File size: 16,463 Bytes
f779a7e |
1 |
{"./data_dir/eval_vidore/arxivqa_test_subsampled": {"ndcg_at_1": 0.648, "ndcg_at_3": 0.71821, "ndcg_at_5": 0.73147, "ndcg_at_10": 0.7444, "ndcg_at_20": 0.75657, "ndcg_at_50": 0.76882, "ndcg_at_100": 0.77437, "map_at_1": 0.648, "map_at_3": 0.701, "map_at_5": 0.7084, "map_at_10": 0.71373, "map_at_20": 0.71708, "map_at_50": 0.71904, "map_at_100": 0.71954, "recall_at_1": 0.648, "recall_at_3": 0.768, "recall_at_5": 0.8, "recall_at_10": 0.84, "recall_at_20": 0.888, "recall_at_50": 0.95, "recall_at_100": 0.984, "precision_at_1": 0.648, "precision_at_3": 0.256, "precision_at_5": 0.16, "precision_at_10": 0.084, "precision_at_20": 0.0444, "precision_at_50": 0.019, "precision_at_100": 0.00984, "mrr_at_1": 0.652, "mrr_at_3": 0.703, "mrr_at_5": 0.7096, "mrr_at_10": 0.7163452380952381, "mrr_at_20": 0.7195925928990171, "mrr_at_50": 0.7213793558075785, "mrr_at_100": 0.7218459925368063, "naucs_at_1_max": 0.5320415548033148, "naucs_at_1_std": -0.09459100164714764, "naucs_at_1_diff1": 0.9007496465766573, "naucs_at_3_max": 0.5235472456346243, "naucs_at_3_std": -0.09979906824566964, "naucs_at_3_diff1": 0.8154895266545747, "naucs_at_5_max": 0.5494205794205791, "naucs_at_5_std": -0.07948051948052083, "naucs_at_5_diff1": 0.7935714285714291, "naucs_at_10_max": 0.6103399607265577, "naucs_at_10_std": -0.012493863524791362, "naucs_at_10_diff1": 0.7950233186057926, "naucs_at_20_max": 0.6541277732683969, "naucs_at_20_std": 0.12420522186147001, "naucs_at_20_diff1": 0.8013984713203455, "naucs_at_50_max": 0.7840896358543438, "naucs_at_50_std": 0.21335200746965483, "naucs_at_50_diff1": 0.8345471521942116, "naucs_at_100_max": 0.8932072829131755, "naucs_at_100_std": 0.2987861811391213, "naucs_at_100_diff1": 0.9183006535947711}, "./data_dir/eval_vidore/docvqa_test_subsampled": {"ndcg_at_1": 0.46785, "ndcg_at_3": 0.53315, "ndcg_at_5": 0.56073, "ndcg_at_10": 0.58063, "ndcg_at_20": 0.59491, "ndcg_at_50": 0.60895, "ndcg_at_100": 0.61612, "map_at_1": 0.46785, "map_at_3": 0.51737, "map_at_5": 0.53278, "map_at_10": 0.5413, "map_at_20": 0.54536, "map_at_50": 0.5476, "map_at_100": 0.54822, "recall_at_1": 0.46785, "recall_at_3": 0.57871, "recall_at_5": 0.64523, "recall_at_10": 0.7051, "recall_at_20": 0.76053, "recall_at_50": 0.83149, "recall_at_100": 0.87583, "precision_at_1": 0.46785, "precision_at_3": 0.1929, "precision_at_5": 0.12905, "precision_at_10": 0.07051, "precision_at_20": 0.03803, "precision_at_50": 0.01663, "precision_at_100": 0.00876, "mrr_at_1": 0.4722838137472284, "mrr_at_3": 0.5203252032520325, "mrr_at_5": 0.5344050258684405, "mrr_at_10": 0.54401418364833, "mrr_at_20": 0.547432357060943, "mrr_at_50": 0.5496278114926427, "mrr_at_100": 0.5502293224133955, "naucs_at_1_max": 0.22191473813157103, "naucs_at_1_std": 0.43792922230870507, "naucs_at_1_diff1": 0.846614423933555, "naucs_at_3_max": 0.14455270946312818, "naucs_at_3_std": 0.5599978573011983, "naucs_at_3_diff1": 0.7812484213095876, "naucs_at_5_max": 0.07904307020932648, "naucs_at_5_std": 0.5873613578608561, "naucs_at_5_diff1": 0.7507308300016783, "naucs_at_10_max": -0.00569065025671971, "naucs_at_10_std": 0.6086341964529604, "naucs_at_10_diff1": 0.7216978702993005, "naucs_at_20_max": -0.08386122115622482, "naucs_at_20_std": 0.6774737565173128, "naucs_at_20_diff1": 0.6877353492008539, "naucs_at_50_max": -0.22730282183008066, "naucs_at_50_std": 0.7387768986158063, "naucs_at_50_diff1": 0.6755005771418886, "naucs_at_100_max": -0.3500275419596926, "naucs_at_100_std": 0.8080829651530356, "naucs_at_100_diff1": 0.6506258023643423}, "./data_dir/eval_vidore/syntheticDocQA_energy_test": {"ndcg_at_1": 0.89, "ndcg_at_3": 0.93286, "ndcg_at_5": 0.94103, "ndcg_at_10": 0.94419, "ndcg_at_20": 0.94419, "ndcg_at_50": 0.94609, "ndcg_at_100": 0.94609, "map_at_1": 0.89, "map_at_3": 0.92333, "map_at_5": 0.92783, "map_at_10": 0.92908, "map_at_20": 0.92908, "map_at_50": 0.92935, "map_at_100": 0.92935, "recall_at_1": 0.89, "recall_at_3": 0.96, "recall_at_5": 0.98, "recall_at_10": 0.99, "recall_at_20": 0.99, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.89, "precision_at_3": 0.32, "precision_at_5": 0.196, "precision_at_10": 0.099, "precision_at_20": 0.0495, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.91, "mrr_at_3": 0.9333333333333332, "mrr_at_5": 0.9378333333333333, "mrr_at_10": 0.9395, "mrr_at_20": 0.9395, "mrr_at_50": 0.9397777777777777, "mrr_at_100": 0.9397777777777777, "naucs_at_1_max": 0.36255476333648357, "naucs_at_1_std": -0.16244308908169428, "naucs_at_1_diff1": 0.9488875526157551, "naucs_at_3_max": 0.6091269841269833, "naucs_at_3_std": 0.12313258636788006, "naucs_at_3_diff1": 0.9305555555555564, "naucs_at_5_max": 0.7957516339869297, "naucs_at_5_std": 0.09337068160597826, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 0.8692810457516413, "naucs_at_10_std": -0.1713352007469681, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 0.8692810457516413, "naucs_at_20_std": -0.1713352007469681, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": NaN, "naucs_at_50_std": NaN, "naucs_at_50_diff1": NaN, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}, "./data_dir/eval_vidore/tatdqa_test": {"ndcg_at_1": 0.62333, "ndcg_at_3": 0.72622, "ndcg_at_5": 0.75278, "ndcg_at_10": 0.774, "ndcg_at_20": 0.78172, "ndcg_at_50": 0.78651, "ndcg_at_100": 0.78807, "map_at_1": 0.62333, "map_at_3": 0.70089, "map_at_5": 0.71565, "map_at_10": 0.72464, "map_at_20": 0.72678, "map_at_50": 0.72758, "map_at_100": 0.72772, "recall_at_1": 0.62333, "recall_at_3": 0.79951, "recall_at_5": 0.86391, "recall_at_10": 0.92831, "recall_at_20": 0.95869, "recall_at_50": 0.98238, "recall_at_100": 0.9921, "precision_at_1": 0.62333, "precision_at_3": 0.2665, "precision_at_5": 0.17278, "precision_at_10": 0.09283, "precision_at_20": 0.04793, "precision_at_50": 0.01965, "precision_at_100": 0.00992, "mrr_at_1": 0.6245443499392467, "mrr_at_3": 0.7006885378695827, "mrr_at_5": 0.7164540299716484, "mrr_at_10": 0.7250101255569056, "mrr_at_20": 0.7271291712945228, "mrr_at_50": 0.7279352643901574, "mrr_at_100": 0.728074338884466, "naucs_at_1_max": 0.18350079884403742, "naucs_at_1_std": -0.24504541716152195, "naucs_at_1_diff1": 0.788310462608764, "naucs_at_3_max": 0.2174478161548601, "naucs_at_3_std": -0.20875995619294158, "naucs_at_3_diff1": 0.6747002810297338, "naucs_at_5_max": 0.20232749356597957, "naucs_at_5_std": -0.2066338463888926, "naucs_at_5_diff1": 0.6660532813247335, "naucs_at_10_max": 0.3515538559874108, "naucs_at_10_std": 0.028956552272515996, "naucs_at_10_diff1": 0.6143796841665605, "naucs_at_20_max": 0.4075526493649992, "naucs_at_20_std": 0.30084166201853924, "naucs_at_20_diff1": 0.569699607771098, "naucs_at_50_max": 0.5357294959420121, "naucs_at_50_std": 0.5086348340492322, "naucs_at_50_diff1": 0.5257337269588929, "naucs_at_100_max": 0.5784455619019262, "naucs_at_100_std": 0.5499580299143445, "naucs_at_100_diff1": 0.6538225911609866}, "./data_dir/eval_vidore/infovqa_test_subsampled": {"ndcg_at_1": 0.76518, "ndcg_at_3": 0.82176, "ndcg_at_5": 0.83004, "ndcg_at_10": 0.84013, "ndcg_at_20": 0.84547, "ndcg_at_50": 0.85151, "ndcg_at_100": 0.85344, "map_at_1": 0.76518, "map_at_3": 0.80837, "map_at_5": 0.81292, "map_at_10": 0.81725, "map_at_20": 0.81883, "map_at_50": 0.81982, "map_at_100": 0.81997, "recall_at_1": 0.76518, "recall_at_3": 0.86032, "recall_at_5": 0.88057, "recall_at_10": 0.91093, "recall_at_20": 0.93117, "recall_at_50": 0.96154, "recall_at_100": 0.97368, "precision_at_1": 0.76518, "precision_at_3": 0.28677, "precision_at_5": 0.17611, "precision_at_10": 0.09109, "precision_at_20": 0.04656, "precision_at_50": 0.01923, "precision_at_100": 0.00974, "mrr_at_1": 0.7692307692307693, "mrr_at_3": 0.8120782726045884, "mrr_at_5": 0.8161268556005399, "mrr_at_10": 0.8204557869031553, "mrr_at_20": 0.8221355547683245, "mrr_at_50": 0.822944584158538, "mrr_at_100": 0.8231490232346909, "naucs_at_1_max": 0.5149096264576689, "naucs_at_1_std": 0.028876109416755995, "naucs_at_1_diff1": 0.8423502810401294, "naucs_at_3_max": 0.5357633382771757, "naucs_at_3_std": 0.006279786216531792, "naucs_at_3_diff1": 0.7627339364022564, "naucs_at_5_max": 0.5194080180551114, "naucs_at_5_std": 0.016244825987351323, "naucs_at_5_diff1": 0.7459578410036671, "naucs_at_10_max": 0.5866947683147792, "naucs_at_10_std": 0.24003115026037278, "naucs_at_10_diff1": 0.7051949282237744, "naucs_at_20_max": 0.5682575132928513, "naucs_at_20_std": 0.2770052352671666, "naucs_at_20_diff1": 0.6887081088160373, "naucs_at_50_max": 0.7594010289684164, "naucs_at_50_std": 0.7008338816869777, "naucs_at_50_diff1": 0.658391093237802, "naucs_at_100_max": 0.8815298509718311, "naucs_at_100_std": 0.8291345987814791, "naucs_at_100_diff1": 0.5748643414505342}, "./data_dir/eval_vidore/syntheticDocQA_healthcare_industry_test": {"ndcg_at_1": 0.9, "ndcg_at_3": 0.96178, "ndcg_at_5": 0.96178, "ndcg_at_10": 0.96178, "ndcg_at_20": 0.96178, "ndcg_at_50": 0.96178, "ndcg_at_100": 0.96178, "map_at_1": 0.9, "map_at_3": 0.94833, "map_at_5": 0.94833, "map_at_10": 0.94833, "map_at_20": 0.94833, "map_at_50": 0.94833, "map_at_100": 0.94833, "recall_at_1": 0.9, "recall_at_3": 1.0, "recall_at_5": 1.0, "recall_at_10": 1.0, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.9, "precision_at_3": 0.33333, "precision_at_5": 0.2, "precision_at_10": 0.1, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.9, "mrr_at_3": 0.9483333333333333, "mrr_at_5": 0.9483333333333333, "mrr_at_10": 0.9483333333333333, "mrr_at_20": 0.9483333333333333, "mrr_at_50": 0.9483333333333333, "mrr_at_100": 0.9483333333333333, "naucs_at_1_max": 0.38183940242763664, "naucs_at_1_std": -0.3154995331465938, "naucs_at_1_diff1": 0.9607843137254906, "naucs_at_3_max": 1.0, "naucs_at_3_std": 1.0, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 1.0, "naucs_at_5_std": 1.0, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 1.0, "naucs_at_10_std": 1.0, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 1.0, "naucs_at_20_std": 1.0, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": NaN, "naucs_at_50_std": NaN, "naucs_at_50_diff1": NaN, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}, "./data_dir/eval_vidore/tabfquad_test_subsampled": {"ndcg_at_1": 0.67143, "ndcg_at_3": 0.72355, "ndcg_at_5": 0.73831, "ndcg_at_10": 0.76065, "ndcg_at_20": 0.77133, "ndcg_at_50": 0.78726, "ndcg_at_100": 0.78967, "map_at_1": 0.67143, "map_at_3": 0.71071, "map_at_5": 0.71893, "map_at_10": 0.72838, "map_at_20": 0.73124, "map_at_50": 0.73393, "map_at_100": 0.73417, "recall_at_1": 0.67143, "recall_at_3": 0.76071, "recall_at_5": 0.79643, "recall_at_10": 0.86429, "recall_at_20": 0.90714, "recall_at_50": 0.98571, "recall_at_100": 1.0, "precision_at_1": 0.67143, "precision_at_3": 0.25357, "precision_at_5": 0.15929, "precision_at_10": 0.08643, "precision_at_20": 0.04536, "precision_at_50": 0.01971, "precision_at_100": 0.01, "mrr_at_1": 0.6714285714285714, "mrr_at_3": 0.7107142857142857, "mrr_at_5": 0.71875, "mrr_at_10": 0.7274503968253968, "mrr_at_20": 0.7311492984354826, "mrr_at_50": 0.7336928024479792, "mrr_at_100": 0.7339324686957099, "naucs_at_1_max": 0.2918131988067234, "naucs_at_1_std": 0.10674613761490172, "naucs_at_1_diff1": 0.7514915961291264, "naucs_at_3_max": 0.2682433845224536, "naucs_at_3_std": 0.08520936427913174, "naucs_at_3_diff1": 0.6522779546035362, "naucs_at_5_max": 0.23648809157215928, "naucs_at_5_std": 0.11365575977265355, "naucs_at_5_diff1": 0.6022576302800486, "naucs_at_10_max": 0.1419792355845087, "naucs_at_10_std": 0.04070181688635578, "naucs_at_10_diff1": 0.6177159142958931, "naucs_at_20_max": 0.15348703583997816, "naucs_at_20_std": -0.004776269482152739, "naucs_at_20_diff1": 0.5820045967104782, "naucs_at_50_max": 0.8558590102707799, "naucs_at_50_std": 0.6790382819794502, "naucs_at_50_diff1": 0.8231792717086844, "naucs_at_100_max": 1.0, "naucs_at_100_std": 1.0, "naucs_at_100_diff1": 1.0}, "./data_dir/eval_vidore/syntheticDocQA_government_reports_test": {"ndcg_at_1": 0.87, "ndcg_at_3": 0.94047, "ndcg_at_5": 0.94478, "ndcg_at_10": 0.94478, "ndcg_at_20": 0.94478, "ndcg_at_50": 0.94478, "ndcg_at_100": 0.94478, "map_at_1": 0.87, "map_at_3": 0.92333, "map_at_5": 0.92583, "map_at_10": 0.92583, "map_at_20": 0.92583, "map_at_50": 0.92583, "map_at_100": 0.92583, "recall_at_1": 0.87, "recall_at_3": 0.99, "recall_at_5": 1.0, "recall_at_10": 1.0, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.87, "precision_at_3": 0.33, "precision_at_5": 0.2, "precision_at_10": 0.1, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.88, "mrr_at_3": 0.93, "mrr_at_5": 0.9325, "mrr_at_10": 0.9325, "mrr_at_20": 0.9325, "mrr_at_50": 0.9325, "mrr_at_100": 0.9325, "naucs_at_1_max": 0.16200103680663608, "naucs_at_1_std": -0.2467229504554533, "naucs_at_1_diff1": 0.8407761238243359, "naucs_at_3_max": 1.0, "naucs_at_3_std": 1.0, "naucs_at_3_diff1": 0.35807656395889226, "naucs_at_5_max": 1.0, "naucs_at_5_std": 1.0, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 1.0, "naucs_at_10_std": 1.0, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 1.0, "naucs_at_20_std": 1.0, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": NaN, "naucs_at_50_std": NaN, "naucs_at_50_diff1": NaN, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}, "./data_dir/eval_vidore/shiftproject_test": {"ndcg_at_1": 0.54, "ndcg_at_3": 0.64309, "ndcg_at_5": 0.69258, "ndcg_at_10": 0.71262, "ndcg_at_20": 0.72955, "ndcg_at_50": 0.73167, "ndcg_at_100": 0.73337, "map_at_1": 0.54, "map_at_3": 0.61667, "map_at_5": 0.64417, "map_at_10": 0.65282, "map_at_20": 0.65707, "map_at_50": 0.65747, "map_at_100": 0.65764, "recall_at_1": 0.54, "recall_at_3": 0.72, "recall_at_5": 0.84, "recall_at_10": 0.9, "recall_at_20": 0.97, "recall_at_50": 0.98, "recall_at_100": 0.99, "precision_at_1": 0.54, "precision_at_3": 0.24, "precision_at_5": 0.168, "precision_at_10": 0.09, "precision_at_20": 0.0485, "precision_at_50": 0.0196, "precision_at_100": 0.0099, "mrr_at_1": 0.54, "mrr_at_3": 0.6233333333333334, "mrr_at_5": 0.6463333333333334, "mrr_at_10": 0.655984126984127, "mrr_at_20": 0.659218253968254, "mrr_at_50": 0.66014898989899, "mrr_at_100": 0.6603275613275613, "naucs_at_1_max": 0.06050123642021599, "naucs_at_1_std": -0.24067127893084392, "naucs_at_1_diff1": 0.586299352615932, "naucs_at_3_max": 0.2029532967032969, "naucs_at_3_std": -0.11450702075702028, "naucs_at_3_diff1": 0.5169413919413928, "naucs_at_5_max": 0.22760186548846306, "naucs_at_5_std": -0.12404884634265974, "naucs_at_5_diff1": 0.44949680903289124, "naucs_at_10_max": 0.06274509803921678, "naucs_at_10_std": -0.3894957983193242, "naucs_at_10_diff1": 0.5401960784313737, "naucs_at_20_max": -0.30376595082477437, "naucs_at_20_std": -0.4344849050731363, "naucs_at_20_diff1": 0.5154061624649885, "naucs_at_50_max": -0.9556489262371552, "naucs_at_50_std": -1.1517273576097031, "naucs_at_50_diff1": 0.3384687208216665, "naucs_at_100_max": -0.17133520074697067, "naucs_at_100_std": -0.5634920634920583, "naucs_at_100_diff1": 0.554154995331464}, "./data_dir/eval_vidore/syntheticDocQA_artificial_intelligence_test": {"ndcg_at_1": 0.95, "ndcg_at_3": 0.98024, "ndcg_at_5": 0.98024, "ndcg_at_10": 0.98024, "ndcg_at_20": 0.98024, "ndcg_at_50": 0.98024, "ndcg_at_100": 0.98024, "map_at_1": 0.95, "map_at_3": 0.97333, "map_at_5": 0.97333, "map_at_10": 0.97333, "map_at_20": 0.97333, "map_at_50": 0.97333, "map_at_100": 0.97333, "recall_at_1": 0.95, "recall_at_3": 1.0, "recall_at_5": 1.0, "recall_at_10": 1.0, "recall_at_20": 1.0, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.95, "precision_at_3": 0.33333, "precision_at_5": 0.2, "precision_at_10": 0.1, "precision_at_20": 0.05, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.96, "mrr_at_3": 0.9783333333333333, "mrr_at_5": 0.9783333333333333, "mrr_at_10": 0.9783333333333333, "mrr_at_20": 0.9783333333333333, "mrr_at_50": 0.9783333333333333, "mrr_at_100": 0.9783333333333333, "naucs_at_1_max": 0.3213818860877688, "naucs_at_1_std": -0.07721755368813897, "naucs_at_1_diff1": 0.9738562091503253, "naucs_at_3_max": 1.0, "naucs_at_3_std": 1.0, "naucs_at_3_diff1": 1.0, "naucs_at_5_max": 1.0, "naucs_at_5_std": 1.0, "naucs_at_5_diff1": 1.0, "naucs_at_10_max": 1.0, "naucs_at_10_std": 1.0, "naucs_at_10_diff1": 1.0, "naucs_at_20_max": 1.0, "naucs_at_20_std": 1.0, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": NaN, "naucs_at_50_std": NaN, "naucs_at_50_diff1": NaN, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}} |