage_at_5pct_error": 0.85,
"coverage_at_1pct_error": 0.8,
"aurc": 0.0169297143647776,
"error_rate_at_0_9": 0.0958904109589041,
"confidence_bias": 0.09551330731355778,
"top_bins": {
"0.9": {
"n": 73,
"errors": 7,
"error_rate": 0.0958904109589041
},
"0.95": {
"n": 71,
"errors": 6,
"error_rate": 0.08450704225352113
},
"0.99": {
"n": 68,
"errors": 3,
"error_rate": 0.04411764705882353
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999941701311645
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9997393381382442
}
},
"score_mae": 0.27589619803682164,
"ranked_probability_score": 0.11777542530670457
},
"emotion": {
"n": 80,
"nll": 1.2094004500158717,
"acc": 0.675,
"ece": 0.19812590622433582,
"brier": 0.4998620899742045,
"mean_conf": 0.7808203087114324,
"confident_error_rate": 0.075,
"coverage_at_0_9": 0.4375,
"accuracy_at_0_9": 0.8285714285714286,
"coverage_at_5pct_error": 0.075,
"coverage_at_1pct_error": 0.075,
"aurc": 0.1832973887490438,
"error_rate_at_0_9": 0.17142857142857143,
"confidence_bias": 0.10582030871143233,
"top_bins": {
"0.9": {
"n": 35,
"errors": 6,
"error_rate": 0.17142857142857143
},
"0.95": {
"n": 30,
"errors": 4,
"error_rate": 0.13333333333333333
},
"0.99": {
"n": 12,
"errors": 1,
"error_rate": 0.08333333333333333
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.8,
"confidence_cutoff": 0.8500919215367448
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.734375,
"confidence_cutoff": 0.5761070047579089
}
}
},
"mmlu": {
"n": 80,
"nll": 0.9027907855123264,
"acc": 0.6875,
"ece": 0.14251964152159943,
"brier": 0.4317941195669751,
"mean_conf": 0.8238540468148077,
"confident_error_rate": 0.075,
"coverage_at_0_9": 0.5,
"accuracy_at_0_9": 0.85,
"coverage_at_5pct_error": 0.375,
"coverage_at_1pct_error": 0.325,
"aurc": 0.11589530494156297,
"error_rate_at_0_9": 0.15,
"confidence_bias": 0.13635404681480767,
"top_bins": {
"0.9": {
"n": 40,
"errors": 6,
"error_rate": 0.15
},
"0.95": {
"n": 30,
"errors": 1,
"error_rate": 0.03333333333333333
},
"0.99": {
"n": 19,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.85,
"confidence_cutoff": 0.9017783630155174
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.8125,
"confidence_cutoff": 0.613701094142181
}
}
},
"paws": {
"n": 80,
"nll": 0.9771542282927816,
"acc": 0.8125,
"ece": 0.1651648000763565,
"brier": 0.31695123775026923,
"mean_conf": 0.9524627544936155,
"confident_error_rate": 0.125,
"coverage_at_0_9": 0.8625,
"accuracy_at_0_9": 0.855072463768116,
"coverage_at_5pct_error": 0.15,
"coverage_at_1pct_error": 0.15,
"aurc": 0.0997884943666906,
"error_rate_at_0_9": 0.14492753623188406,
"confidence_bias": 0.1399627544936155,
"top_bins": {
"0.9": {
"n": 69,
"errors": 10,
"error_rate": 0.14492753623188406
},
"0.95": {
"n": 69,
"errors": 10,
"error_rate": 0.14492753623188406
},
"0.99": {
"n": 61,
"errors": 6,
"error_rate": 0.09836065573770492
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.875,
"confidence_cutoff": 0.9997721781585782
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.890625,
"confidence_cutoff": 0.9846682768581441
}
}
},
"qnli": {
"n": 80,
"nll": 0.32010583310414403,
"acc": 0.925,
"ece": 0.06578826678776872,
"brier": 0.13874220003757964,
"mean_conf": 0.9719094562930305,
"confident_error_rate": 0.05,
"coverage_at_0_9": 0.925,
"accuracy_at_0_9": 0.9459459459459459,
"coverage_at_5pct_error": 0.8625,
"coverage_at_1pct_error": 0.6,
"aurc": 0.015384772492900507,
"error_rate_at_0_9": 0.05405405405405406,
"confidence_bias": 0.04690945629303045,
"top_bins": {
"0.9": {
"n": 74,
"errors": 4,
"error_rate": 0.05405405405405406
},
"0.95": {
"n": 72,
"errors": 4,
"error_rate": 0.05555555555555555
},
"0.99": {
"n": 66,
"errors": 2,
"error_rate": 0.030303030303030304
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9997714445133897
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.96875,
"confidence_cutoff": 0.9937610824296935
}
}
},
"sciq": {
"n": 80,
"nll": 0.00696809639784113,
"acc": 1.0,
"ece": 0.006546558428673463,
"brier": 0.0013901896311441294,
"mean_conf": 0.9934534415713265,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.975,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 1.0,
"aurc": 0.0,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.006546558428673466,
"top_bins": {
"0.9": {
"n": 78,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 76,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 72,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999699221173486
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9997295605534583
}
}
},
"tweet_offensive": {
"n": 80,
"nll": 0.4386151088933339,
"acc": 0.8625,
"ece": 0.10954228653253775,
"brier": 0.2508583693920234,
"mean_conf": 0.926426194311437,
"confident_error_rate": 0.1,
"coverage_at_0_9": 0.8,
"accuracy_at_0_9": 0.875,
"coverage_at_5pct_error": 0.575,
"coverage_at_1pct_error": 0.4875,
"aurc": 0.05180455375068149,
"error_rate_at_0_9": 0.125,
"confidence_bias": 0.06392619431143698,
"top_bins": {
"0.9": {
"n": 64,
"errors": 8,
"error_rate": 0.125
},
"0.95": {
"n": 55,
"errors": 6,
"error_rate": 0.10909090909090909
},
"0.99": {
"n": 35,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.975,
"confidence_cutoff": 0.9842379652550892
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.875,
"confidence_cutoff": 0.902661270668069
}
}
}
},
"types": {
"choice": {
"n": 336,
"nll": 0.7750697655913044,
"acc": 0.8095238095238095,
"ece": 0.09777001761032751,
"brier": 0.2977591883723148,
"mean_conf": 0.8982306263978672,
"confident_error_rate": 0.06845238095238096,
"coverage_at_0_9": 0.7261904761904762,
"accuracy_at_0_9": 0.9057377049180327,
"coverage_at_5pct_error": 0.4583333333333333,
"coverage_at_1pct_error": 0.3630952380952381,
"aurc": 0.05730883047817376,
"error_rate_at_0_9": 0.0942622950819672,
"confidence_bias": 0.08870681687405768,
"top_bins": {
"0.9": {
"n": 244,
"errors": 23,
"error_rate": 0.0942622950819672
},
"0.95": {
"n": 226,
"errors": 16,
"error_rate": 0.07079646017699115
},
"0.99": {
"n": 192,
"errors": 12,
"error_rate": 0.0625
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9464285714285714,
"confidence_cutoff": 0.9979451630958165
},
"0.8": {
"coverage": 0.8005952380952381,
"accuracy": 0.8921933085501859,
"confidence_cutoff": 0.805212947551472
}
}
},
"noul": {
"n": 280,
"nll": 0.4959650021746835,
"acc": 0.8857142857142857,
"ece": 0.08836865713253901,
"brier": 0.20187194492263405,
"mean_conf": 0.9573703050826868,
"confident_error_rate": 0.07857142857142857,
"coverage_at_0_9": 0.8821428571428571,
"accuracy_at_0_9": 0.9109311740890689,
"coverage_at_5pct_error": 0.7535714285714286,
"coverage_at_1pct_error": 0.25,
"aurc": 0.039651831256480245,
"error_rate_at_0_9": 0.08906882591093117,
"confidence_bias": 0.07165601936840116,
"top_bins": {
"0.9": {
"n": 247,
"errors": 22,
"error_rate": 0.08906882591093117
},
"0.95": {
"n": 236,
"errors": 20,
"error_rate": 0.0847457627118644
},
"0.99": {
"n": 202,
"errors": 8,
"error_rate": 0.039603960396039604
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9571428571428572,
"confidence_cutoff": 0.9995423463157226
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9330357142857143,
"confidence_cutoff": 0.9697351853044945
}
}
},
"score": {
"n": 40,
"nll": 1.0231410422892537,
"acc": 0.75,
"ece": 0.26013051552429656,
"brier": 0.46812385113527205,
"mean_conf": 0.9410312892444732,
"confident_error_rate": 0.175,
"coverage_at_0_9": 0.825,
"accuracy_at_0_9": 0.7878787878787878,
"coverage_at_5pct_error": 0.65,
"coverage_at_1pct_error": 0.6,
"aurc": 0.07355999344204343,
"error_rate_at_0_9": 0.21212121212121213,
"confidence_bias": 0.1910312892444732,
"top_bins": {
"0.9": {
"n": 33,
"errors": 7,
"error_rate": 0.21212121212121213
},
"0.95": {
"n": 31,
"errors": 6,
"error_rate": 0.1935483870967742
},
"0.99": {
"n": 28,
"errors": 3,
"error_rate": 0.10714285714285714
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9999132212992297
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.78125,
"confidence_cutoff": 0.9461605303081507
}
},
"score_mae": 0.27589619803682164,
"ranked_probability_score": 0.11777542530670457
}
}
}
},
"temperature": 1.334968103387446
}
],
"kev_4b_security": {
"test": {
"raw": {
"questions": 21634,
"accuracy": 0.5048996949246556,
"mean_class_recall": 0.5647195198307282,
"nll": 1.2328084492499014,
"brier": 0.7147920651499965,
"ece15": 0.3370024824919156,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.44924655634649163,
"error_rate_at_confidence_0_9": 0.34736083959255065,
"ordinal_expected_level_mae": 0.2811370953407191,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 426,
"accuracy": 0.5422535211267606,
"mean_class_recall": 0.7045454545454546,
"nll": 1.2120829748020316,
"brier": 0.6804576908439421,
"ece15": 0.2992356768561654,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.5187793427230047,
"error_rate_at_confidence_0_9": 0.3936651583710407,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 426,
"accuracy": 0.7300469483568075,
"mean_class_recall": 0.8257575757575757,
"nll": 0.6169123027998439,
"brier": 0.3918536224381035,
"ece15": 0.1421670820222778,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.57981220657277,
"error_rate_at_confidence_0_9": 0.15384615384615385,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 800,
"accuracy": 0.79875,
"mean_class_recall": 0.49921875,
"nll": 0.5895657949467273,
"brier": 0.35429205012331033,
"ece15": 0.1133528731473528,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.2175,
"error_rate_at_confidence_0_9": 0.4942528735632184,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 6000,
"accuracy": 0.2628333333333333,
"mean_class_recall": 0.5085555555555555,
"nll": 1.2759496387732032,
"brier": 0.9560488943280168,
"ece15": 0.5505303178793823,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.12266666666666666,
"error_rate_at_confidence_0_9": 0.453804347826087,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 0.6290322580645161,
"mean_class_recall": 0.6714285714285715,
"nll": 1.0817458611434574,
"brier": 0.6236238714203711,
"ece15": 0.30916032986760006,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.75,
"error_rate_at_confidence_0_9": 0.34408602150537637,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 540,
"accuracy": 0.7981481481481482,
"mean_class_recall": 0.7528344671201813,
"nll": 0.4871733371313632,
"brier": 0.29625578551796805,
"ece15": 0.15242774183988111,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.7962962962962963,
"error_rate_at_confidence_0_9": 0.08837209302325581,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 540,
"accuracy": 0.9944444444444445,
"mean_class_recall": 0.9886363636363636,
"nll": 0.01641532065146085,
"brier": 0.007136931607478901,
"ece15": 0.013212748100349914,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9777777777777777,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 3500,
"accuracy": 0.5065714285714286,
"mean_class_recall": 0.6710476190476191,
"nll": 1.1359565067657027,
"brier": 0.6406729439019092,
"ece15": 0.23932739622085006,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.38971428571428574,
"error_rate_at_confidence_0_9": 0.3878299120234604,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 659,
"accuracy": 0.8118361153262519,
"mean_class_recall": 0.794717762064859,
"nll": 0.47493573588084764,
"brier": 0.2794985686136285,
"ece15": 0.05201558678606951,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.4931714719271624,
"error_rate_at_confidence_0_9": 0.04923076923076923,
"ordinal_expected_level_mae": 0.2811370953407191
},
"twins_streamlet:twins_policy_action": {
"questions": 659,
"accuracy": 0.7177541729893778,
"mean_class_recall": 0.6923577773226661,
"nll": 0.5581275002335633,
"brier": 0.3443282053616745,
"ece15": 0.09062810464195983,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.44613050075872535,
"error_rate_at_confidence_0_9": 0.04421768707482993,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 659,
"accuracy": 0.7814871016691958,
"mean_class_recall": 0.7019105077928607,
"nll": 0.4980710054489271,
"brier": 0.31257573538065736,
"ece15": 0.09857831652926007,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.590288315629742,
"error_rate_at_confidence_0_9": 0.09511568123393316,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 659,
"accuracy": 0.9711684370257967,
"mean_class_recall": 0.9693685820406379,
"nll": 0.10259254119815524,
"brier": 0.05091487551434137,
"ece15": 0.030811660924900536,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.8679817905918058,
"error_rate_at_confidence_0_9": 0.006993006993006993,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 3321,
"accuracy": 0.47756699789220114,
"mean_class_recall": 0.5,
"nll": 1.2951885811608232,
"brier": 0.8663028363232907,
"ece15": 0.42951687019840074,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.7633242999096658,
"error_rate_at_confidence_0_9": 0.504930966469428,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 3321,
"accuracy": 0.48087925323697683,
"mean_class_recall": 0.12233932942660039,
"nll": 2.41017951541039,
"brier": 0.8961545633575424,
"ece15": 0.3893600279043549,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.5453176753989762,
"error_rate_at_confidence_0_9": 0.4870237437879624,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.6787694400132313,
"macro_task_balanced_accuracy": 0.671622736838496
},
"recalibrated": {
"questions": 21634,
"accuracy": 0.5048996949246556,
"mean_class_recall": 0.5647195198307282,
"nll": 1.2328084492499014,
"brier": 0.7147920651499965,
"ece15": 0.3370024824919156,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.44924655634649163,
"error_rate_at_confidence_0_9": 0.34736083959255065,
"ordinal_expected_level_mae": 0.2811370953407191,
"tasks": {
"agentdojo_tool_outputs:agentdojo_boundary_action": {
"questions": 426,
"accuracy": 0.5422535211267606,
"mean_class_recall": 0.7045454545454546,
"nll": 1.2120829748020316,
"brier": 0.6804576908439421,
"ece15": 0.2992356768561654,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.5187793427230047,
"error_rate_at_confidence_0_9": 0.3936651583710407,
"ordinal_expected_level_mae": null
},
"agentdojo_tool_outputs:agentdojo_embedded_injection": {
"questions": 426,
"accuracy": 0.7300469483568075,
"mean_class_recall": 0.8257575757575757,
"nll": 0.6169123027998439,
"brier": 0.3918536224381035,
"ece15": 0.1421670820222778,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.57981220657277,
"error_rate_at_confidence_0_9": 0.15384615384615385,
"ordinal_expected_level_mae": null
},
"byzfl_mnist:fl_attack_generated_update": {
"questions": 800,
"accuracy": 0.79875,
"mean_class_recall": 0.49921875,
"nll": 0.5895657949467273,
"brier": 0.35429205012331033,
"ece15": 0.1133528731473528,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.2175,
"error_rate_at_confidence_0_9": 0.4942528735632184,
"ordinal_expected_level_mae": null
},
"ciciot2023:ciciot_attack": {
"questions": 6000,
"accuracy": 0.2628333333333333,
"mean_class_recall": 0.5085555555555555,
"nll": 1.2759496387732032,
"brier": 0.9560488943280168,
"ece15": 0.5505303178793823,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.12266666666666666,
"error_rate_at_confidence_0_9": 0.453804347826087,
"ordinal_expected_level_mae": null
},
"injecagent:injecagent_requested_operation_v13": {
"questions": 124,
"accuracy": 0.6290322580645161,
"mean_class_recall": 0.6714285714285715,
"nll": 1.0817458611434574,
"brier": 0.6236238714203711,
"ece15": 0.30916032986760006,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.75,
"error_rate_at_confidence_0_9": 0.34408602150537637,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_known_logon_action_v16": {
"questions": 540,
"accuracy": 0.7981481481481482,
"mean_class_recall": 0.7528344671201813,
"nll": 0.4871733371313632,
"brier": 0.29625578551796805,
"ece15": 0.15242774183988111,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.7962962962962963,
"error_rate_at_confidence_0_9": 0.08837209302325581,
"ordinal_expected_level_mae": null
},
"lanl_auth_policy:lanl_observable_policy": {
"questions": 540,
"accuracy": 0.9944444444444445,
"mean_class_recall": 0.9886363636363636,
"nll": 0.01641532065146085,
"brier": 0.007136931607478901,
"ece15": 0.013212748100349914,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.9777777777777777,
"error_rate_at_confidence_0_9": 0.0,
"ordinal_expected_level_mae": null
},
"ton_iot_network:ton_attack": {
"questions": 3500,
"accuracy": 0.5065714285714286,
"mean_class_recall": 0.6710476190476191,
"nll": 1.1359565067657027,
"brier": 0.6406729439019092,
"ece15": 0.23932739622085006,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.38971428571428574,
"error_rate_at_confidence_0_9": 0.3878299120234604,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_evidence_level": {
"questions": 659,
"accuracy": 0.8118361153262519,
"mean_class_recall": 0.794717762064859,
"nll": 0.47493573588084764,
"brier": 0.2794985686136285,
"ece15": 0.05201558678606951,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.4931714719271624,
"error_rate_at_confidence_0_9": 0.04923076923076923,
"ordinal_expected_level_mae": 0.2811370953407191
},
"twins_streamlet:twins_policy_action": {
"questions": 659,
"accuracy": 0.7177541729893778,
"mean_class_recall": 0.6923577773226661,
"nll": 0.5581275002335633,
"brier": 0.3443282053616745,
"ece15": 0.09062810464195983,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.44613050075872535,
"error_rate_at_confidence_0_9": 0.04421768707482993,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_quorum": {
"questions": 659,
"accuracy": 0.7814871016691958,
"mean_class_recall": 0.7019105077928607,
"nll": 0.4980710054489271,
"brier": 0.31257573538065736,
"ece15": 0.09857831652926007,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.590288315629742,
"error_rate_at_confidence_0_9": 0.09511568123393316,
"ordinal_expected_level_mae": null
},
"twins_streamlet:twins_visible_conflict": {
"questions": 659,
"accuracy": 0.9711684370257967,
"mean_class_recall": 0.9693685820406379,
"nll": 0.10259254119815524,
"brier": 0.05091487551434137,
"ece15": 0.030811660924900536,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.8679817905918058,
"error_rate_at_confidence_0_9": 0.006993006993006993,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_message_manipulation": {
"questions": 3321,
"accuracy": 0.47756699789220114,
"mean_class_recall": 0.5,
"nll": 1.2951885811608232,
"brier": 0.8663028363232907,
"ece15": 0.42951687019840074,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.7633242999096658,
"error_rate_at_confidence_0_9": 0.504930966469428,
"ordinal_expected_level_mae": null
},
"veremi_nextgen:veremi_supported_attack_type_v16": {
"questions": 3321,
"accuracy": 0.48087925323697683,
"mean_class_recall": 0.12233932942660039,
"nll": 2.41017951541039,
"brier": 0.8961545633575424,
"ece15": 0.3893600279043549,
"temperature": 1.0,
"coverage_at_confidence_0_9": 0.5453176753989762,
"error_rate_at_confidence_0_9": 0.4870237437879624,
"ordinal_expected_level_mae": null
}
},
"macro_task_accuracy": 0.6787694400132313,
"macro_task_balanced_accuracy": 0.671622736838496
}
},
"test_used_for_selection": false
},
"kev_4b_general": {
"decision-v7": {
"raw_clean": {
"n": 1200,
"nll": 0.46776271359373983,
"acc": 0.8708333333333333,
"ece": 0.07324011844924935,
"brier": 0.2068727227531146,
"mean_conf": 0.9391164907697294,
"confident_error_rate": 0.04833333333333333,
"coverage_at_0_9": 0.8166666666666667,
"accuracy_at_0_9": 0.9408163265306122,
"coverage_at_5pct_error": 0.7791666666666667,
"coverage_at_1pct_error": 0.6133333333333333,
"aurc": 0.02837551275061399,
"error_rate_at_0_9": 0.05918367346938776,
"confidence_bias": 0.06828315743639601,
"top_bins": {
"0.9": {
"n": 980,
"errors": 58,
"error_rate": 0.05918367346938776
},
"0.95": {
"n": 915,
"errors": 39,
"error_rate": 0.04262295081967213
},
"0.99": {
"n": 829,
"errors": 21,
"error_rate": 0.025331724969843185
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9916666666666667,
"confidence_cutoff": 0.999598385219589
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.94375,
"confidence_cutoff": 0.9220693757732149
}
},
"score_mae": 0.4635506535819828,
"ranked_probability_score": 0.08574066842436183
},
"as_shipped_clean": {
"n": 1200,
"nll": 0.35957326104404586,
"acc": 0.8708333333333333,
"ece": 0.020190532051177417,
"brier": 0.18914513739887015,
"mean_conf": 0.8656556446393923,
"confident_error_rate": 0.013333333333333334,
"coverage_at_0_9": 0.6583333333333333,
"accuracy_at_0_9": 0.979746835443038,
"coverage_at_5pct_error": 0.7883333333333333,
"coverage_at_1pct_error": 0.585,
"aurc": 0.02903042619721546,
"error_rate_at_0_9": 0.020253164556962026,
"confidence_bias": -0.005177688693941085,
"top_bins": {
"0.9": {
"n": 790,
"errors": 16,
"error_rate": 0.020253164556962026
},
"0.95": {
"n": 697,
"errors": 6,
"error_rate": 0.00860832137733142
},
"0.99": {
"n": 320,
"errors": 2,
"error_rate": 0.00625
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9916666666666667,
"confidence_cutoff": 0.9690438255888087
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.946875,
"confidence_cutoff": 0.7097948924008157
}
},
"score_mae": 0.496047231607107,
"ranked_probability_score": 0.07900573734715889
},
"sources": {
"agnews": {
"n": 240,
"nll": 0.2188339468959484,
"acc": 0.9208333333333333,
"ece": 0.03458717155438394,
"brier": 0.11967225736034369,
"mean_conf": 0.9210081021968138,
"confident_error_rate": 0.0125,
"coverage_at_0_9": 0.7875,
"accuracy_at_0_9": 0.9841269841269841,
"coverage_at_5pct_error": 0.9291666666666667,
"coverage_at_1pct_error": 0.7208333333333333,
"aurc": 0.010595808181650849,
"error_rate_at_0_9": 0.015873015873015872,
"confidence_bias": 0.00017476886348055665,
"top_bins": {
"0.9": {
"n": 189,
"errors": 3,
"error_rate": 0.015873015873015872
},
"0.95": {
"n": 151,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 8,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.962887554254757
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9791666666666666,
"confidence_cutoff": 0.8982351380482543
}
}
},
"amazon": {
"n": 80,
"nll": 1.019521275107054,
"acc": 0.6125,
"ece": 0.06021888255181947,
"brier": 0.5421117796889148,
"mean_conf": 0.5878388840884237,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.0375,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 0.0625,
"coverage_at_1pct_error": 0.0625,
"aurc": 0.26205124362469656,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.02466111591157638,
"top_bins": {
"0.9": {
"n": 3,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 1,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 0,
"errors": 0,
"error_rate": null
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.75,
"confidence_cutoff": 0.5617743497724099
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.640625,
"confidence_cutoff": 0.44733349967779523
}
},
"score_mae": 0.5169737287164902,
"ranked_probability_score": 0.08574565040577714
},
"banking77": {
"n": 80,
"nll": 0.6434976771895786,
"acc": 0.875,
"ece": 0.16139951556317184,
"brier": 0.23632755413887385,
"mean_conf": 0.7261215293796409,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.4,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 0.75,
"coverage_at_1pct_error": 0.675,
"aurc": 0.025355014288770995,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.14887847062035908,
"top_bins": {
"0.9": {
"n": 32,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 24,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 2,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.853940262270331
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.921875,
"confidence_cutoff": 0.4421809473945611
}
}
},
"boolq": {
"n": 80,
"nll": 0.3692803969173892,
"acc": 0.875,
"ece": 0.0748529395052494,
"brier": 0.2143912889240845,
"mean_conf": 0.9094612721753246,
"confident_error_rate": 0.0625,
"coverage_at_0_9": 0.75,
"accuracy_at_0_9": 0.9166666666666666,
"coverage_at_5pct_error": 0.6625,
"coverage_at_1pct_error": 0.075,
"aurc": 0.059468554539862016,
"error_rate_at_0_9": 0.08333333333333333,
"confidence_bias": 0.034461272175324575,
"top_bins": {
"0.9": {
"n": 60,
"errors": 5,
"error_rate": 0.08333333333333333
},
"0.95": {
"n": 50,
"errors": 1,
"error_rate": 0.02
},
"0.99": {
"n": 26,
"errors": 1,
"error_rate": 0.038461538461538464
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.975,
"confidence_cutoff": 0.9805121330496657
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.890625,
"confidence_cutoff": 0.8395732170279804
}
}
},
"contrastive": {
"n": 160,
"nll": 0.06053811563664827,
"acc": 0.96875,
"ece": 0.03645679005941199,
"brier": 0.032513686428614616,
"mean_conf": 0.9645709125518511,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.90625,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.975,
"aurc": 0.0006312206080826334,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.004179087448148877,
"top_bins": {
"0.9": {
"n": 145,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 139,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 83,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9906152293058859
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9714574090050602
}
}
},
"dbpedia14": {
"n": 80,
"nll": 0.031074258317027002,
"acc": 1.0,
"ece": 0.02597689933514287,
"brier": 0.010683464293698132,
"mean_conf": 0.9740231006648571,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.95,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 1.0,
"aurc": 0.0,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.02597689933514291,
"top_bins": {
"0.9": {
"n": 76,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 75,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 57,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9936484074458076
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9825304423953378
}
}
},
"imdb": {
"n": 80,
"nll": 0.0720718883732508,
"acc": 0.95,
"ece": 0.044527070446157675,
"brier": 0.0446787843047757,
"mean_conf": 0.9624677353009415,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.9,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.9375,
"aurc": 0.0017470077113648063,
"error_rate_at_0_9": 0.0,
"confidence_bias": 0.012467735300941585,
"top_bins": {
"0.9": {
"n": 72,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 70,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 50,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9920587271734113
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.979826985587298
}
}
},
"mnli": {
"n": 80,
"nll": 0.27465339009465656,
"acc": 0.9125,
"ece": 0.09412857766900207,
"brier": 0.15288319760619568,
"mean_conf": 0.94639379824153,
"confident_error_rate": 0.0375,
"coverage_at_0_9": 0.875,
"accuracy_at_0_9": 0.9571428571428572,
"coverage_at_5pct_error": 0.875,
"coverage_at_1pct_error": 0.675,
"aurc": 0.01561228714610165,
"error_rate_at_0_9": 0.04285714285714286,
"confidence_bias": 0.03389379824153005,
"top_bins": {
"0.9": {
"n": 70,
"errors": 3,
"error_rate": 0.04285714285714286
},
"0.95": {
"n": 67,
"errors": 2,
"error_rate": 0.029850746268656716
},
"0.99": {
"n": 45,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9920616734287709
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.96875,
"confidence_cutoff": 0.9661281848866967
}
}
},
"sst5": {
"n": 80,
"nll": 0.9636808958022751,
"acc": 0.6125,
"ece": 0.09386529578800333,
"brier": 0.5310525330540982,
"mean_conf": 0.6617158029994963,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.0,
"accuracy_at_0_9": null,
"coverage_at_5pct_error": 0.1625,
"coverage_at_1pct_error": 0.1625,
"aurc": 0.26405329260422244,
"error_rate_at_0_9": null,
"confidence_bias": 0.04921580299949624,
"top_bins": {
"0.9": {
"n": 0,
"errors": 0,
"error_rate": null
},
"0.95": {
"n": 0,
"errors": 0,
"error_rate": null
},
"0.99": {
"n": 0,
"errors": 0,
"error_rate": null
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.65,
"confidence_cutoff": 0.6682213711318762
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.625,
"confidence_cutoff": 0.5940210248093071
}
},
"score_mae": 0.5227774115005795,
"ranked_probability_score": 0.08371156976927716
},
"trec": {
"n": 80,
"nll": 0.16488315000308515,
"acc": 0.9625,
"ece": 0.02170728565832692,
"brier": 0.06057653876714909,
"mean_conf": 0.9707615704179906,
"confident_error_rate": 0.025,
"coverage_at_0_9": 0.9625,
"accuracy_at_0_9": 0.974025974025974,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 0.2625,
"aurc": 0.018315788693627898,
"error_rate_at_0_9": 0.025974025974025976,
"confidence_bias": 0.008261570417990582,
"top_bins": {
"0.9": {
"n": 77,
"errors": 2,
"error_rate": 0.025974025974025976
},
"0.95": {
"n": 73,
"errors": 2,
"error_rate": 0.0273972602739726
},
"0.99": {
"n": 47,
"errors": 1,
"error_rate": 0.02127659574468085
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.975,
"confidence_cutoff": 0.990957410950485
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.984375,
"confidence_cutoff": 0.9755548475921001
}
}
},
"yelp": {
"n": 160,
"nll": 0.5386789559476151,
"acc": 0.78125,
"ece": 0.06183692449171378,
"brier": 0.3102138876335011,
"mean_conf": 0.7769424223142675,
"confident_error_rate": 0.01875,
"coverage_at_0_9": 0.4125,
"accuracy_at_0_9": 0.9545454545454546,
"coverage_at_5pct_error": 0.4375,
"coverage_at_1pct_error": 0.14375,
"aurc": 0.0856358031592544,
"error_rate_at_0_9": 0.045454545454545456,
"confidence_bias": -0.004307577685732533,
"top_bins": {
"0.9": {
"n": 66,
"errors": 3,
"error_rate": 0.045454545454545456
},
"0.95": {
"n": 47,
"errors": 1,
"error_rate": 0.02127659574468085
},
"0.99": {
"n": 2,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9125,
"confidence_cutoff": 0.8216206137745724
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.8515625,
"confidence_cutoff": 0.577455875568763
}
},
"score_mae": 0.44839055460425137,
"ranked_probability_score": 0.0675599918664224
}
},
"types": {
"choice": {
"n": 480,
"nll": 0.2482065801683337,
"acc": 0.9375,
"ece": 0.03877078391929752,
"brier": 0.10960480604725589,
"mean_conf": 0.9179948652716078,
"confident_error_rate": 0.0125,
"coverage_at_0_9": 0.8145833333333333,
"accuracy_at_0_9": 0.9846547314578005,
"coverage_at_5pct_error": 0.9583333333333334,
"coverage_at_1pct_error": 0.69375,
"aurc": 0.01188927247802621,
"error_rate_at_0_9": 0.015345268542199489,
"confidence_bias": -0.019505134728392215,
"top_bins": {
"0.9": {
"n": 391,
"errors": 6,
"error_rate": 0.015345268542199489
},
"0.95": {
"n": 361,
"errors": 4,
"error_rate": 0.0110803324099723
},
"0.99": {
"n": 202,
"errors": 1,
"error_rate": 0.0049504950495049506
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9958333333333333,
"confidence_cutoff": 0.9867396184306336
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9869791666666666,
"confidence_cutoff": 0.9197849614177179
}
}
},
"noul": {
"n": 480,
"nll": 0.18011051231286856,
"acc": 0.9270833333333334,
"ece": 0.014056851970226658,
"brier": 0.10328956268313709,
"mean_conf": 0.9306022039720875,
"confident_error_rate": 0.01875,
"coverage_at_0_9": 0.8166666666666667,
"accuracy_at_0_9": 0.9770408163265306,
"coverage_at_5pct_error": 0.9416666666666667,
"coverage_at_1pct_error": 0.7041666666666667,
"aurc": 0.015597964078219255,
"error_rate_at_0_9": 0.02295918367346939,
"confidence_bias": 0.0035188706387541124,
"top_bins": {
"0.9": {
"n": 392,
"errors": 9,
"error_rate": 0.02295918367346939
},
"0.95": {
"n": 334,
"errors": 2,
"error_rate": 0.005988023952095809
},
"0.99": {
"n": 118,
"errors": 1,
"error_rate": 0.00847457627118644
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9916666666666667,
"confidence_cutoff": 0.9755983279973393
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9765625,
"confidence_cutoff": 0.9172471488342538
}
}
},
"score": {
"n": 240,
"nll": 0.9412321202578248,
"acc": 0.625,
"ece": 0.048867889504322116,
"brier": 0.519936949533565,
"mean_conf": 0.6310840847095708,
"confident_error_rate": 0.004166666666666667,
"coverage_at_0_9": 0.029166666666666667,
"accuracy_at_0_9": 0.8571428571428571,
"coverage_at_5pct_error": 0.016666666666666666,
"coverage_at_1pct_error": 0.016666666666666666,
"aurc": 0.2774440601733873,
"error_rate_at_0_9": 0.14285714285714285,
"confidence_bias": 0.006084084709570781,
"top_bins": {
"0.9": {
"n": 7,
"errors": 1,
"error_rate": 0.14285714285714285
},
"0.95": {
"n": 2,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 0,
"errors": 0,
"error_rate": null
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.7166666666666667,
"confidence_cutoff": 0.6329725841576003
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.6614583333333334,
"confidence_cutoff": 0.4922131178494099
}
},
"score_mae": 0.496047231607107,
"ranked_probability_score": 0.07900573734715889
}
}
},
"transfer-v4": {
"raw_clean": {
"n": 656,
"nll": 0.5640777093137941,
"acc": 0.836890243902439,
"ece": 0.10195634226245451,
"brier": 0.25517441623108567,
"mean_conf": 0.9301396868913653,
"confident_error_rate": 0.06707317073170732,
"coverage_at_0_9": 0.8094512195121951,
"accuracy_at_0_9": 0.9171374764595104,
"coverage_at_5pct_error": 0.6829268292682927,
"coverage_at_1pct_error": 0.2820121951219512,
"aurc": 0.043201879169005745,
"error_rate_at_0_9": 0.08286252354048965,
"confidence_bias": 0.09324944298892623,
"top_bins": {
"0.9": {
"n": 531,
"errors": 44,
"error_rate": 0.08286252354048965
},
"0.95": {
"n": 487,
"errors": 33,
"error_rate": 0.06776180698151951
},
"0.99": {
"n": 406,
"errors": 14,
"error_rate": 0.034482758620689655
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9725609756097561,
"confidence_cutoff": 0.9980492531927935
},
"0.8": {
"coverage": 0.8003048780487805,
"accuracy": 0.92,
"confidence_cutoff": 0.9074193401309704
}
},
"score_mae": 0.24004523406424622,
"ranked_probability_score": 0.09767451376793074
},
"as_shipped_clean": {
"n": 656,
"nll": 0.42587070132965826,
"acc": 0.836890243902439,
"ece": 0.0314949650736429,
"brier": 0.23187082308763576,
"mean_conf": 0.844962353152835,
"confident_error_rate": 0.018292682926829267,
"coverage_at_0_9": 0.5853658536585366,
"accuracy_at_0_9": 0.96875,
"coverage_at_5pct_error": 0.6570121951219512,
"coverage_at_1pct_error": 0.2530487804878049,
"aurc": 0.04507835320830427,
"error_rate_at_0_9": 0.03125,
"confidence_bias": 0.00807210925039592,
"top_bins": {
"0.9": {
"n": 384,
"errors": 12,
"error_rate": 0.03125
},
"0.95": {
"n": 315,
"errors": 8,
"error_rate": 0.025396825396825397
},
"0.99": {
"n": 79,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9725609756097561,
"confidence_cutoff": 0.9441193658749547
},
"0.8": {
"coverage": 0.8003048780487805,
"accuracy": 0.9047619047619048,
"confidence_cutoff": 0.693693819569072
}
},
"score_mae": 0.27196301946029916,
"ranked_probability_score": 0.0837200362891968
},
"sources": {
"composition_holdout": {
"n": 96,
"nll": 0.29615290237046005,
"acc": 0.875,
"ece": 0.06938096946882913,
"brier": 0.1808256349283982,
"mean_conf": 0.9443809694688293,
"confident_error_rate": 0.041666666666666664,
"coverage_at_0_9": 0.8333333333333334,
"accuracy_at_0_9": 0.95,
"coverage_at_5pct_error": 0.8645833333333334,
"coverage_at_1pct_error": 0.6666666666666666,
"aurc": 0.021756148279610994,
"error_rate_at_0_9": 0.05,
"confidence_bias": 0.06938096946882932,
"top_bins": {
"0.9": {
"n": 80,
"errors": 4,
"error_rate": 0.05
},
"0.95": {
"n": 73,
"errors": 3,
"error_rate": 0.0410958904109589
},
"0.99": {
"n": 2,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.980048039700624
},
"0.8": {
"coverage": 0.8020833333333334,
"accuracy": 0.948051948051948,
"confidence_cutoff": 0.9379020298350527
}
}
},
"contrastive": {
"n": 80,
"nll": 0.2412487469898486,
"acc": 0.875,
"ece": 0.054181213136020756,
"brier": 0.15115638127662953,
"mean_conf": 0.9258459913436543,
"confident_error_rate": 0.0125,
"coverage_at_0_9": 0.8125,
"accuracy_at_0_9": 0.9846153846153847,
"coverage_at_5pct_error": 0.875,
"coverage_at_1pct_error": 0.7875,
"aurc": 0.01404271679837261,
"error_rate_at_0_9": 0.015384615384615385,
"confidence_bias": 0.050845991343654307,
"top_bins": {
"0.9": {
"n": 65,
"errors": 1,
"error_rate": 0.015384615384615385
},
"0.95": {
"n": 62,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 26,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9886860005278774
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.984375,
"confidence_cutoff": 0.9353245409640992
}
},
"score_mae": 0.27196301946029916,
"ranked_probability_score": 0.0837200362891968
},
"emotion": {
"n": 80,
"nll": 1.0001023668251066,
"acc": 0.6625,
"ece": 0.14881697863795876,
"brier": 0.4744803648425345,
"mean_conf": 0.591235292513326,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.05,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 0.0625,
"coverage_at_1pct_error": 0.0625,
"aurc": 0.18702686566035498,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.07126470748667402,
"top_bins": {
"0.9": {
"n": 4,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 2,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 0,
"errors": 0,
"error_rate": null
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.8,
"confidence_cutoff": 0.5874489256154185
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.71875,
"confidence_cutoff": 0.3855885748389788
}
}
},
"mmlu": {
"n": 80,
"nll": 0.777539674387988,
"acc": 0.7,
"ece": 0.13070084664599038,
"brier": 0.40568344535429135,
"mean_conf": 0.6636013764250539,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.175,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 0.2625,
"coverage_at_1pct_error": 0.225,
"aurc": 0.12322939721812473,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.036398623574946054,
"top_bins": {
"0.9": {
"n": 14,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 7,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 1,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9,
"confidence_cutoff": 0.6761349246582344
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.796875,
"confidence_cutoff": 0.4466846683613359
}
}
},
"paws": {
"n": 80,
"nll": 0.512348189940992,
"acc": 0.8125,
"ece": 0.10522416975834781,
"brier": 0.2914852770101592,
"mean_conf": 0.9091572173256038,
"confident_error_rate": 0.0875,
"coverage_at_0_9": 0.75,
"accuracy_at_0_9": 0.8833333333333333,
"coverage_at_5pct_error": 0.2625,
"coverage_at_1pct_error": 0.2,
"aurc": 0.09484078957534384,
"error_rate_at_0_9": 0.11666666666666667,
"confidence_bias": 0.09665721732560384,
"top_bins": {
"0.9": {
"n": 60,
"errors": 7,
"error_rate": 0.11666666666666667
},
"0.95": {
"n": 51,
"errors": 5,
"error_rate": 0.09803921568627451
},
"0.99": {
"n": 4,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.875,
"confidence_cutoff": 0.974944024208843
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.875,
"confidence_cutoff": 0.8613523629912301
}
}
},
"qnli": {
"n": 80,
"nll": 0.20826182445785202,
"acc": 0.9125,
"ece": 0.04839798205004245,
"brier": 0.12797478277700258,
"mean_conf": 0.9090029243294062,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.7,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 0.8875,
"coverage_at_1pct_error": 0.7,
"aurc": 0.014269660330833212,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.0034970756705937633,
"top_bins": {
"0.9": {
"n": 56,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 45,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 10,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9595371830584098
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.96875,
"confidence_cutoff": 0.8535858138327058
}
}
},
"sciq": {
"n": 80,
"nll": 0.04197126388379959,
"acc": 1.0,
"ece": 0.03774145345553605,
"brier": 0.010730106398852924,
"mean_conf": 0.962258546544464,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.9,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 1.0,
"coverage_at_1pct_error": 1.0,
"aurc": 0.0,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.037741453455536034,
"top_bins": {
"0.9": {
"n": 72,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 69,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 36,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9883715126062644
},
"0.8": {
"coverage": 0.8,
"accuracy": 1.0,
"confidence_cutoff": 0.9713861020562242
}
}
},
"tweet_offensive": {
"n": 80,
"nll": 0.3552842015730585,
"acc": 0.85,
"ece": 0.056220131039951354,
"brier": 0.22283962974506544,
"mean_conf": 0.8343327840091443,
"confident_error_rate": 0.0,
"coverage_at_0_9": 0.4125,
"accuracy_at_0_9": 1.0,
"coverage_at_5pct_error": 0.6,
"coverage_at_1pct_error": 0.525,
"aurc": 0.045801042430234855,
"error_rate_at_0_9": 0.0,
"confidence_bias": -0.015667215990855632,
"top_bins": {
"0.9": {
"n": 33,
"errors": 0,
"error_rate": 0.0
},
"0.95": {
"n": 6,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 0,
"errors": 0,
"error_rate": null
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.8768513333328267
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.890625,
"confidence_cutoff": 0.7493184208869934
}
}
}
},
"types": {
"choice": {
"n": 336,
"nll": 0.5178563780812967,
"acc": 0.8125,
"ece": 0.04292515250441144,
"brier": 0.26378206631184675,
"mean_conf": 0.797702947344152,
"confident_error_rate": 0.011904761904761904,
"coverage_at_0_9": 0.5059523809523809,
"accuracy_at_0_9": 0.9764705882352941,
"coverage_at_5pct_error": 0.6130952380952381,
"coverage_at_1pct_error": 0.40773809523809523,
"aurc": 0.04890439545080765,
"error_rate_at_0_9": 0.023529411764705882,
"confidence_bias": -0.014797052655847964,
"top_bins": {
"0.9": {
"n": 170,
"errors": 4,
"error_rate": 0.023529411764705882
},
"0.95": {
"n": 151,
"errors": 3,
"error_rate": 0.019867549668874173
},
"0.99": {
"n": 39,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9761904761904762,
"confidence_cutoff": 0.9209057199846711
},
"0.8": {
"coverage": 0.8005952380952381,
"accuracy": 0.8996282527881041,
"confidence_cutoff": 0.5765734433053208
}
}
},
"noul": {
"n": 280,
"nll": 0.3088148767031123,
"acc": 0.8785714285714286,
"ece": 0.02716426081477808,
"brier": 0.18354251467515595,
"mean_conf": 0.8993028632705061,
"confident_error_rate": 0.025,
"coverage_at_0_9": 0.675,
"accuracy_at_0_9": 0.9629629629629629,
"coverage_at_5pct_error": 0.7428571428571429,
"coverage_at_1pct_error": 0.24642857142857144,
"aurc": 0.04037503899248962,
"error_rate_at_0_9": 0.037037037037037035,
"confidence_bias": 0.020731434699077567,
"top_bins": {
"0.9": {
"n": 189,
"errors": 7,
"error_rate": 0.037037037037037035
},
"0.95": {
"n": 142,
"errors": 5,
"error_rate": 0.035211267605633804
},
"0.99": {
"n": 35,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 0.9642857142857143,
"confidence_cutoff": 0.951430429315972
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.9330357142857143,
"confidence_cutoff": 0.8169286372799969
}
}
},
"score": {
"n": 40,
"nll": 0.47258178900171605,
"acc": 0.75,
"ece": 0.11341449184915596,
"brier": 0.3021145388916218,
"mean_conf": 0.8615577911220736,
"confident_error_rate": 0.025,
"coverage_at_0_9": 0.625,
"accuracy_at_0_9": 0.96,
"coverage_at_5pct_error": 0.65,
"coverage_at_1pct_error": 0.575,
"aurc": 0.060906503666048904,
"error_rate_at_0_9": 0.04,
"confidence_bias": 0.11155779112207365,
"top_bins": {
"0.9": {
"n": 25,
"errors": 1,
"error_rate": 0.04
},
"0.95": {
"n": 22,
"errors": 0,
"error_rate": 0.0
},
"0.99": {
"n": 5,
"errors": 0,
"error_rate": 0.0
}
},
"selective": {
"0.5": {
"coverage": 0.5,
"accuracy": 1.0,
"confidence_cutoff": 0.9618942423403596
},
"0.8": {
"coverage": 0.8,
"accuracy": 0.84375,
"confidence_cutoff": 0.7283845543861389
}
},
"score_mae": 0.27196301946029916,
"ranked_probability_score": 0.0837200362891968
}
}
}
},
"base_revision": "1001bb4d826a52d1f399e183466143f4da7b741b",
"test_used_for_selection": false
}