Upload results.json

results.json

@@ -1 +1 @@
-{"validation_set": {"ndcg_at_1": 0.826, "ndcg_at_3": 0.88395, "ndcg_at_5": 0.89135, "ndcg_at_10": 0.89712, "ndcg_at_20": 0.90009, "ndcg_at_50": 0.90242, "ndcg_at_100": 0.90369, "map_at_1": 0.826, "map_at_3": 0.87067, "map_at_5": 0.87477, "map_at_10": 0.87712, "map_at_20": 0.8779, "map_at_50": 0.87825, "map_at_100": 0.87836, "recall_at_1": 0.826, "recall_at_3": 0.922, "recall_at_5": 0.94, "recall_at_10": 0.958, "recall_at_20": 0.97, "recall_at_50": 0.982, "recall_at_100": 0.99, "precision_at_1": 0.826, "precision_at_3": 0.30733, "precision_at_5": 0.188, "precision_at_10": 0.0958, "precision_at_20": 0.0485, "precision_at_50": 0.01964, "precision_at_100": 0.0099, "mrr_at_1": 0.828, "mrr_at_3": 0.869, "mrr_at_5": 0.8742000000000001, "mrr_at_10": 0.8767158730158731, "mrr_at_20": 0.8776762126762127, "mrr_at_50": 0.878013022025774, "mrr_at_100": 0.8781584051922459, "naucs_at_1_max": 0.3432503380797081, "naucs_at_1_std": 0.2095241883246028, "naucs_at_1_diff1": 0.9486630226030438, "naucs_at_3_max": 0.4193540664128913, "naucs_at_3_std": 0.6719935837582925, "naucs_at_3_diff1": 0.9475089180971545, "naucs_at_5_max": 0.4053065670712697, "naucs_at_5_std": 0.7934484905073144, "naucs_at_5_diff1": 0.9540927482103968, "naucs_at_10_max": 0.4086078875994817, "naucs_at_10_std": 0.8241741140900799, "naucs_at_10_diff1": 0.9751011515717399, "naucs_at_20_max": 0.3347650171179612, "naucs_at_20_std": 0.8932461873638364, "naucs_at_20_diff1": 0.9738562091503296, "naucs_at_50_max": 0.3245668637825497, "naucs_at_50_std": 0.865649963689171, "naucs_at_50_diff1": 0.9854756717501884, "naucs_at_100_max": 0.6406162464985967, "naucs_at_100_std": 0.8398692810457434, "naucs_at_100_diff1": 0.9738562091503188}, "syntheticDocQA_energy": {"ndcg_at_1": 0.95, "ndcg_at_3": 0.96762, "ndcg_at_5": 0.96762, "ndcg_at_10": 0.96762, "ndcg_at_20": 0.96762, "ndcg_at_50": 0.97149, "ndcg_at_100": 0.97149, "map_at_1": 0.95, "map_at_3": 0.96333, "map_at_5": 0.96333, "map_at_10": 0.96333, "map_at_20": 0.96333, "map_at_50": 0.96391, "map_at_100": 0.96391, "recall_at_1": 0.95, "recall_at_3": 0.98, "recall_at_5": 0.98, "recall_at_10": 0.98, "recall_at_20": 0.98, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.95, "precision_at_3": 0.32667, "precision_at_5": 0.196, "precision_at_10": 0.098, "precision_at_20": 0.049, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.97, "mrr_at_3": 0.975, "mrr_at_5": 0.975, "mrr_at_10": 0.975, "mrr_at_20": 0.975, "mrr_at_50": 0.9755887445887446, "mrr_at_100": 0.9755887445887446, "naucs_at_1_max": 0.02399626517273991, "naucs_at_1_std": -0.8497665732959815, "naucs_at_1_diff1": 0.9738562091503253, "naucs_at_3_max": 0.9346405228758099, "naucs_at_3_std": 0.19140989729224936, "naucs_at_3_diff1": 0.9346405228758099, "naucs_at_5_max": 0.9346405228758136, "naucs_at_5_std": 0.1914098972922579, "naucs_at_5_diff1": 0.9346405228758136, "naucs_at_10_max": 0.9346405228758136, "naucs_at_10_std": 0.1914098972922579, "naucs_at_10_diff1": 0.9346405228758136, "naucs_at_20_max": 0.9346405228758136, "naucs_at_20_std": 0.1914098972922579, "naucs_at_20_diff1": 0.9346405228758136, "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}, "syntheticDocQA_healthcare_industry": {"ndcg_at_1": 0.96, "ndcg_at_3": 0.98524, "ndcg_at_5": 0.98524, "ndcg_at_10": 0.98524, "ndcg_at_20": 0.98524, "ndcg_at_50": 0.98524, "ndcg_at_100": 0.98524, "map_at_1": 0.96, "map_at_3": 0.98, "map_at_5": 0.98, "map_at_10": 0.98, "map_at_20": 0.98, "map_at_50": 0.98, "map_at_100": 0.98, "recall_at_1": 0.96, "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": 0.96, "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": 0.98, "mrr_at_5": 0.98, "mrr_at_10": 0.98, "mrr_at_20": 0.98, "mrr_at_50": 0.98, "mrr_at_100": 0.98, "naucs_at_1_max": 0.4025443510737659, "naucs_at_1_std": 0.29470121381886283, "naucs_at_1_diff1": 0.96732026143791, "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}, "syntheticDocQA_artificial_intelligence_test": {"ndcg_at_1": 0.96, "ndcg_at_3": 0.97893, "ndcg_at_5": 0.98323, "ndcg_at_10": 0.98323, "ndcg_at_20": 0.98323, "ndcg_at_50": 0.98323, "ndcg_at_100": 0.98323, "map_at_1": 0.96, "map_at_3": 0.975, "map_at_5": 0.9775, "map_at_10": 0.9775, "map_at_20": 0.9775, "map_at_50": 0.9775, "map_at_100": 0.9775, "recall_at_1": 0.96, "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": 0.96, "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": 0.96, "mrr_at_3": 0.975, "mrr_at_5": 0.9775, "mrr_at_10": 0.9775, "mrr_at_20": 0.9775, "mrr_at_50": 0.9775, "mrr_at_100": 0.9775, "naucs_at_1_max": 0.6458916900093364, "naucs_at_1_std": -0.053688141923436454, "naucs_at_1_diff1": 1.0, "naucs_at_3_max": 0.8692810457516356, "naucs_at_3_std": -1.1517273576097802, "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}, "syntheticDocQA_government_reports": {"ndcg_at_1": 0.87, "ndcg_at_3": 0.93678, "ndcg_at_5": 0.94452, "ndcg_at_10": 0.94452, "ndcg_at_20": 0.94452, "ndcg_at_50": 0.94452, "ndcg_at_100": 0.94452, "map_at_1": 0.87, "map_at_3": 0.92167, "map_at_5": 0.92567, "map_at_10": 0.92567, "map_at_20": 0.92567, "map_at_50": 0.92567, "map_at_100": 0.92567, "recall_at_1": 0.87, "recall_at_3": 0.98, "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": 0.32667, "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": 0.9283333333333332, "mrr_at_5": 0.9323333333333333, "mrr_at_10": 0.9323333333333333, "mrr_at_20": 0.9323333333333333, "mrr_at_50": 0.9323333333333333, "mrr_at_100": 0.9323333333333333, "naucs_at_1_max": 0.37621269347552294, "naucs_at_1_std": 0.36136414130193245, "naucs_at_1_diff1": 0.8826186773309634, "naucs_at_3_max": 0.6136788048552655, "naucs_at_3_std": 0.6136788048552655, "naucs_at_3_diff1": 0.7117180205415458, "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}, "infovqa_subsampled": {"ndcg_at_1": 0.878, "ndcg_at_3": 0.90088, "ndcg_at_5": 0.91414, "ndcg_at_10": 0.91797, "ndcg_at_20": 0.91957, "ndcg_at_50": 0.92115, "ndcg_at_100": 0.92343, "map_at_1": 0.878, "map_at_3": 0.895, "map_at_5": 0.9024, "map_at_10": 0.90396, "map_at_20": 0.90444, "map_at_50": 0.90469, "map_at_100": 0.9049, "recall_at_1": 0.878, "recall_at_3": 0.918, "recall_at_5": 0.95, "recall_at_10": 0.962, "recall_at_20": 0.968, "recall_at_50": 0.976, "recall_at_100": 0.99, "precision_at_1": 0.878, "precision_at_3": 0.306, "precision_at_5": 0.19, "precision_at_10": 0.0962, "precision_at_20": 0.0484, "precision_at_50": 0.01952, "precision_at_100": 0.0099, "mrr_at_1": 0.874, "mrr_at_3": 0.8923333333333334, "mrr_at_5": 0.9007333333333334, "mrr_at_10": 0.9020444444444444, "mrr_at_20": 0.9025257278669044, "mrr_at_50": 0.9028050383312147, "mrr_at_100": 0.9029730583888165, "naucs_at_1_max": 0.6417911617626315, "naucs_at_1_std": -0.011788496809894352, "naucs_at_1_diff1": 0.9438792306124684, "naucs_at_3_max": 0.7491858532030709, "naucs_at_3_std": 0.04282526018537632, "naucs_at_3_diff1": 0.9373163899706234, "naucs_at_5_max": 0.8344724556489264, "naucs_at_5_std": 0.2926610644257683, "naucs_at_5_diff1": 0.9470588235294107, "naucs_at_10_max": 0.9240748931151417, "naucs_at_10_std": 0.578775369797045, "naucs_at_10_diff1": 0.9587203302373579, "naucs_at_20_max": 0.9180088702147511, "naucs_at_20_std": 0.5161356209150314, "naucs_at_20_diff1": 0.9591503267973829, "naucs_at_50_max": 0.9015717398070241, "naucs_at_50_std": 0.5081310301898541, "naucs_at_50_diff1": 0.9673202614379068, "naucs_at_100_max": 0.8921568627450854, "naucs_at_100_std": 0.6873015873015698, "naucs_at_100_diff1": 0.9738562091503188}, "docvqa_subsampled": {"ndcg_at_1": 0.472, "ndcg_at_3": 0.54143, "ndcg_at_5": 0.55975, "ndcg_at_10": 0.57821, "ndcg_at_20": 0.59145, "ndcg_at_50": 0.60588, "ndcg_at_100": 0.61495, "map_at_1": 0.472, "map_at_3": 0.524, "map_at_5": 0.5338, "map_at_10": 0.54163, "map_at_20": 0.54531, "map_at_50": 0.54753, "map_at_100": 0.54833, "recall_at_1": 0.472, "recall_at_3": 0.592, "recall_at_5": 0.638, "recall_at_10": 0.694, "recall_at_20": 0.746, "recall_at_50": 0.82, "recall_at_100": 0.876, "precision_at_1": 0.472, "precision_at_3": 0.19733, "precision_at_5": 0.1276, "precision_at_10": 0.0694, "precision_at_20": 0.0373, "precision_at_50": 0.0164, "precision_at_100": 0.00876, "mrr_at_1": 0.478, "mrr_at_3": 0.5263333333333333, "mrr_at_5": 0.5356333333333333, "mrr_at_10": 0.5434126984126985, "mrr_at_20": 0.5472397111058875, "mrr_at_50": 0.5499635119629092, "mrr_at_100": 0.5505736042826437, "naucs_at_1_max": 0.3923025767156111, "naucs_at_1_std": 0.7053772180760662, "naucs_at_1_diff1": 0.9042387552384067, "naucs_at_3_max": 0.29406297063253023, "naucs_at_3_std": 0.759083207831325, "naucs_at_3_diff1": 0.8269248870481934, "naucs_at_5_max": 0.22360003435693662, "naucs_at_5_std": 0.8072734907602107, "naucs_at_5_diff1": 0.8233417210916594, "naucs_at_10_max": 0.158478323509218, "naucs_at_10_std": 0.8277726591986796, "naucs_at_10_diff1": 0.7939532782424016, "naucs_at_20_max": 0.06102847885472823, "naucs_at_20_std": 0.8541052779488288, "naucs_at_20_diff1": 0.7847864809128443, "naucs_at_50_max": 0.04328119836986498, "naucs_at_50_std": 0.89758233285604, "naucs_at_50_diff1": 0.7738242097147262, "naucs_at_100_max": 0.015149408238311582, "naucs_at_100_std": 0.9025215537220722, "naucs_at_100_diff1": 0.789476530391521}, "arxivqa_subsampled": {"ndcg_at_1": 0.822, "ndcg_at_3": 0.87355, "ndcg_at_5": 0.88413, "ndcg_at_10": 0.89092, "ndcg_at_20": 0.89665, "ndcg_at_50": 0.89867, "ndcg_at_100": 0.89967, "map_at_1": 0.822, "map_at_3": 0.86033, "map_at_5": 0.86613, "map_at_10": 0.86912, "map_at_20": 0.87078, "map_at_50": 0.87112, "map_at_100": 0.87122, "recall_at_1": 0.822, "recall_at_3": 0.912, "recall_at_5": 0.938, "recall_at_10": 0.958, "recall_at_20": 0.98, "recall_at_50": 0.99, "recall_at_100": 0.996, "precision_at_1": 0.822, "precision_at_3": 0.304, "precision_at_5": 0.1876, "precision_at_10": 0.0958, "precision_at_20": 0.049, "precision_at_50": 0.0198, "precision_at_100": 0.00996, "mrr_at_1": 0.822, "mrr_at_3": 0.8613333333333334, "mrr_at_5": 0.8663333333333334, "mrr_at_10": 0.8695214285714286, "mrr_at_20": 0.8711353662677191, "mrr_at_50": 0.8715260011883541, "mrr_at_100": 0.8715838702010618, "naucs_at_1_max": 0.7184700475188902, "naucs_at_1_std": -0.08461589824057751, "naucs_at_1_diff1": 0.942910559877141, "naucs_at_3_max": 0.7786796536796564, "naucs_at_3_std": -0.022472625413802825, "naucs_at_3_diff1": 0.9324123588829496, "naucs_at_5_max": 0.7821451161109595, "naucs_at_5_std": -0.10835517002500032, "naucs_at_5_diff1": 0.9530887623866757, "naucs_at_10_max": 0.7989862611711338, "naucs_at_10_std": -0.19698990707394395, "naucs_at_10_diff1": 0.9502023031434798, "naucs_at_20_max": 0.7794584500466787, "naucs_at_20_std": -0.047385620915033094, "naucs_at_20_diff1": 0.934640522875813, "naucs_at_50_max": 0.947712418300658, "naucs_at_50_std": -0.15378151260503836, "naucs_at_50_diff1": 0.9738562091503188, "naucs_at_100_max": 0.9346405228758466, "naucs_at_100_std": -0.3699813258636757, "naucs_at_100_diff1": 1.0}, "tabfquad_subsampled": {"ndcg_at_1": 0.83214, "ndcg_at_3": 0.86849, "ndcg_at_5": 0.88218, "ndcg_at_10": 0.89148, "ndcg_at_20": 0.89908, "ndcg_at_50": 0.90177, "ndcg_at_100": 0.90177, "map_at_1": 0.83214, "map_at_3": 0.85893, "map_at_5": 0.86679, "map_at_10": 0.87065, "map_at_20": 0.87294, "map_at_50": 0.87331, "map_at_100": 0.87331, "recall_at_1": 0.83214, "recall_at_3": 0.89643, "recall_at_5": 0.92857, "recall_at_10": 0.95714, "recall_at_20": 0.98571, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.83214, "precision_at_3": 0.29881, "precision_at_5": 0.18571, "precision_at_10": 0.09571, "precision_at_20": 0.04929, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.8357142857142857, "mrr_at_3": 0.8601190476190477, "mrr_at_5": 0.8690476190476191, "mrr_at_10": 0.872640306122449, "mrr_at_20": 0.8752497021535887, "mrr_at_50": 0.8756202568795441, "mrr_at_100": 0.8756202568795441, "naucs_at_1_max": 0.48693290265746514, "naucs_at_1_std": 0.2166243681439694, "naucs_at_1_diff1": 0.873135893617467, "naucs_at_3_max": 0.5662052459652638, "naucs_at_3_std": 0.4133380769106388, "naucs_at_3_diff1": 0.8562534363983321, "naucs_at_5_max": 0.6014005602240904, "naucs_at_5_std": 0.41097105508870374, "naucs_at_5_diff1": 0.8548319327731102, "naucs_at_10_max": 0.6419623404917498, "naucs_at_10_std": 0.33916900093370245, "naucs_at_10_diff1": 0.8869825708061017, "naucs_at_20_max": 0.865196078431377, "naucs_at_20_std": 0.7496498599439745, "naucs_at_20_diff1": 0.8978758169934754, "naucs_at_50_max": 1.0, "naucs_at_50_std": 1.0, "naucs_at_50_diff1": 1.0, "naucs_at_100_max": 1.0, "naucs_at_100_std": 1.0, "naucs_at_100_diff1": 1.0}, "tatdqa": {"ndcg_at_1": 0.66627, "ndcg_at_3": 0.76591, "ndcg_at_5": 0.78897, "ndcg_at_10": 0.80301, "ndcg_at_20": 0.80898, "ndcg_at_50": 0.81233, "ndcg_at_100": 0.81417, "map_at_1": 0.66627, "map_at_3": 0.74243, "map_at_5": 0.75524, "map_at_10": 0.76106, "map_at_20": 0.7628, "map_at_50": 0.76334, "map_at_100": 0.7635, "recall_at_1": 0.66627, "recall_at_3": 0.83343, "recall_at_5": 0.88936, "recall_at_10": 0.93265, "recall_at_20": 0.9555, "recall_at_50": 0.97234, "recall_at_100": 0.98376, "precision_at_1": 0.66627, "precision_at_3": 0.27781, "precision_at_5": 0.17787, "precision_at_10": 0.09327, "precision_at_20": 0.04778, "precision_at_50": 0.01945, "precision_at_100": 0.00984, "mrr_at_1": 0.6650631389055923, "mrr_at_3": 0.7424333533774304, "mrr_at_5": 0.7541290839847664, "mrr_at_10": 0.7603403678569043, "mrr_at_20": 0.7620416242709359, "mrr_at_50": 0.7625731305022777, "mrr_at_100": 0.7627555296002397, "naucs_at_1_max": 0.2255901193553477, "naucs_at_1_std": -0.17195994632386122, "naucs_at_1_diff1": 0.8184172224887907, "naucs_at_3_max": 0.2738703807942716, "naucs_at_3_std": -0.11962815389024864, "naucs_at_3_diff1": 0.7412672113098526, "naucs_at_5_max": 0.3481139472219308, "naucs_at_5_std": -0.026753182766958084, "naucs_at_5_diff1": 0.703118215452914, "naucs_at_10_max": 0.4767078473369491, "naucs_at_10_std": 0.2645481592876129, "naucs_at_10_diff1": 0.6929624809523717, "naucs_at_20_max": 0.5055479453630981, "naucs_at_20_std": 0.42820219435749896, "naucs_at_20_diff1": 0.6905113082549136, "naucs_at_50_max": 0.5175966319812615, "naucs_at_50_std": 0.518633493957509, "naucs_at_50_diff1": 0.6885449373033483, "naucs_at_100_max": 0.6300622311883954, "naucs_at_100_std": 0.7252919027287567, "naucs_at_100_diff1": 0.7003719897704582}, "shift_project": {"ndcg_at_1": 0.7, "ndcg_at_3": 0.82702, "ndcg_at_5": 0.8352, "ndcg_at_10": 0.84124, "ndcg_at_20": 0.84639, "ndcg_at_50": 0.84836, "ndcg_at_100": 0.84996, "map_at_1": 0.7, "map_at_3": 0.795, "map_at_5": 0.7995, "map_at_10": 0.80175, "map_at_20": 0.80321, "map_at_50": 0.80351, "map_at_100": 0.80365, "recall_at_1": 0.7, "recall_at_3": 0.92, "recall_at_5": 0.94, "recall_at_10": 0.96, "recall_at_20": 0.98, "recall_at_50": 0.99, "recall_at_100": 1.0, "precision_at_1": 0.7, "precision_at_3": 0.30667, "precision_at_5": 0.188, "precision_at_10": 0.096, "precision_at_20": 0.049, "precision_at_50": 0.0198, "precision_at_100": 0.01, "mrr_at_1": 0.7, "mrr_at_3": 0.795, "mrr_at_5": 0.7995, "mrr_at_10": 0.8029999999999999, "mrr_at_20": 0.8036249999999999, "mrr_at_50": 0.8039475806451613, "mrr_at_100": 0.8040827157802964, "naucs_at_1_max": 0.007692307692307685, "naucs_at_1_std": -0.21826697892271701, "naucs_at_1_diff1": 0.6957665285534135, "naucs_at_3_max": 0.2762605042016824, "naucs_at_3_std": -0.2441059757236238, "naucs_at_3_diff1": 0.8978758169934661, "naucs_at_5_max": 0.15561780267662587, "naucs_at_5_std": -0.31738250855898026, "naucs_at_5_diff1": 0.8856209150326813, "naucs_at_10_max": -0.04726890756302253, "naucs_at_10_std": -0.6832399626517155, "naucs_at_10_diff1": 0.9305555555555542, "naucs_at_20_max": -0.5088702147525547, "naucs_at_20_std": -0.5088702147525547, "naucs_at_20_diff1": 1.0, "naucs_at_50_max": 0.7222222222222041, "naucs_at_50_std": 0.7222222222222041, "naucs_at_50_diff1": 1.0, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}}
+{"./data_dir/eval_vidore/tatdqa_test": {"ndcg_at_1": 0.67193, "ndcg_at_3": 0.77075, "ndcg_at_5": 0.79358, "ndcg_at_10": 0.8076, "ndcg_at_20": 0.81423, "ndcg_at_50": 0.81698, "ndcg_at_100": 0.81863, "map_at_1": 0.67193, "map_at_3": 0.74727, "map_at_5": 0.75996, "map_at_10": 0.76579, "map_at_20": 0.7677, "map_at_50": 0.76813, "map_at_100": 0.76827, "recall_at_1": 0.67193, "recall_at_3": 0.8384, "recall_at_5": 0.89368, "recall_at_10": 0.93682, "recall_at_20": 0.96233, "recall_at_50": 0.97631, "recall_at_100": 0.98663, "precision_at_1": 0.67193, "precision_at_3": 0.27947, "precision_at_5": 0.17874, "precision_at_10": 0.09368, "precision_at_20": 0.04812, "precision_at_50": 0.01953, "precision_at_100": 0.00987, "mrr_at_1": 0.6676792223572297, "mrr_at_3": 0.745240988254354, "mrr_at_5": 0.7566018631024706, "mrr_at_10": 0.7630472622422804, "mrr_at_20": 0.7648916939123942, "mrr_at_50": 0.7653111892014712, "mrr_at_100": 0.7654665531953649, "naucs_at_1_max": 0.20771145663189075, "naucs_at_1_std": -0.18773433911267384, "naucs_at_1_diff1": 0.8170194754624222, "naucs_at_3_max": 0.24607774142531255, "naucs_at_3_std": -0.16584110193613, "naucs_at_3_diff1": 0.7378398220281965, "naucs_at_5_max": 0.2989857582497622, "naucs_at_5_std": -0.08436505768312826, "naucs_at_5_diff1": 0.7055812813302946, "naucs_at_10_max": 0.4087086049178199, "naucs_at_10_std": 0.1542914363622356, "naucs_at_10_diff1": 0.6839773195644397, "naucs_at_20_max": 0.45412221479599435, "naucs_at_20_std": 0.3801693627818807, "naucs_at_20_diff1": 0.6874517110683748, "naucs_at_50_max": 0.42104082268255205, "naucs_at_50_std": 0.46988809251941027, "naucs_at_50_diff1": 0.6654484997896868, "naucs_at_100_max": 0.507330323866924, "naucs_at_100_std": 0.6683042871053126, "naucs_at_100_diff1": 0.661034185188578}, "./data_dir/eval_vidore/shiftproject_test": {"ndcg_at_1": 0.71, "ndcg_at_3": 0.82309, "ndcg_at_5": 0.83514, "ndcg_at_10": 0.84118, "ndcg_at_20": 0.84624, "ndcg_at_50": 0.84824, "ndcg_at_100": 0.84986, "map_at_1": 0.71, "map_at_3": 0.79333, "map_at_5": 0.79983, "map_at_10": 0.80208, "map_at_20": 0.80346, "map_at_50": 0.80379, "map_at_100": 0.80393, "recall_at_1": 0.71, "recall_at_3": 0.91, "recall_at_5": 0.94, "recall_at_10": 0.96, "recall_at_20": 0.98, "recall_at_50": 0.99, "recall_at_100": 1.0, "precision_at_1": 0.71, "precision_at_3": 0.30333, "precision_at_5": 0.188, "precision_at_10": 0.096, "precision_at_20": 0.049, "precision_at_50": 0.0198, "precision_at_100": 0.01, "mrr_at_1": 0.71, "mrr_at_3": 0.795, "mrr_at_5": 0.802, "mrr_at_10": 0.80425, "mrr_at_20": 0.8058733766233765, "mrr_at_50": 0.80620670995671, "mrr_at_100": 0.8063516374929418, "naucs_at_1_max": -0.006300613383243949, "naucs_at_1_std": -0.21616663269277062, "naucs_at_1_diff1": 0.6874339825435943, "naucs_at_3_max": 0.35667600373482716, "naucs_at_3_std": -0.1857557837949981, "naucs_at_3_diff1": 0.830635958086938, "naucs_at_5_max": 0.15561780267662587, "naucs_at_5_std": -0.39091192032368627, "naucs_at_5_diff1": 0.7895269218798655, "naucs_at_10_max": -0.04726890756302253, "naucs_at_10_std": -0.7935340802987776, "naucs_at_10_diff1": 0.8231792717086845, "naucs_at_20_max": -0.5088702147525547, "naucs_at_20_std": -0.43534080298785716, "naucs_at_20_diff1": 0.8692810457516353, "naucs_at_50_max": 0.7222222222222041, "naucs_at_50_std": 0.8692810457516374, "naucs_at_50_diff1": 0.8692810457516374, "naucs_at_100_max": NaN, "naucs_at_100_std": NaN, "naucs_at_100_diff1": NaN}, "./data_dir/eval_vidore/syntheticDocQA_artificial_intelligence_test": {"ndcg_at_1": 0.96, "ndcg_at_3": 0.97893, "ndcg_at_5": 0.98323, "ndcg_at_10": 0.98323, "ndcg_at_20": 0.98323, "ndcg_at_50": 0.98323, "ndcg_at_100": 0.98323, "map_at_1": 0.96, "map_at_3": 0.975, "map_at_5": 0.9775, "map_at_10": 0.9775, "map_at_20": 0.9775, "map_at_50": 0.9775, "map_at_100": 0.9775, "recall_at_1": 0.96, "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": 0.96, "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": 0.96, "mrr_at_3": 0.975, "mrr_at_5": 0.9775, "mrr_at_10": 0.9775, "mrr_at_20": 0.9775, "mrr_at_50": 0.9775, "mrr_at_100": 0.9775, "naucs_at_1_max": 0.6458916900093364, "naucs_at_1_std": -0.053688141923436454, "naucs_at_1_diff1": 1.0, "naucs_at_3_max": 0.8692810457516356, "naucs_at_3_std": -1.1517273576097802, "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/syntheticDocQA_government_reports_test": {"ndcg_at_1": 0.88, "ndcg_at_3": 0.94047, "ndcg_at_5": 0.94865, "ndcg_at_10": 0.94865, "ndcg_at_20": 0.94865, "ndcg_at_50": 0.94865, "ndcg_at_100": 0.94865, "map_at_1": 0.88, "map_at_3": 0.92667, "map_at_5": 0.93117, "map_at_10": 0.93117, "map_at_20": 0.93117, "map_at_50": 0.93117, "map_at_100": 0.93117, "recall_at_1": 0.88, "recall_at_3": 0.98, "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": 0.88, "precision_at_3": 0.32667, "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": 0.9283333333333332, "mrr_at_5": 0.9328333333333333, "mrr_at_10": 0.9328333333333333, "mrr_at_20": 0.9328333333333333, "mrr_at_50": 0.9328333333333333, "mrr_at_100": 0.9328333333333333, "naucs_at_1_max": 0.31053133948456807, "naucs_at_1_std": 0.3409163219853637, "naucs_at_1_diff1": 0.8739261851734011, "naucs_at_3_max": 0.6136788048552655, "naucs_at_3_std": 0.6136788048552655, "naucs_at_3_diff1": 0.7117180205415458, "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/arxivqa_test_subsampled": {"ndcg_at_1": 0.824, "ndcg_at_3": 0.87528, "ndcg_at_5": 0.88501, "ndcg_at_10": 0.89231, "ndcg_at_20": 0.89749, "ndcg_at_50": 0.89952, "ndcg_at_100": 0.90052, "map_at_1": 0.824, "map_at_3": 0.862, "map_at_5": 0.8673, "map_at_10": 0.87042, "map_at_20": 0.87191, "map_at_50": 0.87225, "map_at_100": 0.87235, "recall_at_1": 0.824, "recall_at_3": 0.914, "recall_at_5": 0.938, "recall_at_10": 0.96, "recall_at_20": 0.98, "recall_at_50": 0.99, "recall_at_100": 0.996, "precision_at_1": 0.824, "precision_at_3": 0.30467, "precision_at_5": 0.1876, "precision_at_10": 0.096, "precision_at_20": 0.049, "precision_at_50": 0.0198, "precision_at_100": 0.00996, "mrr_at_1": 0.822, "mrr_at_3": 0.8623333333333334, "mrr_at_5": 0.8669333333333333, "mrr_at_10": 0.8700460317460318, "mrr_at_20": 0.8716778804528804, "mrr_at_50": 0.8720248035298035, "mrr_at_100": 0.8721252844768993, "naucs_at_1_max": 0.735077815761065, "naucs_at_1_std": -0.04282115869017602, "naucs_at_1_diff1": 0.9371963835912196, "naucs_at_3_max": 0.7796126202418969, "naucs_at_3_std": -0.025850650337654255, "naucs_at_3_diff1": 0.9369205046359642, "naucs_at_5_max": 0.7821451161109595, "naucs_at_5_std": -0.09443992650824237, "naucs_at_5_diff1": 0.9525616698292219, "naucs_at_10_max": 0.7889355742296903, "naucs_at_10_std": -0.2129785247432343, "naucs_at_10_diff1": 0.9607843137254902, "naucs_at_20_max": 0.7794584500466787, "naucs_at_20_std": -0.02777777777778027, "naucs_at_20_diff1": 0.9477124183006537, "naucs_at_50_max": 0.9215686274509768, "naucs_at_50_std": -0.06750700280111896, "naucs_at_50_diff1": 0.947712418300658, "naucs_at_100_max": 0.9346405228758466, "naucs_at_100_std": -0.3699813258636757, "naucs_at_100_diff1": 1.0}, "./data_dir/eval_vidore/docvqa_test_subsampled": {"ndcg_at_1": 0.5255, "ndcg_at_3": 0.60078, "ndcg_at_5": 0.61862, "ndcg_at_10": 0.63259, "ndcg_at_20": 0.64451, "ndcg_at_50": 0.65881, "ndcg_at_100": 0.66752, "map_at_1": 0.5255, "map_at_3": 0.58167, "map_at_5": 0.59132, "map_at_10": 0.59728, "map_at_20": 0.60064, "map_at_50": 0.60285, "map_at_100": 0.60364, "recall_at_1": 0.5255, "recall_at_3": 0.65632, "recall_at_5": 0.70067, "recall_at_10": 0.74279, "recall_at_20": 0.78936, "recall_at_50": 0.86253, "recall_at_100": 0.91574, "precision_at_1": 0.5255, "precision_at_3": 0.21877, "precision_at_5": 0.14013, "precision_at_10": 0.07428, "precision_at_20": 0.03947, "precision_at_50": 0.01725, "precision_at_100": 0.00916, "mrr_at_1": 0.5277161862527716, "mrr_at_3": 0.5805617147080562, "mrr_at_5": 0.5903178122690318, "mrr_at_10": 0.5972389399218668, "mrr_at_20": 0.6002228297434856, "mrr_at_50": 0.6027282655343267, "mrr_at_100": 0.6033300437903428, "naucs_at_1_max": 0.34919146368331516, "naucs_at_1_std": 0.6755700583274495, "naucs_at_1_diff1": 0.9012163798897747, "naucs_at_3_max": 0.21972764067769676, "naucs_at_3_std": 0.7384263112544703, "naucs_at_3_diff1": 0.8261106084064109, "naucs_at_5_max": 0.13679495064259664, "naucs_at_5_std": 0.7543173419080406, "naucs_at_5_diff1": 0.8038332262621047, "naucs_at_10_max": 0.08213096380071569, "naucs_at_10_std": 0.8025911097016517, "naucs_at_10_diff1": 0.8067908578890264, "naucs_at_20_max": 0.009640010057001363, "naucs_at_20_std": 0.8318658815167514, "naucs_at_20_diff1": 0.7973567600141757, "naucs_at_50_max": -0.05840523108601927, "naucs_at_50_std": 0.8861125417030429, "naucs_at_50_diff1": 0.7971097456566356, "naucs_at_100_max": -0.17013898231075125, "naucs_at_100_std": 0.8872540701354344, "naucs_at_100_diff1": 0.7978782264372887}, "./data_dir/eval_vidore/syntheticDocQA_healthcare_industry_test": {"ndcg_at_1": 0.96, "ndcg_at_3": 0.98524, "ndcg_at_5": 0.98524, "ndcg_at_10": 0.98524, "ndcg_at_20": 0.98524, "ndcg_at_50": 0.98524, "ndcg_at_100": 0.98524, "map_at_1": 0.96, "map_at_3": 0.98, "map_at_5": 0.98, "map_at_10": 0.98, "map_at_20": 0.98, "map_at_50": 0.98, "map_at_100": 0.98, "recall_at_1": 0.96, "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": 0.96, "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": 0.98, "mrr_at_5": 0.98, "mrr_at_10": 0.98, "mrr_at_20": 0.98, "mrr_at_50": 0.98, "mrr_at_100": 0.98, "naucs_at_1_max": 0.47607376283847014, "naucs_at_1_std": 0.29470121381886283, "naucs_at_1_diff1": 1.0, "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/infovqa_test_subsampled": {"ndcg_at_1": 0.88664, "ndcg_at_3": 0.91464, "ndcg_at_5": 0.92466, "ndcg_at_10": 0.92913, "ndcg_at_20": 0.93018, "ndcg_at_50": 0.93097, "ndcg_at_100": 0.9323, "map_at_1": 0.88664, "map_at_3": 0.90756, "map_at_5": 0.91312, "map_at_10": 0.91491, "map_at_20": 0.91521, "map_at_50": 0.91533, "map_at_100": 0.91545, "recall_at_1": 0.88664, "recall_at_3": 0.93522, "recall_at_5": 0.95951, "recall_at_10": 0.97368, "recall_at_20": 0.97773, "recall_at_50": 0.98178, "recall_at_100": 0.98988, "precision_at_1": 0.88664, "precision_at_3": 0.31174, "precision_at_5": 0.1919, "precision_at_10": 0.09737, "precision_at_20": 0.04889, "precision_at_50": 0.01964, "precision_at_100": 0.0099, "mrr_at_1": 0.8825910931174089, "mrr_at_3": 0.9045209176788124, "mrr_at_5": 0.9107962213225371, "mrr_at_10": 0.9126630679262259, "mrr_at_20": 0.912966170396511, "mrr_at_50": 0.9130906841326399, "mrr_at_100": 0.9132104022267463, "naucs_at_1_max": 0.6260502828027972, "naucs_at_1_std": -0.005294178018755926, "naucs_at_1_diff1": 0.9399416231604794, "naucs_at_3_max": 0.7681566860034388, "naucs_at_3_std": 0.09777449128981919, "naucs_at_3_diff1": 0.9365795595509864, "naucs_at_5_max": 0.8274718121903979, "naucs_at_5_std": 0.28017408880961947, "naucs_at_5_diff1": 0.9469328268926711, "naucs_at_10_max": 0.8890624021380364, "naucs_at_10_std": 0.6020867105858281, "naucs_at_10_diff1": 0.9698612029081272, "naucs_at_20_max": 0.8807647892599361, "naucs_at_20_std": 0.5416116992436861, "naucs_at_20_diff1": 0.9643814216187027, "naucs_at_50_max": 0.8687793484360007, "naucs_at_50_std": 0.5014664630218402, "naucs_at_50_diff1": 0.9709774546522744, "naucs_at_100_max": 0.892212470822504, "naucs_at_100_std": 0.6872894779438028, "naucs_at_100_diff1": 1.0}, "./data_dir/eval_vidore/syntheticDocQA_energy_test": {"ndcg_at_1": 0.95, "ndcg_at_3": 0.96762, "ndcg_at_5": 0.96762, "ndcg_at_10": 0.96762, "ndcg_at_20": 0.96762, "ndcg_at_50": 0.97147, "ndcg_at_100": 0.97147, "map_at_1": 0.95, "map_at_3": 0.96333, "map_at_5": 0.96333, "map_at_10": 0.96333, "map_at_20": 0.96333, "map_at_50": 0.9639, "map_at_100": 0.9639, "recall_at_1": 0.95, "recall_at_3": 0.98, "recall_at_5": 0.98, "recall_at_10": 0.98, "recall_at_20": 0.98, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.95, "precision_at_3": 0.32667, "precision_at_5": 0.196, "precision_at_10": 0.098, "precision_at_20": 0.049, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.97, "mrr_at_3": 0.975, "mrr_at_5": 0.975, "mrr_at_10": 0.975, "mrr_at_20": 0.975, "mrr_at_50": 0.9755971479500891, "mrr_at_100": 0.9755971479500891, "naucs_at_1_max": 0.02399626517273991, "naucs_at_1_std": -0.8497665732959815, "naucs_at_1_diff1": 0.9738562091503253, "naucs_at_3_max": 0.9346405228758099, "naucs_at_3_std": 0.19140989729224936, "naucs_at_3_diff1": 0.9346405228758099, "naucs_at_5_max": 0.9346405228758136, "naucs_at_5_std": 0.1914098972922579, "naucs_at_5_diff1": 0.9346405228758136, "naucs_at_10_max": 0.9346405228758136, "naucs_at_10_std": 0.1914098972922579, "naucs_at_10_diff1": 0.9346405228758136, "naucs_at_20_max": 0.9346405228758136, "naucs_at_20_std": 0.1914098972922579, "naucs_at_20_diff1": 0.9346405228758136, "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": 0.83214, "ndcg_at_3": 0.86764, "ndcg_at_5": 0.88133, "ndcg_at_10": 0.89185, "ndcg_at_20": 0.89926, "ndcg_at_50": 0.90195, "ndcg_at_100": 0.90195, "map_at_1": 0.83214, "map_at_3": 0.85893, "map_at_5": 0.86679, "map_at_10": 0.8712, "map_at_20": 0.87333, "map_at_50": 0.8737, "map_at_100": 0.8737, "recall_at_1": 0.83214, "recall_at_3": 0.89286, "recall_at_5": 0.925, "recall_at_10": 0.95714, "recall_at_20": 0.98571, "recall_at_50": 1.0, "recall_at_100": 1.0, "precision_at_1": 0.83214, "precision_at_3": 0.29762, "precision_at_5": 0.185, "precision_at_10": 0.09571, "precision_at_20": 0.04929, "precision_at_50": 0.02, "precision_at_100": 0.01, "mrr_at_1": 0.8321428571428572, "mrr_at_3": 0.8565476190476191, "mrr_at_5": 0.8663690476190476, "mrr_at_10": 0.8704435941043084, "mrr_at_20": 0.8729122974947554, "mrr_at_50": 0.8732756347375155, "mrr_at_100": 0.8732756347375155, "naucs_at_1_max": 0.4951548964909703, "naucs_at_1_std": 0.2199278478092164, "naucs_at_1_diff1": 0.8757052666904376, "naucs_at_3_max": 0.5832810047095762, "naucs_at_3_std": 0.4004395604395596, "naucs_at_3_diff1": 0.8785871271585566, "naucs_at_5_max": 0.605375483526744, "naucs_at_5_std": 0.4247921390778528, "naucs_at_5_diff1": 0.8749722111066649, "naucs_at_10_max": 0.5688608776844056, "naucs_at_10_std": 0.2660675381263621, "naucs_at_10_diff1": 0.922385620915035, "naucs_at_20_max": 0.865196078431377, "naucs_at_20_std": 0.7496498599439745, "naucs_at_20_diff1": 0.8978758169934754, "naucs_at_50_max": 1.0, "naucs_at_50_std": 1.0, "naucs_at_50_diff1": 1.0, "naucs_at_100_max": 1.0, "naucs_at_100_std": 1.0, "naucs_at_100_diff1": 1.0}}