diff --git a/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/command.txt b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/command.txt new file mode 100644 index 0000000000000000000000000000000000000000..9bd04eceac39253b9e35f26e0db2a42a1d78bf55 --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/command.txt @@ -0,0 +1 @@ +python scripts/eval_learned_dominance_selector.py --calibration-input runs/ctt_base_context_obs_train_cal_envclip_k16_rollout_comparison/combined_measured_candidates.json --calibration-target-index data/cil_charts_rgb_refs/train/index.json --eval-input runs/ctt_base_context_obs_test_envclip_k16_rollout_comparison/combined_measured_candidates.json --eval-target-index data/cil_charts_rgb_refs/test/index.json --source-index data/cil_charts_rgb_refs/train/index.json --checkpoint-template runs/ctt_residual_base_context_obs_seed{seed}/model.pt --out-dir runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test --feature-set score_context_chart_compat --selector-chart-feature-mode base_context_obs --target success_weighted_margin --success-bonus 2.0 --threshold-scope task --k 16 --bootstrap-samples 300 --no-markdown-report diff --git a/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/config.yaml b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f6dd790b4a5d3b6d9e35beab9e365d9e1d168515 --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/config.yaml @@ -0,0 +1,18 @@ +bootstrap_samples: 300 +calibration_input: runs/ctt_base_context_obs_train_cal_envclip_k16_rollout_comparison/combined_measured_candidates.json +calibration_target_index: data/cil_charts_rgb_refs/train/index.json +checkpoint_template: runs/ctt_residual_base_context_obs_seed{seed}/model.pt +eval_input: runs/ctt_base_context_obs_test_envclip_k16_rollout_comparison/combined_measured_candidates.json +eval_target_index: data/cil_charts_rgb_refs/test/index.json +feature_set: score_context_chart_compat +fit_objective: pointwise +k: 16 +no_markdown_report: True +out_dir: runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test +pairwise_weight: 1.0 +ridge_lambdas: 0,0.01,0.1,1,10,100 +selector_chart_feature_mode: base_context_obs +source_index: data/cil_charts_rgb_refs/train/index.json +success_bonus: 2.0 +target: success_weighted_margin +threshold_scope: task diff --git a/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/data_hash.txt b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/data_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..e65c158c17e61aaf10422563d2ffdd8922735e6a --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/data_hash.txt @@ -0,0 +1,7 @@ +{ + "calibration_input": "d79e275b896eb98caa128470d4614b76960cab12407faa157c583224be7abe80", + "calibration_target_index": "67973c50072aa87a3b97533b6292b08a8b4d88a95571dc22e91b6e8fd19e0d85", + "eval_input": "541d5a928da5f7ee04ca3dea570e04208830dffde431edfe5b5e38b692634045", + "eval_target_index": "dee1b5225485df6624747dd4c5074bcf514685b54017d05e698d2051638fbf36", + "source_index": "67973c50072aa87a3b97533b6292b08a8b4d88a95571dc22e91b6e8fd19e0d85" +} diff --git a/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/eval.log b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/eval.log new file mode 100644 index 0000000000000000000000000000000000000000..ad44bd37e51f82471f7a17d7372cee5fed00ed3f --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_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/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/git_hash.txt b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/git_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..293c6ad4e775f2f937dca574f1b83a5f93f0bdb1 --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/git_hash.txt @@ -0,0 +1 @@ +14e81292e2004925f81e0bb4810f6f144bfd87b7 diff --git a/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics.json b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..8ce1bec6c587d4e514928392e675de63bbc698d6 --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics.json @@ -0,0 +1,9116 @@ +{ + "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": 0.5300925925925926, + "fallback_rate": 0.4699074074074074, + "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": 0.3541666666666667, + "selected_success_gain_over_base": 0.0763888888888889, + "selected_utility": 0.825786815760862, + "selector_regret": 0.40355927437189226, + "success_selector_gap": 0.2222222222222222, + "success_support_gap": 0.17592592592592593, + "support_gap": 0.35465237537627453 + }, + "calibration_summary": { + "base_success": 0.2777777777777778, + "base_utility": 0.7007981554626086, + "coverage": 0.5300925925925926, + "fallback_rate": 0.4699074074074074, + "hidden_chart_oracle_success": 0.6597222222222222, + "hidden_chart_oracle_utility": 1.4150809271054134, + "outcome_ptr": 0.5462962962962963, + "pairwise_causal_calibration_accuracy": 0.5059771446626995, + "pairwise_causal_calibration_confidence": 0.5662641043006831, + "pairwise_causal_calibration_ece": 0.060286959637983646, + "pairwise_causal_calibration_pairs": 37978.0, + "proposal_oracle_success": 0.5370370370370371, + "proposal_oracle_utility": 1.1440672337592803, + "selected_success": 0.3541666666666667, + "selected_success_gain_over_base": 0.0763888888888889, + "selected_utility": 0.825786815760862, + "selector_regret": 0.40355927437189226, + "success_selector_gap": 0.2222222222222222, + "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": 0.5328936786231301, + "pairwise_causal_calibration_confidence": 0.5663262950707817, + "pairwise_causal_calibration_ece": 0.036522383777275284, + "pairwise_causal_calibration_pairs": 12434.0, + "proposal_oracle_success": 0.5694444444444444, + "proposal_oracle_utility": 1.2159999172436073, + "selected_success": 0.3402777777777778, + "selected_success_gain_over_base": 0.04861111111111111, + "selected_utility": 0.802287638622349, + "selector_regret": 0.4247725029110572, + "success_selector_gap": 0.22916666666666666, + "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, + 0.5, + 0.06635944700460818, + -1.1461792655128946, + -0.8539997540940796, + -0.16156509362744845, + 0.16156509362744845, + 0.5331797235023041, + 0.29217951141881504, + 0.02142224631788315, + -0.009975603144483841, + -0.4435059184964108, + 0.741602963766158, + 0.48853347790731816, + 0.6095744104092821, + 0.37631155688083895, + -0.07796512153248361, + 0.02142224631788315, + -0.009975603144483841, + -0.4435059184964108, + 0.741602963766158, + 0.48853347790731816, + 0.6095744104092821, + 0.37631155688083895, + -0.07796512153248361, + 0.34864732989968916, + -0.484069756935069, + 0.13361525255263118, + 0.30599078341012464, + -0.5450293665853819, + -0.5758453058642928, + 0.502998102466806, + -0.11478810879190798, + 1.0, + 0.2692288306451613, + 0.22799539170506913, + 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 + ], + "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", + "candidate_score_rank_fraction", + "candidate_score_softmax_prob", + "candidate_score_gap_to_best", + "candidate_score_gap_to_second_best", + "candidate_score_gap_to_prev_higher", + "candidate_score_gap_to_next_lower", + "candidate_score_percentile", + "candidate_score_top_margin", + "target_task_hash_00", + "target_task_hash_01", + "target_task_hash_02", + "target_task_hash_03", + "target_task_hash_04", + "target_task_hash_05", + "target_task_hash_06", + "target_task_hash_07", + "source_task_hash_00", + "source_task_hash_01", + "source_task_hash_02", + "source_task_hash_03", + "source_task_hash_04", + "source_task_hash_05", + "source_task_hash_06", + "source_task_hash_07", + "instruction_hash_00", + "instruction_hash_01", + "instruction_hash_02", + "instruction_hash_03", + "instruction_hash_04", + "instruction_hash_05", + "instruction_hash_06", + "instruction_hash_07", + "source_target_same_task", + "instruction_chars_per_128", + "instruction_words_per_32", + "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" + ], + "feature_set": "score_context_chart_compat", + "feature_std": [ + 1.0, + 2.5232884736710055, + 2.2444575887945213, + 0.9999992484766103, + 4.542613580634851, + 0.24891036261589178, + 1.9642314545113713, + 0.37078681697003224, + 0.7171840796063713, + 1.7485561930736828, + 0.30856036102426265, + 0.04867152161828175, + 0.9290082113338436, + 0.8520345362910837, + 0.21685697548301888, + 0.21685697548301888, + 0.2880797360439561, + 0.30393911054414136, + 0.353466344495717, + 0.5934225297989979, + 0.27998337850869326, + 0.2537489327134842, + 0.12495790521337884, + 0.28095509978260164, + 0.35944258338923735, + 0.2407095730397282, + 0.353466344495717, + 0.5934225297989979, + 0.27998337850869326, + 0.2537489327134842, + 0.12495790521337884, + 0.28095509978260164, + 0.35944258338923735, + 0.2407095730397282, + 0.3692654145721106, + 0.32347163055420985, + 0.31016830264374196, + 0.30908751583334115, + 0.40620363629118383, + 0.3797486870020415, + 0.4751352882966128, + 0.7141433326304655, + 1e-06, + 0.026199955886555618, + 0.023776289266070644, + 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 + ], + "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": 0.5059771446626995, + "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": 0.02807611697713175, + "accuracy": 0.5065266841644794, + "confidence": 0.5346028011416112, + "count": 28575, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.13357354156207535, + "accuracy": 0.5054137147440182, + "confidence": 0.6389872563060935, + "count": 7481, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.2322620661921625, + "accuracy": 0.5055079559363526, + "confidence": 0.7377700221285151, + "count": 1634, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.3456248818393146, + "accuracy": 0.4963235294117647, + "confidence": 0.8419484112510793, + "count": 272, + "lower": 0.8, + "upper": 0.9 + }, + { + "abs_gap": 0.9070788791018938, + "accuracy": 0.0, + "confidence": 0.9070788791018938, + "count": 16, + "lower": 0.9, + "upper": 1.0 + } + ], + "ece": 0.060286959637983646, + "mean_confidence": 0.5662641043006831, + "n_bins": 10, + "num_pairs": 37978, + "num_rows": 432 + }, + "eval": { + "accuracy": 0.5328936786231301, + "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": 0.0020224345639368035, + "accuracy": 0.5380080016845652, + "confidence": 0.5359855671206284, + "count": 9498, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.13166615763674738, + "accuracy": 0.5057624113475178, + "confidence": 0.6374285689842651, + "count": 2256, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.1860502661065786, + "accuracy": 0.5493934142114385, + "confidence": 0.7354436803180171, + "count": 577, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.29631438083855477, + "accuracy": 0.5631067961165048, + "confidence": 0.8594211769550596, + "count": 103, + "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": 0.036522383777275284, + "mean_confidence": 0.5663262950707817, + "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": 0.5546218487394958, + "pairwise_causal_calibration_confidence": 0.5281524186595942, + "pairwise_causal_calibration_ece": 0.026469430079901657, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.7217723168310286, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.713559627532959, + "selected_utility_gain_over_base": -0.01004570722579956, + "selector_regret": 1.286440372467041, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "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": 0.5934065934065934, + "pairwise_causal_calibration_confidence": 0.5944461291750907, + "pairwise_causal_calibration_ece": 0.12040324822018197, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 1.0739910273739994, + "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": 0.19761089980602264, + "selected_utility_gain_over_base": -0.4024796634912491, + "selector_regret": 0.01394665241241455, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7884424477815628, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "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": 0.6386554621848739, + "pairwise_causal_calibration_confidence": 0.5373743568439672, + "pairwise_causal_calibration_ece": 0.10128110534090673, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5566626770858889, + "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": 0.49469300263486815, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "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": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5478322151873722, + "pairwise_causal_calibration_ece": 0.04803994731985866, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.34122628210022654, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/train0", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.725701630115509, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.274298369884491, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "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": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.6008428753783649, + "pairwise_causal_calibration_ece": 0.21902469356018284, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 1.1052022133857848, + "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": 0.20773810148239136, + "selected_utility_gain_over_base": -0.4922170639038086, + "selector_regret": 1.7922618985176086, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "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": 0.7222222222222222, + "pairwise_causal_calibration_confidence": 0.5639487552532058, + "pairwise_causal_calibration_ece": 0.15827346696901654, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.3519897595066688, + "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": 0.8901452015560949, + "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": 0.7045359481010607, + "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": 0.3663750360591518, + "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": 0.48717948717948717, + "pairwise_causal_calibration_confidence": 0.5167227660116596, + "pairwise_causal_calibration_ece": 0.029543278832172393, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.4458593850215422, + "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": 0.3663750360591518, + "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": 0.8166666666666667, + "pairwise_causal_calibration_confidence": 0.5411312415303838, + "pairwise_causal_calibration_ece": 0.2755354251362828, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.3670742388145669, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06019836664199829, + "seed": "1461069713/train0", + "selected_candidate_index": 7, + "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": 0.3663750360591518, + "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": 0.7333333333333333, + "pairwise_causal_calibration_confidence": 0.6964450425935876, + "pairwise_causal_calibration_ece": 0.1276439226740295, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.7072857822793743, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.010183374397456646, + "seed": "626486813/train0", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.005017320159822702, + "selected_utility_gain_over_base": 0.0030878185061737895, + "selector_regret": 0.0051660542376339436, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24323733616620302, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.25, + "pairwise_causal_calibration_confidence": 0.566620580797147, + "pairwise_causal_calibration_ece": 0.31662058079714683, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.119907953469152, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15456630289554596, + "seed": "2054899126/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.12878061830997467, + "selected_utility_gain_over_base": -1.8712193816900253, + "selector_regret": 0.02578568458557129, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.845433697104454, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.8823529411764706, + "pairwise_causal_calibration_confidence": 0.637846163378886, + "pairwise_causal_calibration_ece": 0.24450677779758456, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": 1.7549902847176035, + "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": 0.3663750360591518, + "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": 0.4, + "pairwise_causal_calibration_confidence": 0.5992638172231477, + "pairwise_causal_calibration_ece": 0.19926381722314773, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.011655385654258055, + "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": 0.3663750360591518, + "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": 0.7583333333333333, + "pairwise_causal_calibration_confidence": 0.6483849827472722, + "pairwise_causal_calibration_ece": 0.10994835058606132, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0401359558411656, + "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": 0.0739455446600914, + "selected_utility_gain_over_base": 0.03805923834443092, + "selector_regret": 0.07276710122823715, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08110463619232178, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.6521739130434783, + "pairwise_causal_calibration_confidence": 0.5356510938834426, + "pairwise_causal_calibration_ece": 0.1165228191600357, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": 1.3305286793016613, + "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": 0.3663750360591518, + "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": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5782036666019458, + "pairwise_causal_calibration_ece": 0.08438422646141949, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7032815711510096, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07920757681131363, + "seed": "626486813/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.009500582702457905, + "selected_utility_gain_over_base": 0.0012333989143371582, + "selector_regret": 0.06970699410885572, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.7416666666666667, + "pairwise_causal_calibration_confidence": 0.5410498110420184, + "pairwise_causal_calibration_ece": 0.2006168556246484, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.34444568460277836, + "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": 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": 0.3663750360591518, + "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": 0.30952380952380953, + "pairwise_causal_calibration_confidence": 0.5734358732826558, + "pairwise_causal_calibration_ece": 0.2639120637588461, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.937964471343637, + "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": 0.3663750360591518, + "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": 0.31666666666666665, + "pairwise_causal_calibration_confidence": 0.563739392345968, + "pairwise_causal_calibration_ece": 0.24707272567930166, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6540918192032681, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "seed": "141984615/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.00405565882101655, + "selected_utility_gain_over_base": -0.09877858823165298, + "selector_regret": 0.011122246738523245, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "train_seed": "0" + }, + { + "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": 0.7, + "pairwise_causal_calibration_confidence": 0.5427511555617247, + "pairwise_causal_calibration_ece": 0.15724884443827536, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.185301885809487, + "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": 0.11457405239343643, + "selected_utility_gain_over_base": -0.02813490480184555, + "selector_regret": 0.10201253741979599, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7834134101867676, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.529875300645354, + "pairwise_causal_calibration_ece": 0.04450430387060281, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0013727088350777, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.08826130628585815, + "seed": "636680137/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.056092143058776855, + "selected_utility_gain_over_base": -0.027042903006076813, + "selector_regret": 0.0321691632270813, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0823030173778534, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.36666666666666664, + "pairwise_causal_calibration_confidence": 0.6059976479353057, + "pairwise_causal_calibration_ece": 0.3080264917794725, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.40216870081900424, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.019201835617423058, + "seed": "767747424/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007196406368166208, + "selected_utility_gain_over_base": -0.046926537062972784, + "selector_regret": 0.01200542924925685, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17790568806231022, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.64, + "pairwise_causal_calibration_confidence": 0.5585290828327523, + "pairwise_causal_calibration_ece": 0.08147091716724786, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.0461275740764826, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/train0", + "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": 0.3663750360591518, + "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": 0.27586206896551724, + "pairwise_causal_calibration_confidence": 0.6084782956989382, + "pairwise_causal_calibration_ece": 0.33261622673342073, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.5405611437535534, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/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": 0.3663750360591518, + "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": 0.5553719618949108, + "pairwise_causal_calibration_ece": 0.05063916429908746, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.2813433025246468, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.041394393891096115, + "seed": "750448296/train0", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.019331812858581543, + "selected_utility_gain_over_base": -0.07366567850112915, + "selector_regret": 0.022062581032514572, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.18259509280323982, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 1.0506886042100234, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train0", + "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": 0.3663750360591518, + "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": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.555978759483347, + "pairwise_causal_calibration_ece": 0.04733473753037369, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6951613484301653, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10839563608169556, + "seed": "747260437/train0", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07023122161626816, + "selected_utility_gain_over_base": -1.9297687783837318, + "selector_regret": 0.0381644144654274, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17026761174201965, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.6868686868686869, + "pairwise_causal_calibration_confidence": 0.5843043332170215, + "pairwise_causal_calibration_ece": 0.10256435365166547, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.061290053517613075, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train0", + "selected_candidate_index": 14, + "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": 0.3663750360591518, + "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": 0.36666666666666664, + "pairwise_causal_calibration_confidence": 0.5627369782509959, + "pairwise_causal_calibration_ece": 0.19607031158432947, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.171258504104145, + "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": 0.3663750360591518, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5864304711853732, + "pairwise_causal_calibration_ece": 0.08360394561276052, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 1.5192545920029907, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06247224286198616, + "selected_utility_gain_over_base": -0.009443063288927078, + "selector_regret": 1.9375277571380138, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "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": 0.5238095238095238, + "pairwise_causal_calibration_confidence": 0.5452287538806504, + "pairwise_causal_calibration_ece": 0.02141923007112652, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.6648270886527712, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0944695845246315, + "selected_utility_gain_over_base": -0.022603414952754974, + "selector_regret": 1.9055304154753685, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.6206896551724138, + "pairwise_causal_calibration_confidence": 0.5455151446938427, + "pairwise_causal_calibration_ece": 0.07517451047857106, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.5732830552633406, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train0", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.55, + "pairwise_causal_calibration_confidence": 0.5260274890742292, + "pairwise_causal_calibration_ece": 0.023972510925770796, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.8008980492120834, + "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": 1.311950902009979, + "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": 0.3, + "pairwise_causal_calibration_confidence": 0.519098923420412, + "pairwise_causal_calibration_ece": 0.21909892342041198, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.5993442072252648, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train0", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.87523253262043, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.46464646464646464, + "pairwise_causal_calibration_confidence": 0.5719896658662266, + "pairwise_causal_calibration_ece": 0.12912993334954873, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 1.1198710760724104, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/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": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.6470588235294118, + "pairwise_causal_calibration_confidence": 0.5508070209680679, + "pairwise_causal_calibration_ece": 0.09625180256134375, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 2.08491547386598, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/train0", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11310504376888275, + "selected_utility_gain_over_base": 0.11310482025146484, + "selector_regret": 1.8868949562311172, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.8080808080808081, + "pairwise_causal_calibration_confidence": 0.5911741147277916, + "pairwise_causal_calibration_ece": 0.21690669335301668, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 2.149778411340016, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train0", + "selected_candidate_index": 0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9129659235477448, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.5277777777777778, + "pairwise_causal_calibration_confidence": 0.5745387895297428, + "pairwise_causal_calibration_ece": 0.04676101175196509, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 1.0205351759283516, + "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": 1.4393298012330682, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "chart_id": "ms-15fc8cc6f74cf589056f", + "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": 0.5555555555555556, + "pairwise_causal_calibration_confidence": 0.6048190953611235, + "pairwise_causal_calibration_ece": 0.08237033905417408, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 1.095636223788052, + "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": 0.16300667822360992, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8626713156700134, + "task_id": "StackCube-v1", + "tau": 1.4393298012330682, + "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": 0.5701414698923739, + "pairwise_causal_calibration_ece": 0.06599981030811651, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.4419532287444907, + "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": 1.4393298012330682, + "train_seed": "0" + }, + { + "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": 0.7094017094017094, + "pairwise_causal_calibration_confidence": 0.5852719048004434, + "pairwise_causal_calibration_ece": 0.12412980460126585, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 1.369766946850774, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train0", + "selected_candidate_index": 12, + "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": 1.4393298012330682, + "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": 0.5222222222222223, + "pairwise_causal_calibration_confidence": 0.5411679300565454, + "pairwise_causal_calibration_ece": 0.018945707834323134, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.7690911091840036, + "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": 1.4393298012330682, + "train_seed": "0" + }, + { + "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": 0.6218487394957983, + "pairwise_causal_calibration_confidence": 0.5746201886713925, + "pairwise_causal_calibration_ece": 0.04722855082440577, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.2882788003210806, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "211245842/train0", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8090915828943253, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 1.4393298012330682, + "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": 0.5142857142857142, + "pairwise_causal_calibration_confidence": 0.5667135771731647, + "pairwise_causal_calibration_ece": 0.05242786288745051, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.024503811643319, + "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": 1.4393298012330682, + "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": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5755791782863937, + "pairwise_causal_calibration_ece": 0.147007749714965, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.36820169505830336, + "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": 1.4393298012330682, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "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": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5376448138211073, + "pairwise_causal_calibration_ece": 0.2519305281068216, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 0.6729506809560988, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/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": "StackCube-v1", + "tau": 1.4393298012330682, + "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": 0.65, + "pairwise_causal_calibration_confidence": 0.5385704504068652, + "pairwise_causal_calibration_ece": 0.11142954959313478, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 1.0933587971168492, + "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": 1.4393298012330682, + "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": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5351557804986392, + "pairwise_causal_calibration_ece": 0.17801292335578203, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.9988148345478187, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train0", + "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": "StackCube-v1", + "tau": 1.4393298012330682, + "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": 0.6083333333333333, + "pairwise_causal_calibration_confidence": 0.5310859795550439, + "pairwise_causal_calibration_ece": 0.0772473537782894, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5565021996855251, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7183246612548828, + "selected_utility_gain_over_base": -0.005280673503875732, + "selector_regret": 1.2816753387451172, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "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": 0.5384615384615384, + "pairwise_causal_calibration_confidence": 0.5937223391260118, + "pairwise_causal_calibration_ece": 0.0607943275756009, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.6874983921047111, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5876073241233826, + "seed": "307068879/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19690512120723724, + "selected_utility_gain_over_base": -0.4031854420900345, + "selector_regret": 0.3907022029161453, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7b229b60d7f64fb624cf", + "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": 0.6833333333333333, + "pairwise_causal_calibration_confidence": 0.5322990000175534, + "pairwise_causal_calibration_ece": 0.15103433331577998, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3872764726300346, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/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": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "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": 0.5534688429350546, + "pairwise_causal_calibration_ece": 0.03000164602114203, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.1770980989610429, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/train1", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.725701630115509, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.274298369884491, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "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": 0.6181818181818182, + "pairwise_causal_calibration_confidence": 0.6026494132350357, + "pairwise_causal_calibration_ece": 0.04900842512659715, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.6838040349024017, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7233588695526123, + "seed": "441046872/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21283115446567535, + "selected_utility_gain_over_base": -0.4871240109205246, + "selector_regret": 0.510527715086937, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "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": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5732579849122883, + "pairwise_causal_calibration_ece": 0.10103576269006627, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.7344242898918623, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07014383375644684, + "seed": "1522272686/train1", + "selected_candidate_index": 7, + "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": 0.8901452015560949, + "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": 0.6796904941310151, + "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": 0.3663750360591518, + "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": 0.40336134453781514, + "pairwise_causal_calibration_confidence": 0.5306915219990296, + "pairwise_causal_calibration_ece": 0.12733017746121444, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.2750543443626805, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train1", + "selected_candidate_index": 5, + "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": 0.3663750360591518, + "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": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5331472452449045, + "pairwise_causal_calibration_ece": 0.06648057857823786, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.16101647169097077, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "seed": "1461069713/train1", + "selected_candidate_index": 1, + "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": 0.3663750360591518, + "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": 0.26666666666666666, + "pairwise_causal_calibration_confidence": 0.6879506284014189, + "pairwise_causal_calibration_ece": 0.42128396173475235, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.5275938383554237, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.009895960800349712, + "seed": "626486813/train1", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.00289076566696167, + "selected_utility_gain_over_base": 0.000961264013312757, + "selector_regret": 0.0070051951333880424, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24352474976330996, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.6578947368421053, + "pairwise_causal_calibration_confidence": 0.563744413130429, + "pairwise_causal_calibration_ece": 0.0941503237116762, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 0.8669420362744713, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train1", + "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": 0.3663750360591518, + "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": 0.45714285714285713, + "pairwise_causal_calibration_confidence": 0.629165943748097, + "pairwise_causal_calibration_ece": 0.1720230866052398, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": 1.7117888933582819, + "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": 0.3663750360591518, + "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": 0.5833333333333334, + "pairwise_causal_calibration_confidence": 0.5812739255194401, + "pairwise_causal_calibration_ece": 0.08560483155863768, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.04654875866683028, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.20903043448925018, + "seed": "767747424/train1", + "selected_candidate_index": 2, + "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": 0.3663750360591518, + "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": 0.55, + "pairwise_causal_calibration_confidence": 0.6338433401507874, + "pairwise_causal_calibration_ece": 0.1366595206157522, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.1007077562191654, + "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": 0.3663750360591518, + "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": 0.6153846153846154, + "pairwise_causal_calibration_confidence": 0.5374675838313216, + "pairwise_causal_calibration_ece": 0.07791703155329377, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": 1.412757419493385, + "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": 0.3663750360591518, + "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": 0.4, + "pairwise_causal_calibration_confidence": 0.5694762803447845, + "pairwise_causal_calibration_ece": 0.16947628034478399, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8786000190247165, + "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": 0.3663750360591518, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7e295b5859c07c90487f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "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": 0.5282761616966329, + "pairwise_causal_calibration_ece": 0.03582640240593116, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.47482806790000787, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train1", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.002724091289564967, + "selected_utility_gain_over_base": -1.997275908710435, + "selector_regret": 1.997275908710435, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.56, + "pairwise_causal_calibration_confidence": 0.5673349851937346, + "pairwise_causal_calibration_ece": 0.17050541777278239, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.867255607957766, + "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": 0.3663750360591518, + "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": 0.625, + "pairwise_causal_calibration_confidence": 0.581566420974051, + "pairwise_causal_calibration_ece": 0.10416729357646315, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8936304139520725, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.037011127918958664, + "seed": "141984615/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.027254482731223106, + "selected_utility_gain_over_base": -0.07557976432144642, + "selector_regret": 0.009756645187735558, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24876614287495613, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "train_seed": "1" + }, + { + "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": 0.4, + "pairwise_causal_calibration_confidence": 0.5378957345206259, + "pairwise_causal_calibration_ece": 0.13789573452062587, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7048231595494219, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06543686240911484, + "selected_utility_gain_over_base": -0.07727209478616714, + "selector_regret": 1.9345631375908852, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.475, + "pairwise_causal_calibration_confidence": 0.5159628167606266, + "pairwise_causal_calibration_ece": 0.04096281676062663, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.40455029586141933, + "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": 0.3663750360591518, + "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": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.6186182445886487, + "pairwise_causal_calibration_ece": 0.24311264053469925, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7758264949510902, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.042781930416822433, + "seed": "767747424/train1", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.010063628666102886, + "selected_utility_gain_over_base": -0.044059314765036106, + "selector_regret": 0.03271830175071955, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15432559326291084, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.5610084990880762, + "pairwise_causal_calibration_ece": 0.03469270961439179, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.45720538390197873, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/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": 0.3663750360591518, + "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": 0.6260111821626294, + "pairwise_causal_calibration_ece": 0.39213662401689425, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.2016013806419839, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/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": 0.3663750360591518, + "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": 0.4, + "pairwise_causal_calibration_confidence": 0.5537352168629893, + "pairwise_causal_calibration_ece": 0.15373521686298958, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8853855149370524, + "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": 0.3663750360591518, + "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": 1.3054770183052646, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train1", + "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": 0.3663750360591518, + "train_seed": "1" + }, + { + "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": 0.675, + "pairwise_causal_calibration_confidence": 0.5687872987376212, + "pairwise_causal_calibration_ece": 0.2478180852749098, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5927568927858111, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06328952312469482, + "seed": "747260437/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.05181918665766716, + "selected_utility_gain_over_base": -1.9481808133423328, + "selector_regret": 0.011470336467027664, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.21537372469902039, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.6630434782608695, + "pairwise_causal_calibration_confidence": 0.5889473917896182, + "pairwise_causal_calibration_ece": 0.16711174361051992, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.12845068122756942, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train1", + "selected_candidate_index": 1, + "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": 0.3663750360591518, + "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": 0.35833333333333334, + "pairwise_causal_calibration_confidence": 0.5720588559847943, + "pairwise_causal_calibration_ece": 0.24463158134758306, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8208688776506676, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "seed": "1549324805/train1", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07181540876626968, + "selected_utility_gain_over_base": -0.032699428498744965, + "selector_regret": 0.12710610777139664, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.6530612244897959, + "pairwise_causal_calibration_confidence": 0.6090277928253506, + "pairwise_causal_calibration_ece": 0.04403343166444537, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": 1.4922290708776522, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/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": 1.9280846938490868, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "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": 0.49230769230769234, + "pairwise_causal_calibration_confidence": 0.5481079742764919, + "pairwise_causal_calibration_ece": 0.055800281968799464, + "pairwise_causal_calibration_pairs": 65.0, + "predicted_margin": 0.6538089264105622, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8829270005226135, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5341777572159363, + "pairwise_causal_calibration_ece": 0.0008444238826029737, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 1.4913443146002, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train1", + "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": 1.311950902009979, + "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": 0.7058823529411765, + "pairwise_causal_calibration_confidence": 0.5257069664610932, + "pairwise_causal_calibration_ece": 0.18017538648008335, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.4072117500373107, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/train1", + "selected_candidate_index": 3, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.89352947473526, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.3508771929824561, + "pairwise_causal_calibration_confidence": 0.5176749012095792, + "pairwise_causal_calibration_ece": 0.16679770822712303, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 1.622658467644122, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10585354268550873, + "selected_utility_gain_over_base": -0.01891392469406128, + "selector_regret": 1.8941464573144913, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.509090909090909, + "pairwise_causal_calibration_confidence": 0.5680754050042826, + "pairwise_causal_calibration_ece": 0.05898449591337354, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.0083103168677, + "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": 1.311950902009979, + "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": 0.5434550291857321, + "pairwise_causal_calibration_ece": 0.1537570975036694, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.9268668914168288, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/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": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.5798319327731093, + "pairwise_causal_calibration_confidence": 0.5569826639894806, + "pairwise_causal_calibration_ece": 0.02590153431562172, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.6034012607502497, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/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": 1.9129659235477448, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.3888888888888889, + "pairwise_causal_calibration_confidence": 0.5699891820408323, + "pairwise_causal_calibration_ece": 0.1811002931519435, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.9728419884347872, + "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": 0.05225050449371338, + "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": 1.4393298012330682, + "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": 0.6276117908365466, + "pairwise_causal_calibration_ece": 0.15419473151003862, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 1.4832101206592103, + "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": 1.4393298012330682, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5714411258007887, + "pairwise_causal_calibration_ece": 0.1327098335615991, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.428095135833086, + "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": 1.4393298012330682, + "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": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5793795802536641, + "pairwise_causal_calibration_ece": 0.12483412570820954, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.5329062844260477, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.1366594284772873, + "selected_utility_gain_over_base": -0.6874770671129227, + "selector_regret": 1.8633405715227127, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 1.4393298012330682, + "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": 0.25396825396825395, + "pairwise_causal_calibration_confidence": 0.5436154155208672, + "pairwise_causal_calibration_ece": 0.2896471615526132, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": 0.8299756115727561, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train1", + "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": 1.4393298012330682, + "train_seed": "1" + }, + { + "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": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.5692105391837949, + "pairwise_causal_calibration_ece": 0.08786394286037044, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.3428657851681152, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "seed": "211245842/train1", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02204875648021698, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7870428264141083, + "task_id": "StackCube-v1", + "tau": 1.4393298012330682, + "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": 0.37142857142857144, + "pairwise_causal_calibration_confidence": 0.5683151855386959, + "pairwise_causal_calibration_ece": 0.19688661411012442, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.0689882649948501, + "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": 1.4393298012330682, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5725133632951149, + "pairwise_causal_calibration_ece": 0.05201497258156908, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.0393222781022331, + "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": 1.4393298012330682, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "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": 0.5347371571140561, + "pairwise_causal_calibration_ece": 0.060500938124039116, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.7189098900525153, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/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": 1.4393298012330682, + "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": 0.8461538461538461, + "pairwise_causal_calibration_confidence": 0.5524013537256702, + "pairwise_causal_calibration_ece": 0.2937524924281759, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": 0.9728013354434921, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/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": 1.4393298012330682, + "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": 0.7428571428571429, + "pairwise_causal_calibration_confidence": 0.5308223915074378, + "pairwise_causal_calibration_ece": 0.21203475134970512, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.9255492206908755, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train1", + "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": 1.4393298012330682, + "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": 0.7282608695652174, + "pairwise_causal_calibration_confidence": 0.5365603796062945, + "pairwise_causal_calibration_ece": 0.19170048995892286, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.7477936348864276, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train2", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.2763946652412415, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "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": 0.4175824175824176, + "pairwise_causal_calibration_confidence": 0.589313207577152, + "pairwise_causal_calibration_ece": 0.1717307899947341, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 1.1349062127057792, + "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": 0.22035211324691772, + "selected_utility_gain_over_base": -0.379738450050354, + "selector_regret": 0.009861335158348083, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7697865515947342, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "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": 0.6428571428571429, + "pairwise_causal_calibration_confidence": 0.5451426904730965, + "pairwise_causal_calibration_ece": 0.09771445238404626, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.5303575557953266, + "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": 0.49469300263486815, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "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": 0.55, + "pairwise_causal_calibration_confidence": 0.5597255439272126, + "pairwise_causal_calibration_ece": 0.029011213993528073, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3122233671552621, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7117276191711426, + "seed": "608016290/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.725701630115509, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "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": 0.4909090909090909, + "pairwise_causal_calibration_confidence": 0.6088372831949117, + "pairwise_causal_calibration_ece": 0.20077855305415088, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 1.0158360314379022, + "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": 0.7130289077758789, + "selected_utility_gain_over_base": 0.013073742389678955, + "selector_regret": 0.006105959415435791, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "task_id": "LiftPegUpright-v1", + "tau": 0.49469300263486815, + "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": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5874115349169855, + "pairwise_causal_calibration_ece": 0.11518931269476336, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.4801934854321459, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14937306940555573, + "seed": "1522272686/train2", + "selected_candidate_index": 7, + "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": 0.8901452015560949, + "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": 0.7495662010387569, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/train2", + "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": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "train_seed": "2" + }, + { + "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": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.534001150570229, + "pairwise_causal_calibration_ece": 0.0794442275810315, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.40711517481418724, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8383140712976456, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.4083333333333333, + "pairwise_causal_calibration_confidence": 0.527311702364046, + "pairwise_causal_calibration_ece": 0.11897836903071268, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.14534023869673263, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "seed": "1461069713/train2", + "selected_candidate_index": 5, + "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": 0.3663750360591518, + "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": 0.7833333333333333, + "pairwise_causal_calibration_confidence": 0.6919635004574665, + "pairwise_causal_calibration_ece": 0.15399256329494126, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.3682760396231402, + "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": 0.017679253593087196, + "selected_utility_gain_over_base": 0.015749751939438283, + "selector_regret": 0.02117268182337284, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.21456877514719963, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.5630252100840336, + "pairwise_causal_calibration_confidence": 0.585134673600266, + "pairwise_causal_calibration_ece": 0.07742887022115333, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.9016945871378184, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.10418763011693954, + "selected_utility_gain_over_base": -1.8958123698830605, + "selector_regret": 1.8958123698830605, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "train_seed": "2" + }, + { + "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": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.6358538481639073, + "pairwise_causal_calibration_ece": 0.30252051483057396, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": 1.5731937235091635, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train2", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.08514729887247086, + "selected_utility_gain_over_base": -1.9148527011275291, + "selector_regret": 1.9148527011275291, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.3, + "pairwise_causal_calibration_confidence": 0.596820997580881, + "pairwise_causal_calibration_ece": 0.31958408189904103, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.15783058549316564, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21886132657527924, + "seed": "767747424/train2", + "selected_candidate_index": 1, + "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": 0.3663750360591518, + "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": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.6451490796833242, + "pairwise_causal_calibration_ece": 0.18877975835907373, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0549404942953586, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2718978226184845, + "seed": "278008524/train2", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14274199306964874, + "selected_utility_gain_over_base": 0.10685568675398827, + "selector_regret": 0.12915582954883575, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.72, + "pairwise_causal_calibration_confidence": 0.5506543992540976, + "pairwise_causal_calibration_ece": 0.16934560074590232, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": 1.3688073876119233, + "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": 0.3663750360591518, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5835775699590143, + "pairwise_causal_calibration_ece": 0.08357756995901419, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7315728847619027, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05094657465815544, + "seed": "626486813/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.016867956146597862, + "selected_utility_gain_over_base": 0.008600772358477116, + "selector_regret": 0.03407861851155758, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.026797477155923843, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.5726495726495726, + "pairwise_causal_calibration_confidence": 0.53232955534045, + "pairwise_causal_calibration_ece": 0.0403200173091226, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.3418595665842563, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "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": 0.3663750360591518, + "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": 0.6181818181818182, + "pairwise_causal_calibration_confidence": 0.5698793670685108, + "pairwise_causal_calibration_ece": 0.04830245111330746, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.0810514424099793, + "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": 0.3663750360591518, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5761565714740622, + "pairwise_causal_calibration_ece": 0.15948990480739544, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4503895097168673, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058951716870069504, + "seed": "141984615/train2", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.025202473625540733, + "selected_utility_gain_over_base": -0.07763177342712879, + "selector_regret": 0.03374924324452877, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.55, + "pairwise_causal_calibration_confidence": 0.5397172086215737, + "pairwise_causal_calibration_ece": 0.010282791378426288, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.1503265444695998, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train2", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12231997400522232, + "selected_utility_gain_over_base": -0.020388983190059662, + "selector_regret": 1.8776800259947777, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.4583333333333333, + "pairwise_causal_calibration_confidence": 0.5181751903900925, + "pairwise_causal_calibration_ece": 0.05984185705675921, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.531328083388189, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0756516233086586, + "seed": "636680137/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.015303333289921284, + "selected_utility_gain_over_base": -0.06783171277493238, + "selector_regret": 0.060348290018737316, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09491270035505295, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.5583333333333333, + "pairwise_causal_calibration_confidence": 0.6249243743560945, + "pairwise_causal_calibration_ece": 0.17421772277691577, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9201943816824425, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04980705305933952, + "seed": "767747424/train2", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0314757414162159, + "selected_utility_gain_over_base": -0.022647202014923096, + "selector_regret": 0.018331311643123627, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14730047062039375, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.5676443863088845, + "pairwise_causal_calibration_ece": 0.18712668693111412, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.34776849268222776, + "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": 0.3663750360591518, + "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": 0.6127037139516429, + "pairwise_causal_calibration_ece": 0.20866330991123888, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 1.5465686912481509, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/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": 0.3663750360591518, + "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": 0.475, + "pairwise_causal_calibration_confidence": 0.5450385226353394, + "pairwise_causal_calibration_ece": 0.07393992471005516, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8340847040373909, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.051346004009246826, + "seed": "750448296/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.03456125780940056, + "selected_utility_gain_over_base": -0.058436233550310135, + "selector_regret": 0.016784746199846268, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1726434826850891, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 1.474390672677842, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "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": 0.3663750360591518, + "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": 0.7333333333333333, + "pairwise_causal_calibration_confidence": 0.5528896350884385, + "pairwise_causal_calibration_ece": 0.18044369824489515, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8567886898938823, + "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": 0.3663750360591518, + "train_seed": "2" + }, + { + "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": 0.6060606060606061, + "pairwise_causal_calibration_confidence": 0.5909489391258794, + "pairwise_causal_calibration_ece": 0.048165522914242594, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.42430446773174424, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train2", + "selected_candidate_index": 15, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9504163265228271, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.35833333333333334, + "pairwise_causal_calibration_confidence": 0.5710713293995606, + "pairwise_causal_calibration_ece": 0.2127379960662275, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.420225409082441, + "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": 0.3663750360591518, + "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": 0.425, + "pairwise_causal_calibration_confidence": 0.5868090410812485, + "pairwise_causal_calibration_ece": 0.16180904108124847, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 1.5163699975204636, + "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": 0.0730673298239708, + "selected_utility_gain_over_base": 0.0011520236730575562, + "selector_regret": 1.9269326701760292, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "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": 0.40476190476190477, + "pairwise_causal_calibration_confidence": 0.5608779912997348, + "pairwise_causal_calibration_ece": 0.15611608653783016, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.5146204567061616, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train2", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10672565549612045, + "selected_utility_gain_over_base": -0.010347343981266022, + "selector_regret": 1.8932743445038795, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.3663750360591518, + "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": 0.46296296296296297, + "pairwise_causal_calibration_confidence": 0.5311030945024883, + "pairwise_causal_calibration_ece": 0.0681401315395253, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.424277496369104, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train2", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.527876777683192, + "pairwise_causal_calibration_ece": 0.017577767771353447, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.8877017595011019, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/train2", + "selected_candidate_index": 2, + "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": 1.311950902009979, + "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": 0.4636363636363636, + "pairwise_causal_calibration_confidence": 0.5178387112441525, + "pairwise_causal_calibration_ece": 0.05420234760778886, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.4403189606632267, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.1341685652732849, + "selected_utility_gain_over_base": 0.009401097893714905, + "selector_regret": 1.865831434726715, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.6309523809523809, + "pairwise_causal_calibration_confidence": 0.5619754350016658, + "pairwise_causal_calibration_ece": 0.068976945950715, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.0993338824963033, + "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": 1.311950902009979, + "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": 0.5520785359696122, + "pairwise_causal_calibration_ece": 0.1334130916939903, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.7820896027638113, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 5.424022674560547e-06, + "seed": "509430392/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.682209014892578e-07, + "selected_utility_gain_over_base": 4.470348358154297e-08, + "selector_regret": 5.155801773071289e-06, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9999945759773254, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "chart_id": "ms-d4c37ed44fb338c1d551", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5966386554621849, + "pairwise_causal_calibration_confidence": 0.5507966193765133, + "pairwise_causal_calibration_ece": 0.04584203608567172, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.300578618886508, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08703407645225525, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9129659235477448, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 1.311950902009979, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.5914469514823854, + "pairwise_causal_calibration_ece": 0.14700250703794093, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.8245519344630416, + "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": 1.4393298012330682, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "chart_id": "ms-15fc8cc6f74cf589056f", + "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": 0.6012392567929711, + "pairwise_causal_calibration_ece": 0.18457259012630453, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 1.3520359199432512, + "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": 0.16300667822360992, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.05146662890911102, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.785526692867279, + "task_id": "StackCube-v1", + "tau": 1.4393298012330682, + "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": 0.75, + "pairwise_causal_calibration_confidence": 0.5817900691636045, + "pairwise_causal_calibration_ece": 0.16820993083639552, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.32091844016441584, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "seed": "1113371261/train2", + "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": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "task_id": "StackCube-v1", + "tau": 1.4393298012330682, + "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": 0.4957983193277311, + "pairwise_causal_calibration_confidence": 0.5773391698248656, + "pairwise_causal_calibration_ece": 0.08154085049713428, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.4783739869715653, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15504932403564453, + "selected_utility_gain_over_base": -0.6690871715545654, + "selector_regret": 1.8449506759643555, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 1.4393298012330682, + "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": 0.5555555555555556, + "pairwise_causal_calibration_confidence": 0.5426912569382093, + "pairwise_causal_calibration_ece": 0.02770067302918524, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.8506856511161374, + "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": 1.4393298012330682, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5702656129330281, + "pairwise_causal_calibration_ece": 0.07026561293302824, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.2817098708877097, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "seed": "211245842/train2", + "selected_candidate_index": 6, + "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": 1.4393298012330682, + "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": 0.4380952380952381, + "pairwise_causal_calibration_confidence": 0.565958352249713, + "pairwise_causal_calibration_ece": 0.1278631141544748, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.0039423360080715, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "seed": "1151410346/train2", + "selected_candidate_index": 4, + "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": 1.4393298012330682, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5840738619592613, + "pairwise_causal_calibration_ece": 0.1404771894262721, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3115767566062857, + "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": 0.09220491349697113, + "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": 1.4393298012330682, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "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": 0.5740740740740741, + "pairwise_causal_calibration_confidence": 0.5391779065928842, + "pairwise_causal_calibration_ece": 0.05737097122205889, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.841493235581737, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/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": 1.4393298012330682, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5411176357911663, + "pairwise_causal_calibration_ece": 0.04111763579116634, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.0057266984542679, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/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": "StackCube-v1", + "tau": 1.4393298012330682, + "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": 0.3684210526315789, + "pairwise_causal_calibration_confidence": 0.528258651563379, + "pairwise_causal_calibration_ece": 0.15983759893180005, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": 1.0109377403715456, + "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": 1.4393298012330682, + "train_seed": "2" + } + ], + "schema_version": 1, + "selected_lambda": 0.01, + "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": 2.0, + "summary": { + "base_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.38958333333333334, + "low": 0.22083333333333333, + "mean": 0.30625, + "n": 120, + "num_samples": 300 + }, + "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": 0.3038787878787878, + "low": 0.09454545454545454, + "mean": 0.19987878787878788, + "n": 5, + "num_samples": 300 + }, + "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": 0.3680555555555556, + "low": 0.2152777777777778, + "mean": 0.2916666666666667, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "base_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.9549208872815751, + "low": 0.6291768466153169, + "mean": 0.7901180622977944, + "n": 120, + "num_samples": 300 + }, + "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": 0.879990507716313, + "low": 0.5455794048082704, + "mean": 0.695559797174015, + "n": 5, + "num_samples": 300 + }, + "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": 0.8711436447313948, + "low": 0.6016125785051069, + "mean": 0.7418195864423373, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "coverage": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.6666666666666666, + "low": 0.4791666666666667, + "mean": 0.5791666666666667, + "n": 120, + "num_samples": 300 + }, + "mean": 0.5791666666666667, + "per_seed": { + "1073161778/train0": 1.0, + "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": 0.0, + "1261930456/train1": 0.0, + "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": 1.0, + "141984615/train1": 1.0, + "141984615/train2": 1.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "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": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 0.0, + "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": 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": 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": 0.5, + "278008524/train1": 0.5, + "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": 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": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "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": 0.0, + "655884605/train1": 1.0, + "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": 1.0, + "750448296/train1": 1.0, + "750448296/train2": 1.0, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7653333333333333, + "low": 0.1808080808080808, + "mean": 0.4719595959595959, + "n": 5, + "num_samples": 300 + }, + "mean": 0.4719595959595959, + "per_task": { + "LiftPegUpright-v1": 0.7333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": 0.8133333333333334, + "PushCube-v1": 0.7222222222222222, + "StackCube-v1": 0.09090909090909091 + } + }, + "metric": "coverage", + "micro": { + "confidence": 0.95, + "high": 0.6944444444444444, + "low": 0.5347222222222222, + "mean": 0.6111111111111112, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "execute_generated": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.6666666666666666, + "low": 0.4791666666666667, + "mean": 0.5791666666666667, + "n": 120, + "num_samples": 300 + }, + "mean": 0.5791666666666667, + "per_seed": { + "1073161778/train0": 1.0, + "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": 0.0, + "1261930456/train1": 0.0, + "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": 1.0, + "141984615/train1": 1.0, + "141984615/train2": 1.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "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": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 0.0, + "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": 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": 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": 0.5, + "278008524/train1": 0.5, + "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": 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": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "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": 0.0, + "655884605/train1": 1.0, + "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": 1.0, + "750448296/train1": 1.0, + "750448296/train2": 1.0, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7653333333333333, + "low": 0.1808080808080808, + "mean": 0.4719595959595959, + "n": 5, + "num_samples": 300 + }, + "mean": 0.4719595959595959, + "per_task": { + "LiftPegUpright-v1": 0.7333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": 0.8133333333333334, + "PushCube-v1": 0.7222222222222222, + "StackCube-v1": 0.09090909090909091 + } + }, + "metric": "execute_generated", + "micro": { + "confidence": 0.95, + "high": 0.6944444444444444, + "low": 0.5347222222222222, + "mean": 0.6111111111111112, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "fallback_rate": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5208333333333334, + "low": 0.3333333333333333, + "mean": 0.42083333333333334, + "n": 120, + "num_samples": 300 + }, + "mean": 0.42083333333333334, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 1.0, + "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": 1.0, + "1261930456/train1": 1.0, + "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": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 1.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 1.0, + "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": 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": 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": 0.0, + "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": 0.5, + "278008524/train1": 0.5, + "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": 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": 0.0, + "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": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.8191919191919192, + "low": 0.23466666666666666, + "mean": 0.5280404040404041, + "n": 5, + "num_samples": 300 + }, + "mean": 0.5280404040404041, + "per_task": { + "LiftPegUpright-v1": 0.26666666666666666, + "PickCube-v1": 1.0, + "PullCube-v1": 0.18666666666666668, + "PushCube-v1": 0.2777777777777778, + "StackCube-v1": 0.9090909090909091 + } + }, + "metric": "fallback_rate", + "micro": { + "confidence": 0.95, + "high": 0.4652777777777778, + "low": 0.3055555555555556, + "mean": 0.3888888888888889, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "hidden_chart_oracle_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.8597222222222223, + "low": 0.7097222222222221, + "mean": 0.7854166666666668, + "n": 120, + "num_samples": 300 + }, + "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": 1.0, + "low": 0.6786666666666668, + "mean": 0.8706666666666667, + "n": 5, + "num_samples": 300 + }, + "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": 0.7986111111111112, + "low": 0.6597222222222222, + "mean": 0.7291666666666666, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "hidden_chart_oracle_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.7487430048278638, + "low": 1.487454852130678, + "mean": 1.6190625454609595, + "n": 120, + "num_samples": 300 + }, + "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": 2.0, + "low": 1.428388743619124, + "mean": 1.7696735892097153, + "n": 5, + "num_samples": 300 + }, + "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": 1.6414945029343169, + "low": 1.3965875044878986, + "mean": 1.5183161981403828, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "outcome_ptr": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.6180555555555555, + "low": 0.44861111111111107, + "mean": 0.5319444444444444, + "n": 120, + "num_samples": 300 + }, + "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": 0.7151515151515151, + "low": 0.23393939393939395, + "mean": 0.49393939393939396, + "n": 5, + "num_samples": 300 + }, + "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": 0.6388888888888888, + "low": 0.4722222222222222, + "mean": 0.5486111111111112, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "pairwise_causal_calibration_accuracy": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5578284570165397, + "low": 0.5089962643805661, + "mean": 0.5350816292923953, + "n": 117, + "num_samples": 300 + }, + "mean": 0.5350816292923953, + "per_seed": { + "1073161778/train0": 0.6386554621848739, + "1073161778/train1": 0.6833333333333333, + "1073161778/train2": 0.6428571428571429, + "1113371261/train0": 0.5357142857142857, + "1113371261/train1": 0.6071428571428571, + "1113371261/train2": 0.75, + "1151410346/train0": 0.5142857142857142, + "1151410346/train1": 0.37142857142857144, + "1151410346/train2": 0.4380952380952381, + "1261930456/train0": 0.46464646464646464, + "1261930456/train1": 0.509090909090909, + "1261930456/train2": 0.6309523809523809, + "1324133122/train0": 0.42857142857142855, + "1324133122/train1": 0.6071428571428571, + "1324133122/train2": 0.4642857142857143, + "1324642411/train0": 0.55, + "1324642411/train1": 0.7058823529411765, + "1324642411/train2": 0.5454545454545454, + "1375492786/train0": 0.64, + "1375492786/train1": 0.5263157894736842, + "1375492786/train2": 0.7380952380952381, + "141984615/train0": 0.31666666666666665, + "141984615/train1": 0.625, + "141984615/train2": 0.4166666666666667, + "1461069713/train0": 0.8166666666666667, + "1461069713/train1": 0.4666666666666667, + "1461069713/train2": 0.4083333333333333, + "1480856810/train0": 0.8080808080808081, + "1480856810/train1": 0.5798319327731093, + "1480856810/train2": 0.5966386554621849, + "1509980664/train0": 0.5222222222222223, + "1509980664/train1": 0.25396825396825395, + "1509980664/train2": 0.5555555555555556, + "1522272686/train0": 0.7222222222222222, + "1522272686/train1": 0.4722222222222222, + "1522272686/train2": 0.4722222222222222, + "1549324805/train0": 0.36666666666666664, + "1549324805/train1": 0.35833333333333334, + "1549324805/train2": 0.35833333333333334, + "1578990937/train0": 0.7094017094017094, + "1578990937/train1": 0.45454545454545453, + "1578990937/train2": 0.4957983193277311, + "1790188787/train0": 0.3, + "1790188787/train1": 0.3508771929824561, + "1790188787/train2": 0.4636363636363636, + "1822210471/train0": 0.6, + "1822210471/train1": 0.6530612244897959, + "1822210471/train2": 0.425, + "1845489013/train0": 0.6521739130434783, + "1845489013/train1": 0.6153846153846154, + "1845489013/train2": 0.72, + "1999022574/train0": 0.35714285714285715, + "1999022574/train1": 0.7428571428571429, + "1999022574/train2": 0.3684210526315789, + "2012211287/train0": 0.5546218487394958, + "2012211287/train1": 0.6083333333333333, + "2012211287/train2": 0.7282608695652174, + "2038594191/train0": 0.27586206896551724, + "2038594191/train1": 0.3148148148148148, + "2038594191/train2": 0.40404040404040403, + "2054899126/train0": 0.25, + "2054899126/train1": 0.6578947368421053, + "2054899126/train2": 0.5630252100840336, + "2061635888/train0": 0.5277777777777778, + "2061635888/train1": 0.3888888888888889, + "2061635888/train2": 0.4444444444444444, + "211245842/train0": 0.6218487394957983, + "211245842/train1": 0.6166666666666667, + "211245842/train2": 0.5, + "278008524/train0": 0.7226010101010101, + "278008524/train1": 0.6065217391304347, + "278008524/train2": 0.5363636363636364, + "307068879/train0": 0.5934065934065934, + "307068879/train1": 0.5384615384615384, + "307068879/train2": 0.4175824175824176, + "412102563/train0": 0.65, + "412102563/train1": 0.8461538461538461, + "412102563/train2": 0.5, + "440577711/train0": 0.6206896551724138, + "440577711/train1": 0.5333333333333333, + "440577711/train2": 0.46296296296296297, + "441046872/train0": 0.38181818181818183, + "441046872/train1": 0.6181818181818182, + "441046872/train2": 0.4909090909090909, + "509430392/train0": 0.6470588235294118, + "509430392/train1": 0.6, + "509430392/train2": 0.6470588235294118, + "585904130/train0": 0.30952380952380953, + "585904130/train1": 0.56, + "585904130/train2": 0.6181818181818182, + "608016290/train0": 0.5333333333333333, + "608016290/train1": 0.5523809523809524, + "608016290/train2": 0.55, + "626486813/train0": 0.6499999999999999, + "626486813/train1": 0.33333333333333337, + "626486813/train2": 0.6416666666666666, + "636680137/train0": 0.49166666666666664, + "636680137/train1": 0.475, + "636680137/train2": 0.4583333333333333, + "655884605/train0": 0.5555555555555556, + "655884605/train1": 0.5, + "655884605/train2": 0.4166666666666667, + "672328262/train0": 0.8823529411764706, + "672328262/train1": 0.45714285714285713, + "672328262/train2": 0.3333333333333333, + "701667486/train0": 0.2857142857142857, + "701667486/train1": 0.5952380952380952, + "701667486/train2": 0.5740740740740741, + "747260437/train0": 0.5166666666666667, + "747260437/train1": 0.675, + "747260437/train2": 0.7333333333333333, + "750448296/train0": 0.5857142857142857, + "750448296/train1": 0.4307692307692308, + "750448296/train2": 0.4765873015873015, + "767747424/train0": 0.49887820512820513, + "767747424/train1": 0.5168659771600949, + "767747424/train2": 0.5111070710335417 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5610369044795278, + "low": 0.5297194800901149, + "mean": 0.545468165643421, + "n": 5, + "num_samples": 300 + }, + "mean": 0.545468165643421, + "per_task": { + "LiftPegUpright-v1": 0.5688090610724882, + "PickCube-v1": 0.5555555555555555, + "PullCube-v1": 0.5253206716900769, + "PushCube-v1": 0.5564552891415517, + "StackCube-v1": 0.5212002507574325 + } + }, + "metric": "pairwise_causal_calibration_accuracy", + "micro": { + "confidence": 0.95, + "high": 0.5564154651873677, + "low": 0.5136261705086657, + "mean": 0.5337806696298005, + "n": 138, + "num_samples": 300 + }, + "n": 138 + }, + "pairwise_causal_calibration_confidence": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5729068871628751, + "low": 0.5607975783294857, + "mean": 0.566333866388919, + "n": 117, + "num_samples": 300 + }, + "mean": 0.566333866388919, + "per_seed": { + "1073161778/train0": 0.5373743568439672, + "1073161778/train1": 0.5322990000175534, + "1073161778/train2": 0.5451426904730965, + "1113371261/train0": 0.5701414698923739, + "1113371261/train1": 0.5714411258007887, + "1113371261/train2": 0.5817900691636045, + "1151410346/train0": 0.5667135771731647, + "1151410346/train1": 0.5683151855386959, + "1151410346/train2": 0.565958352249713, + "1261930456/train0": 0.5719896658662266, + "1261930456/train1": 0.5680754050042826, + "1261930456/train2": 0.5619754350016658, + "1324133122/train0": 0.5755791782863937, + "1324133122/train1": 0.5725133632951149, + "1324133122/train2": 0.5840738619592613, + "1324642411/train0": 0.5260274890742292, + "1324642411/train1": 0.5257069664610932, + "1324642411/train2": 0.527876777683192, + "1375492786/train0": 0.5585290828327523, + "1375492786/train1": 0.5610084990880762, + "1375492786/train2": 0.5676443863088845, + "141984615/train0": 0.563739392345968, + "141984615/train1": 0.581566420974051, + "141984615/train2": 0.5761565714740622, + "1461069713/train0": 0.5411312415303838, + "1461069713/train1": 0.5331472452449045, + "1461069713/train2": 0.527311702364046, + "1480856810/train0": 0.5911741147277916, + "1480856810/train1": 0.5569826639894806, + "1480856810/train2": 0.5507966193765133, + "1509980664/train0": 0.5411679300565454, + "1509980664/train1": 0.5436154155208672, + "1509980664/train2": 0.5426912569382093, + "1522272686/train0": 0.5639487552532058, + "1522272686/train1": 0.5732579849122883, + "1522272686/train2": 0.5874115349169855, + "1549324805/train0": 0.5627369782509959, + "1549324805/train1": 0.5720588559847943, + "1549324805/train2": 0.5710713293995606, + "1578990937/train0": 0.5852719048004434, + "1578990937/train1": 0.5793795802536641, + "1578990937/train2": 0.5773391698248656, + "1790188787/train0": 0.519098923420412, + "1790188787/train1": 0.5176749012095792, + "1790188787/train2": 0.5178387112441525, + "1822210471/train0": 0.5864304711853732, + "1822210471/train1": 0.6090277928253506, + "1822210471/train2": 0.5868090410812485, + "1845489013/train0": 0.5356510938834426, + "1845489013/train1": 0.5374675838313216, + "1845489013/train2": 0.5506543992540976, + "1999022574/train0": 0.5351557804986392, + "1999022574/train1": 0.5308223915074378, + "1999022574/train2": 0.528258651563379, + "2012211287/train0": 0.5281524186595942, + "2012211287/train1": 0.5310859795550439, + "2012211287/train2": 0.5365603796062945, + "2038594191/train0": 0.6084782956989382, + "2038594191/train1": 0.6260111821626294, + "2038594191/train2": 0.6127037139516429, + "2054899126/train0": 0.566620580797147, + "2054899126/train1": 0.563744413130429, + "2054899126/train2": 0.585134673600266, + "2061635888/train0": 0.5745387895297428, + "2061635888/train1": 0.5699891820408323, + "2061635888/train2": 0.5914469514823854, + "211245842/train0": 0.5746201886713925, + "211245842/train1": 0.5692105391837949, + "211245842/train2": 0.5702656129330281, + "278008524/train0": 0.6163446579821468, + "278008524/train1": 0.6113953659702027, + "278008524/train2": 0.6180490094046018, + "307068879/train0": 0.5944461291750907, + "307068879/train1": 0.5937223391260118, + "307068879/train2": 0.589313207577152, + "412102563/train0": 0.5385704504068652, + "412102563/train1": 0.5524013537256702, + "412102563/train2": 0.5411176357911663, + "440577711/train0": 0.5455151446938427, + "440577711/train1": 0.5341777572159363, + "440577711/train2": 0.5311030945024883, + "441046872/train0": 0.6008428753783649, + "441046872/train1": 0.6026494132350357, + "441046872/train2": 0.6088372831949117, + "509430392/train0": 0.5508070209680679, + "509430392/train1": 0.5434550291857321, + "509430392/train2": 0.5520785359696122, + "585904130/train0": 0.5734358732826558, + "585904130/train1": 0.5673349851937346, + "585904130/train2": 0.5698793670685108, + "608016290/train0": 0.5478322151873722, + "608016290/train1": 0.5534688429350546, + "608016290/train2": 0.5597255439272126, + "626486813/train0": 0.6373243545977667, + "626486813/train1": 0.6287134543731017, + "626486813/train2": 0.6377705352082403, + "636680137/train0": 0.529875300645354, + "636680137/train1": 0.5159628167606266, + "636680137/train2": 0.5181751903900925, + "655884605/train0": 0.6048190953611235, + "655884605/train1": 0.6276117908365466, + "655884605/train2": 0.6012392567929711, + "672328262/train0": 0.637846163378886, + "672328262/train1": 0.629165943748097, + "672328262/train2": 0.6358538481639073, + "701667486/train0": 0.5376448138211073, + "701667486/train1": 0.5347371571140561, + "701667486/train2": 0.5391779065928842, + "747260437/train0": 0.555978759483347, + "747260437/train1": 0.5687872987376212, + "747260437/train2": 0.5528896350884385, + "750448296/train0": 0.5477839571124287, + "750448296/train1": 0.5465796418867024, + "750448296/train2": 0.5485445741855494, + "767747424/train0": 0.5657585105530328, + "767747424/train1": 0.5647149634509379, + "767747424/train2": 0.5720190194619137 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5736887848240381, + "low": 0.554205729190532, + "mean": 0.56491201755332, + "n": 5, + "num_samples": 300 + }, + "mean": 0.56491201755332, + "per_task": { + "LiftPegUpright-v1": 0.5640968449927838, + "PickCube-v1": 0.5748727583608265, + "PullCube-v1": 0.5764914402555266, + "PushCube-v1": 0.5440196808663499, + "StackCube-v1": 0.5650793632911131 + } + }, + "metric": "pairwise_causal_calibration_confidence", + "micro": { + "confidence": 0.95, + "high": 0.5748534101753632, + "low": 0.5622579599806611, + "mean": 0.5681445911000474, + "n": 138, + "num_samples": 300 + }, + "n": 138 + }, + "pairwise_causal_calibration_ece": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.1386958105817993, + "low": 0.1109564386070653, + "mean": 0.12511151754183172, + "n": 120, + "num_samples": 300 + }, + "mean": 0.12511151754183172, + "per_seed": { + "1073161778/train0": 0.10128110534090673, + "1073161778/train1": 0.15103433331577998, + "1073161778/train2": 0.09771445238404626, + "1113371261/train0": 0.06599981030811651, + "1113371261/train1": 0.1327098335615991, + "1113371261/train2": 0.16820993083639552, + "1151410346/train0": 0.05242786288745051, + "1151410346/train1": 0.19688661411012442, + "1151410346/train2": 0.1278631141544748, + "1261930456/train0": 0.12912993334954873, + "1261930456/train1": 0.05898449591337354, + "1261930456/train2": 0.068976945950715, + "1324133122/train0": 0.147007749714965, + "1324133122/train1": 0.05201497258156908, + "1324133122/train2": 0.1404771894262721, + "1324642411/train0": 0.023972510925770796, + "1324642411/train1": 0.18017538648008335, + "1324642411/train2": 0.017577767771353447, + "1375492786/train0": 0.08147091716724786, + "1375492786/train1": 0.03469270961439179, + "1375492786/train2": 0.18712668693111412, + "141984615/train0": 0.24707272567930166, + "141984615/train1": 0.10416729357646315, + "141984615/train2": 0.15948990480739544, + "1461069713/train0": 0.2755354251362828, + "1461069713/train1": 0.06648057857823786, + "1461069713/train2": 0.11897836903071268, + "1480856810/train0": 0.21690669335301668, + "1480856810/train1": 0.02590153431562172, + "1480856810/train2": 0.04584203608567172, + "1509980664/train0": 0.018945707834323134, + "1509980664/train1": 0.2896471615526132, + "1509980664/train2": 0.02770067302918524, + "1522272686/train0": 0.15827346696901654, + "1522272686/train1": 0.10103576269006627, + "1522272686/train2": 0.11518931269476336, + "1549324805/train0": 0.19607031158432947, + "1549324805/train1": 0.24463158134758306, + "1549324805/train2": 0.2127379960662275, + "1578990937/train0": 0.12412980460126585, + "1578990937/train1": 0.12483412570820954, + "1578990937/train2": 0.08154085049713428, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.21909892342041198, + "1790188787/train1": 0.16679770822712303, + "1790188787/train2": 0.05420234760778886, + "1822210471/train0": 0.08360394561276052, + "1822210471/train1": 0.04403343166444537, + "1822210471/train2": 0.16180904108124847, + "1845489013/train0": 0.1165228191600357, + "1845489013/train1": 0.07791703155329377, + "1845489013/train2": 0.16934560074590232, + "1999022574/train0": 0.17801292335578203, + "1999022574/train1": 0.21203475134970512, + "1999022574/train2": 0.15983759893180005, + "2012211287/train0": 0.026469430079901657, + "2012211287/train1": 0.0772473537782894, + "2012211287/train2": 0.19170048995892286, + "2038594191/train0": 0.33261622673342073, + "2038594191/train1": 0.39213662401689425, + "2038594191/train2": 0.20866330991123888, + "2054899126/train0": 0.15831029039857342, + "2054899126/train1": 0.0470751618558381, + "2054899126/train2": 0.038714435110576664, + "2061635888/train0": 0.04676101175196509, + "2061635888/train1": 0.1811002931519435, + "2061635888/train2": 0.14700250703794093, + "211245842/train0": 0.04722855082440577, + "211245842/train1": 0.08786394286037044, + "211245842/train2": 0.07026561293302824, + "278008524/train0": 0.10625635211886339, + "278008524/train1": 0.15188563211313605, + "278008524/train2": 0.11847264063665816, + "307068879/train0": 0.12040324822018197, + "307068879/train1": 0.0607943275756009, + "307068879/train2": 0.1717307899947341, + "412102563/train0": 0.11142954959313478, + "412102563/train1": 0.2937524924281759, + "412102563/train2": 0.04111763579116634, + "440577711/train0": 0.07517451047857106, + "440577711/train1": 0.0008444238826029737, + "440577711/train2": 0.0681401315395253, + "441046872/train0": 0.21902469356018284, + "441046872/train1": 0.04900842512659715, + "441046872/train2": 0.20077855305415088, + "509430392/train0": 0.09625180256134375, + "509430392/train1": 0.1537570975036694, + "509430392/train2": 0.1334130916939903, + "585904130/train0": 0.2639120637588461, + "585904130/train1": 0.17050541777278239, + "585904130/train2": 0.04830245111330746, + "608016290/train0": 0.04803994731985866, + "608016290/train1": 0.03000164602114203, + "608016290/train2": 0.029011213993528073, + "626486813/train0": 0.10601407456772449, + "626486813/train1": 0.2953801210397682, + "626486813/train2": 0.11878506662697773, + "636680137/train0": 0.04450430387060281, + "636680137/train1": 0.04096281676062663, + "636680137/train2": 0.05984185705675921, + "655884605/train0": 0.08237033905417408, + "655884605/train1": 0.15419473151003862, + "655884605/train2": 0.18457259012630453, + "672328262/train0": 0.24450677779758456, + "672328262/train1": 0.1720230866052398, + "672328262/train2": 0.30252051483057396, + "701667486/train0": 0.2519305281068216, + "701667486/train1": 0.060500938124039116, + "701667486/train2": 0.05737097122205889, + "747260437/train0": 0.04733473753037369, + "747260437/train1": 0.2478180852749098, + "747260437/train2": 0.18044369824489515, + "750448296/train0": 0.07643574626949645, + "750448296/train1": 0.11581041111747163, + "750448296/train2": 0.08011293420877054, + "767747424/train0": 0.18436261086486025, + "767747424/train1": 0.12296851299012063, + "767747424/train2": 0.1533915123915277 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.12970277798794536, + "low": 0.10379468098858013, + "mean": 0.11760395543608722, + "n": 5, + "num_samples": 300 + }, + "mean": 0.11760395543608722, + "per_task": { + "LiftPegUpright-v1": 0.10494933398158825, + "PickCube-v1": 0.12483284745128205, + "PullCube-v1": 0.13706044945766013, + "PushCube-v1": 0.09639707450334344, + "StackCube-v1": 0.12478007178656221 + } + }, + "metric": "pairwise_causal_calibration_ece", + "micro": { + "confidence": 0.95, + "high": 0.14253205323029616, + "low": 0.11178952992711648, + "mean": 0.12556362480152028, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "pairwise_causal_calibration_pairs": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 88.32847222222222, + "low": 75.17222222222223, + "mean": 82.21180555555556, + "n": 120, + "num_samples": 300 + }, + "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": 95.32888888888888, + "low": 54.05090909090909, + "mean": 77.60379797979797, + "n": 5, + "num_samples": 300 + }, + "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": 92.36111111111111, + "low": 80.19444444444444, + "mean": 86.34722222222223, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "predicted_margin": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.018094850319079, + "low": 0.8291033309118777, + "mean": 0.9314191444954871, + "n": 120, + "num_samples": 300 + }, + "mean": 0.9314191444954871, + "per_seed": { + "1073161778/train0": 0.5566626770858889, + "1073161778/train1": 0.3872764726300346, + "1073161778/train2": 0.5303575557953266, + "1113371261/train0": 0.4419532287444907, + "1113371261/train1": 0.428095135833086, + "1113371261/train2": 0.32091844016441584, + "1151410346/train0": 1.024503811643319, + "1151410346/train1": 1.0689882649948501, + "1151410346/train2": 1.0039423360080715, + "1261930456/train0": 1.1198710760724104, + "1261930456/train1": 1.0083103168677, + "1261930456/train2": 1.0993338824963033, + "1324133122/train0": -0.36820169505830336, + "1324133122/train1": 0.0393222781022331, + "1324133122/train2": -0.3115767566062857, + "1324642411/train0": 1.8008980492120834, + "1324642411/train1": 1.4072117500373107, + "1324642411/train2": 0.8877017595011019, + "1375492786/train0": 0.0461275740764826, + "1375492786/train1": -0.45720538390197873, + "1375492786/train2": -0.34776849268222776, + "141984615/train0": 0.6540918192032681, + "141984615/train1": 0.8936304139520725, + "141984615/train2": 0.4503895097168673, + "1461069713/train0": -0.3670742388145669, + "1461069713/train1": -0.16101647169097077, + "1461069713/train2": -0.14534023869673263, + "1480856810/train0": 2.149778411340016, + "1480856810/train1": 1.6034012607502497, + "1480856810/train2": 1.300578618886508, + "1509980664/train0": 0.7690911091840036, + "1509980664/train1": 0.8299756115727561, + "1509980664/train2": 0.8506856511161374, + "1522272686/train0": 0.3519897595066688, + "1522272686/train1": 0.7344242898918623, + "1522272686/train2": 0.4801934854321459, + "1549324805/train0": 1.171258504104145, + "1549324805/train1": 0.8208688776506676, + "1549324805/train2": 1.420225409082441, + "1578990937/train0": 1.369766946850774, + "1578990937/train1": 1.5329062844260477, + "1578990937/train2": 1.4783739869715653, + "1747178030/train0": 0.7045359481010607, + "1747178030/train1": 0.6796904941310151, + "1747178030/train2": 0.7495662010387569, + "1790188787/train0": 1.5993442072252648, + "1790188787/train1": 1.622658467644122, + "1790188787/train2": 1.4403189606632267, + "1822210471/train0": 1.5192545920029907, + "1822210471/train1": 1.4922290708776522, + "1822210471/train2": 1.5163699975204636, + "1845489013/train0": 1.3305286793016613, + "1845489013/train1": 1.412757419493385, + "1845489013/train2": 1.3688073876119233, + "1999022574/train0": 0.9988148345478187, + "1999022574/train1": 0.9255492206908755, + "1999022574/train2": 1.0109377403715456, + "2012211287/train0": 0.7217723168310286, + "2012211287/train1": 0.5565021996855251, + "2012211287/train2": 0.7477936348864276, + "2038594191/train0": 1.5405611437535534, + "2038594191/train1": 1.2016013806419839, + "2038594191/train2": 1.5465686912481509, + "2054899126/train0": 1.0852982788395877, + "2054899126/train1": 1.086209527289868, + "2054899126/train2": 1.1880426299078302, + "2061635888/train0": 1.0205351759283516, + "2061635888/train1": 0.9728419884347872, + "2061635888/train2": 0.8245519344630416, + "211245842/train0": 1.2882788003210806, + "211245842/train1": 1.3428657851681152, + "211245842/train2": 1.2817098708877097, + "278008524/train0": 0.5507130046793893, + "278008524/train1": 0.6145792187233674, + "278008524/train2": 0.7396224810135514, + "307068879/train0": 1.0739910273739994, + "307068879/train1": 0.6874983921047111, + "307068879/train2": 1.1349062127057792, + "412102563/train0": 1.0933587971168492, + "412102563/train1": 0.9728013354434921, + "412102563/train2": 1.0057266984542679, + "440577711/train0": 1.5732830552633406, + "440577711/train1": 1.4913443146002, + "440577711/train2": 1.424277496369104, + "441046872/train0": 1.1052022133857848, + "441046872/train1": 0.6838040349024017, + "441046872/train2": 1.0158360314379022, + "509430392/train0": 2.08491547386598, + "509430392/train1": 1.9268668914168288, + "509430392/train2": 1.7820896027638113, + "585904130/train0": 0.937964471343637, + "585904130/train1": 0.867255607957766, + "585904130/train2": 1.0810514424099793, + "608016290/train0": 0.34122628210022654, + "608016290/train1": 0.1770980989610429, + "608016290/train2": 0.3122233671552621, + "626486813/train0": 1.205283676715192, + "626486813/train1": 1.2030969286900701, + "626486813/train2": 1.0499244621925214, + "636680137/train0": 1.0013727088350777, + "636680137/train1": 0.40455029586141933, + "636680137/train2": 0.531328083388189, + "655884605/train0": 1.095636223788052, + "655884605/train1": 1.4832101206592103, + "655884605/train2": 1.3520359199432512, + "672328262/train0": 1.7549902847176035, + "672328262/train1": 1.7117888933582819, + "672328262/train2": 1.5731937235091635, + "701667486/train0": 0.6729506809560988, + "701667486/train1": 0.7189098900525153, + "701667486/train2": 0.841493235581737, + "747260437/train0": 0.6951613484301653, + "747260437/train1": 0.5927568927858111, + "747260437/train2": 0.8567886898938823, + "750448296/train0": 1.0438240923289683, + "750448296/train1": 0.7480058669656788, + "750448296/train2": 0.8330105684043841, + "767747424/train0": 0.3010322890243957, + "767747424/train1": 0.3930644164701522, + "767747424/train2": 0.3778346343969301 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.2225678411915908, + "low": 0.6531389496686, + "mean": 0.8815997696403448, + "n": 5, + "num_samples": 300 + }, + "mean": 0.8815997696403448, + "per_task": { + "LiftPegUpright-v1": 0.6688100344694228, + "PickCube-v1": 0.5222025116102257, + "PullCube-v1": 0.8087553078530508, + "PushCube-v1": 1.5178990886097534, + "StackCube-v1": 0.8903319056592717 + } + }, + "metric": "predicted_margin", + "micro": { + "confidence": 0.95, + "high": 0.9847565531698795, + "low": 0.8118648188702267, + "mean": 0.8955454348790438, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "proposal_oracle_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.673611111111111, + "low": 0.4958333333333333, + "mean": 0.5847222222222223, + "n": 120, + "num_samples": 300 + }, + "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": 0.7563636363636363, + "low": 0.22303030303030305, + "mean": 0.5074747474747474, + "n": 5, + "num_samples": 300 + }, + "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": 0.6458333333333334, + "low": 0.4861111111111111, + "mean": 0.5694444444444444, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "proposal_oracle_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.4087191828565362, + "low": 1.0967581196351806, + "mean": 1.251578581963097, + "n": 120, + "num_samples": 300 + }, + "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": 1.5757816767742236, + "low": 0.5943610860542816, + "mean": 1.1196794969856438, + "n": 5, + "num_samples": 300 + }, + "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": 1.3550500385204538, + "low": 1.0635484514318199, + "mean": 1.2159999172436073, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_candidate_index": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 6.372916666666667, + "low": 4.951388888888889, + "mean": 5.649305555555555, + "n": 120, + "num_samples": 300 + }, + "mean": 5.649305555555555, + "per_seed": { + "1073161778/train0": 6.0, + "1073161778/train1": 8.0, + "1073161778/train2": 6.0, + "1113371261/train0": 2.0, + "1113371261/train1": 5.0, + "1113371261/train2": 5.0, + "1151410346/train0": 6.0, + "1151410346/train1": 2.0, + "1151410346/train2": 4.0, + "1261930456/train0": 1.0, + "1261930456/train1": 2.0, + "1261930456/train2": 4.0, + "1324133122/train0": 6.0, + "1324133122/train1": 5.0, + "1324133122/train2": 7.0, + "1324642411/train0": 11.0, + "1324642411/train1": 3.0, + "1324642411/train2": 2.0, + "1375492786/train0": 3.0, + "1375492786/train1": 4.0, + "1375492786/train2": 2.0, + "141984615/train0": 2.0, + "141984615/train1": 5.0, + "141984615/train2": 10.0, + "1461069713/train0": 7.0, + "1461069713/train1": 1.0, + "1461069713/train2": 5.0, + "1480856810/train0": 0.0, + "1480856810/train1": 10.0, + "1480856810/train2": 4.0, + "1509980664/train0": 10.0, + "1509980664/train1": 9.0, + "1509980664/train2": 9.0, + "1522272686/train0": 8.0, + "1522272686/train1": 7.0, + "1522272686/train2": 7.0, + "1549324805/train0": 5.0, + "1549324805/train1": 10.0, + "1549324805/train2": 1.0, + "1578990937/train0": 12.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 9.0, + "1747178030/train1": 3.0, + "1747178030/train2": 15.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 3.0, + "1822210471/train0": 5.0, + "1822210471/train1": 2.0, + "1822210471/train2": 7.0, + "1845489013/train0": 5.0, + "1845489013/train1": 7.0, + "1845489013/train2": 8.0, + "1999022574/train0": 3.0, + "1999022574/train1": 5.0, + "1999022574/train2": 13.0, + "2012211287/train0": 7.0, + "2012211287/train1": 4.0, + "2012211287/train2": 10.0, + "2038594191/train0": 1.0, + "2038594191/train1": 4.0, + "2038594191/train2": 2.0, + "2054899126/train0": 4.0, + "2054899126/train1": 11.0, + "2054899126/train2": 6.5, + "2061635888/train0": 1.0, + "2061635888/train1": 5.0, + "2061635888/train2": 2.0, + "211245842/train0": 11.0, + "211245842/train1": 11.0, + "211245842/train2": 6.0, + "278008524/train0": 9.0, + "278008524/train1": 4.5, + "278008524/train2": 15.0, + "307068879/train0": 11.0, + "307068879/train1": 6.0, + "307068879/train2": 3.0, + "412102563/train0": 10.0, + "412102563/train1": 8.0, + "412102563/train2": 9.0, + "440577711/train0": 1.0, + "440577711/train1": 4.0, + "440577711/train2": 2.0, + "441046872/train0": 7.0, + "441046872/train1": 6.0, + "441046872/train2": 1.0, + "509430392/train0": 15.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 4.0, + "585904130/train1": 15.0, + "585904130/train2": 1.0, + "608016290/train0": 11.0, + "608016290/train1": 3.0, + "608016290/train2": 1.0, + "626486813/train0": 9.5, + "626486813/train1": 6.0, + "626486813/train2": 2.0, + "636680137/train0": 0.0, + "636680137/train1": 1.0, + "636680137/train2": 1.0, + "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": 9.0, + "701667486/train1": 12.0, + "701667486/train2": 13.0, + "747260437/train0": 9.0, + "747260437/train1": 4.0, + "747260437/train2": 14.0, + "750448296/train0": 4.0, + "750448296/train1": 3.6666666666666665, + "750448296/train2": 8.0, + "767747424/train0": 5.25, + "767747424/train1": 5.5, + "767747424/train2": 8.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 6.856, + "low": 4.656, + "mean": 5.856, + "n": 5, + "num_samples": 300 + }, + "mean": 5.856, + "per_task": { + "LiftPegUpright-v1": 6.0, + "PickCube-v1": 7.333333333333333, + "PullCube-v1": 5.946666666666666, + "PushCube-v1": 3.6666666666666665, + "StackCube-v1": 6.333333333333333 + } + }, + "metric": "selected_candidate_index", + "micro": { + "confidence": 0.95, + "high": 6.451388888888889, + "low": 5.076388888888889, + "mean": 5.784722222222222, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.4444444444444444, + "low": 0.2736111111111111, + "mean": 0.35694444444444445, + "n": 120, + "num_samples": 300 + }, + "mean": 0.35694444444444445, + "per_seed": { + "1073161778/train0": 0.0, + "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": 1.0, + "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": 1.0, + "1480856810/train1": 1.0, + "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": 1.0, + "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": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 0.5, + "2054899126/train1": 1.0, + "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": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "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": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 0.0, + "509430392/train1": 1.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 1.0, + "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": 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": 0.0, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.0, + "767747424/train0": 0.5, + "767747424/train1": 0.0, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.476, + "low": 0.12387878787878788, + "mean": 0.2861010101010101, + "n": 5, + "num_samples": 300 + }, + "mean": 0.2861010101010101, + "per_task": { + "LiftPegUpright-v1": 0.2, + "PickCube-v1": 0.0, + "PullCube-v1": 0.3466666666666667, + "PushCube-v1": 0.6111111111111112, + "StackCube-v1": 0.2727272727272727 + } + }, + "metric": "selected_success", + "micro": { + "confidence": 0.95, + "high": 0.4166666666666667, + "low": 0.2638888888888889, + "mean": 0.3402777777777778, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_success_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.1138888888888889, + "low": -0.014583333333333334, + "mean": 0.050694444444444445, + "n": 120, + "num_samples": 300 + }, + "mean": 0.050694444444444445, + "per_seed": { + "1073161778/train0": -1.0, + "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": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "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": 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": 1.0, + "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": 1.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": -0.5, + "2054899126/train1": 0.0, + "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": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "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": 0.0, + "509430392/train1": 1.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 1.0, + "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": 0.0, + "672328262/train1": 0.0, + "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": 0.0, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.0, + "767747424/train0": 0.25, + "767747424/train1": -0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.26666666666666666, + "low": -0.008, + "mean": 0.08622222222222223, + "n": 5, + "num_samples": 300 + }, + "mean": 0.08622222222222223, + "per_task": { + "LiftPegUpright-v1": 0.0, + "PickCube-v1": 0.0, + "PullCube-v1": -0.013333333333333334, + "PushCube-v1": 0.4444444444444444, + "StackCube-v1": 0.0 + } + }, + "metric": "selected_success_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.11805555555555555, + "low": -0.020833333333333332, + "mean": 0.04861111111111111, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.022063064091425, + "low": 0.7004546924930765, + "mean": 0.8510480198283201, + "n": 120, + "num_samples": 300 + }, + "mean": 0.8510480198283201, + "per_seed": { + "1073161778/train0": 0.67014479637146, + "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": 0.12164360284805298, + "1324642411/train1": 2.0, + "1324642411/train2": 0.10647052526473999, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.00405565882101655, + "141984615/train1": 0.027254482731223106, + "141984615/train2": 0.025202473625540733, + "1461069713/train0": 0.0939030647277832, + "1461069713/train1": 0.0939030647277832, + "1461069713/train2": 0.0939030647277832, + "1480856810/train0": 2.0, + "1480856810/train1": 2.0, + "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": 0.03462250903248787, + "1549324805/train1": 0.07181540876626968, + "1549324805/train2": 0.04690708592534065, + "1578990937/train0": 0.82413649559021, + "1578990937/train1": 0.1366594284772873, + "1578990937/train2": 0.15504932403564453, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 2.0, + "1790188787/train1": 0.10585354268550873, + "1790188787/train2": 0.1341685652732849, + "1822210471/train0": 0.06247224286198616, + "1822210471/train1": 2.0, + "1822210471/train2": 0.0730673298239708, + "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": 0.713559627532959, + "2012211287/train1": 0.7183246612548828, + "2012211287/train2": 2.0, + "2038594191/train0": 2.0, + "2038594191/train1": 2.0, + "2038594191/train2": 2.0, + "2054899126/train0": 1.0643903091549873, + "2054899126/train1": 2.0, + "2054899126/train2": 1.0520938150584698, + "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": 0.061764609068632126, + "278008524/train1": 0.034389604814350605, + "278008524/train2": 1.0713709965348244, + "307068879/train0": 0.19761089980602264, + "307068879/train1": 0.19690512120723724, + "307068879/train2": 0.22035211324691772, + "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": 0.20773810148239136, + "441046872/train1": 0.21283115446567535, + "441046872/train2": 0.7130289077758789, + "509430392/train0": 0.11310504376888275, + "509430392/train1": 2.0, + "509430392/train2": 2.682209014892578e-07, + "585904130/train0": 0.09919407218694687, + "585904130/train1": 0.04792279005050659, + "585904130/train2": 2.0, + "608016290/train0": 0.725701630115509, + "608016290/train1": 0.725701630115509, + "608016290/train2": 0.725701630115509, + "626486813/train0": 0.007258951431140304, + "626486813/train1": 0.011037707328796387, + "626486813/train2": 0.01727360486984253, + "636680137/train0": 0.056092143058776855, + "636680137/train1": 0.017789265140891075, + "636680137/train2": 0.015303333289921284, + "655884605/train0": 0.16300667822360992, + "655884605/train1": 0.11176733672618866, + "655884605/train2": 0.16300667822360992, + "672328262/train0": 2.0, + "672328262/train1": 2.0, + "672328262/train2": 0.08514729887247086, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 0.07023122161626816, + "747260437/train1": 0.05181918665766716, + "747260437/train2": 0.0880771279335022, + "750448296/train0": 0.07612514992554982, + "750448296/train1": 0.7115031480789185, + "750448296/train2": 0.08786896243691444, + "767747424/train0": 1.021046225563623, + "767747424/train1": 0.06286553613608703, + "767747424/train2": 1.0271160593256354 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.062293044694389, + "low": 0.6574902632200357, + "mean": 0.8220922048931207, + "n": 5, + "num_samples": 300 + }, + "mean": 0.8220922048931207, + "per_task": { + "LiftPegUpright-v1": 0.8018400182326635, + "PickCube-v1": 0.5839591026306152, + "PullCube-v1": 0.7315792678203433, + "PushCube-v1": 1.2593486458063126, + "StackCube-v1": 0.7337339899756692 + } + }, + "metric": "selected_utility", + "micro": { + "confidence": 0.95, + "high": 0.9381401165980302, + "low": 0.6609127892410874, + "mean": 0.802287638622349, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_utility_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.17575063188786672, + "low": -0.0658762492805383, + "mean": 0.06092995753052592, + "n": 120, + "num_samples": 300 + }, + "mean": 0.06092995753052592, + "per_seed": { + "1073161778/train0": -1.32985520362854, + "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": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.015173077583312988, + "1324642411/train1": 1.89352947473526, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": -0.09877858823165298, + "141984615/train1": -0.07557976432144642, + "141984615/train2": -0.07763177342712879, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.9129659235477448, + "1480856810/train1": 1.9129659235477448, + "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": -0.06989232823252678, + "1549324805/train1": -0.032699428498744965, + "1549324805/train2": -0.057607751339673996, + "1578990937/train0": 0.0, + "1578990937/train1": -0.6874770671129227, + "1578990937/train2": -0.6690871715545654, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.87523253262043, + "1790188787/train1": -0.01891392469406128, + "1790188787/train2": 0.009401097893714905, + "1822210471/train0": -0.009443063288927078, + "1822210471/train1": 1.9280846938490868, + "1822210471/train2": 0.0011520236730575562, + "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": -0.01004570722579956, + "2012211287/train1": -0.005280673503875732, + "2012211287/train2": 1.2763946652412415, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": -0.9356096908450127, + "2054899126/train1": 0.0, + "2054899126/train2": -0.9479061849415302, + "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": 0.01902961917221546, + "278008524/train1": -0.008345385082066059, + "278008524/train2": 1.0286360066384077, + "307068879/train0": -0.4024796634912491, + "307068879/train1": -0.4031854420900345, + "307068879/train2": -0.379738450050354, + "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": -0.4922170639038086, + "441046872/train1": -0.4871240109205246, + "441046872/train2": 0.013073742389678955, + "509430392/train0": 0.11310482025146484, + "509430392/train1": 1.999999776482582, + "509430392/train2": 4.470348358154297e-08, + "585904130/train0": -0.0825214758515358, + "585904130/train1": -0.13379275798797607, + "585904130/train2": 1.8182844519615173, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.002160608710255474, + "626486813/train1": 0.005939364607911557, + "626486813/train2": 0.0121752621489577, + "636680137/train0": -0.027042903006076813, + "636680137/train1": -0.0653457809239626, + "636680137/train2": -0.06783171277493238, + "655884605/train0": 0.0, + "655884605/train1": -0.051239341497421265, + "655884605/train2": 0.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": -1.9148527011275291, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": -1.9297687783837318, + "747260437/train1": -1.9481808133423328, + "747260437/train2": -1.9119228720664978, + "750448296/train0": -0.04146799941857656, + "750448296/train1": 0.5939099987347921, + "750448296/train2": -0.02972418690721194, + "767747424/train0": 0.4478468835586682, + "767747424/train1": -0.5103338058688678, + "767747424/train2": 0.4539167173206806 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4905605761210124, + "low": -0.10606869387804445, + "mean": 0.12653240771910573, + "n": 5, + "num_samples": 300 + }, + "mean": 0.12653240771910573, + "per_task": { + "LiftPegUpright-v1": -0.14803052047888438, + "PickCube-v1": 0.0, + "PullCube-v1": -0.04359119340311736, + "PushCube-v1": 0.8669444670279821, + "StackCube-v1": -0.0426607145504518 + } + }, + "metric": "selected_utility_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.19171025126787652, + "low": -0.08498874931117623, + "mean": 0.060468052180011805, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selector_regret": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.543745311871074, + "low": 0.3002178259030593, + "mean": 0.4138028312825352, + "n": 120, + "num_samples": 300 + }, + "mean": 0.4138028312825352, + "per_seed": { + "1073161778/train0": 1.32985520362854, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "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": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.04025374352931976, + "1324133122/train2": 0.10070674121379852, + "1324642411/train0": 0.02241644263267517, + "1324642411/train1": 0.0, + "1324642411/train2": 1.89352947473526, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.011122246738523245, + "141984615/train1": 0.009756645187735558, + "141984615/train2": 0.03374924324452877, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.006073415279388428, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 1.9129659235477448, + "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": 0.12710610777139664, + "1549324805/train2": 0.06619475409388542, + "1578990937/train0": 1.17586350440979, + "1578990937/train1": 1.8633405715227127, + "1578990937/train2": 1.8449506759643555, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.8941464573144913, + "1790188787/train2": 1.865831434726715, + "1822210471/train0": 1.9375277571380138, + "1822210471/train1": 0.0, + "1822210471/train2": 1.9269326701760292, + "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": 1.286440372467041, + "2012211287/train1": 1.2816753387451172, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.012892842292785645, + "2054899126/train1": 0.0, + "2054899126/train2": 0.9479061849415302, + "2061635888/train0": 0.04574880003929138, + "2061635888/train1": 0.02355286478996277, + "2061635888/train2": 0.05880081653594971, + "211245842/train0": 1.8090915828943253, + "211245842/train1": 0.02204875648021698, + "211245842/train2": 0.04297435283660889, + "278008524/train0": 1.0115917138755322, + "278008524/train1": 1.0032849898561835, + "278008524/train2": 0.06457791477441788, + "307068879/train0": 0.01394665241241455, + "307068879/train1": 0.3907022029161453, + "307068879/train2": 0.009861335158348083, + "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": 1.7922618985176086, + "441046872/train1": 0.510527715086937, + "441046872/train2": 0.006105959415435791, + "509430392/train0": 1.8868949562311172, + "509430392/train1": 0.0, + "509430392/train2": 5.155801773071289e-06, + "585904130/train0": 1.9008059278130531, + "585904130/train1": 1.9520772099494934, + "585904130/train2": 0.0, + "608016290/train0": 1.274298369884491, + "608016290/train1": 1.274298369884491, + "608016290/train2": 0.0, + "626486813/train0": 0.037436524173244834, + "626486813/train1": 0.024790167342871428, + "626486813/train2": 0.02762565016746521, + "636680137/train0": 0.0321691632270813, + "636680137/train1": 0.04962191917002201, + "636680137/train2": 0.060348290018737316, + "655884605/train0": 0.0, + "655884605/train1": 0.7011084705591202, + "655884605/train2": 0.05146662890911102, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 1.9148527011275291, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0381644144654274, + "747260437/train1": 0.011470336467027664, + "747260437/train2": 0.03702475130558014, + "750448296/train0": 0.676535177975893, + "750448296/train1": 0.6590001260240873, + "750448296/train2": 1.2625797055661678, + "767747424/train0": 0.021382962237112224, + "767747424/train1": 1.0000875550904311, + "767747424/train2": 0.04005103558301926 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.540436565328126, + "low": 0.1811315929842078, + "mean": 0.3887287384043401, + "n": 5, + "num_samples": 300 + }, + "mean": 0.3887287384043401, + "per_task": { + "LiftPegUpright-v1": 0.6113315612077713, + "PickCube-v1": 0.0, + "PullCube-v1": 0.3792251957549403, + "PushCube-v1": 0.5264327691660987, + "StackCube-v1": 0.42665416589289 + } + }, + "metric": "selector_regret", + "micro": { + "confidence": 0.95, + "high": 0.5418494971333226, + "low": 0.3011564811555824, + "mean": 0.4247725029110572, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "success_selector_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.3097222222222223, + "low": 0.1652777777777778, + "mean": 0.22777777777777777, + "n": 120, + "num_samples": 300 + }, + "mean": 0.22777777777777777, + "per_seed": { + "1073161778/train0": 1.0, + "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": 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": 0.0, + "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": 0.0, + "1480856810/train1": 0.0, + "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": 0.0, + "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": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.5, + "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": 0.0, + "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": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 1.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 0.0, + "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": 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": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.0, + "767747424/train1": 0.5, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.3258181818181819, + "low": 0.09696969696969697, + "mean": 0.22137373737373736, + "n": 5, + "num_samples": 300 + }, + "mean": 0.22137373737373736, + "per_task": { + "LiftPegUpright-v1": 0.4, + "PickCube-v1": 0.0, + "PullCube-v1": 0.18666666666666668, + "PushCube-v1": 0.2777777777777778, + "StackCube-v1": 0.24242424242424243 + } + }, + "metric": "success_selector_gap", + "micro": { + "confidence": 0.95, + "high": 0.2986111111111111, + "low": 0.1527777777777778, + "mean": 0.22916666666666666, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "success_support_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.3055555555555555, + "low": 0.15833333333333333, + "mean": 0.23194444444444443, + "n": 120, + "num_samples": 300 + }, + "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": 0.7023030303030303, + "low": 0.10133333333333334, + "mean": 0.40452525252525257, + "n": 5, + "num_samples": 300 + }, + "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": 0.2708333333333333, + "low": 0.14583333333333334, + "mean": 0.2013888888888889, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "support_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5403802280805798, + "low": 0.2915130193099483, + "mean": 0.4232947583840643, + "n": 120, + "num_samples": 300 + }, + "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": 1.3095993929816916, + "low": 0.20360193743308388, + "mean": 0.7236978183567826, + "n": 5, + "num_samples": 300 + }, + "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": 0.49125471889015493, + "low": 0.27718254707158646, + "mean": 0.3761285081272945, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "tau": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.9210821320668933, + "low": 0.7471692078395286, + "mean": 0.8324079763339929, + "n": 120, + "num_samples": 300 + }, + "mean": 0.8324079763339929, + "per_seed": { + "1073161778/train0": 0.49469300263486815, + "1073161778/train1": 0.49469300263486815, + "1073161778/train2": 0.49469300263486815, + "1113371261/train0": 1.4393298012330682, + "1113371261/train1": 1.4393298012330682, + "1113371261/train2": 1.4393298012330682, + "1151410346/train0": 1.4393298012330682, + "1151410346/train1": 1.4393298012330682, + "1151410346/train2": 1.4393298012330682, + "1261930456/train0": 1.311950902009979, + "1261930456/train1": 1.311950902009979, + "1261930456/train2": 1.311950902009979, + "1324133122/train0": 1.4393298012330682, + "1324133122/train1": 1.4393298012330682, + "1324133122/train2": 1.4393298012330682, + "1324642411/train0": 1.311950902009979, + "1324642411/train1": 1.311950902009979, + "1324642411/train2": 1.311950902009979, + "1375492786/train0": 0.3663750360591518, + "1375492786/train1": 0.3663750360591518, + "1375492786/train2": 0.3663750360591518, + "141984615/train0": 0.3663750360591518, + "141984615/train1": 0.3663750360591518, + "141984615/train2": 0.3663750360591518, + "1461069713/train0": 0.3663750360591518, + "1461069713/train1": 0.3663750360591518, + "1461069713/train2": 0.3663750360591518, + "1480856810/train0": 1.311950902009979, + "1480856810/train1": 1.311950902009979, + "1480856810/train2": 1.311950902009979, + "1509980664/train0": 1.4393298012330682, + "1509980664/train1": 1.4393298012330682, + "1509980664/train2": 1.4393298012330682, + "1522272686/train0": 0.8901452015560949, + "1522272686/train1": 0.8901452015560949, + "1522272686/train2": 0.8901452015560949, + "1549324805/train0": 0.3663750360591518, + "1549324805/train1": 0.3663750360591518, + "1549324805/train2": 0.3663750360591518, + "1578990937/train0": 1.4393298012330682, + "1578990937/train1": 1.4393298012330682, + "1578990937/train2": 1.4393298012330682, + "1747178030/train0": 0.3663750360591518, + "1747178030/train1": 0.3663750360591518, + "1747178030/train2": 0.3663750360591518, + "1790188787/train0": 1.311950902009979, + "1790188787/train1": 1.311950902009979, + "1790188787/train2": 1.311950902009979, + "1822210471/train0": 0.3663750360591518, + "1822210471/train1": 0.3663750360591518, + "1822210471/train2": 0.3663750360591518, + "1845489013/train0": 0.3663750360591518, + "1845489013/train1": 0.3663750360591518, + "1845489013/train2": 0.3663750360591518, + "1999022574/train0": 1.4393298012330682, + "1999022574/train1": 1.4393298012330682, + "1999022574/train2": 1.4393298012330682, + "2012211287/train0": 0.49469300263486815, + "2012211287/train1": 0.49469300263486815, + "2012211287/train2": 0.49469300263486815, + "2038594191/train0": 0.3663750360591518, + "2038594191/train1": 0.3663750360591518, + "2038594191/train2": 0.3663750360591518, + "2054899126/train0": 0.3663750360591518, + "2054899126/train1": 0.3663750360591518, + "2054899126/train2": 0.3663750360591518, + "2061635888/train0": 1.4393298012330682, + "2061635888/train1": 1.4393298012330682, + "2061635888/train2": 1.4393298012330682, + "211245842/train0": 1.4393298012330682, + "211245842/train1": 1.4393298012330682, + "211245842/train2": 1.4393298012330682, + "278008524/train0": 0.3663750360591518, + "278008524/train1": 0.3663750360591518, + "278008524/train2": 0.3663750360591518, + "307068879/train0": 0.49469300263486815, + "307068879/train1": 0.49469300263486815, + "307068879/train2": 0.49469300263486815, + "412102563/train0": 1.4393298012330682, + "412102563/train1": 1.4393298012330682, + "412102563/train2": 1.4393298012330682, + "440577711/train0": 1.311950902009979, + "440577711/train1": 1.311950902009979, + "440577711/train2": 1.311950902009979, + "441046872/train0": 0.49469300263486815, + "441046872/train1": 0.49469300263486815, + "441046872/train2": 0.49469300263486815, + "509430392/train0": 1.311950902009979, + "509430392/train1": 1.311950902009979, + "509430392/train2": 1.311950902009979, + "585904130/train0": 0.3663750360591518, + "585904130/train1": 0.3663750360591518, + "585904130/train2": 0.3663750360591518, + "608016290/train0": 0.49469300263486815, + "608016290/train1": 0.49469300263486815, + "608016290/train2": 0.49469300263486815, + "626486813/train0": 0.3663750360591518, + "626486813/train1": 0.3663750360591518, + "626486813/train2": 0.3663750360591518, + "636680137/train0": 0.3663750360591518, + "636680137/train1": 0.3663750360591518, + "636680137/train2": 0.3663750360591518, + "655884605/train0": 1.4393298012330682, + "655884605/train1": 1.4393298012330682, + "655884605/train2": 1.4393298012330682, + "672328262/train0": 0.3663750360591518, + "672328262/train1": 0.3663750360591518, + "672328262/train2": 0.3663750360591518, + "701667486/train0": 1.4393298012330682, + "701667486/train1": 1.4393298012330682, + "701667486/train2": 1.4393298012330682, + "747260437/train0": 0.3663750360591518, + "747260437/train1": 0.3663750360591518, + "747260437/train2": 0.3663750360591518, + "750448296/train0": 0.3663750360591518, + "750448296/train1": 0.3663750360591518, + "750448296/train2": 0.3663750360591518, + "767747424/train0": 0.3663750360591518, + "767747424/train1": 0.3663750360591518, + "767747424/train2": 0.3663750360591518 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.3040171014530562, + "low": 0.5224562557888273, + "mean": 0.9004987886986328, + "n": 5, + "num_samples": 300 + }, + "mean": 0.9004987886986328, + "per_task": { + "LiftPegUpright-v1": 0.49469300263486815, + "PickCube-v1": 0.8901452015560949, + "PullCube-v1": 0.3663750360591525, + "PushCube-v1": 1.311950902009979, + "StackCube-v1": 1.439329801233069 + } + }, + "metric": "tau", + "micro": { + "confidence": 0.95, + "high": 0.8442411854780398, + "low": 0.6815616160413008, + "mean": 0.7547358196215194, + "n": 144, + "num_samples": 300 + }, + "n": 144 + } + }, + "target": "success_weighted_margin", + "tau": { + "LiftPegUpright-v1": 0.49469300263486815, + "PickCube-v1": 0.8901452015560949, + "PullCube-v1": 0.3663750360591518, + "PushCube-v1": 1.311950902009979, + "StackCube-v1": 1.4393298012330682, + "__global__": 1.342806745945893 + }, + "threshold_scope": "task", + "weights": [ + 0.48450353350029784, + 0.07164325084566325, + 0.13384891787780617, + 0.14277476518903853, + -0.25142644745940856, + -0.06071005391438436, + -0.08510435948035511, + -0.35257424080968924, + 0.22771891138681077, + -0.01053488041108075, + -0.1905714409687107, + 0.10986510306533881, + -0.11189561324454189, + -0.11115529153539683, + 0.010621379429637479, + -0.012675082078364097, + -0.524921903059294, + 0.0304133002373419, + -0.018416761369735468, + -0.02909537162486231, + -0.06343099385412643, + 0.0541815552933258, + 0.018513799321226543, + 0.04677287738013547, + 0.06664463090419304, + -0.07562482165667461, + -0.018416759434330743, + -0.029095372061419766, + -0.06343098849174567, + 0.054181555642571766, + 0.018513798971071083, + 0.04677287723461632, + 0.06664463120978326, + -0.07562481825152645, + -0.02804355804619263, + -0.05128446951130172, + -0.039079363508790266, + -0.042589451928506605, + -0.050692557691945694, + -0.0391266424121568, + 0.053025433837319724, + -0.05561228844453581, + -6.139736871028727e-14, + -0.03801878647209378, + -0.056499726000765804, + 5.335937097464887e-14, + -17.308496856983766, + -16.184493665148693, + 33.40724765684445, + 2.547233946090273, + -0.2126175289873471, + -2.3716070393174675, + -3.2050477294519624, + 18.4319833381669, + 17.08916851083086, + -30.865822416550145, + 0.10526076975656906, + -1.1068176241469609e-13, + 0.5207000416908709, + 0.7446261919426078, + -0.010244430584805286, + -0.1800777625408756, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] +} diff --git a/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_seed.json b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_seed.json new file mode 100644 index 0000000000000000000000000000000000000000..9f32a59d024eb39d32ca550fc57db9e10cd3f7c2 --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_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": 0.6386554621848739, + "pairwise_causal_calibration_confidence": 0.5373743568439672, + "pairwise_causal_calibration_ece": 0.10128110534090673, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5566626770858889, + "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": 0.49469300263486815 + }, + "1073161778/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": 0.6833333333333333, + "pairwise_causal_calibration_confidence": 0.5322990000175534, + "pairwise_causal_calibration_ece": 0.15103433331577998, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3872764726300346, + "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": 0.49469300263486815 + }, + "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": 0.6428571428571429, + "pairwise_causal_calibration_confidence": 0.5451426904730965, + "pairwise_causal_calibration_ece": 0.09771445238404626, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.5303575557953266, + "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": 0.49469300263486815 + }, + "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": 0.5701414698923739, + "pairwise_causal_calibration_ece": 0.06599981030811651, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.4419532287444907, + "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": 1.4393298012330682 + }, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5714411258007887, + "pairwise_causal_calibration_ece": 0.1327098335615991, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.428095135833086, + "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": 1.4393298012330682 + }, + "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": 0.75, + "pairwise_causal_calibration_confidence": 0.5817900691636045, + "pairwise_causal_calibration_ece": 0.16820993083639552, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.32091844016441584, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "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": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "tau": 1.4393298012330682 + }, + "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": 0.5142857142857142, + "pairwise_causal_calibration_confidence": 0.5667135771731647, + "pairwise_causal_calibration_ece": 0.05242786288745051, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.024503811643319, + "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": 1.4393298012330682 + }, + "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": 0.37142857142857144, + "pairwise_causal_calibration_confidence": 0.5683151855386959, + "pairwise_causal_calibration_ece": 0.19688661411012442, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.0689882649948501, + "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": 1.4393298012330682 + }, + "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": 0.4380952380952381, + "pairwise_causal_calibration_confidence": 0.565958352249713, + "pairwise_causal_calibration_ece": 0.1278631141544748, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.0039423360080715, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "selected_candidate_index": 4.0, + "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": 1.4393298012330682 + }, + "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": 0.46464646464646464, + "pairwise_causal_calibration_confidence": 0.5719896658662266, + "pairwise_causal_calibration_ece": 0.12912993334954873, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 1.1198710760724104, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "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": 1.311950902009979 + }, + "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": 0.509090909090909, + "pairwise_causal_calibration_confidence": 0.5680754050042826, + "pairwise_causal_calibration_ece": 0.05898449591337354, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.0083103168677, + "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": 1.311950902009979 + }, + "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": 0.6309523809523809, + "pairwise_causal_calibration_confidence": 0.5619754350016658, + "pairwise_causal_calibration_ece": 0.068976945950715, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.0993338824963033, + "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": 1.311950902009979 + }, + "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": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5755791782863937, + "pairwise_causal_calibration_ece": 0.147007749714965, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.36820169505830336, + "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": 1.4393298012330682 + }, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5725133632951149, + "pairwise_causal_calibration_ece": 0.05201497258156908, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.0393222781022331, + "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": 1.4393298012330682 + }, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5840738619592613, + "pairwise_causal_calibration_ece": 0.1404771894262721, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3115767566062857, + "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": 0.09220491349697113, + "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": 1.4393298012330682 + }, + "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": 0.55, + "pairwise_causal_calibration_confidence": 0.5260274890742292, + "pairwise_causal_calibration_ece": 0.023972510925770796, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.8008980492120834, + "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": 1.311950902009979 + }, + "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": 0.7058823529411765, + "pairwise_causal_calibration_confidence": 0.5257069664610932, + "pairwise_causal_calibration_ece": 0.18017538648008335, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.4072117500373107, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 3.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.89352947473526, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "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": 0.527876777683192, + "pairwise_causal_calibration_ece": 0.017577767771353447, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.8877017595011019, + "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": 0.10647052526473999, + "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": 1.311950902009979 + }, + "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": 0.64, + "pairwise_causal_calibration_confidence": 0.5585290828327523, + "pairwise_causal_calibration_ece": 0.08147091716724786, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.0461275740764826, + "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": 0.3663750360591518 + }, + "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": 0.5610084990880762, + "pairwise_causal_calibration_ece": 0.03469270961439179, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.45720538390197873, + "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": 0.3663750360591518 + }, + "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": 0.5676443863088845, + "pairwise_causal_calibration_ece": 0.18712668693111412, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.34776849268222776, + "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": 0.3663750360591518 + }, + "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": 0.31666666666666665, + "pairwise_causal_calibration_confidence": 0.563739392345968, + "pairwise_causal_calibration_ece": 0.24707272567930166, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6540918192032681, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.00405565882101655, + "selected_utility_gain_over_base": -0.09877858823165298, + "selector_regret": 0.011122246738523245, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "tau": 0.3663750360591518 + }, + "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": 0.625, + "pairwise_causal_calibration_confidence": 0.581566420974051, + "pairwise_causal_calibration_ece": 0.10416729357646315, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8936304139520725, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.037011127918958664, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.027254482731223106, + "selected_utility_gain_over_base": -0.07557976432144642, + "selector_regret": 0.009756645187735558, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24876614287495613, + "tau": 0.3663750360591518 + }, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5761565714740622, + "pairwise_causal_calibration_ece": 0.15948990480739544, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4503895097168673, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058951716870069504, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.025202473625540733, + "selected_utility_gain_over_base": -0.07763177342712879, + "selector_regret": 0.03374924324452877, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "tau": 0.3663750360591518 + }, + "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": 0.8166666666666667, + "pairwise_causal_calibration_confidence": 0.5411312415303838, + "pairwise_causal_calibration_ece": 0.2755354251362828, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.3670742388145669, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06019836664199829, + "selected_candidate_index": 7.0, + "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": 0.3663750360591518 + }, + "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": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5331472452449045, + "pairwise_causal_calibration_ece": 0.06648057857823786, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.16101647169097077, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "selected_candidate_index": 1.0, + "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": 0.3663750360591518 + }, + "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": 0.4083333333333333, + "pairwise_causal_calibration_confidence": 0.527311702364046, + "pairwise_causal_calibration_ece": 0.11897836903071268, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.14534023869673263, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "selected_candidate_index": 5.0, + "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": 0.3663750360591518 + }, + "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": 0.8080808080808081, + "pairwise_causal_calibration_confidence": 0.5911741147277916, + "pairwise_causal_calibration_ece": 0.21690669335301668, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 2.149778411340016, + "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": 1.9129659235477448, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "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": 0.5798319327731093, + "pairwise_causal_calibration_confidence": 0.5569826639894806, + "pairwise_causal_calibration_ece": 0.02590153431562172, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.6034012607502497, + "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": 1.9129659235477448, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "1480856810/train2": { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5966386554621849, + "pairwise_causal_calibration_confidence": 0.5507966193765133, + "pairwise_causal_calibration_ece": 0.04584203608567172, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.300578618886508, + "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": 0.08703407645225525, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9129659235477448, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "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": 0.5222222222222223, + "pairwise_causal_calibration_confidence": 0.5411679300565454, + "pairwise_causal_calibration_ece": 0.018945707834323134, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.7690911091840036, + "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": 1.4393298012330682 + }, + "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": 0.25396825396825395, + "pairwise_causal_calibration_confidence": 0.5436154155208672, + "pairwise_causal_calibration_ece": 0.2896471615526132, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": 0.8299756115727561, + "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": 1.4393298012330682 + }, + "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": 0.5555555555555556, + "pairwise_causal_calibration_confidence": 0.5426912569382093, + "pairwise_causal_calibration_ece": 0.02770067302918524, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.8506856511161374, + "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": 1.4393298012330682 + }, + "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": 0.7222222222222222, + "pairwise_causal_calibration_confidence": 0.5639487552532058, + "pairwise_causal_calibration_ece": 0.15827346696901654, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.3519897595066688, + "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": 0.8901452015560949 + }, + "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": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5732579849122883, + "pairwise_causal_calibration_ece": 0.10103576269006627, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.7344242898918623, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07014383375644684, + "selected_candidate_index": 7.0, + "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": 0.8901452015560949 + }, + "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": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5874115349169855, + "pairwise_causal_calibration_ece": 0.11518931269476336, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.4801934854321459, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14937306940555573, + "selected_candidate_index": 7.0, + "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": 0.8901452015560949 + }, + "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": 0.36666666666666664, + "pairwise_causal_calibration_confidence": 0.5627369782509959, + "pairwise_causal_calibration_ece": 0.19607031158432947, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.171258504104145, + "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": 0.3663750360591518 + }, + "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": 0.35833333333333334, + "pairwise_causal_calibration_confidence": 0.5720588559847943, + "pairwise_causal_calibration_ece": 0.24463158134758306, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8208688776506676, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07181540876626968, + "selected_utility_gain_over_base": -0.032699428498744965, + "selector_regret": 0.12710610777139664, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "tau": 0.3663750360591518 + }, + "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": 0.35833333333333334, + "pairwise_causal_calibration_confidence": 0.5710713293995606, + "pairwise_causal_calibration_ece": 0.2127379960662275, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.420225409082441, + "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": 0.3663750360591518 + }, + "1578990937/train0": { + "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": 0.7094017094017094, + "pairwise_causal_calibration_confidence": 0.5852719048004434, + "pairwise_causal_calibration_ece": 0.12412980460126585, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 1.369766946850774, + "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": 0.82413649559021, + "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": 1.4393298012330682 + }, + "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": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5793795802536641, + "pairwise_causal_calibration_ece": 0.12483412570820954, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.5329062844260477, + "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": 0.1366594284772873, + "selected_utility_gain_over_base": -0.6874770671129227, + "selector_regret": 1.8633405715227127, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.4393298012330682 + }, + "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": 0.4957983193277311, + "pairwise_causal_calibration_confidence": 0.5773391698248656, + "pairwise_causal_calibration_ece": 0.08154085049713428, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.4783739869715653, + "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": 0.15504932403564453, + "selected_utility_gain_over_base": -0.6690871715545654, + "selector_regret": 1.8449506759643555, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.4393298012330682 + }, + "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": 0.7045359481010607, + "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": 0.3663750360591518 + }, + "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": 0.6796904941310151, + "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": 0.3663750360591518 + }, + "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": 0.7495662010387569, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "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": 0.0, + "tau": 0.3663750360591518 + }, + "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": 0.3, + "pairwise_causal_calibration_confidence": 0.519098923420412, + "pairwise_causal_calibration_ece": 0.21909892342041198, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.5993442072252648, + "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": 1.87523253262043, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "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": 0.3508771929824561, + "pairwise_causal_calibration_confidence": 0.5176749012095792, + "pairwise_causal_calibration_ece": 0.16679770822712303, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 1.622658467644122, + "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": 0.10585354268550873, + "selected_utility_gain_over_base": -0.01891392469406128, + "selector_regret": 1.8941464573144913, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "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": 0.4636363636363636, + "pairwise_causal_calibration_confidence": 0.5178387112441525, + "pairwise_causal_calibration_ece": 0.05420234760778886, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.4403189606632267, + "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": 0.1341685652732849, + "selected_utility_gain_over_base": 0.009401097893714905, + "selector_regret": 1.865831434726715, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5864304711853732, + "pairwise_causal_calibration_ece": 0.08360394561276052, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 1.5192545920029907, + "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": 0.06247224286198616, + "selected_utility_gain_over_base": -0.009443063288927078, + "selector_regret": 1.9375277571380138, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.3663750360591518 + }, + "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": 0.6530612244897959, + "pairwise_causal_calibration_confidence": 0.6090277928253506, + "pairwise_causal_calibration_ece": 0.04403343166444537, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": 1.4922290708776522, + "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": 1.9280846938490868, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.3663750360591518 + }, + "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": 0.425, + "pairwise_causal_calibration_confidence": 0.5868090410812485, + "pairwise_causal_calibration_ece": 0.16180904108124847, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 1.5163699975204636, + "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": 0.0730673298239708, + "selected_utility_gain_over_base": 0.0011520236730575562, + "selector_regret": 1.9269326701760292, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.3663750360591518 + }, + "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": 0.6521739130434783, + "pairwise_causal_calibration_confidence": 0.5356510938834426, + "pairwise_causal_calibration_ece": 0.1165228191600357, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": 1.3305286793016613, + "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": 0.3663750360591518 + }, + "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": 0.6153846153846154, + "pairwise_causal_calibration_confidence": 0.5374675838313216, + "pairwise_causal_calibration_ece": 0.07791703155329377, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": 1.412757419493385, + "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": 0.3663750360591518 + }, + "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": 0.72, + "pairwise_causal_calibration_confidence": 0.5506543992540976, + "pairwise_causal_calibration_ece": 0.16934560074590232, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": 1.3688073876119233, + "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": 0.3663750360591518 + }, + "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": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5351557804986392, + "pairwise_causal_calibration_ece": 0.17801292335578203, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.9988148345478187, + "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": 1.4393298012330682 + }, + "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": 0.7428571428571429, + "pairwise_causal_calibration_confidence": 0.5308223915074378, + "pairwise_causal_calibration_ece": 0.21203475134970512, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.9255492206908755, + "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": 1.4393298012330682 + }, + "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": 0.3684210526315789, + "pairwise_causal_calibration_confidence": 0.528258651563379, + "pairwise_causal_calibration_ece": 0.15983759893180005, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": 1.0109377403715456, + "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": 1.4393298012330682 + }, + "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": 0.5546218487394958, + "pairwise_causal_calibration_confidence": 0.5281524186595942, + "pairwise_causal_calibration_ece": 0.026469430079901657, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.7217723168310286, + "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": 0.713559627532959, + "selected_utility_gain_over_base": -0.01004570722579956, + "selector_regret": 1.286440372467041, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.49469300263486815 + }, + "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": 0.6083333333333333, + "pairwise_causal_calibration_confidence": 0.5310859795550439, + "pairwise_causal_calibration_ece": 0.0772473537782894, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5565021996855251, + "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": 0.7183246612548828, + "selected_utility_gain_over_base": -0.005280673503875732, + "selector_regret": 1.2816753387451172, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.49469300263486815 + }, + "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": 0.7282608695652174, + "pairwise_causal_calibration_confidence": 0.5365603796062945, + "pairwise_causal_calibration_ece": 0.19170048995892286, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.7477936348864276, + "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": 1.2763946652412415, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.49469300263486815 + }, + "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": 0.27586206896551724, + "pairwise_causal_calibration_confidence": 0.6084782956989382, + "pairwise_causal_calibration_ece": 0.33261622673342073, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.5405611437535534, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "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": 0.3663750360591518 + }, + "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": 0.6260111821626294, + "pairwise_causal_calibration_ece": 0.39213662401689425, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.2016013806419839, + "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": 0.3663750360591518 + }, + "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": 0.6127037139516429, + "pairwise_causal_calibration_ece": 0.20866330991123888, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 1.5465686912481509, + "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": 0.3663750360591518 + }, + "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": 0.25, + "pairwise_causal_calibration_confidence": 0.566620580797147, + "pairwise_causal_calibration_ece": 0.15831029039857342, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 1.0852982788395877, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.077283151447773, + "selected_candidate_index": 4.0, + "selected_success": 0.5, + "selected_success_gain_over_base": -0.5, + "selected_utility": 1.0643903091549873, + "selected_utility_gain_over_base": -0.9356096908450127, + "selector_regret": 0.012892842292785645, + "success_selector_gap": 0.0, + "success_support_gap": 0.5, + "support_gap": 0.922716848552227, + "tau": 0.3663750360591518 + }, + "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": 0.6578947368421053, + "pairwise_causal_calibration_confidence": 0.563744413130429, + "pairwise_causal_calibration_ece": 0.0470751618558381, + "pairwise_causal_calibration_pairs": 57.0, + "predicted_margin": 1.086209527289868, + "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": 0.3663750360591518 + }, + "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": 0.5630252100840336, + "pairwise_causal_calibration_confidence": 0.585134673600266, + "pairwise_causal_calibration_ece": 0.038714435110576664, + "pairwise_causal_calibration_pairs": 59.5, + "predicted_margin": 1.1880426299078302, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 6.5, + "selected_success": 0.5, + "selected_success_gain_over_base": -0.5, + "selected_utility": 1.0520938150584698, + "selected_utility_gain_over_base": -0.9479061849415302, + "selector_regret": 0.9479061849415302, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.3663750360591518 + }, + "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": 0.5277777777777778, + "pairwise_causal_calibration_confidence": 0.5745387895297428, + "pairwise_causal_calibration_ece": 0.04676101175196509, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 1.0205351759283516, + "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": 1.4393298012330682 + }, + "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": 0.3888888888888889, + "pairwise_causal_calibration_confidence": 0.5699891820408323, + "pairwise_causal_calibration_ece": 0.1811002931519435, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.9728419884347872, + "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": 0.05225050449371338, + "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": 1.4393298012330682 + }, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.5914469514823854, + "pairwise_causal_calibration_ece": 0.14700250703794093, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.8245519344630416, + "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": 1.4393298012330682 + }, + "211245842/train0": { + "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": 0.6218487394957983, + "pairwise_causal_calibration_confidence": 0.5746201886713925, + "pairwise_causal_calibration_ece": 0.04722855082440577, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.2882788003210806, + "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": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8090915828943253, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.4393298012330682 + }, + "211245842/train1": { + "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": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.5692105391837949, + "pairwise_causal_calibration_ece": 0.08786394286037044, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.3428657851681152, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02204875648021698, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7870428264141083, + "tau": 1.4393298012330682 + }, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5702656129330281, + "pairwise_causal_calibration_ece": 0.07026561293302824, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.2817098708877097, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "selected_candidate_index": 6.0, + "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": 1.4393298012330682 + }, + "278008524/train0": { + "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": 0.7226010101010101, + "pairwise_causal_calibration_confidence": 0.6163446579821468, + "pairwise_causal_calibration_ece": 0.10625635211886339, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.5507130046793893, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0733563229441643, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.061764609068632126, + "selected_utility_gain_over_base": 0.01902961917221546, + "selector_regret": 1.0115917138755322, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.04055231809616089, + "tau": 0.3663750360591518 + }, + "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": 0.6065217391304347, + "pairwise_causal_calibration_confidence": 0.6113953659702027, + "pairwise_causal_calibration_ece": 0.15188563211313605, + "pairwise_causal_calibration_pairs": 106.0, + "predicted_margin": 0.6145792187233674, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0376745946705341, + "selected_candidate_index": 4.5, + "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": 0.3663750360591518 + }, + "278008524/train2": { + "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": 0.5363636363636364, + "pairwise_causal_calibration_confidence": 0.6180490094046018, + "pairwise_causal_calibration_ece": 0.11847264063665816, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.7396224810135514, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.1359489113092422, + "selected_candidate_index": 15.0, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.5, + "selected_utility": 1.0713709965348244, + "selected_utility_gain_over_base": 1.0286360066384077, + "selector_regret": 0.06457791477441788, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.3663750360591518 + }, + "307068879/train0": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "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": 0.5934065934065934, + "pairwise_causal_calibration_confidence": 0.5944461291750907, + "pairwise_causal_calibration_ece": 0.12040324822018197, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 1.0739910273739994, + "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": 0.19761089980602264, + "selected_utility_gain_over_base": -0.4024796634912491, + "selector_regret": 0.01394665241241455, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7884424477815628, + "tau": 0.49469300263486815 + }, + "307068879/train1": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "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": 0.5384615384615384, + "pairwise_causal_calibration_confidence": 0.5937223391260118, + "pairwise_causal_calibration_ece": 0.0607943275756009, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.6874983921047111, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5876073241233826, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19690512120723724, + "selected_utility_gain_over_base": -0.4031854420900345, + "selector_regret": 0.3907022029161453, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "tau": 0.49469300263486815 + }, + "307068879/train2": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "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": 0.4175824175824176, + "pairwise_causal_calibration_confidence": 0.589313207577152, + "pairwise_causal_calibration_ece": 0.1717307899947341, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 1.1349062127057792, + "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": 0.22035211324691772, + "selected_utility_gain_over_base": -0.379738450050354, + "selector_regret": 0.009861335158348083, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7697865515947342, + "tau": 0.49469300263486815 + }, + "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": 0.65, + "pairwise_causal_calibration_confidence": 0.5385704504068652, + "pairwise_causal_calibration_ece": 0.11142954959313478, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 1.0933587971168492, + "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": 1.4393298012330682 + }, + "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": 0.8461538461538461, + "pairwise_causal_calibration_confidence": 0.5524013537256702, + "pairwise_causal_calibration_ece": 0.2937524924281759, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": 0.9728013354434921, + "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": 1.4393298012330682 + }, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5411176357911663, + "pairwise_causal_calibration_ece": 0.04111763579116634, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.0057266984542679, + "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": 1.4393298012330682 + }, + "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": 0.6206896551724138, + "pairwise_causal_calibration_confidence": 0.5455151446938427, + "pairwise_causal_calibration_ece": 0.07517451047857106, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.5732830552633406, + "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": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "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": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5341777572159363, + "pairwise_causal_calibration_ece": 0.0008444238826029737, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 1.4913443146002, + "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": 1.311950902009979 + }, + "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": 0.46296296296296297, + "pairwise_causal_calibration_confidence": 0.5311030945024883, + "pairwise_causal_calibration_ece": 0.0681401315395253, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.424277496369104, + "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": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "441046872/train0": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "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": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.6008428753783649, + "pairwise_causal_calibration_ece": 0.21902469356018284, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 1.1052022133857848, + "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": 0.20773810148239136, + "selected_utility_gain_over_base": -0.4922170639038086, + "selector_regret": 1.7922618985176086, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.49469300263486815 + }, + "441046872/train1": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "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": 0.6181818181818182, + "pairwise_causal_calibration_confidence": 0.6026494132350357, + "pairwise_causal_calibration_ece": 0.04900842512659715, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.6838040349024017, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7233588695526123, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21283115446567535, + "selected_utility_gain_over_base": -0.4871240109205246, + "selector_regret": 0.510527715086937, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "tau": 0.49469300263486815 + }, + "441046872/train2": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "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": 0.4909090909090909, + "pairwise_causal_calibration_confidence": 0.6088372831949117, + "pairwise_causal_calibration_ece": 0.20077855305415088, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 1.0158360314379022, + "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": 0.7130289077758789, + "selected_utility_gain_over_base": 0.013073742389678955, + "selector_regret": 0.006105959415435791, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "tau": 0.49469300263486815 + }, + "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": 0.6470588235294118, + "pairwise_causal_calibration_confidence": 0.5508070209680679, + "pairwise_causal_calibration_ece": 0.09625180256134375, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 2.08491547386598, + "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": 0.11310504376888275, + "selected_utility_gain_over_base": 0.11310482025146484, + "selector_regret": 1.8868949562311172, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "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": 0.5434550291857321, + "pairwise_causal_calibration_ece": 0.1537570975036694, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.9268668914168288, + "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": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.311950902009979 + }, + "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": 0.5520785359696122, + "pairwise_causal_calibration_ece": 0.1334130916939903, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.7820896027638113, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 5.424022674560547e-06, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.682209014892578e-07, + "selected_utility_gain_over_base": 4.470348358154297e-08, + "selector_regret": 5.155801773071289e-06, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9999945759773254, + "tau": 1.311950902009979 + }, + "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": 0.30952380952380953, + "pairwise_causal_calibration_confidence": 0.5734358732826558, + "pairwise_causal_calibration_ece": 0.2639120637588461, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.937964471343637, + "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": 0.3663750360591518 + }, + "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": 0.56, + "pairwise_causal_calibration_confidence": 0.5673349851937346, + "pairwise_causal_calibration_ece": 0.17050541777278239, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.867255607957766, + "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": 0.3663750360591518 + }, + "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": 0.6181818181818182, + "pairwise_causal_calibration_confidence": 0.5698793670685108, + "pairwise_causal_calibration_ece": 0.04830245111330746, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.0810514424099793, + "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": 0.3663750360591518 + }, + "608016290/train0": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "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": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5478322151873722, + "pairwise_causal_calibration_ece": 0.04803994731985866, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.34122628210022654, + "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": 0.725701630115509, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.274298369884491, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.49469300263486815 + }, + "608016290/train1": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "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": 0.5534688429350546, + "pairwise_causal_calibration_ece": 0.03000164602114203, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.1770980989610429, + "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": 0.725701630115509, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.274298369884491, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.49469300263486815 + }, + "608016290/train2": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "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": 0.55, + "pairwise_causal_calibration_confidence": 0.5597255439272126, + "pairwise_causal_calibration_ece": 0.029011213993528073, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3122233671552621, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7117276191711426, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.725701630115509, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "tau": 0.49469300263486815 + }, + "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": 0.6499999999999999, + "pairwise_causal_calibration_confidence": 0.6373243545977667, + "pairwise_causal_calibration_ece": 0.10601407456772449, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.205283676715192, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04469547560438514, + "selected_candidate_index": 9.5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007258951431140304, + "selected_utility_gain_over_base": 0.002160608710255474, + "selector_regret": 0.037436524173244834, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12161866808310151, + "tau": 0.3663750360591518 + }, + "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": 0.33333333333333337, + "pairwise_causal_calibration_confidence": 0.6287134543731017, + "pairwise_causal_calibration_ece": 0.2953801210397682, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.2030969286900701, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.035827874671667814, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.011037707328796387, + "selected_utility_gain_over_base": 0.005939364607911557, + "selector_regret": 0.024790167342871428, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12975450651720166, + "tau": 0.3663750360591518 + }, + "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": 0.6416666666666666, + "pairwise_causal_calibration_confidence": 0.6377705352082403, + "pairwise_causal_calibration_ece": 0.11878506662697773, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0499244621925214, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04489925503730774, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.01727360486984253, + "selected_utility_gain_over_base": 0.0121752621489577, + "selector_regret": 0.02762565016746521, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12068312615156174, + "tau": 0.3663750360591518 + }, + "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": 0.529875300645354, + "pairwise_causal_calibration_ece": 0.04450430387060281, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0013727088350777, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.08826130628585815, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.056092143058776855, + "selected_utility_gain_over_base": -0.027042903006076813, + "selector_regret": 0.0321691632270813, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0823030173778534, + "tau": 0.3663750360591518 + }, + "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": 0.475, + "pairwise_causal_calibration_confidence": 0.5159628167606266, + "pairwise_causal_calibration_ece": 0.04096281676062663, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.40455029586141933, + "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": 0.3663750360591518 + }, + "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": 0.4583333333333333, + "pairwise_causal_calibration_confidence": 0.5181751903900925, + "pairwise_causal_calibration_ece": 0.05984185705675921, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.531328083388189, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0756516233086586, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.015303333289921284, + "selected_utility_gain_over_base": -0.06783171277493238, + "selector_regret": 0.060348290018737316, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09491270035505295, + "tau": 0.3663750360591518 + }, + "655884605/train0": { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "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": 0.5555555555555556, + "pairwise_causal_calibration_confidence": 0.6048190953611235, + "pairwise_causal_calibration_ece": 0.08237033905417408, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 1.095636223788052, + "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": 0.16300667822360992, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8626713156700134, + "tau": 1.4393298012330682 + }, + "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": 0.6276117908365466, + "pairwise_causal_calibration_ece": 0.15419473151003862, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 1.4832101206592103, + "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": 1.4393298012330682 + }, + "655884605/train2": { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "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": 0.6012392567929711, + "pairwise_causal_calibration_ece": 0.18457259012630453, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 1.3520359199432512, + "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": 0.16300667822360992, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.05146662890911102, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.785526692867279, + "tau": 1.4393298012330682 + }, + "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": 0.8823529411764706, + "pairwise_causal_calibration_confidence": 0.637846163378886, + "pairwise_causal_calibration_ece": 0.24450677779758456, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": 1.7549902847176035, + "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": 0.3663750360591518 + }, + "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": 0.45714285714285713, + "pairwise_causal_calibration_confidence": 0.629165943748097, + "pairwise_causal_calibration_ece": 0.1720230866052398, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": 1.7117888933582819, + "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": 0.3663750360591518 + }, + "672328262/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": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.6358538481639073, + "pairwise_causal_calibration_ece": 0.30252051483057396, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": 1.5731937235091635, + "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": 0.08514729887247086, + "selected_utility_gain_over_base": -1.9148527011275291, + "selector_regret": 1.9148527011275291, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.3663750360591518 + }, + "701667486/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": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5376448138211073, + "pairwise_causal_calibration_ece": 0.2519305281068216, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 0.6729506809560988, + "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": 1.4393298012330682 + }, + "701667486/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": 0.5952380952380952, + "pairwise_causal_calibration_confidence": 0.5347371571140561, + "pairwise_causal_calibration_ece": 0.060500938124039116, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.7189098900525153, + "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": 1.4393298012330682 + }, + "701667486/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": 0.5740740740740741, + "pairwise_causal_calibration_confidence": 0.5391779065928842, + "pairwise_causal_calibration_ece": 0.05737097122205889, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.841493235581737, + "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": 1.4393298012330682 + }, + "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": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.555978759483347, + "pairwise_causal_calibration_ece": 0.04733473753037369, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6951613484301653, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10839563608169556, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07023122161626816, + "selected_utility_gain_over_base": -1.9297687783837318, + "selector_regret": 0.0381644144654274, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17026761174201965, + "tau": 0.3663750360591518 + }, + "747260437/train1": { + "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": 0.675, + "pairwise_causal_calibration_confidence": 0.5687872987376212, + "pairwise_causal_calibration_ece": 0.2478180852749098, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5927568927858111, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06328952312469482, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.05181918665766716, + "selected_utility_gain_over_base": -1.9481808133423328, + "selector_regret": 0.011470336467027664, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.21537372469902039, + "tau": 0.3663750360591518 + }, + "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": 0.7333333333333333, + "pairwise_causal_calibration_confidence": 0.5528896350884385, + "pairwise_causal_calibration_ece": 0.18044369824489515, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8567886898938823, + "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": 0.3663750360591518 + }, + "750448296/train0": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.5857142857142857, + "pairwise_causal_calibration_confidence": 0.5477839571124287, + "pairwise_causal_calibration_ece": 0.07643574626949645, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": 1.0438240923289683, + "proposal_oracle_success": 0.3333333333333333, + "proposal_oracle_utility": 0.7526603279014429, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07612514992554982, + "selected_utility_gain_over_base": -0.04146799941857656, + "selector_regret": 0.676535177975893, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.3333333333333333, + "support_gap": 0.6553361676633358, + "tau": 0.3663750360591518 + }, + "750448296/train1": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4307692307692308, + "pairwise_causal_calibration_confidence": 0.5465796418867024, + "pairwise_causal_calibration_ece": 0.11581041111747163, + "pairwise_causal_calibration_pairs": 101.66666666666667, + "predicted_margin": 0.7480058669656788, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3705032741030057, + "selected_candidate_index": 3.6666666666666665, + "selected_success": 0.3333333333333333, + "selected_success_gain_over_base": 0.3333333333333333, + "selected_utility": 0.7115031480789185, + "selected_utility_gain_over_base": 0.5939099987347921, + "selector_regret": 0.6590001260240873, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.0, + "support_gap": 0.03749322146177292, + "tau": 0.3663750360591518 + }, + "750448296/train2": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.4765873015873015, + "pairwise_causal_calibration_confidence": 0.5485445741855494, + "pairwise_causal_calibration_ece": 0.08011293420877054, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": 0.8330105684043841, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3504486680030823, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08786896243691444, + "selected_utility_gain_over_base": -0.02972418690721194, + "selector_regret": 1.2625797055661678, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.05754782756169637, + "tau": 0.3663750360591518 + }, + "767747424/train0": { + "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": 0.49887820512820513, + "pairwise_causal_calibration_confidence": 0.5657585105530328, + "pairwise_causal_calibration_ece": 0.18436261086486025, + "pairwise_causal_calibration_pairs": 119.25, + "predicted_margin": 0.3010322890243957, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0424291878007352, + "selected_candidate_index": 5.25, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.25, + "selected_utility": 1.021046225563623, + "selected_utility_gain_over_base": 0.4478468835586682, + "selector_regret": 0.021382962237112224, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.06562284985557199, + "tau": 0.3663750360591518 + }, + "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": 0.5168659771600949, + "pairwise_causal_calibration_confidence": 0.5647149634509379, + "pairwise_causal_calibration_ece": 0.12296851299012063, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.3930644164701522, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0629530912265182, + "selected_candidate_index": 5.5, + "selected_success": 0.0, + "selected_success_gain_over_base": -0.25, + "selected_utility": 0.06286553613608703, + "selected_utility_gain_over_base": -0.5103338058688678, + "selector_regret": 1.0000875550904311, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.045098946429789066, + "tau": 0.3663750360591518 + }, + "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": 0.5111070710335417, + "pairwise_causal_calibration_confidence": 0.5720190194619137, + "pairwise_causal_calibration_ece": 0.1533915123915277, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.3778346343969301, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0671670949086547, + "selected_candidate_index": 8.0, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.25, + "selected_utility": 1.0271160593256354, + "selected_utility_gain_over_base": 0.4539167173206806, + "selector_regret": 0.04005103558301926, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.04088494274765253, + "tau": 0.3663750360591518 + } +} diff --git a/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_task.json b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_task.json new file mode 100644 index 0000000000000000000000000000000000000000..24fceafaaab3170007f3ac75e24b359fe3babbb5 --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_task.json @@ -0,0 +1,137 @@ +{ + "LiftPegUpright-v1": { + "base_success": 0.2, + "base_utility": 0.9498705387115478, + "coverage": 0.7333333333333333, + "execute_generated": 0.7333333333333333, + "fallback_rate": 0.26666666666666666, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.5333333333333333, + "pairwise_causal_calibration_accuracy": 0.5688090610724882, + "pairwise_causal_calibration_confidence": 0.5640968449927838, + "pairwise_causal_calibration_ece": 0.10494933398158825, + "pairwise_causal_calibration_pairs": 94.8, + "predicted_margin": 0.6688100344694228, + "proposal_oracle_success": 0.6, + "proposal_oracle_utility": 1.4122399787108104, + "selected_candidate_index": 6.0, + "selected_success": 0.2, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8018400182326635, + "selected_utility_gain_over_base": -0.14803052047888438, + "selector_regret": 0.6113315612077713, + "success_selector_gap": 0.4, + "success_support_gap": 0.4, + "support_gap": 0.5877600212891897, + "tau": 0.49469300263486815 + }, + "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": 0.5555555555555555, + "pairwise_causal_calibration_confidence": 0.5748727583608265, + "pairwise_causal_calibration_ece": 0.12483284745128205, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.5222025116102257, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13425676027933756, + "selected_candidate_index": 7.333333333333333, + "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": 0.8901452015560949 + }, + "PullCube-v1": { + "base_success": 0.36, + "base_utility": 0.7751704612234608, + "coverage": 0.8133333333333334, + "execute_generated": 0.8133333333333334, + "fallback_rate": 0.18666666666666668, + "hidden_chart_oracle_success": 0.52, + "hidden_chart_oracle_utility": 1.1467878860235214, + "outcome_ptr": 0.4666666666666667, + "pairwise_causal_calibration_accuracy": 0.5253206716900769, + "pairwise_causal_calibration_confidence": 0.5764914402555266, + "pairwise_causal_calibration_ece": 0.13706044945766013, + "pairwise_causal_calibration_pairs": 92.32, + "predicted_margin": 0.8087553078530508, + "proposal_oracle_success": 0.5333333333333333, + "proposal_oracle_utility": 1.1093241602430741, + "selected_candidate_index": 5.946666666666666, + "selected_success": 0.3466666666666667, + "selected_success_gain_over_base": -0.013333333333333334, + "selected_utility": 0.7315792678203433, + "selected_utility_gain_over_base": -0.04359119340311736, + "selector_regret": 0.3792251957549403, + "success_selector_gap": 0.18666666666666668, + "success_support_gap": 0.02666666666666667, + "support_gap": 0.1075624164690574, + "tau": 0.3663750360591525 + }, + "PushCube-v1": { + "base_success": 0.16666666666666666, + "base_utility": 0.3924041787783305, + "coverage": 0.7222222222222222, + "execute_generated": 0.7222222222222222, + "fallback_rate": 0.2777777777777778, + "hidden_chart_oracle_success": 0.8333333333333334, + "hidden_chart_oracle_utility": 1.7015800600250561, + "outcome_ptr": 0.8333333333333334, + "pairwise_causal_calibration_accuracy": 0.5564552891415517, + "pairwise_causal_calibration_confidence": 0.5440196808663499, + "pairwise_causal_calibration_ece": 0.09639707450334344, + "pairwise_causal_calibration_pairs": 97.44444444444444, + "predicted_margin": 1.5178990886097534, + "proposal_oracle_success": 0.8888888888888888, + "proposal_oracle_utility": 1.7857814149724112, + "selected_candidate_index": 3.6666666666666665, + "selected_success": 0.6111111111111112, + "selected_success_gain_over_base": 0.4444444444444444, + "selected_utility": 1.2593486458063126, + "selected_utility_gain_over_base": 0.8669444670279821, + "selector_regret": 0.5264327691660987, + "success_selector_gap": 0.2777777777777778, + "success_support_gap": 0.1111111111111111, + "support_gap": 0.21421858502758873, + "tau": 1.311950902009979 + }, + "StackCube-v1": { + "base_success": 0.2727272727272727, + "base_utility": 0.7763947045261209, + "coverage": 0.09090909090909091, + "execute_generated": 0.09090909090909091, + "fallback_rate": 0.9090909090909091, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.6363636363636364, + "pairwise_causal_calibration_accuracy": 0.5212002507574325, + "pairwise_causal_calibration_confidence": 0.5650793632911131, + "pairwise_causal_calibration_ece": 0.12478007178656221, + "pairwise_causal_calibration_pairs": 67.45454545454545, + "predicted_margin": 0.8903319056592717, + "proposal_oracle_success": 0.5151515151515151, + "proposal_oracle_utility": 1.1567951707225856, + "selected_candidate_index": 6.333333333333333, + "selected_success": 0.2727272727272727, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7337339899756692, + "selected_utility_gain_over_base": -0.0426607145504518, + "selector_regret": 0.42665416589289, + "success_selector_gap": 0.24242424242424243, + "success_support_gap": 0.48484848484848486, + "support_gap": 0.8432048292774142, + "tau": 1.439329801233069 + } +} diff --git a/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/split_hash.txt b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/split_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..865ad41e869c1bfbc57e4db65ed79e91e6d93081 --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/split_hash.txt @@ -0,0 +1,14 @@ +{ + "calibration_target_index": { + "content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "retrieval_index_allowed": true, + "split": "train", + "split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372" + }, + "eval_target_index": { + "content_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "retrieval_index_allowed": false, + "split": "test", + "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f" + } +} diff --git a/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/table.tex b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/table.tex new file mode 100644 index 0000000000000000000000000000000000000000..99b58c64cfae3dd2319b27f33d3e83edc0b28537 --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_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 & 0.3403 & 0.5694 & 0.5486 & 0.2014 & 0.2292 & 0.0365 \\ +\bottomrule +\end{tabular} diff --git a/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/train.log b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/train.log new file mode 100644 index 0000000000000000000000000000000000000000..f53ad2d2dc23f831010e8f476476b5c57d9740df --- /dev/null +++ b/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_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=0.01 +tau={"LiftPegUpright-v1": 0.494693, "PickCube-v1": 0.890145, "PullCube-v1": 0.366375, "PushCube-v1": 1.311951, "StackCube-v1": 1.43933, "__global__": 1.342807} +calibration_input=runs/ctt_base_context_obs_train_cal_envclip_k16_rollout_comparison/combined_measured_candidates.json