diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/eval.log b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/eval.log new file mode 100644 index 0000000000000000000000000000000000000000..ad44bd37e51f82471f7a17d7372cee5fed00ed3f --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/eval.log @@ -0,0 +1,3 @@ +evaluated learned fallback rule on held-out measured rollout rows +eval_input=runs/ctt_base_context_obs_test_envclip_k16_rollout_comparison/combined_measured_candidates.json +num_eval_rows=144 diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/metrics.json b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..aef9026ef8603ed98bfcf4888e4a7b5c6f019eba --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/metrics.json @@ -0,0 +1,9052 @@ +{ + "calibration_input": "runs/ctt_base_context_obs_train_cal_envclip_k16_rollout_comparison/combined_measured_candidates.json", + "calibration_model_selection": { + "base_success": 0.2777777777777778, + "base_utility": 0.7007981554626086, + "coverage": 0.5601851851851852, + "fallback_rate": 0.4398148148148148, + "hidden_chart_oracle_success": 0.6597222222222222, + "hidden_chart_oracle_utility": 1.4150809271054134, + "outcome_ptr": 0.5462962962962963, + "proposal_oracle_success": 0.5370370370370371, + "proposal_oracle_utility": 1.1440672337592803, + "selected_success": 0.37962962962962965, + "selected_success_gain_over_base": 0.10185185185185185, + "selected_utility": 0.8782783362403279, + "selector_regret": 0.36712626344739907, + "success_selector_gap": 0.20601851851851852, + "success_support_gap": 0.17592592592592593, + "support_gap": 0.35465237537627453 + }, + "calibration_summary": { + "base_success": 0.2777777777777778, + "base_utility": 0.7007981554626086, + "coverage": 0.5601851851851852, + "fallback_rate": 0.4398148148148148, + "hidden_chart_oracle_success": 0.6597222222222222, + "hidden_chart_oracle_utility": 1.4150809271054134, + "outcome_ptr": 0.5462962962962963, + "pairwise_causal_calibration_accuracy": 0.506767075675391, + "pairwise_causal_calibration_confidence": 0.5434667958528854, + "pairwise_causal_calibration_ece": 0.03669972017749429, + "pairwise_causal_calibration_pairs": 37978.0, + "proposal_oracle_success": 0.5370370370370371, + "proposal_oracle_utility": 1.1440672337592803, + "selected_success": 0.37962962962962965, + "selected_success_gain_over_base": 0.10185185185185185, + "selected_utility": 0.8782783362403279, + "selector_regret": 0.36712626344739907, + "success_selector_gap": 0.20601851851851852, + "success_support_gap": 0.17592592592592593, + "support_gap": 0.35465237537627453 + }, + "calibration_target_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "calibration_target_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "chart_feature_mode": "base_context_obs", + "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "eval_input": "runs/ctt_base_context_obs_test_envclip_k16_rollout_comparison/combined_measured_candidates.json", + "eval_summary": { + "base_success": 0.2916666666666667, + "base_utility": 0.7418195864423373, + "coverage": 0.6111111111111112, + "fallback_rate": 0.3888888888888889, + "hidden_chart_oracle_success": 0.7291666666666666, + "hidden_chart_oracle_utility": 1.5183161981403828, + "outcome_ptr": 0.5486111111111112, + "pairwise_causal_calibration_accuracy": 0.5160045037799582, + "pairwise_causal_calibration_confidence": 0.5414611045062885, + "pairwise_causal_calibration_ece": 0.025456600726330422, + "pairwise_causal_calibration_pairs": 12434.0, + "proposal_oracle_success": 0.5694444444444444, + "proposal_oracle_utility": 1.2159999172436073, + "selected_success": 0.3125, + "selected_success_gain_over_base": 0.020833333333333332, + "selected_utility": 0.7690221180495832, + "selector_regret": 0.501692300305068, + "success_selector_gap": 0.2777777777777778, + "success_support_gap": 0.2013888888888889, + "support_gap": 0.3761285081272945 + }, + "eval_target_content_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "eval_target_split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "feature_mean": [ + 0.0, + 4.602051421289207, + 1.8887871984592688, + -1.985099232970326e-17, + 7.187557603686636, + 0.0663594470046083, + 2.0903225806451613, + 0.7930022866760579, + 1.9819565879721794, + 15.375115207373272, + 1.0, + 1.044349800585736, + 1.0509106460665045, + 0.8002031215841607, + 0.5811250768510506, + 2.5434513049028507, + 0.34326541665726357, + 0.8966401857535892, + 1.2182893520781068, + 1.2263923089918751, + 0.786961656617233, + 0.6987682628846764, + 1.0, + 0.9974076898437442, + 0.02008356847617045, + 0.39137378168005876, + 0.3910654018162905, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 2.788372394309843, + 5.701382488479263, + 12.81136712749616, + 0.40686157587079336, + 0.8936925909045862, + 0.24551906257091227, + 0.12993606097190302, + 0.1850963060060051, + 0.0663594470046083, + 3.944723778791356, + 4.364277845709234, + 4.603718512944113, + 5.60843294300425, + 4.854070660522273, + 0.3680969062939123, + 1.0, + 0.3217802000866517, + 4.613985697387828, + 0.5009397088659758 + ], + "feature_names": [ + "bias", + "candidate_score", + "candidate_score_minus_base_score", + "row_score_z", + "candidate_index", + "is_top_score_candidate", + "source_chart_rank", + "tangent_rms_norm", + "tangent_linf_norm", + "num_candidates", + "source_chart_feature_found", + "target_chart_feature_norm", + "source_chart_feature_norm", + "target_source_chart_cosine", + "target_source_chart_rms", + "target_source_chart_linf", + "target_source_chart_absmean", + "target_source_chart_dot_per_dim", + "target_base_action_norm", + "source_base_action_norm", + "target_source_base_cosine", + "target_source_base_rms", + "target_source_obs_available", + "target_source_obs_cosine", + "target_source_obs_rms", + "target_obs_norm", + "source_obs_norm", + "target_source_obj_available", + "target_source_obj_cosine", + "target_source_obj_rms", + "target_obj_norm", + "source_obj_norm", + "bundle_num_candidates_log", + "bundle_neighbor_count_r020", + "bundle_neighbor_count_r040", + "bundle_neighbor_frac_r020", + "bundle_neighbor_frac_r040", + "bundle_mean_peer_rms", + "bundle_min_peer_rms", + "bundle_medoid_rms", + "bundle_is_medoid", + "bundle_peer_score_mean_r020", + "bundle_peer_score_max_r020", + "bundle_peer_score_mean_r040", + "bundle_peer_score_max_r040", + "bundle_unique_source_count_r040", + "bundle_unique_source_frac_r040", + "bundle_unique_task_count_r040", + "bundle_same_source_frac_r040", + "bundle_score_density_r040", + "bundle_rank_density_r040" + ], + "feature_set": "chart_bundle_consensus", + "feature_std": [ + 1.0, + 2.5232884736710055, + 2.2444575887945213, + 0.9999992484766103, + 4.542613580634851, + 0.24891036261589178, + 1.9642314545113713, + 0.37078681697003224, + 0.7171840796063713, + 1.7485561930736828, + 1e-06, + 0.14319094131293478, + 0.14491309378768996, + 0.19298897178584146, + 0.32495613500512616, + 1.3413619446954763, + 0.2065176097090355, + 0.3376678168884184, + 0.1766334065498482, + 0.1790096989966568, + 0.20687067948130008, + 0.3907071033474156, + 1e-06, + 0.005866370796541541, + 0.02376663855951178, + 0.015525196749693172, + 0.017227249939028074, + 1e-06, + 1e-06, + 1e-06, + 1e-06, + 1e-06, + 0.12987037363894735, + 4.878517299430629, + 3.217377946148024, + 0.346567593200359, + 0.19990387562321757, + 0.09641872137117102, + 0.07067998005028978, + 0.11663352509384882, + 0.2489103626158918, + 2.8138515595422326, + 2.965671838859836, + 2.4297066599961332, + 2.38657189782847, + 2.243933495797979, + 0.19723478821626303, + 1e-06, + 0.28108282108993726, + 2.4172277778814877, + 0.31354917609205285 + ], + "fit_design": { + "fit_objective": "pointwise", + "num_candidate_rows": 6510, + "num_fit_rows": 6510, + "num_pairwise_rows": 0, + "num_pointwise_rows": 6510, + "pairwise_weight": 0.0 + }, + "fit_objective": "pointwise", + "k": 16, + "num_calibration_candidates": 6510, + "num_calibration_rows": 432, + "num_eval_candidates": 2115, + "num_eval_rows": 144, + "pairwise_causal_calibration": { + "calibration": { + "accuracy": 0.506767075675391, + "bins": [ + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.0, + "upper": 0.1 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.1, + "upper": 0.2 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.2, + "upper": 0.3 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.3, + "upper": 0.4 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.4, + "upper": 0.5 + }, + { + "abs_gap": 0.027608275119104242, + "accuracy": 0.5067967933077727, + "confidence": 0.534405068426877, + "count": 34428, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.11624444768750941, + "accuracy": 0.511539585159217, + "confidence": 0.6277840328467265, + "count": 3423, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.3573191548504935, + "accuracy": 0.3700787401574803, + "confidence": 0.7273978950079738, + "count": 127, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.8, + "upper": 0.9 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.9, + "upper": 1.0 + } + ], + "ece": 0.03669972017749429, + "mean_confidence": 0.5434667958528854, + "n_bins": 10, + "num_pairs": 37978, + "num_rows": 432 + }, + "eval": { + "accuracy": 0.5160045037799582, + "bins": [ + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.0, + "upper": 0.1 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.1, + "upper": 0.2 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.2, + "upper": 0.3 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.3, + "upper": 0.4 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.4, + "upper": 0.5 + }, + { + "abs_gap": 0.018781001293858512, + "accuracy": 0.5147045913440436, + "confidence": 0.5334855926379021, + "count": 11391, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.09843975341018307, + "accuracy": 0.529126213592233, + "confidence": 0.6275659670024161, + "count": 1030, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.09231089848935525, + "accuracy": 0.6153846153846154, + "confidence": 0.7076955138739707, + "count": 13, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.8, + "upper": 0.9 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.9, + "upper": 1.0 + } + ], + "ece": 0.025456600726330422, + "mean_confidence": 0.5414611045062885, + "n_bins": 10, + "num_pairs": 12434, + "num_rows": 144 + } + }, + "pairwise_weight": 1.0, + "report_type": "learned_dominance_selector_eval", + "ridge_lambdas": [ + 0.0, + 0.01, + 0.1, + 1.0, + 10.0, + 100.0 + ], + "rows": [ + { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "chart_id": "ms-5799584ed40e211c50a5", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.226890756302521, + "pairwise_causal_calibration_confidence": 0.5363365754560178, + "pairwise_causal_calibration_ece": 0.3094458191534968, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.3161539044218536, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train0", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6949572563171387, + "selected_utility_gain_over_base": -0.028648078441619873, + "selector_regret": 1.3050427436828613, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6153846153846154, + "pairwise_causal_calibration_confidence": 0.5488728190719444, + "pairwise_causal_calibration_ece": 0.06651179631267089, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": -0.14233288142692663, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2115575522184372, + "seed": "307068879/train0", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7884424477815628, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7b229b60d7f64fb624cf", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5362042832267765, + "pairwise_causal_calibration_ece": 0.07639944182594278, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.24085858055337617, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.67014479637146, + "selected_utility_gain_over_base": -1.32985520362854, + "selector_regret": 1.32985520362854, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5619047619047619, + "pairwise_causal_calibration_confidence": 0.5498437058747719, + "pairwise_causal_calibration_ece": 0.01206105602998994, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.48611772128013164, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/train0", + "selected_candidate_index": 12, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.41818181818181815, + "pairwise_causal_calibration_confidence": 0.5379829028529084, + "pairwise_causal_calibration_ece": 0.11980108467109023, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.024167399539588352, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "441046872/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.3000448346138, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5681060848637505, + "pairwise_causal_calibration_ece": 0.07816118582099628, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.4821765811639407, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18325337767601013, + "seed": "1522272686/train0", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8167466223239899, + "task_id": "PickCube-v1", + "tau": 0.3611323919348086, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-06babf453b22ba8f2407", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.2685368177679842, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/train0", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.16168592870235443, + "chart_id": "ms-0e1ff8d16372796dac97", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.39316239316239315, + "pairwise_causal_calibration_confidence": 0.5290202473461273, + "pairwise_causal_calibration_ece": 0.13585785418373414, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.11402875379307875, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.025471270084381104, + "selected_utility_gain_over_base": -0.13621465861797333, + "selector_regret": 1.974528729915619, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.275, + "pairwise_causal_calibration_confidence": 0.5282766199734893, + "pairwise_causal_calibration_ece": 0.25327661997348927, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3059149126148908, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06019836664199829, + "seed": "1461069713/train0", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.011984944343566895, + "selected_utility_gain_over_base": -0.08191812038421631, + "selector_regret": 0.048213422298431396, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0991440862417221, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.001929501653648913, + "chart_id": "ms-34726202b475d29cdf38", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.25342071056365967, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.725, + "pairwise_causal_calibration_confidence": 0.5861390643686207, + "pairwise_causal_calibration_ece": 0.13886093563137974, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.780703524543997, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.010183374397456646, + "seed": "626486813/train0", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007048348430544138, + "selected_utility_gain_over_base": 0.005118846776895225, + "selector_regret": 0.003135025966912508, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24323733616620302, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-4cc40587ec490257b1d2", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.30833333333333335, + "pairwise_causal_calibration_confidence": 0.5238354835309519, + "pairwise_causal_calibration_ece": 0.21550215019761854, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.18790832822817682, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15456630289554596, + "seed": "2054899126/train0", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07831952720880508, + "selected_utility_gain_over_base": -1.921680472791195, + "selector_regret": 0.07624677568674088, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.845433697104454, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-62f8c18a15c566ec4b00", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7647058823529411, + "pairwise_causal_calibration_confidence": 0.542991093593322, + "pairwise_causal_calibration_ece": 0.22171478875961914, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": -0.14745117870449231, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.45, + "pairwise_causal_calibration_confidence": 0.5719337795347729, + "pairwise_causal_calibration_ece": 0.12193377953477302, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3049399036443091, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15051491558551788, + "seed": "767747424/train0", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08540759235620499, + "selected_utility_gain_over_base": 0.00841909646987915, + "selector_regret": 0.0651073232293129, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08458571135997772, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.03588630631566048, + "chart_id": "ms-6b789356d65b3312fbae", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22781728208065033, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6833333333333333, + "pairwise_causal_calibration_confidence": 0.5568021501755852, + "pairwise_causal_calibration_ece": 0.12653118315774806, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.45248840961956327, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14671264588832855, + "seed": "278008524/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05741874501109123, + "selected_utility_gain_over_base": 0.021532438695430756, + "selector_regret": 0.08929390087723732, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08110463619232178, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5869565217391305, + "pairwise_causal_calibration_confidence": 0.5198092453508348, + "pairwise_causal_calibration_ece": 0.06714727638829565, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": 0.08565911209810279, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1845489013/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.008267183788120747, + "chart_id": "ms-7df455efd413d77e6ce3", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.07774405181407928, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5494843505241936, + "pairwise_causal_calibration_ece": 0.049484350524193546, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.36207319396878956, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07920757681131363, + "seed": "626486813/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.012751738540828228, + "selected_utility_gain_over_base": 0.004484554752707481, + "selector_regret": 0.0664558382704854, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7e295b5859c07c90487f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.26271408796310425, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5297418005916671, + "pairwise_causal_calibration_ece": 0.03692486607499967, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.26665208050643885, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "chart_id": "ms-8577816da2917a2bb63d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4523809523809524, + "pairwise_causal_calibration_confidence": 0.5292002900283871, + "pairwise_causal_calibration_ece": 0.0768193376474347, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.23094602840169032, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "585904130/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09919407218694687, + "selected_utility_gain_over_base": -0.0825214758515358, + "selector_regret": 1.9008059278130531, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "chart_id": "ms-8e56b0eaa2604db6749e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.49166666666666664, + "pairwise_causal_calibration_confidence": 0.5402158636214657, + "pairwise_causal_calibration_ece": 0.048549196954798944, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.30976598855930915, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "seed": "141984615/train0", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.00942655373364687, + "selected_utility_gain_over_base": -0.09340769331902266, + "selector_regret": 0.005751351825892925, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.14270895719528198, + "chart_id": "ms-972c5be72d0b24be7cd5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5245009617743884, + "pairwise_causal_calibration_ece": 0.04216570489227822, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.07070455387890318, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21658658981323242, + "seed": "750448296/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14270895719528198, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.07387763261795044, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7834134101867676, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "chart_id": "ms-9757fc494e79b5543bb6", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.49166666666666664, + "pairwise_causal_calibration_confidence": 0.532502870801132, + "pairwise_causal_calibration_ece": 0.04083620413446535, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5991343617786198, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.08826130628585815, + "seed": "636680137/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08826130628585815, + "selected_utility_gain_over_base": 0.005126260221004486, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0823030173778534, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.05412294343113899, + "chart_id": "ms-a8731059713697ce5e2a", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.19710752367973328, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5539312638819843, + "pairwise_causal_calibration_ece": 0.13726459721531775, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.26132902007960485, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.019201835617423058, + "seed": "767747424/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007447937969118357, + "selected_utility_gain_over_base": -0.046675005462020636, + "selector_regret": 0.0117538976483047, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17790568806231022, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-ac94b48febe523da15a5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6533333333333333, + "pairwise_causal_calibration_confidence": 0.5447046728594015, + "pairwise_causal_calibration_ece": 0.10862866047393183, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": -0.28428690828432274, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b49043f15f4846de5b13", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3793103448275862, + "pairwise_causal_calibration_confidence": 0.6018790607135653, + "pairwise_causal_calibration_ece": 0.22256871588597915, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.28192292963769827, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04285776615142822, + "selected_utility_gain_over_base": -1.9571422338485718, + "selector_regret": 1.9571422338485718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.0929974913597107, + "chart_id": "ms-b7d7021b782f73816083", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22398948669433594, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.475, + "pairwise_causal_calibration_confidence": 0.5286800216420143, + "pairwise_causal_calibration_ece": 0.05368002164201435, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6038009462868253, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.041394393891096115, + "seed": "750448296/train0", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.015753567218780518, + "selected_utility_gain_over_base": -0.07724392414093018, + "selector_regret": 0.025640826672315598, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.18259509280323982, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.613885358282616, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train0", + "selected_candidate_index": 6, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8fc1568ce66c45c8d8b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4583333333333333, + "pairwise_causal_calibration_confidence": 0.5237942697642416, + "pairwise_causal_calibration_ece": 0.06546093643090828, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.03550968246605479, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10839563608169556, + "seed": "747260437/train0", + "selected_candidate_index": 15, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17026761174201965, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.04958367347717285, + "chart_id": "ms-da555a03dd382d93e119", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3838383838383838, + "pairwise_causal_calibration_confidence": 0.5387657741422159, + "pairwise_causal_calibration_ece": 0.15492739030383204, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.12355403937307244, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07767494767904282, + "selected_utility_gain_over_base": 0.028091274201869965, + "selector_regret": 1.9223250523209572, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "chart_id": "ms-f5dd1c43ebc4c0cfc3f1", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7083333333333334, + "pairwise_causal_calibration_confidence": 0.5299868735462288, + "pairwise_causal_calibration_ece": 0.17834645978710464, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8008610893558498, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.060871422290802, + "seed": "1549324805/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.03462250903248787, + "selected_utility_gain_over_base": -0.06989232823252678, + "selector_regret": 0.026248913258314133, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.23658466339111328, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "chart_id": "ms-f9740a08a98cb251b761", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.575, + "pairwise_causal_calibration_confidence": 0.5565662285154201, + "pairwise_causal_calibration_ece": 0.15721195241478056, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 0.013297597406553785, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07191530615091324, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9280846938490868, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5238095238095238, + "pairwise_causal_calibration_confidence": 0.5289802004330936, + "pairwise_causal_calibration_ece": 0.0051706766235697366, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.5084711391061645, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "chart_id": "ms-05f34a90fa7355262b8e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7586206896551724, + "pairwise_causal_calibration_confidence": 0.5267661600012826, + "pairwise_causal_calibration_ece": 0.23185452965388975, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.7982832696690534, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train0", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "chart_id": "ms-2413d1155c87f0442485", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5337665023974802, + "pairwise_causal_calibration_ece": 0.032055347656845035, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.711002878905307, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14406004548072815, + "seed": "1324642411/train0", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12164360284805298, + "selected_utility_gain_over_base": 0.015173077583312988, + "selector_regret": 0.02241644263267517, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8559399545192719, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "chart_id": "ms-54346a8f79bd9dca1dfe", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6363636363636364, + "pairwise_causal_calibration_confidence": 0.5392236442621141, + "pairwise_causal_calibration_ece": 0.09713999210152216, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.4590932780625647, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.87523253262043, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-64f98d163b6bff29332b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.47474747474747475, + "pairwise_causal_calibration_confidence": 0.5407472634490431, + "pairwise_causal_calibration_ece": 0.06599978870156839, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.09035295152992723, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train0", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "chart_id": "ms-c2ae2b78ab7891e74c80", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6050420168067226, + "pairwise_causal_calibration_confidence": 0.5254427085184519, + "pairwise_causal_calibration_ece": 0.07959930828827078, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5885139744079059, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/train0", + "selected_candidate_index": 11, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "chart_id": "ms-d4c37ed44fb338c1d551", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.47474747474747475, + "pairwise_causal_calibration_confidence": 0.5212194722509627, + "pairwise_causal_calibration_ece": 0.046471997503487905, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.6891134156079126, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train0", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07608599960803986, + "selected_utility_gain_over_base": -0.010948076844215393, + "selector_regret": 1.9239140003919601, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "chart_id": "ms-146ff6c3f22b3a49af3a", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5833333333333334, + "pairwise_causal_calibration_confidence": 0.5351503713504386, + "pairwise_causal_calibration_ece": 0.048182961982894734, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.2859560598368813, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09799930453300476, + "seed": "2061635888/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07349006831645966, + "selected_utility_gain_over_base": 0.021239563822746277, + "selector_regret": 0.024509236216545105, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9020006954669952, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "chart_id": "ms-15fc8cc6f74cf589056f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.568990751323036, + "pairwise_causal_calibration_ece": 0.09767591534363074, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.4522143975120525, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13732868432998657, + "seed": "655884605/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12222731113433838, + "selected_utility_gain_over_base": -0.040779367089271545, + "selector_regret": 0.015101373195648193, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8626713156700134, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "chart_id": "ms-36bc1364d2439099e9ed", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5401592730365951, + "pairwise_causal_calibration_ece": 0.04015927303659509, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.002655981453428158, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10274861752986908, + "seed": "1113371261/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.897251382470131, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "chart_id": "ms-420280f15a522ce21c72", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6153846153846154, + "pairwise_causal_calibration_confidence": 0.544349322443362, + "pairwise_causal_calibration_ece": 0.07103529294125341, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.10915469051695448, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14271311461925507, + "selected_utility_gain_over_base": -0.6814233809709549, + "selector_regret": 1.857286885380745, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "chart_id": "ms-4aed1197d0906ae51c34", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5320927425921708, + "pairwise_causal_calibration_ece": 0.07422005046880752, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.2172662931516871, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7058823529411765, + "pairwise_causal_calibration_confidence": 0.5545781832114899, + "pairwise_causal_calibration_ece": 0.1513041697296866, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.05542394657973021, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "211245842/train0", + "selected_candidate_index": 11, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8090915828943253, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "chart_id": "ms-77536801924a2a78688e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5047619047619047, + "pairwise_causal_calibration_confidence": 0.5241846653555288, + "pairwise_causal_calibration_ece": 0.019422760593624022, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.11341562423740303, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8663521409034729, + "seed": "1151410346/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16260439157485962, + "selected_utility_gain_over_base": -0.06339690089225769, + "selector_regret": 0.7037477493286133, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.133647859096527, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "chart_id": "ms-9dd1ff4c297a81b13071", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5500982965860671, + "pairwise_causal_calibration_ece": 0.08581258230035274, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.07548993896412504, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.054989203810691833, + "seed": "1324133122/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.036974772810935974, + "selected_utility_gain_over_base": -0.055230140686035156, + "selector_regret": 0.01801443099975586, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9450107961893082, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.40476190476190477, + "pairwise_causal_calibration_confidence": 0.5617516698950917, + "pairwise_causal_calibration_ece": 0.31044146560402797, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 0.2186075457961687, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/train0", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d776afa14e6322cadedd", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5413119316483767, + "pairwise_causal_calibration_ece": 0.007978598315043392, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": -0.07414481146410684, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/train0", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-e887dbfa0133af46829d", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3, + "pairwise_causal_calibration_confidence": 0.5344493626989616, + "pairwise_causal_calibration_ece": 0.23444936269896155, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": -0.11161114265900707, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "chart_id": "ms-5799584ed40e211c50a5", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5335084608302744, + "pairwise_causal_calibration_ece": 0.07199480463689167, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.24432178181307274, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train1", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6900332570075989, + "selected_utility_gain_over_base": -0.03357207775115967, + "selector_regret": 1.3099667429924011, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5934065934065934, + "pairwise_causal_calibration_confidence": 0.5348399847973542, + "pairwise_causal_calibration_ece": 0.06342054136618626, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": -0.13522650945094028, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5876073241233826, + "seed": "307068879/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7b229b60d7f64fb624cf", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5416666666666666, + "pairwise_causal_calibration_confidence": 0.5477005041063188, + "pairwise_causal_calibration_ece": 0.12867242656969963, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.11864476132183588, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/train1", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.6947963237762451, + "selected_utility_gain_over_base": -1.3052036762237549, + "selector_regret": 1.3052036762237549, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.47619047619047616, + "pairwise_causal_calibration_confidence": 0.5706245241886934, + "pairwise_causal_calibration_ece": 0.09443404799821745, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.2549086170487004, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/train1", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7062739729881287, + "selected_utility_gain_over_base": -0.01942765712738037, + "selector_regret": 1.2937260270118713, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5818181818181818, + "pairwise_causal_calibration_confidence": 0.5337021369933691, + "pairwise_causal_calibration_ece": 0.04811604482481282, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.11694270182946509, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7233588695526123, + "seed": "441046872/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.023403704166412354, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.19444444444444445, + "pairwise_causal_calibration_confidence": 0.5689701709118037, + "pairwise_causal_calibration_ece": 0.37452572646735927, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.25520062215278894, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07014383375644684, + "seed": "1522272686/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9298561662435532, + "task_id": "PickCube-v1", + "tau": 0.3611323919348086, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-06babf453b22ba8f2407", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.27732487886731, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/train1", + "selected_candidate_index": 7, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.16168592870235443, + "chart_id": "ms-0e1ff8d16372796dac97", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42016806722689076, + "pairwise_causal_calibration_confidence": 0.5409061540955996, + "pairwise_causal_calibration_ece": 0.12073808686870902, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.008845977198187305, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train1", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16168592870235443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8383140712976456, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.43333333333333335, + "pairwise_causal_calibration_confidence": 0.5243338816656252, + "pairwise_causal_calibration_ece": 0.09100054833229188, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3007404756181269, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "seed": "1461069713/train1", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007486680988222361, + "selected_utility_gain_over_base": -0.08641638373956084, + "selector_regret": 0.009098331909626722, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14275743998587132, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.001929501653648913, + "chart_id": "ms-34726202b475d29cdf38", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.25342071056365967, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.35, + "pairwise_causal_calibration_confidence": 0.5844898439579203, + "pairwise_causal_calibration_ece": 0.2859099968854081, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5953498952583893, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.009895960800349712, + "seed": "626486813/train1", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.003232975723221898, + "selected_utility_gain_over_base": 0.0013034740695729852, + "selector_regret": 0.006662985077127814, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24352474976330996, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-4cc40587ec490257b1d2", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6052631578947368, + "pairwise_causal_calibration_confidence": 0.5244611135518833, + "pairwise_causal_calibration_ece": 0.08080204434285354, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 0.049921997064514445, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train1", + "selected_candidate_index": 14, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-62f8c18a15c566ec4b00", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6571428571428571, + "pairwise_causal_calibration_confidence": 0.5517445085212526, + "pairwise_causal_calibration_ece": 0.10539834862160463, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": -0.21177052566683674, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train1", + "selected_candidate_index": 4, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5916666666666667, + "pairwise_causal_calibration_confidence": 0.5740013522613887, + "pairwise_causal_calibration_ece": 0.06212237425755651, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3629064243590099, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.20903043448925018, + "seed": "767747424/train1", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11253257840871811, + "selected_utility_gain_over_base": 0.03554408252239227, + "selector_regret": 0.09649785608053207, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.026070192456245422, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.03588630631566048, + "chart_id": "ms-6b789356d65b3312fbae", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22781728208065033, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.44166666666666665, + "pairwise_causal_calibration_confidence": 0.5528011474359009, + "pairwise_causal_calibration_ece": 0.11113448076923427, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6558361682983371, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07534918934106827, + "seed": "278008524/train1", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.01919553615152836, + "selected_utility_gain_over_base": -0.016690770164132118, + "selector_regret": 0.05615365318953991, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15246809273958206, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7692307692307693, + "pairwise_causal_calibration_confidence": 0.519320720354459, + "pairwise_causal_calibration_ece": 0.2499100488763103, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": 0.10538760252719387, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1845489013/train1", + "selected_candidate_index": 7, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.008267183788120747, + "chart_id": "ms-7df455efd413d77e6ce3", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.07774405181407928, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.43333333333333335, + "pairwise_causal_calibration_confidence": 0.5446342348545582, + "pairwise_causal_calibration_ece": 0.11130090152122507, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5162704610434539, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.061759788542985916, + "seed": "626486813/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.019184648990631104, + "selected_utility_gain_over_base": 0.010917465202510357, + "selector_regret": 0.04257513955235481, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.01598426327109337, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7e295b5859c07c90487f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.26271408796310425, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5213675213675214, + "pairwise_causal_calibration_confidence": 0.5264743811072923, + "pairwise_causal_calibration_ece": 0.005106859739770875, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.006995236123196025, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train1", + "selected_candidate_index": 12, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "chart_id": "ms-8577816da2917a2bb63d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5466666666666666, + "pairwise_causal_calibration_confidence": 0.5343887712270259, + "pairwise_causal_calibration_ece": 0.050834006610117985, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.22575925280556675, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "585904130/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04792279005050659, + "selected_utility_gain_over_base": -0.13379275798797607, + "selector_regret": 1.9520772099494934, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "chart_id": "ms-8e56b0eaa2604db6749e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5473004480108934, + "pairwise_causal_calibration_ece": 0.046271089584524104, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.39033675889099473, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.037011127918958664, + "seed": "141984615/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.030119141563773155, + "selected_utility_gain_over_base": -0.07271510548889637, + "selector_regret": 0.006891986355185509, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24876614287495613, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.14270895719528198, + "chart_id": "ms-972c5be72d0b24be7cd5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.532366840305153, + "pairwise_causal_calibration_ece": 0.11570017363848628, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.12459883652924558, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14270895719528198, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.857291042804718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "chart_id": "ms-9757fc494e79b5543bb6", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5173967032259332, + "pairwise_causal_calibration_ece": 0.015936630107400096, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.19155725528837464, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06741118431091309, + "seed": "636680137/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.017789265140891075, + "selected_utility_gain_over_base": -0.0653457809239626, + "selector_regret": 0.04962191917002201, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10315313935279846, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.05412294343113899, + "chart_id": "ms-a8731059713697ce5e2a", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.19710752367973328, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5707296437991907, + "pairwise_causal_calibration_ece": 0.037396310465857494, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5964692231627363, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.042781930416822433, + "seed": "767747424/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.006599446292966604, + "selected_utility_gain_over_base": -0.04752349713817239, + "selector_regret": 0.03618248412385583, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15432559326291084, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-ac94b48febe523da15a5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5614035087719298, + "pairwise_causal_calibration_confidence": 0.5446679554502398, + "pairwise_causal_calibration_ece": 0.01673555332169, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.6782841630171567, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/train1", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b49043f15f4846de5b13", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.42592592592592593, + "pairwise_causal_calibration_confidence": 0.5909326842120879, + "pairwise_causal_calibration_ece": 0.1851395001152688, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.21139315187633212, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07674666494131088, + "selected_utility_gain_over_base": -1.9232533350586891, + "selector_regret": 1.9232533350586891, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.0929974913597107, + "chart_id": "ms-b7d7021b782f73816083", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22398948669433594, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.25833333333333336, + "pairwise_causal_calibration_confidence": 0.5321487037817768, + "pairwise_causal_calibration_ece": 0.27381537044844345, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3642338419356729, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11150982230901718, + "seed": "750448296/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06907258182764053, + "selected_utility_gain_over_base": -0.02392490953207016, + "selector_regret": 0.04243724048137665, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.11247966438531876, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.5975395941482049, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train1", + "selected_candidate_index": 6, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8fc1568ce66c45c8d8b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5173328988662427, + "pairwise_causal_calibration_ece": 0.04933376780042398, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.27755861197582443, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06328952312469482, + "seed": "747260437/train1", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.21537372469902039, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.04958367347717285, + "chart_id": "ms-da555a03dd382d93e119", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6195652173913043, + "pairwise_causal_calibration_confidence": 0.5430261259354103, + "pairwise_causal_calibration_ece": 0.07653909145589391, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.0643969314465564, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train1", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04958367347717285, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9504163265228271, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "chart_id": "ms-f5dd1c43ebc4c0cfc3f1", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4083333333333333, + "pairwise_causal_calibration_confidence": 0.5265093687069327, + "pairwise_causal_calibration_ece": 0.11817603537359933, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4059182168351142, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "seed": "1549324805/train1", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08073369413614273, + "selected_utility_gain_over_base": -0.023781143128871918, + "selector_regret": 0.11818782240152359, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "chart_id": "ms-f9740a08a98cb251b761", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5510204081632653, + "pairwise_causal_calibration_confidence": 0.5508218552014704, + "pairwise_causal_calibration_ece": 0.14397334425216246, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": 0.041781127110214866, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07191530615091324, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9280846938490868, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.38461538461538464, + "pairwise_causal_calibration_confidence": 0.5545436137717367, + "pairwise_causal_calibration_ece": 0.16992822915635225, + "pairwise_causal_calibration_pairs": 65.0, + "predicted_margin": -0.3332857847632774, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "chart_id": "ms-05f34a90fa7355262b8e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.5289194284606322, + "pairwise_causal_calibration_ece": 0.07108057153936775, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 0.862893346101075, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train1", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "chart_id": "ms-2413d1155c87f0442485", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5798319327731093, + "pairwise_causal_calibration_confidence": 0.5288122689723307, + "pairwise_causal_calibration_ece": 0.0510196638007786, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.38447100568728765, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.1378108412027359, + "selected_utility_gain_over_base": 0.03134031593799591, + "selector_regret": 1.862189158797264, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "chart_id": "ms-54346a8f79bd9dca1dfe", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5350877192982456, + "pairwise_causal_calibration_confidence": 0.5396885778611449, + "pairwise_causal_calibration_ece": 0.004600858562899268, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 0.33925681284155934, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12476746737957001, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.87523253262043, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-64f98d163b6bff29332b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6181818181818182, + "pairwise_causal_calibration_confidence": 0.5491787800871848, + "pairwise_causal_calibration_ece": 0.0690030380946333, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": -0.02356230018007266, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train1", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "chart_id": "ms-c2ae2b78ab7891e74c80", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5904761904761905, + "pairwise_causal_calibration_confidence": 0.5382862696863612, + "pairwise_causal_calibration_ece": 0.052189920789829286, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.5296709902746183, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/train1", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "chart_id": "ms-d4c37ed44fb338c1d551", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.48739495798319327, + "pairwise_causal_calibration_confidence": 0.5158258523764001, + "pairwise_causal_calibration_ece": 0.02843089439320684, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5636198101475918, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21981161832809448, + "selected_utility_gain_over_base": 0.13277754187583923, + "selector_regret": 1.7801883816719055, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "chart_id": "ms-146ff6c3f22b3a49af3a", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.518040614785329, + "pairwise_causal_calibration_ece": 0.018040614785329034, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.14710078956282402, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07580336928367615, + "seed": "2061635888/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0713038444519043, + "selected_utility_gain_over_base": 0.019053339958190918, + "selector_regret": 0.004499524831771851, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9241966307163239, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "chart_id": "ms-15fc8cc6f74cf589056f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5500689384066487, + "pairwise_causal_calibration_ece": 0.07784671618442661, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.446772375365045, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8128758072853088, + "seed": "655884605/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11176733672618866, + "selected_utility_gain_over_base": -0.051239341497421265, + "selector_regret": 0.7011084705591202, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.1871241927146912, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "chart_id": "ms-36bc1364d2439099e9ed", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5380625270265227, + "pairwise_causal_calibration_ece": 0.038062527026522686, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.08193066148519779, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18093222379684448, + "seed": "1113371261/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.1746392846107483, + "selected_utility_gain_over_base": 0.016215860843658447, + "selector_regret": 0.006292939186096191, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8190677762031555, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "chart_id": "ms-420280f15a522ce21c72", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4727272727272727, + "pairwise_causal_calibration_confidence": 0.5543665049371578, + "pairwise_causal_calibration_ece": 0.081639232209885, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.10231812655593348, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train1", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.17168906331062317, + "selected_utility_gain_over_base": -0.6524474322795868, + "selector_regret": 1.8283109366893768, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "chart_id": "ms-4aed1197d0906ae51c34", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.2222222222222222, + "pairwise_causal_calibration_confidence": 0.5391700207769446, + "pairwise_causal_calibration_ece": 0.31694779855472244, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": -0.2497394332434142, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train1", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7916666666666666, + "pairwise_causal_calibration_confidence": 0.5498984873423876, + "pairwise_causal_calibration_ece": 0.2417681793242794, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2276424215366616, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "seed": "211245842/train1", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19474327564239502, + "selected_utility_gain_over_base": 0.003834858536720276, + "selector_regret": 0.018213897943496704, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7870428264141083, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "chart_id": "ms-77536801924a2a78688e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5513917700725205, + "pairwise_causal_calibration_ece": 0.08472510340585364, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.027308203722147817, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1151410346/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8135483264923096, + "selected_utility_gain_over_base": 0.5875470340251923, + "selector_regret": 1.1864516735076904, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "chart_id": "ms-9dd1ff4c297a81b13071", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5502823354081949, + "pairwise_causal_calibration_ece": 0.015677218286420276, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.3962004123380809, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1324586570262909, + "seed": "1324133122/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07714252173900604, + "selected_utility_gain_over_base": -0.015062391757965088, + "selector_regret": 0.05531613528728485, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.867541342973709, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.546009069568378, + "pairwise_causal_calibration_ece": 0.13256235900305033, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.20451752246432947, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/train1", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d776afa14e6322cadedd", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5384615384615384, + "pairwise_causal_calibration_confidence": 0.5307516643020463, + "pairwise_causal_calibration_ece": 0.007709874159492092, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": -0.10576298261783212, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/train1", + "selected_candidate_index": 12, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-e887dbfa0133af46829d", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4714285714285714, + "pairwise_causal_calibration_confidence": 0.5395820832404352, + "pairwise_causal_calibration_ece": 0.06815351181186355, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": -0.1454763298079958, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train1", + "selected_candidate_index": 8, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "chart_id": "ms-5799584ed40e211c50a5", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6086956521739131, + "pairwise_causal_calibration_confidence": 0.5293148633692549, + "pairwise_causal_calibration_ece": 0.07938078880465815, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.3345646470523239, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train2", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6934840679168701, + "selected_utility_gain_over_base": -0.030121266841888428, + "selector_regret": 1.3065159320831299, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4945054945054945, + "pairwise_causal_calibration_confidence": 0.5640014996133708, + "pairwise_causal_calibration_ece": 0.06949600510787626, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": -0.02519869144581416, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2302134484052658, + "seed": "307068879/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7697865515947342, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7b229b60d7f64fb624cf", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.8095238095238095, + "pairwise_causal_calibration_confidence": 0.5547882621085445, + "pairwise_causal_calibration_ece": 0.25473554741526483, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.24206048374838196, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/train2", + "selected_candidate_index": 6, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.48333333333333334, + "pairwise_causal_calibration_confidence": 0.5615509166352838, + "pairwise_causal_calibration_ece": 0.0782175833019504, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4202802138182712, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7117276191711426, + "seed": "608016290/train2", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.22811375558376312, + "selected_utility_gain_over_base": -0.4975878745317459, + "selector_regret": 0.48361386358737946, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7272727272727273, + "pairwise_causal_calibration_confidence": 0.5454083397120673, + "pairwise_causal_calibration_ece": 0.18186438756066006, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.12738205708571582, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7191348671913147, + "seed": "441046872/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019179701805114746, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "task_id": "LiftPegUpright-v1", + "tau": 0.0940318405174132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.545306820402758, + "pairwise_causal_calibration_ece": 0.04530682040275796, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.43379366859847046, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14937306940555573, + "seed": "1522272686/train2", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8506269305944443, + "task_id": "PickCube-v1", + "tau": 0.3611323919348086, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-06babf453b22ba8f2407", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.3315076869895872, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/train2", + "selected_candidate_index": 11, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.16168592870235443, + "chart_id": "ms-0e1ff8d16372796dac97", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5126050420168067, + "pairwise_causal_calibration_confidence": 0.5335372983831308, + "pairwise_causal_calibration_ece": 0.02093225636632412, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.08505583152382623, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16168592870235443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8383140712976456, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6583333333333333, + "pairwise_causal_calibration_confidence": 0.5225830368338087, + "pairwise_causal_calibration_ece": 0.13575029649952464, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2553375780680003, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "seed": "1461069713/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09997648000717163, + "selected_utility_gain_over_base": 0.006073415279388428, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05936597287654877, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.001929501653648913, + "chart_id": "ms-34726202b475d29cdf38", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.25342071056365967, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.5750839572021335, + "pairwise_causal_calibration_ece": 0.04158270946453317, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4690776598726668, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.03885193541646004, + "seed": "626486813/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.02248450182378292, + "selected_utility_gain_over_base": 0.020555000170134008, + "selector_regret": 0.016367433592677116, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.21456877514719963, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-4cc40587ec490257b1d2", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.47058823529411764, + "pairwise_causal_calibration_confidence": 0.5297648676632524, + "pairwise_causal_calibration_ece": 0.05917663236913473, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.00965012140270699, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "selected_candidate_index": 11, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-62f8c18a15c566ec4b00", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.16666666666666666, + "pairwise_causal_calibration_confidence": 0.5462121445015077, + "pairwise_causal_calibration_ece": 0.37954547783484094, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": -0.19816182889430234, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train2", + "selected_candidate_index": 4, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.25, + "pairwise_causal_calibration_confidence": 0.5592804818587975, + "pairwise_causal_calibration_ece": 0.30928048185879753, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.1702189610695619, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21886132657527924, + "seed": "767747424/train2", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.026480933651328087, + "selected_utility_gain_over_base": -0.05050756223499775, + "selector_regret": 0.19238039292395115, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.01623930037021637, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.03588630631566048, + "chart_id": "ms-6b789356d65b3312fbae", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22781728208065033, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.38333333333333336, + "pairwise_causal_calibration_confidence": 0.5652706387896659, + "pairwise_causal_calibration_ece": 0.18193730545633274, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.41994817119419847, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2718978226184845, + "seed": "278008524/train2", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04554526135325432, + "selected_utility_gain_over_base": 0.009658955037593842, + "selector_regret": 0.22635256126523018, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.52, + "pairwise_causal_calibration_confidence": 0.5338919568560658, + "pairwise_causal_calibration_ece": 0.013891956856065812, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": 0.14874188314117123, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1845489013/train2", + "selected_candidate_index": 8, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.008267183788120747, + "chart_id": "ms-7df455efd413d77e6ce3", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.07774405181407928, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.43333333333333335, + "pairwise_causal_calibration_confidence": 0.5475369153231996, + "pairwise_causal_calibration_ece": 0.11420358198986626, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.30168313132887237, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05094657465815544, + "seed": "626486813/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.009525339119136333, + "selected_utility_gain_over_base": 0.0012581553310155869, + "selector_regret": 0.04142123553901911, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.026797477155923843, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7e295b5859c07c90487f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.26271408796310425, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4188034188034188, + "pairwise_causal_calibration_confidence": 0.5214197977010018, + "pairwise_causal_calibration_ece": 0.10261637889758296, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": -0.2624403378904797, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "selected_candidate_index": 8, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "chart_id": "ms-8577816da2917a2bb63d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7272727272727273, + "pairwise_causal_calibration_confidence": 0.5282487650578878, + "pairwise_causal_calibration_ece": 0.19902396221483953, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.18843222973041862, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "585904130/train2", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8182844519615173, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "chart_id": "ms-8e56b0eaa2604db6749e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5416666666666666, + "pairwise_causal_calibration_confidence": 0.5440555881176564, + "pairwise_causal_calibration_ece": 0.03902364539059017, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.29555407226838537, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058951716870069504, + "seed": "141984615/train2", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.058951716870069504, + "selected_utility_gain_over_base": -0.04388253018260002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.14270895719528198, + "chart_id": "ms-972c5be72d0b24be7cd5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.531606386735559, + "pairwise_causal_calibration_ece": 0.0350602799311076, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.008678557833403544, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train2", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14270895719528198, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.857291042804718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "chart_id": "ms-9757fc494e79b5543bb6", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6583333333333333, + "pairwise_causal_calibration_confidence": 0.5151979568732741, + "pairwise_causal_calibration_ece": 0.14313537646005925, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.27475298254295666, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0756516233086586, + "seed": "636680137/train2", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.039896368980407715, + "selected_utility_gain_over_base": -0.04323867708444595, + "selector_regret": 0.035755254328250885, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09491270035505295, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.05412294343113899, + "chart_id": "ms-a8731059713697ce5e2a", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.19710752367973328, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.48333333333333334, + "pairwise_causal_calibration_confidence": 0.5601528764543879, + "pairwise_causal_calibration_ece": 0.07768936366580693, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5596205731602828, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04980705305933952, + "seed": "767747424/train2", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.028320351615548134, + "selected_utility_gain_over_base": -0.02580259181559086, + "selector_regret": 0.02148670144379139, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14730047062039375, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-ac94b48febe523da15a5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7380952380952381, + "pairwise_causal_calibration_confidence": 0.565958547539462, + "pairwise_causal_calibration_ece": 0.1721366905557764, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.559803911595051, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/train2", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b49043f15f4846de5b13", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.41414141414141414, + "pairwise_causal_calibration_confidence": 0.5664818478408624, + "pairwise_causal_calibration_ece": 0.1523404336994486, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.2451404604489238, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.12977690994739532, + "selected_utility_gain_over_base": -1.8702230900526047, + "selector_regret": 1.8702230900526047, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.0929974913597107, + "chart_id": "ms-b7d7021b782f73816083", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22398948669433594, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.30833333333333335, + "pairwise_causal_calibration_confidence": 0.5224162480703396, + "pairwise_causal_calibration_ece": 0.21408291473700625, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.34844644909874656, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.051346004009246826, + "seed": "750448296/train2", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.035547394305467606, + "selected_utility_gain_over_base": -0.05745009705424309, + "selector_regret": 0.01579860970377922, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1726434826850891, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.6946523201836585, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8fc1568ce66c45c8d8b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.625, + "pairwise_causal_calibration_confidence": 0.5276406049368979, + "pairwise_causal_calibration_ece": 0.0973593950631021, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.061753054977374155, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.12510187923908234, + "seed": "747260437/train2", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15356136858463287, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.04958367347717285, + "chart_id": "ms-da555a03dd382d93e119", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.5403790398403705, + "pairwise_causal_calibration_ece": 0.2070457065070372, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.02036982454775252, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train2", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04958367347717285, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9504163265228271, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "chart_id": "ms-f5dd1c43ebc4c0cfc3f1", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.2833333333333333, + "pairwise_causal_calibration_confidence": 0.541054902787583, + "pairwise_causal_calibration_ece": 0.25772156945424957, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8716902723485557, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11310184001922607, + "seed": "1549324805/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04690708592534065, + "selected_utility_gain_over_base": -0.057607751339673996, + "selector_regret": 0.06619475409388542, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1843542456626892, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "chart_id": "ms-f9740a08a98cb251b761", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.525, + "pairwise_causal_calibration_confidence": 0.5302663723099226, + "pairwise_causal_calibration_ece": 0.118390411016323, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": -0.09918008580701754, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07191530615091324, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9280846938490868, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5300035557325993, + "pairwise_causal_calibration_ece": 0.07503274438189782, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.5452751058050467, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.10332537601561143, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "chart_id": "ms-05f34a90fa7355262b8e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6111111111111112, + "pairwise_causal_calibration_confidence": 0.5489590881909069, + "pairwise_causal_calibration_ece": 0.06215202292020417, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.7945693500962356, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train2", + "selected_candidate_index": 13, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "chart_id": "ms-2413d1155c87f0442485", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7, + "pairwise_causal_calibration_confidence": 0.528083048643476, + "pairwise_causal_calibration_ece": 0.17191695135652396, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.15208916465183764, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/train2", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10647052526473999, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.89352947473526, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "chart_id": "ms-54346a8f79bd9dca1dfe", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.44545454545454544, + "pairwise_causal_calibration_confidence": 0.5413352723129059, + "pairwise_causal_calibration_ece": 0.09588072685836047, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.297516022662714, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train2", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12476746737957001, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.87523253262043, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-64f98d163b6bff29332b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5397391671472074, + "pairwise_causal_calibration_ece": 0.12692749951945942, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.04072195094915404, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train2", + "selected_candidate_index": 4, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "chart_id": "ms-c2ae2b78ab7891e74c80", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.5437150122698521, + "pairwise_causal_calibration_ece": 0.06973036588140841, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4810616128884794, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 5.424022674560547e-06, + "seed": "509430392/train2", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 5.424022674560547e-06, + "selected_utility_gain_over_base": 5.200505256652832e-06, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9999945759773254, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "chart_id": "ms-d4c37ed44fb338c1d551", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.680672268907563, + "pairwise_causal_calibration_confidence": 0.5204768609791875, + "pairwise_causal_calibration_ece": 0.1601954079283755, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4905519223760793, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21615806221961975, + "selected_utility_gain_over_base": 0.1291239857673645, + "selector_regret": 1.7838419377803802, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.35241155802994517, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "chart_id": "ms-146ff6c3f22b3a49af3a", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5252975568525992, + "pairwise_causal_calibration_ece": 0.025297556852599157, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.15541062212467605, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11105132102966309, + "seed": "2061635888/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05500011146068573, + "selected_utility_gain_over_base": 0.002749606966972351, + "selector_regret": 0.056051209568977356, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.888948678970337, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "chart_id": "ms-15fc8cc6f74cf589056f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3888888888888889, + "pairwise_causal_calibration_confidence": 0.556089617835855, + "pairwise_causal_calibration_ece": 0.16720072894696625, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.407721921578567, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21447330713272095, + "seed": "655884605/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09131735563278198, + "selected_utility_gain_over_base": -0.07168932259082794, + "selector_regret": 0.12315595149993896, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.785526692867279, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "chart_id": "ms-36bc1364d2439099e9ed", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.5487087011529058, + "pairwise_causal_calibration_ece": 0.1655770131328084, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.15407578821846468, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "seed": "1113371261/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "chart_id": "ms-420280f15a522ce21c72", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42016806722689076, + "pairwise_causal_calibration_confidence": 0.540574554638564, + "pairwise_causal_calibration_ece": 0.1204064874116733, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.08662696338330045, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.82413649559021, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.17586350440979, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "chart_id": "ms-4aed1197d0906ae51c34", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5111111111111111, + "pairwise_causal_calibration_confidence": 0.5394504298273115, + "pairwise_causal_calibration_ece": 0.02833931871620072, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.3123388737013393, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train2", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.55, + "pairwise_causal_calibration_confidence": 0.5458205639974503, + "pairwise_causal_calibration_ece": 0.026780026272463814, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.05502138612288987, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "seed": "211245842/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04297435283660889, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7661172300577164, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "chart_id": "ms-77536801924a2a78688e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5313847796953075, + "pairwise_causal_calibration_ece": 0.06471811302864083, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.06561892171053907, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "seed": "1151410346/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.20950870215892792, + "selected_utility_gain_over_base": -0.016492590308189392, + "selector_regret": 0.34269602596759796, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4477952718734741, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "chart_id": "ms-9dd1ff4c297a81b13071", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.545535241490244, + "pairwise_causal_calibration_ece": 0.07681453301793482, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.0670111567575864, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19291165471076965, + "seed": "1324133122/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13575458526611328, + "selected_utility_gain_over_base": 0.04354967176914215, + "selector_regret": 0.05715706944465637, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8070883452892303, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.8333333333333334, + "pairwise_causal_calibration_confidence": 0.5807247911175116, + "pairwise_causal_calibration_ece": 0.25260854221582163, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.22431661105910145, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/train2", + "selected_candidate_index": 11, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d776afa14e6322cadedd", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5952380952380952, + "pairwise_causal_calibration_confidence": 0.5400007019106284, + "pairwise_causal_calibration_ece": 0.05523739332746688, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.10754995958912039, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/train2", + "selected_candidate_index": 12, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-e887dbfa0133af46829d", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5394736842105263, + "pairwise_causal_calibration_confidence": 0.5438956195294256, + "pairwise_causal_calibration_ece": 0.05003241707833195, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": -0.12321088891053344, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train2", + "selected_candidate_index": 13, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.0009112489076126273, + "train_seed": "2" + } + ], + "schema_version": 1, + "selected_lambda": 10.0, + "selector_calibration_target_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "selector_calibration_target_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "selector_chart_feature_mode": "base_context_obs", + "selector_eval_target_content_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "selector_eval_target_split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "selector_source_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "selector_source_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "source_content_hash": null, + "source_index": null, + "source_split_hash": null, + "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "success_bonus": 1.0, + "summary": { + "base_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.26666666666666666, + "low": 0.26666666666666666, + "mean": 0.30625, + "n": 120, + "num_samples": 1 + }, + "mean": 0.30625, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 1.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 0.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 1.0, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 1.0, + "747260437/train1": 1.0, + "747260437/train2": 1.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.20654545454545453, + "low": 0.20654545454545453, + "mean": 0.19987878787878788, + "n": 5, + "num_samples": 1 + }, + "mean": 0.19987878787878788, + "per_task": { + "LiftPegUpright-v1": 0.2, + "PickCube-v1": 0.0, + "PullCube-v1": 0.36, + "PushCube-v1": 0.16666666666666666, + "StackCube-v1": 0.2727272727272727 + } + }, + "metric": "base_success", + "micro": { + "confidence": 0.95, + "high": 0.2777777777777778, + "low": 0.2777777777777778, + "mean": 0.2916666666666667, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "base_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.7473616204298904, + "low": 0.7473616204298904, + "mean": 0.7901180622977944, + "n": 120, + "num_samples": 1 + }, + "mean": 0.7901180622977944, + "per_seed": { + "1073161778/train0": 2.0, + "1073161778/train1": 2.0, + "1073161778/train2": 2.0, + "1113371261/train0": 0.15842342376708984, + "1113371261/train1": 0.15842342376708984, + "1113371261/train2": 0.15842342376708984, + "1151410346/train0": 0.2260012924671173, + "1151410346/train1": 0.2260012924671173, + "1151410346/train2": 0.2260012924671173, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 2.0, + "1324133122/train0": 0.09220491349697113, + "1324133122/train1": 0.09220491349697113, + "1324133122/train2": 0.09220491349697113, + "1324642411/train0": 0.10647052526473999, + "1324642411/train1": 0.10647052526473999, + "1324642411/train2": 0.10647052526473999, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.10283424705266953, + "141984615/train1": 0.10283424705266953, + "141984615/train2": 0.10283424705266953, + "1461069713/train0": 0.0939030647277832, + "1461069713/train1": 0.0939030647277832, + "1461069713/train2": 0.0939030647277832, + "1480856810/train0": 0.08703407645225525, + "1480856810/train1": 0.08703407645225525, + "1480856810/train2": 0.08703407645225525, + "1509980664/train0": 0.8334100246429443, + "1509980664/train1": 0.8334100246429443, + "1509980664/train2": 0.8334100246429443, + "1522272686/train0": 0.5839591026306152, + "1522272686/train1": 0.5839591026306152, + "1522272686/train2": 0.5839591026306152, + "1549324805/train0": 0.10451483726501465, + "1549324805/train1": 0.10451483726501465, + "1549324805/train2": 0.10451483726501465, + "1578990937/train0": 0.82413649559021, + "1578990937/train1": 0.82413649559021, + "1578990937/train2": 0.82413649559021, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 0.12476746737957001, + "1790188787/train1": 0.12476746737957001, + "1790188787/train2": 0.12476746737957001, + "1822210471/train0": 0.07191530615091324, + "1822210471/train1": 0.07191530615091324, + "1822210471/train2": 0.07191530615091324, + "1845489013/train0": 2.0, + "1845489013/train1": 2.0, + "1845489013/train2": 2.0, + "1999022574/train0": 2.0, + "1999022574/train1": 2.0, + "1999022574/train2": 2.0, + "2012211287/train0": 0.7236053347587585, + "2012211287/train1": 0.7236053347587585, + "2012211287/train2": 0.7236053347587585, + "2038594191/train0": 2.0, + "2038594191/train1": 2.0, + "2038594191/train2": 2.0, + "2054899126/train0": 2.0, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 0.05225050449371338, + "2061635888/train1": 0.05225050449371338, + "2061635888/train2": 0.05225050449371338, + "211245842/train0": 0.19090841710567474, + "211245842/train1": 0.19090841710567474, + "211245842/train2": 0.19090841710567474, + "278008524/train0": 0.042734989896416664, + "278008524/train1": 0.042734989896416664, + "278008524/train2": 0.042734989896416664, + "307068879/train0": 0.6000905632972717, + "307068879/train1": 0.6000905632972717, + "307068879/train2": 0.6000905632972717, + "412102563/train0": 2.0, + "412102563/train1": 2.0, + "412102563/train2": 2.0, + "440577711/train0": 0.036152780055999756, + "440577711/train1": 0.036152780055999756, + "440577711/train2": 0.036152780055999756, + "441046872/train0": 0.6999551653862, + "441046872/train1": 0.6999551653862, + "441046872/train2": 0.6999551653862, + "509430392/train0": 2.2351741790771484e-07, + "509430392/train1": 2.2351741790771484e-07, + "509430392/train2": 2.2351741790771484e-07, + "585904130/train0": 0.18171554803848267, + "585904130/train1": 0.18171554803848267, + "585904130/train2": 0.18171554803848267, + "608016290/train0": 0.725701630115509, + "608016290/train1": 0.725701630115509, + "608016290/train2": 0.725701630115509, + "626486813/train0": 0.00509834272088483, + "626486813/train1": 0.00509834272088483, + "626486813/train2": 0.00509834272088483, + "636680137/train0": 0.08313504606485367, + "636680137/train1": 0.08313504606485367, + "636680137/train2": 0.08313504606485367, + "655884605/train0": 0.16300667822360992, + "655884605/train1": 0.16300667822360992, + "655884605/train2": 0.16300667822360992, + "672328262/train0": 2.0, + "672328262/train1": 2.0, + "672328262/train2": 2.0, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 2.0, + "747260437/train1": 2.0, + "747260437/train2": 2.0, + "750448296/train0": 0.11759314934412639, + "750448296/train1": 0.11759314934412639, + "750448296/train2": 0.11759314934412639, + "767747424/train0": 0.5731993420049548, + "767747424/train1": 0.5731993420049548, + "767747424/train2": 0.5731993420049548 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.8070530691606586, + "low": 0.8070530691606586, + "mean": 0.695559797174015, + "n": 5, + "num_samples": 1 + }, + "mean": 0.695559797174015, + "per_task": { + "LiftPegUpright-v1": 0.9498705387115478, + "PickCube-v1": 0.5839591026306152, + "PullCube-v1": 0.7751704612234608, + "PushCube-v1": 0.3924041787783305, + "StackCube-v1": 0.7763947045261209 + } + }, + "metric": "base_utility", + "micro": { + "confidence": 0.95, + "high": 0.6946634994641904, + "low": 0.6946634994641904, + "mean": 0.7418195864423373, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "coverage": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5576388888888889, + "low": 0.5576388888888889, + "mean": 0.6145833333333333, + "n": 120, + "num_samples": 1 + }, + "mean": 0.6145833333333333, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 1.0, + "1113371261/train2": 0.0, + "1151410346/train0": 1.0, + "1151410346/train1": 1.0, + "1151410346/train2": 1.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 1.0, + "1324133122/train1": 1.0, + "1324133122/train2": 1.0, + "1324642411/train0": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 1.0, + "141984615/train1": 1.0, + "141984615/train2": 1.0, + "1461069713/train0": 1.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 0.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 1.0, + "2054899126/train1": 0.5, + "2054899126/train2": 0.5, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 0.0, + "278008524/train0": 1.0, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 1.0, + "636680137/train0": 1.0, + "636680137/train1": 1.0, + "636680137/train2": 1.0, + "655884605/train0": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.3333333333333333, + "767747424/train0": 0.75, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4865454545454545, + "low": 0.4865454545454545, + "mean": 0.4998787878787878, + "n": 5, + "num_samples": 1 + }, + "mean": 0.4998787878787878, + "per_task": { + "LiftPegUpright-v1": 0.6, + "PickCube-v1": 0.0, + "PullCube-v1": 0.6266666666666667, + "PushCube-v1": 0.6666666666666666, + "StackCube-v1": 0.6060606060606061 + } + }, + "metric": "coverage", + "micro": { + "confidence": 0.95, + "high": 0.6527777777777778, + "low": 0.6527777777777778, + "mean": 0.6111111111111112, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "execute_generated": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5576388888888889, + "low": 0.5576388888888889, + "mean": 0.6145833333333333, + "n": 120, + "num_samples": 1 + }, + "mean": 0.6145833333333333, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 1.0, + "1113371261/train2": 0.0, + "1151410346/train0": 1.0, + "1151410346/train1": 1.0, + "1151410346/train2": 1.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 1.0, + "1324133122/train1": 1.0, + "1324133122/train2": 1.0, + "1324642411/train0": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 1.0, + "141984615/train1": 1.0, + "141984615/train2": 1.0, + "1461069713/train0": 1.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 0.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 1.0, + "2054899126/train1": 0.5, + "2054899126/train2": 0.5, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 0.0, + "278008524/train0": 1.0, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 1.0, + "636680137/train0": 1.0, + "636680137/train1": 1.0, + "636680137/train2": 1.0, + "655884605/train0": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.3333333333333333, + "767747424/train0": 0.75, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4865454545454545, + "low": 0.4865454545454545, + "mean": 0.4998787878787878, + "n": 5, + "num_samples": 1 + }, + "mean": 0.4998787878787878, + "per_task": { + "LiftPegUpright-v1": 0.6, + "PickCube-v1": 0.0, + "PullCube-v1": 0.6266666666666667, + "PushCube-v1": 0.6666666666666666, + "StackCube-v1": 0.6060606060606061 + } + }, + "metric": "execute_generated", + "micro": { + "confidence": 0.95, + "high": 0.6527777777777778, + "low": 0.6527777777777778, + "mean": 0.6111111111111112, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "fallback_rate": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.44236111111111104, + "low": 0.44236111111111104, + "mean": 0.38541666666666663, + "n": 120, + "num_samples": 1 + }, + "mean": 0.38541666666666663, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 1.0, + "1113371261/train1": 0.0, + "1113371261/train2": 1.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 1.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 1.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 1.0, + "1522272686/train1": 1.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 1.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 1.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.0, + "2054899126/train1": 0.5, + "2054899126/train2": 0.5, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 1.0, + "278008524/train0": 0.0, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 1.0, + "307068879/train1": 1.0, + "307068879/train2": 1.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 1.0, + "441046872/train1": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 1.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 1.0, + "747260437/train1": 1.0, + "747260437/train2": 1.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.25, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5134545454545455, + "low": 0.5134545454545455, + "mean": 0.5001212121212121, + "n": 5, + "num_samples": 1 + }, + "mean": 0.5001212121212121, + "per_task": { + "LiftPegUpright-v1": 0.4, + "PickCube-v1": 1.0, + "PullCube-v1": 0.37333333333333335, + "PushCube-v1": 0.3333333333333333, + "StackCube-v1": 0.3939393939393939 + } + }, + "metric": "fallback_rate", + "micro": { + "confidence": 0.95, + "high": 0.3472222222222222, + "low": 0.3472222222222222, + "mean": 0.3888888888888889, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "hidden_chart_oracle_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.7944444444444445, + "low": 0.7944444444444445, + "mean": 0.7854166666666668, + "n": 120, + "num_samples": 1 + }, + "mean": 0.7854166666666668, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 1.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 1.0, + "1151410346/train1": 1.0, + "1151410346/train2": 1.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 1.0, + "1324133122/train0": 1.0, + "1324133122/train1": 1.0, + "1324133122/train2": 1.0, + "1324642411/train0": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 1.0, + "1522272686/train1": 1.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 1.0, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 1.0, + "278008524/train0": 0.5, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 1.0, + "307068879/train1": 1.0, + "307068879/train2": 1.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 1.0, + "441046872/train1": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.9039999999999999, + "low": 0.9039999999999999, + "mean": 0.8706666666666667, + "n": 5, + "num_samples": 1 + }, + "mean": 0.8706666666666667, + "per_task": { + "LiftPegUpright-v1": 1.0, + "PickCube-v1": 1.0, + "PullCube-v1": 0.52, + "PushCube-v1": 0.8333333333333334, + "StackCube-v1": 1.0 + } + }, + "metric": "hidden_chart_oracle_success", + "micro": { + "confidence": 0.95, + "high": 0.7083333333333334, + "low": 0.7083333333333334, + "mean": 0.7291666666666666, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "hidden_chart_oracle_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.6374536744836306, + "low": 1.6374536744836306, + "mean": 1.6190625454609595, + "n": 120, + "num_samples": 1 + }, + "mean": 1.6190625454609595, + "per_seed": { + "1073161778/train0": 2.0, + "1073161778/train1": 2.0, + "1073161778/train2": 2.0, + "1113371261/train0": 2.0, + "1113371261/train1": 2.0, + "1113371261/train2": 2.0, + "1151410346/train0": 2.0, + "1151410346/train1": 2.0, + "1151410346/train2": 2.0, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 2.0, + "1324133122/train0": 2.0, + "1324133122/train1": 2.0, + "1324133122/train2": 2.0, + "1324642411/train0": 2.0, + "1324642411/train1": 2.0, + "1324642411/train2": 2.0, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.2857772707939148, + "141984615/train1": 0.2857772707939148, + "141984615/train2": 0.2857772707939148, + "1461069713/train0": 0.1593424528837204, + "1461069713/train1": 0.1593424528837204, + "1461069713/train2": 0.1593424528837204, + "1480856810/train0": 0.20948036015033722, + "1480856810/train1": 0.20948036015033722, + "1480856810/train2": 0.20948036015033722, + "1509980664/train0": 2.0, + "1509980664/train1": 2.0, + "1509980664/train2": 2.0, + "1522272686/train0": 2.0, + "1522272686/train1": 2.0, + "1522272686/train2": 2.0, + "1549324805/train0": 0.2974560856819153, + "1549324805/train1": 0.2974560856819153, + "1549324805/train2": 0.2974560856819153, + "1578990937/train0": 2.0, + "1578990937/train1": 2.0, + "1578990937/train2": 2.0, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 2.0, + "1790188787/train1": 2.0, + "1790188787/train2": 2.0, + "1822210471/train0": 2.0, + "1822210471/train1": 2.0, + "1822210471/train2": 2.0, + "1845489013/train0": 2.0, + "1845489013/train1": 2.0, + "1845489013/train2": 2.0, + "1999022574/train0": 2.0, + "1999022574/train1": 2.0, + "1999022574/train2": 2.0, + "2012211287/train0": 2.0, + "2012211287/train1": 2.0, + "2012211287/train2": 2.0, + "2038594191/train0": 2.0, + "2038594191/train1": 2.0, + "2038594191/train2": 2.0, + "2054899126/train0": 2.0, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 2.0, + "2061635888/train1": 2.0, + "2061635888/train2": 2.0, + "211245842/train0": 2.0, + "211245842/train1": 2.0, + "211245842/train2": 2.0, + "278008524/train0": 1.1139086410403252, + "278008524/train1": 1.1139086410403252, + "278008524/train2": 1.1139086410403252, + "307068879/train0": 2.0, + "307068879/train1": 2.0, + "307068879/train2": 2.0, + "412102563/train0": 2.0, + "412102563/train1": 2.0, + "412102563/train2": 2.0, + "440577711/train0": 2.0, + "440577711/train1": 2.0, + "440577711/train2": 2.0, + "441046872/train0": 2.0, + "441046872/train1": 2.0, + "441046872/train2": 2.0, + "509430392/train0": 2.0, + "509430392/train1": 2.0, + "509430392/train2": 2.0, + "585904130/train0": 2.0, + "585904130/train1": 2.0, + "585904130/train2": 2.0, + "608016290/train0": 2.0, + "608016290/train1": 2.0, + "608016290/train2": 2.0, + "626486813/train0": 0.16558238118886948, + "626486813/train1": 0.16558238118886948, + "626486813/train2": 0.16558238118886948, + "636680137/train0": 0.17056432366371155, + "636680137/train1": 0.17056432366371155, + "636680137/train2": 0.17056432366371155, + "655884605/train0": 2.0, + "655884605/train1": 2.0, + "655884605/train2": 2.0, + "672328262/train0": 2.0, + "672328262/train1": 2.0, + "672328262/train2": 2.0, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 0.2786632478237152, + "747260437/train1": 0.2786632478237152, + "747260437/train2": 0.2786632478237152, + "750448296/train0": 1.4079964955647786, + "750448296/train1": 1.4079964955647786, + "750448296/train2": 1.4079964955647786, + "767747424/train0": 0.6737305596470833, + "767747424/train1": 0.6737305596470833, + "767747424/train2": 0.6737305596470833 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.8293575772047042, + "low": 1.8293575772047042, + "mean": 1.7696735892097153, + "n": 5, + "num_samples": 1 + }, + "mean": 1.7696735892097153, + "per_task": { + "LiftPegUpright-v1": 2.0, + "PickCube-v1": 2.0, + "PullCube-v1": 1.1467878860235214, + "PushCube-v1": 1.7015800600250561, + "StackCube-v1": 2.0 + } + }, + "metric": "hidden_chart_oracle_utility", + "micro": { + "confidence": 0.95, + "high": 1.4819050737553172, + "low": 1.4819050737553172, + "mean": 1.5183161981403828, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "outcome_ptr": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5611111111111111, + "low": 0.5611111111111111, + "mean": 0.5319444444444444, + "n": 120, + "num_samples": 1 + }, + "mean": 0.5319444444444444, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 1.0, + "1151410346/train1": 1.0, + "1151410346/train2": 1.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 1.0, + "1324133122/train2": 1.0, + "1324642411/train0": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 1.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 1.0, + "278008524/train0": 1.0, + "278008524/train1": 1.0, + "278008524/train2": 1.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 1.0, + "441046872/train1": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 0.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 1.0, + "636680137/train0": 1.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 1.0, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4339393939393939, + "low": 0.4339393939393939, + "mean": 0.49393939393939396, + "n": 5, + "num_samples": 1 + }, + "mean": 0.49393939393939396, + "per_task": { + "LiftPegUpright-v1": 0.5333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": 0.4666666666666667, + "PushCube-v1": 0.8333333333333334, + "StackCube-v1": 0.6363636363636364 + } + }, + "metric": "outcome_ptr", + "micro": { + "confidence": 0.95, + "high": 0.5694444444444444, + "low": 0.5694444444444444, + "mean": 0.5486111111111112, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "pairwise_causal_calibration_accuracy": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5105254043373397, + "low": 0.5105254043373397, + "mean": 0.5314638149834829, + "n": 117, + "num_samples": 1 + }, + "mean": 0.5314638149834829, + "per_seed": { + "1073161778/train0": 0.5714285714285714, + "1073161778/train1": 0.5416666666666666, + "1073161778/train2": 0.8095238095238095, + "1113371261/train0": 0.5, + "1113371261/train1": 0.5, + "1113371261/train2": 0.7142857142857143, + "1151410346/train0": 0.5047619047619047, + "1151410346/train1": 0.4666666666666667, + "1151410346/train2": 0.4666666666666667, + "1261930456/train0": 0.47474747474747475, + "1261930456/train1": 0.6181818181818182, + "1261930456/train2": 0.6666666666666666, + "1324133122/train0": 0.4642857142857143, + "1324133122/train1": 0.5357142857142857, + "1324133122/train2": 0.6071428571428571, + "1324642411/train0": 0.5083333333333333, + "1324642411/train1": 0.5798319327731093, + "1324642411/train2": 0.7, + "1375492786/train0": 0.6533333333333333, + "1375492786/train1": 0.5614035087719298, + "1375492786/train2": 0.7380952380952381, + "141984615/train0": 0.49166666666666664, + "141984615/train1": 0.5083333333333333, + "141984615/train2": 0.5416666666666666, + "1461069713/train0": 0.275, + "1461069713/train1": 0.43333333333333335, + "1461069713/train2": 0.6583333333333333, + "1480856810/train0": 0.47474747474747475, + "1480856810/train1": 0.48739495798319327, + "1480856810/train2": 0.680672268907563, + "1509980664/train0": 0.4666666666666667, + "1509980664/train1": 0.2222222222222222, + "1509980664/train2": 0.5111111111111111, + "1522272686/train0": 0.5, + "1522272686/train1": 0.19444444444444445, + "1522272686/train2": 0.5, + "1549324805/train0": 0.7083333333333334, + "1549324805/train1": 0.4083333333333333, + "1549324805/train2": 0.2833333333333333, + "1578990937/train0": 0.6153846153846154, + "1578990937/train1": 0.4727272727272727, + "1578990937/train2": 0.42016806722689076, + "1790188787/train0": 0.6363636363636364, + "1790188787/train1": 0.5350877192982456, + "1790188787/train2": 0.44545454545454544, + "1822210471/train0": 0.575, + "1822210471/train1": 0.5510204081632653, + "1822210471/train2": 0.525, + "1845489013/train0": 0.5869565217391305, + "1845489013/train1": 0.7692307692307693, + "1845489013/train2": 0.52, + "1999022574/train0": 0.3, + "1999022574/train1": 0.4714285714285714, + "1999022574/train2": 0.5394736842105263, + "2012211287/train0": 0.226890756302521, + "2012211287/train1": 0.4666666666666667, + "2012211287/train2": 0.6086956521739131, + "2038594191/train0": 0.3793103448275862, + "2038594191/train1": 0.42592592592592593, + "2038594191/train2": 0.41414141414141414, + "2054899126/train0": 0.30833333333333335, + "2054899126/train1": 0.6052631578947368, + "2054899126/train2": 0.47058823529411764, + "2061635888/train0": 0.5833333333333334, + "2061635888/train1": 0.5, + "2061635888/train2": 0.5, + "211245842/train0": 0.7058823529411765, + "211245842/train1": 0.7916666666666666, + "211245842/train2": 0.55, + "278008524/train0": 0.5335858585858586, + "278008524/train1": 0.5306159420289855, + "278008524/train2": 0.35833333333333334, + "307068879/train0": 0.6153846153846154, + "307068879/train1": 0.5934065934065934, + "307068879/train2": 0.4945054945054945, + "412102563/train0": 0.5333333333333333, + "412102563/train1": 0.5384615384615384, + "412102563/train2": 0.5952380952380952, + "440577711/train0": 0.7586206896551724, + "440577711/train1": 0.6, + "440577711/train2": 0.6111111111111112, + "441046872/train0": 0.41818181818181815, + "441046872/train1": 0.5818181818181818, + "441046872/train2": 0.7272727272727273, + "509430392/train0": 0.6050420168067226, + "509430392/train1": 0.5904761904761905, + "509430392/train2": 0.6134453781512605, + "585904130/train0": 0.4523809523809524, + "585904130/train1": 0.5466666666666666, + "585904130/train2": 0.7272727272727273, + "608016290/train0": 0.5619047619047619, + "608016290/train1": 0.47619047619047616, + "608016290/train2": 0.48333333333333334, + "626486813/train0": 0.6125, + "626486813/train1": 0.39166666666666666, + "626486813/train2": 0.525, + "636680137/train0": 0.49166666666666664, + "636680137/train1": 0.5333333333333333, + "636680137/train2": 0.6583333333333333, + "655884605/train0": 0.6666666666666666, + "655884605/train1": 0.4722222222222222, + "655884605/train2": 0.3888888888888889, + "672328262/train0": 0.7647058823529411, + "672328262/train1": 0.6571428571428571, + "672328262/train2": 0.16666666666666666, + "701667486/train0": 0.40476190476190477, + "701667486/train1": 0.6785714285714286, + "701667486/train2": 0.8333333333333334, + "747260437/train0": 0.4583333333333333, + "747260437/train1": 0.5666666666666667, + "747260437/train2": 0.625, + "750448296/train0": 0.5218253968253967, + "750448296/train1": 0.3532051282051282, + "750448296/train2": 0.44642857142857145, + "767747424/train0": 0.4566239316239316, + "767747424/train1": 0.516633897148603, + "767747424/train2": 0.41618544853838974 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5037682368460075, + "low": 0.5037682368460075, + "mean": 0.5123674931231779, + "n": 5, + "num_samples": 1 + }, + "mean": 0.5123674931231779, + "per_task": { + "LiftPegUpright-v1": 0.54512467498401, + "PickCube-v1": 0.39814814814814814, + "PullCube-v1": 0.4995022986920425, + "PushCube-v1": 0.588120956369862, + "StackCube-v1": 0.5309413874218266 + } + }, + "metric": "pairwise_causal_calibration_accuracy", + "micro": { + "confidence": 0.95, + "high": 0.5238010329409267, + "low": 0.5238010329409267, + "mean": 0.5213348998444443, + "n": 138, + "num_samples": 1 + }, + "n": 138 + }, + "pairwise_causal_calibration_confidence": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5441632520903402, + "low": 0.5441632520903402, + "mean": 0.5417425001495095, + "n": 117, + "num_samples": 1 + }, + "mean": 0.5417425001495095, + "per_seed": { + "1073161778/train0": 0.5362042832267765, + "1073161778/train1": 0.5477005041063188, + "1073161778/train2": 0.5547882621085445, + "1113371261/train0": 0.5401592730365951, + "1113371261/train1": 0.5380625270265227, + "1113371261/train2": 0.5487087011529058, + "1151410346/train0": 0.5241846653555288, + "1151410346/train1": 0.5513917700725205, + "1151410346/train2": 0.5313847796953075, + "1261930456/train0": 0.5407472634490431, + "1261930456/train1": 0.5491787800871848, + "1261930456/train2": 0.5397391671472074, + "1324133122/train0": 0.5500982965860671, + "1324133122/train1": 0.5502823354081949, + "1324133122/train2": 0.545535241490244, + "1324642411/train0": 0.5337665023974802, + "1324642411/train1": 0.5288122689723307, + "1324642411/train2": 0.528083048643476, + "1375492786/train0": 0.5447046728594015, + "1375492786/train1": 0.5446679554502398, + "1375492786/train2": 0.565958547539462, + "141984615/train0": 0.5402158636214657, + "141984615/train1": 0.5473004480108934, + "141984615/train2": 0.5440555881176564, + "1461069713/train0": 0.5282766199734893, + "1461069713/train1": 0.5243338816656252, + "1461069713/train2": 0.5225830368338087, + "1480856810/train0": 0.5212194722509627, + "1480856810/train1": 0.5158258523764001, + "1480856810/train2": 0.5204768609791875, + "1509980664/train0": 0.5320927425921708, + "1509980664/train1": 0.5391700207769446, + "1509980664/train2": 0.5394504298273115, + "1522272686/train0": 0.5681060848637505, + "1522272686/train1": 0.5689701709118037, + "1522272686/train2": 0.545306820402758, + "1549324805/train0": 0.5299868735462288, + "1549324805/train1": 0.5265093687069327, + "1549324805/train2": 0.541054902787583, + "1578990937/train0": 0.544349322443362, + "1578990937/train1": 0.5543665049371578, + "1578990937/train2": 0.540574554638564, + "1790188787/train0": 0.5392236442621141, + "1790188787/train1": 0.5396885778611449, + "1790188787/train2": 0.5413352723129059, + "1822210471/train0": 0.5565662285154201, + "1822210471/train1": 0.5508218552014704, + "1822210471/train2": 0.5302663723099226, + "1845489013/train0": 0.5198092453508348, + "1845489013/train1": 0.519320720354459, + "1845489013/train2": 0.5338919568560658, + "1999022574/train0": 0.5344493626989616, + "1999022574/train1": 0.5395820832404352, + "1999022574/train2": 0.5438956195294256, + "2012211287/train0": 0.5363365754560178, + "2012211287/train1": 0.5335084608302744, + "2012211287/train2": 0.5293148633692549, + "2038594191/train0": 0.6018790607135653, + "2038594191/train1": 0.5909326842120879, + "2038594191/train2": 0.5664818478408624, + "2054899126/train0": 0.5238354835309519, + "2054899126/train1": 0.5244611135518833, + "2054899126/train2": 0.5297648676632524, + "2061635888/train0": 0.5351503713504386, + "2061635888/train1": 0.518040614785329, + "2061635888/train2": 0.5252975568525992, + "211245842/train0": 0.5545781832114899, + "211245842/train1": 0.5498984873423876, + "211245842/train2": 0.5458205639974503, + "278008524/train0": 0.5477839621589006, + "278008524/train1": 0.5479136366856556, + "278008524/train2": 0.5528248393150181, + "307068879/train0": 0.5488728190719444, + "307068879/train1": 0.5348399847973542, + "307068879/train2": 0.5640014996133708, + "412102563/train0": 0.5413119316483767, + "412102563/train1": 0.5307516643020463, + "412102563/train2": 0.5400007019106284, + "440577711/train0": 0.5267661600012826, + "440577711/train1": 0.5289194284606322, + "440577711/train2": 0.5489590881909069, + "441046872/train0": 0.5379829028529084, + "441046872/train1": 0.5337021369933691, + "441046872/train2": 0.5454083397120673, + "509430392/train0": 0.5254427085184519, + "509430392/train1": 0.5382862696863612, + "509430392/train2": 0.5437150122698521, + "585904130/train0": 0.5292002900283871, + "585904130/train1": 0.5343887712270259, + "585904130/train2": 0.5282487650578878, + "608016290/train0": 0.5498437058747719, + "608016290/train1": 0.5706245241886934, + "608016290/train2": 0.5615509166352838, + "626486813/train0": 0.5678117074464072, + "626486813/train1": 0.5645620394062392, + "626486813/train2": 0.5613104362626665, + "636680137/train0": 0.532502870801132, + "636680137/train1": 0.5173967032259332, + "636680137/train2": 0.5151979568732741, + "655884605/train0": 0.568990751323036, + "655884605/train1": 0.5500689384066487, + "655884605/train2": 0.556089617835855, + "672328262/train0": 0.542991093593322, + "672328262/train1": 0.5517445085212526, + "672328262/train2": 0.5462121445015077, + "701667486/train0": 0.5617516698950917, + "701667486/train1": 0.546009069568378, + "701667486/train2": 0.5807247911175116, + "747260437/train0": 0.5237942697642416, + "747260437/train1": 0.5173328988662427, + "747260437/train2": 0.5276406049368979, + "750448296/train0": 0.5273870612831654, + "750448296/train1": 0.5396863859528889, + "750448296/train2": 0.5280087301794993, + "767747424/train0": 0.5461567728386378, + "767747424/train1": 0.5530278828158679, + "767747424/train2": 0.5435976135993295 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5476679716398777, + "low": 0.5476679716398777, + "mean": 0.5453187454539064, + "n": 5, + "num_samples": 1 + }, + "mean": 0.5453187454539064, + "per_task": { + "LiftPegUpright-v1": 0.54564531858913, + "PickCube-v1": 0.5607943587261041, + "PullCube-v1": 0.5422481003539483, + "PushCube-v1": 0.5338991876592735, + "StackCube-v1": 0.5440067619410754 + } + }, + "metric": "pairwise_causal_calibration_confidence", + "micro": { + "confidence": 0.95, + "high": 0.5405627434399822, + "low": 0.5405627434399822, + "mean": 0.5423521036330443, + "n": 138, + "num_samples": 1 + }, + "n": 138 + }, + "pairwise_causal_calibration_ece": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.10143819375783508, + "low": 0.10143819375783508, + "mean": 0.10620714166278308, + "n": 120, + "num_samples": 1 + }, + "mean": 0.10620714166278308, + "per_seed": { + "1073161778/train0": 0.07639944182594278, + "1073161778/train1": 0.12867242656969963, + "1073161778/train2": 0.25473554741526483, + "1113371261/train0": 0.04015927303659509, + "1113371261/train1": 0.038062527026522686, + "1113371261/train2": 0.1655770131328084, + "1151410346/train0": 0.019422760593624022, + "1151410346/train1": 0.08472510340585364, + "1151410346/train2": 0.06471811302864083, + "1261930456/train0": 0.06599978870156839, + "1261930456/train1": 0.0690030380946333, + "1261930456/train2": 0.12692749951945942, + "1324133122/train0": 0.08581258230035274, + "1324133122/train1": 0.015677218286420276, + "1324133122/train2": 0.07681453301793482, + "1324642411/train0": 0.032055347656845035, + "1324642411/train1": 0.0510196638007786, + "1324642411/train2": 0.17191695135652396, + "1375492786/train0": 0.10862866047393183, + "1375492786/train1": 0.01673555332169, + "1375492786/train2": 0.1721366905557764, + "141984615/train0": 0.048549196954798944, + "141984615/train1": 0.046271089584524104, + "141984615/train2": 0.03902364539059017, + "1461069713/train0": 0.25327661997348927, + "1461069713/train1": 0.09100054833229188, + "1461069713/train2": 0.13575029649952464, + "1480856810/train0": 0.046471997503487905, + "1480856810/train1": 0.02843089439320684, + "1480856810/train2": 0.1601954079283755, + "1509980664/train0": 0.07422005046880752, + "1509980664/train1": 0.31694779855472244, + "1509980664/train2": 0.02833931871620072, + "1522272686/train0": 0.07816118582099628, + "1522272686/train1": 0.37452572646735927, + "1522272686/train2": 0.04530682040275796, + "1549324805/train0": 0.17834645978710464, + "1549324805/train1": 0.11817603537359933, + "1549324805/train2": 0.25772156945424957, + "1578990937/train0": 0.07103529294125341, + "1578990937/train1": 0.081639232209885, + "1578990937/train2": 0.1204064874116733, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.09713999210152216, + "1790188787/train1": 0.004600858562899268, + "1790188787/train2": 0.09588072685836047, + "1822210471/train0": 0.15721195241478056, + "1822210471/train1": 0.14397334425216246, + "1822210471/train2": 0.118390411016323, + "1845489013/train0": 0.06714727638829565, + "1845489013/train1": 0.2499100488763103, + "1845489013/train2": 0.013891956856065812, + "1999022574/train0": 0.23444936269896155, + "1999022574/train1": 0.06815351181186355, + "1999022574/train2": 0.05003241707833195, + "2012211287/train0": 0.3094458191534968, + "2012211287/train1": 0.07199480463689167, + "2012211287/train2": 0.07938078880465815, + "2038594191/train0": 0.22256871588597915, + "2038594191/train1": 0.1851395001152688, + "2038594191/train2": 0.1523404336994486, + "2054899126/train0": 0.10775107509880927, + "2054899126/train1": 0.04040102217142677, + "2054899126/train2": 0.029588316184567365, + "2061635888/train0": 0.048182961982894734, + "2061635888/train1": 0.018040614785329034, + "2061635888/train2": 0.025297556852599157, + "211245842/train0": 0.1513041697296866, + "211245842/train1": 0.2417681793242794, + "211245842/train2": 0.026780026272463814, + "278008524/train0": 0.14072928673079005, + "278008524/train1": 0.09383678611256409, + "278008524/train2": 0.19449150598168496, + "307068879/train0": 0.06651179631267089, + "307068879/train1": 0.06342054136618626, + "307068879/train2": 0.06949600510787626, + "412102563/train0": 0.007978598315043392, + "412102563/train1": 0.007709874159492092, + "412102563/train2": 0.05523739332746688, + "440577711/train0": 0.23185452965388975, + "440577711/train1": 0.07108057153936775, + "440577711/train2": 0.06215202292020417, + "441046872/train0": 0.11980108467109023, + "441046872/train1": 0.04811604482481282, + "441046872/train2": 0.18186438756066006, + "509430392/train0": 0.07959930828827078, + "509430392/train1": 0.052189920789829286, + "509430392/train2": 0.06973036588140841, + "585904130/train0": 0.0768193376474347, + "585904130/train1": 0.050834006610117985, + "585904130/train2": 0.19902396221483953, + "608016290/train0": 0.01206105602998994, + "608016290/train1": 0.09443404799821745, + "608016290/train2": 0.0782175833019504, + "626486813/train0": 0.09417264307778664, + "626486813/train1": 0.19860544920331658, + "626486813/train2": 0.07789314572719971, + "636680137/train0": 0.04083620413446535, + "636680137/train1": 0.015936630107400096, + "636680137/train2": 0.14313537646005925, + "655884605/train0": 0.09767591534363074, + "655884605/train1": 0.07784671618442661, + "655884605/train2": 0.16720072894696625, + "672328262/train0": 0.22171478875961914, + "672328262/train1": 0.10539834862160463, + "672328262/train2": 0.37954547783484094, + "701667486/train0": 0.31044146560402797, + "701667486/train1": 0.13256235900305033, + "701667486/train2": 0.25260854221582163, + "747260437/train0": 0.06546093643090828, + "747260437/train1": 0.04933376780042398, + "747260437/train2": 0.0973593950631021, + "750448296/train0": 0.0336721343859541, + "750448296/train1": 0.18648125774776067, + "750448296/train2": 0.1080586463500039, + "767747424/train0": 0.10799527425220615, + "767747424/train1": 0.05634090783297348, + "767747424/train2": 0.1276296201971279 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.1192700379370925, + "low": 0.1192700379370925, + "mean": 0.11405656276881852, + "n": 5, + "num_samples": 1 + }, + "mean": 0.11405656276881852, + "per_task": { + "LiftPegUpright-v1": 0.11030342503862721, + "PickCube-v1": 0.16599791089703783, + "PullCube-v1": 0.11105368029396932, + "PushCube-v1": 0.08423604919725729, + "StackCube-v1": 0.09869174841720094 + } + }, + "metric": "pairwise_causal_calibration_ece", + "micro": { + "confidence": 0.95, + "high": 0.10849662633491666, + "low": 0.10849662633491666, + "mean": 0.10593505356691993, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "pairwise_causal_calibration_pairs": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 87.40624999999999, + "low": 87.40624999999999, + "mean": 82.21180555555556, + "n": 120, + "num_samples": 1 + }, + "mean": 82.21180555555556, + "per_seed": { + "1073161778/train0": 119.0, + "1073161778/train1": 120.0, + "1073161778/train2": 84.0, + "1113371261/train0": 28.0, + "1113371261/train1": 28.0, + "1113371261/train2": 28.0, + "1151410346/train0": 105.0, + "1151410346/train1": 105.0, + "1151410346/train2": 105.0, + "1261930456/train0": 99.0, + "1261930456/train1": 110.0, + "1261930456/train2": 84.0, + "1324133122/train0": 28.0, + "1324133122/train1": 28.0, + "1324133122/train2": 28.0, + "1324642411/train0": 120.0, + "1324642411/train1": 119.0, + "1324642411/train2": 110.0, + "1375492786/train0": 75.0, + "1375492786/train1": 114.0, + "1375492786/train2": 84.0, + "141984615/train0": 120.0, + "141984615/train1": 120.0, + "141984615/train2": 120.0, + "1461069713/train0": 120.0, + "1461069713/train1": 120.0, + "1461069713/train2": 120.0, + "1480856810/train0": 99.0, + "1480856810/train1": 119.0, + "1480856810/train2": 119.0, + "1509980664/train0": 90.0, + "1509980664/train1": 63.0, + "1509980664/train2": 90.0, + "1522272686/train0": 36.0, + "1522272686/train1": 36.0, + "1522272686/train2": 36.0, + "1549324805/train0": 120.0, + "1549324805/train1": 120.0, + "1549324805/train2": 120.0, + "1578990937/train0": 117.0, + "1578990937/train1": 110.0, + "1578990937/train2": 119.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 110.0, + "1790188787/train1": 114.0, + "1790188787/train2": 110.0, + "1822210471/train0": 40.0, + "1822210471/train1": 49.0, + "1822210471/train2": 40.0, + "1845489013/train0": 46.0, + "1845489013/train1": 13.0, + "1845489013/train2": 25.0, + "1999022574/train0": 70.0, + "1999022574/train1": 70.0, + "1999022574/train2": 76.0, + "2012211287/train0": 119.0, + "2012211287/train1": 120.0, + "2012211287/train2": 92.0, + "2038594191/train0": 29.0, + "2038594191/train1": 54.0, + "2038594191/train2": 99.0, + "2054899126/train0": 60.0, + "2054899126/train1": 57.0, + "2054899126/train2": 59.5, + "2061635888/train0": 36.0, + "2061635888/train1": 36.0, + "2061635888/train2": 36.0, + "211245842/train0": 119.0, + "211245842/train1": 120.0, + "211245842/train2": 120.0, + "278008524/train0": 109.5, + "278008524/train1": 106.0, + "278008524/train2": 109.5, + "307068879/train0": 91.0, + "307068879/train1": 91.0, + "307068879/train2": 91.0, + "412102563/train0": 60.0, + "412102563/train1": 39.0, + "412102563/train2": 84.0, + "440577711/train0": 29.0, + "440577711/train1": 15.0, + "440577711/train2": 54.0, + "441046872/train0": 55.0, + "441046872/train1": 55.0, + "441046872/train2": 55.0, + "509430392/train0": 119.0, + "509430392/train1": 105.0, + "509430392/train2": 119.0, + "585904130/train0": 84.0, + "585904130/train1": 75.0, + "585904130/train2": 110.0, + "608016290/train0": 105.0, + "608016290/train1": 105.0, + "608016290/train2": 120.0, + "626486813/train0": 120.0, + "626486813/train1": 120.0, + "626486813/train2": 120.0, + "636680137/train0": 120.0, + "636680137/train1": 120.0, + "636680137/train2": 120.0, + "655884605/train0": 36.0, + "655884605/train1": 36.0, + "655884605/train2": 36.0, + "672328262/train0": 17.0, + "672328262/train1": 35.0, + "672328262/train2": 30.0, + "701667486/train0": 42.0, + "701667486/train1": 84.0, + "701667486/train2": 54.0, + "747260437/train0": 120.0, + "747260437/train1": 120.0, + "747260437/train2": 120.0, + "750448296/train0": 108.0, + "750448296/train1": 101.66666666666667, + "750448296/train2": 108.0, + "767747424/train0": 119.25, + "767747424/train1": 119.0, + "767747424/train2": 119.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 77.07490909090909, + "low": 77.07490909090909, + "mean": 77.60379797979797, + "n": 5, + "num_samples": 1 + }, + "mean": 77.60379797979797, + "per_task": { + "LiftPegUpright-v1": 94.8, + "PickCube-v1": 36.0, + "PullCube-v1": 92.32, + "PushCube-v1": 97.44444444444444, + "StackCube-v1": 67.45454545454545 + } + }, + "metric": "pairwise_causal_calibration_pairs", + "micro": { + "confidence": 0.95, + "high": 82.16666666666667, + "low": 82.16666666666667, + "mean": 86.34722222222223, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "predicted_margin": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.15866427339723024, + "low": 0.15866427339723024, + "mean": 0.16961423206737078, + "n": 120, + "num_samples": 1 + }, + "mean": 0.16961423206737078, + "per_seed": { + "1073161778/train0": 0.24085858055337617, + "1073161778/train1": 0.11864476132183588, + "1073161778/train2": 0.24206048374838196, + "1113371261/train0": -0.002655981453428158, + "1113371261/train1": 0.08193066148519779, + "1113371261/train2": -0.15407578821846468, + "1151410346/train0": 0.11341562423740303, + "1151410346/train1": 0.027308203722147817, + "1151410346/train2": 0.06561892171053907, + "1261930456/train0": 0.09035295152992723, + "1261930456/train1": -0.02356230018007266, + "1261930456/train2": 0.04072195094915404, + "1324133122/train0": 0.07548993896412504, + "1324133122/train1": 0.3962004123380809, + "1324133122/train2": 0.0670111567575864, + "1324642411/train0": 0.711002878905307, + "1324642411/train1": 0.38447100568728765, + "1324642411/train2": 0.15208916465183764, + "1375492786/train0": -0.28428690828432274, + "1375492786/train1": -0.6782841630171567, + "1375492786/train2": -0.559803911595051, + "141984615/train0": 0.30976598855930915, + "141984615/train1": 0.39033675889099473, + "141984615/train2": 0.29555407226838537, + "1461069713/train0": 0.3059149126148908, + "1461069713/train1": 0.3007404756181269, + "1461069713/train2": 0.2553375780680003, + "1480856810/train0": 0.6891134156079126, + "1480856810/train1": 0.5636198101475918, + "1480856810/train2": 0.4905519223760793, + "1509980664/train0": -0.2172662931516871, + "1509980664/train1": -0.2497394332434142, + "1509980664/train2": -0.3123388737013393, + "1522272686/train0": -0.4821765811639407, + "1522272686/train1": -0.25520062215278894, + "1522272686/train2": -0.43379366859847046, + "1549324805/train0": 0.8008610893558498, + "1549324805/train1": 0.4059182168351142, + "1549324805/train2": 0.8716902723485557, + "1578990937/train0": 0.10915469051695448, + "1578990937/train1": 0.10231812655593348, + "1578990937/train2": -0.08662696338330045, + "1747178030/train0": 0.2685368177679842, + "1747178030/train1": 0.27732487886731, + "1747178030/train2": 0.3315076869895872, + "1790188787/train0": 0.4590932780625647, + "1790188787/train1": 0.33925681284155934, + "1790188787/train2": 0.297516022662714, + "1822210471/train0": 0.013297597406553785, + "1822210471/train1": 0.041781127110214866, + "1822210471/train2": -0.09918008580701754, + "1845489013/train0": 0.08565911209810279, + "1845489013/train1": 0.10538760252719387, + "1845489013/train2": 0.14874188314117123, + "1999022574/train0": -0.11161114265900707, + "1999022574/train1": -0.1454763298079958, + "1999022574/train2": -0.12321088891053344, + "2012211287/train0": 0.3161539044218536, + "2012211287/train1": 0.24432178181307274, + "2012211287/train2": 0.3345646470523239, + "2038594191/train0": 0.28192292963769827, + "2038594191/train1": 0.21139315187633212, + "2038594191/train2": 0.2451404604489238, + "2054899126/train0": 0.40089684325539643, + "2054899126/train1": 0.32373079560635964, + "2054899126/train2": 0.3425010993904758, + "2061635888/train0": 0.2859560598368813, + "2061635888/train1": 0.14710078956282402, + "2061635888/train2": 0.15541062212467605, + "211245842/train0": 0.05542394657973021, + "211245842/train1": 0.2276424215366616, + "211245842/train2": -0.05502138612288987, + "278008524/train0": 0.28802122449631784, + "278008524/train1": 0.3601165498724468, + "278008524/train2": 0.22015899787097548, + "307068879/train0": -0.14233288142692663, + "307068879/train1": -0.13522650945094028, + "307068879/train2": -0.02519869144581416, + "412102563/train0": -0.07414481146410684, + "412102563/train1": -0.10576298261783212, + "412102563/train2": -0.10754995958912039, + "440577711/train0": 0.7982832696690534, + "440577711/train1": 0.862893346101075, + "440577711/train2": 0.7945693500962356, + "441046872/train0": 0.024167399539588352, + "441046872/train1": -0.11694270182946509, + "441046872/train2": -0.12738205708571582, + "509430392/train0": 0.5885139744079059, + "509430392/train1": 0.5296709902746183, + "509430392/train2": 0.4810616128884794, + "585904130/train0": 0.23094602840169032, + "585904130/train1": 0.22575925280556675, + "585904130/train2": 0.18843222973041862, + "608016290/train0": 0.48611772128013164, + "608016290/train1": 0.2549086170487004, + "608016290/train2": 0.4202802138182712, + "626486813/train0": 0.5713883592563933, + "626486813/train1": 0.5558101781509216, + "626486813/train2": 0.3853803956007696, + "636680137/train0": 0.5991343617786198, + "636680137/train1": 0.19155725528837464, + "636680137/train2": 0.27475298254295666, + "655884605/train0": 0.4522143975120525, + "655884605/train1": 0.446772375365045, + "655884605/train2": 0.407721921578567, + "672328262/train0": -0.14745117870449231, + "672328262/train1": -0.21177052566683674, + "672328262/train2": -0.19816182889430234, + "701667486/train0": 0.2186075457961687, + "701667486/train1": 0.20451752246432947, + "701667486/train2": 0.22431661105910145, + "747260437/train0": -0.03550968246605479, + "747260437/train1": -0.27755861197582443, + "747260437/train2": 0.061753054977374155, + "750448296/train0": 0.008208417767252563, + "750448296/train1": -0.031216926452283373, + "750448296/train2": -0.06850240484656789, + "767747424/train0": 0.10341139925263847, + "767747424/train1": 0.23938122661168873, + "767747424/train2": 0.09558584120388469 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.027967807901998244, + "low": 0.027967807901998244, + "mean": 0.09115919827365179, + "n": 5, + "num_samples": 1 + }, + "mean": 0.09115919827365179, + "per_task": { + "LiftPegUpright-v1": 0.14233301795724493, + "PickCube-v1": -0.39039029063840003, + "PullCube-v1": 0.18136174528296545, + "PushCube-v1": 0.4582899698155127, + "StackCube-v1": 0.06420154895093595 + } + }, + "metric": "predicted_margin", + "micro": { + "confidence": 0.95, + "high": 0.19839291110429633, + "low": 0.19839291110429633, + "mean": 0.17315156851198613, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "proposal_oracle_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.6430555555555557, + "low": 0.6430555555555557, + "mean": 0.5847222222222223, + "n": 120, + "num_samples": 1 + }, + "mean": 0.5847222222222223, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 1.0, + "1151410346/train2": 0.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 1.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 0.5, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.5, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 1.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 0.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4496969696969697, + "low": 0.4496969696969697, + "mean": 0.5074747474747474, + "n": 5, + "num_samples": 1 + }, + "mean": 0.5074747474747474, + "per_task": { + "LiftPegUpright-v1": 0.6, + "PickCube-v1": 0.0, + "PullCube-v1": 0.5333333333333333, + "PushCube-v1": 0.8888888888888888, + "StackCube-v1": 0.5151515151515151 + } + }, + "metric": "proposal_oracle_success", + "micro": { + "confidence": 0.95, + "high": 0.5416666666666666, + "low": 0.5416666666666666, + "mean": 0.5694444444444444, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "proposal_oracle_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.342259801506427, + "low": 1.342259801506427, + "mean": 1.251578581963097, + "n": 120, + "num_samples": 1 + }, + "mean": 1.251578581963097, + "per_seed": { + "1073161778/train0": 2.0, + "1073161778/train1": 2.0, + "1073161778/train2": 2.0, + "1113371261/train0": 0.10274861752986908, + "1113371261/train1": 0.18093222379684448, + "1113371261/train2": 0.19527167081832886, + "1151410346/train0": 0.8663521409034729, + "1151410346/train1": 2.0, + "1151410346/train2": 0.5522047281265259, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 2.0, + "1324133122/train0": 0.054989203810691833, + "1324133122/train1": 0.1324586570262909, + "1324133122/train2": 0.19291165471076965, + "1324642411/train0": 0.14406004548072815, + "1324642411/train1": 2.0, + "1324642411/train2": 2.0, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.015177905559539795, + "141984615/train1": 0.037011127918958664, + "141984615/train2": 0.058951716870069504, + "1461069713/train0": 0.06019836664199829, + "1461069713/train1": 0.016585012897849083, + "1461069713/train2": 0.09997648000717163, + "1480856810/train0": 2.0, + "1480856810/train1": 2.0, + "1480856810/train2": 2.0, + "1509980664/train0": 2.0, + "1509980664/train1": 2.0, + "1509980664/train2": 2.0, + "1522272686/train0": 0.18325337767601013, + "1522272686/train1": 0.07014383375644684, + "1522272686/train2": 0.14937306940555573, + "1549324805/train0": 0.060871422290802, + "1549324805/train1": 0.19892151653766632, + "1549324805/train2": 0.11310184001922607, + "1578990937/train0": 2.0, + "1578990937/train1": 2.0, + "1578990937/train2": 2.0, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 2.0, + "1790188787/train1": 2.0, + "1790188787/train2": 2.0, + "1822210471/train0": 2.0, + "1822210471/train1": 2.0, + "1822210471/train2": 2.0, + "1845489013/train0": 2.0, + "1845489013/train1": 2.0, + "1845489013/train2": 2.0, + "1999022574/train0": 2.0, + "1999022574/train1": 2.0, + "1999022574/train2": 2.0, + "2012211287/train0": 2.0, + "2012211287/train1": 2.0, + "2012211287/train2": 2.0, + "2038594191/train0": 2.0, + "2038594191/train1": 2.0, + "2038594191/train2": 2.0, + "2054899126/train0": 1.077283151447773, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 0.09799930453300476, + "2061635888/train1": 0.07580336928367615, + "2061635888/train2": 0.11105132102966309, + "211245842/train0": 2.0, + "211245842/train1": 0.21295717358589172, + "211245842/train2": 0.23388276994228363, + "278008524/train0": 1.0733563229441643, + "278008524/train1": 1.0376745946705341, + "278008524/train2": 1.1359489113092422, + "307068879/train0": 0.2115575522184372, + "307068879/train1": 0.5876073241233826, + "307068879/train2": 0.2302134484052658, + "412102563/train0": 2.0, + "412102563/train1": 2.0, + "412102563/train2": 2.0, + "440577711/train0": 2.0, + "440577711/train1": 2.0, + "440577711/train2": 2.0, + "441046872/train0": 2.0, + "441046872/train1": 0.7233588695526123, + "441046872/train2": 0.7191348671913147, + "509430392/train0": 2.0, + "509430392/train1": 2.0, + "509430392/train2": 5.424022674560547e-06, + "585904130/train0": 2.0, + "585904130/train1": 2.0, + "585904130/train2": 2.0, + "608016290/train0": 2.0, + "608016290/train1": 2.0, + "608016290/train2": 0.7117276191711426, + "626486813/train0": 0.04469547560438514, + "626486813/train1": 0.035827874671667814, + "626486813/train2": 0.04489925503730774, + "636680137/train0": 0.08826130628585815, + "636680137/train1": 0.06741118431091309, + "636680137/train2": 0.0756516233086586, + "655884605/train0": 0.13732868432998657, + "655884605/train1": 0.8128758072853088, + "655884605/train2": 0.21447330713272095, + "672328262/train0": 2.0, + "672328262/train1": 2.0, + "672328262/train2": 2.0, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 0.10839563608169556, + "747260437/train1": 0.06328952312469482, + "747260437/train2": 0.12510187923908234, + "750448296/train0": 0.7526603279014429, + "750448296/train1": 1.3705032741030057, + "750448296/train2": 1.3504486680030823, + "767747424/train0": 1.0424291878007352, + "767747424/train1": 1.0629530912265182, + "767747424/train2": 1.0671670949086547 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.0449712097333237, + "low": 1.0449712097333237, + "mean": 1.1196794969856438, + "n": 5, + "num_samples": 1 + }, + "mean": 1.1196794969856438, + "per_task": { + "LiftPegUpright-v1": 1.4122399787108104, + "PickCube-v1": 0.13425676027933756, + "PullCube-v1": 1.1093241602430741, + "PushCube-v1": 1.7857814149724112, + "StackCube-v1": 1.1567951707225856 + } + }, + "metric": "proposal_oracle_utility", + "micro": { + "confidence": 0.95, + "high": 1.1602533682404708, + "low": 1.1602533682404708, + "mean": 1.2159999172436073, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selected_candidate_index": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 7.271527777777779, + "low": 7.271527777777779, + "mean": 6.847222222222222, + "n": 120, + "num_samples": 1 + }, + "mean": 6.847222222222222, + "per_seed": { + "1073161778/train0": 6.0, + "1073161778/train1": 8.0, + "1073161778/train2": 6.0, + "1113371261/train0": 2.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 6.0, + "1151410346/train1": 2.0, + "1151410346/train2": 1.0, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 4.0, + "1324133122/train0": 6.0, + "1324133122/train1": 2.0, + "1324133122/train2": 7.0, + "1324642411/train0": 11.0, + "1324642411/train1": 5.0, + "1324642411/train2": 9.0, + "1375492786/train0": 5.0, + "1375492786/train1": 9.0, + "1375492786/train2": 2.0, + "141984615/train0": 14.0, + "141984615/train1": 15.0, + "141984615/train2": 11.0, + "1461069713/train0": 15.0, + "1461069713/train1": 10.0, + "1461069713/train2": 3.0, + "1480856810/train0": 9.0, + "1480856810/train1": 9.0, + "1480856810/train2": 1.0, + "1509980664/train0": 10.0, + "1509980664/train1": 12.0, + "1509980664/train2": 9.0, + "1522272686/train0": 8.0, + "1522272686/train1": 2.0, + "1522272686/train2": 5.0, + "1549324805/train0": 5.0, + "1549324805/train1": 12.0, + "1549324805/train2": 1.0, + "1578990937/train0": 5.0, + "1578990937/train1": 11.0, + "1578990937/train2": 7.0, + "1747178030/train0": 9.0, + "1747178030/train1": 7.0, + "1747178030/train2": 11.0, + "1790188787/train0": 5.0, + "1790188787/train1": 6.0, + "1790188787/train2": 8.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 7.0, + "1845489013/train0": 5.0, + "1845489013/train1": 7.0, + "1845489013/train2": 8.0, + "1999022574/train0": 5.0, + "1999022574/train1": 8.0, + "1999022574/train2": 13.0, + "2012211287/train0": 9.0, + "2012211287/train1": 13.0, + "2012211287/train2": 14.0, + "2038594191/train0": 5.0, + "2038594191/train1": 9.0, + "2038594191/train2": 3.0, + "2054899126/train0": 4.5, + "2054899126/train1": 10.0, + "2054899126/train2": 6.5, + "2061635888/train0": 1.0, + "2061635888/train1": 5.0, + "2061635888/train2": 2.0, + "211245842/train0": 11.0, + "211245842/train1": 7.0, + "211245842/train2": 7.0, + "278008524/train0": 8.0, + "278008524/train1": 8.0, + "278008524/train2": 11.0, + "307068879/train0": 11.0, + "307068879/train1": 5.0, + "307068879/train2": 3.0, + "412102563/train0": 10.0, + "412102563/train1": 12.0, + "412102563/train2": 12.0, + "440577711/train0": 10.0, + "440577711/train1": 10.0, + "440577711/train2": 13.0, + "441046872/train0": 7.0, + "441046872/train1": 5.0, + "441046872/train2": 1.0, + "509430392/train0": 11.0, + "509430392/train1": 2.0, + "509430392/train2": 10.0, + "585904130/train0": 4.0, + "585904130/train1": 15.0, + "585904130/train2": 1.0, + "608016290/train0": 12.0, + "608016290/train1": 8.0, + "608016290/train2": 6.0, + "626486813/train0": 11.0, + "626486813/train1": 7.5, + "626486813/train2": 3.0, + "636680137/train0": 2.0, + "636680137/train1": 1.0, + "636680137/train2": 10.0, + "655884605/train0": 0.0, + "655884605/train1": 6.0, + "655884605/train2": 1.0, + "672328262/train0": 5.0, + "672328262/train1": 4.0, + "672328262/train2": 4.0, + "701667486/train0": 2.0, + "701667486/train1": 10.0, + "701667486/train2": 11.0, + "747260437/train0": 15.0, + "747260437/train1": 10.0, + "747260437/train2": 2.0, + "750448296/train0": 6.666666666666667, + "750448296/train1": 7.333333333333333, + "750448296/train2": 6.666666666666667, + "767747424/train0": 5.75, + "767747424/train1": 8.5, + "767747424/train2": 8.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 6.719757575757576, + "low": 6.719757575757576, + "mean": 6.610868686868686, + "n": 5, + "num_samples": 1 + }, + "mean": 6.610868686868686, + "per_task": { + "LiftPegUpright-v1": 7.6, + "PickCube-v1": 5.0, + "PullCube-v1": 7.1866666666666665, + "PushCube-v1": 7.055555555555555, + "StackCube-v1": 6.212121212121212 + } + }, + "metric": "selected_candidate_index", + "micro": { + "confidence": 0.95, + "high": 6.729166666666667, + "low": 6.729166666666667, + "mean": 6.944444444444445, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selected_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.32083333333333336, + "low": 0.32083333333333336, + "mean": 0.33541666666666664, + "n": 120, + "num_samples": 1 + }, + "mean": 0.33541666666666664, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 1.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.5, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 1.0, + "747260437/train1": 1.0, + "747260437/train2": 1.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.17793939393939393, + "low": 0.17793939393939393, + "mean": 0.25127272727272726, + "n": 5, + "num_samples": 1 + }, + "mean": 0.25127272727272726, + "per_task": { + "LiftPegUpright-v1": 0.13333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": 0.32, + "PushCube-v1": 0.5, + "StackCube-v1": 0.30303030303030304 + } + }, + "metric": "selected_success", + "micro": { + "confidence": 0.95, + "high": 0.3263888888888889, + "low": 0.3263888888888889, + "mean": 0.3125, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selected_success_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.05416666666666667, + "low": 0.05416666666666667, + "mean": 0.029166666666666667, + "n": 120, + "num_samples": 1 + }, + "mean": 0.029166666666666667, + "per_seed": { + "1073161778/train0": -1.0, + "1073161778/train1": -1.0, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": -1.0, + "2038594191/train1": -1.0, + "2038594191/train2": -1.0, + "2054899126/train0": -0.5, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.0, + "767747424/train1": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": -0.0286060606060606, + "low": -0.0286060606060606, + "mean": 0.051393939393939395, + "n": 5, + "num_samples": 1 + }, + "mean": 0.051393939393939395, + "per_task": { + "LiftPegUpright-v1": -0.06666666666666667, + "PickCube-v1": 0.0, + "PullCube-v1": -0.04, + "PushCube-v1": 0.3333333333333333, + "StackCube-v1": 0.030303030303030304 + } + }, + "metric": "selected_success_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.04861111111111111, + "low": 0.04861111111111111, + "mean": 0.020833333333333332, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selected_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.8341566584464293, + "low": 0.8341566584464293, + "mean": 0.8329322801880253, + "n": 120, + "num_samples": 1 + }, + "mean": 0.8329322801880253, + "per_seed": { + "1073161778/train0": 0.67014479637146, + "1073161778/train1": 0.6947963237762451, + "1073161778/train2": 2.0, + "1113371261/train0": 0.15842342376708984, + "1113371261/train1": 0.1746392846107483, + "1113371261/train2": 0.15842342376708984, + "1151410346/train0": 0.16260439157485962, + "1151410346/train1": 0.8135483264923096, + "1151410346/train2": 0.20950870215892792, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 2.0, + "1324133122/train0": 0.036974772810935974, + "1324133122/train1": 0.07714252173900604, + "1324133122/train2": 0.13575458526611328, + "1324642411/train0": 0.12164360284805298, + "1324642411/train1": 0.1378108412027359, + "1324642411/train2": 0.10647052526473999, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.00942655373364687, + "141984615/train1": 0.030119141563773155, + "141984615/train2": 0.058951716870069504, + "1461069713/train0": 0.011984944343566895, + "1461069713/train1": 0.007486680988222361, + "1461069713/train2": 0.09997648000717163, + "1480856810/train0": 0.07608599960803986, + "1480856810/train1": 0.21981161832809448, + "1480856810/train2": 0.21615806221961975, + "1509980664/train0": 0.8334100246429443, + "1509980664/train1": 0.8334100246429443, + "1509980664/train2": 0.8334100246429443, + "1522272686/train0": 0.5839591026306152, + "1522272686/train1": 0.5839591026306152, + "1522272686/train2": 0.5839591026306152, + "1549324805/train0": 0.03462250903248787, + "1549324805/train1": 0.08073369413614273, + "1549324805/train2": 0.04690708592534065, + "1578990937/train0": 0.14271311461925507, + "1578990937/train1": 0.17168906331062317, + "1578990937/train2": 0.82413649559021, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 2.0, + "1790188787/train1": 0.12476746737957001, + "1790188787/train2": 0.12476746737957001, + "1822210471/train0": 0.07191530615091324, + "1822210471/train1": 0.07191530615091324, + "1822210471/train2": 0.07191530615091324, + "1845489013/train0": 2.0, + "1845489013/train1": 2.0, + "1845489013/train2": 2.0, + "1999022574/train0": 2.0, + "1999022574/train1": 2.0, + "1999022574/train2": 2.0, + "2012211287/train0": 0.6949572563171387, + "2012211287/train1": 0.6900332570075989, + "2012211287/train2": 0.6934840679168701, + "2038594191/train0": 0.04285776615142822, + "2038594191/train1": 0.07674666494131088, + "2038594191/train2": 0.12977690994739532, + "2054899126/train0": 1.0391597636044025, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 0.07349006831645966, + "2061635888/train1": 0.0713038444519043, + "2061635888/train2": 0.05500011146068573, + "211245842/train0": 2.0, + "211245842/train1": 0.19474327564239502, + "211245842/train2": 0.19090841710567474, + "278008524/train0": 0.06754684634506702, + "278008524/train1": 0.034389604814350605, + "278008524/train2": 0.047564467415213585, + "307068879/train0": 0.6000905632972717, + "307068879/train1": 0.6000905632972717, + "307068879/train2": 0.6000905632972717, + "412102563/train0": 2.0, + "412102563/train1": 2.0, + "412102563/train2": 2.0, + "440577711/train0": 2.0, + "440577711/train1": 2.0, + "440577711/train2": 2.0, + "441046872/train0": 0.6999551653862, + "441046872/train1": 0.6999551653862, + "441046872/train2": 0.6999551653862, + "509430392/train0": 2.0, + "509430392/train1": 2.0, + "509430392/train2": 5.424022674560547e-06, + "585904130/train0": 0.09919407218694687, + "585904130/train1": 0.04792279005050659, + "585904130/train2": 2.0, + "608016290/train0": 2.0, + "608016290/train1": 0.7062739729881287, + "608016290/train2": 0.22811375558376312, + "626486813/train0": 0.009900043485686183, + "626486813/train1": 0.0112088123569265, + "626486813/train2": 0.016004920471459627, + "636680137/train0": 0.08826130628585815, + "636680137/train1": 0.017789265140891075, + "636680137/train2": 0.039896368980407715, + "655884605/train0": 0.12222731113433838, + "655884605/train1": 0.11176733672618866, + "655884605/train2": 0.09131735563278198, + "672328262/train0": 2.0, + "672328262/train1": 2.0, + "672328262/train2": 2.0, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 2.0, + "747260437/train1": 2.0, + "747260437/train2": 2.0, + "750448296/train0": 0.09184517463048299, + "750448296/train1": 0.109618179500103, + "750448296/train2": 0.09844311699271202, + "767747424/train0": 0.5295817001024261, + "767747424/train1": 0.5702044883510098, + "767747424/train2": 0.5541218034923077 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7409482852775378, + "low": 0.7409482852775378, + "mean": 0.7897704216001951, + "n": 5, + "num_samples": 1 + }, + "mean": 0.7897704216001951, + "per_task": { + "LiftPegUpright-v1": 0.8185293744007747, + "PickCube-v1": 0.5839591026306152, + "PullCube-v1": 0.6814040022250265, + "PushCube-v1": 1.062640056014061, + "StackCube-v1": 0.8023195727304979 + } + }, + "metric": "selected_utility", + "micro": { + "confidence": 0.95, + "high": 0.781853584986594, + "low": 0.781853584986594, + "mean": 0.7690221180495832, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selected_utility_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.08679503801653886, + "low": 0.08679503801653886, + "mean": 0.042814217890231196, + "n": 120, + "num_samples": 1 + }, + "mean": 0.042814217890231196, + "per_seed": { + "1073161778/train0": -1.32985520362854, + "1073161778/train1": -1.3052036762237549, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.016215860843658447, + "1113371261/train2": 0.0, + "1151410346/train0": -0.06339690089225769, + "1151410346/train1": 0.5875470340251923, + "1151410346/train2": -0.016492590308189392, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": -0.055230140686035156, + "1324133122/train1": -0.015062391757965088, + "1324133122/train2": 0.04354967176914215, + "1324642411/train0": 0.015173077583312988, + "1324642411/train1": 0.03134031593799591, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": -0.09340769331902266, + "141984615/train1": -0.07271510548889637, + "141984615/train2": -0.04388253018260002, + "1461069713/train0": -0.08191812038421631, + "1461069713/train1": -0.08641638373956084, + "1461069713/train2": 0.006073415279388428, + "1480856810/train0": -0.010948076844215393, + "1480856810/train1": 0.13277754187583923, + "1480856810/train2": 0.1291239857673645, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": -0.06989232823252678, + "1549324805/train1": -0.023781143128871918, + "1549324805/train2": -0.057607751339673996, + "1578990937/train0": -0.6814233809709549, + "1578990937/train1": -0.6524474322795868, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.87523253262043, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": -0.028648078441619873, + "2012211287/train1": -0.03357207775115967, + "2012211287/train2": -0.030121266841888428, + "2038594191/train0": -1.9571422338485718, + "2038594191/train1": -1.9232533350586891, + "2038594191/train2": -1.8702230900526047, + "2054899126/train0": -0.9608402363955975, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.021239563822746277, + "2061635888/train1": 0.019053339958190918, + "2061635888/train2": 0.002749606966972351, + "211245842/train0": 1.8090915828943253, + "211245842/train1": 0.003834858536720276, + "211245842/train2": 0.0, + "278008524/train0": 0.02481185644865036, + "278008524/train1": -0.008345385082066059, + "278008524/train2": 0.004829477518796921, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 1.9638472199440002, + "440577711/train1": 1.9638472199440002, + "440577711/train2": 1.9638472199440002, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.999999776482582, + "509430392/train1": 1.999999776482582, + "509430392/train2": 5.200505256652832e-06, + "585904130/train0": -0.0825214758515358, + "585904130/train1": -0.13379275798797607, + "585904130/train2": 1.8182844519615173, + "608016290/train0": 1.274298369884491, + "608016290/train1": -0.01942765712738037, + "608016290/train2": -0.4975878745317459, + "626486813/train0": 0.004801700764801353, + "626486813/train1": 0.006110469636041671, + "626486813/train2": 0.010906577750574797, + "636680137/train0": 0.005126260221004486, + "636680137/train1": -0.0653457809239626, + "636680137/train2": -0.04323867708444595, + "655884605/train0": -0.040779367089271545, + "655884605/train1": -0.051239341497421265, + "655884605/train2": -0.07168932259082794, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": -0.02574797471364339, + "750448296/train1": -0.007974969844023386, + "750448296/train2": -0.019150032351414364, + "767747424/train0": -0.0436176419025287, + "767747424/train1": -0.0029948536539450288, + "767747424/train2": -0.019077538512647152 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": -0.06610478388312076, + "low": -0.06610478388312076, + "mean": 0.09421062442617999, + "n": 5, + "num_samples": 1 + }, + "mean": 0.09421062442617999, + "per_task": { + "LiftPegUpright-v1": -0.13134116431077322, + "PickCube-v1": 0.0, + "PullCube-v1": -0.09376645899843425, + "PushCube-v1": 0.6702358772357305, + "StackCube-v1": 0.025924868204376915 + } + }, + "metric": "selected_utility_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.08719008552240363, + "low": 0.08719008552240363, + "mean": 0.027202531607245974, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selector_regret": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5733484660673033, + "low": 0.5733484660673033, + "mean": 0.4843037031083239, + "n": 120, + "num_samples": 1 + }, + "mean": 0.4843037031083239, + "per_seed": { + "1073161778/train0": 1.32985520362854, + "1073161778/train1": 1.3052036762237549, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.006292939186096191, + "1113371261/train2": 0.036848247051239014, + "1151410346/train0": 0.7037477493286133, + "1151410346/train1": 1.1864516735076904, + "1151410346/train2": 0.34269602596759796, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.01801443099975586, + "1324133122/train1": 0.05531613528728485, + "1324133122/train2": 0.05715706944465637, + "1324642411/train0": 0.02241644263267517, + "1324642411/train1": 1.862189158797264, + "1324642411/train2": 1.89352947473526, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.005751351825892925, + "141984615/train1": 0.006891986355185509, + "141984615/train2": 0.0, + "1461069713/train0": 0.048213422298431396, + "1461069713/train1": 0.009098331909626722, + "1461069713/train2": 0.0, + "1480856810/train0": 1.9239140003919601, + "1480856810/train1": 1.7801883816719055, + "1480856810/train2": 1.7838419377803802, + "1509980664/train0": 1.1665899753570557, + "1509980664/train1": 1.1665899753570557, + "1509980664/train2": 1.1665899753570557, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.026248913258314133, + "1549324805/train1": 0.11818782240152359, + "1549324805/train2": 0.06619475409388542, + "1578990937/train0": 1.857286885380745, + "1578990937/train1": 1.8283109366893768, + "1578990937/train2": 1.17586350440979, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.87523253262043, + "1790188787/train2": 1.87523253262043, + "1822210471/train0": 1.9280846938490868, + "1822210471/train1": 1.9280846938490868, + "1822210471/train2": 1.9280846938490868, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 1.3050427436828613, + "2012211287/train1": 1.3099667429924011, + "2012211287/train2": 1.3065159320831299, + "2038594191/train0": 1.9571422338485718, + "2038594191/train1": 1.9232533350586891, + "2038594191/train2": 1.8702230900526047, + "2054899126/train0": 0.03812338784337044, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.024509236216545105, + "2061635888/train1": 0.004499524831771851, + "2061635888/train2": 0.056051209568977356, + "211245842/train0": 0.0, + "211245842/train1": 0.018213897943496704, + "211245842/train2": 0.04297435283660889, + "278008524/train0": 1.0058094765990973, + "278008524/train1": 1.0032849898561835, + "278008524/train2": 1.0883844438940287, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 1.3000448346138, + "441046872/train1": 0.023403704166412354, + "441046872/train2": 0.019179701805114746, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 1.9008059278130531, + "585904130/train1": 1.9520772099494934, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 1.2937260270118713, + "608016290/train2": 0.48361386358737946, + "626486813/train0": 0.034795432118698955, + "626486813/train1": 0.024619062314741313, + "626486813/train2": 0.028894334565848112, + "636680137/train0": 0.0, + "636680137/train1": 0.04962191917002201, + "636680137/train2": 0.035755254328250885, + "655884605/train0": 0.015101373195648193, + "655884605/train1": 0.7011084705591202, + "655884605/train2": 0.12315595149993896, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.6608151532709599, + "750448296/train1": 1.2608850946029027, + "750448296/train2": 1.2520055510103703, + "767747424/train0": 0.5128474876983091, + "767747424/train1": 0.49274860287550837, + "767747424/train2": 0.513045291416347 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.43006654029548397, + "low": 0.43006654029548397, + "mean": 0.4456741130232171, + "n": 5, + "num_samples": 1 + }, + "mean": 0.4456741130232171, + "per_task": { + "LiftPegUpright-v1": 0.6451034953196844, + "PickCube-v1": 0.0, + "PullCube-v1": 0.5039629975054413, + "PushCube-v1": 0.7231413589583503, + "StackCube-v1": 0.3561627133326097 + } + }, + "metric": "selector_regret", + "micro": { + "confidence": 0.95, + "high": 0.4535791254343672, + "low": 0.4535791254343672, + "mean": 0.501692300305068, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "success_selector_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.34722222222222215, + "low": 0.34722222222222215, + "mean": 0.2743055555555555, + "n": 120, + "num_samples": 1 + }, + "mean": 0.2743055555555555, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 1.0, + "1151410346/train2": 0.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.5, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 1.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 1.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.27975757575757576, + "low": 0.27975757575757576, + "mean": 0.2642020202020202, + "n": 5, + "num_samples": 1 + }, + "mean": 0.2642020202020202, + "per_task": { + "LiftPegUpright-v1": 0.4666666666666667, + "PickCube-v1": 0.0, + "PullCube-v1": 0.25333333333333335, + "PushCube-v1": 0.3888888888888889, + "StackCube-v1": 0.21212121212121213 + } + }, + "metric": "success_selector_gap", + "micro": { + "confidence": 0.95, + "high": 0.25, + "low": 0.25, + "mean": 0.2777777777777778, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "success_support_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.18472222222222223, + "low": 0.18472222222222223, + "mean": 0.23194444444444443, + "n": 120, + "num_samples": 1 + }, + "mean": 0.23194444444444443, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 1.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 1.0, + "1151410346/train1": 0.0, + "1151410346/train2": 1.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 1.0, + "1324133122/train1": 1.0, + "1324133122/train2": 1.0, + "1324642411/train0": 1.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 1.0, + "1522272686/train1": 1.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.5, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 0.0, + "211245842/train1": 1.0, + "211245842/train2": 1.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 1.0, + "307068879/train1": 1.0, + "307068879/train2": 1.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 0.0, + "441046872/train1": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 1.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 1.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.0, + "767747424/train1": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4623030303030303, + "low": 0.4623030303030303, + "mean": 0.40452525252525257, + "n": 5, + "num_samples": 1 + }, + "mean": 0.40452525252525257, + "per_task": { + "LiftPegUpright-v1": 0.4, + "PickCube-v1": 1.0, + "PullCube-v1": 0.02666666666666667, + "PushCube-v1": 0.1111111111111111, + "StackCube-v1": 0.48484848484848486 + } + }, + "metric": "success_support_gap", + "micro": { + "confidence": 0.95, + "high": 0.19444444444444445, + "low": 0.19444444444444445, + "mean": 0.2013888888888889, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "support_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.3546240135134819, + "low": 0.3546240135134819, + "mean": 0.4232947583840643, + "n": 120, + "num_samples": 1 + }, + "mean": 0.4232947583840643, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 1.897251382470131, + "1113371261/train1": 1.8190677762031555, + "1113371261/train2": 1.8047283291816711, + "1151410346/train0": 1.133647859096527, + "1151410346/train1": 0.0, + "1151410346/train2": 1.4477952718734741, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 1.9450107961893082, + "1324133122/train1": 1.867541342973709, + "1324133122/train2": 1.8070883452892303, + "1324642411/train0": 1.8559399545192719, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.270599365234375, + "141984615/train1": 0.24876614287495613, + "141984615/train2": 0.2268255539238453, + "1461069713/train0": 0.0991440862417221, + "1461069713/train1": 0.14275743998587132, + "1461069713/train2": 0.05936597287654877, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 1.8167466223239899, + "1522272686/train1": 1.9298561662435532, + "1522272686/train2": 1.8506269305944443, + "1549324805/train0": 0.23658466339111328, + "1549324805/train1": 0.09853456914424896, + "1549324805/train2": 0.1843542456626892, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.922716848552227, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 1.9020006954669952, + "2061635888/train1": 1.9241966307163239, + "2061635888/train2": 1.888948678970337, + "211245842/train0": 0.0, + "211245842/train1": 1.7870428264141083, + "211245842/train2": 1.7661172300577164, + "278008524/train0": 0.04055231809616089, + "278008524/train1": 0.07623404636979103, + "278008524/train2": 0.0, + "307068879/train0": 1.7884424477815628, + "307068879/train1": 1.4123926758766174, + "307068879/train2": 1.7697865515947342, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 0.0, + "441046872/train1": 1.2766411304473877, + "441046872/train2": 1.2808651328086853, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 1.9999945759773254, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 1.2882723808288574, + "626486813/train0": 0.12161866808310151, + "626486813/train1": 0.12975450651720166, + "626486813/train2": 0.12068312615156174, + "636680137/train0": 0.0823030173778534, + "636680137/train1": 0.10315313935279846, + "636680137/train2": 0.09491270035505295, + "655884605/train0": 1.8626713156700134, + "655884605/train1": 1.1871241927146912, + "655884605/train2": 1.785526692867279, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.17026761174201965, + "747260437/train1": 0.21537372469902039, + "747260437/train2": 0.15356136858463287, + "750448296/train0": 0.6553361676633358, + "750448296/train1": 0.03749322146177292, + "750448296/train2": 0.05754782756169637, + "767747424/train0": 0.06562284985557199, + "767747424/train1": 0.045098946429789066, + "767747424/train2": 0.04088494274765253 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7984061056091026, + "low": 0.7984061056091026, + "mean": 0.7236978183567826, + "n": 5, + "num_samples": 1 + }, + "mean": 0.7236978183567826, + "per_task": { + "LiftPegUpright-v1": 0.5877600212891897, + "PickCube-v1": 1.8657432397206624, + "PullCube-v1": 0.1075624164690574, + "PushCube-v1": 0.21421858502758873, + "StackCube-v1": 0.8432048292774142 + } + }, + "metric": "support_gap", + "micro": { + "confidence": 0.95, + "high": 0.37098560981555945, + "low": 0.37098560981555945, + "mean": 0.3761285081272945, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "tau": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.11806497094383707, + "low": 0.11806497094383707, + "mean": 0.11780790182376706, + "n": 120, + "num_samples": 1 + }, + "mean": 0.11780790182376706, + "per_seed": { + "1073161778/train0": 0.0940318405174132, + "1073161778/train1": 0.0940318405174132, + "1073161778/train2": 0.0940318405174132, + "1113371261/train0": 0.0009112489076126273, + "1113371261/train1": 0.0009112489076126273, + "1113371261/train2": 0.0009112489076126273, + "1151410346/train0": 0.0009112489076126273, + "1151410346/train1": 0.0009112489076126273, + "1151410346/train2": 0.0009112489076126273, + "1261930456/train0": 0.35241155802994517, + "1261930456/train1": 0.35241155802994517, + "1261930456/train2": 0.35241155802994517, + "1324133122/train0": 0.0009112489076126273, + "1324133122/train1": 0.0009112489076126273, + "1324133122/train2": 0.0009112489076126273, + "1324642411/train0": 0.35241155802994517, + "1324642411/train1": 0.35241155802994517, + "1324642411/train2": 0.35241155802994517, + "1375492786/train0": 0.10332537601561143, + "1375492786/train1": 0.10332537601561143, + "1375492786/train2": 0.10332537601561143, + "141984615/train0": 0.10332537601561143, + "141984615/train1": 0.10332537601561143, + "141984615/train2": 0.10332537601561143, + "1461069713/train0": 0.10332537601561143, + "1461069713/train1": 0.10332537601561143, + "1461069713/train2": 0.10332537601561143, + "1480856810/train0": 0.35241155802994517, + "1480856810/train1": 0.35241155802994517, + "1480856810/train2": 0.35241155802994517, + "1509980664/train0": 0.0009112489076126273, + "1509980664/train1": 0.0009112489076126273, + "1509980664/train2": 0.0009112489076126273, + "1522272686/train0": 0.3611323919348086, + "1522272686/train1": 0.3611323919348086, + "1522272686/train2": 0.3611323919348086, + "1549324805/train0": 0.10332537601561143, + "1549324805/train1": 0.10332537601561143, + "1549324805/train2": 0.10332537601561143, + "1578990937/train0": 0.0009112489076126273, + "1578990937/train1": 0.0009112489076126273, + "1578990937/train2": 0.0009112489076126273, + "1747178030/train0": 0.10332537601561143, + "1747178030/train1": 0.10332537601561143, + "1747178030/train2": 0.10332537601561143, + "1790188787/train0": 0.35241155802994517, + "1790188787/train1": 0.35241155802994517, + "1790188787/train2": 0.35241155802994517, + "1822210471/train0": 0.10332537601561143, + "1822210471/train1": 0.10332537601561143, + "1822210471/train2": 0.10332537601561143, + "1845489013/train0": 0.10332537601561143, + "1845489013/train1": 0.10332537601561143, + "1845489013/train2": 0.10332537601561143, + "1999022574/train0": 0.0009112489076126273, + "1999022574/train1": 0.0009112489076126273, + "1999022574/train2": 0.0009112489076126273, + "2012211287/train0": 0.0940318405174132, + "2012211287/train1": 0.0940318405174132, + "2012211287/train2": 0.0940318405174132, + "2038594191/train0": 0.10332537601561143, + "2038594191/train1": 0.10332537601561143, + "2038594191/train2": 0.10332537601561143, + "2054899126/train0": 0.10332537601561143, + "2054899126/train1": 0.10332537601561143, + "2054899126/train2": 0.10332537601561143, + "2061635888/train0": 0.0009112489076126273, + "2061635888/train1": 0.0009112489076126273, + "2061635888/train2": 0.0009112489076126273, + "211245842/train0": 0.0009112489076126273, + "211245842/train1": 0.0009112489076126273, + "211245842/train2": 0.0009112489076126273, + "278008524/train0": 0.10332537601561143, + "278008524/train1": 0.10332537601561143, + "278008524/train2": 0.10332537601561143, + "307068879/train0": 0.0940318405174132, + "307068879/train1": 0.0940318405174132, + "307068879/train2": 0.0940318405174132, + "412102563/train0": 0.0009112489076126273, + "412102563/train1": 0.0009112489076126273, + "412102563/train2": 0.0009112489076126273, + "440577711/train0": 0.35241155802994517, + "440577711/train1": 0.35241155802994517, + "440577711/train2": 0.35241155802994517, + "441046872/train0": 0.0940318405174132, + "441046872/train1": 0.0940318405174132, + "441046872/train2": 0.0940318405174132, + "509430392/train0": 0.35241155802994517, + "509430392/train1": 0.35241155802994517, + "509430392/train2": 0.35241155802994517, + "585904130/train0": 0.10332537601561143, + "585904130/train1": 0.10332537601561143, + "585904130/train2": 0.10332537601561143, + "608016290/train0": 0.0940318405174132, + "608016290/train1": 0.0940318405174132, + "608016290/train2": 0.0940318405174132, + "626486813/train0": 0.10332537601561143, + "626486813/train1": 0.10332537601561143, + "626486813/train2": 0.10332537601561143, + "636680137/train0": 0.10332537601561143, + "636680137/train1": 0.10332537601561143, + "636680137/train2": 0.10332537601561143, + "655884605/train0": 0.0009112489076126273, + "655884605/train1": 0.0009112489076126273, + "655884605/train2": 0.0009112489076126273, + "672328262/train0": 0.10332537601561143, + "672328262/train1": 0.10332537601561143, + "672328262/train2": 0.10332537601561143, + "701667486/train0": 0.0009112489076126273, + "701667486/train1": 0.0009112489076126273, + "701667486/train2": 0.0009112489076126273, + "747260437/train0": 0.10332537601561143, + "747260437/train1": 0.10332537601561143, + "747260437/train2": 0.10332537601561143, + "750448296/train0": 0.10332537601561143, + "750448296/train1": 0.10332537601561143, + "750448296/train2": 0.10332537601561143, + "767747424/train0": 0.10332537601561143, + "767747424/train1": 0.10332537601561143, + "767747424/train2": 0.10332537601561143 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.13068653957857182, + "low": 0.13068653957857182, + "mean": 0.1823624830810782, + "n": 5, + "num_samples": 1 + }, + "mean": 0.1823624830810782, + "per_task": { + "LiftPegUpright-v1": 0.0940318405174132, + "PickCube-v1": 0.3611323919348086, + "PullCube-v1": 0.10332537601561143, + "PushCube-v1": 0.35241155802994506, + "StackCube-v1": 0.0009112489076126273 + } + }, + "metric": "tau", + "micro": { + "confidence": 0.95, + "high": 0.11515133924254574, + "low": 0.11515133924254574, + "mean": 0.11539414752240787, + "n": 144, + "num_samples": 1 + }, + "n": 144 + } + }, + "target": "utility_margin", + "tau": { + "LiftPegUpright-v1": 0.0940318405174132, + "PickCube-v1": 0.3611323919348086, + "PullCube-v1": 0.10332537601561143, + "PushCube-v1": 0.35241155802994517, + "StackCube-v1": 0.0009112489076126273, + "__global__": 0.1132010828047253 + }, + "threshold_scope": "task", + "weights": [ + -0.06419078293957127, + 0.12788216234575267, + 0.051170976803687845, + -0.028420492753592114, + 0.03337635695444696, + 0.002369880315031214, + -0.09672508477366085, + -0.2068227001071573, + 0.08161746543544754, + 0.4524576094090067, + 2.527134440574412e-15, + 0.039458049241851256, + -0.02308751172060397, + 0.4238617940529128, + 0.6954725703846307, + -0.10180357933288309, + -0.8182556836231556, + -0.005163758843360622, + -0.08034723257745924, + -0.15452403971399065, + 0.035208002578563935, + 0.6951010350739004, + 1.3594537975155778e-14, + 0.3026471745950607, + 0.36129349622109597, + -0.004826842662609465, + -0.1183715558662008, + 2.2072065050528713e-14, + -1.3225619485371112e-14, + -1.3278969877370364e-15, + -1.9202943211584072e-14, + -5.69884778305397e-15, + -0.4187017119716985, + 0.06103537160295325, + -0.17027060234492541, + 0.14957242823254863, + 0.09943095405665581, + -0.05548909714357808, + 0.0012514703250900427, + 0.02057053122431371, + -0.0042858025214529, + 0.13563555684566087, + -0.2120341171786455, + -0.2453339010385296, + 0.325688065400646, + -0.23292124141248305, + -0.06582709621579497, + 1.774283918510806e-15, + -0.06923220170961447, + -0.08290059793536919, + -0.004361530595525798 + ] +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/metrics_by_seed.json b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/metrics_by_seed.json new file mode 100644 index 0000000000000000000000000000000000000000..a6e3abde08c23a49763ad274219175d7091f6453 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/metrics_by_seed.json @@ -0,0 +1,3236 @@ +{ + "1073161778/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5362042832267765, + "pairwise_causal_calibration_ece": 0.07639944182594278, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.24085858055337617, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.67014479637146, + "selected_utility_gain_over_base": -1.32985520362854, + "selector_regret": 1.32985520362854, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0940318405174132 + }, + "1073161778/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5416666666666666, + "pairwise_causal_calibration_confidence": 0.5477005041063188, + "pairwise_causal_calibration_ece": 0.12867242656969963, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.11864476132183588, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.6947963237762451, + "selected_utility_gain_over_base": -1.3052036762237549, + "selector_regret": 1.3052036762237549, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0940318405174132 + }, + "1073161778/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.8095238095238095, + "pairwise_causal_calibration_confidence": 0.5547882621085445, + "pairwise_causal_calibration_ece": 0.25473554741526483, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.24206048374838196, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 6.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0940318405174132 + }, + "1113371261/train0": { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5401592730365951, + "pairwise_causal_calibration_ece": 0.04015927303659509, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.002655981453428158, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10274861752986908, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.897251382470131, + "tau": 0.0009112489076126273 + }, + "1113371261/train1": { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5380625270265227, + "pairwise_causal_calibration_ece": 0.038062527026522686, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.08193066148519779, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18093222379684448, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.1746392846107483, + "selected_utility_gain_over_base": 0.016215860843658447, + "selector_regret": 0.006292939186096191, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8190677762031555, + "tau": 0.0009112489076126273 + }, + "1113371261/train2": { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.5487087011529058, + "pairwise_causal_calibration_ece": 0.1655770131328084, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.15407578821846468, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "tau": 0.0009112489076126273 + }, + "1151410346/train0": { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5047619047619047, + "pairwise_causal_calibration_confidence": 0.5241846653555288, + "pairwise_causal_calibration_ece": 0.019422760593624022, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.11341562423740303, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8663521409034729, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16260439157485962, + "selected_utility_gain_over_base": -0.06339690089225769, + "selector_regret": 0.7037477493286133, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.133647859096527, + "tau": 0.0009112489076126273 + }, + "1151410346/train1": { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5513917700725205, + "pairwise_causal_calibration_ece": 0.08472510340585364, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.027308203722147817, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8135483264923096, + "selected_utility_gain_over_base": 0.5875470340251923, + "selector_regret": 1.1864516735076904, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "1151410346/train2": { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5313847796953075, + "pairwise_causal_calibration_ece": 0.06471811302864083, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.06561892171053907, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.20950870215892792, + "selected_utility_gain_over_base": -0.016492590308189392, + "selector_regret": 0.34269602596759796, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4477952718734741, + "tau": 0.0009112489076126273 + }, + "1261930456/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.47474747474747475, + "pairwise_causal_calibration_confidence": 0.5407472634490431, + "pairwise_causal_calibration_ece": 0.06599978870156839, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.09035295152992723, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1261930456/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6181818181818182, + "pairwise_causal_calibration_confidence": 0.5491787800871848, + "pairwise_causal_calibration_ece": 0.0690030380946333, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": -0.02356230018007266, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1261930456/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5397391671472074, + "pairwise_causal_calibration_ece": 0.12692749951945942, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.04072195094915404, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1324133122/train0": { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5500982965860671, + "pairwise_causal_calibration_ece": 0.08581258230035274, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.07548993896412504, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.054989203810691833, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.036974772810935974, + "selected_utility_gain_over_base": -0.055230140686035156, + "selector_regret": 0.01801443099975586, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9450107961893082, + "tau": 0.0009112489076126273 + }, + "1324133122/train1": { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5502823354081949, + "pairwise_causal_calibration_ece": 0.015677218286420276, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.3962004123380809, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1324586570262909, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07714252173900604, + "selected_utility_gain_over_base": -0.015062391757965088, + "selector_regret": 0.05531613528728485, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.867541342973709, + "tau": 0.0009112489076126273 + }, + "1324133122/train2": { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.545535241490244, + "pairwise_causal_calibration_ece": 0.07681453301793482, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.0670111567575864, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19291165471076965, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13575458526611328, + "selected_utility_gain_over_base": 0.04354967176914215, + "selector_regret": 0.05715706944465637, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8070883452892303, + "tau": 0.0009112489076126273 + }, + "1324642411/train0": { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5337665023974802, + "pairwise_causal_calibration_ece": 0.032055347656845035, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.711002878905307, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14406004548072815, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12164360284805298, + "selected_utility_gain_over_base": 0.015173077583312988, + "selector_regret": 0.02241644263267517, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8559399545192719, + "tau": 0.35241155802994517 + }, + "1324642411/train1": { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5798319327731093, + "pairwise_causal_calibration_confidence": 0.5288122689723307, + "pairwise_causal_calibration_ece": 0.0510196638007786, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.38447100568728765, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.1378108412027359, + "selected_utility_gain_over_base": 0.03134031593799591, + "selector_regret": 1.862189158797264, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1324642411/train2": { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7, + "pairwise_causal_calibration_confidence": 0.528083048643476, + "pairwise_causal_calibration_ece": 0.17191695135652396, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.15208916465183764, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10647052526473999, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.89352947473526, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1375492786/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6533333333333333, + "pairwise_causal_calibration_confidence": 0.5447046728594015, + "pairwise_causal_calibration_ece": 0.10862866047393183, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": -0.28428690828432274, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1375492786/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5614035087719298, + "pairwise_causal_calibration_confidence": 0.5446679554502398, + "pairwise_causal_calibration_ece": 0.01673555332169, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.6782841630171567, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1375492786/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7380952380952381, + "pairwise_causal_calibration_confidence": 0.565958547539462, + "pairwise_causal_calibration_ece": 0.1721366905557764, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.559803911595051, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "141984615/train0": { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.49166666666666664, + "pairwise_causal_calibration_confidence": 0.5402158636214657, + "pairwise_causal_calibration_ece": 0.048549196954798944, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.30976598855930915, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "selected_candidate_index": 14.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.00942655373364687, + "selected_utility_gain_over_base": -0.09340769331902266, + "selector_regret": 0.005751351825892925, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "tau": 0.10332537601561143 + }, + "141984615/train1": { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5473004480108934, + "pairwise_causal_calibration_ece": 0.046271089584524104, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.39033675889099473, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.037011127918958664, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.030119141563773155, + "selected_utility_gain_over_base": -0.07271510548889637, + "selector_regret": 0.006891986355185509, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24876614287495613, + "tau": 0.10332537601561143 + }, + "141984615/train2": { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5416666666666666, + "pairwise_causal_calibration_confidence": 0.5440555881176564, + "pairwise_causal_calibration_ece": 0.03902364539059017, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.29555407226838537, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058951716870069504, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.058951716870069504, + "selected_utility_gain_over_base": -0.04388253018260002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "tau": 0.10332537601561143 + }, + "1461069713/train0": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.275, + "pairwise_causal_calibration_confidence": 0.5282766199734893, + "pairwise_causal_calibration_ece": 0.25327661997348927, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3059149126148908, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06019836664199829, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.011984944343566895, + "selected_utility_gain_over_base": -0.08191812038421631, + "selector_regret": 0.048213422298431396, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0991440862417221, + "tau": 0.10332537601561143 + }, + "1461069713/train1": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.43333333333333335, + "pairwise_causal_calibration_confidence": 0.5243338816656252, + "pairwise_causal_calibration_ece": 0.09100054833229188, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3007404756181269, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007486680988222361, + "selected_utility_gain_over_base": -0.08641638373956084, + "selector_regret": 0.009098331909626722, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14275743998587132, + "tau": 0.10332537601561143 + }, + "1461069713/train2": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6583333333333333, + "pairwise_causal_calibration_confidence": 0.5225830368338087, + "pairwise_causal_calibration_ece": 0.13575029649952464, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2553375780680003, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09997648000717163, + "selected_utility_gain_over_base": 0.006073415279388428, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05936597287654877, + "tau": 0.10332537601561143 + }, + "1480856810/train0": { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.47474747474747475, + "pairwise_causal_calibration_confidence": 0.5212194722509627, + "pairwise_causal_calibration_ece": 0.046471997503487905, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.6891134156079126, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07608599960803986, + "selected_utility_gain_over_base": -0.010948076844215393, + "selector_regret": 1.9239140003919601, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1480856810/train1": { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.48739495798319327, + "pairwise_causal_calibration_confidence": 0.5158258523764001, + "pairwise_causal_calibration_ece": 0.02843089439320684, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5636198101475918, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21981161832809448, + "selected_utility_gain_over_base": 0.13277754187583923, + "selector_regret": 1.7801883816719055, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1480856810/train2": { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.680672268907563, + "pairwise_causal_calibration_confidence": 0.5204768609791875, + "pairwise_causal_calibration_ece": 0.1601954079283755, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4905519223760793, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21615806221961975, + "selected_utility_gain_over_base": 0.1291239857673645, + "selector_regret": 1.7838419377803802, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1509980664/train0": { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5320927425921708, + "pairwise_causal_calibration_ece": 0.07422005046880752, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.2172662931516871, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "1509980664/train1": { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.2222222222222222, + "pairwise_causal_calibration_confidence": 0.5391700207769446, + "pairwise_causal_calibration_ece": 0.31694779855472244, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": -0.2497394332434142, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "1509980664/train2": { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5111111111111111, + "pairwise_causal_calibration_confidence": 0.5394504298273115, + "pairwise_causal_calibration_ece": 0.02833931871620072, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.3123388737013393, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "1522272686/train0": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5681060848637505, + "pairwise_causal_calibration_ece": 0.07816118582099628, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.4821765811639407, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18325337767601013, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8167466223239899, + "tau": 0.3611323919348086 + }, + "1522272686/train1": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.19444444444444445, + "pairwise_causal_calibration_confidence": 0.5689701709118037, + "pairwise_causal_calibration_ece": 0.37452572646735927, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.25520062215278894, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07014383375644684, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9298561662435532, + "tau": 0.3611323919348086 + }, + "1522272686/train2": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.545306820402758, + "pairwise_causal_calibration_ece": 0.04530682040275796, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.43379366859847046, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14937306940555573, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8506269305944443, + "tau": 0.3611323919348086 + }, + "1549324805/train0": { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7083333333333334, + "pairwise_causal_calibration_confidence": 0.5299868735462288, + "pairwise_causal_calibration_ece": 0.17834645978710464, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8008610893558498, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.060871422290802, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.03462250903248787, + "selected_utility_gain_over_base": -0.06989232823252678, + "selector_regret": 0.026248913258314133, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.23658466339111328, + "tau": 0.10332537601561143 + }, + "1549324805/train1": { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4083333333333333, + "pairwise_causal_calibration_confidence": 0.5265093687069327, + "pairwise_causal_calibration_ece": 0.11817603537359933, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4059182168351142, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "selected_candidate_index": 12.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08073369413614273, + "selected_utility_gain_over_base": -0.023781143128871918, + "selector_regret": 0.11818782240152359, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "tau": 0.10332537601561143 + }, + "1549324805/train2": { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.2833333333333333, + "pairwise_causal_calibration_confidence": 0.541054902787583, + "pairwise_causal_calibration_ece": 0.25772156945424957, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8716902723485557, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11310184001922607, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04690708592534065, + "selected_utility_gain_over_base": -0.057607751339673996, + "selector_regret": 0.06619475409388542, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1843542456626892, + "tau": 0.10332537601561143 + }, + "1578990937/train0": { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6153846153846154, + "pairwise_causal_calibration_confidence": 0.544349322443362, + "pairwise_causal_calibration_ece": 0.07103529294125341, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.10915469051695448, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14271311461925507, + "selected_utility_gain_over_base": -0.6814233809709549, + "selector_regret": 1.857286885380745, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "1578990937/train1": { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4727272727272727, + "pairwise_causal_calibration_confidence": 0.5543665049371578, + "pairwise_causal_calibration_ece": 0.081639232209885, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.10231812655593348, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.17168906331062317, + "selected_utility_gain_over_base": -0.6524474322795868, + "selector_regret": 1.8283109366893768, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "1578990937/train2": { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42016806722689076, + "pairwise_causal_calibration_confidence": 0.540574554638564, + "pairwise_causal_calibration_ece": 0.1204064874116733, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.08662696338330045, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.82413649559021, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.17586350440979, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "1747178030/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.2685368177679842, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1747178030/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.27732487886731, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1747178030/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.3315076869895872, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 11.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1790188787/train0": { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6363636363636364, + "pairwise_causal_calibration_confidence": 0.5392236442621141, + "pairwise_causal_calibration_ece": 0.09713999210152216, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.4590932780625647, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.87523253262043, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1790188787/train1": { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5350877192982456, + "pairwise_causal_calibration_confidence": 0.5396885778611449, + "pairwise_causal_calibration_ece": 0.004600858562899268, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 0.33925681284155934, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12476746737957001, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.87523253262043, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1790188787/train2": { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.44545454545454544, + "pairwise_causal_calibration_confidence": 0.5413352723129059, + "pairwise_causal_calibration_ece": 0.09588072685836047, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.297516022662714, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12476746737957001, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.87523253262043, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "1822210471/train0": { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.575, + "pairwise_causal_calibration_confidence": 0.5565662285154201, + "pairwise_causal_calibration_ece": 0.15721195241478056, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 0.013297597406553785, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07191530615091324, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9280846938490868, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1822210471/train1": { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5510204081632653, + "pairwise_causal_calibration_confidence": 0.5508218552014704, + "pairwise_causal_calibration_ece": 0.14397334425216246, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": 0.041781127110214866, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07191530615091324, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9280846938490868, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1822210471/train2": { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.525, + "pairwise_causal_calibration_confidence": 0.5302663723099226, + "pairwise_causal_calibration_ece": 0.118390411016323, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": -0.09918008580701754, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07191530615091324, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9280846938490868, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1845489013/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5869565217391305, + "pairwise_causal_calibration_confidence": 0.5198092453508348, + "pairwise_causal_calibration_ece": 0.06714727638829565, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": 0.08565911209810279, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1845489013/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7692307692307693, + "pairwise_causal_calibration_confidence": 0.519320720354459, + "pairwise_causal_calibration_ece": 0.2499100488763103, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": 0.10538760252719387, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1845489013/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.52, + "pairwise_causal_calibration_confidence": 0.5338919568560658, + "pairwise_causal_calibration_ece": 0.013891956856065812, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": 0.14874188314117123, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 8.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "1999022574/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3, + "pairwise_causal_calibration_confidence": 0.5344493626989616, + "pairwise_causal_calibration_ece": 0.23444936269896155, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": -0.11161114265900707, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "1999022574/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4714285714285714, + "pairwise_causal_calibration_confidence": 0.5395820832404352, + "pairwise_causal_calibration_ece": 0.06815351181186355, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": -0.1454763298079958, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 8.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "1999022574/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5394736842105263, + "pairwise_causal_calibration_confidence": 0.5438956195294256, + "pairwise_causal_calibration_ece": 0.05003241707833195, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": -0.12321088891053344, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "2012211287/train0": { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.226890756302521, + "pairwise_causal_calibration_confidence": 0.5363365754560178, + "pairwise_causal_calibration_ece": 0.3094458191534968, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.3161539044218536, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6949572563171387, + "selected_utility_gain_over_base": -0.028648078441619873, + "selector_regret": 1.3050427436828613, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0940318405174132 + }, + "2012211287/train1": { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5335084608302744, + "pairwise_causal_calibration_ece": 0.07199480463689167, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.24432178181307274, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6900332570075989, + "selected_utility_gain_over_base": -0.03357207775115967, + "selector_regret": 1.3099667429924011, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0940318405174132 + }, + "2012211287/train2": { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6086956521739131, + "pairwise_causal_calibration_confidence": 0.5293148633692549, + "pairwise_causal_calibration_ece": 0.07938078880465815, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.3345646470523239, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 14.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6934840679168701, + "selected_utility_gain_over_base": -0.030121266841888428, + "selector_regret": 1.3065159320831299, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0940318405174132 + }, + "2038594191/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3793103448275862, + "pairwise_causal_calibration_confidence": 0.6018790607135653, + "pairwise_causal_calibration_ece": 0.22256871588597915, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.28192292963769827, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04285776615142822, + "selected_utility_gain_over_base": -1.9571422338485718, + "selector_regret": 1.9571422338485718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "2038594191/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.42592592592592593, + "pairwise_causal_calibration_confidence": 0.5909326842120879, + "pairwise_causal_calibration_ece": 0.1851395001152688, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.21139315187633212, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07674666494131088, + "selected_utility_gain_over_base": -1.9232533350586891, + "selector_regret": 1.9232533350586891, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "2038594191/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.41414141414141414, + "pairwise_causal_calibration_confidence": 0.5664818478408624, + "pairwise_causal_calibration_ece": 0.1523404336994486, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.2451404604489238, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.12977690994739532, + "selected_utility_gain_over_base": -1.8702230900526047, + "selector_regret": 1.8702230900526047, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "2054899126/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.30833333333333335, + "pairwise_causal_calibration_confidence": 0.5238354835309519, + "pairwise_causal_calibration_ece": 0.10775107509880927, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 0.40089684325539643, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.077283151447773, + "selected_candidate_index": 4.5, + "selected_success": 0.5, + "selected_success_gain_over_base": -0.5, + "selected_utility": 1.0391597636044025, + "selected_utility_gain_over_base": -0.9608402363955975, + "selector_regret": 0.03812338784337044, + "success_selector_gap": 0.0, + "success_support_gap": 0.5, + "support_gap": 0.922716848552227, + "tau": 0.10332537601561143 + }, + "2054899126/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.5, + "execute_generated": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6052631578947368, + "pairwise_causal_calibration_confidence": 0.5244611135518833, + "pairwise_causal_calibration_ece": 0.04040102217142677, + "pairwise_causal_calibration_pairs": 57.0, + "predicted_margin": 0.32373079560635964, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "2054899126/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.5, + "execute_generated": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.47058823529411764, + "pairwise_causal_calibration_confidence": 0.5297648676632524, + "pairwise_causal_calibration_ece": 0.029588316184567365, + "pairwise_causal_calibration_pairs": 59.5, + "predicted_margin": 0.3425010993904758, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 6.5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "2061635888/train0": { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5833333333333334, + "pairwise_causal_calibration_confidence": 0.5351503713504386, + "pairwise_causal_calibration_ece": 0.048182961982894734, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.2859560598368813, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09799930453300476, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07349006831645966, + "selected_utility_gain_over_base": 0.021239563822746277, + "selector_regret": 0.024509236216545105, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9020006954669952, + "tau": 0.0009112489076126273 + }, + "2061635888/train1": { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.518040614785329, + "pairwise_causal_calibration_ece": 0.018040614785329034, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.14710078956282402, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07580336928367615, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0713038444519043, + "selected_utility_gain_over_base": 0.019053339958190918, + "selector_regret": 0.004499524831771851, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9241966307163239, + "tau": 0.0009112489076126273 + }, + "2061635888/train2": { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5252975568525992, + "pairwise_causal_calibration_ece": 0.025297556852599157, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.15541062212467605, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11105132102966309, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05500011146068573, + "selected_utility_gain_over_base": 0.002749606966972351, + "selector_regret": 0.056051209568977356, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.888948678970337, + "tau": 0.0009112489076126273 + }, + "211245842/train0": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7058823529411765, + "pairwise_causal_calibration_confidence": 0.5545781832114899, + "pairwise_causal_calibration_ece": 0.1513041697296866, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.05542394657973021, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 11.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8090915828943253, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "211245842/train1": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7916666666666666, + "pairwise_causal_calibration_confidence": 0.5498984873423876, + "pairwise_causal_calibration_ece": 0.2417681793242794, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2276424215366616, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19474327564239502, + "selected_utility_gain_over_base": 0.003834858536720276, + "selector_regret": 0.018213897943496704, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7870428264141083, + "tau": 0.0009112489076126273 + }, + "211245842/train2": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.55, + "pairwise_causal_calibration_confidence": 0.5458205639974503, + "pairwise_causal_calibration_ece": 0.026780026272463814, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.05502138612288987, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04297435283660889, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7661172300577164, + "tau": 0.0009112489076126273 + }, + "278008524/train0": { + "base_success": 0.0, + "base_utility": 0.042734989896416664, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5335858585858586, + "pairwise_causal_calibration_confidence": 0.5477839621589006, + "pairwise_causal_calibration_ece": 0.14072928673079005, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.28802122449631784, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0733563229441643, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06754684634506702, + "selected_utility_gain_over_base": 0.02481185644865036, + "selector_regret": 1.0058094765990973, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.04055231809616089, + "tau": 0.10332537601561143 + }, + "278008524/train1": { + "base_success": 0.0, + "base_utility": 0.042734989896416664, + "coverage": 0.5, + "execute_generated": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5306159420289855, + "pairwise_causal_calibration_confidence": 0.5479136366856556, + "pairwise_causal_calibration_ece": 0.09383678611256409, + "pairwise_causal_calibration_pairs": 106.0, + "predicted_margin": 0.3601165498724468, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0376745946705341, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.034389604814350605, + "selected_utility_gain_over_base": -0.008345385082066059, + "selector_regret": 1.0032849898561835, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.07623404636979103, + "tau": 0.10332537601561143 + }, + "278008524/train2": { + "base_success": 0.0, + "base_utility": 0.042734989896416664, + "coverage": 0.5, + "execute_generated": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.35833333333333334, + "pairwise_causal_calibration_confidence": 0.5528248393150181, + "pairwise_causal_calibration_ece": 0.19449150598168496, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.22015899787097548, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.1359489113092422, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.047564467415213585, + "selected_utility_gain_over_base": 0.004829477518796921, + "selector_regret": 1.0883844438940287, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "307068879/train0": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6153846153846154, + "pairwise_causal_calibration_confidence": 0.5488728190719444, + "pairwise_causal_calibration_ece": 0.06651179631267089, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": -0.14233288142692663, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2115575522184372, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7884424477815628, + "tau": 0.0940318405174132 + }, + "307068879/train1": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5934065934065934, + "pairwise_causal_calibration_confidence": 0.5348399847973542, + "pairwise_causal_calibration_ece": 0.06342054136618626, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": -0.13522650945094028, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5876073241233826, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "tau": 0.0940318405174132 + }, + "307068879/train2": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4945054945054945, + "pairwise_causal_calibration_confidence": 0.5640014996133708, + "pairwise_causal_calibration_ece": 0.06949600510787626, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": -0.02519869144581416, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2302134484052658, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7697865515947342, + "tau": 0.0940318405174132 + }, + "412102563/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5413119316483767, + "pairwise_causal_calibration_ece": 0.007978598315043392, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": -0.07414481146410684, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "412102563/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5384615384615384, + "pairwise_causal_calibration_confidence": 0.5307516643020463, + "pairwise_causal_calibration_ece": 0.007709874159492092, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": -0.10576298261783212, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "412102563/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5952380952380952, + "pairwise_causal_calibration_confidence": 0.5400007019106284, + "pairwise_causal_calibration_ece": 0.05523739332746688, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.10754995958912039, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "440577711/train0": { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7586206896551724, + "pairwise_causal_calibration_confidence": 0.5267661600012826, + "pairwise_causal_calibration_ece": 0.23185452965388975, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.7982832696690534, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "440577711/train1": { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.5289194284606322, + "pairwise_causal_calibration_ece": 0.07108057153936775, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 0.862893346101075, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "440577711/train2": { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6111111111111112, + "pairwise_causal_calibration_confidence": 0.5489590881909069, + "pairwise_causal_calibration_ece": 0.06215202292020417, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.7945693500962356, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "441046872/train0": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.41818181818181815, + "pairwise_causal_calibration_confidence": 0.5379829028529084, + "pairwise_causal_calibration_ece": 0.11980108467109023, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.024167399539588352, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.3000448346138, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0940318405174132 + }, + "441046872/train1": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5818181818181818, + "pairwise_causal_calibration_confidence": 0.5337021369933691, + "pairwise_causal_calibration_ece": 0.04811604482481282, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.11694270182946509, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7233588695526123, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.023403704166412354, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "tau": 0.0940318405174132 + }, + "441046872/train2": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7272727272727273, + "pairwise_causal_calibration_confidence": 0.5454083397120673, + "pairwise_causal_calibration_ece": 0.18186438756066006, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.12738205708571582, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7191348671913147, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019179701805114746, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "tau": 0.0940318405174132 + }, + "509430392/train0": { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6050420168067226, + "pairwise_causal_calibration_confidence": 0.5254427085184519, + "pairwise_causal_calibration_ece": 0.07959930828827078, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5885139744079059, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 11.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "509430392/train1": { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5904761904761905, + "pairwise_causal_calibration_confidence": 0.5382862696863612, + "pairwise_causal_calibration_ece": 0.052189920789829286, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.5296709902746183, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.35241155802994517 + }, + "509430392/train2": { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.5437150122698521, + "pairwise_causal_calibration_ece": 0.06973036588140841, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4810616128884794, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 5.424022674560547e-06, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 5.424022674560547e-06, + "selected_utility_gain_over_base": 5.200505256652832e-06, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9999945759773254, + "tau": 0.35241155802994517 + }, + "585904130/train0": { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4523809523809524, + "pairwise_causal_calibration_confidence": 0.5292002900283871, + "pairwise_causal_calibration_ece": 0.0768193376474347, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.23094602840169032, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09919407218694687, + "selected_utility_gain_over_base": -0.0825214758515358, + "selector_regret": 1.9008059278130531, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "585904130/train1": { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5466666666666666, + "pairwise_causal_calibration_confidence": 0.5343887712270259, + "pairwise_causal_calibration_ece": 0.050834006610117985, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.22575925280556675, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04792279005050659, + "selected_utility_gain_over_base": -0.13379275798797607, + "selector_regret": 1.9520772099494934, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "585904130/train2": { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7272727272727273, + "pairwise_causal_calibration_confidence": 0.5282487650578878, + "pairwise_causal_calibration_ece": 0.19902396221483953, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.18843222973041862, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8182844519615173, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "608016290/train0": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5619047619047619, + "pairwise_causal_calibration_confidence": 0.5498437058747719, + "pairwise_causal_calibration_ece": 0.01206105602998994, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.48611772128013164, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0940318405174132 + }, + "608016290/train1": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.47619047619047616, + "pairwise_causal_calibration_confidence": 0.5706245241886934, + "pairwise_causal_calibration_ece": 0.09443404799821745, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.2549086170487004, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7062739729881287, + "selected_utility_gain_over_base": -0.01942765712738037, + "selector_regret": 1.2937260270118713, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0940318405174132 + }, + "608016290/train2": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.48333333333333334, + "pairwise_causal_calibration_confidence": 0.5615509166352838, + "pairwise_causal_calibration_ece": 0.0782175833019504, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4202802138182712, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7117276191711426, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.22811375558376312, + "selected_utility_gain_over_base": -0.4975878745317459, + "selector_regret": 0.48361386358737946, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "tau": 0.0940318405174132 + }, + "626486813/train0": { + "base_success": 0.0, + "base_utility": 0.00509834272088483, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.16558238118886948, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6125, + "pairwise_causal_calibration_confidence": 0.5678117074464072, + "pairwise_causal_calibration_ece": 0.09417264307778664, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5713883592563933, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04469547560438514, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.009900043485686183, + "selected_utility_gain_over_base": 0.004801700764801353, + "selector_regret": 0.034795432118698955, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12161866808310151, + "tau": 0.10332537601561143 + }, + "626486813/train1": { + "base_success": 0.0, + "base_utility": 0.00509834272088483, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.16558238118886948, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.39166666666666666, + "pairwise_causal_calibration_confidence": 0.5645620394062392, + "pairwise_causal_calibration_ece": 0.19860544920331658, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5558101781509216, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.035827874671667814, + "selected_candidate_index": 7.5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0112088123569265, + "selected_utility_gain_over_base": 0.006110469636041671, + "selector_regret": 0.024619062314741313, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12975450651720166, + "tau": 0.10332537601561143 + }, + "626486813/train2": { + "base_success": 0.0, + "base_utility": 0.00509834272088483, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.16558238118886948, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.525, + "pairwise_causal_calibration_confidence": 0.5613104362626665, + "pairwise_causal_calibration_ece": 0.07789314572719971, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3853803956007696, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04489925503730774, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.016004920471459627, + "selected_utility_gain_over_base": 0.010906577750574797, + "selector_regret": 0.028894334565848112, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12068312615156174, + "tau": 0.10332537601561143 + }, + "636680137/train0": { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.49166666666666664, + "pairwise_causal_calibration_confidence": 0.532502870801132, + "pairwise_causal_calibration_ece": 0.04083620413446535, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5991343617786198, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.08826130628585815, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08826130628585815, + "selected_utility_gain_over_base": 0.005126260221004486, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0823030173778534, + "tau": 0.10332537601561143 + }, + "636680137/train1": { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5173967032259332, + "pairwise_causal_calibration_ece": 0.015936630107400096, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.19155725528837464, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06741118431091309, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.017789265140891075, + "selected_utility_gain_over_base": -0.0653457809239626, + "selector_regret": 0.04962191917002201, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10315313935279846, + "tau": 0.10332537601561143 + }, + "636680137/train2": { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6583333333333333, + "pairwise_causal_calibration_confidence": 0.5151979568732741, + "pairwise_causal_calibration_ece": 0.14313537646005925, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.27475298254295666, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0756516233086586, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.039896368980407715, + "selected_utility_gain_over_base": -0.04323867708444595, + "selector_regret": 0.035755254328250885, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09491270035505295, + "tau": 0.10332537601561143 + }, + "655884605/train0": { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.568990751323036, + "pairwise_causal_calibration_ece": 0.09767591534363074, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.4522143975120525, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13732868432998657, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12222731113433838, + "selected_utility_gain_over_base": -0.040779367089271545, + "selector_regret": 0.015101373195648193, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8626713156700134, + "tau": 0.0009112489076126273 + }, + "655884605/train1": { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5500689384066487, + "pairwise_causal_calibration_ece": 0.07784671618442661, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.446772375365045, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8128758072853088, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11176733672618866, + "selected_utility_gain_over_base": -0.051239341497421265, + "selector_regret": 0.7011084705591202, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.1871241927146912, + "tau": 0.0009112489076126273 + }, + "655884605/train2": { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3888888888888889, + "pairwise_causal_calibration_confidence": 0.556089617835855, + "pairwise_causal_calibration_ece": 0.16720072894696625, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.407721921578567, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21447330713272095, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09131735563278198, + "selected_utility_gain_over_base": -0.07168932259082794, + "selector_regret": 0.12315595149993896, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.785526692867279, + "tau": 0.0009112489076126273 + }, + "672328262/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7647058823529411, + "pairwise_causal_calibration_confidence": 0.542991093593322, + "pairwise_causal_calibration_ece": 0.22171478875961914, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": -0.14745117870449231, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "672328262/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6571428571428571, + "pairwise_causal_calibration_confidence": 0.5517445085212526, + "pairwise_causal_calibration_ece": 0.10539834862160463, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": -0.21177052566683674, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "672328262/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.16666666666666666, + "pairwise_causal_calibration_confidence": 0.5462121445015077, + "pairwise_causal_calibration_ece": 0.37954547783484094, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": -0.19816182889430234, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.10332537601561143 + }, + "701667486/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.40476190476190477, + "pairwise_causal_calibration_confidence": 0.5617516698950917, + "pairwise_causal_calibration_ece": 0.31044146560402797, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 0.2186075457961687, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "701667486/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.546009069568378, + "pairwise_causal_calibration_ece": 0.13256235900305033, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.20451752246432947, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "701667486/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.8333333333333334, + "pairwise_causal_calibration_confidence": 0.5807247911175116, + "pairwise_causal_calibration_ece": 0.25260854221582163, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.22431661105910145, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 11.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.0009112489076126273 + }, + "747260437/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4583333333333333, + "pairwise_causal_calibration_confidence": 0.5237942697642416, + "pairwise_causal_calibration_ece": 0.06546093643090828, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.03550968246605479, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10839563608169556, + "selected_candidate_index": 15.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17026761174201965, + "tau": 0.10332537601561143 + }, + "747260437/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5173328988662427, + "pairwise_causal_calibration_ece": 0.04933376780042398, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.27755861197582443, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06328952312469482, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.21537372469902039, + "tau": 0.10332537601561143 + }, + "747260437/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.625, + "pairwise_causal_calibration_confidence": 0.5276406049368979, + "pairwise_causal_calibration_ece": 0.0973593950631021, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.061753054977374155, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.12510187923908234, + "selected_candidate_index": 2.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15356136858463287, + "tau": 0.10332537601561143 + }, + "750448296/train0": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.5218253968253967, + "pairwise_causal_calibration_confidence": 0.5273870612831654, + "pairwise_causal_calibration_ece": 0.0336721343859541, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": 0.008208417767252563, + "proposal_oracle_success": 0.3333333333333333, + "proposal_oracle_utility": 0.7526603279014429, + "selected_candidate_index": 6.666666666666667, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09184517463048299, + "selected_utility_gain_over_base": -0.02574797471364339, + "selector_regret": 0.6608151532709599, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.3333333333333333, + "support_gap": 0.6553361676633358, + "tau": 0.10332537601561143 + }, + "750448296/train1": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3532051282051282, + "pairwise_causal_calibration_confidence": 0.5396863859528889, + "pairwise_causal_calibration_ece": 0.18648125774776067, + "pairwise_causal_calibration_pairs": 101.66666666666667, + "predicted_margin": -0.031216926452283373, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3705032741030057, + "selected_candidate_index": 7.333333333333333, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.109618179500103, + "selected_utility_gain_over_base": -0.007974969844023386, + "selector_regret": 1.2608850946029027, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.03749322146177292, + "tau": 0.10332537601561143 + }, + "750448296/train2": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.44642857142857145, + "pairwise_causal_calibration_confidence": 0.5280087301794993, + "pairwise_causal_calibration_ece": 0.1080586463500039, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": -0.06850240484656789, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3504486680030823, + "selected_candidate_index": 6.666666666666667, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09844311699271202, + "selected_utility_gain_over_base": -0.019150032351414364, + "selector_regret": 1.2520055510103703, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.05754782756169637, + "tau": 0.10332537601561143 + }, + "767747424/train0": { + "base_success": 0.25, + "base_utility": 0.5731993420049548, + "coverage": 0.75, + "execute_generated": 0.75, + "fallback_rate": 0.25, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.4566239316239316, + "pairwise_causal_calibration_confidence": 0.5461567728386378, + "pairwise_causal_calibration_ece": 0.10799527425220615, + "pairwise_causal_calibration_pairs": 119.25, + "predicted_margin": 0.10341139925263847, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0424291878007352, + "selected_candidate_index": 5.75, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5295817001024261, + "selected_utility_gain_over_base": -0.0436176419025287, + "selector_regret": 0.5128474876983091, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.06562284985557199, + "tau": 0.10332537601561143 + }, + "767747424/train1": { + "base_success": 0.25, + "base_utility": 0.5731993420049548, + "coverage": 0.5, + "execute_generated": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.516633897148603, + "pairwise_causal_calibration_confidence": 0.5530278828158679, + "pairwise_causal_calibration_ece": 0.05634090783297348, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.23938122661168873, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0629530912265182, + "selected_candidate_index": 8.5, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5702044883510098, + "selected_utility_gain_over_base": -0.0029948536539450288, + "selector_regret": 0.49274860287550837, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.045098946429789066, + "tau": 0.10332537601561143 + }, + "767747424/train2": { + "base_success": 0.25, + "base_utility": 0.5731993420049548, + "coverage": 0.5, + "execute_generated": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.41618544853838974, + "pairwise_causal_calibration_confidence": 0.5435976135993295, + "pairwise_causal_calibration_ece": 0.1276296201971279, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.09558584120388469, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0671670949086547, + "selected_candidate_index": 8.25, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5541218034923077, + "selected_utility_gain_over_base": -0.019077538512647152, + "selector_regret": 0.513045291416347, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.04088494274765253, + "tau": 0.10332537601561143 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/metrics_by_task.json b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/metrics_by_task.json new file mode 100644 index 0000000000000000000000000000000000000000..4fe3a7d563acda13b2c5cd83aae88ae811d7cbd5 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/metrics_by_task.json @@ -0,0 +1,137 @@ +{ + "LiftPegUpright-v1": { + "base_success": 0.2, + "base_utility": 0.9498705387115478, + "coverage": 0.6, + "execute_generated": 0.6, + "fallback_rate": 0.4, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.5333333333333333, + "pairwise_causal_calibration_accuracy": 0.54512467498401, + "pairwise_causal_calibration_confidence": 0.54564531858913, + "pairwise_causal_calibration_ece": 0.11030342503862721, + "pairwise_causal_calibration_pairs": 94.8, + "predicted_margin": 0.14233301795724493, + "proposal_oracle_success": 0.6, + "proposal_oracle_utility": 1.4122399787108104, + "selected_candidate_index": 7.6, + "selected_success": 0.13333333333333333, + "selected_success_gain_over_base": -0.06666666666666667, + "selected_utility": 0.8185293744007747, + "selected_utility_gain_over_base": -0.13134116431077322, + "selector_regret": 0.6451034953196844, + "success_selector_gap": 0.4666666666666667, + "success_support_gap": 0.4, + "support_gap": 0.5877600212891897, + "tau": 0.0940318405174132 + }, + "PickCube-v1": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.39814814814814814, + "pairwise_causal_calibration_confidence": 0.5607943587261041, + "pairwise_causal_calibration_ece": 0.16599791089703783, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.39039029063840003, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13425676027933756, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8657432397206624, + "tau": 0.3611323919348086 + }, + "PullCube-v1": { + "base_success": 0.36, + "base_utility": 0.7751704612234608, + "coverage": 0.6266666666666667, + "execute_generated": 0.6266666666666667, + "fallback_rate": 0.37333333333333335, + "hidden_chart_oracle_success": 0.52, + "hidden_chart_oracle_utility": 1.1467878860235214, + "outcome_ptr": 0.4666666666666667, + "pairwise_causal_calibration_accuracy": 0.4995022986920425, + "pairwise_causal_calibration_confidence": 0.5422481003539483, + "pairwise_causal_calibration_ece": 0.11105368029396932, + "pairwise_causal_calibration_pairs": 92.32, + "predicted_margin": 0.18136174528296545, + "proposal_oracle_success": 0.5333333333333333, + "proposal_oracle_utility": 1.1093241602430741, + "selected_candidate_index": 7.1866666666666665, + "selected_success": 0.32, + "selected_success_gain_over_base": -0.04, + "selected_utility": 0.6814040022250265, + "selected_utility_gain_over_base": -0.09376645899843425, + "selector_regret": 0.5039629975054413, + "success_selector_gap": 0.25333333333333335, + "success_support_gap": 0.02666666666666667, + "support_gap": 0.1075624164690574, + "tau": 0.10332537601561143 + }, + "PushCube-v1": { + "base_success": 0.16666666666666666, + "base_utility": 0.3924041787783305, + "coverage": 0.6666666666666666, + "execute_generated": 0.6666666666666666, + "fallback_rate": 0.3333333333333333, + "hidden_chart_oracle_success": 0.8333333333333334, + "hidden_chart_oracle_utility": 1.7015800600250561, + "outcome_ptr": 0.8333333333333334, + "pairwise_causal_calibration_accuracy": 0.588120956369862, + "pairwise_causal_calibration_confidence": 0.5338991876592735, + "pairwise_causal_calibration_ece": 0.08423604919725729, + "pairwise_causal_calibration_pairs": 97.44444444444444, + "predicted_margin": 0.4582899698155127, + "proposal_oracle_success": 0.8888888888888888, + "proposal_oracle_utility": 1.7857814149724112, + "selected_candidate_index": 7.055555555555555, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.3333333333333333, + "selected_utility": 1.062640056014061, + "selected_utility_gain_over_base": 0.6702358772357305, + "selector_regret": 0.7231413589583503, + "success_selector_gap": 0.3888888888888889, + "success_support_gap": 0.1111111111111111, + "support_gap": 0.21421858502758873, + "tau": 0.35241155802994506 + }, + "StackCube-v1": { + "base_success": 0.2727272727272727, + "base_utility": 0.7763947045261209, + "coverage": 0.6060606060606061, + "execute_generated": 0.6060606060606061, + "fallback_rate": 0.3939393939393939, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.6363636363636364, + "pairwise_causal_calibration_accuracy": 0.5309413874218266, + "pairwise_causal_calibration_confidence": 0.5440067619410754, + "pairwise_causal_calibration_ece": 0.09869174841720094, + "pairwise_causal_calibration_pairs": 67.45454545454545, + "predicted_margin": 0.06420154895093595, + "proposal_oracle_success": 0.5151515151515151, + "proposal_oracle_utility": 1.1567951707225856, + "selected_candidate_index": 6.212121212121212, + "selected_success": 0.30303030303030304, + "selected_success_gain_over_base": 0.030303030303030304, + "selected_utility": 0.8023195727304979, + "selected_utility_gain_over_base": 0.025924868204376915, + "selector_regret": 0.3561627133326097, + "success_selector_gap": 0.21212121212121213, + "success_support_gap": 0.48484848484848486, + "support_gap": 0.8432048292774142, + "tau": 0.0009112489076126273 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/table.tex b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/table.tex new file mode 100644 index 0000000000000000000000000000000000000000..a978c01941f3fd288a56813cb9e3b6683f6dc926 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/table.tex @@ -0,0 +1,8 @@ +% Auto-generated by scripts/eval_learned_dominance_selector.py +\begin{tabular}{lrrrrrrrrr} +\toprule +Rows & Coverage & Fallback & Base succ. & Selected succ. & Oracle succ. & OutcomePTR & Succ. support gap & Succ. selector gap & Cal. ECE \\ +\midrule +144 & 0.6111 & 0.3889 & 0.2917 & 0.3125 & 0.5694 & 0.5486 & 0.2014 & 0.2778 & 0.0255 \\ +\bottomrule +\end{tabular} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/train.log b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/train.log new file mode 100644 index 0000000000000000000000000000000000000000..58bd86f710dd9473c8bcb7619233a37739c0bfb3 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_bundle_explore_envclip_k16_train_to_test/train.log @@ -0,0 +1,5 @@ +trained ridge dominance calibrator on calibration measured rows only +fit_objective=pointwise +selected_lambda=10.0 +tau={"LiftPegUpright-v1": 0.094032, "PickCube-v1": 0.361132, "PullCube-v1": 0.103325, "PushCube-v1": 0.352412, "StackCube-v1": 0.000911, "__global__": 0.113201} +calibration_input=runs/ctt_base_context_obs_train_cal_envclip_k16_rollout_comparison/combined_measured_candidates.json diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/eval.log b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/eval.log new file mode 100644 index 0000000000000000000000000000000000000000..ad44bd37e51f82471f7a17d7372cee5fed00ed3f --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/eval.log @@ -0,0 +1,3 @@ +evaluated learned fallback rule on held-out measured rollout rows +eval_input=runs/ctt_base_context_obs_test_envclip_k16_rollout_comparison/combined_measured_candidates.json +num_eval_rows=144 diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/metrics.json b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..65ce221d8d5a0706646a8c8f5ca3ffb89c8efecb --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/metrics.json @@ -0,0 +1,9124 @@ +{ + "calibration_input": "runs/ctt_base_context_obs_train_cal_envclip_k16_rollout_comparison/combined_measured_candidates.json", + "calibration_model_selection": { + "base_success": 0.2777777777777778, + "base_utility": 0.7007981554626086, + "coverage": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 0.6597222222222222, + "hidden_chart_oracle_utility": 1.4150809271054134, + "outcome_ptr": 0.5462962962962963, + "proposal_oracle_success": 0.5370370370370371, + "proposal_oracle_utility": 1.1440672337592803, + "selected_success": 0.375, + "selected_success_gain_over_base": 0.09722222222222222, + "selected_utility": 0.8586226551326237, + "selector_regret": 0.3815252628321307, + "success_selector_gap": 0.20601851851851852, + "success_support_gap": 0.17592592592592593, + "support_gap": 0.35465237537627453 + }, + "calibration_summary": { + "base_success": 0.2777777777777778, + "base_utility": 0.7007981554626086, + "coverage": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 0.6597222222222222, + "hidden_chart_oracle_utility": 1.4150809271054134, + "outcome_ptr": 0.5462962962962963, + "pairwise_causal_calibration_accuracy": 0.5077676549581337, + "pairwise_causal_calibration_confidence": 0.5876765208114785, + "pairwise_causal_calibration_ece": 0.079908865853345, + "pairwise_causal_calibration_pairs": 37978.0, + "proposal_oracle_success": 0.5370370370370371, + "proposal_oracle_utility": 1.1440672337592803, + "selected_success": 0.375, + "selected_success_gain_over_base": 0.09722222222222222, + "selected_utility": 0.8586226551326237, + "selector_regret": 0.3815252628321307, + "success_selector_gap": 0.20601851851851852, + "success_support_gap": 0.17592592592592593, + "support_gap": 0.35465237537627453 + }, + "calibration_target_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "calibration_target_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "chart_feature_mode": "base_context_obs", + "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "eval_input": "runs/ctt_base_context_obs_test_envclip_k16_rollout_comparison/combined_measured_candidates.json", + "eval_summary": { + "base_success": 0.2916666666666667, + "base_utility": 0.7418195864423373, + "coverage": 0.5625, + "fallback_rate": 0.4375, + "hidden_chart_oracle_success": 0.7291666666666666, + "hidden_chart_oracle_utility": 1.5183161981403828, + "outcome_ptr": 0.5486111111111112, + "pairwise_causal_calibration_accuracy": 0.5206691330223581, + "pairwise_causal_calibration_confidence": 0.5817740213754673, + "pairwise_causal_calibration_ece": 0.06110488835310939, + "pairwise_causal_calibration_pairs": 12434.0, + "proposal_oracle_success": 0.5694444444444444, + "proposal_oracle_utility": 1.2159999172436073, + "selected_success": 0.2916666666666667, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7314973677291872, + "selector_regret": 0.5157032941917552, + "success_selector_gap": 0.2847222222222222, + "success_support_gap": 0.2013888888888889, + "support_gap": 0.3761285081272945 + }, + "eval_target_content_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "eval_target_split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "feature_mean": [ + 0.0, + 4.602051421289207, + 1.8887871984592688, + -1.985099232970326e-17, + 7.187557603686636, + 0.0663594470046083, + 2.0903225806451613, + 0.7930022866760579, + 1.9819565879721794, + 15.375115207373272, + 1.0, + 1.712312153966603, + 1.2801339190863705, + 2.7725887222400836, + 0.37403993855606793, + 1.1886197294610712, + 0.756002662910042, + 1.4434513973681609, + 1.0108343327560374, + 0.676036866359447, + 0.44369032273409875, + 0.7674145310087138, + 0.567144010021948, + 1.0, + 0.0, + 0.6065368202143773, + 0.46997871591933094, + 0.5076804915514593, + 1.0, + 1.044349800585736, + 1.0509106460665045, + 0.8002031215841607, + 0.5811250768510506, + 2.5434513049028507, + 0.34326541665726357, + 0.8966401857535892, + 1.2182893520781068, + 1.2263923089918751, + 0.786961656617233, + 0.6987682628846764, + 1.0, + 0.9974076898437442, + 0.02008356847617045, + 0.39137378168005876, + 0.3910654018162905, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 2.788372394309843, + 5.701382488479263, + 12.81136712749616, + 0.40686157587079336, + 0.8936925909045862, + 0.24551906257091227, + 0.12993606097190302, + 0.1850963060060051, + 0.0663594470046083, + 3.944723778791356, + 4.364277845709234, + 4.603718512944113, + 5.60843294300425, + 4.854070660522273, + 0.3680969062939123, + 1.0, + 0.3217802000866517, + 4.613985697387828, + 0.5009397088659758 + ], + "feature_names": [ + "bias", + "candidate_score", + "candidate_score_minus_base_score", + "row_score_z", + "candidate_index", + "is_top_score_candidate", + "source_chart_rank", + "tangent_rms_norm", + "tangent_linf_norm", + "num_candidates", + "source_chart_found", + "source_positive_count_log", + "source_negative_count_log", + "source_nonbase_count_log", + "source_positive_rate", + "source_best_delta_utility", + "source_mean_delta_utility", + "source_best_utility", + "source_mean_utility", + "source_best_success", + "source_mean_success", + "source_best_progress", + "source_mean_progress", + "source_positive_safety_known_rate", + "source_positive_unsafe_rate_known", + "generated_to_source_positive_min_rms", + "generated_to_source_negative_min_rms", + "generated_source_pos_closer_than_neg", + "source_chart_feature_found", + "target_chart_feature_norm", + "source_chart_feature_norm", + "target_source_chart_cosine", + "target_source_chart_rms", + "target_source_chart_linf", + "target_source_chart_absmean", + "target_source_chart_dot_per_dim", + "target_base_action_norm", + "source_base_action_norm", + "target_source_base_cosine", + "target_source_base_rms", + "target_source_obs_available", + "target_source_obs_cosine", + "target_source_obs_rms", + "target_obs_norm", + "source_obs_norm", + "target_source_obj_available", + "target_source_obj_cosine", + "target_source_obj_rms", + "target_obj_norm", + "source_obj_norm", + "bundle_num_candidates_log", + "bundle_neighbor_count_r020", + "bundle_neighbor_count_r040", + "bundle_neighbor_frac_r020", + "bundle_neighbor_frac_r040", + "bundle_mean_peer_rms", + "bundle_min_peer_rms", + "bundle_medoid_rms", + "bundle_is_medoid", + "bundle_peer_score_mean_r020", + "bundle_peer_score_max_r020", + "bundle_peer_score_mean_r040", + "bundle_peer_score_max_r040", + "bundle_unique_source_count_r040", + "bundle_unique_source_frac_r040", + "bundle_unique_task_count_r040", + "bundle_same_source_frac_r040", + "bundle_score_density_r040", + "bundle_rank_density_r040" + ], + "feature_set": "chart_source_bundle_consensus", + "feature_std": [ + 1.0, + 2.5232884736710055, + 2.2444575887945213, + 0.9999992484766103, + 4.542613580634851, + 0.24891036261589178, + 1.9642314545113713, + 0.37078681697003224, + 0.7171840796063713, + 1.7485561930736828, + 1e-06, + 0.6172259236004526, + 1.0207348750234775, + 1.0000003024247519e-06, + 0.247259211359892, + 0.7394816262084946, + 0.599569803694898, + 0.811233687088667, + 0.7790947397311471, + 0.46798713407060977, + 0.4330181331210625, + 0.3530091411010039, + 0.35507719727234843, + 1e-06, + 1e-06, + 0.3893883017336776, + 0.4345787301086313, + 0.49994200656928256, + 1e-06, + 0.14319094131293478, + 0.14491309378768996, + 0.19298897178584146, + 0.32495613500512616, + 1.3413619446954763, + 0.2065176097090355, + 0.3376678168884184, + 0.1766334065498482, + 0.1790096989966568, + 0.20687067948130008, + 0.3907071033474156, + 1e-06, + 0.005866370796541541, + 0.02376663855951178, + 0.015525196749693172, + 0.017227249939028074, + 1e-06, + 1e-06, + 1e-06, + 1e-06, + 1e-06, + 0.12987037363894735, + 4.878517299430629, + 3.217377946148024, + 0.346567593200359, + 0.19990387562321757, + 0.09641872137117102, + 0.07067998005028978, + 0.11663352509384882, + 0.2489103626158918, + 2.8138515595422326, + 2.965671838859836, + 2.4297066599961332, + 2.38657189782847, + 2.243933495797979, + 0.19723478821626303, + 1e-06, + 0.28108282108993726, + 2.4172277778814877, + 0.31354917609205285 + ], + "fit_design": { + "fit_objective": "pointwise", + "num_candidate_rows": 6510, + "num_fit_rows": 6510, + "num_pairwise_rows": 0, + "num_pointwise_rows": 6510, + "pairwise_weight": 0.0 + }, + "fit_objective": "pointwise", + "k": 16, + "num_calibration_candidates": 6510, + "num_calibration_rows": 432, + "num_eval_candidates": 2115, + "num_eval_rows": 144, + "pairwise_causal_calibration": { + "calibration": { + "accuracy": 0.5077676549581337, + "bins": [ + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.0, + "upper": 0.1 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.1, + "upper": 0.2 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.2, + "upper": 0.3 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.3, + "upper": 0.4 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.4, + "upper": 0.5 + }, + { + "abs_gap": 0.03786159949288581, + "accuracy": 0.503346291027131, + "confidence": 0.5412078905200168, + "count": 24953, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.12472863532485967, + "accuracy": 0.5185390149418926, + "confidence": 0.6432676502667523, + "count": 9035, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.22639637511815436, + "accuracy": 0.5140240412135089, + "confidence": 0.7404204163316632, + "count": 3494, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.3485853017848903, + "accuracy": 0.48268398268398266, + "confidence": 0.8312692844688729, + "count": 462, + "lower": 0.8, + "upper": 0.9 + }, + { + "abs_gap": 0.3241132642588024, + "accuracy": 0.5882352941176471, + "confidence": 0.9123485583764495, + "count": 34, + "lower": 0.9, + "upper": 1.0 + } + ], + "ece": 0.079908865853345, + "mean_confidence": 0.5876765208114785, + "n_bins": 10, + "num_pairs": 37978, + "num_rows": 432 + }, + "eval": { + "accuracy": 0.5206691330223581, + "bins": [ + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.0, + "upper": 0.1 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.1, + "upper": 0.2 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.2, + "upper": 0.3 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.3, + "upper": 0.4 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.4, + "upper": 0.5 + }, + { + "abs_gap": 0.023211338336958764, + "accuracy": 0.5176347340239786, + "confidence": 0.5408460723609374, + "count": 8591, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.10911026547809066, + "accuracy": 0.5314349112426036, + "confidence": 0.6405451767206942, + "count": 2704, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.22631325440901795, + "accuracy": 0.5166666666666667, + "confidence": 0.7429799210756847, + "count": 1020, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.2898814855444869, + "accuracy": 0.5294117647058824, + "confidence": 0.8192932502503693, + "count": 119, + "lower": 0.8, + "upper": 0.9 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.9, + "upper": 1.0 + } + ], + "ece": 0.06110488835310939, + "mean_confidence": 0.5817740213754673, + "n_bins": 10, + "num_pairs": 12434, + "num_rows": 144 + } + }, + "pairwise_weight": 1.0, + "report_type": "learned_dominance_selector_eval", + "ridge_lambdas": [ + 0.0, + 0.01, + 0.1, + 1.0, + 10.0, + 100.0 + ], + "rows": [ + { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "chart_id": "ms-5799584ed40e211c50a5", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5042016806722689, + "pairwise_causal_calibration_confidence": 0.5508434661236096, + "pairwise_causal_calibration_ece": 0.08899904912532385, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.2153078550922283, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7106714248657227, + "selected_utility_gain_over_base": -0.012933909893035889, + "selector_regret": 1.2893285751342773, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7032967032967034, + "pairwise_causal_calibration_confidence": 0.5795355714423935, + "pairwise_causal_calibration_ece": 0.1237611318543097, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.18766723699772747, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2115575522184372, + "seed": "307068879/train0", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7884424477815628, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7b229b60d7f64fb624cf", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5516482746314322, + "pairwise_causal_calibration_ece": 0.1125523173289727, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.197390750232301, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.7067229747772217, + "selected_utility_gain_over_base": -1.2932770252227783, + "selector_regret": 1.2932770252227783, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6571428571428571, + "pairwise_causal_calibration_confidence": 0.5840070772055741, + "pairwise_causal_calibration_ece": 0.07313577993728299, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.2718540448404614, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/train0", + "selected_candidate_index": 12, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.5622784804732108, + "pairwise_causal_calibration_ece": 0.1804602986550292, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.422930196465892, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "441046872/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.3000448346138, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.19444444444444445, + "pairwise_causal_calibration_confidence": 0.6053552926240098, + "pairwise_causal_calibration_ece": 0.4109108481795651, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.7902719177719542, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18325337767601013, + "seed": "1522272686/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8167466223239899, + "task_id": "PickCube-v1", + "tau": 1.280136139242778, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-06babf453b22ba8f2407", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 1.512536567520637, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/train0", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.16168592870235443, + "chart_id": "ms-0e1ff8d16372796dac97", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5128205128205128, + "pairwise_causal_calibration_confidence": 0.5427636297408412, + "pairwise_causal_calibration_ece": 0.05199413529520332, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 1.0958317448801518, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train0", + "selected_candidate_index": 3, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8383140712976456, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.275, + "pairwise_causal_calibration_confidence": 0.5596745083517122, + "pairwise_causal_calibration_ece": 0.2846745083517123, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5914274057916176, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06019836664199829, + "seed": "1461069713/train0", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0939030647277832, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0991440862417221, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.001929501653648913, + "chart_id": "ms-34726202b475d29cdf38", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.25342071056365967, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.6243032484138443, + "pairwise_causal_calibration_ece": 0.06612558013009637, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0207049192558117, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.010183374397456646, + "seed": "626486813/train0", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007048348430544138, + "selected_utility_gain_over_base": 0.005118846776895225, + "selector_regret": 0.003135025966912508, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24323733616620302, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-4cc40587ec490257b1d2", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.275, + "pairwise_causal_calibration_confidence": 0.5896370284261352, + "pairwise_causal_calibration_ece": 0.3146370284261351, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.6346761015834912, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15456630289554596, + "seed": "2054899126/train0", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07831952720880508, + "selected_utility_gain_over_base": -1.921680472791195, + "selector_regret": 0.07624677568674088, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.845433697104454, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-62f8c18a15c566ec4b00", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6470588235294118, + "pairwise_causal_calibration_confidence": 0.6057740013777554, + "pairwise_causal_calibration_ece": 0.08437381833742622, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": 1.0088593202243494, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train0", + "selected_candidate_index": 7, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5856319439826799, + "pairwise_causal_calibration_ece": 0.11896527731601296, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.616048277343447, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15051491558551788, + "seed": "767747424/train0", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07698849588632584, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.07352641969919205, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08458571135997772, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.03588630631566048, + "chart_id": "ms-6b789356d65b3312fbae", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22781728208065033, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7166666666666667, + "pairwise_causal_calibration_confidence": 0.5931535905005015, + "pairwise_causal_calibration_ece": 0.12351307616616536, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6180032029224511, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14671264588832855, + "seed": "278008524/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.03588630631566048, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.11082633957266808, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08110463619232178, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5652173913043478, + "pairwise_causal_calibration_confidence": 0.5667479107448529, + "pairwise_causal_calibration_ece": 0.025257247762570406, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": 1.340214222459458, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1845489013/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.008267183788120747, + "chart_id": "ms-7df455efd413d77e6ce3", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.07774405181407928, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.2833333333333333, + "pairwise_causal_calibration_confidence": 0.5253581386999512, + "pairwise_causal_calibration_ece": 0.24202480536661786, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5133002339982102, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07920757681131363, + "seed": "626486813/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.008267183788120747, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.07094039302319288, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7e295b5859c07c90487f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.26271408796310425, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6916666666666667, + "pairwise_causal_calibration_confidence": 0.5592734793015516, + "pairwise_causal_calibration_ece": 0.13239318736511552, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3220928111639759, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train0", + "selected_candidate_index": 0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "chart_id": "ms-8577816da2917a2bb63d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5597842433887508, + "pairwise_causal_calibration_ece": 0.08130419295423207, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.4903299974496993, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "585904130/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09919407218694687, + "selected_utility_gain_over_base": -0.0825214758515358, + "selector_regret": 1.9008059278130531, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "chart_id": "ms-8e56b0eaa2604db6749e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5416666666666666, + "pairwise_causal_calibration_confidence": 0.5684739620257733, + "pairwise_causal_calibration_ece": 0.0268072953591066, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6161038657373066, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "seed": "141984615/train0", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10283424705266953, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.14270895719528198, + "chart_id": "ms-972c5be72d0b24be7cd5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.5384807944311416, + "pairwise_causal_calibration_ece": 0.04158052457553095, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8210246755902229, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21658658981323242, + "seed": "750448296/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14270895719528198, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.07387763261795044, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7834134101867676, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "chart_id": "ms-9757fc494e79b5543bb6", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5986660164452574, + "pairwise_causal_calibration_ece": 0.18530981522251466, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.70117503855084, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.08826130628585815, + "seed": "636680137/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08826130628585815, + "selected_utility_gain_over_base": 0.005126260221004486, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0823030173778534, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.05412294343113899, + "chart_id": "ms-a8731059713697ce5e2a", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.19710752367973328, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.35833333333333334, + "pairwise_causal_calibration_confidence": 0.6230145863265159, + "pairwise_causal_calibration_ece": 0.30650057427120686, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5724868358709978, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.019201835617423058, + "seed": "767747424/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05412294343113899, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17790568806231022, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-ac94b48febe523da15a5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4533333333333333, + "pairwise_causal_calibration_confidence": 0.5366248224651405, + "pairwise_causal_calibration_ece": 0.08329148913180733, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.3369485573884625, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b49043f15f4846de5b13", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3448275862068966, + "pairwise_causal_calibration_confidence": 0.6255201768318296, + "pairwise_causal_calibration_ece": 0.3473762370913279, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.1448828501178339, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04285776615142822, + "selected_utility_gain_over_base": -1.9571422338485718, + "selector_regret": 1.9571422338485718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.0929974913597107, + "chart_id": "ms-b7d7021b782f73816083", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22398948669433594, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.6190746261431673, + "pairwise_causal_calibration_ece": 0.08763058562166437, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.8443995043743224, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.041394393891096115, + "seed": "750448296/train0", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.015753567218780518, + "selected_utility_gain_over_base": -0.07724392414093018, + "selector_regret": 0.025640826672315598, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.18259509280323982, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 1.9153901564980373, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train0", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8fc1568ce66c45c8d8b", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.45, + "pairwise_causal_calibration_confidence": 0.5630309153002454, + "pairwise_causal_calibration_ece": 0.11303091530024553, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0110612897022033, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10839563608169556, + "seed": "747260437/train0", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.08987965434789658, + "selected_utility_gain_over_base": -1.9101203456521034, + "selector_regret": 0.01851598173379898, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17026761174201965, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.04958367347717285, + "chart_id": "ms-da555a03dd382d93e119", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6565656565656566, + "pairwise_causal_calibration_confidence": 0.6041498943194046, + "pairwise_causal_calibration_ece": 0.07832126548147574, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.13084136604112917, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04958367347717285, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9504163265228271, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "chart_id": "ms-f5dd1c43ebc4c0cfc3f1", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.75, + "pairwise_causal_calibration_confidence": 0.5625148880216412, + "pairwise_causal_calibration_ece": 0.18748511197835893, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.530739092243051, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.060871422290802, + "seed": "1549324805/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.03462250903248787, + "selected_utility_gain_over_base": -0.06989232823252678, + "selector_regret": 0.026248913258314133, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.23658466339111328, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "chart_id": "ms-f9740a08a98cb251b761", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.625, + "pairwise_causal_calibration_confidence": 0.5988235112981328, + "pairwise_causal_calibration_ece": 0.11653574747746832, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 1.3378277706579857, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06306678056716919, + "selected_utility_gain_over_base": -0.008848525583744049, + "selector_regret": 1.9369332194328308, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5595238095238095, + "pairwise_causal_calibration_confidence": 0.5919797870987649, + "pairwise_causal_calibration_ece": 0.07576099703696844, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.007158829388102639, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "chart_id": "ms-05f34a90fa7355262b8e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7586206896551724, + "pairwise_causal_calibration_confidence": 0.5841905535939795, + "pairwise_causal_calibration_ece": 0.17443013606119306, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.839399049033097, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train0", + "selected_candidate_index": 4, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "chart_id": "ms-2413d1155c87f0442485", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5583333333333333, + "pairwise_causal_calibration_confidence": 0.5894684248151024, + "pairwise_causal_calibration_ece": 0.03846474579568359, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 2.1833463985406447, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14406004548072815, + "seed": "1324642411/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.054066285490989685, + "selected_utility_gain_over_base": -0.052404239773750305, + "selector_regret": 0.08999375998973846, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8559399545192719, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "chart_id": "ms-54346a8f79bd9dca1dfe", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6272727272727273, + "pairwise_causal_calibration_confidence": 0.5867290403888722, + "pairwise_causal_calibration_ece": 0.04054368688385518, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.9875622503266075, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.87523253262043, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-64f98d163b6bff29332b", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.494949494949495, + "pairwise_causal_calibration_confidence": 0.6511354281579875, + "pairwise_causal_calibration_ece": 0.3204647685440312, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 1.471050707973179, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.05754852294921875, + "selected_utility_gain_over_base": -1.9424514770507812, + "selector_regret": 1.9424514770507812, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "chart_id": "ms-c2ae2b78ab7891e74c80", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6050420168067226, + "pairwise_causal_calibration_confidence": 0.5998439015094801, + "pairwise_causal_calibration_ece": 0.07491598280692124, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 2.434169345753335, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/train0", + "selected_candidate_index": 11, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "chart_id": "ms-d4c37ed44fb338c1d551", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.26262626262626265, + "pairwise_causal_calibration_confidence": 0.5720642439373731, + "pairwise_causal_calibration_ece": 0.30943798131111067, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 2.1158648725170077, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train0", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07608599960803986, + "selected_utility_gain_over_base": -0.010948076844215393, + "selector_regret": 1.9239140003919601, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "chart_id": "ms-146ff6c3f22b3a49af3a", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5740299161789739, + "pairwise_causal_calibration_ece": 0.1173721779867959, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.7792589748622651, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09799930453300476, + "seed": "2061635888/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04574880003929138, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9020006954669952, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "chart_id": "ms-15fc8cc6f74cf589056f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6388888888888888, + "pairwise_causal_calibration_confidence": 0.605234361994273, + "pairwise_causal_calibration_ece": 0.09534125846142073, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.9134188949984638, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13732868432998657, + "seed": "655884605/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12222731113433838, + "selected_utility_gain_over_base": -0.040779367089271545, + "selector_regret": 0.015101373195648193, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8626713156700134, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "chart_id": "ms-36bc1364d2439099e9ed", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5411387050524964, + "pairwise_causal_calibration_ece": 0.025500013084200863, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.008407185293451747, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10274861752986908, + "seed": "1113371261/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.897251382470131, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "chart_id": "ms-420280f15a522ce21c72", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4358974358974359, + "pairwise_causal_calibration_confidence": 0.6103413950316227, + "pairwise_causal_calibration_ece": 0.17444395913418703, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 1.1959458964053806, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train0", + "selected_candidate_index": 14, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.17586350440979, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "chart_id": "ms-4aed1197d0906ae51c34", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4888888888888889, + "pairwise_causal_calibration_confidence": 0.5462356837170409, + "pairwise_causal_calibration_ece": 0.057346794828152005, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.5816690584072294, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.5845752206379711, + "pairwise_causal_calibration_ece": 0.04745706568944737, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.9680028345756826, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "211245842/train0", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8090915828943253, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "chart_id": "ms-77536801924a2a78688e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5619047619047619, + "pairwise_causal_calibration_confidence": 0.5447959029344634, + "pairwise_causal_calibration_ece": 0.05152312612164488, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.8492504414490131, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8663521409034729, + "seed": "1151410346/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.6403508484363556, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.133647859096527, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "chart_id": "ms-9dd1ff4c297a81b13071", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5778483044050591, + "pairwise_causal_calibration_ece": 0.11356259011934491, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41376711119187704, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.054989203810691833, + "seed": "1324133122/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9450107961893082, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5760031182899459, + "pairwise_causal_calibration_ece": 0.20253601281087805, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 1.0829493379419586, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d776afa14e6322cadedd", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.6052287842223839, + "pairwise_causal_calibration_ece": 0.08071518251541011, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 1.4315714539962017, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/train0", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-e887dbfa0133af46829d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6714285714285714, + "pairwise_causal_calibration_confidence": 0.6012777562735464, + "pairwise_causal_calibration_ece": 0.07015081515502501, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 1.3600458578396741, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train0", + "selected_candidate_index": 8, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "chart_id": "ms-5799584ed40e211c50a5", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.475, + "pairwise_causal_calibration_confidence": 0.5273415392712859, + "pairwise_causal_calibration_ece": 0.05234153927128593, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7823168935976765, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7191179394721985, + "selected_utility_gain_over_base": -0.004487395286560059, + "selector_regret": 1.2808820605278015, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7692307692307693, + "pairwise_causal_calibration_confidence": 0.5687146745124166, + "pairwise_causal_calibration_ece": 0.2005160947183524, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.12417642494936398, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5876073241233826, + "seed": "307068879/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7b229b60d7f64fb624cf", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.39166666666666666, + "pairwise_causal_calibration_confidence": 0.5665909321275702, + "pairwise_causal_calibration_ece": 0.17492426546090348, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.057082412822946, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/train1", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.7057674527168274, + "selected_utility_gain_over_base": -1.2942325472831726, + "selector_regret": 1.2942325472831726, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.6130721925494387, + "pairwise_causal_calibration_ece": 0.157312300699128, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.9190150483991828, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/train1", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7279855608940125, + "selected_utility_gain_over_base": 0.002283930778503418, + "selector_regret": 1.2720144391059875, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5636363636363636, + "pairwise_causal_calibration_confidence": 0.5559054729794208, + "pairwise_causal_calibration_ece": 0.02996166181944273, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.2197157946327053, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7233588695526123, + "seed": "441046872/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.023403704166412354, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3055555555555556, + "pairwise_causal_calibration_confidence": 0.6354932924180177, + "pairwise_causal_calibration_ece": 0.32993773686246214, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.20562222929953555, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07014383375644684, + "seed": "1522272686/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9298561662435532, + "task_id": "PickCube-v1", + "tau": 1.280136139242778, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-06babf453b22ba8f2407", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 1.639637290759596, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/train1", + "selected_candidate_index": 3, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.16168592870235443, + "chart_id": "ms-0e1ff8d16372796dac97", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.44537815126050423, + "pairwise_causal_calibration_confidence": 0.5652758800428274, + "pairwise_causal_calibration_ece": 0.11989772878232334, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.7484857304335504, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16168592870235443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8383140712976456, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.375, + "pairwise_causal_calibration_confidence": 0.5561091294148327, + "pairwise_causal_calibration_ece": 0.1811091294148327, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6317591128320483, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "seed": "1461069713/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0939030647277832, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14275743998587132, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.001929501653648913, + "chart_id": "ms-34726202b475d29cdf38", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.25342071056365967, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.23333333333333334, + "pairwise_causal_calibration_confidence": 0.6225071103783514, + "pairwise_causal_calibration_ece": 0.38917377704501765, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9240538283867041, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.009895960800349712, + "seed": "626486813/train1", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.001929501653648913, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0079664591467008, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24352474976330996, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-4cc40587ec490257b1d2", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6228070175438597, + "pairwise_causal_calibration_confidence": 0.5635260175535555, + "pairwise_causal_calibration_ece": 0.05928099999030417, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 1.408040645988902, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train1", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.11194747686386108, + "selected_utility_gain_over_base": -1.888052523136139, + "selector_regret": 1.888052523136139, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-62f8c18a15c566ec4b00", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.6362714146835213, + "pairwise_causal_calibration_ece": 0.05483727064861237, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": 1.1331098956489296, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train1", + "selected_candidate_index": 4, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5916666666666667, + "pairwise_causal_calibration_confidence": 0.589009514704837, + "pairwise_causal_calibration_ece": 0.16556659496285603, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.734254641128912, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.20903043448925018, + "seed": "767747424/train1", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07698849588632584, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.13204193860292435, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.026070192456245422, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.03588630631566048, + "chart_id": "ms-6b789356d65b3312fbae", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22781728208065033, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5916666666666667, + "pairwise_causal_calibration_confidence": 0.5847627797475756, + "pairwise_causal_calibration_ece": 0.07886929809787337, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7353933081725526, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07534918934106827, + "seed": "278008524/train1", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.03588630631566048, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.03946288302540779, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15246809273958206, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.46153846153846156, + "pairwise_causal_calibration_confidence": 0.5409684680296629, + "pairwise_causal_calibration_ece": 0.11303375788512275, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": 1.2941213610673934, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1845489013/train1", + "selected_candidate_index": 13, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.008267183788120747, + "chart_id": "ms-7df455efd413d77e6ce3", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.07774405181407928, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6416666666666667, + "pairwise_causal_calibration_confidence": 0.5526245318298203, + "pairwise_causal_calibration_ece": 0.08904213483684603, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9222355806937221, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.061759788542985916, + "seed": "626486813/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.008267183788120747, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.05349260475486517, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.01598426327109337, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7e295b5859c07c90487f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.26271408796310425, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5641025641025641, + "pairwise_causal_calibration_confidence": 0.5418426001200853, + "pairwise_causal_calibration_ece": 0.022259963982478842, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.5727868921893446, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train1", + "selected_candidate_index": 12, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "chart_id": "ms-8577816da2917a2bb63d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4266666666666667, + "pairwise_causal_calibration_confidence": 0.5738533196584167, + "pairwise_causal_calibration_ece": 0.14718665299174993, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 1.4765054459187703, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "585904130/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04792279005050659, + "selected_utility_gain_over_base": -0.13379275798797607, + "selector_regret": 1.9520772099494934, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "chart_id": "ms-8e56b0eaa2604db6749e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.5676327735960262, + "pairwise_causal_calibration_ece": 0.0509661069293595, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7260454977260915, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.037011127918958664, + "seed": "141984615/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10283424705266953, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24876614287495613, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.14270895719528198, + "chart_id": "ms-972c5be72d0b24be7cd5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.325, + "pairwise_causal_calibration_confidence": 0.5670206661736661, + "pairwise_causal_calibration_ece": 0.2420206661736662, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8690092233397244, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14270895719528198, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.857291042804718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "chart_id": "ms-9757fc494e79b5543bb6", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6833333333333333, + "pairwise_causal_calibration_confidence": 0.5644217102464404, + "pairwise_causal_calibration_ece": 0.11891162308689304, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9868011033647903, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06741118431091309, + "seed": "636680137/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.017789265140891075, + "selected_utility_gain_over_base": -0.0653457809239626, + "selector_regret": 0.04962191917002201, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10315313935279846, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.05412294343113899, + "chart_id": "ms-a8731059713697ce5e2a", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.19710752367973328, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.44166666666666665, + "pairwise_causal_calibration_confidence": 0.665505380834036, + "pairwise_causal_calibration_ece": 0.2549196296057593, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.543283265167321, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.042781930416822433, + "seed": "767747424/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.006599446292966604, + "selected_utility_gain_over_base": -0.04752349713817239, + "selector_regret": 0.03618248412385583, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15432559326291084, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-ac94b48febe523da15a5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5263157894736842, + "pairwise_causal_calibration_confidence": 0.5675437560777116, + "pairwise_causal_calibration_ece": 0.04122796660402744, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.11389548031987307, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/train1", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b49043f15f4846de5b13", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3148148148148148, + "pairwise_causal_calibration_confidence": 0.6741121240452155, + "pairwise_causal_calibration_ece": 0.3592973092304006, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.6895335177823154, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07674666494131088, + "selected_utility_gain_over_base": -1.9232533350586891, + "selector_regret": 1.9232533350586891, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.0929974913597107, + "chart_id": "ms-b7d7021b782f73816083", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22398948669433594, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5840682757408601, + "pairwise_causal_calibration_ece": 0.07650195015295802, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.4807216759053028, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11150982230901718, + "seed": "750448296/train1", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04923723265528679, + "selected_utility_gain_over_base": -0.043760258704423904, + "selector_regret": 0.06227258965373039, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.11247966438531876, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 2.1287174207133073, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train1", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8fc1568ce66c45c8d8b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3416666666666667, + "pairwise_causal_calibration_confidence": 0.5662548497567926, + "pairwise_causal_calibration_ece": 0.2245881830901258, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6563621545761416, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06328952312469482, + "seed": "747260437/train1", + "selected_candidate_index": 13, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.21537372469902039, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.04958367347717285, + "chart_id": "ms-da555a03dd382d93e119", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6847826086956522, + "pairwise_causal_calibration_confidence": 0.5795301941627359, + "pairwise_causal_calibration_ece": 0.1576643171073477, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.04149374232098668, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train1", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04958367347717285, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9504163265228271, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "chart_id": "ms-f5dd1c43ebc4c0cfc3f1", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5833333333333334, + "pairwise_causal_calibration_confidence": 0.5360665332231191, + "pairwise_causal_calibration_ece": 0.07111703737429365, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.2279357729216895, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "seed": "1549324805/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0303483996540308, + "selected_utility_gain_over_base": -0.07416643761098385, + "selector_regret": 0.16857311688363552, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "chart_id": "ms-f9740a08a98cb251b761", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.6072514522495137, + "pairwise_causal_calibration_ece": 0.1057225969843389, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": 1.2817760714600654, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train1", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9280846938490868, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3230769230769231, + "pairwise_causal_calibration_confidence": 0.6144157349372877, + "pairwise_causal_calibration_ece": 0.2913388118603647, + "pairwise_causal_calibration_pairs": 65.0, + "predicted_margin": -0.029555234625273652, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "chart_id": "ms-05f34a90fa7355262b8e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4, + "pairwise_causal_calibration_confidence": 0.5669827903966217, + "pairwise_causal_calibration_ece": 0.16698279039662164, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 2.0021154026255705, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train1", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "chart_id": "ms-2413d1155c87f0442485", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42016806722689076, + "pairwise_causal_calibration_confidence": 0.6037811380752174, + "pairwise_causal_calibration_ece": 0.18361307084832634, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.7807138675039442, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10393516719341278, + "selected_utility_gain_over_base": -0.0025353580713272095, + "selector_regret": 1.8960648328065872, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "chart_id": "ms-54346a8f79bd9dca1dfe", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5263157894736842, + "pairwise_causal_calibration_confidence": 0.5937233509400742, + "pairwise_causal_calibration_ece": 0.10426662637283049, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 1.9148838772749794, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13741976022720337, + "selected_utility_gain_over_base": 0.012652292847633362, + "selector_regret": 1.8625802397727966, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-64f98d163b6bff29332b", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5636363636363636, + "pairwise_causal_calibration_confidence": 0.6560658348856379, + "pairwise_causal_calibration_ece": 0.305306037200651, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.3365043159479286, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train1", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07495865225791931, + "selected_utility_gain_over_base": -1.9250413477420807, + "selector_regret": 1.9250413477420807, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "chart_id": "ms-c2ae2b78ab7891e74c80", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.6128506472019413, + "pairwise_causal_calibration_ece": 0.1382932703018193, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 2.300921534892476, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/train1", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "chart_id": "ms-d4c37ed44fb338c1d551", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4957983193277311, + "pairwise_causal_calibration_confidence": 0.5496493820116296, + "pairwise_causal_calibration_ece": 0.05385106268389854, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.8867636980590103, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train1", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19651852548122406, + "selected_utility_gain_over_base": 0.10948444902896881, + "selector_regret": 1.803481474518776, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "chart_id": "ms-146ff6c3f22b3a49af3a", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.2777777777777778, + "pairwise_causal_calibration_confidence": 0.555294391767556, + "pairwise_causal_calibration_ece": 0.27751661398977834, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.7080441039530474, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07580336928367615, + "seed": "2061635888/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02355286478996277, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9241966307163239, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "chart_id": "ms-15fc8cc6f74cf589056f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5888894132525583, + "pairwise_causal_calibration_ece": 0.08888941325255821, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 1.0866492061529267, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8128758072853088, + "seed": "655884605/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11176733672618866, + "selected_utility_gain_over_base": -0.051239341497421265, + "selector_regret": 0.7011084705591202, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.1871241927146912, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "chart_id": "ms-36bc1364d2439099e9ed", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.53349783320347, + "pairwise_causal_calibration_ece": 0.10492640463204145, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.2579825916955958, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18093222379684448, + "seed": "1113371261/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02250880002975464, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8190677762031555, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "chart_id": "ms-420280f15a522ce21c72", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42727272727272725, + "pairwise_causal_calibration_confidence": 0.6263703949729311, + "pairwise_causal_calibration_ece": 0.1990976677002039, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.2514512085861837, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train1", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13703244924545288, + "selected_utility_gain_over_base": -0.6871040463447571, + "selector_regret": 1.8629675507545471, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "chart_id": "ms-4aed1197d0906ae51c34", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.1746031746031746, + "pairwise_causal_calibration_confidence": 0.5479465839142913, + "pairwise_causal_calibration_ece": 0.37334340931111676, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": 0.5626192514371053, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train1", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5416666666666666, + "pairwise_causal_calibration_confidence": 0.5914410404762934, + "pairwise_causal_calibration_ece": 0.10787248047280223, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.097151631086689, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "seed": "211245842/train1", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19218182563781738, + "selected_utility_gain_over_base": 0.0012734085321426392, + "selector_regret": 0.02077534794807434, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7870428264141083, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "chart_id": "ms-77536801924a2a78688e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5428571428571428, + "pairwise_causal_calibration_confidence": 0.5928488173780985, + "pairwise_causal_calibration_ece": 0.05599809010651253, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.8553272798880048, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1151410346/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.7739987075328827, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "chart_id": "ms-9dd1ff4c297a81b13071", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5835206510970223, + "pairwise_causal_calibration_ece": 0.11178165495215062, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.4382011802480461, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1324586570262909, + "seed": "1324133122/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04025374352931976, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.867541342973709, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5952380952380952, + "pairwise_causal_calibration_confidence": 0.5668881781988476, + "pairwise_causal_calibration_ece": 0.02834991703924743, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.0614915153045223, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/train1", + "selected_candidate_index": 6, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d776afa14e6322cadedd", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5954452533432482, + "pairwise_causal_calibration_ece": 0.1017872953636167, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": 1.4327881584967936, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/train1", + "selected_candidate_index": 12, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-e887dbfa0133af46829d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7428571428571429, + "pairwise_causal_calibration_confidence": 0.6153268689343157, + "pairwise_causal_calibration_ece": 0.1374436377245066, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 1.363778237267606, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train1", + "selected_candidate_index": 12, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "chart_id": "ms-5799584ed40e211c50a5", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3804347826086957, + "pairwise_causal_calibration_confidence": 0.5546047203835289, + "pairwise_causal_calibration_ece": 0.1741699377748331, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 1.285926111665494, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7165453433990479, + "selected_utility_gain_over_base": -0.007059991359710693, + "selector_regret": 1.2834546566009521, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5947805123965023, + "pairwise_causal_calibration_ece": 0.045119558117126445, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.6469329014391034, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2302134484052658, + "seed": "307068879/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7697865515947342, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7b229b60d7f64fb624cf", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7619047619047619, + "pairwise_causal_calibration_confidence": 0.5700167609515335, + "pairwise_causal_calibration_ece": 0.19188800095322855, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.0317041974020489, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/train2", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.7130528092384338, + "selected_utility_gain_over_base": -1.2869471907615662, + "selector_regret": 1.2869471907615662, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.6026118586438607, + "pairwise_causal_calibration_ece": 0.0859451919771938, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.080232004048101, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7117276191711426, + "seed": "608016290/train2", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.22811375558376312, + "selected_utility_gain_over_base": -0.4975878745317459, + "selector_regret": 0.48361386358737946, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6181818181818182, + "pairwise_causal_calibration_confidence": 0.599086589574069, + "pairwise_causal_calibration_ece": 0.05893542172411783, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.3023349173152088, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7191348671913147, + "seed": "441046872/train2", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019179701805114746, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "task_id": "LiftPegUpright-v1", + "tau": 0.7478358458742458, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5663336315117011, + "pairwise_causal_calibration_ece": 0.08285581154113539, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.21722683006216226, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14937306940555573, + "seed": "1522272686/train2", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8506269305944443, + "task_id": "PickCube-v1", + "tau": 1.280136139242778, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-06babf453b22ba8f2407", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 1.8087128419207654, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/train2", + "selected_candidate_index": 11, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.16168592870235443, + "chart_id": "ms-0e1ff8d16372796dac97", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4369747899159664, + "pairwise_causal_calibration_confidence": 0.5433509927973336, + "pairwise_causal_calibration_ece": 0.10637620288136701, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.6353590329652259, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16168592870235443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8383140712976456, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.575, + "pairwise_causal_calibration_confidence": 0.5593899568017219, + "pairwise_causal_calibration_ece": 0.043120160988310785, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5260931359239889, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "seed": "1461069713/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0939030647277832, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.006073415279388428, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05936597287654877, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.001929501653648913, + "chart_id": "ms-34726202b475d29cdf38", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.25342071056365967, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.6200598937703816, + "pairwise_causal_calibration_ece": 0.02602328863996555, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4681106330126257, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.03885193541646004, + "seed": "626486813/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.001929501653648913, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036922433762811124, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.21456877514719963, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-4cc40587ec490257b1d2", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5210084033613446, + "pairwise_causal_calibration_confidence": 0.6169594875657232, + "pairwise_causal_calibration_ece": 0.09595108420437885, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.3818302310808093, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "selected_candidate_index": 14, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-62f8c18a15c566ec4b00", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.26666666666666666, + "pairwise_causal_calibration_confidence": 0.6313448801933792, + "pairwise_causal_calibration_ece": 0.3646782135267123, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": 0.8920977858545905, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train2", + "selected_candidate_index": 6, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.25, + "pairwise_causal_calibration_confidence": 0.5716458039521563, + "pairwise_causal_calibration_ece": 0.3486728786988602, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2513066342792345, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21886132657527924, + "seed": "767747424/train2", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07698849588632584, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.1418728306889534, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.01623930037021637, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.03588630631566048, + "chart_id": "ms-6b789356d65b3312fbae", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22781728208065033, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.6004519862037204, + "pairwise_causal_calibration_ece": 0.17643094206268192, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6996223306430975, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2718978226184845, + "seed": "278008524/train2", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.03588630631566048, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.23601151630282402, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.64, + "pairwise_causal_calibration_confidence": 0.5658799464871863, + "pairwise_causal_calibration_ece": 0.07971403378879252, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": 1.3003248867416755, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1845489013/train2", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.008267183788120747, + "chart_id": "ms-7df455efd413d77e6ce3", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.07774405181407928, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5201602929563037, + "pairwise_causal_calibration_ece": 0.020160292956303727, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.28364725281282954, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05094657465815544, + "seed": "626486813/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.008267183788120747, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.042679390870034695, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.026797477155923843, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7e295b5859c07c90487f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.26271408796310425, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5299145299145299, + "pairwise_causal_calibration_confidence": 0.5379909929329658, + "pairwise_causal_calibration_ece": 0.010407836919206468, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.23865231334213255, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "selected_candidate_index": 8, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "chart_id": "ms-8577816da2917a2bb63d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6363636363636364, + "pairwise_causal_calibration_confidence": 0.5450229275953111, + "pairwise_causal_calibration_ece": 0.09134070876832519, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.1772480571547321, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "585904130/train2", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8182844519615173, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "chart_id": "ms-8e56b0eaa2604db6749e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.525, + "pairwise_causal_calibration_confidence": 0.5698959198273771, + "pairwise_causal_calibration_ece": 0.06285173835176751, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5103624253528025, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058951716870069504, + "seed": "141984615/train2", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10283424705266953, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.14270895719528198, + "chart_id": "ms-972c5be72d0b24be7cd5", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5416666666666666, + "pairwise_causal_calibration_confidence": 0.5852475221522373, + "pairwise_causal_calibration_ece": 0.04358085548557049, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0890374052491354, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train2", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08332619816064835, + "selected_utility_gain_over_base": -0.059382759034633636, + "selector_regret": 1.9166738018393517, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "chart_id": "ms-9757fc494e79b5543bb6", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.551260515476342, + "pairwise_causal_calibration_ece": 0.06540615119032467, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9766279269153202, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0756516233086586, + "seed": "636680137/train2", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.041415613144636154, + "selected_utility_gain_over_base": -0.041719432920217514, + "selector_regret": 0.034236010164022446, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09491270035505295, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.05412294343113899, + "chart_id": "ms-a8731059713697ce5e2a", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.19710752367973328, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.525, + "pairwise_causal_calibration_confidence": 0.6426743208998875, + "pairwise_causal_calibration_ece": 0.19841177468421542, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.036160954547161, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04980705305933952, + "seed": "767747424/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.02992103435099125, + "selected_utility_gain_over_base": -0.024201909080147743, + "selector_regret": 0.019886018708348274, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14730047062039375, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-ac94b48febe523da15a5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5607087307240273, + "pairwise_causal_calibration_ece": 0.029809709781793304, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.07382412594656773, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/train2", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b49043f15f4846de5b13", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.40404040404040403, + "pairwise_causal_calibration_confidence": 0.5932032968584295, + "pairwise_causal_calibration_ece": 0.18916289281802573, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 1.2250107783962947, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train2", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04917396977543831, + "selected_utility_gain_over_base": -1.9508260302245617, + "selector_regret": 1.9508260302245617, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.0929974913597107, + "chart_id": "ms-b7d7021b782f73816083", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22398948669433594, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.48333333333333334, + "pairwise_causal_calibration_confidence": 0.582109700780416, + "pairwise_causal_calibration_ece": 0.10055984124716723, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.6042564099737049, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.051346004009246826, + "seed": "750448296/train2", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.03218920901417732, + "selected_utility_gain_over_base": -0.06080828234553337, + "selector_regret": 0.019156794995069504, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1726434826850891, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 1.879935987237566, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8fc1568ce66c45c8d8b", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.5582193713806621, + "pairwise_causal_calibration_ece": 0.07511396195267138, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.353137376402291, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.12510187923908234, + "seed": "747260437/train2", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.0880771279335022, + "selected_utility_gain_over_base": -1.9119228720664978, + "selector_regret": 0.03702475130558014, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15356136858463287, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.04958367347717285, + "chart_id": "ms-da555a03dd382d93e119", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.41414141414141414, + "pairwise_causal_calibration_confidence": 0.5704888369602925, + "pairwise_causal_calibration_ece": 0.15634742281887823, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": -0.017971861448823123, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train2", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04958367347717285, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9504163265228271, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "chart_id": "ms-f5dd1c43ebc4c0cfc3f1", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5498885743535615, + "pairwise_causal_calibration_ece": 0.10670850028777285, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.8289911704354562, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11310184001922607, + "seed": "1549324805/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04690708592534065, + "selected_utility_gain_over_base": -0.057607751339673996, + "selector_regret": 0.06619475409388542, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1843542456626892, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "chart_id": "ms-f9740a08a98cb251b761", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.325, + "pairwise_causal_calibration_confidence": 0.5927326895223773, + "pairwise_causal_calibration_ece": 0.2677326895223773, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 1.1202233214608746, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train2", + "selected_candidate_index": 0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9280846938490868, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6547619047619048, + "pairwise_causal_calibration_confidence": 0.5644513573496062, + "pairwise_causal_calibration_ece": 0.12058941098976882, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.016143196834125018, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.9470671281950391, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "chart_id": "ms-05f34a90fa7355262b8e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6111111111111112, + "pairwise_causal_calibration_confidence": 0.6090600639786808, + "pairwise_causal_calibration_ece": 0.2232694799037939, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.913369843520469, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train2", + "selected_candidate_index": 13, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "chart_id": "ms-2413d1155c87f0442485", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5454545454545454, + "pairwise_causal_calibration_confidence": 0.593444506997112, + "pairwise_causal_calibration_ece": 0.14259508480466812, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.0767725563667239, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/train2", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10647052526473999, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.89352947473526, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "chart_id": "ms-54346a8f79bd9dca1dfe", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.6001715319238825, + "pairwise_causal_calibration_ece": 0.16451829055105932, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.8332407953089316, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train2", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13743820786476135, + "selected_utility_gain_over_base": 0.012670740485191345, + "selector_regret": 1.8625617921352386, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-64f98d163b6bff29332b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5952380952380952, + "pairwise_causal_calibration_confidence": 0.6424422270814019, + "pairwise_causal_calibration_ece": 0.130927476896239, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.168316549791102, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train2", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "chart_id": "ms-c2ae2b78ab7891e74c80", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6470588235294118, + "pairwise_causal_calibration_confidence": 0.6259952855490101, + "pairwise_causal_calibration_ece": 0.15999043325783527, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 2.088591462744573, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 5.424022674560547e-06, + "seed": "509430392/train2", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 5.424022674560547e-06, + "selected_utility_gain_over_base": 5.200505256652832e-06, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9999945759773254, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "chart_id": "ms-d4c37ed44fb338c1d551", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6050420168067226, + "pairwise_causal_calibration_confidence": 0.5634795931375042, + "pairwise_causal_calibration_ece": 0.06596647647042775, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.8428291167331592, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21615806221961975, + "selected_utility_gain_over_base": 0.1291239857673645, + "selector_regret": 1.7838419377803802, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.2264360294338528, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "chart_id": "ms-146ff6c3f22b3a49af3a", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.5650883149039513, + "pairwise_causal_calibration_ece": 0.23175498157061808, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.6784286329948439, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11105132102966309, + "seed": "2061635888/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.05880081653594971, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.888948678970337, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "chart_id": "ms-15fc8cc6f74cf589056f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.25, + "pairwise_causal_calibration_confidence": 0.5949151821754458, + "pairwise_causal_calibration_ece": 0.34491518217544576, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.9421624207634716, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21447330713272095, + "seed": "655884605/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09131735563278198, + "selected_utility_gain_over_base": -0.07168932259082794, + "selector_regret": 0.12315595149993896, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.785526692867279, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "chart_id": "ms-36bc1364d2439099e9ed", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5791187642133746, + "pairwise_causal_calibration_ece": 0.20659552150091112, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.213636373075774, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "seed": "1113371261/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "chart_id": "ms-420280f15a522ce21c72", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7226890756302521, + "pairwise_causal_calibration_confidence": 0.6132925793293023, + "pairwise_causal_calibration_ece": 0.10939649630095027, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.0745551734071643, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19338026642799377, + "selected_utility_gain_over_base": -0.6307562291622162, + "selector_regret": 1.8066197335720062, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "chart_id": "ms-4aed1197d0906ae51c34", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.5594453160124555, + "pairwise_causal_calibration_ece": 0.06722278178918951, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.6223762395341456, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train2", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.575, + "pairwise_causal_calibration_confidence": 0.5804806828596304, + "pairwise_causal_calibration_ece": 0.020991444611033885, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7949698501065345, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "seed": "211245842/train2", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04297435283660889, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7661172300577164, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "chart_id": "ms-77536801924a2a78688e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5523809523809524, + "pairwise_causal_calibration_confidence": 0.5611842303245104, + "pairwise_causal_calibration_ece": 0.03328359958616477, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.8355344562806973, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "seed": "1151410346/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.32620343565940857, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4477952718734741, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "chart_id": "ms-9dd1ff4c297a81b13071", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5825263036735857, + "pairwise_causal_calibration_ece": 0.20318798204070004, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.24745057375120538, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19291165471076965, + "seed": "1324133122/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.10070674121379852, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8070883452892303, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7407407407407407, + "pairwise_causal_calibration_confidence": 0.5964473013091497, + "pairwise_causal_calibration_ece": 0.14429343943159095, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.1464704263550276, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/train2", + "selected_candidate_index": 4, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d776afa14e6322cadedd", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.6315411122591507, + "pairwise_causal_calibration_ece": 0.2148744455924842, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.5236864695041792, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/train2", + "selected_candidate_index": 12, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-e887dbfa0133af46829d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.25, + "pairwise_causal_calibration_confidence": 0.6423780754545177, + "pairwise_causal_calibration_ece": 0.3923780754545179, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": 1.4709100971517652, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train2", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.8183539509773254, + "selected_utility_gain_over_base": -1.1816460490226746, + "selector_regret": 1.1816460490226746, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.893807936331132, + "train_seed": "2" + } + ], + "schema_version": 1, + "selected_lambda": 1.0, + "selector_calibration_target_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "selector_calibration_target_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "selector_chart_feature_mode": "base_context_obs", + "selector_eval_target_content_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "selector_eval_target_split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "selector_source_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "selector_source_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "source_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "source_index": "data/cil_charts_rgb_refs/train/index.json", + "source_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "success_bonus": 2.0, + "summary": { + "base_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.26666666666666666, + "low": 0.26666666666666666, + "mean": 0.30625, + "n": 120, + "num_samples": 1 + }, + "mean": 0.30625, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 1.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 0.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 1.0, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 1.0, + "747260437/train1": 1.0, + "747260437/train2": 1.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.20654545454545453, + "low": 0.20654545454545453, + "mean": 0.19987878787878788, + "n": 5, + "num_samples": 1 + }, + "mean": 0.19987878787878788, + "per_task": { + "LiftPegUpright-v1": 0.2, + "PickCube-v1": 0.0, + "PullCube-v1": 0.36, + "PushCube-v1": 0.16666666666666666, + "StackCube-v1": 0.2727272727272727 + } + }, + "metric": "base_success", + "micro": { + "confidence": 0.95, + "high": 0.2777777777777778, + "low": 0.2777777777777778, + "mean": 0.2916666666666667, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "base_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.7473616204298904, + "low": 0.7473616204298904, + "mean": 0.7901180622977944, + "n": 120, + "num_samples": 1 + }, + "mean": 0.7901180622977944, + "per_seed": { + "1073161778/train0": 2.0, + "1073161778/train1": 2.0, + "1073161778/train2": 2.0, + "1113371261/train0": 0.15842342376708984, + "1113371261/train1": 0.15842342376708984, + "1113371261/train2": 0.15842342376708984, + "1151410346/train0": 0.2260012924671173, + "1151410346/train1": 0.2260012924671173, + "1151410346/train2": 0.2260012924671173, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 2.0, + "1324133122/train0": 0.09220491349697113, + "1324133122/train1": 0.09220491349697113, + "1324133122/train2": 0.09220491349697113, + "1324642411/train0": 0.10647052526473999, + "1324642411/train1": 0.10647052526473999, + "1324642411/train2": 0.10647052526473999, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.10283424705266953, + "141984615/train1": 0.10283424705266953, + "141984615/train2": 0.10283424705266953, + "1461069713/train0": 0.0939030647277832, + "1461069713/train1": 0.0939030647277832, + "1461069713/train2": 0.0939030647277832, + "1480856810/train0": 0.08703407645225525, + "1480856810/train1": 0.08703407645225525, + "1480856810/train2": 0.08703407645225525, + "1509980664/train0": 0.8334100246429443, + "1509980664/train1": 0.8334100246429443, + "1509980664/train2": 0.8334100246429443, + "1522272686/train0": 0.5839591026306152, + "1522272686/train1": 0.5839591026306152, + "1522272686/train2": 0.5839591026306152, + "1549324805/train0": 0.10451483726501465, + "1549324805/train1": 0.10451483726501465, + "1549324805/train2": 0.10451483726501465, + "1578990937/train0": 0.82413649559021, + "1578990937/train1": 0.82413649559021, + "1578990937/train2": 0.82413649559021, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 0.12476746737957001, + "1790188787/train1": 0.12476746737957001, + "1790188787/train2": 0.12476746737957001, + "1822210471/train0": 0.07191530615091324, + "1822210471/train1": 0.07191530615091324, + "1822210471/train2": 0.07191530615091324, + "1845489013/train0": 2.0, + "1845489013/train1": 2.0, + "1845489013/train2": 2.0, + "1999022574/train0": 2.0, + "1999022574/train1": 2.0, + "1999022574/train2": 2.0, + "2012211287/train0": 0.7236053347587585, + "2012211287/train1": 0.7236053347587585, + "2012211287/train2": 0.7236053347587585, + "2038594191/train0": 2.0, + "2038594191/train1": 2.0, + "2038594191/train2": 2.0, + "2054899126/train0": 2.0, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 0.05225050449371338, + "2061635888/train1": 0.05225050449371338, + "2061635888/train2": 0.05225050449371338, + "211245842/train0": 0.19090841710567474, + "211245842/train1": 0.19090841710567474, + "211245842/train2": 0.19090841710567474, + "278008524/train0": 0.042734989896416664, + "278008524/train1": 0.042734989896416664, + "278008524/train2": 0.042734989896416664, + "307068879/train0": 0.6000905632972717, + "307068879/train1": 0.6000905632972717, + "307068879/train2": 0.6000905632972717, + "412102563/train0": 2.0, + "412102563/train1": 2.0, + "412102563/train2": 2.0, + "440577711/train0": 0.036152780055999756, + "440577711/train1": 0.036152780055999756, + "440577711/train2": 0.036152780055999756, + "441046872/train0": 0.6999551653862, + "441046872/train1": 0.6999551653862, + "441046872/train2": 0.6999551653862, + "509430392/train0": 2.2351741790771484e-07, + "509430392/train1": 2.2351741790771484e-07, + "509430392/train2": 2.2351741790771484e-07, + "585904130/train0": 0.18171554803848267, + "585904130/train1": 0.18171554803848267, + "585904130/train2": 0.18171554803848267, + "608016290/train0": 0.725701630115509, + "608016290/train1": 0.725701630115509, + "608016290/train2": 0.725701630115509, + "626486813/train0": 0.00509834272088483, + "626486813/train1": 0.00509834272088483, + "626486813/train2": 0.00509834272088483, + "636680137/train0": 0.08313504606485367, + "636680137/train1": 0.08313504606485367, + "636680137/train2": 0.08313504606485367, + "655884605/train0": 0.16300667822360992, + "655884605/train1": 0.16300667822360992, + "655884605/train2": 0.16300667822360992, + "672328262/train0": 2.0, + "672328262/train1": 2.0, + "672328262/train2": 2.0, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 2.0, + "747260437/train1": 2.0, + "747260437/train2": 2.0, + "750448296/train0": 0.11759314934412639, + "750448296/train1": 0.11759314934412639, + "750448296/train2": 0.11759314934412639, + "767747424/train0": 0.5731993420049548, + "767747424/train1": 0.5731993420049548, + "767747424/train2": 0.5731993420049548 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.8070530691606586, + "low": 0.8070530691606586, + "mean": 0.695559797174015, + "n": 5, + "num_samples": 1 + }, + "mean": 0.695559797174015, + "per_task": { + "LiftPegUpright-v1": 0.9498705387115478, + "PickCube-v1": 0.5839591026306152, + "PullCube-v1": 0.7751704612234608, + "PushCube-v1": 0.3924041787783305, + "StackCube-v1": 0.7763947045261209 + } + }, + "metric": "base_utility", + "micro": { + "confidence": 0.95, + "high": 0.6946634994641904, + "low": 0.6946634994641904, + "mean": 0.7418195864423373, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "coverage": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.6263888888888888, + "low": 0.6263888888888888, + "mean": 0.6048611111111111, + "n": 120, + "num_samples": 1 + }, + "mean": 0.6048611111111111, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 1.0, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 0.5, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 1.0, + "636680137/train1": 1.0, + "636680137/train2": 1.0, + "655884605/train0": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 0.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 1.0, + "747260437/train1": 0.0, + "747260437/train2": 1.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.44703030303030306, + "low": 0.44703030303030306, + "mean": 0.5048080808080808, + "n": 5, + "num_samples": 1 + }, + "mean": 0.5048080808080808, + "per_task": { + "LiftPegUpright-v1": 0.6, + "PickCube-v1": 0.0, + "PullCube-v1": 0.52, + "PushCube-v1": 0.8888888888888888, + "StackCube-v1": 0.5151515151515151 + } + }, + "metric": "coverage", + "micro": { + "confidence": 0.95, + "high": 0.5277777777777778, + "low": 0.5277777777777778, + "mean": 0.5625, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "execute_generated": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.6263888888888888, + "low": 0.6263888888888888, + "mean": 0.6048611111111111, + "n": 120, + "num_samples": 1 + }, + "mean": 0.6048611111111111, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 1.0, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 0.5, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 1.0, + "636680137/train1": 1.0, + "636680137/train2": 1.0, + "655884605/train0": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 0.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 1.0, + "747260437/train1": 0.0, + "747260437/train2": 1.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.44703030303030306, + "low": 0.44703030303030306, + "mean": 0.5048080808080808, + "n": 5, + "num_samples": 1 + }, + "mean": 0.5048080808080808, + "per_task": { + "LiftPegUpright-v1": 0.6, + "PickCube-v1": 0.0, + "PullCube-v1": 0.52, + "PushCube-v1": 0.8888888888888888, + "StackCube-v1": 0.5151515151515151 + } + }, + "metric": "execute_generated", + "micro": { + "confidence": 0.95, + "high": 0.5277777777777778, + "low": 0.5277777777777778, + "mean": 0.5625, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "fallback_rate": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.37361111111111106, + "low": 0.37361111111111106, + "mean": 0.3951388888888889, + "n": 120, + "num_samples": 1 + }, + "mean": 0.3951388888888889, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 1.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 1.0, + "1151410346/train1": 1.0, + "1151410346/train2": 1.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 1.0, + "1324133122/train0": 1.0, + "1324133122/train1": 1.0, + "1324133122/train2": 1.0, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 1.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 1.0, + "141984615/train1": 1.0, + "141984615/train2": 1.0, + "1461069713/train0": 1.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 1.0, + "1522272686/train1": 1.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 1.0, + "278008524/train0": 1.0, + "278008524/train1": 1.0, + "278008524/train2": 1.0, + "307068879/train0": 1.0, + "307068879/train1": 1.0, + "307068879/train2": 1.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 1.0, + "441046872/train1": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.5, + "626486813/train1": 1.0, + "626486813/train2": 1.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 1.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 1.0, + "747260437/train2": 0.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.3333333333333333, + "767747424/train0": 0.75, + "767747424/train1": 0.75, + "767747424/train2": 0.75 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5529696969696969, + "low": 0.5529696969696969, + "mean": 0.4951919191919192, + "n": 5, + "num_samples": 1 + }, + "mean": 0.4951919191919192, + "per_task": { + "LiftPegUpright-v1": 0.4, + "PickCube-v1": 1.0, + "PullCube-v1": 0.48, + "PushCube-v1": 0.1111111111111111, + "StackCube-v1": 0.48484848484848486 + } + }, + "metric": "fallback_rate", + "micro": { + "confidence": 0.95, + "high": 0.4722222222222222, + "low": 0.4722222222222222, + "mean": 0.4375, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "hidden_chart_oracle_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.7944444444444445, + "low": 0.7944444444444445, + "mean": 0.7854166666666668, + "n": 120, + "num_samples": 1 + }, + "mean": 0.7854166666666668, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 1.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 1.0, + "1151410346/train1": 1.0, + "1151410346/train2": 1.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 1.0, + "1324133122/train0": 1.0, + "1324133122/train1": 1.0, + "1324133122/train2": 1.0, + "1324642411/train0": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 1.0, + "1522272686/train1": 1.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 1.0, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 1.0, + "278008524/train0": 0.5, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 1.0, + "307068879/train1": 1.0, + "307068879/train2": 1.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 1.0, + "441046872/train1": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.9039999999999999, + "low": 0.9039999999999999, + "mean": 0.8706666666666667, + "n": 5, + "num_samples": 1 + }, + "mean": 0.8706666666666667, + "per_task": { + "LiftPegUpright-v1": 1.0, + "PickCube-v1": 1.0, + "PullCube-v1": 0.52, + "PushCube-v1": 0.8333333333333334, + "StackCube-v1": 1.0 + } + }, + "metric": "hidden_chart_oracle_success", + "micro": { + "confidence": 0.95, + "high": 0.7083333333333334, + "low": 0.7083333333333334, + "mean": 0.7291666666666666, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "hidden_chart_oracle_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.6374536744836306, + "low": 1.6374536744836306, + "mean": 1.6190625454609595, + "n": 120, + "num_samples": 1 + }, + "mean": 1.6190625454609595, + "per_seed": { + "1073161778/train0": 2.0, + "1073161778/train1": 2.0, + "1073161778/train2": 2.0, + "1113371261/train0": 2.0, + "1113371261/train1": 2.0, + "1113371261/train2": 2.0, + "1151410346/train0": 2.0, + "1151410346/train1": 2.0, + "1151410346/train2": 2.0, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 2.0, + "1324133122/train0": 2.0, + "1324133122/train1": 2.0, + "1324133122/train2": 2.0, + "1324642411/train0": 2.0, + "1324642411/train1": 2.0, + "1324642411/train2": 2.0, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.2857772707939148, + "141984615/train1": 0.2857772707939148, + "141984615/train2": 0.2857772707939148, + "1461069713/train0": 0.1593424528837204, + "1461069713/train1": 0.1593424528837204, + "1461069713/train2": 0.1593424528837204, + "1480856810/train0": 0.20948036015033722, + "1480856810/train1": 0.20948036015033722, + "1480856810/train2": 0.20948036015033722, + "1509980664/train0": 2.0, + "1509980664/train1": 2.0, + "1509980664/train2": 2.0, + "1522272686/train0": 2.0, + "1522272686/train1": 2.0, + "1522272686/train2": 2.0, + "1549324805/train0": 0.2974560856819153, + "1549324805/train1": 0.2974560856819153, + "1549324805/train2": 0.2974560856819153, + "1578990937/train0": 2.0, + "1578990937/train1": 2.0, + "1578990937/train2": 2.0, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 2.0, + "1790188787/train1": 2.0, + "1790188787/train2": 2.0, + "1822210471/train0": 2.0, + "1822210471/train1": 2.0, + "1822210471/train2": 2.0, + "1845489013/train0": 2.0, + "1845489013/train1": 2.0, + "1845489013/train2": 2.0, + "1999022574/train0": 2.0, + "1999022574/train1": 2.0, + "1999022574/train2": 2.0, + "2012211287/train0": 2.0, + "2012211287/train1": 2.0, + "2012211287/train2": 2.0, + "2038594191/train0": 2.0, + "2038594191/train1": 2.0, + "2038594191/train2": 2.0, + "2054899126/train0": 2.0, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 2.0, + "2061635888/train1": 2.0, + "2061635888/train2": 2.0, + "211245842/train0": 2.0, + "211245842/train1": 2.0, + "211245842/train2": 2.0, + "278008524/train0": 1.1139086410403252, + "278008524/train1": 1.1139086410403252, + "278008524/train2": 1.1139086410403252, + "307068879/train0": 2.0, + "307068879/train1": 2.0, + "307068879/train2": 2.0, + "412102563/train0": 2.0, + "412102563/train1": 2.0, + "412102563/train2": 2.0, + "440577711/train0": 2.0, + "440577711/train1": 2.0, + "440577711/train2": 2.0, + "441046872/train0": 2.0, + "441046872/train1": 2.0, + "441046872/train2": 2.0, + "509430392/train0": 2.0, + "509430392/train1": 2.0, + "509430392/train2": 2.0, + "585904130/train0": 2.0, + "585904130/train1": 2.0, + "585904130/train2": 2.0, + "608016290/train0": 2.0, + "608016290/train1": 2.0, + "608016290/train2": 2.0, + "626486813/train0": 0.16558238118886948, + "626486813/train1": 0.16558238118886948, + "626486813/train2": 0.16558238118886948, + "636680137/train0": 0.17056432366371155, + "636680137/train1": 0.17056432366371155, + "636680137/train2": 0.17056432366371155, + "655884605/train0": 2.0, + "655884605/train1": 2.0, + "655884605/train2": 2.0, + "672328262/train0": 2.0, + "672328262/train1": 2.0, + "672328262/train2": 2.0, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 0.2786632478237152, + "747260437/train1": 0.2786632478237152, + "747260437/train2": 0.2786632478237152, + "750448296/train0": 1.4079964955647786, + "750448296/train1": 1.4079964955647786, + "750448296/train2": 1.4079964955647786, + "767747424/train0": 0.6737305596470833, + "767747424/train1": 0.6737305596470833, + "767747424/train2": 0.6737305596470833 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.8293575772047042, + "low": 1.8293575772047042, + "mean": 1.7696735892097153, + "n": 5, + "num_samples": 1 + }, + "mean": 1.7696735892097153, + "per_task": { + "LiftPegUpright-v1": 2.0, + "PickCube-v1": 2.0, + "PullCube-v1": 1.1467878860235214, + "PushCube-v1": 1.7015800600250561, + "StackCube-v1": 2.0 + } + }, + "metric": "hidden_chart_oracle_utility", + "micro": { + "confidence": 0.95, + "high": 1.4819050737553172, + "low": 1.4819050737553172, + "mean": 1.5183161981403828, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "outcome_ptr": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5611111111111111, + "low": 0.5611111111111111, + "mean": 0.5319444444444444, + "n": 120, + "num_samples": 1 + }, + "mean": 0.5319444444444444, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 1.0, + "1151410346/train1": 1.0, + "1151410346/train2": 1.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 1.0, + "1324133122/train2": 1.0, + "1324642411/train0": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 1.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 1.0, + "278008524/train0": 1.0, + "278008524/train1": 1.0, + "278008524/train2": 1.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 1.0, + "441046872/train1": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 0.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 1.0, + "636680137/train0": 1.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 1.0, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4339393939393939, + "low": 0.4339393939393939, + "mean": 0.49393939393939396, + "n": 5, + "num_samples": 1 + }, + "mean": 0.49393939393939396, + "per_task": { + "LiftPegUpright-v1": 0.5333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": 0.4666666666666667, + "PushCube-v1": 0.8333333333333334, + "StackCube-v1": 0.6363636363636364 + } + }, + "metric": "outcome_ptr", + "micro": { + "confidence": 0.95, + "high": 0.5694444444444444, + "low": 0.5694444444444444, + "mean": 0.5486111111111112, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "pairwise_causal_calibration_accuracy": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.4891761152890268, + "low": 0.4891761152890268, + "mean": 0.520337438676491, + "n": 117, + "num_samples": 1 + }, + "mean": 0.520337438676491, + "per_seed": { + "1073161778/train0": 0.5714285714285714, + "1073161778/train1": 0.39166666666666666, + "1073161778/train2": 0.7619047619047619, + "1113371261/train0": 0.5357142857142857, + "1113371261/train1": 0.42857142857142855, + "1113371261/train2": 0.7857142857142857, + "1151410346/train0": 0.5619047619047619, + "1151410346/train1": 0.5428571428571428, + "1151410346/train2": 0.5523809523809524, + "1261930456/train0": 0.494949494949495, + "1261930456/train1": 0.5636363636363636, + "1261930456/train2": 0.5952380952380952, + "1324133122/train0": 0.4642857142857143, + "1324133122/train1": 0.5357142857142857, + "1324133122/train2": 0.7857142857142857, + "1324642411/train0": 0.5583333333333333, + "1324642411/train1": 0.42016806722689076, + "1324642411/train2": 0.5454545454545454, + "1375492786/train0": 0.4533333333333333, + "1375492786/train1": 0.5263157894736842, + "1375492786/train2": 0.5357142857142857, + "141984615/train0": 0.5416666666666666, + "141984615/train1": 0.5166666666666667, + "141984615/train2": 0.525, + "1461069713/train0": 0.275, + "1461069713/train1": 0.375, + "1461069713/train2": 0.575, + "1480856810/train0": 0.26262626262626265, + "1480856810/train1": 0.4957983193277311, + "1480856810/train2": 0.6050420168067226, + "1509980664/train0": 0.4888888888888889, + "1509980664/train1": 0.1746031746031746, + "1509980664/train2": 0.6, + "1522272686/train0": 0.19444444444444445, + "1522272686/train1": 0.3055555555555556, + "1522272686/train2": 0.5, + "1549324805/train0": 0.75, + "1549324805/train1": 0.5833333333333334, + "1549324805/train2": 0.5083333333333333, + "1578990937/train0": 0.4358974358974359, + "1578990937/train1": 0.42727272727272725, + "1578990937/train2": 0.7226890756302521, + "1790188787/train0": 0.6272727272727273, + "1790188787/train1": 0.5263157894736842, + "1790188787/train2": 0.45454545454545453, + "1822210471/train0": 0.625, + "1822210471/train1": 0.5714285714285714, + "1822210471/train2": 0.325, + "1845489013/train0": 0.5652173913043478, + "1845489013/train1": 0.46153846153846156, + "1845489013/train2": 0.64, + "1999022574/train0": 0.6714285714285714, + "1999022574/train1": 0.7428571428571429, + "1999022574/train2": 0.25, + "2012211287/train0": 0.5042016806722689, + "2012211287/train1": 0.475, + "2012211287/train2": 0.3804347826086957, + "2038594191/train0": 0.3448275862068966, + "2038594191/train1": 0.3148148148148148, + "2038594191/train2": 0.40404040404040403, + "2054899126/train0": 0.275, + "2054899126/train1": 0.6228070175438597, + "2054899126/train2": 0.5210084033613446, + "2061635888/train0": 0.4722222222222222, + "2061635888/train1": 0.2777777777777778, + "2061635888/train2": 0.3333333333333333, + "211245842/train0": 0.6134453781512605, + "211245842/train1": 0.5416666666666666, + "211245842/train2": 0.575, + "278008524/train0": 0.6866161616161617, + "278008524/train1": 0.6382246376811594, + "278008524/train2": 0.47373737373737373, + "307068879/train0": 0.7032967032967034, + "307068879/train1": 0.7692307692307693, + "307068879/train2": 0.5714285714285714, + "412102563/train0": 0.6166666666666667, + "412102563/train1": 0.6666666666666666, + "412102563/train2": 0.4166666666666667, + "440577711/train0": 0.7586206896551724, + "440577711/train1": 0.4, + "440577711/train2": 0.6111111111111112, + "441046872/train0": 0.38181818181818183, + "441046872/train1": 0.5636363636363636, + "441046872/train2": 0.6181818181818182, + "509430392/train0": 0.6050420168067226, + "509430392/train1": 0.6, + "509430392/train2": 0.6470588235294118, + "585904130/train0": 0.5714285714285714, + "585904130/train1": 0.4266666666666667, + "585904130/train2": 0.6363636363636364, + "608016290/train0": 0.6571428571428571, + "608016290/train1": 0.4666666666666667, + "608016290/train2": 0.5166666666666667, + "626486813/train0": 0.475, + "626486813/train1": 0.4375, + "626486813/train2": 0.55, + "636680137/train0": 0.4166666666666667, + "636680137/train1": 0.6833333333333333, + "636680137/train2": 0.6166666666666667, + "655884605/train0": 0.6388888888888888, + "655884605/train1": 0.5, + "655884605/train2": 0.25, + "672328262/train0": 0.6470588235294118, + "672328262/train1": 0.6, + "672328262/train2": 0.26666666666666666, + "701667486/train0": 0.5, + "701667486/train1": 0.5952380952380952, + "701667486/train2": 0.7407407407407407, + "747260437/train0": 0.45, + "747260437/train1": 0.3416666666666667, + "747260437/train2": 0.6333333333333333, + "750448296/train0": 0.5365079365079365, + "750448296/train1": 0.3938034188034188, + "750448296/train2": 0.5599206349206348, + "767747424/train0": 0.5073717948717948, + "767747424/train1": 0.5107035121741004, + "767747424/train2": 0.4354723299576241 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4958531380101441, + "low": 0.4958531380101441, + "mean": 0.4933194384254135, + "n": 5, + "num_samples": 1 + }, + "mean": 0.4933194384254135, + "per_task": { + "LiftPegUpright-v1": 0.5555136707566375, + "PickCube-v1": 0.3333333333333333, + "PullCube-v1": 0.5062744920994355, + "PushCube-v1": 0.5428451728329846, + "StackCube-v1": 0.5286305231046764 + } + }, + "metric": "pairwise_causal_calibration_accuracy", + "micro": { + "confidence": 0.95, + "high": 0.5211886747045793, + "low": 0.5211886747045793, + "mean": 0.5179830825337585, + "n": 138, + "num_samples": 1 + }, + "n": 138 + }, + "pairwise_causal_calibration_confidence": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5868813051184918, + "low": 0.5868813051184918, + "mean": 0.5831914877849582, + "n": 117, + "num_samples": 1 + }, + "mean": 0.5831914877849582, + "per_seed": { + "1073161778/train0": 0.5516482746314322, + "1073161778/train1": 0.5665909321275702, + "1073161778/train2": 0.5700167609515335, + "1113371261/train0": 0.5411387050524964, + "1113371261/train1": 0.53349783320347, + "1113371261/train2": 0.5791187642133746, + "1151410346/train0": 0.5447959029344634, + "1151410346/train1": 0.5928488173780985, + "1151410346/train2": 0.5611842303245104, + "1261930456/train0": 0.6511354281579875, + "1261930456/train1": 0.6560658348856379, + "1261930456/train2": 0.6424422270814019, + "1324133122/train0": 0.5778483044050591, + "1324133122/train1": 0.5835206510970223, + "1324133122/train2": 0.5825263036735857, + "1324642411/train0": 0.5894684248151024, + "1324642411/train1": 0.6037811380752174, + "1324642411/train2": 0.593444506997112, + "1375492786/train0": 0.5366248224651405, + "1375492786/train1": 0.5675437560777116, + "1375492786/train2": 0.5607087307240273, + "141984615/train0": 0.5684739620257733, + "141984615/train1": 0.5676327735960262, + "141984615/train2": 0.5698959198273771, + "1461069713/train0": 0.5596745083517122, + "1461069713/train1": 0.5561091294148327, + "1461069713/train2": 0.5593899568017219, + "1480856810/train0": 0.5720642439373731, + "1480856810/train1": 0.5496493820116296, + "1480856810/train2": 0.5634795931375042, + "1509980664/train0": 0.5462356837170409, + "1509980664/train1": 0.5479465839142913, + "1509980664/train2": 0.5594453160124555, + "1522272686/train0": 0.6053552926240098, + "1522272686/train1": 0.6354932924180177, + "1522272686/train2": 0.5663336315117011, + "1549324805/train0": 0.5625148880216412, + "1549324805/train1": 0.5360665332231191, + "1549324805/train2": 0.5498885743535615, + "1578990937/train0": 0.6103413950316227, + "1578990937/train1": 0.6263703949729311, + "1578990937/train2": 0.6132925793293023, + "1790188787/train0": 0.5867290403888722, + "1790188787/train1": 0.5937233509400742, + "1790188787/train2": 0.6001715319238825, + "1822210471/train0": 0.5988235112981328, + "1822210471/train1": 0.6072514522495137, + "1822210471/train2": 0.5927326895223773, + "1845489013/train0": 0.5667479107448529, + "1845489013/train1": 0.5409684680296629, + "1845489013/train2": 0.5658799464871863, + "1999022574/train0": 0.6012777562735464, + "1999022574/train1": 0.6153268689343157, + "1999022574/train2": 0.6423780754545177, + "2012211287/train0": 0.5508434661236096, + "2012211287/train1": 0.5273415392712859, + "2012211287/train2": 0.5546047203835289, + "2038594191/train0": 0.6255201768318296, + "2038594191/train1": 0.6741121240452155, + "2038594191/train2": 0.5932032968584295, + "2054899126/train0": 0.5896370284261352, + "2054899126/train1": 0.5635260175535555, + "2054899126/train2": 0.6169594875657232, + "2061635888/train0": 0.5740299161789739, + "2061635888/train1": 0.555294391767556, + "2061635888/train2": 0.5650883149039513, + "211245842/train0": 0.5845752206379711, + "211245842/train1": 0.5914410404762934, + "211245842/train2": 0.5804806828596304, + "278008524/train0": 0.5986517424099531, + "278008524/train1": 0.5821464869551558, + "278008524/train2": 0.5854704115820064, + "307068879/train0": 0.5795355714423935, + "307068879/train1": 0.5687146745124166, + "307068879/train2": 0.5947805123965023, + "412102563/train0": 0.6052287842223839, + "412102563/train1": 0.5954452533432482, + "412102563/train2": 0.6315411122591507, + "440577711/train0": 0.5841905535939795, + "440577711/train1": 0.5669827903966217, + "440577711/train2": 0.6090600639786808, + "441046872/train0": 0.5622784804732108, + "441046872/train1": 0.5559054729794208, + "441046872/train2": 0.599086589574069, + "509430392/train0": 0.5998439015094801, + "509430392/train1": 0.6128506472019413, + "509430392/train2": 0.6259952855490101, + "585904130/train0": 0.5597842433887508, + "585904130/train1": 0.5738533196584167, + "585904130/train2": 0.5450229275953111, + "608016290/train0": 0.5840070772055741, + "608016290/train1": 0.6130721925494387, + "608016290/train2": 0.6026118586438607, + "626486813/train0": 0.5748306935568978, + "626486813/train1": 0.5875658211040858, + "626486813/train2": 0.5701100933633427, + "636680137/train0": 0.5986660164452574, + "636680137/train1": 0.5644217102464404, + "636680137/train2": 0.551260515476342, + "655884605/train0": 0.605234361994273, + "655884605/train1": 0.5888894132525583, + "655884605/train2": 0.5949151821754458, + "672328262/train0": 0.6057740013777554, + "672328262/train1": 0.6362714146835213, + "672328262/train2": 0.6313448801933792, + "701667486/train0": 0.5760031182899459, + "701667486/train1": 0.5668881781988476, + "701667486/train2": 0.5964473013091497, + "747260437/train0": 0.5630309153002454, + "747260437/train1": 0.5662548497567926, + "747260437/train2": 0.5582193713806621, + "750448296/train0": 0.5831784025576913, + "750448296/train1": 0.5885015589506045, + "750448296/train2": 0.5772695267607532, + "767747424/train0": 0.5776709098378972, + "767747424/train1": 0.5904083439254464, + "767747424/train2": 0.5739155276455858 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5820289127889391, + "low": 0.5820289127889391, + "mean": 0.5876270483074112, + "n": 5, + "num_samples": 1 + }, + "mean": 0.5876270483074112, + "per_task": { + "LiftPegUpright-v1": 0.5720692082177231, + "PickCube-v1": 0.6023940721845762, + "PullCube-v1": 0.5796546075128097, + "PushCube-v1": 0.6000598858100837, + "StackCube-v1": 0.5839574678118631 + } + }, + "metric": "pairwise_causal_calibration_confidence", + "micro": { + "confidence": 0.95, + "high": 0.5839591163438812, + "low": 0.5839591163438812, + "mean": 0.5830149466708441, + "n": 138, + "num_samples": 1 + }, + "n": 138 + }, + "pairwise_causal_calibration_ece": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.11508507420044621, + "low": 0.11508507420044621, + "mean": 0.13498521085291207, + "n": 120, + "num_samples": 1 + }, + "mean": 0.13498521085291207, + "per_seed": { + "1073161778/train0": 0.1125523173289727, + "1073161778/train1": 0.17492426546090348, + "1073161778/train2": 0.19188800095322855, + "1113371261/train0": 0.025500013084200863, + "1113371261/train1": 0.10492640463204145, + "1113371261/train2": 0.20659552150091112, + "1151410346/train0": 0.05152312612164488, + "1151410346/train1": 0.05599809010651253, + "1151410346/train2": 0.03328359958616477, + "1261930456/train0": 0.3204647685440312, + "1261930456/train1": 0.305306037200651, + "1261930456/train2": 0.130927476896239, + "1324133122/train0": 0.11356259011934491, + "1324133122/train1": 0.11178165495215062, + "1324133122/train2": 0.20318798204070004, + "1324642411/train0": 0.03846474579568359, + "1324642411/train1": 0.18361307084832634, + "1324642411/train2": 0.14259508480466812, + "1375492786/train0": 0.08329148913180733, + "1375492786/train1": 0.04122796660402744, + "1375492786/train2": 0.029809709781793304, + "141984615/train0": 0.0268072953591066, + "141984615/train1": 0.0509661069293595, + "141984615/train2": 0.06285173835176751, + "1461069713/train0": 0.2846745083517123, + "1461069713/train1": 0.1811091294148327, + "1461069713/train2": 0.043120160988310785, + "1480856810/train0": 0.30943798131111067, + "1480856810/train1": 0.05385106268389854, + "1480856810/train2": 0.06596647647042775, + "1509980664/train0": 0.057346794828152005, + "1509980664/train1": 0.37334340931111676, + "1509980664/train2": 0.06722278178918951, + "1522272686/train0": 0.4109108481795651, + "1522272686/train1": 0.32993773686246214, + "1522272686/train2": 0.08285581154113539, + "1549324805/train0": 0.18748511197835893, + "1549324805/train1": 0.07111703737429365, + "1549324805/train2": 0.10670850028777285, + "1578990937/train0": 0.17444395913418703, + "1578990937/train1": 0.1990976677002039, + "1578990937/train2": 0.10939649630095027, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.04054368688385518, + "1790188787/train1": 0.10426662637283049, + "1790188787/train2": 0.16451829055105932, + "1822210471/train0": 0.11653574747746832, + "1822210471/train1": 0.1057225969843389, + "1822210471/train2": 0.2677326895223773, + "1845489013/train0": 0.025257247762570406, + "1845489013/train1": 0.11303375788512275, + "1845489013/train2": 0.07971403378879252, + "1999022574/train0": 0.07015081515502501, + "1999022574/train1": 0.1374436377245066, + "1999022574/train2": 0.3923780754545179, + "2012211287/train0": 0.08899904912532385, + "2012211287/train1": 0.05234153927128593, + "2012211287/train2": 0.1741699377748331, + "2038594191/train0": 0.3473762370913279, + "2038594191/train1": 0.3592973092304006, + "2038594191/train2": 0.18916289281802573, + "2054899126/train0": 0.15731851421306756, + "2054899126/train1": 0.029640499995152086, + "2054899126/train2": 0.04797554210218943, + "2061635888/train0": 0.1173721779867959, + "2061635888/train1": 0.27751661398977834, + "2061635888/train2": 0.23175498157061808, + "211245842/train0": 0.04745706568944737, + "211245842/train1": 0.10787248047280223, + "211245842/train2": 0.020991444611033885, + "278008524/train0": 0.10091717082382054, + "278008524/train1": 0.11826680760261055, + "278008524/train2": 0.16638918244078008, + "307068879/train0": 0.1237611318543097, + "307068879/train1": 0.2005160947183524, + "307068879/train2": 0.045119558117126445, + "412102563/train0": 0.08071518251541011, + "412102563/train1": 0.1017872953636167, + "412102563/train2": 0.2148744455924842, + "440577711/train0": 0.17443013606119306, + "440577711/train1": 0.16698279039662164, + "440577711/train2": 0.2232694799037939, + "441046872/train0": 0.1804602986550292, + "441046872/train1": 0.02996166181944273, + "441046872/train2": 0.05893542172411783, + "509430392/train0": 0.07491598280692124, + "509430392/train1": 0.1382932703018193, + "509430392/train2": 0.15999043325783527, + "585904130/train0": 0.08130419295423207, + "585904130/train1": 0.14718665299174993, + "585904130/train2": 0.09134070876832519, + "608016290/train0": 0.07313577993728299, + "608016290/train1": 0.157312300699128, + "608016290/train2": 0.0859451919771938, + "626486813/train0": 0.15407519274835713, + "626486813/train1": 0.23910795594093184, + "626486813/train2": 0.02309179079813464, + "636680137/train0": 0.18530981522251466, + "636680137/train1": 0.11891162308689304, + "636680137/train2": 0.06540615119032467, + "655884605/train0": 0.09534125846142073, + "655884605/train1": 0.08888941325255821, + "655884605/train2": 0.34491518217544576, + "672328262/train0": 0.08437381833742622, + "672328262/train1": 0.05483727064861237, + "672328262/train2": 0.3646782135267123, + "701667486/train0": 0.20253601281087805, + "701667486/train1": 0.02834991703924743, + "701667486/train2": 0.14429343943159095, + "747260437/train0": 0.11303091530024553, + "747260437/train1": 0.2245881830901258, + "747260437/train2": 0.07511396195267138, + "750448296/train0": 0.06832403574472125, + "750448296/train1": 0.2032871427289963, + "750448296/train2": 0.08824336924083552, + "767747424/train0": 0.15246329356188465, + "767747424/train1": 0.1406609793333544, + "767747424/train2": 0.1659671732959123 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.15439315889174393, + "low": 0.15439315889174393, + "mean": 0.16214660713386758, + "n": 5, + "num_samples": 1 + }, + "mean": 0.16214660713386758, + "per_task": { + "LiftPegUpright-v1": 0.11666816996110206, + "PickCube-v1": 0.2745681321943875, + "PullCube-v1": 0.12491436687229007, + "PushCube-v1": 0.15543541117172033, + "StackCube-v1": 0.13914695546983782 + } + }, + "metric": "pairwise_causal_calibration_ece", + "micro": { + "confidence": 0.95, + "high": 0.13599733064338715, + "low": 0.13599733064338715, + "mean": 0.13424994022928516, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "pairwise_causal_calibration_pairs": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 87.40624999999999, + "low": 87.40624999999999, + "mean": 82.21180555555556, + "n": 120, + "num_samples": 1 + }, + "mean": 82.21180555555556, + "per_seed": { + "1073161778/train0": 119.0, + "1073161778/train1": 120.0, + "1073161778/train2": 84.0, + "1113371261/train0": 28.0, + "1113371261/train1": 28.0, + "1113371261/train2": 28.0, + "1151410346/train0": 105.0, + "1151410346/train1": 105.0, + "1151410346/train2": 105.0, + "1261930456/train0": 99.0, + "1261930456/train1": 110.0, + "1261930456/train2": 84.0, + "1324133122/train0": 28.0, + "1324133122/train1": 28.0, + "1324133122/train2": 28.0, + "1324642411/train0": 120.0, + "1324642411/train1": 119.0, + "1324642411/train2": 110.0, + "1375492786/train0": 75.0, + "1375492786/train1": 114.0, + "1375492786/train2": 84.0, + "141984615/train0": 120.0, + "141984615/train1": 120.0, + "141984615/train2": 120.0, + "1461069713/train0": 120.0, + "1461069713/train1": 120.0, + "1461069713/train2": 120.0, + "1480856810/train0": 99.0, + "1480856810/train1": 119.0, + "1480856810/train2": 119.0, + "1509980664/train0": 90.0, + "1509980664/train1": 63.0, + "1509980664/train2": 90.0, + "1522272686/train0": 36.0, + "1522272686/train1": 36.0, + "1522272686/train2": 36.0, + "1549324805/train0": 120.0, + "1549324805/train1": 120.0, + "1549324805/train2": 120.0, + "1578990937/train0": 117.0, + "1578990937/train1": 110.0, + "1578990937/train2": 119.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 110.0, + "1790188787/train1": 114.0, + "1790188787/train2": 110.0, + "1822210471/train0": 40.0, + "1822210471/train1": 49.0, + "1822210471/train2": 40.0, + "1845489013/train0": 46.0, + "1845489013/train1": 13.0, + "1845489013/train2": 25.0, + "1999022574/train0": 70.0, + "1999022574/train1": 70.0, + "1999022574/train2": 76.0, + "2012211287/train0": 119.0, + "2012211287/train1": 120.0, + "2012211287/train2": 92.0, + "2038594191/train0": 29.0, + "2038594191/train1": 54.0, + "2038594191/train2": 99.0, + "2054899126/train0": 60.0, + "2054899126/train1": 57.0, + "2054899126/train2": 59.5, + "2061635888/train0": 36.0, + "2061635888/train1": 36.0, + "2061635888/train2": 36.0, + "211245842/train0": 119.0, + "211245842/train1": 120.0, + "211245842/train2": 120.0, + "278008524/train0": 109.5, + "278008524/train1": 106.0, + "278008524/train2": 109.5, + "307068879/train0": 91.0, + "307068879/train1": 91.0, + "307068879/train2": 91.0, + "412102563/train0": 60.0, + "412102563/train1": 39.0, + "412102563/train2": 84.0, + "440577711/train0": 29.0, + "440577711/train1": 15.0, + "440577711/train2": 54.0, + "441046872/train0": 55.0, + "441046872/train1": 55.0, + "441046872/train2": 55.0, + "509430392/train0": 119.0, + "509430392/train1": 105.0, + "509430392/train2": 119.0, + "585904130/train0": 84.0, + "585904130/train1": 75.0, + "585904130/train2": 110.0, + "608016290/train0": 105.0, + "608016290/train1": 105.0, + "608016290/train2": 120.0, + "626486813/train0": 120.0, + "626486813/train1": 120.0, + "626486813/train2": 120.0, + "636680137/train0": 120.0, + "636680137/train1": 120.0, + "636680137/train2": 120.0, + "655884605/train0": 36.0, + "655884605/train1": 36.0, + "655884605/train2": 36.0, + "672328262/train0": 17.0, + "672328262/train1": 35.0, + "672328262/train2": 30.0, + "701667486/train0": 42.0, + "701667486/train1": 84.0, + "701667486/train2": 54.0, + "747260437/train0": 120.0, + "747260437/train1": 120.0, + "747260437/train2": 120.0, + "750448296/train0": 108.0, + "750448296/train1": 101.66666666666667, + "750448296/train2": 108.0, + "767747424/train0": 119.25, + "767747424/train1": 119.0, + "767747424/train2": 119.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 77.07490909090909, + "low": 77.07490909090909, + "mean": 77.60379797979797, + "n": 5, + "num_samples": 1 + }, + "mean": 77.60379797979797, + "per_task": { + "LiftPegUpright-v1": 94.8, + "PickCube-v1": 36.0, + "PullCube-v1": 92.32, + "PushCube-v1": 97.44444444444444, + "StackCube-v1": 67.45454545454545 + } + }, + "metric": "pairwise_causal_calibration_pairs", + "micro": { + "confidence": 0.95, + "high": 82.16666666666667, + "low": 82.16666666666667, + "mean": 86.34722222222223, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "predicted_margin": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.0572726011192959, + "low": 1.0572726011192959, + "mean": 1.032678135142672, + "n": 120, + "num_samples": 1 + }, + "mean": 1.032678135142672, + "per_seed": { + "1073161778/train0": 1.197390750232301, + "1073161778/train1": 1.057082412822946, + "1073161778/train2": 1.0317041974020489, + "1113371261/train0": -0.008407185293451747, + "1113371261/train1": 0.2579825916955958, + "1113371261/train2": -0.213636373075774, + "1151410346/train0": 0.8492504414490131, + "1151410346/train1": 0.8553272798880048, + "1151410346/train2": 0.8355344562806973, + "1261930456/train0": 1.471050707973179, + "1261930456/train1": 1.3365043159479286, + "1261930456/train2": 1.168316549791102, + "1324133122/train0": -0.41376711119187704, + "1324133122/train1": 0.4382011802480461, + "1324133122/train2": -0.24745057375120538, + "1324642411/train0": 2.1833463985406447, + "1324642411/train1": 1.7807138675039442, + "1324642411/train2": 1.0767725563667239, + "1375492786/train0": 0.3369485573884625, + "1375492786/train1": -0.11389548031987307, + "1375492786/train2": -0.07382412594656773, + "141984615/train0": 0.6161038657373066, + "141984615/train1": 0.7260454977260915, + "141984615/train2": 0.5103624253528025, + "1461069713/train0": 0.5914274057916176, + "1461069713/train1": 0.6317591128320483, + "1461069713/train2": 0.5260931359239889, + "1480856810/train0": 2.1158648725170077, + "1480856810/train1": 1.8867636980590103, + "1480856810/train2": 1.8428291167331592, + "1509980664/train0": 0.5816690584072294, + "1509980664/train1": 0.5626192514371053, + "1509980664/train2": 0.6223762395341456, + "1522272686/train0": -0.7902719177719542, + "1522272686/train1": 0.20562222929953555, + "1522272686/train2": 0.21722683006216226, + "1549324805/train0": 1.530739092243051, + "1549324805/train1": 1.2279357729216895, + "1549324805/train2": 1.8289911704354562, + "1578990937/train0": 1.1959458964053806, + "1578990937/train1": 1.2514512085861837, + "1578990937/train2": 1.0745551734071643, + "1747178030/train0": 1.512536567520637, + "1747178030/train1": 1.639637290759596, + "1747178030/train2": 1.8087128419207654, + "1790188787/train0": 1.9875622503266075, + "1790188787/train1": 1.9148838772749794, + "1790188787/train2": 1.8332407953089316, + "1822210471/train0": 1.3378277706579857, + "1822210471/train1": 1.2817760714600654, + "1822210471/train2": 1.1202233214608746, + "1845489013/train0": 1.340214222459458, + "1845489013/train1": 1.2941213610673934, + "1845489013/train2": 1.3003248867416755, + "1999022574/train0": 1.3600458578396741, + "1999022574/train1": 1.363778237267606, + "1999022574/train2": 1.4709100971517652, + "2012211287/train0": 1.2153078550922283, + "2012211287/train1": 0.7823168935976765, + "2012211287/train2": 1.285926111665494, + "2038594191/train0": 1.1448828501178339, + "2038594191/train1": 1.6895335177823154, + "2038594191/train2": 1.2250107783962947, + "2054899126/train0": 1.7750331290407644, + "2054899126/train1": 1.7683790333511047, + "2054899126/train2": 1.6308831091591878, + "2061635888/train0": 0.7792589748622651, + "2061635888/train1": 0.7080441039530474, + "2061635888/train2": 0.6784286329948439, + "211245842/train0": 0.9680028345756826, + "211245842/train1": 1.097151631086689, + "211245842/train2": 0.7949698501065345, + "278008524/train0": 0.37442228448179016, + "278008524/train1": 0.3884435252467696, + "278008524/train2": 0.3408252345971372, + "307068879/train0": 0.18766723699772747, + "307068879/train1": 0.12417642494936398, + "307068879/train2": 0.6469329014391034, + "412102563/train0": 1.4315714539962017, + "412102563/train1": 1.4327881584967936, + "412102563/train2": 1.5236864695041792, + "440577711/train0": 1.839399049033097, + "440577711/train1": 2.0021154026255705, + "440577711/train2": 1.913369843520469, + "441046872/train0": 0.422930196465892, + "441046872/train1": 0.2197157946327053, + "441046872/train2": 0.3023349173152088, + "509430392/train0": 2.434169345753335, + "509430392/train1": 2.300921534892476, + "509430392/train2": 2.088591462744573, + "585904130/train0": 1.4903299974496993, + "585904130/train1": 1.4765054459187703, + "585904130/train2": 1.1772480571547321, + "608016290/train0": 1.2718540448404614, + "608016290/train1": 0.9190150483991828, + "608016290/train2": 1.080232004048101, + "626486813/train0": 0.767002576627011, + "626486813/train1": 0.9231447045402131, + "626486813/train2": 0.3758789429127276, + "636680137/train0": 1.70117503855084, + "636680137/train1": 0.9868011033647903, + "636680137/train2": 0.9766279269153202, + "655884605/train0": 0.9134188949984638, + "655884605/train1": 1.0866492061529267, + "655884605/train2": 0.9421624207634716, + "672328262/train0": 1.0088593202243494, + "672328262/train1": 1.1331098956489296, + "672328262/train2": 0.8920977858545905, + "701667486/train0": 1.0829493379419586, + "701667486/train1": 1.0614915153045223, + "701667486/train2": 1.1464704263550276, + "747260437/train0": 1.0110612897022033, + "747260437/train1": 0.6563621545761416, + "747260437/train2": 1.353137376402291, + "750448296/train0": 0.8860884501921475, + "750448296/train1": 0.7733918882065844, + "750448296/train2": 0.892383539462905, + "767747424/train0": 0.6516149173146432, + "767747424/train1": 0.8997026322297819, + "767747424/train2": 0.5403697337834386 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.6462128151272032, + "low": 0.6462128151272032, + "mean": 0.8582451650942277, + "n": 5, + "num_samples": 1 + }, + "mean": 0.8582451650942277, + "per_task": { + "LiftPegUpright-v1": 0.7829724526600295, + "PickCube-v1": -0.12247428613675211, + "PullCube-v1": 0.9547622553200451, + "PushCube-v1": 1.843134202495152, + "StackCube-v1": 0.832831201132664 + } + }, + "metric": "predicted_margin", + "micro": { + "confidence": 0.95, + "high": 1.0141786521896266, + "low": 1.0141786521896266, + "mean": 0.9975290164082237, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "proposal_oracle_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.6430555555555557, + "low": 0.6430555555555557, + "mean": 0.5847222222222223, + "n": 120, + "num_samples": 1 + }, + "mean": 0.5847222222222223, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 1.0, + "1151410346/train2": 0.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 1.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 0.5, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.5, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 1.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 0.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4496969696969697, + "low": 0.4496969696969697, + "mean": 0.5074747474747474, + "n": 5, + "num_samples": 1 + }, + "mean": 0.5074747474747474, + "per_task": { + "LiftPegUpright-v1": 0.6, + "PickCube-v1": 0.0, + "PullCube-v1": 0.5333333333333333, + "PushCube-v1": 0.8888888888888888, + "StackCube-v1": 0.5151515151515151 + } + }, + "metric": "proposal_oracle_success", + "micro": { + "confidence": 0.95, + "high": 0.5416666666666666, + "low": 0.5416666666666666, + "mean": 0.5694444444444444, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "proposal_oracle_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.342259801506427, + "low": 1.342259801506427, + "mean": 1.251578581963097, + "n": 120, + "num_samples": 1 + }, + "mean": 1.251578581963097, + "per_seed": { + "1073161778/train0": 2.0, + "1073161778/train1": 2.0, + "1073161778/train2": 2.0, + "1113371261/train0": 0.10274861752986908, + "1113371261/train1": 0.18093222379684448, + "1113371261/train2": 0.19527167081832886, + "1151410346/train0": 0.8663521409034729, + "1151410346/train1": 2.0, + "1151410346/train2": 0.5522047281265259, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 2.0, + "1324133122/train0": 0.054989203810691833, + "1324133122/train1": 0.1324586570262909, + "1324133122/train2": 0.19291165471076965, + "1324642411/train0": 0.14406004548072815, + "1324642411/train1": 2.0, + "1324642411/train2": 2.0, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.015177905559539795, + "141984615/train1": 0.037011127918958664, + "141984615/train2": 0.058951716870069504, + "1461069713/train0": 0.06019836664199829, + "1461069713/train1": 0.016585012897849083, + "1461069713/train2": 0.09997648000717163, + "1480856810/train0": 2.0, + "1480856810/train1": 2.0, + "1480856810/train2": 2.0, + "1509980664/train0": 2.0, + "1509980664/train1": 2.0, + "1509980664/train2": 2.0, + "1522272686/train0": 0.18325337767601013, + "1522272686/train1": 0.07014383375644684, + "1522272686/train2": 0.14937306940555573, + "1549324805/train0": 0.060871422290802, + "1549324805/train1": 0.19892151653766632, + "1549324805/train2": 0.11310184001922607, + "1578990937/train0": 2.0, + "1578990937/train1": 2.0, + "1578990937/train2": 2.0, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 2.0, + "1790188787/train1": 2.0, + "1790188787/train2": 2.0, + "1822210471/train0": 2.0, + "1822210471/train1": 2.0, + "1822210471/train2": 2.0, + "1845489013/train0": 2.0, + "1845489013/train1": 2.0, + "1845489013/train2": 2.0, + "1999022574/train0": 2.0, + "1999022574/train1": 2.0, + "1999022574/train2": 2.0, + "2012211287/train0": 2.0, + "2012211287/train1": 2.0, + "2012211287/train2": 2.0, + "2038594191/train0": 2.0, + "2038594191/train1": 2.0, + "2038594191/train2": 2.0, + "2054899126/train0": 1.077283151447773, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 0.09799930453300476, + "2061635888/train1": 0.07580336928367615, + "2061635888/train2": 0.11105132102966309, + "211245842/train0": 2.0, + "211245842/train1": 0.21295717358589172, + "211245842/train2": 0.23388276994228363, + "278008524/train0": 1.0733563229441643, + "278008524/train1": 1.0376745946705341, + "278008524/train2": 1.1359489113092422, + "307068879/train0": 0.2115575522184372, + "307068879/train1": 0.5876073241233826, + "307068879/train2": 0.2302134484052658, + "412102563/train0": 2.0, + "412102563/train1": 2.0, + "412102563/train2": 2.0, + "440577711/train0": 2.0, + "440577711/train1": 2.0, + "440577711/train2": 2.0, + "441046872/train0": 2.0, + "441046872/train1": 0.7233588695526123, + "441046872/train2": 0.7191348671913147, + "509430392/train0": 2.0, + "509430392/train1": 2.0, + "509430392/train2": 5.424022674560547e-06, + "585904130/train0": 2.0, + "585904130/train1": 2.0, + "585904130/train2": 2.0, + "608016290/train0": 2.0, + "608016290/train1": 2.0, + "608016290/train2": 0.7117276191711426, + "626486813/train0": 0.04469547560438514, + "626486813/train1": 0.035827874671667814, + "626486813/train2": 0.04489925503730774, + "636680137/train0": 0.08826130628585815, + "636680137/train1": 0.06741118431091309, + "636680137/train2": 0.0756516233086586, + "655884605/train0": 0.13732868432998657, + "655884605/train1": 0.8128758072853088, + "655884605/train2": 0.21447330713272095, + "672328262/train0": 2.0, + "672328262/train1": 2.0, + "672328262/train2": 2.0, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 0.10839563608169556, + "747260437/train1": 0.06328952312469482, + "747260437/train2": 0.12510187923908234, + "750448296/train0": 0.7526603279014429, + "750448296/train1": 1.3705032741030057, + "750448296/train2": 1.3504486680030823, + "767747424/train0": 1.0424291878007352, + "767747424/train1": 1.0629530912265182, + "767747424/train2": 1.0671670949086547 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.0449712097333237, + "low": 1.0449712097333237, + "mean": 1.1196794969856438, + "n": 5, + "num_samples": 1 + }, + "mean": 1.1196794969856438, + "per_task": { + "LiftPegUpright-v1": 1.4122399787108104, + "PickCube-v1": 0.13425676027933756, + "PullCube-v1": 1.1093241602430741, + "PushCube-v1": 1.7857814149724112, + "StackCube-v1": 1.1567951707225856 + } + }, + "metric": "proposal_oracle_utility", + "micro": { + "confidence": 0.95, + "high": 1.1602533682404708, + "low": 1.1602533682404708, + "mean": 1.2159999172436073, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selected_candidate_index": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 6.812499999999999, + "low": 6.812499999999999, + "mean": 6.477083333333334, + "n": 120, + "num_samples": 1 + }, + "mean": 6.477083333333334, + "per_seed": { + "1073161778/train0": 2.0, + "1073161778/train1": 12.0, + "1073161778/train2": 0.0, + "1113371261/train0": 2.0, + "1113371261/train1": 5.0, + "1113371261/train2": 2.0, + "1151410346/train0": 6.0, + "1151410346/train1": 2.0, + "1151410346/train2": 1.0, + "1261930456/train0": 0.0, + "1261930456/train1": 10.0, + "1261930456/train2": 2.0, + "1324133122/train0": 6.0, + "1324133122/train1": 5.0, + "1324133122/train2": 4.0, + "1324642411/train0": 7.0, + "1324642411/train1": 1.0, + "1324642411/train2": 5.0, + "1375492786/train0": 5.0, + "1375492786/train1": 9.0, + "1375492786/train2": 2.0, + "141984615/train0": 14.0, + "141984615/train1": 15.0, + "141984615/train2": 11.0, + "1461069713/train0": 15.0, + "1461069713/train1": 15.0, + "1461069713/train2": 3.0, + "1480856810/train0": 9.0, + "1480856810/train1": 8.0, + "1480856810/train2": 1.0, + "1509980664/train0": 10.0, + "1509980664/train1": 12.0, + "1509980664/train2": 12.0, + "1522272686/train0": 1.0, + "1522272686/train1": 0.0, + "1522272686/train2": 5.0, + "1549324805/train0": 5.0, + "1549324805/train1": 0.0, + "1549324805/train2": 1.0, + "1578990937/train0": 14.0, + "1578990937/train1": 7.0, + "1578990937/train2": 3.0, + "1747178030/train0": 9.0, + "1747178030/train1": 3.0, + "1747178030/train2": 11.0, + "1790188787/train0": 5.0, + "1790188787/train1": 6.0, + "1790188787/train2": 8.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 0.0, + "1845489013/train0": 5.0, + "1845489013/train1": 13.0, + "1845489013/train2": 9.0, + "1999022574/train0": 8.0, + "1999022574/train1": 12.0, + "1999022574/train2": 11.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 2.0, + "2038594191/train0": 5.0, + "2038594191/train1": 9.0, + "2038594191/train2": 0.0, + "2054899126/train0": 2.0, + "2054899126/train1": 7.5, + "2054899126/train2": 8.0, + "2061635888/train0": 1.0, + "2061635888/train1": 4.0, + "2061635888/train2": 2.0, + "211245842/train0": 9.0, + "211245842/train1": 12.0, + "211245842/train2": 12.0, + "278008524/train0": 7.0, + "278008524/train1": 8.0, + "278008524/train2": 10.0, + "307068879/train0": 9.0, + "307068879/train1": 1.0, + "307068879/train2": 2.0, + "412102563/train0": 10.0, + "412102563/train1": 12.0, + "412102563/train2": 12.0, + "440577711/train0": 4.0, + "440577711/train1": 10.0, + "440577711/train2": 13.0, + "441046872/train0": 7.0, + "441046872/train1": 9.0, + "441046872/train2": 6.0, + "509430392/train0": 11.0, + "509430392/train1": 2.0, + "509430392/train2": 10.0, + "585904130/train0": 4.0, + "585904130/train1": 15.0, + "585904130/train2": 1.0, + "608016290/train0": 12.0, + "608016290/train1": 7.0, + "608016290/train2": 6.0, + "626486813/train0": 11.0, + "626486813/train1": 9.0, + "626486813/train2": 3.0, + "636680137/train0": 2.0, + "636680137/train1": 1.0, + "636680137/train2": 9.0, + "655884605/train0": 0.0, + "655884605/train1": 6.0, + "655884605/train2": 1.0, + "672328262/train0": 7.0, + "672328262/train1": 4.0, + "672328262/train2": 6.0, + "701667486/train0": 5.0, + "701667486/train1": 6.0, + "701667486/train2": 4.0, + "747260437/train0": 15.0, + "747260437/train1": 13.0, + "747260437/train2": 14.0, + "750448296/train0": 6.0, + "750448296/train1": 10.666666666666666, + "750448296/train2": 8.333333333333334, + "767747424/train0": 4.75, + "767747424/train1": 8.25, + "767747424/train2": 8.75 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 5.227878787878788, + "low": 5.227878787878788, + "mean": 5.445656565656566, + "n": 5, + "num_samples": 1 + }, + "mean": 5.445656565656566, + "per_task": { + "LiftPegUpright-v1": 5.133333333333334, + "PickCube-v1": 2.0, + "PullCube-v1": 7.266666666666667, + "PushCube-v1": 6.222222222222222, + "StackCube-v1": 6.606060606060606 + } + }, + "metric": "selected_candidate_index", + "micro": { + "confidence": 0.95, + "high": 6.354166666666667, + "low": 6.354166666666667, + "mean": 6.652777777777778, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selected_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.28958333333333336, + "low": 0.28958333333333336, + "mean": 0.30833333333333335, + "n": 120, + "num_samples": 1 + }, + "mean": 0.30833333333333335, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 1.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.5, + "2054899126/train1": 0.5, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 1.0, + "672328262/train1": 1.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 0.0, + "747260437/train1": 1.0, + "747260437/train2": 0.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.5, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.15127272727272728, + "low": 0.15127272727272728, + "mean": 0.2157171717171717, + "n": 5, + "num_samples": 1 + }, + "mean": 0.2157171717171717, + "per_task": { + "LiftPegUpright-v1": 0.06666666666666667, + "PickCube-v1": 0.0, + "PullCube-v1": 0.32, + "PushCube-v1": 0.3888888888888889, + "StackCube-v1": 0.30303030303030304 + } + }, + "metric": "selected_success", + "micro": { + "confidence": 0.95, + "high": 0.3055555555555556, + "low": 0.3055555555555556, + "mean": 0.2916666666666667, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selected_success_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.022916666666666665, + "low": 0.022916666666666665, + "mean": 0.0020833333333333333, + "n": 120, + "num_samples": 1 + }, + "mean": 0.0020833333333333333, + "per_seed": { + "1073161778/train0": -1.0, + "1073161778/train1": -1.0, + "1073161778/train2": -1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": -1.0, + "1261930456/train1": -1.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": -1.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": -1.0, + "2038594191/train1": -1.0, + "2038594191/train2": -1.0, + "2054899126/train0": -0.5, + "2054899126/train1": -0.5, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": -1.0, + "747260437/train1": 0.0, + "747260437/train2": -1.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.25, + "767747424/train1": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": -0.05527272727272727, + "low": -0.05527272727272727, + "mean": 0.015838383838383836, + "n": 5, + "num_samples": 1 + }, + "mean": 0.015838383838383836, + "per_task": { + "LiftPegUpright-v1": -0.13333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": -0.04, + "PushCube-v1": 0.2222222222222222, + "StackCube-v1": 0.030303030303030304 + } + }, + "metric": "selected_success_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.027777777777777776, + "low": 0.027777777777777776, + "mean": 0.0, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selected_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.7686271689956712, + "low": 0.7686271689956712, + "mean": 0.7838827531079813, + "n": 120, + "num_samples": 1 + }, + "mean": 0.7838827531079813, + "per_seed": { + "1073161778/train0": 0.7067229747772217, + "1073161778/train1": 0.7057674527168274, + "1073161778/train2": 0.7130528092384338, + "1113371261/train0": 0.15842342376708984, + "1113371261/train1": 0.15842342376708984, + "1113371261/train2": 0.15842342376708984, + "1151410346/train0": 0.2260012924671173, + "1151410346/train1": 0.2260012924671173, + "1151410346/train2": 0.2260012924671173, + "1261930456/train0": 0.05754852294921875, + "1261930456/train1": 0.07495865225791931, + "1261930456/train2": 2.0, + "1324133122/train0": 0.09220491349697113, + "1324133122/train1": 0.09220491349697113, + "1324133122/train2": 0.09220491349697113, + "1324642411/train0": 0.054066285490989685, + "1324642411/train1": 0.10393516719341278, + "1324642411/train2": 0.10647052526473999, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.10283424705266953, + "141984615/train1": 0.10283424705266953, + "141984615/train2": 0.10283424705266953, + "1461069713/train0": 0.0939030647277832, + "1461069713/train1": 0.0939030647277832, + "1461069713/train2": 0.0939030647277832, + "1480856810/train0": 0.07608599960803986, + "1480856810/train1": 0.19651852548122406, + "1480856810/train2": 0.21615806221961975, + "1509980664/train0": 0.8334100246429443, + "1509980664/train1": 0.8334100246429443, + "1509980664/train2": 0.8334100246429443, + "1522272686/train0": 0.5839591026306152, + "1522272686/train1": 0.5839591026306152, + "1522272686/train2": 0.5839591026306152, + "1549324805/train0": 0.03462250903248787, + "1549324805/train1": 0.0303483996540308, + "1549324805/train2": 0.04690708592534065, + "1578990937/train0": 2.0, + "1578990937/train1": 0.13703244924545288, + "1578990937/train2": 0.19338026642799377, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 2.0, + "1790188787/train1": 0.13741976022720337, + "1790188787/train2": 0.13743820786476135, + "1822210471/train0": 0.06306678056716919, + "1822210471/train1": 2.0, + "1822210471/train2": 2.0, + "1845489013/train0": 2.0, + "1845489013/train1": 2.0, + "1845489013/train2": 2.0, + "1999022574/train0": 2.0, + "1999022574/train1": 2.0, + "1999022574/train2": 0.8183539509773254, + "2012211287/train0": 0.7106714248657227, + "2012211287/train1": 0.7191179394721985, + "2012211287/train2": 0.7165453433990479, + "2038594191/train0": 0.04285776615142822, + "2038594191/train1": 0.07674666494131088, + "2038594191/train2": 0.04917396977543831, + "2054899126/train0": 1.0391597636044025, + "2054899126/train1": 1.0559737384319305, + "2054899126/train2": 2.0, + "2061635888/train0": 0.05225050449371338, + "2061635888/train1": 0.05225050449371338, + "2061635888/train2": 0.05225050449371338, + "211245842/train0": 2.0, + "211245842/train1": 0.19218182563781738, + "211245842/train2": 0.19090841710567474, + "278008524/train0": 0.042734989896416664, + "278008524/train1": 0.042734989896416664, + "278008524/train2": 0.042734989896416664, + "307068879/train0": 0.6000905632972717, + "307068879/train1": 0.6000905632972717, + "307068879/train2": 0.6000905632972717, + "412102563/train0": 2.0, + "412102563/train1": 2.0, + "412102563/train2": 2.0, + "440577711/train0": 2.0, + "440577711/train1": 2.0, + "440577711/train2": 2.0, + "441046872/train0": 0.6999551653862, + "441046872/train1": 0.6999551653862, + "441046872/train2": 0.6999551653862, + "509430392/train0": 2.0, + "509430392/train1": 2.0, + "509430392/train2": 5.424022674560547e-06, + "585904130/train0": 0.09919407218694687, + "585904130/train1": 0.04792279005050659, + "585904130/train2": 2.0, + "608016290/train0": 2.0, + "608016290/train1": 0.7279855608940125, + "608016290/train2": 0.22811375558376312, + "626486813/train0": 0.007657766109332442, + "626486813/train1": 0.00509834272088483, + "626486813/train2": 0.00509834272088483, + "636680137/train0": 0.08826130628585815, + "636680137/train1": 0.017789265140891075, + "636680137/train2": 0.041415613144636154, + "655884605/train0": 0.12222731113433838, + "655884605/train1": 0.11176733672618866, + "655884605/train2": 0.09131735563278198, + "672328262/train0": 2.0, + "672328262/train1": 2.0, + "672328262/train2": 2.0, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 0.08987965434789658, + "747260437/train1": 2.0, + "747260437/train2": 0.0880771279335022, + "750448296/train0": 0.09184517463048299, + "750448296/train1": 0.10300639644265175, + "750448296/train2": 0.07752946888407071, + "767747424/train0": 1.0327778598293662, + "767747424/train1": 0.5613184677204117, + "767747424/train2": 0.5671488647349179 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7077141006093923, + "low": 0.7077141006093923, + "mean": 0.727790409455866, + "n": 5, + "num_samples": 1 + }, + "mean": 0.727790409455866, + "per_task": { + "LiftPegUpright-v1": 0.7418742964665095, + "PickCube-v1": 0.5839591026306152, + "PullCube-v1": 0.6846797956805676, + "PushCube-v1": 0.8422558406988779, + "StackCube-v1": 0.7861830118027601 + } + }, + "metric": "selected_utility", + "micro": { + "confidence": 0.95, + "high": 0.7456324993949642, + "low": 0.7456324993949642, + "mean": 0.7314973677291872, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selected_utility_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.02126554856578068, + "low": 0.02126554856578068, + "mean": -0.006235309189812849, + "n": 120, + "num_samples": 1 + }, + "mean": -0.006235309189812849, + "per_seed": { + "1073161778/train0": -1.2932770252227783, + "1073161778/train1": -1.2942325472831726, + "1073161778/train2": -1.2869471907615662, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": -1.9424514770507812, + "1261930456/train1": -1.9250413477420807, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": -0.052404239773750305, + "1324642411/train1": -0.0025353580713272095, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": -0.010948076844215393, + "1480856810/train1": 0.10948444902896881, + "1480856810/train2": 0.1291239857673645, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": -0.06989232823252678, + "1549324805/train1": -0.07416643761098385, + "1549324805/train2": -0.057607751339673996, + "1578990937/train0": 1.17586350440979, + "1578990937/train1": -0.6871040463447571, + "1578990937/train2": -0.6307562291622162, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.87523253262043, + "1790188787/train1": 0.012652292847633362, + "1790188787/train2": 0.012670740485191345, + "1822210471/train0": -0.008848525583744049, + "1822210471/train1": 1.9280846938490868, + "1822210471/train2": 1.9280846938490868, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": -1.1816460490226746, + "2012211287/train0": -0.012933909893035889, + "2012211287/train1": -0.004487395286560059, + "2012211287/train2": -0.007059991359710693, + "2038594191/train0": -1.9571422338485718, + "2038594191/train1": -1.9232533350586891, + "2038594191/train2": -1.9508260302245617, + "2054899126/train0": -0.9608402363955975, + "2054899126/train1": -0.9440262615680695, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.8090915828943253, + "211245842/train1": 0.0012734085321426392, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 1.9638472199440002, + "440577711/train1": 1.9638472199440002, + "440577711/train2": 1.9638472199440002, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.999999776482582, + "509430392/train1": 1.999999776482582, + "509430392/train2": 5.200505256652832e-06, + "585904130/train0": -0.0825214758515358, + "585904130/train1": -0.13379275798797607, + "585904130/train2": 1.8182844519615173, + "608016290/train0": 1.274298369884491, + "608016290/train1": 0.002283930778503418, + "608016290/train2": -0.4975878745317459, + "626486813/train0": 0.0025594233884476125, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.005126260221004486, + "636680137/train1": -0.0653457809239626, + "636680137/train2": -0.041719432920217514, + "655884605/train0": -0.040779367089271545, + "655884605/train1": -0.051239341497421265, + "655884605/train2": -0.07168932259082794, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": -1.9101203456521034, + "747260437/train1": 0.0, + "747260437/train2": -1.9119228720664978, + "750448296/train0": -0.02574797471364339, + "750448296/train1": -0.014586752901474634, + "750448296/train2": -0.04006368046005567, + "767747424/train0": 0.4595785178244114, + "767747424/train1": -0.011880874284543097, + "767747424/train2": -0.006050477270036936 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": -0.09933896855126614, + "low": -0.09933896855126614, + "mean": 0.03223061228185102, + "n": 5, + "num_samples": 1 + }, + "mean": 0.03223061228185102, + "per_task": { + "LiftPegUpright-v1": -0.20799624224503835, + "PickCube-v1": 0.0, + "PullCube-v1": -0.09049066554289312, + "PushCube-v1": 0.4498516619205475, + "StackCube-v1": 0.00978830727663907 + } + }, + "metric": "selected_utility_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.05096899993077386, + "low": 0.05096899993077386, + "mean": -0.0103222187131501, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "selector_regret": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5961417103811982, + "low": 0.5961417103811982, + "mean": 0.5049184655440816, + "n": 120, + "num_samples": 1 + }, + "mean": 0.5049184655440816, + "per_seed": { + "1073161778/train0": 1.2932770252227783, + "1073161778/train1": 1.2942325472831726, + "1073161778/train2": 1.2869471907615662, + "1113371261/train0": 0.0, + "1113371261/train1": 0.02250880002975464, + "1113371261/train2": 0.036848247051239014, + "1151410346/train0": 0.6403508484363556, + "1151410346/train1": 1.7739987075328827, + "1151410346/train2": 0.32620343565940857, + "1261930456/train0": 1.9424514770507812, + "1261930456/train1": 1.9250413477420807, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.04025374352931976, + "1324133122/train2": 0.10070674121379852, + "1324642411/train0": 0.08999375998973846, + "1324642411/train1": 1.8960648328065872, + "1324642411/train2": 1.89352947473526, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.006073415279388428, + "1480856810/train0": 1.9239140003919601, + "1480856810/train1": 1.803481474518776, + "1480856810/train2": 1.7838419377803802, + "1509980664/train0": 1.1665899753570557, + "1509980664/train1": 1.1665899753570557, + "1509980664/train2": 1.1665899753570557, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.026248913258314133, + "1549324805/train1": 0.16857311688363552, + "1549324805/train2": 0.06619475409388542, + "1578990937/train0": 0.0, + "1578990937/train1": 1.8629675507545471, + "1578990937/train2": 1.8066197335720062, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.8625802397727966, + "1790188787/train2": 1.8625617921352386, + "1822210471/train0": 1.9369332194328308, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 1.1816460490226746, + "2012211287/train0": 1.2893285751342773, + "2012211287/train1": 1.2808820605278015, + "2012211287/train2": 1.2834546566009521, + "2038594191/train0": 1.9571422338485718, + "2038594191/train1": 1.9232533350586891, + "2038594191/train2": 1.9508260302245617, + "2054899126/train0": 0.03812338784337044, + "2054899126/train1": 0.9440262615680695, + "2054899126/train2": 0.0, + "2061635888/train0": 0.04574880003929138, + "2061635888/train1": 0.02355286478996277, + "2061635888/train2": 0.05880081653594971, + "211245842/train0": 0.0, + "211245842/train1": 0.02077534794807434, + "211245842/train2": 0.04297435283660889, + "278008524/train0": 1.0306213330477476, + "278008524/train1": 0.9949396047741175, + "278008524/train2": 1.0932139214128256, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 1.3000448346138, + "441046872/train1": 0.023403704166412354, + "441046872/train2": 0.019179701805114746, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 1.9008059278130531, + "585904130/train1": 1.9520772099494934, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 1.2720144391059875, + "608016290/train2": 0.48361386358737946, + "626486813/train0": 0.037037709495052695, + "626486813/train1": 0.030729531950782984, + "626486813/train2": 0.03980091231642291, + "636680137/train0": 0.0, + "636680137/train1": 0.04962191917002201, + "636680137/train2": 0.034236010164022446, + "655884605/train0": 0.015101373195648193, + "655884605/train1": 0.7011084705591202, + "655884605/train2": 0.12315595149993896, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.01851598173379898, + "747260437/train1": 0.0, + "747260437/train2": 0.03702475130558014, + "750448296/train0": 0.6608151532709599, + "750448296/train1": 1.267496877660354, + "750448296/train2": 1.2729191991190116, + "767747424/train0": 0.018381604924798012, + "767747424/train1": 0.5016346235061064, + "767747424/train2": 0.5000182301737368 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4543977503562628, + "low": 0.4543977503562628, + "mean": 0.49875115056017966, + "n": 5, + "num_samples": 1 + }, + "mean": 0.49875115056017966, + "per_task": { + "LiftPegUpright-v1": 0.7217585732539494, + "PickCube-v1": 0.0, + "PullCube-v1": 0.45504458223469557, + "PushCube-v1": 0.9435255742735333, + "StackCube-v1": 0.37342702303871966 + } + }, + "metric": "selector_regret", + "micro": { + "confidence": 0.95, + "high": 0.45732345881858943, + "low": 0.45732345881858943, + "mean": 0.5157032941917552, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "success_selector_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.35347222222222213, + "low": 0.35347222222222213, + "mean": 0.2847222222222222, + "n": 120, + "num_samples": 1 + }, + "mean": 0.2847222222222222, + "per_seed": { + "1073161778/train0": 1.0, + "1073161778/train1": 1.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 1.0, + "1151410346/train2": 0.0, + "1261930456/train0": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 1.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 0.0, + "2054899126/train1": 0.5, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.5, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 1.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 1.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.0, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.3010909090909091, + "low": 0.3010909090909091, + "mean": 0.2944242424242424, + "n": 5, + "num_samples": 1 + }, + "mean": 0.2944242424242424, + "per_task": { + "LiftPegUpright-v1": 0.5333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": 0.22666666666666666, + "PushCube-v1": 0.5, + "StackCube-v1": 0.21212121212121213 + } + }, + "metric": "success_selector_gap", + "micro": { + "confidence": 0.95, + "high": 0.25, + "low": 0.25, + "mean": 0.2847222222222222, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "success_support_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.18472222222222223, + "low": 0.18472222222222223, + "mean": 0.23194444444444443, + "n": 120, + "num_samples": 1 + }, + "mean": 0.23194444444444443, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 1.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 1.0, + "1151410346/train1": 0.0, + "1151410346/train2": 1.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 1.0, + "1324133122/train1": 1.0, + "1324133122/train2": 1.0, + "1324642411/train0": 1.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 1.0, + "1522272686/train1": 1.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.5, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 0.0, + "211245842/train1": 1.0, + "211245842/train2": 1.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 1.0, + "307068879/train1": 1.0, + "307068879/train2": 1.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 0.0, + "441046872/train1": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 1.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 1.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.0, + "767747424/train1": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4623030303030303, + "low": 0.4623030303030303, + "mean": 0.40452525252525257, + "n": 5, + "num_samples": 1 + }, + "mean": 0.40452525252525257, + "per_task": { + "LiftPegUpright-v1": 0.4, + "PickCube-v1": 1.0, + "PullCube-v1": 0.02666666666666667, + "PushCube-v1": 0.1111111111111111, + "StackCube-v1": 0.48484848484848486 + } + }, + "metric": "success_support_gap", + "micro": { + "confidence": 0.95, + "high": 0.19444444444444445, + "low": 0.19444444444444445, + "mean": 0.2013888888888889, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "support_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.3546240135134819, + "low": 0.3546240135134819, + "mean": 0.4232947583840643, + "n": 120, + "num_samples": 1 + }, + "mean": 0.4232947583840643, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 1.897251382470131, + "1113371261/train1": 1.8190677762031555, + "1113371261/train2": 1.8047283291816711, + "1151410346/train0": 1.133647859096527, + "1151410346/train1": 0.0, + "1151410346/train2": 1.4477952718734741, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 1.9450107961893082, + "1324133122/train1": 1.867541342973709, + "1324133122/train2": 1.8070883452892303, + "1324642411/train0": 1.8559399545192719, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.270599365234375, + "141984615/train1": 0.24876614287495613, + "141984615/train2": 0.2268255539238453, + "1461069713/train0": 0.0991440862417221, + "1461069713/train1": 0.14275743998587132, + "1461069713/train2": 0.05936597287654877, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 1.8167466223239899, + "1522272686/train1": 1.9298561662435532, + "1522272686/train2": 1.8506269305944443, + "1549324805/train0": 0.23658466339111328, + "1549324805/train1": 0.09853456914424896, + "1549324805/train2": 0.1843542456626892, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.922716848552227, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 1.9020006954669952, + "2061635888/train1": 1.9241966307163239, + "2061635888/train2": 1.888948678970337, + "211245842/train0": 0.0, + "211245842/train1": 1.7870428264141083, + "211245842/train2": 1.7661172300577164, + "278008524/train0": 0.04055231809616089, + "278008524/train1": 0.07623404636979103, + "278008524/train2": 0.0, + "307068879/train0": 1.7884424477815628, + "307068879/train1": 1.4123926758766174, + "307068879/train2": 1.7697865515947342, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 0.0, + "441046872/train1": 1.2766411304473877, + "441046872/train2": 1.2808651328086853, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 1.9999945759773254, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 1.2882723808288574, + "626486813/train0": 0.12161866808310151, + "626486813/train1": 0.12975450651720166, + "626486813/train2": 0.12068312615156174, + "636680137/train0": 0.0823030173778534, + "636680137/train1": 0.10315313935279846, + "636680137/train2": 0.09491270035505295, + "655884605/train0": 1.8626713156700134, + "655884605/train1": 1.1871241927146912, + "655884605/train2": 1.785526692867279, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.17026761174201965, + "747260437/train1": 0.21537372469902039, + "747260437/train2": 0.15356136858463287, + "750448296/train0": 0.6553361676633358, + "750448296/train1": 0.03749322146177292, + "750448296/train2": 0.05754782756169637, + "767747424/train0": 0.06562284985557199, + "767747424/train1": 0.045098946429789066, + "767747424/train2": 0.04088494274765253 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7984061056091026, + "low": 0.7984061056091026, + "mean": 0.7236978183567826, + "n": 5, + "num_samples": 1 + }, + "mean": 0.7236978183567826, + "per_task": { + "LiftPegUpright-v1": 0.5877600212891897, + "PickCube-v1": 1.8657432397206624, + "PullCube-v1": 0.1075624164690574, + "PushCube-v1": 0.21421858502758873, + "StackCube-v1": 0.8432048292774142 + } + }, + "metric": "support_gap", + "micro": { + "confidence": 0.95, + "high": 0.37098560981555945, + "low": 0.37098560981555945, + "mean": 0.3761285081272945, + "n": 144, + "num_samples": 1 + }, + "n": 144 + }, + "tau": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.9529843839369022, + "low": 0.9529843839369022, + "mean": 0.9577490006043817, + "n": 120, + "num_samples": 1 + }, + "mean": 0.9577490006043817, + "per_seed": { + "1073161778/train0": 0.7478358458742458, + "1073161778/train1": 0.7478358458742458, + "1073161778/train2": 0.7478358458742458, + "1113371261/train0": 0.893807936331132, + "1113371261/train1": 0.893807936331132, + "1113371261/train2": 0.893807936331132, + "1151410346/train0": 0.893807936331132, + "1151410346/train1": 0.893807936331132, + "1151410346/train2": 0.893807936331132, + "1261930456/train0": 1.2264360294338528, + "1261930456/train1": 1.2264360294338528, + "1261930456/train2": 1.2264360294338528, + "1324133122/train0": 0.893807936331132, + "1324133122/train1": 0.893807936331132, + "1324133122/train2": 0.893807936331132, + "1324642411/train0": 1.2264360294338528, + "1324642411/train1": 1.2264360294338528, + "1324642411/train2": 1.2264360294338528, + "1375492786/train0": 0.9470671281950391, + "1375492786/train1": 0.9470671281950391, + "1375492786/train2": 0.9470671281950391, + "141984615/train0": 0.9470671281950391, + "141984615/train1": 0.9470671281950391, + "141984615/train2": 0.9470671281950391, + "1461069713/train0": 0.9470671281950391, + "1461069713/train1": 0.9470671281950391, + "1461069713/train2": 0.9470671281950391, + "1480856810/train0": 1.2264360294338528, + "1480856810/train1": 1.2264360294338528, + "1480856810/train2": 1.2264360294338528, + "1509980664/train0": 0.893807936331132, + "1509980664/train1": 0.893807936331132, + "1509980664/train2": 0.893807936331132, + "1522272686/train0": 1.280136139242778, + "1522272686/train1": 1.280136139242778, + "1522272686/train2": 1.280136139242778, + "1549324805/train0": 0.9470671281950391, + "1549324805/train1": 0.9470671281950391, + "1549324805/train2": 0.9470671281950391, + "1578990937/train0": 0.893807936331132, + "1578990937/train1": 0.893807936331132, + "1578990937/train2": 0.893807936331132, + "1747178030/train0": 0.9470671281950391, + "1747178030/train1": 0.9470671281950391, + "1747178030/train2": 0.9470671281950391, + "1790188787/train0": 1.2264360294338528, + "1790188787/train1": 1.2264360294338528, + "1790188787/train2": 1.2264360294338528, + "1822210471/train0": 0.9470671281950391, + "1822210471/train1": 0.9470671281950391, + "1822210471/train2": 0.9470671281950391, + "1845489013/train0": 0.9470671281950391, + "1845489013/train1": 0.9470671281950391, + "1845489013/train2": 0.9470671281950391, + "1999022574/train0": 0.893807936331132, + "1999022574/train1": 0.893807936331132, + "1999022574/train2": 0.893807936331132, + "2012211287/train0": 0.7478358458742458, + "2012211287/train1": 0.7478358458742458, + "2012211287/train2": 0.7478358458742458, + "2038594191/train0": 0.9470671281950391, + "2038594191/train1": 0.9470671281950391, + "2038594191/train2": 0.9470671281950391, + "2054899126/train0": 0.9470671281950391, + "2054899126/train1": 0.9470671281950391, + "2054899126/train2": 0.9470671281950391, + "2061635888/train0": 0.893807936331132, + "2061635888/train1": 0.893807936331132, + "2061635888/train2": 0.893807936331132, + "211245842/train0": 0.893807936331132, + "211245842/train1": 0.893807936331132, + "211245842/train2": 0.893807936331132, + "278008524/train0": 0.9470671281950391, + "278008524/train1": 0.9470671281950391, + "278008524/train2": 0.9470671281950391, + "307068879/train0": 0.7478358458742458, + "307068879/train1": 0.7478358458742458, + "307068879/train2": 0.7478358458742458, + "412102563/train0": 0.893807936331132, + "412102563/train1": 0.893807936331132, + "412102563/train2": 0.893807936331132, + "440577711/train0": 1.2264360294338528, + "440577711/train1": 1.2264360294338528, + "440577711/train2": 1.2264360294338528, + "441046872/train0": 0.7478358458742458, + "441046872/train1": 0.7478358458742458, + "441046872/train2": 0.7478358458742458, + "509430392/train0": 1.2264360294338528, + "509430392/train1": 1.2264360294338528, + "509430392/train2": 1.2264360294338528, + "585904130/train0": 0.9470671281950391, + "585904130/train1": 0.9470671281950391, + "585904130/train2": 0.9470671281950391, + "608016290/train0": 0.7478358458742458, + "608016290/train1": 0.7478358458742458, + "608016290/train2": 0.7478358458742458, + "626486813/train0": 0.9470671281950391, + "626486813/train1": 0.9470671281950391, + "626486813/train2": 0.9470671281950391, + "636680137/train0": 0.9470671281950391, + "636680137/train1": 0.9470671281950391, + "636680137/train2": 0.9470671281950391, + "655884605/train0": 0.893807936331132, + "655884605/train1": 0.893807936331132, + "655884605/train2": 0.893807936331132, + "672328262/train0": 0.9470671281950391, + "672328262/train1": 0.9470671281950391, + "672328262/train2": 0.9470671281950391, + "701667486/train0": 0.893807936331132, + "701667486/train1": 0.893807936331132, + "701667486/train2": 0.893807936331132, + "747260437/train0": 0.9470671281950391, + "747260437/train1": 0.9470671281950391, + "747260437/train2": 0.9470671281950391, + "750448296/train0": 0.9470671281950391, + "750448296/train1": 0.9470671281950391, + "750448296/train2": 0.9470671281950391, + "767747424/train0": 0.9470671281950391, + "767747424/train1": 0.9470671281950391, + "767747424/train2": 0.9470671281950391 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.923336579103488, + "low": 0.923336579103488, + "mean": 1.0190566158154093, + "n": 5, + "num_samples": 1 + }, + "mean": 1.0190566158154093, + "per_task": { + "LiftPegUpright-v1": 0.7478358458742457, + "PickCube-v1": 1.280136139242778, + "PullCube-v1": 0.9470671281950384, + "PushCube-v1": 1.2264360294338525, + "StackCube-v1": 0.8938079363311323 + } + }, + "metric": "tau", + "micro": { + "confidence": 0.95, + "high": 0.9561518882917827, + "low": 0.9561518882917827, + "mean": 0.9559686885361595, + "n": 144, + "num_samples": 1 + }, + "n": 144 + } + }, + "target": "success_weighted_margin", + "tau": { + "LiftPegUpright-v1": 0.7478358458742458, + "PickCube-v1": 1.280136139242778, + "PullCube-v1": 0.9470671281950391, + "PushCube-v1": 1.2264360294338528, + "StackCube-v1": 0.893807936331132, + "__global__": 0.9063413508752075 + }, + "threshold_scope": "task", + "weights": [ + 0.48450353349747766, + 0.1202532444167339, + 0.155682061283909, + -0.02441177660309185, + 0.024380137939416402, + -0.003640887143329219, + -0.022064299856128927, + -0.5289186063935553, + 0.2831664990097297, + 0.1697769359307344, + -1.4311502652615907e-12, + 0.5632558633953502, + 0.01891643765131379, + -3.812772905557871e-12, + -0.4745788255595529, + 0.0223358572828829, + 0.17886238148821576, + -0.07857526126457515, + 0.034629655768952716, + -0.3694378263109286, + -0.03692550319544807, + 0.30919666931845313, + 0.12101367094479087, + -3.367420661491832e-14, + 7.831214649569506e-13, + 0.17614436781573706, + 0.009569206047353118, + 0.05351390524412675, + -3.200995637988036e-13, + -2.3096275173741674, + -2.0562522677684, + 2.7748361386071565, + 1.0887750665501699, + -0.2504127200851598, + -1.8636001512197575, + -0.18745991293015152, + 2.3526103961524427, + 1.9162466598226047, + -2.2408820301440073, + 1.0575725982390622, + -1.252532793179583e-13, + 0.5344620971340412, + 0.7544142449206772, + -0.13811908568580122, + -0.21306501765016073, + -1.0967599879902776e-13, + -1.401784574525504e-13, + 3.329992400197128e-14, + 6.0642218037049375e-15, + -1.1068642358440333e-13, + 0.2255812123126617, + -0.27427416007010486, + -0.6751629064421536, + 0.5451814470601999, + 0.5737151916015826, + -0.27708589461000876, + 0.05294407275731161, + 0.06763997647593922, + 0.011213033280449689, + 0.19071598573872706, + -0.30965862623116536, + -0.4429815477074821, + 0.6911700190093208, + -0.5590717051247991, + 0.043405984945490195, + -1.0952605455157922e-14, + -0.06191438693708173, + -0.230246623624919, + 0.0225452491316821 + ] +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/metrics_by_seed.json b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/metrics_by_seed.json new file mode 100644 index 0000000000000000000000000000000000000000..4f09f13dd35d5f02d11f54c4b0c2c1ba206b6ef1 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/metrics_by_seed.json @@ -0,0 +1,3236 @@ +{ + "1073161778/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5516482746314322, + "pairwise_causal_calibration_ece": 0.1125523173289727, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.197390750232301, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.7067229747772217, + "selected_utility_gain_over_base": -1.2932770252227783, + "selector_regret": 1.2932770252227783, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.7478358458742458 + }, + "1073161778/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.39166666666666666, + "pairwise_causal_calibration_confidence": 0.5665909321275702, + "pairwise_causal_calibration_ece": 0.17492426546090348, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.057082412822946, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.7057674527168274, + "selected_utility_gain_over_base": -1.2942325472831726, + "selector_regret": 1.2942325472831726, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.7478358458742458 + }, + "1073161778/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7619047619047619, + "pairwise_causal_calibration_confidence": 0.5700167609515335, + "pairwise_causal_calibration_ece": 0.19188800095322855, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.0317041974020489, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.7130528092384338, + "selected_utility_gain_over_base": -1.2869471907615662, + "selector_regret": 1.2869471907615662, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.7478358458742458 + }, + "1113371261/train0": { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5411387050524964, + "pairwise_causal_calibration_ece": 0.025500013084200863, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.008407185293451747, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10274861752986908, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.897251382470131, + "tau": 0.893807936331132 + }, + "1113371261/train1": { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.53349783320347, + "pairwise_causal_calibration_ece": 0.10492640463204145, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.2579825916955958, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18093222379684448, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02250880002975464, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8190677762031555, + "tau": 0.893807936331132 + }, + "1113371261/train2": { + "base_success": 0.0, + "base_utility": 0.15842342376708984, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5791187642133746, + "pairwise_causal_calibration_ece": 0.20659552150091112, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.213636373075774, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "tau": 0.893807936331132 + }, + "1151410346/train0": { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5619047619047619, + "pairwise_causal_calibration_confidence": 0.5447959029344634, + "pairwise_causal_calibration_ece": 0.05152312612164488, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.8492504414490131, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8663521409034729, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.6403508484363556, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.133647859096527, + "tau": 0.893807936331132 + }, + "1151410346/train1": { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5428571428571428, + "pairwise_causal_calibration_confidence": 0.5928488173780985, + "pairwise_causal_calibration_ece": 0.05599809010651253, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.8553272798880048, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.7739987075328827, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "1151410346/train2": { + "base_success": 0.0, + "base_utility": 0.2260012924671173, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5523809523809524, + "pairwise_causal_calibration_confidence": 0.5611842303245104, + "pairwise_causal_calibration_ece": 0.03328359958616477, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.8355344562806973, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.32620343565940857, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4477952718734741, + "tau": 0.893807936331132 + }, + "1261930456/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.494949494949495, + "pairwise_causal_calibration_confidence": 0.6511354281579875, + "pairwise_causal_calibration_ece": 0.3204647685440312, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 1.471050707973179, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.05754852294921875, + "selected_utility_gain_over_base": -1.9424514770507812, + "selector_regret": 1.9424514770507812, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1261930456/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5636363636363636, + "pairwise_causal_calibration_confidence": 0.6560658348856379, + "pairwise_causal_calibration_ece": 0.305306037200651, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.3365043159479286, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07495865225791931, + "selected_utility_gain_over_base": -1.9250413477420807, + "selector_regret": 1.9250413477420807, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1261930456/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5952380952380952, + "pairwise_causal_calibration_confidence": 0.6424422270814019, + "pairwise_causal_calibration_ece": 0.130927476896239, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.168316549791102, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1324133122/train0": { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5778483044050591, + "pairwise_causal_calibration_ece": 0.11356259011934491, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41376711119187704, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.054989203810691833, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9450107961893082, + "tau": 0.893807936331132 + }, + "1324133122/train1": { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5835206510970223, + "pairwise_causal_calibration_ece": 0.11178165495215062, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.4382011802480461, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1324586570262909, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04025374352931976, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.867541342973709, + "tau": 0.893807936331132 + }, + "1324133122/train2": { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5825263036735857, + "pairwise_causal_calibration_ece": 0.20318798204070004, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.24745057375120538, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19291165471076965, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.10070674121379852, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8070883452892303, + "tau": 0.893807936331132 + }, + "1324642411/train0": { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5583333333333333, + "pairwise_causal_calibration_confidence": 0.5894684248151024, + "pairwise_causal_calibration_ece": 0.03846474579568359, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 2.1833463985406447, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14406004548072815, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.054066285490989685, + "selected_utility_gain_over_base": -0.052404239773750305, + "selector_regret": 0.08999375998973846, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8559399545192719, + "tau": 1.2264360294338528 + }, + "1324642411/train1": { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42016806722689076, + "pairwise_causal_calibration_confidence": 0.6037811380752174, + "pairwise_causal_calibration_ece": 0.18361307084832634, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.7807138675039442, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10393516719341278, + "selected_utility_gain_over_base": -0.0025353580713272095, + "selector_regret": 1.8960648328065872, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1324642411/train2": { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5454545454545454, + "pairwise_causal_calibration_confidence": 0.593444506997112, + "pairwise_causal_calibration_ece": 0.14259508480466812, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.0767725563667239, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10647052526473999, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.89352947473526, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1375492786/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4533333333333333, + "pairwise_causal_calibration_confidence": 0.5366248224651405, + "pairwise_causal_calibration_ece": 0.08329148913180733, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.3369485573884625, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1375492786/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5263157894736842, + "pairwise_causal_calibration_confidence": 0.5675437560777116, + "pairwise_causal_calibration_ece": 0.04122796660402744, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.11389548031987307, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1375492786/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5607087307240273, + "pairwise_causal_calibration_ece": 0.029809709781793304, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.07382412594656773, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "141984615/train0": { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5416666666666666, + "pairwise_causal_calibration_confidence": 0.5684739620257733, + "pairwise_causal_calibration_ece": 0.0268072953591066, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6161038657373066, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "selected_candidate_index": 14.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10283424705266953, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "tau": 0.9470671281950391 + }, + "141984615/train1": { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.5676327735960262, + "pairwise_causal_calibration_ece": 0.0509661069293595, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7260454977260915, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.037011127918958664, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10283424705266953, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24876614287495613, + "tau": 0.9470671281950391 + }, + "141984615/train2": { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.525, + "pairwise_causal_calibration_confidence": 0.5698959198273771, + "pairwise_causal_calibration_ece": 0.06285173835176751, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5103624253528025, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058951716870069504, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10283424705266953, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "tau": 0.9470671281950391 + }, + "1461069713/train0": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.275, + "pairwise_causal_calibration_confidence": 0.5596745083517122, + "pairwise_causal_calibration_ece": 0.2846745083517123, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5914274057916176, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06019836664199829, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0939030647277832, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0991440862417221, + "tau": 0.9470671281950391 + }, + "1461069713/train1": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.375, + "pairwise_causal_calibration_confidence": 0.5561091294148327, + "pairwise_causal_calibration_ece": 0.1811091294148327, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6317591128320483, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0939030647277832, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14275743998587132, + "tau": 0.9470671281950391 + }, + "1461069713/train2": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.575, + "pairwise_causal_calibration_confidence": 0.5593899568017219, + "pairwise_causal_calibration_ece": 0.043120160988310785, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5260931359239889, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0939030647277832, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.006073415279388428, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05936597287654877, + "tau": 0.9470671281950391 + }, + "1480856810/train0": { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.26262626262626265, + "pairwise_causal_calibration_confidence": 0.5720642439373731, + "pairwise_causal_calibration_ece": 0.30943798131111067, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 2.1158648725170077, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07608599960803986, + "selected_utility_gain_over_base": -0.010948076844215393, + "selector_regret": 1.9239140003919601, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1480856810/train1": { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4957983193277311, + "pairwise_causal_calibration_confidence": 0.5496493820116296, + "pairwise_causal_calibration_ece": 0.05385106268389854, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.8867636980590103, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19651852548122406, + "selected_utility_gain_over_base": 0.10948444902896881, + "selector_regret": 1.803481474518776, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1480856810/train2": { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6050420168067226, + "pairwise_causal_calibration_confidence": 0.5634795931375042, + "pairwise_causal_calibration_ece": 0.06596647647042775, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.8428291167331592, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21615806221961975, + "selected_utility_gain_over_base": 0.1291239857673645, + "selector_regret": 1.7838419377803802, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1509980664/train0": { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4888888888888889, + "pairwise_causal_calibration_confidence": 0.5462356837170409, + "pairwise_causal_calibration_ece": 0.057346794828152005, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.5816690584072294, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "1509980664/train1": { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.1746031746031746, + "pairwise_causal_calibration_confidence": 0.5479465839142913, + "pairwise_causal_calibration_ece": 0.37334340931111676, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": 0.5626192514371053, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "1509980664/train2": { + "base_success": 0.0, + "base_utility": 0.8334100246429443, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.5594453160124555, + "pairwise_causal_calibration_ece": 0.06722278178918951, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.6223762395341456, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "1522272686/train0": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.19444444444444445, + "pairwise_causal_calibration_confidence": 0.6053552926240098, + "pairwise_causal_calibration_ece": 0.4109108481795651, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.7902719177719542, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18325337767601013, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8167466223239899, + "tau": 1.280136139242778 + }, + "1522272686/train1": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3055555555555556, + "pairwise_causal_calibration_confidence": 0.6354932924180177, + "pairwise_causal_calibration_ece": 0.32993773686246214, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.20562222929953555, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07014383375644684, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9298561662435532, + "tau": 1.280136139242778 + }, + "1522272686/train2": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5663336315117011, + "pairwise_causal_calibration_ece": 0.08285581154113539, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.21722683006216226, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14937306940555573, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8506269305944443, + "tau": 1.280136139242778 + }, + "1549324805/train0": { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.75, + "pairwise_causal_calibration_confidence": 0.5625148880216412, + "pairwise_causal_calibration_ece": 0.18748511197835893, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.530739092243051, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.060871422290802, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.03462250903248787, + "selected_utility_gain_over_base": -0.06989232823252678, + "selector_regret": 0.026248913258314133, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.23658466339111328, + "tau": 0.9470671281950391 + }, + "1549324805/train1": { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5833333333333334, + "pairwise_causal_calibration_confidence": 0.5360665332231191, + "pairwise_causal_calibration_ece": 0.07111703737429365, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.2279357729216895, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0303483996540308, + "selected_utility_gain_over_base": -0.07416643761098385, + "selector_regret": 0.16857311688363552, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "tau": 0.9470671281950391 + }, + "1549324805/train2": { + "base_success": 0.0, + "base_utility": 0.10451483726501465, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5498885743535615, + "pairwise_causal_calibration_ece": 0.10670850028777285, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.8289911704354562, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11310184001922607, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04690708592534065, + "selected_utility_gain_over_base": -0.057607751339673996, + "selector_regret": 0.06619475409388542, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1843542456626892, + "tau": 0.9470671281950391 + }, + "1578990937/train0": { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4358974358974359, + "pairwise_causal_calibration_confidence": 0.6103413950316227, + "pairwise_causal_calibration_ece": 0.17444395913418703, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 1.1959458964053806, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 14.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.17586350440979, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "1578990937/train1": { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42727272727272725, + "pairwise_causal_calibration_confidence": 0.6263703949729311, + "pairwise_causal_calibration_ece": 0.1990976677002039, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.2514512085861837, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13703244924545288, + "selected_utility_gain_over_base": -0.6871040463447571, + "selector_regret": 1.8629675507545471, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "1578990937/train2": { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7226890756302521, + "pairwise_causal_calibration_confidence": 0.6132925793293023, + "pairwise_causal_calibration_ece": 0.10939649630095027, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.0745551734071643, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19338026642799377, + "selected_utility_gain_over_base": -0.6307562291622162, + "selector_regret": 1.8066197335720062, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "1747178030/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 1.512536567520637, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1747178030/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 1.639637290759596, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 3.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1747178030/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 1.8087128419207654, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 11.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1790188787/train0": { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6272727272727273, + "pairwise_causal_calibration_confidence": 0.5867290403888722, + "pairwise_causal_calibration_ece": 0.04054368688385518, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.9875622503266075, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.87523253262043, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1790188787/train1": { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5263157894736842, + "pairwise_causal_calibration_confidence": 0.5937233509400742, + "pairwise_causal_calibration_ece": 0.10426662637283049, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 1.9148838772749794, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13741976022720337, + "selected_utility_gain_over_base": 0.012652292847633362, + "selector_regret": 1.8625802397727966, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1790188787/train2": { + "base_success": 0.0, + "base_utility": 0.12476746737957001, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.6001715319238825, + "pairwise_causal_calibration_ece": 0.16451829055105932, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.8332407953089316, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13743820786476135, + "selected_utility_gain_over_base": 0.012670740485191345, + "selector_regret": 1.8625617921352386, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "1822210471/train0": { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.625, + "pairwise_causal_calibration_confidence": 0.5988235112981328, + "pairwise_causal_calibration_ece": 0.11653574747746832, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 1.3378277706579857, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06306678056716919, + "selected_utility_gain_over_base": -0.008848525583744049, + "selector_regret": 1.9369332194328308, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1822210471/train1": { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.6072514522495137, + "pairwise_causal_calibration_ece": 0.1057225969843389, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": 1.2817760714600654, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9280846938490868, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1822210471/train2": { + "base_success": 0.0, + "base_utility": 0.07191530615091324, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.325, + "pairwise_causal_calibration_confidence": 0.5927326895223773, + "pairwise_causal_calibration_ece": 0.2677326895223773, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 1.1202233214608746, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9280846938490868, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1845489013/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5652173913043478, + "pairwise_causal_calibration_confidence": 0.5667479107448529, + "pairwise_causal_calibration_ece": 0.025257247762570406, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": 1.340214222459458, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1845489013/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.46153846153846156, + "pairwise_causal_calibration_confidence": 0.5409684680296629, + "pairwise_causal_calibration_ece": 0.11303375788512275, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": 1.2941213610673934, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1845489013/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.64, + "pairwise_causal_calibration_confidence": 0.5658799464871863, + "pairwise_causal_calibration_ece": 0.07971403378879252, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": 1.3003248867416755, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "1999022574/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6714285714285714, + "pairwise_causal_calibration_confidence": 0.6012777562735464, + "pairwise_causal_calibration_ece": 0.07015081515502501, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 1.3600458578396741, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 8.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "1999022574/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7428571428571429, + "pairwise_causal_calibration_confidence": 0.6153268689343157, + "pairwise_causal_calibration_ece": 0.1374436377245066, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 1.363778237267606, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "1999022574/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.25, + "pairwise_causal_calibration_confidence": 0.6423780754545177, + "pairwise_causal_calibration_ece": 0.3923780754545179, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": 1.4709100971517652, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.8183539509773254, + "selected_utility_gain_over_base": -1.1816460490226746, + "selector_regret": 1.1816460490226746, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "2012211287/train0": { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5042016806722689, + "pairwise_causal_calibration_confidence": 0.5508434661236096, + "pairwise_causal_calibration_ece": 0.08899904912532385, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.2153078550922283, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7106714248657227, + "selected_utility_gain_over_base": -0.012933909893035889, + "selector_regret": 1.2893285751342773, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.7478358458742458 + }, + "2012211287/train1": { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.475, + "pairwise_causal_calibration_confidence": 0.5273415392712859, + "pairwise_causal_calibration_ece": 0.05234153927128593, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7823168935976765, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7191179394721985, + "selected_utility_gain_over_base": -0.004487395286560059, + "selector_regret": 1.2808820605278015, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.7478358458742458 + }, + "2012211287/train2": { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3804347826086957, + "pairwise_causal_calibration_confidence": 0.5546047203835289, + "pairwise_causal_calibration_ece": 0.1741699377748331, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 1.285926111665494, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7165453433990479, + "selected_utility_gain_over_base": -0.007059991359710693, + "selector_regret": 1.2834546566009521, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.7478358458742458 + }, + "2038594191/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3448275862068966, + "pairwise_causal_calibration_confidence": 0.6255201768318296, + "pairwise_causal_calibration_ece": 0.3473762370913279, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.1448828501178339, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04285776615142822, + "selected_utility_gain_over_base": -1.9571422338485718, + "selector_regret": 1.9571422338485718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "2038594191/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3148148148148148, + "pairwise_causal_calibration_confidence": 0.6741121240452155, + "pairwise_causal_calibration_ece": 0.3592973092304006, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.6895335177823154, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07674666494131088, + "selected_utility_gain_over_base": -1.9232533350586891, + "selector_regret": 1.9232533350586891, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "2038594191/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.40404040404040403, + "pairwise_causal_calibration_confidence": 0.5932032968584295, + "pairwise_causal_calibration_ece": 0.18916289281802573, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 1.2250107783962947, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04917396977543831, + "selected_utility_gain_over_base": -1.9508260302245617, + "selector_regret": 1.9508260302245617, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "2054899126/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.275, + "pairwise_causal_calibration_confidence": 0.5896370284261352, + "pairwise_causal_calibration_ece": 0.15731851421306756, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 1.7750331290407644, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.077283151447773, + "selected_candidate_index": 2.0, + "selected_success": 0.5, + "selected_success_gain_over_base": -0.5, + "selected_utility": 1.0391597636044025, + "selected_utility_gain_over_base": -0.9608402363955975, + "selector_regret": 0.03812338784337044, + "success_selector_gap": 0.0, + "success_support_gap": 0.5, + "support_gap": 0.922716848552227, + "tau": 0.9470671281950391 + }, + "2054899126/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6228070175438597, + "pairwise_causal_calibration_confidence": 0.5635260175535555, + "pairwise_causal_calibration_ece": 0.029640499995152086, + "pairwise_causal_calibration_pairs": 57.0, + "predicted_margin": 1.7683790333511047, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.5, + "selected_success": 0.5, + "selected_success_gain_over_base": -0.5, + "selected_utility": 1.0559737384319305, + "selected_utility_gain_over_base": -0.9440262615680695, + "selector_regret": 0.9440262615680695, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "2054899126/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5210084033613446, + "pairwise_causal_calibration_confidence": 0.6169594875657232, + "pairwise_causal_calibration_ece": 0.04797554210218943, + "pairwise_causal_calibration_pairs": 59.5, + "predicted_margin": 1.6308831091591878, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 8.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "2061635888/train0": { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5740299161789739, + "pairwise_causal_calibration_ece": 0.1173721779867959, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.7792589748622651, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09799930453300476, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04574880003929138, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9020006954669952, + "tau": 0.893807936331132 + }, + "2061635888/train1": { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.2777777777777778, + "pairwise_causal_calibration_confidence": 0.555294391767556, + "pairwise_causal_calibration_ece": 0.27751661398977834, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.7080441039530474, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07580336928367615, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02355286478996277, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9241966307163239, + "tau": 0.893807936331132 + }, + "2061635888/train2": { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.5650883149039513, + "pairwise_causal_calibration_ece": 0.23175498157061808, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.6784286329948439, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11105132102966309, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.05880081653594971, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.888948678970337, + "tau": 0.893807936331132 + }, + "211245842/train0": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.5845752206379711, + "pairwise_causal_calibration_ece": 0.04745706568944737, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.9680028345756826, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8090915828943253, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "211245842/train1": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5416666666666666, + "pairwise_causal_calibration_confidence": 0.5914410404762934, + "pairwise_causal_calibration_ece": 0.10787248047280223, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.097151631086689, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "selected_candidate_index": 12.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19218182563781738, + "selected_utility_gain_over_base": 0.0012734085321426392, + "selector_regret": 0.02077534794807434, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7870428264141083, + "tau": 0.893807936331132 + }, + "211245842/train2": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.575, + "pairwise_causal_calibration_confidence": 0.5804806828596304, + "pairwise_causal_calibration_ece": 0.020991444611033885, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7949698501065345, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "selected_candidate_index": 12.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04297435283660889, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7661172300577164, + "tau": 0.893807936331132 + }, + "278008524/train0": { + "base_success": 0.0, + "base_utility": 0.042734989896416664, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6866161616161617, + "pairwise_causal_calibration_confidence": 0.5986517424099531, + "pairwise_causal_calibration_ece": 0.10091717082382054, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.37442228448179016, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0733563229441643, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.042734989896416664, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.0306213330477476, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.04055231809616089, + "tau": 0.9470671281950391 + }, + "278008524/train1": { + "base_success": 0.0, + "base_utility": 0.042734989896416664, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6382246376811594, + "pairwise_causal_calibration_confidence": 0.5821464869551558, + "pairwise_causal_calibration_ece": 0.11826680760261055, + "pairwise_causal_calibration_pairs": 106.0, + "predicted_margin": 0.3884435252467696, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0376745946705341, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.042734989896416664, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.9949396047741175, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.07623404636979103, + "tau": 0.9470671281950391 + }, + "278008524/train2": { + "base_success": 0.0, + "base_utility": 0.042734989896416664, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.47373737373737373, + "pairwise_causal_calibration_confidence": 0.5854704115820064, + "pairwise_causal_calibration_ece": 0.16638918244078008, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.3408252345971372, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.1359489113092422, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.042734989896416664, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.0932139214128256, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "307068879/train0": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7032967032967034, + "pairwise_causal_calibration_confidence": 0.5795355714423935, + "pairwise_causal_calibration_ece": 0.1237611318543097, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.18766723699772747, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2115575522184372, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7884424477815628, + "tau": 0.7478358458742458 + }, + "307068879/train1": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7692307692307693, + "pairwise_causal_calibration_confidence": 0.5687146745124166, + "pairwise_causal_calibration_ece": 0.2005160947183524, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.12417642494936398, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5876073241233826, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "tau": 0.7478358458742458 + }, + "307068879/train2": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5947805123965023, + "pairwise_causal_calibration_ece": 0.045119558117126445, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.6469329014391034, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2302134484052658, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7697865515947342, + "tau": 0.7478358458742458 + }, + "412102563/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.6052287842223839, + "pairwise_causal_calibration_ece": 0.08071518251541011, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 1.4315714539962017, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "412102563/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5954452533432482, + "pairwise_causal_calibration_ece": 0.1017872953636167, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": 1.4327881584967936, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "412102563/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.6315411122591507, + "pairwise_causal_calibration_ece": 0.2148744455924842, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.5236864695041792, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "440577711/train0": { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7586206896551724, + "pairwise_causal_calibration_confidence": 0.5841905535939795, + "pairwise_causal_calibration_ece": 0.17443013606119306, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.839399049033097, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "440577711/train1": { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4, + "pairwise_causal_calibration_confidence": 0.5669827903966217, + "pairwise_causal_calibration_ece": 0.16698279039662164, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 2.0021154026255705, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "440577711/train2": { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6111111111111112, + "pairwise_causal_calibration_confidence": 0.6090600639786808, + "pairwise_causal_calibration_ece": 0.2232694799037939, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.913369843520469, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "441046872/train0": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.5622784804732108, + "pairwise_causal_calibration_ece": 0.1804602986550292, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.422930196465892, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.3000448346138, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.7478358458742458 + }, + "441046872/train1": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5636363636363636, + "pairwise_causal_calibration_confidence": 0.5559054729794208, + "pairwise_causal_calibration_ece": 0.02996166181944273, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.2197157946327053, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7233588695526123, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.023403704166412354, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "tau": 0.7478358458742458 + }, + "441046872/train2": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6181818181818182, + "pairwise_causal_calibration_confidence": 0.599086589574069, + "pairwise_causal_calibration_ece": 0.05893542172411783, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.3023349173152088, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7191348671913147, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019179701805114746, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "tau": 0.7478358458742458 + }, + "509430392/train0": { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6050420168067226, + "pairwise_causal_calibration_confidence": 0.5998439015094801, + "pairwise_causal_calibration_ece": 0.07491598280692124, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 2.434169345753335, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 11.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "509430392/train1": { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.6128506472019413, + "pairwise_causal_calibration_ece": 0.1382932703018193, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 2.300921534892476, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 2.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.2264360294338528 + }, + "509430392/train2": { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6470588235294118, + "pairwise_causal_calibration_confidence": 0.6259952855490101, + "pairwise_causal_calibration_ece": 0.15999043325783527, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 2.088591462744573, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 5.424022674560547e-06, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 5.424022674560547e-06, + "selected_utility_gain_over_base": 5.200505256652832e-06, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9999945759773254, + "tau": 1.2264360294338528 + }, + "585904130/train0": { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5597842433887508, + "pairwise_causal_calibration_ece": 0.08130419295423207, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.4903299974496993, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09919407218694687, + "selected_utility_gain_over_base": -0.0825214758515358, + "selector_regret": 1.9008059278130531, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "585904130/train1": { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4266666666666667, + "pairwise_causal_calibration_confidence": 0.5738533196584167, + "pairwise_causal_calibration_ece": 0.14718665299174993, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 1.4765054459187703, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04792279005050659, + "selected_utility_gain_over_base": -0.13379275798797607, + "selector_regret": 1.9520772099494934, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "585904130/train2": { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6363636363636364, + "pairwise_causal_calibration_confidence": 0.5450229275953111, + "pairwise_causal_calibration_ece": 0.09134070876832519, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.1772480571547321, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8182844519615173, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "608016290/train0": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6571428571428571, + "pairwise_causal_calibration_confidence": 0.5840070772055741, + "pairwise_causal_calibration_ece": 0.07313577993728299, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.2718540448404614, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 12.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.7478358458742458 + }, + "608016290/train1": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.6130721925494387, + "pairwise_causal_calibration_ece": 0.157312300699128, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.9190150483991828, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7279855608940125, + "selected_utility_gain_over_base": 0.002283930778503418, + "selector_regret": 1.2720144391059875, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.7478358458742458 + }, + "608016290/train2": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.6026118586438607, + "pairwise_causal_calibration_ece": 0.0859451919771938, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.080232004048101, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7117276191711426, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.22811375558376312, + "selected_utility_gain_over_base": -0.4975878745317459, + "selector_regret": 0.48361386358737946, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "tau": 0.7478358458742458 + }, + "626486813/train0": { + "base_success": 0.0, + "base_utility": 0.00509834272088483, + "coverage": 0.5, + "execute_generated": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.16558238118886948, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.475, + "pairwise_causal_calibration_confidence": 0.5748306935568978, + "pairwise_causal_calibration_ece": 0.15407519274835713, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.767002576627011, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04469547560438514, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007657766109332442, + "selected_utility_gain_over_base": 0.0025594233884476125, + "selector_regret": 0.037037709495052695, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12161866808310151, + "tau": 0.9470671281950391 + }, + "626486813/train1": { + "base_success": 0.0, + "base_utility": 0.00509834272088483, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.16558238118886948, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4375, + "pairwise_causal_calibration_confidence": 0.5875658211040858, + "pairwise_causal_calibration_ece": 0.23910795594093184, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9231447045402131, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.035827874671667814, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.00509834272088483, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.030729531950782984, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12975450651720166, + "tau": 0.9470671281950391 + }, + "626486813/train2": { + "base_success": 0.0, + "base_utility": 0.00509834272088483, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.16558238118886948, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.55, + "pairwise_causal_calibration_confidence": 0.5701100933633427, + "pairwise_causal_calibration_ece": 0.02309179079813464, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3758789429127276, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04489925503730774, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.00509834272088483, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.03980091231642291, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12068312615156174, + "tau": 0.9470671281950391 + }, + "636680137/train0": { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5986660164452574, + "pairwise_causal_calibration_ece": 0.18530981522251466, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.70117503855084, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.08826130628585815, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08826130628585815, + "selected_utility_gain_over_base": 0.005126260221004486, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0823030173778534, + "tau": 0.9470671281950391 + }, + "636680137/train1": { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6833333333333333, + "pairwise_causal_calibration_confidence": 0.5644217102464404, + "pairwise_causal_calibration_ece": 0.11891162308689304, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9868011033647903, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06741118431091309, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.017789265140891075, + "selected_utility_gain_over_base": -0.0653457809239626, + "selector_regret": 0.04962191917002201, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10315313935279846, + "tau": 0.9470671281950391 + }, + "636680137/train2": { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.551260515476342, + "pairwise_causal_calibration_ece": 0.06540615119032467, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9766279269153202, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0756516233086586, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.041415613144636154, + "selected_utility_gain_over_base": -0.041719432920217514, + "selector_regret": 0.034236010164022446, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09491270035505295, + "tau": 0.9470671281950391 + }, + "655884605/train0": { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6388888888888888, + "pairwise_causal_calibration_confidence": 0.605234361994273, + "pairwise_causal_calibration_ece": 0.09534125846142073, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.9134188949984638, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13732868432998657, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12222731113433838, + "selected_utility_gain_over_base": -0.040779367089271545, + "selector_regret": 0.015101373195648193, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8626713156700134, + "tau": 0.893807936331132 + }, + "655884605/train1": { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5888894132525583, + "pairwise_causal_calibration_ece": 0.08888941325255821, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 1.0866492061529267, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8128758072853088, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11176733672618866, + "selected_utility_gain_over_base": -0.051239341497421265, + "selector_regret": 0.7011084705591202, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.1871241927146912, + "tau": 0.893807936331132 + }, + "655884605/train2": { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.25, + "pairwise_causal_calibration_confidence": 0.5949151821754458, + "pairwise_causal_calibration_ece": 0.34491518217544576, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.9421624207634716, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21447330713272095, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09131735563278198, + "selected_utility_gain_over_base": -0.07168932259082794, + "selector_regret": 0.12315595149993896, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.785526692867279, + "tau": 0.893807936331132 + }, + "672328262/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6470588235294118, + "pairwise_causal_calibration_confidence": 0.6057740013777554, + "pairwise_causal_calibration_ece": 0.08437381833742622, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": 1.0088593202243494, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "672328262/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.6362714146835213, + "pairwise_causal_calibration_ece": 0.05483727064861237, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": 1.1331098956489296, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "672328262/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.26666666666666666, + "pairwise_causal_calibration_confidence": 0.6313448801933792, + "pairwise_causal_calibration_ece": 0.3646782135267123, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": 0.8920977858545905, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 6.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.9470671281950391 + }, + "701667486/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5760031182899459, + "pairwise_causal_calibration_ece": 0.20253601281087805, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 1.0829493379419586, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "701667486/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5952380952380952, + "pairwise_causal_calibration_confidence": 0.5668881781988476, + "pairwise_causal_calibration_ece": 0.02834991703924743, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.0614915153045223, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 6.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "701667486/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7407407407407407, + "pairwise_causal_calibration_confidence": 0.5964473013091497, + "pairwise_causal_calibration_ece": 0.14429343943159095, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.1464704263550276, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.893807936331132 + }, + "747260437/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.45, + "pairwise_causal_calibration_confidence": 0.5630309153002454, + "pairwise_causal_calibration_ece": 0.11303091530024553, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0110612897022033, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10839563608169556, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.08987965434789658, + "selected_utility_gain_over_base": -1.9101203456521034, + "selector_regret": 0.01851598173379898, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17026761174201965, + "tau": 0.9470671281950391 + }, + "747260437/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3416666666666667, + "pairwise_causal_calibration_confidence": 0.5662548497567926, + "pairwise_causal_calibration_ece": 0.2245881830901258, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6563621545761416, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06328952312469482, + "selected_candidate_index": 13.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.21537372469902039, + "tau": 0.9470671281950391 + }, + "747260437/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.5582193713806621, + "pairwise_causal_calibration_ece": 0.07511396195267138, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.353137376402291, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.12510187923908234, + "selected_candidate_index": 14.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.0880771279335022, + "selected_utility_gain_over_base": -1.9119228720664978, + "selector_regret": 0.03702475130558014, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15356136858463287, + "tau": 0.9470671281950391 + }, + "750448296/train0": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.5365079365079365, + "pairwise_causal_calibration_confidence": 0.5831784025576913, + "pairwise_causal_calibration_ece": 0.06832403574472125, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": 0.8860884501921475, + "proposal_oracle_success": 0.3333333333333333, + "proposal_oracle_utility": 0.7526603279014429, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09184517463048299, + "selected_utility_gain_over_base": -0.02574797471364339, + "selector_regret": 0.6608151532709599, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.3333333333333333, + "support_gap": 0.6553361676633358, + "tau": 0.9470671281950391 + }, + "750448296/train1": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3938034188034188, + "pairwise_causal_calibration_confidence": 0.5885015589506045, + "pairwise_causal_calibration_ece": 0.2032871427289963, + "pairwise_causal_calibration_pairs": 101.66666666666667, + "predicted_margin": 0.7733918882065844, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3705032741030057, + "selected_candidate_index": 10.666666666666666, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10300639644265175, + "selected_utility_gain_over_base": -0.014586752901474634, + "selector_regret": 1.267496877660354, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.03749322146177292, + "tau": 0.9470671281950391 + }, + "750448296/train2": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.6666666666666666, + "execute_generated": 0.6666666666666666, + "fallback_rate": 0.3333333333333333, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.5599206349206348, + "pairwise_causal_calibration_confidence": 0.5772695267607532, + "pairwise_causal_calibration_ece": 0.08824336924083552, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": 0.892383539462905, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3504486680030823, + "selected_candidate_index": 8.333333333333334, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07752946888407071, + "selected_utility_gain_over_base": -0.04006368046005567, + "selector_regret": 1.2729191991190116, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.05754782756169637, + "tau": 0.9470671281950391 + }, + "767747424/train0": { + "base_success": 0.25, + "base_utility": 0.5731993420049548, + "coverage": 0.25, + "execute_generated": 0.25, + "fallback_rate": 0.75, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.5073717948717948, + "pairwise_causal_calibration_confidence": 0.5776709098378972, + "pairwise_causal_calibration_ece": 0.15246329356188465, + "pairwise_causal_calibration_pairs": 119.25, + "predicted_margin": 0.6516149173146432, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0424291878007352, + "selected_candidate_index": 4.75, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.25, + "selected_utility": 1.0327778598293662, + "selected_utility_gain_over_base": 0.4595785178244114, + "selector_regret": 0.018381604924798012, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.06562284985557199, + "tau": 0.9470671281950391 + }, + "767747424/train1": { + "base_success": 0.25, + "base_utility": 0.5731993420049548, + "coverage": 0.25, + "execute_generated": 0.25, + "fallback_rate": 0.75, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.5107035121741004, + "pairwise_causal_calibration_confidence": 0.5904083439254464, + "pairwise_causal_calibration_ece": 0.1406609793333544, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.8997026322297819, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0629530912265182, + "selected_candidate_index": 8.25, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5613184677204117, + "selected_utility_gain_over_base": -0.011880874284543097, + "selector_regret": 0.5016346235061064, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.045098946429789066, + "tau": 0.9470671281950391 + }, + "767747424/train2": { + "base_success": 0.25, + "base_utility": 0.5731993420049548, + "coverage": 0.25, + "execute_generated": 0.25, + "fallback_rate": 0.75, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.4354723299576241, + "pairwise_causal_calibration_confidence": 0.5739155276455858, + "pairwise_causal_calibration_ece": 0.1659671732959123, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5403697337834386, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0671670949086547, + "selected_candidate_index": 8.75, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5671488647349179, + "selected_utility_gain_over_base": -0.006050477270036936, + "selector_regret": 0.5000182301737368, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.04088494274765253, + "tau": 0.9470671281950391 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/metrics_by_task.json b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/metrics_by_task.json new file mode 100644 index 0000000000000000000000000000000000000000..a43c26cf0006477824b9771c010e9a15eac1789b --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/metrics_by_task.json @@ -0,0 +1,137 @@ +{ + "LiftPegUpright-v1": { + "base_success": 0.2, + "base_utility": 0.9498705387115478, + "coverage": 0.6, + "execute_generated": 0.6, + "fallback_rate": 0.4, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.5333333333333333, + "pairwise_causal_calibration_accuracy": 0.5555136707566375, + "pairwise_causal_calibration_confidence": 0.5720692082177231, + "pairwise_causal_calibration_ece": 0.11666816996110206, + "pairwise_causal_calibration_pairs": 94.8, + "predicted_margin": 0.7829724526600295, + "proposal_oracle_success": 0.6, + "proposal_oracle_utility": 1.4122399787108104, + "selected_candidate_index": 5.133333333333334, + "selected_success": 0.06666666666666667, + "selected_success_gain_over_base": -0.13333333333333333, + "selected_utility": 0.7418742964665095, + "selected_utility_gain_over_base": -0.20799624224503835, + "selector_regret": 0.7217585732539494, + "success_selector_gap": 0.5333333333333333, + "success_support_gap": 0.4, + "support_gap": 0.5877600212891897, + "tau": 0.7478358458742457 + }, + "PickCube-v1": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.6023940721845762, + "pairwise_causal_calibration_ece": 0.2745681321943875, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.12247428613675211, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13425676027933756, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8657432397206624, + "tau": 1.280136139242778 + }, + "PullCube-v1": { + "base_success": 0.36, + "base_utility": 0.7751704612234608, + "coverage": 0.52, + "execute_generated": 0.52, + "fallback_rate": 0.48, + "hidden_chart_oracle_success": 0.52, + "hidden_chart_oracle_utility": 1.1467878860235214, + "outcome_ptr": 0.4666666666666667, + "pairwise_causal_calibration_accuracy": 0.5062744920994355, + "pairwise_causal_calibration_confidence": 0.5796546075128097, + "pairwise_causal_calibration_ece": 0.12491436687229007, + "pairwise_causal_calibration_pairs": 92.32, + "predicted_margin": 0.9547622553200451, + "proposal_oracle_success": 0.5333333333333333, + "proposal_oracle_utility": 1.1093241602430741, + "selected_candidate_index": 7.266666666666667, + "selected_success": 0.32, + "selected_success_gain_over_base": -0.04, + "selected_utility": 0.6846797956805676, + "selected_utility_gain_over_base": -0.09049066554289312, + "selector_regret": 0.45504458223469557, + "success_selector_gap": 0.22666666666666666, + "success_support_gap": 0.02666666666666667, + "support_gap": 0.1075624164690574, + "tau": 0.9470671281950384 + }, + "PushCube-v1": { + "base_success": 0.16666666666666666, + "base_utility": 0.3924041787783305, + "coverage": 0.8888888888888888, + "execute_generated": 0.8888888888888888, + "fallback_rate": 0.1111111111111111, + "hidden_chart_oracle_success": 0.8333333333333334, + "hidden_chart_oracle_utility": 1.7015800600250561, + "outcome_ptr": 0.8333333333333334, + "pairwise_causal_calibration_accuracy": 0.5428451728329846, + "pairwise_causal_calibration_confidence": 0.6000598858100837, + "pairwise_causal_calibration_ece": 0.15543541117172033, + "pairwise_causal_calibration_pairs": 97.44444444444444, + "predicted_margin": 1.843134202495152, + "proposal_oracle_success": 0.8888888888888888, + "proposal_oracle_utility": 1.7857814149724112, + "selected_candidate_index": 6.222222222222222, + "selected_success": 0.3888888888888889, + "selected_success_gain_over_base": 0.2222222222222222, + "selected_utility": 0.8422558406988779, + "selected_utility_gain_over_base": 0.4498516619205475, + "selector_regret": 0.9435255742735333, + "success_selector_gap": 0.5, + "success_support_gap": 0.1111111111111111, + "support_gap": 0.21421858502758873, + "tau": 1.2264360294338525 + }, + "StackCube-v1": { + "base_success": 0.2727272727272727, + "base_utility": 0.7763947045261209, + "coverage": 0.5151515151515151, + "execute_generated": 0.5151515151515151, + "fallback_rate": 0.48484848484848486, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.6363636363636364, + "pairwise_causal_calibration_accuracy": 0.5286305231046764, + "pairwise_causal_calibration_confidence": 0.5839574678118631, + "pairwise_causal_calibration_ece": 0.13914695546983782, + "pairwise_causal_calibration_pairs": 67.45454545454545, + "predicted_margin": 0.832831201132664, + "proposal_oracle_success": 0.5151515151515151, + "proposal_oracle_utility": 1.1567951707225856, + "selected_candidate_index": 6.606060606060606, + "selected_success": 0.30303030303030304, + "selected_success_gain_over_base": 0.030303030303030304, + "selected_utility": 0.7861830118027601, + "selected_utility_gain_over_base": 0.00978830727663907, + "selector_regret": 0.37342702303871966, + "success_selector_gap": 0.21212121212121213, + "success_support_gap": 0.48484848484848486, + "support_gap": 0.8432048292774142, + "tau": 0.8938079363311323 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/table.tex b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/table.tex new file mode 100644 index 0000000000000000000000000000000000000000..0fd442f231f2d6b91e6407ffb73747c446bf854c --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/table.tex @@ -0,0 +1,8 @@ +% Auto-generated by scripts/eval_learned_dominance_selector.py +\begin{tabular}{lrrrrrrrrr} +\toprule +Rows & Coverage & Fallback & Base succ. & Selected succ. & Oracle succ. & OutcomePTR & Succ. support gap & Succ. selector gap & Cal. ECE \\ +\midrule +144 & 0.5625 & 0.4375 & 0.2917 & 0.2917 & 0.5694 & 0.5486 & 0.2014 & 0.2847 & 0.0611 \\ +\bottomrule +\end{tabular} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/train.log b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/train.log new file mode 100644 index 0000000000000000000000000000000000000000..7dda9703b7f04afbbdf2513328d6f52da93ef305 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_chart_source_bundle_explore_envclip_k16_train_to_test/train.log @@ -0,0 +1,5 @@ +trained ridge dominance calibrator on calibration measured rows only +fit_objective=pointwise +selected_lambda=1.0 +tau={"LiftPegUpright-v1": 0.747836, "PickCube-v1": 1.280136, "PullCube-v1": 0.947067, "PushCube-v1": 1.226436, "StackCube-v1": 0.893808, "__global__": 0.906341} +calibration_input=runs/ctt_base_context_obs_train_cal_envclip_k16_rollout_comparison/combined_measured_candidates.json