diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/command.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/command.txt new file mode 100644 index 0000000000000000000000000000000000000000..dd18d1098592bc0bd372bc1f401a67f816c38197 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_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_chartcompat_successbonus2_task_envclip_k16_train_to_test --feature-set score_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/config.yaml b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cfabeae8a9f08286bdbc68a972265e7d4d8fe0b3 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_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_chart_compat +fit_objective: pointwise +k: 16 +no_markdown_report: True +out_dir: runs/ctt_base_context_obs_learned_dominance_score_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/data_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/data_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..e65c158c17e61aaf10422563d2ffdd8922735e6a --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/eval.log b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/eval.log new file mode 100644 index 0000000000000000000000000000000000000000..ad44bd37e51f82471f7a17d7372cee5fed00ed3f --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/git_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/git_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..293c6ad4e775f2f937dca574f1b83a5f93f0bdb1 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/git_hash.txt @@ -0,0 +1 @@ +14e81292e2004925f81e0bb4810f6f144bfd87b7 diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics.json b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..1909c5fc2efda51578c0d723a6f48a295c892dd4 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics.json @@ -0,0 +1,9008 @@ +{ + "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": 0.5138888888888888, + "fallback_rate": 0.4861111111111111, + "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": 0.35185185185185186, + "selected_success_gain_over_base": 0.07407407407407407, + "selected_utility": 0.8191827951105316, + "selector_regret": 0.41054302532871517, + "success_selector_gap": 0.22453703703703703, + "success_support_gap": 0.17592592592592593, + "support_gap": 0.35465237537627453 + }, + "calibration_summary": { + "base_success": 0.2777777777777778, + "base_utility": 0.7007981554626086, + "coverage": 0.5138888888888888, + "fallback_rate": 0.4861111111111111, + "hidden_chart_oracle_success": 0.6597222222222222, + "hidden_chart_oracle_utility": 1.4150809271054134, + "outcome_ptr": 0.5462962962962963, + "pairwise_causal_calibration_accuracy": 0.5058718205276739, + "pairwise_causal_calibration_confidence": 0.5568104969283495, + "pairwise_causal_calibration_ece": 0.05093867640067546, + "pairwise_causal_calibration_pairs": 37978.0, + "proposal_oracle_success": 0.5370370370370371, + "proposal_oracle_utility": 1.1440672337592803, + "selected_success": 0.35185185185185186, + "selected_success_gain_over_base": 0.07407407407407407, + "selected_utility": 0.8191827951105316, + "selector_regret": 0.41054302532871517, + "success_selector_gap": 0.22453703703703703, + "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": 0.5486111111111112, + "fallback_rate": 0.4513888888888889, + "hidden_chart_oracle_success": 0.7291666666666666, + "hidden_chart_oracle_utility": 1.5183161981403828, + "outcome_ptr": 0.5486111111111112, + "pairwise_causal_calibration_accuracy": 0.5247707897699855, + "pairwise_causal_calibration_confidence": 0.5549377694154455, + "pairwise_causal_calibration_ece": 0.030166979645460227, + "pairwise_causal_calibration_pairs": 12434.0, + "proposal_oracle_success": 0.5694444444444444, + "proposal_oracle_utility": 1.2159999172436073, + "selected_success": 0.3194444444444444, + "selected_success_gain_over_base": 0.027777777777777776, + "selected_utility": 0.7728274115272344, + "selector_regret": 0.45799514941932606, + "success_selector_gap": 0.25, + "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, + 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", + "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_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, + 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": 0.5058718205276739, + "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": 0.02938225092245028, + "accuracy": 0.5073257124456609, + "confidence": 0.5367079633681112, + "count": 31055, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.13873592791765804, + "accuracy": 0.49385749385749383, + "confidence": 0.6325934217751519, + "count": 6105, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.15430795591393243, + "accuracy": 0.5841726618705037, + "confidence": 0.7384806177844361, + "count": 695, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.5458597609917754, + "accuracy": 0.2975206611570248, + "confidence": 0.8433804221488002, + "count": 121, + "lower": 0.8, + "upper": 0.9 + }, + { + "abs_gap": 0.9036747853344398, + "accuracy": 0.0, + "confidence": 0.9036747853344398, + "count": 2, + "lower": 0.9, + "upper": 1.0 + } + ], + "ece": 0.05093867640067546, + "mean_confidence": 0.5568104969283495, + "n_bins": 10, + "num_pairs": 37978, + "num_rows": 432 + }, + "eval": { + "accuracy": 0.5247707897699855, + "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": 0.009271938591821294, + "accuracy": 0.5264476346468832, + "confidence": 0.5357195732387044, + "count": 10379, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.11527287642897033, + "accuracy": 0.5194585448392555, + "confidence": 0.6347314212682258, + "count": 1773, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.28892814277768075, + "accuracy": 0.44329896907216493, + "confidence": 0.7322271118498457, + "count": 194, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.20945346204210047, + "accuracy": 0.6136363636363636, + "confidence": 0.8230898256784641, + "count": 88, + "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": 0.030166979645460227, + "mean_confidence": 0.5549377694154455, + "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": 0.48739495798319327, + "pairwise_causal_calibration_confidence": 0.533600718072752, + "pairwise_causal_calibration_ece": 0.04620576008955871, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.107133532936841, + "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": 0.87335466614541, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5494505494505495, + "pairwise_causal_calibration_confidence": 0.5728850640530763, + "pairwise_causal_calibration_ece": 0.1199981805489268, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.7747416665744656, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2115575522184372, + "seed": "307068879/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7884424477815628, + "task_id": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "train_seed": "0" + }, + { + "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": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.5462334122382229, + "pairwise_causal_calibration_ece": 0.06721196591303755, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.8315109279797959, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/train0", + "selected_candidate_index": 6, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4857142857142857, + "pairwise_causal_calibration_confidence": 0.594216772727526, + "pairwise_causal_calibration_ece": 0.11412216411454103, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.589765155203597, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.707331120967865, + "selected_utility_gain_over_base": -0.018370509147644043, + "selector_regret": 1.292668879032135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.5664245191730715, + "pairwise_causal_calibration_ece": 0.19522652344582997, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.6069194020559678, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "441046872/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.3000448346138, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.550550217481388, + "pairwise_causal_calibration_ece": 0.11611644918527875, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.3705665707855397, + "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": 1.2894892762347252, + "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": 0.6807136864797709, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/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": 0.5825884328057473, + "train_seed": "0" + }, + { + "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": 0.49572649572649574, + "pairwise_causal_calibration_confidence": 0.5264649509158664, + "pairwise_causal_calibration_ece": 0.030738455189370684, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.5168928978077507, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train0", + "selected_candidate_index": 14, + "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": 0.5825884328057473, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.523550311048989, + "pairwise_causal_calibration_ece": 0.14311635561767766, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.06695685840699794, + "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": 0.5825884328057473, + "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": 0.75, + "pairwise_causal_calibration_confidence": 0.6566064262371367, + "pairwise_causal_calibration_ece": 0.27387444019257834, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.3361085944561055, + "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": 0.5825884328057473, + "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": 0.26666666666666666, + "pairwise_causal_calibration_confidence": 0.5291897348449885, + "pairwise_causal_calibration_ece": 0.2625230681783219, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8451839925989519, + "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": 0.5825884328057473, + "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": 0.8235294117647058, + "pairwise_causal_calibration_confidence": 0.5940991028072462, + "pairwise_causal_calibration_ece": 0.22943030895745967, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": 0.7871393331842744, + "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": 0.5825884328057473, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5787709662074795, + "pairwise_causal_calibration_ece": 0.16210429954081287, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4206163793097437, + "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": 0.5825884328057473, + "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": 0.7333333333333333, + "pairwise_causal_calibration_confidence": 0.6097360202282281, + "pairwise_causal_calibration_ece": 0.12359731310510547, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8291893276454425, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14671264588832855, + "seed": "278008524/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05741874501109123, + "selected_utility_gain_over_base": 0.021532438695430756, + "selector_regret": 0.08929390087723732, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08110463619232178, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.41304347826086957, + "pairwise_causal_calibration_confidence": 0.51605426349952, + "pairwise_causal_calibration_ece": 0.10301078523865043, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": 0.798908624848506, + "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": 0.5825884328057473, + "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": 0.55, + "pairwise_causal_calibration_confidence": 0.5751237107284101, + "pairwise_causal_calibration_ece": 0.07811172592180064, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5841434390229071, + "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": 0.5825884328057473, + "train_seed": "0" + }, + { + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5491820139582639, + "pairwise_causal_calibration_ece": 0.11748465270840278, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5904121339165793, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train0", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.001021603704430163, + "selected_utility_gain_over_base": -1.9989783962955698, + "selector_regret": 1.9989783962955698, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.36904761904761907, + "pairwise_causal_calibration_confidence": 0.5472546704459165, + "pairwise_causal_calibration_ece": 0.17820705139829746, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.9073249211388359, + "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": 0.5825884328057473, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "chart_id": "ms-8e56b0eaa2604db6749e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.2833333333333333, + "pairwise_causal_calibration_confidence": 0.5448471499959241, + "pairwise_causal_calibration_ece": 0.2615138166625908, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5415586302800286, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "seed": "141984615/train0", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10283424705266953, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.5341284371266161, + "pairwise_causal_calibration_ece": 0.09920489620671744, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9654271375816255, + "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": 0.5825884328057473, + "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": 0.425, + "pairwise_causal_calibration_confidence": 0.540894445846163, + "pairwise_causal_calibration_ece": 0.11589444584616293, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0758545824030454, + "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": 0.5825884328057473, + "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": 0.38333333333333336, + "pairwise_causal_calibration_confidence": 0.5826379337146643, + "pairwise_causal_calibration_ece": 0.19930460038133097, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6551650928365002, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.019201835617423058, + "seed": "767747424/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007447937969118357, + "selected_utility_gain_over_base": -0.046675005462020636, + "selector_regret": 0.0117538976483047, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17790568806231022, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.5866666666666667, + "pairwise_causal_calibration_confidence": 0.5192841401886225, + "pairwise_causal_calibration_ece": 0.06738252647804421, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.13235098862844796, + "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": 0.5825884328057473, + "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": 0.5879562438719218, + "pairwise_causal_calibration_ece": 0.31209417490640445, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.2008765491740636, + "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": 0.5825884328057473, + "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": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.5409741214707926, + "pairwise_causal_calibration_ece": 0.07569254519587415, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.3177872369672483, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.041394393891096115, + "seed": "750448296/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.041394393891096115, + "selected_utility_gain_over_base": -0.05160309746861458, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.18259509280323982, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 1.1837230041552602, + "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": 0.5825884328057473, + "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": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5322967641018285, + "pairwise_causal_calibration_ece": 0.046545342019779196, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7403769985406177, + "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": 0.5825884328057473, + "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": 0.5050505050505051, + "pairwise_causal_calibration_confidence": 0.5720678596122971, + "pairwise_causal_calibration_ece": 0.14539609879445076, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": -0.27774835711899776, + "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": 0.5825884328057473, + "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": 0.43333333333333335, + "pairwise_causal_calibration_confidence": 0.5370476153908349, + "pairwise_causal_calibration_ece": 0.10371428205750155, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.332576441483642, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.060871422290802, + "seed": "1549324805/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0260902252048254, + "selected_utility_gain_over_base": -0.07842461206018925, + "selector_regret": 0.0347811970859766, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.23658466339111328, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.5678773715560055, + "pairwise_causal_calibration_ece": 0.0719433679479887, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 0.7418095681624295, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06306678056716919, + "selected_utility_gain_over_base": -0.008848525583744049, + "selector_regret": 1.9369332194328308, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4880952380952381, + "pairwise_causal_calibration_confidence": 0.5328695070998236, + "pairwise_causal_calibration_ece": 0.04477426900458553, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.11786003398422906, + "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": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.6551724137931034, + "pairwise_causal_calibration_confidence": 0.5624630563402808, + "pairwise_causal_calibration_ece": 0.09270935745282255, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.7706684649919293, + "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": 0.8468573198810986, + "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": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5461265642044052, + "pairwise_causal_calibration_ece": 0.020540102462261362, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.8003040878491947, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14406004548072815, + "seed": "1324642411/train0", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12164360284805298, + "selected_utility_gain_over_base": 0.015173077583312988, + "selector_regret": 0.02241644263267517, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8559399545192719, + "task_id": "PushCube-v1", + "tau": 0.8468573198810986, + "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": 0.2818181818181818, + "pairwise_causal_calibration_confidence": 0.5215209331218725, + "pairwise_causal_calibration_ece": 0.23970275130369073, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.1652542261433299, + "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": 0.8468573198810986, + "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": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5514987071348044, + "pairwise_causal_calibration_ece": 0.09695325258934998, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.7394668700396011, + "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": 0.8468573198810986, + "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": 0.6386554621848739, + "pairwise_causal_calibration_confidence": 0.5365694240669033, + "pairwise_causal_calibration_ece": 0.10561970831698192, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.7250273458520244, + "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": 0.8468573198810986, + "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": 0.8282828282828283, + "pairwise_causal_calibration_confidence": 0.5709539167236203, + "pairwise_causal_calibration_ece": 0.2573289115592079, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 2.2046095259547753, + "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": 0.8468573198810986, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.5639694431421344, + "pairwise_causal_calibration_ece": 0.130957329325739, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.39584763737906137, + "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": 1.0718654827809488, + "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": 0.5833333333333334, + "pairwise_causal_calibration_confidence": 0.5958635862373126, + "pairwise_causal_calibration_ece": 0.10068633489228643, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.5292851601669687, + "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": 1.0718654827809488, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5542145473380933, + "pairwise_causal_calibration_ece": 0.08992883305237898, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.028473507842246004, + "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": 1.0718654827809488, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "chart_id": "ms-420280f15a522ce21c72", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7350427350427351, + "pairwise_causal_calibration_confidence": 0.575799443033985, + "pairwise_causal_calibration_ece": 0.15924329200874998, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 1.3204376505810713, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train0", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.17586350440979, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.5523718802888111, + "pairwise_causal_calibration_ece": 0.10792743584436665, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.6801689818523341, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train0", + "selected_candidate_index": 2, + "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": 1.0718654827809488, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6302521008403361, + "pairwise_causal_calibration_confidence": 0.5426955700305087, + "pairwise_causal_calibration_ece": 0.08755653080982741, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.1272780451605109, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "211245842/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8592818975448608, + "selected_utility_gain_over_base": 0.6683734804391861, + "selector_regret": 1.1407181024551392, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.5428571428571428, + "pairwise_causal_calibration_confidence": 0.5504892981594165, + "pairwise_causal_calibration_ece": 0.007632155302273683, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.8265443858364611, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8663521409034729, + "seed": "1151410346/train0", + "selected_candidate_index": 12, + "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": 1.0718654827809488, + "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": 0.5765433563593303, + "pairwise_causal_calibration_ece": 0.1479719277879018, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3146091278603853, + "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": 1.0718654827809488, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.38095238095238093, + "pairwise_causal_calibration_confidence": 0.5278746131288938, + "pairwise_causal_calibration_ece": 0.1469222321765129, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 1.1356123561511042, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/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": 1.0718654827809488, + "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": 0.5493461173759878, + "pairwise_causal_calibration_ece": 0.10951912707521982, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 1.0435143603129777, + "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": 1.0718654827809488, + "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": 0.543208103565328, + "pairwise_causal_calibration_ece": 0.1860652464224708, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.8291022763943762, + "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": 1.0718654827809488, + "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": 0.5379143734950628, + "pairwise_causal_calibration_ece": 0.07041895983827043, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9306280702125403, + "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": 0.87335466614541, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5274725274725275, + "pairwise_causal_calibration_confidence": 0.5704103700900957, + "pairwise_causal_calibration_ece": 0.07309450817078264, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.3600826916101192, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5876073241233826, + "seed": "307068879/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "task_id": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "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": 0.5414540870914354, + "pairwise_causal_calibration_ece": 0.14187924624189804, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.696723804413939, + "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": 0.87335466614541, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5047619047619047, + "pairwise_causal_calibration_confidence": 0.5978828235074669, + "pairwise_causal_calibration_ece": 0.20716363895507375, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.3392733370703795, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/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": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6181818181818182, + "pairwise_causal_calibration_confidence": 0.5652066435973622, + "pairwise_causal_calibration_ece": 0.09055210003018804, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.1629836409186699, + "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": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.023403704166412354, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "task_id": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5518834340111771, + "pairwise_causal_calibration_ece": 0.13521676734451044, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.8136424781392986, + "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": 1.2894892762347252, + "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": 0.7412627528047919, + "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": 0.5825884328057473, + "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": 0.4369747899159664, + "pairwise_causal_calibration_confidence": 0.5350494099395171, + "pairwise_causal_calibration_ece": 0.09807462002355061, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.3100122906270366, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16168592870235443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8383140712976456, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.5236824574508954, + "pairwise_causal_calibration_ece": 0.19034912411756205, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.24721792540099097, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "seed": "1461069713/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0939030647277832, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14275743998587132, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.6443129578291337, + "pairwise_causal_calibration_ece": 0.37764629116246684, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.1814917850355002, + "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": 0.5825884328057473, + "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": 0.6754385964912281, + "pairwise_causal_calibration_confidence": 0.5263065559429049, + "pairwise_causal_calibration_ece": 0.14913204054832319, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 0.6246851300643428, + "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": 0.5825884328057473, + "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": 0.5428571428571428, + "pairwise_causal_calibration_confidence": 0.5924504110795782, + "pairwise_causal_calibration_ece": 0.07756824536937998, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": 0.7852289914727505, + "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": 0.5825884328057473, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5916666666666667, + "pairwise_causal_calibration_confidence": 0.5516469249954096, + "pairwise_causal_calibration_ece": 0.09795148252740986, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.47195461369487435, + "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": 0.5825884328057473, + "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": 0.5583333333333333, + "pairwise_causal_calibration_confidence": 0.5898531385273096, + "pairwise_causal_calibration_ece": 0.06222000578143905, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9399097065125667, + "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": 0.5825884328057473, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7692307692307693, + "pairwise_causal_calibration_confidence": 0.5252519047235767, + "pairwise_causal_calibration_ece": 0.24397886450719253, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": 0.8663205932771525, + "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": 0.5825884328057473, + "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": 0.39166666666666666, + "pairwise_causal_calibration_confidence": 0.5647193558303677, + "pairwise_causal_calibration_ece": 0.17305268916370073, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7391401538679906, + "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": 0.5825884328057473, + "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": 0.7350427350427351, + "pairwise_causal_calibration_confidence": 0.5263645115117457, + "pairwise_causal_calibration_ece": 0.20867822353098942, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.5895406979096317, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/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": 0.5825884328057473, + "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": 0.64, + "pairwise_causal_calibration_confidence": 0.549320518336562, + "pairwise_causal_calibration_ece": 0.19650416069860938, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.8486373501760932, + "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": 0.5825884328057473, + "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": 0.6083333333333333, + "pairwise_causal_calibration_confidence": 0.5656089960064071, + "pairwise_causal_calibration_ece": 0.14136934515423885, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8524230828537844, + "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": 0.5825884328057473, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.14270895719528198, + "chart_id": "ms-972c5be72d0b24be7cd5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.425, + "pairwise_causal_calibration_confidence": 0.5283042091940678, + "pairwise_causal_calibration_ece": 0.10330420919406785, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.297807510881273, + "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": 0.14270895719528198, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.857291042804718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "chart_id": "ms-9757fc494e79b5543bb6", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.625, + "pairwise_causal_calibration_confidence": 0.5301042033378723, + "pairwise_causal_calibration_ece": 0.09489579666212766, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.44432064636845636, + "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": 0.08313504606485367, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10315313935279846, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.597444625657663, + "pairwise_causal_calibration_ece": 0.21983253135176184, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0400199103730006, + "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": 0.5825884328057473, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5386998051668176, + "pairwise_causal_calibration_ece": 0.03869980516681758, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.48662334960024145, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/train1", + "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": 0.5825884328057473, + "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": 0.24074074074074073, + "pairwise_causal_calibration_confidence": 0.6117747139844449, + "pairwise_causal_calibration_ece": 0.37103397324370413, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.112157225492381, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07674666494131088, + "selected_utility_gain_over_base": -1.9232533350586891, + "selector_regret": 1.9232533350586891, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.425, + "pairwise_causal_calibration_confidence": 0.5397340717215419, + "pairwise_causal_calibration_ece": 0.11729596961499984, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7678920642673333, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11150982230901718, + "seed": "750448296/train1", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06182505562901497, + "selected_utility_gain_over_base": -0.031172435730695724, + "selector_regret": 0.04968476668000221, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.11247966438531876, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 1.5233706918468892, + "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": 0.5825884328057473, + "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": 0.5502411276617256, + "pairwise_causal_calibration_ece": 0.2890079292574854, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6210780034951271, + "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": 0.5825884328057473, + "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": 0.6086956521739131, + "pairwise_causal_calibration_confidence": 0.5797446400488626, + "pairwise_causal_calibration_ece": 0.07560374307224685, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": -0.22025978406278973, + "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": 0.5825884328057473, + "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": 0.30833333333333335, + "pairwise_causal_calibration_confidence": 0.5437634159530779, + "pairwise_causal_calibration_ece": 0.2354300826197445, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9503772340326878, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "seed": "1549324805/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0303483996540308, + "selected_utility_gain_over_base": -0.07416643761098385, + "selector_regret": 0.16857311688363552, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.6938775510204082, + "pairwise_causal_calibration_confidence": 0.592873970483461, + "pairwise_causal_calibration_ece": 0.10100358053694718, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": 0.7078283118607503, + "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": 0.5825884328057473, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.47692307692307695, + "pairwise_causal_calibration_confidence": 0.5453125451571335, + "pairwise_causal_calibration_ece": 0.06838946823405649, + "pairwise_causal_calibration_pairs": 65.0, + "predicted_margin": 0.1509547219477273, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.5471459072582541, + "pairwise_causal_calibration_ece": 0.3939678976832891, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 1.808986225138223, + "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": 0.8468573198810986, + "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": 0.6218487394957983, + "pairwise_causal_calibration_confidence": 0.5434572117117997, + "pairwise_causal_calibration_ece": 0.07839152778399854, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.3876858687209335, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11658626794815063, + "selected_utility_gain_over_base": 0.010115742683410645, + "selector_regret": 1.8834137320518494, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.8468573198810986, + "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": 0.517198535735028, + "pairwise_causal_calibration_ece": 0.16632134275257193, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 1.1907212928313995, + "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": 0.8468573198810986, + "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": 0.4909090909090909, + "pairwise_causal_calibration_confidence": 0.5481788586308872, + "pairwise_causal_calibration_ece": 0.05726976772179637, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.7193134962639322, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train1", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.8468573198810986, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5326653211281538, + "pairwise_causal_calibration_ece": 0.038763250300417725, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.5703557578982594, + "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": 0.8468573198810986, + "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": 0.5882352941176471, + "pairwise_causal_calibration_confidence": 0.543456481052637, + "pairwise_causal_calibration_ece": 0.04477881306501008, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.735589926228614, + "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": 0.8468573198810986, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5581364424191588, + "pairwise_causal_calibration_ece": 0.1414697757524921, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.3705725365164993, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07580336928367615, + "seed": "2061635888/train1", + "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": 0.02355286478996277, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9241966307163239, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "train_seed": "1" + }, + { + "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": 0.5277777777777778, + "pairwise_causal_calibration_confidence": 0.6117746577593474, + "pairwise_causal_calibration_ece": 0.228982599451425, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.8854182759065323, + "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": 0.16300667822360992, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.6498691290616989, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.1871241927146912, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5534020512884718, + "pairwise_causal_calibration_ece": 0.12516937728295682, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.018305243090398104, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18093222379684448, + "seed": "1113371261/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "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": 1.0718654827809488, + "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": 0.4090909090909091, + "pairwise_causal_calibration_confidence": 0.5644259196254204, + "pairwise_causal_calibration_ece": 0.15533501053451115, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.2385033751871737, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12732218205928802, + "selected_utility_gain_over_base": -0.6968143135309219, + "selector_regret": 1.872677817940712, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.19047619047619047, + "pairwise_causal_calibration_confidence": 0.5517045024701317, + "pairwise_causal_calibration_ece": 0.36122831199394134, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": 0.7807195223473116, + "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": 1.0718654827809488, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.65, + "pairwise_causal_calibration_confidence": 0.5351083386938548, + "pairwise_causal_calibration_ece": 0.12495817852344455, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.2265101367583866, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "seed": "211245842/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09285585582256317, + "selected_utility_gain_over_base": -0.09805256128311157, + "selector_regret": 0.12010131776332855, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7870428264141083, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.38095238095238093, + "pairwise_causal_calibration_confidence": 0.5497312288122934, + "pairwise_causal_calibration_ece": 0.16877884785991246, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.9117051526313378, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1151410346/train1", + "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": 1.7739987075328827, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5703079922305777, + "pairwise_causal_calibration_ece": 0.07659820557682445, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.05209384637709236, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1324586570262909, + "seed": "1324133122/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "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": 1.0718654827809488, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5308683756394741, + "pairwise_causal_calibration_ece": 0.004845910074811617, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.234980143424118, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/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": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.7948717948717948, + "pairwise_causal_calibration_confidence": 0.5593981622164304, + "pairwise_causal_calibration_ece": 0.23547363265536436, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": 0.949568847497014, + "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": 1.0718654827809488, + "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": 0.6714285714285714, + "pairwise_causal_calibration_confidence": 0.5400358084963361, + "pairwise_causal_calibration_ece": 0.13139276293223526, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.7563954831040236, + "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": 1.0718654827809488, + "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": 0.7717391304347826, + "pairwise_causal_calibration_confidence": 0.5403873973296214, + "pairwise_causal_calibration_ece": 0.2313517331051612, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 1.1151401969224932, + "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": 0.87335466614541, + "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": 0.3956043956043956, + "pairwise_causal_calibration_confidence": 0.5723883106472546, + "pairwise_causal_calibration_ece": 0.1767839150428592, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.979654396381198, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2302134484052658, + "seed": "307068879/train2", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2021283358335495, + "selected_utility_gain_over_base": -0.39796222746372223, + "selector_regret": 0.02808511257171631, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7697865515947342, + "task_id": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "train_seed": "2" + }, + { + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5552324860178256, + "pairwise_causal_calibration_ece": 0.051910371125031565, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.8010166202874374, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/train2", + "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": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.55, + "pairwise_causal_calibration_confidence": 0.6068596136210345, + "pairwise_causal_calibration_ece": 0.0651220182950748, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.4772393565445312, + "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": 0.6789124011993408, + "selected_utility_gain_over_base": -0.04678922891616821, + "selector_regret": 0.03281521797180176, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "task_id": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5755444444938004, + "pairwise_causal_calibration_ece": 0.13114432369738743, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.5214657402479244, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7191348671913147, + "seed": "441046872/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019179701805114746, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "task_id": "LiftPegUpright-v1", + "tau": 0.87335466614541, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5635063449255251, + "pairwise_causal_calibration_ece": 0.06840618279050634, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.5686852402790281, + "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": 1.2894892762347252, + "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": 0.7399153467725452, + "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": 0.5825884328057473, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.16168592870235443, + "chart_id": "ms-0e1ff8d16372796dac97", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6554621848739496, + "pairwise_causal_calibration_confidence": 0.5336391081875892, + "pairwise_causal_calibration_ece": 0.12182307668636036, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.40469624443133256, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "selected_candidate_index": 9, + "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": 0.5825884328057473, + "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": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5276472177677529, + "pairwise_causal_calibration_ece": 0.019313884434419615, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2696456255951446, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "seed": "1461069713/train2", + "selected_candidate_index": 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, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.7416666666666667, + "pairwise_causal_calibration_confidence": 0.6498155430714577, + "pairwise_causal_calibration_ece": 0.2291282500464284, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9688416841155103, + "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": 0.5825884328057473, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-4cc40587ec490257b1d2", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5210084033613446, + "pairwise_causal_calibration_confidence": 0.5536973137638624, + "pairwise_causal_calibration_ece": 0.05340737629836459, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5730407111707803, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "selected_candidate_index": 4, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.2, + "pairwise_causal_calibration_confidence": 0.6012678269199632, + "pairwise_causal_calibration_ece": 0.4012678269199631, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": 0.6365332388231544, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04013325646519661, + "selected_utility_gain_over_base": -1.9598667435348034, + "selector_regret": 1.9598667435348034, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.3416666666666667, + "pairwise_causal_calibration_confidence": 0.5761802746367215, + "pairwise_causal_calibration_ece": 0.23451360797005488, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.24727067758313925, + "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": 0.5825884328057473, + "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": 0.4583333333333333, + "pairwise_causal_calibration_confidence": 0.6068247037197843, + "pairwise_causal_calibration_ece": 0.148491370386451, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8202350551439472, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2718978226184845, + "seed": "278008524/train2", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04554526135325432, + "selected_utility_gain_over_base": 0.009658955037593842, + "selector_regret": 0.22635256126523018, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.5274458415050698, + "pairwise_causal_calibration_ece": 0.19255415849493018, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": 0.8229653113797162, + "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": 0.5825884328057473, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.008267183788120747, + "chart_id": "ms-7df455efd413d77e6ce3", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.07774405181407928, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5786084601365713, + "pairwise_causal_calibration_ece": 0.07027512680323779, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.578326404421488, + "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": 0.008267183788120747, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.042679390870034695, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.026797477155923843, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.5384615384615384, + "pairwise_causal_calibration_confidence": 0.5293057577168471, + "pairwise_causal_calibration_ece": 0.009155780744691366, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.491258714195548, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "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": "PullCube-v1", + "tau": 0.5825884328057473, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "chart_id": "ms-8577816da2917a2bb63d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7272727272727273, + "pairwise_causal_calibration_confidence": 0.5513841945828967, + "pairwise_causal_calibration_ece": 0.1758885326898305, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.9433176675944015, + "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": 0.5825884328057473, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "chart_id": "ms-8e56b0eaa2604db6749e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.425, + "pairwise_causal_calibration_confidence": 0.5603553061579338, + "pairwise_causal_calibration_ece": 0.1353553061579339, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3015705513281932, + "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": 0.10283424705266953, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.5285042510062523, + "pairwise_causal_calibration_ece": 0.021495748993747718, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8104586454289818, + "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": 0.5825884328057473, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "chart_id": "ms-9757fc494e79b5543bb6", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": 0.5238615170185813, + "pairwise_causal_calibration_ece": 0.048861517018581324, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5548863156073045, + "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": 0.08313504606485367, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09491270035505295, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.5416666666666666, + "pairwise_causal_calibration_confidence": 0.6039307444274978, + "pairwise_causal_calibration_ece": 0.06615831252348275, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.1187950154023634, + "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": 0.5825884328057473, + "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": 0.7023809523809523, + "pairwise_causal_calibration_confidence": 0.5315351438061215, + "pairwise_causal_calibration_ece": 0.17084580857483078, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.36262329065526533, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/train2", + "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": 0.5825884328057473, + "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": 0.5954970669139762, + "pairwise_causal_calibration_ece": 0.19145666287357221, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 1.2364955407573137, + "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": 0.5825884328057473, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.0929974913597107, + "chart_id": "ms-b7d7021b782f73816083", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22398948669433594, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.48333333333333334, + "pairwise_causal_calibration_confidence": 0.5278506560476054, + "pairwise_causal_calibration_ece": 0.05093124656435263, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.776315255538415, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.051346004009246826, + "seed": "750448296/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04953896999359131, + "selected_utility_gain_over_base": -0.043458521366119385, + "selector_regret": 0.0018070340156555176, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1726434826850891, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 1.5690353615664272, + "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": 0.5825884328057473, + "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": 0.7083333333333334, + "pairwise_causal_calibration_confidence": 0.5279460992108118, + "pairwise_causal_calibration_ece": 0.1803872341225216, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9133990308397049, + "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": 0.5825884328057473, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.04958367347717285, + "chart_id": "ms-da555a03dd382d93e119", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5959595959595959, + "pairwise_causal_calibration_confidence": 0.5820771575937359, + "pairwise_causal_calibration_ece": 0.04630465959545786, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.13031225474580455, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train2", + "selected_candidate_index": 15, + "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": 0.5825884328057473, + "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": 0.4583333333333333, + "pairwise_causal_calibration_confidence": 0.540090634882582, + "pairwise_causal_calibration_ece": 0.08175730154924889, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.4498268017490537, + "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": 0.5825884328057473, + "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": 0.4, + "pairwise_causal_calibration_confidence": 0.5677066182834786, + "pairwise_causal_calibration_ece": 0.16770661828347844, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 0.714483922998961, + "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": 0.5825884328057473, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5565463429998946, + "pairwise_causal_calibration_ece": 0.13987967633322793, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.07483677485698936, + "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": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": 0.5825884328057473, + "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": 0.5185185185185185, + "pairwise_causal_calibration_confidence": 0.5475473443233374, + "pairwise_causal_calibration_ece": 0.17413687717265497, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.7310319203842686, + "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": 0.8468573198810986, + "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": 0.509090909090909, + "pairwise_causal_calibration_confidence": 0.5436846632267275, + "pairwise_causal_calibration_ece": 0.03459375413581838, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.7578621304129007, + "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": 0.8468573198810986, + "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": 0.43636363636363634, + "pairwise_causal_calibration_confidence": 0.5206471195621108, + "pairwise_causal_calibration_ece": 0.08428348319847451, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.0616754451066321, + "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": 0.8468573198810986, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5450693806396976, + "pairwise_causal_calibration_ece": 0.0620734765031595, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.7862881124399304, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train2", + "selected_candidate_index": 4, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.8468573198810986, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "chart_id": "ms-c2ae2b78ab7891e74c80", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.5403636942457117, + "pairwise_causal_calibration_ece": 0.14212516552457483, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.3794654730008236, + "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": 0.8468573198810986, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "chart_id": "ms-d4c37ed44fb338c1d551", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5370429153145398, + "pairwise_causal_calibration_ece": 0.03438565611403155, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.359251321625822, + "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": 0.21894000470638275, + "selected_utility_gain_over_base": 0.1319059282541275, + "selector_regret": 1.7810599952936172, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.8468573198810986, + "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": 0.5794706736680671, + "pairwise_causal_calibration_ece": 0.13502622922362276, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.345599063353308, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11105132102966309, + "seed": "2061635888/train2", + "selected_candidate_index": 7, + "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": 1.0718654827809488, + "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": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5899703889806722, + "pairwise_causal_calibration_ece": 0.19094254082377798, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.7967317370820184, + "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": 1.0718654827809488, + "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": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5590997153025572, + "pairwise_causal_calibration_ece": 0.11947171326887146, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.16653094820357445, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "seed": "1113371261/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5644061750173082, + "pairwise_causal_calibration_ece": 0.13583474644587978, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.1890371636445307, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train2", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13359522819519043, + "selected_utility_gain_over_base": -0.6905412673950195, + "selector_regret": 1.8664047718048096, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.5444444444444444, + "pairwise_causal_calibration_confidence": 0.5506485429344871, + "pairwise_causal_calibration_ece": 0.024528598759442233, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.7965612115773528, + "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": 1.0718654827809488, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.525, + "pairwise_causal_calibration_confidence": 0.534871348160612, + "pairwise_causal_calibration_ece": 0.009871348160611883, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.1256343724003985, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "seed": "211245842/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13431161642074585, + "selected_utility_gain_over_base": -0.056596800684928894, + "selector_regret": 0.09957115352153778, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7661172300577164, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.3904761904761905, + "pairwise_causal_calibration_confidence": 0.5495148445011372, + "pairwise_causal_calibration_ece": 0.15903865402494677, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.8848141434142904, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "seed": "1151410346/train2", + "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": 0.32620343565940857, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4477952718734741, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5728588815635123, + "pairwise_causal_calibration_ece": 0.21571602442065527, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.31434308061981353, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19291165471076965, + "seed": "1324133122/train2", + "selected_candidate_index": 1, + "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": 1.0718654827809488, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.5331361711412006, + "pairwise_causal_calibration_ece": 0.08869172669675614, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.3518684532465013, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/train2", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.8125, + "selected_utility_gain_over_base": -1.1875, + "selector_regret": 1.1875, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 1.0718654827809488, + "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": 0.4880952380952381, + "pairwise_causal_calibration_confidence": 0.5532244830443317, + "pairwise_causal_calibration_ece": 0.0651292449490937, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.9787258921638226, + "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": 1.0718654827809488, + "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": 0.3815789473684211, + "pairwise_causal_calibration_confidence": 0.5298970718699219, + "pairwise_causal_calibration_ece": 0.14831812450150075, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": 0.8419564487563823, + "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": 1.0718654827809488, + "train_seed": "2" + } + ], + "schema_version": 1, + "selected_lambda": 10.0, + "selector_calibration_target_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "selector_calibration_target_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "selector_chart_feature_mode": "base_context_obs", + "selector_eval_target_content_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "selector_eval_target_split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "selector_source_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "selector_source_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "source_content_hash": null, + "source_index": null, + "source_split_hash": null, + "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "success_bonus": 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": 0.6326388888888889, + "low": 0.46388888888888885, + "mean": 0.545138888888889, + "n": 120, + "num_samples": 300 + }, + "mean": 0.545138888888889, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 1.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 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": 0.5, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "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": 0.0, + "307068879/train1": 0.0, + "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": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 0.5, + "636680137/train0": 1.0, + "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": 0.6666666666666666, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.6906666666666667, + "low": 0.18521212121212122, + "mean": 0.43410101010101004, + "n": 5, + "num_samples": 300 + }, + "mean": 0.43410101010101004, + "per_task": { + "LiftPegUpright-v1": 0.4666666666666667, + "PickCube-v1": 0.0, + "PullCube-v1": 0.6533333333333333, + "PushCube-v1": 0.7777777777777778, + "StackCube-v1": 0.2727272727272727 + } + }, + "metric": "coverage", + "micro": { + "confidence": 0.95, + "high": 0.6319444444444444, + "low": 0.4722222222222222, + "mean": 0.5486111111111112, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "execute_generated": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.6326388888888889, + "low": 0.46388888888888885, + "mean": 0.545138888888889, + "n": 120, + "num_samples": 300 + }, + "mean": 0.545138888888889, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 1.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 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": 0.5, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "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": 0.0, + "307068879/train1": 0.0, + "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": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 0.5, + "636680137/train0": 1.0, + "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": 0.6666666666666666, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.6906666666666667, + "low": 0.18521212121212122, + "mean": 0.43410101010101004, + "n": 5, + "num_samples": 300 + }, + "mean": 0.43410101010101004, + "per_task": { + "LiftPegUpright-v1": 0.4666666666666667, + "PickCube-v1": 0.0, + "PullCube-v1": 0.6533333333333333, + "PushCube-v1": 0.7777777777777778, + "StackCube-v1": 0.2727272727272727 + } + }, + "metric": "execute_generated", + "micro": { + "confidence": 0.95, + "high": 0.6319444444444444, + "low": 0.4722222222222222, + "mean": 0.5486111111111112, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "fallback_rate": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5361111111111112, + "low": 0.3673611111111111, + "mean": 0.4548611111111111, + "n": 120, + "num_samples": 300 + }, + "mean": 0.4548611111111111, + "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": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 1.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 1.0, + "141984615/train1": 0.0, + "141984615/train2": 1.0, + "1461069713/train0": 1.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 1.0, + "1522272686/train1": 1.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 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": 0.5, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.5, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 1.0, + "307068879/train1": 1.0, + "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": 1.0, + "441046872/train1": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.5, + "636680137/train0": 0.0, + "636680137/train1": 1.0, + "636680137/train2": 1.0, + "655884605/train0": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.3333333333333333, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.75 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.8147878787878788, + "low": 0.30933333333333335, + "mean": 0.5658989898989899, + "n": 5, + "num_samples": 300 + }, + "mean": 0.5658989898989899, + "per_task": { + "LiftPegUpright-v1": 0.5333333333333333, + "PickCube-v1": 1.0, + "PullCube-v1": 0.3466666666666667, + "PushCube-v1": 0.2222222222222222, + "StackCube-v1": 0.7272727272727273 + } + }, + "metric": "fallback_rate", + "micro": { + "confidence": 0.95, + "high": 0.5277777777777778, + "low": 0.3680555555555556, + "mean": 0.4513888888888889, + "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": 0.5448702331354158, + "low": 0.49797706822658805, + "mean": 0.5231682660617707, + "n": 117, + "num_samples": 300 + }, + "mean": 0.5231682660617707, + "per_seed": { + "1073161778/train0": 0.6134453781512605, + "1073161778/train1": 0.6833333333333333, + "1073161778/train2": 0.6071428571428571, + "1113371261/train0": 0.4642857142857143, + "1113371261/train1": 0.6785714285714286, + "1113371261/train2": 0.6785714285714286, + "1151410346/train0": 0.5428571428571428, + "1151410346/train1": 0.38095238095238093, + "1151410346/train2": 0.3904761904761905, + "1261930456/train0": 0.45454545454545453, + "1261930456/train1": 0.4909090909090909, + "1261930456/train2": 0.6071428571428571, + "1324133122/train0": 0.42857142857142855, + "1324133122/train1": 0.5714285714285714, + "1324133122/train2": 0.35714285714285715, + "1324642411/train0": 0.5666666666666667, + "1324642411/train1": 0.6218487394957983, + "1324642411/train2": 0.509090909090909, + "1375492786/train0": 0.5866666666666667, + "1375492786/train1": 0.5, + "1375492786/train2": 0.7023809523809523, + "141984615/train0": 0.2833333333333333, + "141984615/train1": 0.6083333333333333, + "141984615/train2": 0.425, + "1461069713/train0": 0.6666666666666666, + "1461069713/train1": 0.3333333333333333, + "1461069713/train2": 0.5083333333333333, + "1480856810/train0": 0.8282828282828283, + "1480856810/train1": 0.5882352941176471, + "1480856810/train2": 0.5714285714285714, + "1509980664/train0": 0.4444444444444444, + "1509980664/train1": 0.19047619047619047, + "1509980664/train2": 0.5444444444444444, + "1522272686/train0": 0.6666666666666666, + "1522272686/train1": 0.4166666666666667, + "1522272686/train2": 0.5, + "1549324805/train0": 0.43333333333333335, + "1549324805/train1": 0.30833333333333335, + "1549324805/train2": 0.4583333333333333, + "1578990937/train0": 0.7350427350427351, + "1578990937/train1": 0.4090909090909091, + "1578990937/train2": 0.42857142857142855, + "1790188787/train0": 0.2818181818181818, + "1790188787/train1": 0.3508771929824561, + "1790188787/train2": 0.43636363636363634, + "1822210471/train0": 0.575, + "1822210471/train1": 0.6938775510204082, + "1822210471/train2": 0.4, + "1845489013/train0": 0.41304347826086957, + "1845489013/train1": 0.7692307692307693, + "1845489013/train2": 0.72, + "1999022574/train0": 0.35714285714285715, + "1999022574/train1": 0.6714285714285714, + "1999022574/train2": 0.3815789473684211, + "2012211287/train0": 0.48739495798319327, + "2012211287/train1": 0.6083333333333333, + "2012211287/train2": 0.7717391304347826, + "2038594191/train0": 0.27586206896551724, + "2038594191/train1": 0.24074074074074073, + "2038594191/train2": 0.40404040404040403, + "2054899126/train0": 0.26666666666666666, + "2054899126/train1": 0.6754385964912281, + "2054899126/train2": 0.5210084033613446, + "2061635888/train0": 0.4444444444444444, + "2061635888/train1": 0.4166666666666667, + "2061635888/train2": 0.4444444444444444, + "211245842/train0": 0.6302521008403361, + "211245842/train1": 0.65, + "211245842/train2": 0.525, + "278008524/train0": 0.6191919191919192, + "278008524/train1": 0.5835144927536232, + "278008524/train2": 0.5271464646464646, + "307068879/train0": 0.5494505494505495, + "307068879/train1": 0.5274725274725275, + "307068879/train2": 0.3956043956043956, + "412102563/train0": 0.65, + "412102563/train1": 0.7948717948717948, + "412102563/train2": 0.4880952380952381, + "440577711/train0": 0.6551724137931034, + "440577711/train1": 0.5333333333333333, + "440577711/train2": 0.5185185185185185, + "441046872/train0": 0.38181818181818183, + "441046872/train1": 0.6181818181818182, + "441046872/train2": 0.45454545454545453, + "509430392/train0": 0.6386554621848739, + "509430392/train1": 0.5714285714285714, + "509430392/train2": 0.6134453781512605, + "585904130/train0": 0.36904761904761907, + "585904130/train1": 0.64, + "585904130/train2": 0.7272727272727273, + "608016290/train0": 0.4857142857142857, + "608016290/train1": 0.5047619047619047, + "608016290/train2": 0.55, + "626486813/train0": 0.65, + "626486813/train1": 0.32916666666666666, + "626486813/train2": 0.625, + "636680137/train0": 0.425, + "636680137/train1": 0.625, + "636680137/train2": 0.475, + "655884605/train0": 0.5833333333333334, + "655884605/train1": 0.5277777777777778, + "655884605/train2": 0.4722222222222222, + "672328262/train0": 0.8235294117647058, + "672328262/train1": 0.5428571428571428, + "672328262/train2": 0.2, + "701667486/train0": 0.38095238095238093, + "701667486/train1": 0.5357142857142857, + "701667486/train2": 0.4444444444444444, + "747260437/train0": 0.5083333333333333, + "747260437/train1": 0.675, + "747260437/train2": 0.7083333333333334, + "750448296/train0": 0.5793650793650794, + "750448296/train1": 0.4423076923076923, + "750448296/train2": 0.4833333333333334, + "767747424/train0": 0.49059829059829063, + "767747424/train1": 0.5700877145730087, + "767747424/train2": 0.5193142641672053 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5431888679749804, + "low": 0.5159382160536211, + "mean": 0.5299078728020017, + "n": 5, + "num_samples": 300 + }, + "mean": 0.5299078728020017, + "per_task": { + "LiftPegUpright-v1": 0.5492625405285252, + "PickCube-v1": 0.5277777777777778, + "PullCube-v1": 0.5216143238307843, + "PushCube-v1": 0.5465423944585421, + "StackCube-v1": 0.5043423274143793 + } + }, + "metric": "pairwise_causal_calibration_accuracy", + "micro": { + "confidence": 0.95, + "high": 0.5451467233807504, + "low": 0.5030967497956031, + "mean": 0.5238747803660404, + "n": 138, + "num_samples": 300 + }, + "n": 138 + }, + "pairwise_causal_calibration_confidence": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5604558890777185, + "low": 0.550776431898817, + "mean": 0.5552154770559953, + "n": 117, + "num_samples": 300 + }, + "mean": 0.5552154770559953, + "per_seed": { + "1073161778/train0": 0.5462334122382229, + "1073161778/train1": 0.5414540870914354, + "1073161778/train2": 0.5552324860178256, + "1113371261/train0": 0.5542145473380933, + "1113371261/train1": 0.5534020512884718, + "1113371261/train2": 0.5590997153025572, + "1151410346/train0": 0.5504892981594165, + "1151410346/train1": 0.5497312288122934, + "1151410346/train2": 0.5495148445011372, + "1261930456/train0": 0.5514987071348044, + "1261930456/train1": 0.5481788586308872, + "1261930456/train2": 0.5450693806396976, + "1324133122/train0": 0.5765433563593303, + "1324133122/train1": 0.5703079922305777, + "1324133122/train2": 0.5728588815635123, + "1324642411/train0": 0.5461265642044052, + "1324642411/train1": 0.5434572117117997, + "1324642411/train2": 0.5436846632267275, + "1375492786/train0": 0.5192841401886225, + "1375492786/train1": 0.5386998051668176, + "1375492786/train2": 0.5315351438061215, + "141984615/train0": 0.5448471499959241, + "141984615/train1": 0.5656089960064071, + "141984615/train2": 0.5603553061579338, + "1461069713/train0": 0.523550311048989, + "1461069713/train1": 0.5236824574508954, + "1461069713/train2": 0.5276472177677529, + "1480856810/train0": 0.5709539167236203, + "1480856810/train1": 0.543456481052637, + "1480856810/train2": 0.5370429153145398, + "1509980664/train0": 0.5523718802888111, + "1509980664/train1": 0.5517045024701317, + "1509980664/train2": 0.5506485429344871, + "1522272686/train0": 0.550550217481388, + "1522272686/train1": 0.5518834340111771, + "1522272686/train2": 0.5635063449255251, + "1549324805/train0": 0.5370476153908349, + "1549324805/train1": 0.5437634159530779, + "1549324805/train2": 0.540090634882582, + "1578990937/train0": 0.575799443033985, + "1578990937/train1": 0.5644259196254204, + "1578990937/train2": 0.5644061750173082, + "1790188787/train0": 0.5215209331218725, + "1790188787/train1": 0.517198535735028, + "1790188787/train2": 0.5206471195621108, + "1822210471/train0": 0.5678773715560055, + "1822210471/train1": 0.592873970483461, + "1822210471/train2": 0.5677066182834786, + "1845489013/train0": 0.51605426349952, + "1845489013/train1": 0.5252519047235767, + "1845489013/train2": 0.5274458415050698, + "1999022574/train0": 0.543208103565328, + "1999022574/train1": 0.5400358084963361, + "1999022574/train2": 0.5298970718699219, + "2012211287/train0": 0.533600718072752, + "2012211287/train1": 0.5379143734950628, + "2012211287/train2": 0.5403873973296214, + "2038594191/train0": 0.5879562438719218, + "2038594191/train1": 0.6117747139844449, + "2038594191/train2": 0.5954970669139762, + "2054899126/train0": 0.5291897348449885, + "2054899126/train1": 0.5263065559429049, + "2054899126/train2": 0.5536973137638624, + "2061635888/train0": 0.5639694431421344, + "2061635888/train1": 0.5581364424191588, + "2061635888/train2": 0.5794706736680671, + "211245842/train0": 0.5426955700305087, + "211245842/train1": 0.5351083386938548, + "211245842/train2": 0.534871348160612, + "278008524/train0": 0.5909019399202626, + "278008524/train1": 0.5847988892880861, + "278008524/train2": 0.59445093065676, + "307068879/train0": 0.5728850640530763, + "307068879/train1": 0.5704103700900957, + "307068879/train2": 0.5723883106472546, + "412102563/train0": 0.5493461173759878, + "412102563/train1": 0.5593981622164304, + "412102563/train2": 0.5532244830443317, + "440577711/train0": 0.5624630563402808, + "440577711/train1": 0.5471459072582541, + "440577711/train2": 0.5475473443233374, + "441046872/train0": 0.5664245191730715, + "441046872/train1": 0.5652066435973622, + "441046872/train2": 0.5755444444938004, + "509430392/train0": 0.5365694240669033, + "509430392/train1": 0.5326653211281538, + "509430392/train2": 0.5403636942457117, + "585904130/train0": 0.5472546704459165, + "585904130/train1": 0.549320518336562, + "585904130/train2": 0.5513841945828967, + "608016290/train0": 0.594216772727526, + "608016290/train1": 0.5978828235074669, + "608016290/train2": 0.6068596136210345, + "626486813/train0": 0.6158650684827733, + "626486813/train1": 0.6045161568297507, + "626486813/train2": 0.6142120016040145, + "636680137/train0": 0.540894445846163, + "636680137/train1": 0.5301042033378723, + "636680137/train2": 0.5238615170185813, + "655884605/train0": 0.5958635862373126, + "655884605/train1": 0.6117746577593474, + "655884605/train2": 0.5899703889806722, + "672328262/train0": 0.5940991028072462, + "672328262/train1": 0.5924504110795782, + "672328262/train2": 0.6012678269199632, + "701667486/train0": 0.5278746131288938, + "701667486/train1": 0.5308683756394741, + "701667486/train2": 0.5331361711412006, + "747260437/train0": 0.5322967641018285, + "747260437/train1": 0.5502411276617256, + "747260437/train2": 0.5279460992108118, + "750448296/train0": 0.5359906885657441, + "750448296/train1": 0.5377836086909143, + "750448296/train2": 0.5376337500179175, + "767747424/train0": 0.5592639661990686, + "767747424/train1": 0.5526263680260839, + "767747424/train2": 0.5607639712421639 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5621158691203029, + "low": 0.5492708813906398, + "mean": 0.5555302194109235, + "n": 5, + "num_samples": 300 + }, + "mean": 0.5555302194109235, + "per_task": { + "LiftPegUpright-v1": 0.5651094024103738, + "PickCube-v1": 0.5553133321393634, + "PullCube-v1": 0.5584521161129652, + "PushCube-v1": 0.5419772241344875, + "StackCube-v1": 0.5567990222574275 + } + }, + "metric": "pairwise_causal_calibration_confidence", + "micro": { + "confidence": 0.95, + "high": 0.5613581362340928, + "low": 0.5522925498793144, + "mean": 0.5565632957484361, + "n": 138, + "num_samples": 300 + }, + "n": 138 + }, + "pairwise_causal_calibration_ece": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.14460797017020918, + "low": 0.11646635724751547, + "mean": 0.13076108178779555, + "n": 120, + "num_samples": 300 + }, + "mean": 0.13076108178779555, + "per_seed": { + "1073161778/train0": 0.06721196591303755, + "1073161778/train1": 0.14187924624189804, + "1073161778/train2": 0.051910371125031565, + "1113371261/train0": 0.08992883305237898, + "1113371261/train1": 0.12516937728295682, + "1113371261/train2": 0.11947171326887146, + "1151410346/train0": 0.007632155302273683, + "1151410346/train1": 0.16877884785991246, + "1151410346/train2": 0.15903865402494677, + "1261930456/train0": 0.09695325258934998, + "1261930456/train1": 0.05726976772179637, + "1261930456/train2": 0.0620734765031595, + "1324133122/train0": 0.1479719277879018, + "1324133122/train1": 0.07659820557682445, + "1324133122/train2": 0.21571602442065527, + "1324642411/train0": 0.020540102462261362, + "1324642411/train1": 0.07839152778399854, + "1324642411/train2": 0.03459375413581838, + "1375492786/train0": 0.06738252647804421, + "1375492786/train1": 0.03869980516681758, + "1375492786/train2": 0.17084580857483078, + "141984615/train0": 0.2615138166625908, + "141984615/train1": 0.14136934515423885, + "141984615/train2": 0.1353553061579339, + "1461069713/train0": 0.14311635561767766, + "1461069713/train1": 0.19034912411756205, + "1461069713/train2": 0.019313884434419615, + "1480856810/train0": 0.2573289115592079, + "1480856810/train1": 0.04477881306501008, + "1480856810/train2": 0.03438565611403155, + "1509980664/train0": 0.10792743584436665, + "1509980664/train1": 0.36122831199394134, + "1509980664/train2": 0.024528598759442233, + "1522272686/train0": 0.11611644918527875, + "1522272686/train1": 0.13521676734451044, + "1522272686/train2": 0.06840618279050634, + "1549324805/train0": 0.10371428205750155, + "1549324805/train1": 0.2354300826197445, + "1549324805/train2": 0.08175730154924889, + "1578990937/train0": 0.15924329200874998, + "1578990937/train1": 0.15533501053451115, + "1578990937/train2": 0.13583474644587978, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.23970275130369073, + "1790188787/train1": 0.16632134275257193, + "1790188787/train2": 0.08428348319847451, + "1822210471/train0": 0.0719433679479887, + "1822210471/train1": 0.10100358053694718, + "1822210471/train2": 0.16770661828347844, + "1845489013/train0": 0.10301078523865043, + "1845489013/train1": 0.24397886450719253, + "1845489013/train2": 0.19255415849493018, + "1999022574/train0": 0.1860652464224708, + "1999022574/train1": 0.13139276293223526, + "1999022574/train2": 0.14831812450150075, + "2012211287/train0": 0.04620576008955871, + "2012211287/train1": 0.07041895983827043, + "2012211287/train2": 0.2313517331051612, + "2038594191/train0": 0.31209417490640445, + "2038594191/train1": 0.37103397324370413, + "2038594191/train2": 0.19145666287357221, + "2054899126/train0": 0.13126153408916094, + "2054899126/train1": 0.07456602027416159, + "2054899126/train2": 0.026703688149182295, + "2061635888/train0": 0.130957329325739, + "2061635888/train1": 0.1414697757524921, + "2061635888/train2": 0.13502622922362276, + "211245842/train0": 0.08755653080982741, + "211245842/train1": 0.12495817852344455, + "211245842/train2": 0.009871348160611883, + "278008524/train0": 0.1344967059497781, + "278008524/train1": 0.06891187442684295, + "278008524/train2": 0.09739801499095443, + "307068879/train0": 0.1199981805489268, + "307068879/train1": 0.07309450817078264, + "307068879/train2": 0.1767839150428592, + "412102563/train0": 0.10951912707521982, + "412102563/train1": 0.23547363265536436, + "412102563/train2": 0.0651292449490937, + "440577711/train0": 0.09270935745282255, + "440577711/train1": 0.3939678976832891, + "440577711/train2": 0.17413687717265497, + "441046872/train0": 0.19522652344582997, + "441046872/train1": 0.09055210003018804, + "441046872/train2": 0.13114432369738743, + "509430392/train0": 0.10561970831698192, + "509430392/train1": 0.038763250300417725, + "509430392/train2": 0.14212516552457483, + "585904130/train0": 0.17820705139829746, + "585904130/train1": 0.19650416069860938, + "585904130/train2": 0.1758885326898305, + "608016290/train0": 0.11412216411454103, + "608016290/train1": 0.20716363895507375, + "608016290/train2": 0.0651220182950748, + "626486813/train0": 0.1759930830571895, + "626486813/train1": 0.2753494901630838, + "626486813/train2": 0.14970168842483308, + "636680137/train0": 0.11589444584616293, + "636680137/train1": 0.09489579666212766, + "636680137/train2": 0.048861517018581324, + "655884605/train0": 0.10068633489228643, + "655884605/train1": 0.228982599451425, + "655884605/train2": 0.19094254082377798, + "672328262/train0": 0.22943030895745967, + "672328262/train1": 0.07756824536937998, + "672328262/train2": 0.4012678269199631, + "701667486/train0": 0.1469222321765129, + "701667486/train1": 0.004845910074811617, + "701667486/train2": 0.08869172669675614, + "747260437/train0": 0.046545342019779196, + "747260437/train1": 0.2890079292574854, + "747260437/train2": 0.1803872341225216, + "750448296/train0": 0.07322390346905903, + "750448296/train1": 0.09632988234770806, + "750448296/train2": 0.07076889063044275, + "767747424/train0": 0.1274080019549793, + "767747424/train1": 0.15613421435842795, + "767747424/train2": 0.10791269448114733 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.1292576743373383, + "low": 0.11373643350416622, + "mean": 0.12135669260871884, + "n": 5, + "num_samples": 300 + }, + "mean": 0.12135669260871884, + "per_task": { + "LiftPegUpright-v1": 0.11881236057424142, + "PickCube-v1": 0.10657979977343184, + "PullCube-v1": 0.13244853449538532, + "PushCube-v1": 0.11799694975778399, + "StackCube-v1": 0.13094581844275163 + } + }, + "metric": "pairwise_causal_calibration_ece", + "micro": { + "confidence": 0.95, + "high": 0.1433183357132694, + "low": 0.11511341593299673, + "mean": 0.12833834721763002, + "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": 0.9056800281312739, + "low": 0.7479934311770733, + "mean": 0.8247400227037968, + "n": 120, + "num_samples": 300 + }, + "mean": 0.8247400227037968, + "per_seed": { + "1073161778/train0": 0.8315109279797959, + "1073161778/train1": 0.696723804413939, + "1073161778/train2": 0.8010166202874374, + "1113371261/train0": 0.028473507842246004, + "1113371261/train1": 0.018305243090398104, + "1113371261/train2": -0.16653094820357445, + "1151410346/train0": 0.8265443858364611, + "1151410346/train1": 0.9117051526313378, + "1151410346/train2": 0.8848141434142904, + "1261930456/train0": 0.7394668700396011, + "1261930456/train1": 0.7193134962639322, + "1261930456/train2": 0.7862881124399304, + "1324133122/train0": -0.3146091278603853, + "1324133122/train1": 0.05209384637709236, + "1324133122/train2": -0.31434308061981353, + "1324642411/train0": 1.8003040878491947, + "1324642411/train1": 1.3876858687209335, + "1324642411/train2": 0.7578621304129007, + "1375492786/train0": 0.13235098862844796, + "1375492786/train1": -0.48662334960024145, + "1375492786/train2": -0.36262329065526533, + "141984615/train0": 0.5415586302800286, + "141984615/train1": 0.8524230828537844, + "141984615/train2": 0.3015705513281932, + "1461069713/train0": -0.06695685840699794, + "1461069713/train1": 0.24721792540099097, + "1461069713/train2": 0.2696456255951446, + "1480856810/train0": 2.2046095259547753, + "1480856810/train1": 1.735589926228614, + "1480856810/train2": 1.359251321625822, + "1509980664/train0": 0.6801689818523341, + "1509980664/train1": 0.7807195223473116, + "1509980664/train2": 0.7965612115773528, + "1522272686/train0": 0.3705665707855397, + "1522272686/train1": 0.8136424781392986, + "1522272686/train2": 0.5686852402790281, + "1549324805/train0": 1.332576441483642, + "1549324805/train1": 0.9503772340326878, + "1549324805/train2": 1.4498268017490537, + "1578990937/train0": 1.3204376505810713, + "1578990937/train1": 1.2385033751871737, + "1578990937/train2": 1.1890371636445307, + "1747178030/train0": 0.6807136864797709, + "1747178030/train1": 0.7412627528047919, + "1747178030/train2": 0.7399153467725452, + "1790188787/train0": 1.1652542261433299, + "1790188787/train1": 1.1907212928313995, + "1790188787/train2": 1.0616754451066321, + "1822210471/train0": 0.7418095681624295, + "1822210471/train1": 0.7078283118607503, + "1822210471/train2": 0.714483922998961, + "1845489013/train0": 0.798908624848506, + "1845489013/train1": 0.8663205932771525, + "1845489013/train2": 0.8229653113797162, + "1999022574/train0": 0.8291022763943762, + "1999022574/train1": 0.7563954831040236, + "1999022574/train2": 0.8419564487563823, + "2012211287/train0": 1.107133532936841, + "2012211287/train1": 0.9306280702125403, + "2012211287/train2": 1.1151401969224932, + "2038594191/train0": 1.2008765491740636, + "2038594191/train1": 1.112157225492381, + "2038594191/train2": 1.2364955407573137, + "2054899126/train0": 1.014453498377106, + "2054899126/train1": 1.074027910955616, + "2054899126/train2": 1.0710380363686038, + "2061635888/train0": 0.39584763737906137, + "2061635888/train1": 0.3705725365164993, + "2061635888/train2": 0.345599063353308, + "211245842/train0": 1.1272780451605109, + "211245842/train1": 1.2265101367583866, + "211245842/train2": 1.1256343724003985, + "278008524/train0": 0.27572048526322235, + "278008524/train1": 0.3598249612248885, + "278008524/train2": 0.4752736549448759, + "307068879/train0": 0.7747416665744656, + "307068879/train1": 0.3600826916101192, + "307068879/train2": 0.979654396381198, + "412102563/train0": 1.0435143603129777, + "412102563/train1": 0.949568847497014, + "412102563/train2": 0.9787258921638226, + "440577711/train0": 1.7706684649919293, + "440577711/train1": 1.808986225138223, + "440577711/train2": 1.7310319203842686, + "441046872/train0": 0.6069194020559678, + "441046872/train1": 0.1629836409186699, + "441046872/train2": 0.5214657402479244, + "509430392/train0": 1.7250273458520244, + "509430392/train1": 1.5703557578982594, + "509430392/train2": 1.3794654730008236, + "585904130/train0": 0.9073249211388359, + "585904130/train1": 0.8486373501760932, + "585904130/train2": 0.9433176675944015, + "608016290/train0": 1.589765155203597, + "608016290/train1": 1.3392733370703795, + "608016290/train2": 1.4772393565445312, + "626486813/train0": 0.9601260167395063, + "626486813/train1": 0.9603159694517454, + "626486813/train2": 0.7735840442684991, + "636680137/train0": 1.0758545824030454, + "636680137/train1": 0.44432064636845636, + "636680137/train2": 0.5548863156073045, + "655884605/train0": 0.5292851601669687, + "655884605/train1": 0.8854182759065323, + "655884605/train2": 0.7967317370820184, + "672328262/train0": 0.7871393331842744, + "672328262/train1": 0.7852289914727505, + "672328262/train2": 0.6365332388231544, + "701667486/train0": 1.1356123561511042, + "701667486/train1": 1.234980143424118, + "701667486/train2": 1.3518684532465013, + "747260437/train0": 0.7403769985406177, + "747260437/train1": 0.6210780034951271, + "747260437/train2": 0.9133990308397049, + "750448296/train0": 0.8003581361777009, + "750448296/train1": 0.4055514323654445, + "750448296/train2": 0.5039790420368024, + "767747424/train0": 0.5457716259676435, + "767747424/train1": 0.6028818781511358, + "767747424/train2": 0.5655051629030958 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.1422375180911708, + "low": 0.6446955243124963, + "mean": 0.8504917118375032, + "n": 5, + "num_samples": 300 + }, + "mean": 0.8504917118375032, + "per_task": { + "LiftPegUpright-v1": 0.8862852359573267, + "PickCube-v1": 0.5842980964012888, + "PullCube-v1": 0.6759761060180944, + "PushCube-v1": 1.3829754161601442, + "StackCube-v1": 0.7229237046506615 + } + }, + "metric": "predicted_margin", + "micro": { + "confidence": 0.95, + "high": 0.8833200719102705, + "low": 0.7199132535269486, + "mean": 0.795107086974134, + "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": 6.045833333333333, + "low": 4.645833333333333, + "mean": 5.317361111111111, + "n": 120, + "num_samples": 300 + }, + "mean": 5.317361111111111, + "per_seed": { + "1073161778/train0": 6.0, + "1073161778/train1": 8.0, + "1073161778/train2": 3.0, + "1113371261/train0": 2.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 12.0, + "1151410346/train1": 4.0, + "1151410346/train2": 6.0, + "1261930456/train0": 1.0, + "1261930456/train1": 2.0, + "1261930456/train2": 4.0, + "1324133122/train0": 6.0, + "1324133122/train1": 2.0, + "1324133122/train2": 1.0, + "1324642411/train0": 11.0, + "1324642411/train1": 4.0, + "1324642411/train2": 2.0, + "1375492786/train0": 3.0, + "1375492786/train1": 15.0, + "1375492786/train2": 5.0, + "141984615/train0": 3.0, + "141984615/train1": 5.0, + "141984615/train2": 10.0, + "1461069713/train0": 7.0, + "1461069713/train1": 15.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 10.0, + "1480856810/train2": 4.0, + "1509980664/train0": 2.0, + "1509980664/train1": 9.0, + "1509980664/train2": 9.0, + "1522272686/train0": 8.0, + "1522272686/train1": 7.0, + "1522272686/train2": 7.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 1.0, + "1578990937/train0": 9.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 3.0, + "1747178030/train1": 3.0, + "1747178030/train2": 15.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 3.0, + "1822210471/train0": 1.0, + "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": 9.0, + "2038594191/train2": 2.0, + "2054899126/train0": 4.0, + "2054899126/train1": 11.0, + "2054899126/train2": 2.0, + "2061635888/train0": 1.0, + "2061635888/train1": 2.0, + "2061635888/train2": 7.0, + "211245842/train0": 5.0, + "211245842/train1": 1.0, + "211245842/train2": 7.0, + "278008524/train0": 10.0, + "278008524/train1": 4.5, + "278008524/train2": 14.0, + "307068879/train0": 4.0, + "307068879/train1": 9.0, + "307068879/train2": 10.0, + "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": 1.0, + "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": 4.0, + "701667486/train0": 10.0, + "701667486/train1": 11.0, + "701667486/train2": 15.0, + "747260437/train0": 9.0, + "747260437/train1": 4.0, + "747260437/train2": 14.0, + "750448296/train0": 0.0, + "750448296/train1": 7.333333333333333, + "750448296/train2": 9.0, + "767747424/train0": 8.25, + "767747424/train1": 3.75, + "767747424/train2": 8.75 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 6.545454545454545, + "low": 4.497333333333333, + "mean": 5.55389898989899, + "n": 5, + "num_samples": 300 + }, + "mean": 5.55389898989899, + "per_task": { + "LiftPegUpright-v1": 5.333333333333333, + "PickCube-v1": 7.333333333333333, + "PullCube-v1": 5.986666666666666, + "PushCube-v1": 3.7222222222222223, + "StackCube-v1": 5.393939393939394 + } + }, + "metric": "selected_candidate_index", + "micro": { + "confidence": 0.95, + "high": 6.263888888888889, + "low": 4.784722222222222, + "mean": 5.527777777777778, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.44166666666666665, + "low": 0.26666666666666666, + "mean": 0.35, + "n": 120, + "num_samples": 300 + }, + "mean": 0.35, + "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": 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": 1.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 1.0, + "1822210471/train2": 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": 0.0, + "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": 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": 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": 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": 0.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.0, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4545454545454545, + "low": 0.12121212121212119, + "mean": 0.290989898989899, + "n": 5, + "num_samples": 300 + }, + "mean": 0.290989898989899, + "per_task": { + "LiftPegUpright-v1": 0.3333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": 0.29333333333333333, + "PushCube-v1": 0.5555555555555556, + "StackCube-v1": 0.2727272727272727 + } + }, + "metric": "selected_success", + "micro": { + "confidence": 0.95, + "high": 0.3819444444444444, + "low": 0.24305555555555555, + "mean": 0.3194444444444444, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_success_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.10625, + "low": -0.025, + "mean": 0.04375, + "n": 120, + "num_samples": 300 + }, + "mean": 0.04375, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 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": 1.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 1.0, + "1822210471/train2": 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": -1.0, + "2038594191/train2": 0.0, + "2054899126/train0": -0.5, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 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": 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": 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": -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": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.26, + "low": -0.026666666666666665, + "mean": 0.09111111111111111, + "n": 5, + "num_samples": 300 + }, + "mean": 0.09111111111111111, + "per_task": { + "LiftPegUpright-v1": 0.13333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": -0.06666666666666667, + "PushCube-v1": 0.3888888888888889, + "StackCube-v1": 0.0 + } + }, + "metric": "selected_success_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.09027777777777778, + "low": -0.027777777777777776, + "mean": 0.027777777777777776, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.0168445986038326, + "low": 0.6897832043572433, + "mean": 0.8491037837639548, + "n": 120, + "num_samples": 300 + }, + "mean": 0.8491037837639548, + "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": 0.12164360284805298, + "1324642411/train1": 0.11658626794815063, + "1324642411/train2": 0.10647052526473999, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.10283424705266953, + "141984615/train1": 0.027254482731223106, + "141984615/train2": 0.10283424705266953, + "1461069713/train0": 0.0939030647277832, + "1461069713/train1": 0.0939030647277832, + "1461069713/train2": 0.0939030647277832, + "1480856810/train0": 2.0, + "1480856810/train1": 2.0, + "1480856810/train2": 0.21894000470638275, + "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": 0.0260902252048254, + "1549324805/train1": 0.0303483996540308, + "1549324805/train2": 0.04690708592534065, + "1578990937/train0": 2.0, + "1578990937/train1": 0.12732218205928802, + "1578990937/train2": 0.13359522819519043, + "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": 0.06306678056716919, + "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": 0.07674666494131088, + "2038594191/train2": 2.0, + "2054899126/train0": 1.0643903091549873, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 0.05225050449371338, + "2061635888/train1": 0.05225050449371338, + "2061635888/train2": 0.05225050449371338, + "211245842/train0": 0.8592818975448608, + "211245842/train1": 0.09285585582256317, + "211245842/train2": 0.13431161642074585, + "278008524/train0": 0.05350120924413204, + "278008524/train1": 0.034389604814350605, + "278008524/train2": 0.047564467415213585, + "307068879/train0": 0.6000905632972717, + "307068879/train1": 0.6000905632972717, + "307068879/train2": 0.2021283358335495, + "412102563/train0": 2.0, + "412102563/train1": 2.0, + "412102563/train2": 2.0, + "440577711/train0": 2.0, + "440577711/train1": 2.0, + "440577711/train2": 2.0, + "441046872/train0": 0.6999551653862, + "441046872/train1": 0.6999551653862, + "441046872/train2": 0.6999551653862, + "509430392/train0": 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": 0.707331120967865, + "608016290/train1": 2.0, + "608016290/train2": 0.6789124011993408, + "626486813/train0": 0.007258951431140304, + "626486813/train1": 0.011037707328796387, + "626486813/train2": 0.012973218690603971, + "636680137/train0": 0.056092143058776855, + "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": 0.04013325646519661, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 0.8125, + "747260437/train0": 0.07023122161626816, + "747260437/train1": 0.05181918665766716, + "747260437/train2": 0.0880771279335022, + "750448296/train0": 0.09101381525397301, + "750448296/train1": 0.10720233743389447, + "750448296/train2": 0.09631064782540004, + "767747424/train0": 0.0617859915655572, + "767747424/train1": 0.5621845133136958, + "767747424/train2": 0.567537541501224 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.0499388170904582, + "low": 0.6472050376432016, + "mean": 0.8435512482026176, + "n": 5, + "num_samples": 300 + }, + "mean": 0.8435512482026176, + "per_task": { + "LiftPegUpright-v1": 1.088020184636116, + "PickCube-v1": 0.5839591026306152, + "PullCube-v1": 0.6341658942727372, + "PushCube-v1": 1.162042656706439, + "StackCube-v1": 0.7495684027671814 + } + }, + "metric": "selected_utility", + "micro": { + "confidence": 0.95, + "high": 0.9005737514243164, + "low": 0.6412263065835254, + "mean": 0.7728274115272344, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_utility_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.16965195490823437, + "low": -0.06579780244865105, + "mean": 0.058985721466160615, + "n": 120, + "num_samples": 300 + }, + "mean": 0.058985721466160615, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.015173077583312988, + "1324642411/train1": 0.010115742683410645, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": -0.07557976432144642, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 1.9129659235477448, + "1480856810/train1": 1.9129659235477448, + "1480856810/train2": 0.1319059282541275, + "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": -0.07842461206018925, + "1549324805/train1": -0.07416643761098385, + "1549324805/train2": -0.057607751339673996, + "1578990937/train0": 1.17586350440979, + "1578990937/train1": -0.6968143135309219, + "1578990937/train2": -0.6905412673950195, + "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": -0.008848525583744049, + "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": -1.9232533350586891, + "2038594191/train2": 0.0, + "2054899126/train0": -0.9356096908450127, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.6683734804391861, + "211245842/train1": -0.09805256128311157, + "211245842/train2": -0.056596800684928894, + "278008524/train0": 0.010766219347715378, + "278008524/train1": -0.008345385082066059, + "278008524/train2": 0.004829477518796921, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": -0.39796222746372223, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 1.9638472199440002, + "440577711/train1": 1.9638472199440002, + "440577711/train2": 1.9638472199440002, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 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": -0.018370509147644043, + "608016290/train1": 1.274298369884491, + "608016290/train2": -0.04678922891616821, + "626486813/train0": 0.002160608710255474, + "626486813/train1": 0.005939364607911557, + "626486813/train2": 0.007874875969719142, + "636680137/train0": -0.027042903006076813, + "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": -1.9598667435348034, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": -1.1875, + "747260437/train0": -1.9297687783837318, + "747260437/train1": -1.9481808133423328, + "747260437/train2": -1.9119228720664978, + "750448296/train0": -0.026579334090153377, + "750448296/train1": -0.010390811910231909, + "750448296/train2": -0.02128250151872635, + "767747424/train0": -0.5114133504393976, + "767747424/train1": -0.011014828691259027, + "767747424/train2": -0.005661800503730774 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4840477555899909, + "low": -0.06713234748386523, + "mean": 0.14799145102860276, + "n": 5, + "num_samples": 300 + }, + "mean": 0.14799145102860276, + "per_task": { + "LiftPegUpright-v1": 0.1381496459245682, + "PickCube-v1": 0.0, + "PullCube-v1": -0.14100456695072353, + "PushCube-v1": 0.7696384779281087, + "StackCube-v1": -0.02682630175893957 + } + }, + "metric": "selected_utility_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.13739498196405797, + "low": -0.08578009277456052, + "mean": 0.031007825084897276, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selector_regret": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5576142528169536, + "low": 0.31175716476257853, + "mean": 0.4202619706426857, + "n": 120, + "num_samples": 300 + }, + "mean": 0.4202619706426857, + "per_seed": { + "1073161778/train0": 0.0, + "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": 1.8834137320518494, + "1324642411/train2": 1.89352947473526, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.009756645187735558, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.006073415279388428, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 1.7810599952936172, + "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": 0.0347811970859766, + "1549324805/train1": 0.16857311688363552, + "1549324805/train2": 0.06619475409388542, + "1578990937/train0": 0.0, + "1578990937/train1": 1.872677817940712, + "1578990937/train2": 1.8664047718048096, + "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": 1.9369332194328308, + "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": 1.9232533350586891, + "2038594191/train2": 0.0, + "2054899126/train0": 0.012892842292785645, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.04574880003929138, + "2061635888/train1": 0.02355286478996277, + "2061635888/train2": 0.05880081653594971, + "211245842/train0": 1.1407181024551392, + "211245842/train1": 0.12010131776332855, + "211245842/train2": 0.09957115352153778, + "278008524/train0": 1.0198551137000322, + "278008524/train1": 1.0032849898561835, + "278008524/train2": 1.0883844438940287, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.02808511257171631, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 1.3000448346138, + "441046872/train1": 0.023403704166412354, + "441046872/train2": 0.019179701805114746, + "509430392/train0": 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": 1.292668879032135, + "608016290/train1": 0.0, + "608016290/train2": 0.03281521797180176, + "626486813/train0": 0.037436524173244834, + "626486813/train1": 0.024790167342871428, + "626486813/train2": 0.03192603634670377, + "636680137/train0": 0.0321691632270813, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 0.0, + "655884605/train1": 0.6498691290616989, + "655884605/train2": 0.05146662890911102, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 1.9598667435348034, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 1.1875, + "747260437/train0": 0.0381644144654274, + "747260437/train1": 0.011470336467027664, + "747260437/train2": 0.03702475130558014, + "750448296/train0": 0.6616465126474699, + "750448296/train1": 1.2633009366691113, + "750448296/train2": 1.2541380201776822, + "767747424/train0": 0.980643196235178, + "767747424/train1": 0.5007685779128224, + "767747424/train2": 0.49962955340743065 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5367166141374037, + "low": 0.1779043208972805, + "mean": 0.3728429147021167, + "n": 5, + "num_samples": 300 + }, + "mean": 0.3728429147021167, + "per_task": { + "LiftPegUpright-v1": 0.3509542107582092, + "PickCube-v1": 0.0, + "PullCube-v1": 0.4787018513850247, + "PushCube-v1": 0.6237387582659721, + "StackCube-v1": 0.41081975310137775 + } + }, + "metric": "selector_regret", + "micro": { + "confidence": 0.95, + "high": 0.587600458064925, + "low": 0.3392519335482373, + "mean": 0.45799514941932606, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "success_selector_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.3125, + "low": 0.16597222222222227, + "mean": 0.23472222222222222, + "n": 120, + "num_samples": 300 + }, + "mean": 0.23472222222222222, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 1.0, + "1151410346/train2": 0.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 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": 0.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 0.0, + "1822210471/train2": 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": 1.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 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": 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": 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": 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": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.3013333333333333, + "low": 0.10181818181818181, + "mean": 0.21648484848484845, + "n": 5, + "num_samples": 300 + }, + "mean": 0.21648484848484845, + "per_task": { + "LiftPegUpright-v1": 0.26666666666666666, + "PickCube-v1": 0.0, + "PullCube-v1": 0.24, + "PushCube-v1": 0.3333333333333333, + "StackCube-v1": 0.24242424242424243 + } + }, + "metric": "success_selector_gap", + "micro": { + "confidence": 0.95, + "high": 0.3194444444444444, + "low": 0.1875, + "mean": 0.25, + "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": 0.8498021009856326, + "low": 0.7752087490002378, + "mean": 0.8107982548634141, + "n": 120, + "num_samples": 300 + }, + "mean": 0.8107982548634141, + "per_seed": { + "1073161778/train0": 0.87335466614541, + "1073161778/train1": 0.87335466614541, + "1073161778/train2": 0.87335466614541, + "1113371261/train0": 1.0718654827809488, + "1113371261/train1": 1.0718654827809488, + "1113371261/train2": 1.0718654827809488, + "1151410346/train0": 1.0718654827809488, + "1151410346/train1": 1.0718654827809488, + "1151410346/train2": 1.0718654827809488, + "1261930456/train0": 0.8468573198810986, + "1261930456/train1": 0.8468573198810986, + "1261930456/train2": 0.8468573198810986, + "1324133122/train0": 1.0718654827809488, + "1324133122/train1": 1.0718654827809488, + "1324133122/train2": 1.0718654827809488, + "1324642411/train0": 0.8468573198810986, + "1324642411/train1": 0.8468573198810986, + "1324642411/train2": 0.8468573198810986, + "1375492786/train0": 0.5825884328057473, + "1375492786/train1": 0.5825884328057473, + "1375492786/train2": 0.5825884328057473, + "141984615/train0": 0.5825884328057473, + "141984615/train1": 0.5825884328057473, + "141984615/train2": 0.5825884328057473, + "1461069713/train0": 0.5825884328057473, + "1461069713/train1": 0.5825884328057473, + "1461069713/train2": 0.5825884328057473, + "1480856810/train0": 0.8468573198810986, + "1480856810/train1": 0.8468573198810986, + "1480856810/train2": 0.8468573198810986, + "1509980664/train0": 1.0718654827809488, + "1509980664/train1": 1.0718654827809488, + "1509980664/train2": 1.0718654827809488, + "1522272686/train0": 1.2894892762347252, + "1522272686/train1": 1.2894892762347252, + "1522272686/train2": 1.2894892762347252, + "1549324805/train0": 0.5825884328057473, + "1549324805/train1": 0.5825884328057473, + "1549324805/train2": 0.5825884328057473, + "1578990937/train0": 1.0718654827809488, + "1578990937/train1": 1.0718654827809488, + "1578990937/train2": 1.0718654827809488, + "1747178030/train0": 0.5825884328057473, + "1747178030/train1": 0.5825884328057473, + "1747178030/train2": 0.5825884328057473, + "1790188787/train0": 0.8468573198810986, + "1790188787/train1": 0.8468573198810986, + "1790188787/train2": 0.8468573198810986, + "1822210471/train0": 0.5825884328057473, + "1822210471/train1": 0.5825884328057473, + "1822210471/train2": 0.5825884328057473, + "1845489013/train0": 0.5825884328057473, + "1845489013/train1": 0.5825884328057473, + "1845489013/train2": 0.5825884328057473, + "1999022574/train0": 1.0718654827809488, + "1999022574/train1": 1.0718654827809488, + "1999022574/train2": 1.0718654827809488, + "2012211287/train0": 0.87335466614541, + "2012211287/train1": 0.87335466614541, + "2012211287/train2": 0.87335466614541, + "2038594191/train0": 0.5825884328057473, + "2038594191/train1": 0.5825884328057473, + "2038594191/train2": 0.5825884328057473, + "2054899126/train0": 0.5825884328057473, + "2054899126/train1": 0.5825884328057473, + "2054899126/train2": 0.5825884328057473, + "2061635888/train0": 1.0718654827809488, + "2061635888/train1": 1.0718654827809488, + "2061635888/train2": 1.0718654827809488, + "211245842/train0": 1.0718654827809488, + "211245842/train1": 1.0718654827809488, + "211245842/train2": 1.0718654827809488, + "278008524/train0": 0.5825884328057473, + "278008524/train1": 0.5825884328057473, + "278008524/train2": 0.5825884328057473, + "307068879/train0": 0.87335466614541, + "307068879/train1": 0.87335466614541, + "307068879/train2": 0.87335466614541, + "412102563/train0": 1.0718654827809488, + "412102563/train1": 1.0718654827809488, + "412102563/train2": 1.0718654827809488, + "440577711/train0": 0.8468573198810986, + "440577711/train1": 0.8468573198810986, + "440577711/train2": 0.8468573198810986, + "441046872/train0": 0.87335466614541, + "441046872/train1": 0.87335466614541, + "441046872/train2": 0.87335466614541, + "509430392/train0": 0.8468573198810986, + "509430392/train1": 0.8468573198810986, + "509430392/train2": 0.8468573198810986, + "585904130/train0": 0.5825884328057473, + "585904130/train1": 0.5825884328057473, + "585904130/train2": 0.5825884328057473, + "608016290/train0": 0.87335466614541, + "608016290/train1": 0.87335466614541, + "608016290/train2": 0.87335466614541, + "626486813/train0": 0.5825884328057473, + "626486813/train1": 0.5825884328057473, + "626486813/train2": 0.5825884328057473, + "636680137/train0": 0.5825884328057473, + "636680137/train1": 0.5825884328057473, + "636680137/train2": 0.5825884328057473, + "655884605/train0": 1.0718654827809488, + "655884605/train1": 1.0718654827809488, + "655884605/train2": 1.0718654827809488, + "672328262/train0": 0.5825884328057473, + "672328262/train1": 0.5825884328057473, + "672328262/train2": 0.5825884328057473, + "701667486/train0": 1.0718654827809488, + "701667486/train1": 1.0718654827809488, + "701667486/train2": 1.0718654827809488, + "747260437/train0": 0.5825884328057473, + "747260437/train1": 0.5825884328057473, + "747260437/train2": 0.5825884328057473, + "750448296/train0": 0.5825884328057473, + "750448296/train1": 0.5825884328057473, + "750448296/train2": 0.5825884328057473, + "767747424/train0": 0.5825884328057473, + "767747424/train1": 0.5825884328057473, + "767747424/train2": 0.5825884328057473 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.1192128368353518, + "low": 0.7332976202158582, + "mean": 0.9328310355695862, + "n": 5, + "num_samples": 300 + }, + "mean": 0.9328310355695862, + "per_task": { + "LiftPegUpright-v1": 0.87335466614541, + "PickCube-v1": 1.2894892762347252, + "PullCube-v1": 0.5825884328057477, + "PushCube-v1": 0.8468573198810989, + "StackCube-v1": 1.0718654827809493 + } + }, + "metric": "tau", + "micro": { + "confidence": 0.95, + "high": 0.8080168440864098, + "low": 0.7418667305422796, + "mean": 0.7727632845204696, + "n": 144, + "num_samples": 300 + }, + "n": 144 + } + }, + "target": "success_weighted_margin", + "tau": { + "LiftPegUpright-v1": 0.87335466614541, + "PickCube-v1": 1.2894892762347252, + "PullCube-v1": 0.5825884328057473, + "PushCube-v1": 0.8468573198810986, + "StackCube-v1": 1.0718654827809488, + "__global__": 0.9334263687578273 + }, + "threshold_scope": "task", + "weights": [ + 0.4845035334970376, + 0.06838088727345144, + 0.16678883668530634, + 0.1363771742142653, + -0.1715406741040887, + -0.05724106927282932, + -0.13180922932833544, + -0.4557284010396998, + 0.2706929796541009, + 0.12341385944262218, + 0.12946277027885256, + 0.09769273471009002, + -0.11888136575346309, + -0.11238405821382269, + 0.011849448176968633, + -0.011698642767114149, + -0.11263857629083951, + 0.04832107985993872, + 1.4789877601250618e-14, + -0.6680233898598331, + -0.7755119424783743, + 0.6029408716686788, + 0.8450649960404633, + -0.0803300694570015, + -1.4306019886895758, + -0.45181180635753737, + 0.7706978279797312, + 0.5941190290408258, + 0.23021888951801373, + 0.8402692970746983, + -1.1288945246160321e-14, + 0.5380600170663836, + 0.6962065975153082, + -0.13222563070253313, + -0.22016576761284312, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_seed.json b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_seed.json new file mode 100644 index 0000000000000000000000000000000000000000..d45998d9bea68a48ab05e34d6b01ce7d02316632 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_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": 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": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.5462334122382229, + "pairwise_causal_calibration_ece": 0.06721196591303755, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.8315109279797959, + "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": 0.87335466614541 + }, + "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": 0.5414540870914354, + "pairwise_causal_calibration_ece": 0.14187924624189804, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.696723804413939, + "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": 0.87335466614541 + }, + "1073161778/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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5552324860178256, + "pairwise_causal_calibration_ece": 0.051910371125031565, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.8010166202874374, + "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": 0.87335466614541 + }, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5542145473380933, + "pairwise_causal_calibration_ece": 0.08992883305237898, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.028473507842246004, + "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": 1.0718654827809488 + }, + "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": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5534020512884718, + "pairwise_causal_calibration_ece": 0.12516937728295682, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.018305243090398104, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18093222379684448, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "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": 1.0718654827809488 + }, + "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": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5590997153025572, + "pairwise_causal_calibration_ece": 0.11947171326887146, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.16653094820357445, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "tau": 1.0718654827809488 + }, + "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": 0.5428571428571428, + "pairwise_causal_calibration_confidence": 0.5504892981594165, + "pairwise_causal_calibration_ece": 0.007632155302273683, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.8265443858364611, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8663521409034729, + "selected_candidate_index": 12.0, + "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": 1.0718654827809488 + }, + "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": 0.38095238095238093, + "pairwise_causal_calibration_confidence": 0.5497312288122934, + "pairwise_causal_calibration_ece": 0.16877884785991246, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.9117051526313378, + "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": 0.2260012924671173, + "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": 1.0718654827809488 + }, + "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": 0.3904761904761905, + "pairwise_causal_calibration_confidence": 0.5495148445011372, + "pairwise_causal_calibration_ece": 0.15903865402494677, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.8848141434142904, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "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": 0.32620343565940857, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4477952718734741, + "tau": 1.0718654827809488 + }, + "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": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5514987071348044, + "pairwise_causal_calibration_ece": 0.09695325258934998, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.7394668700396011, + "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": 0.8468573198810986 + }, + "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": 0.4909090909090909, + "pairwise_causal_calibration_confidence": 0.5481788586308872, + "pairwise_causal_calibration_ece": 0.05726976772179637, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.7193134962639322, + "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": 0.8468573198810986 + }, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5450693806396976, + "pairwise_causal_calibration_ece": 0.0620734765031595, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.7862881124399304, + "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": 0.8468573198810986 + }, + "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": 0.5765433563593303, + "pairwise_causal_calibration_ece": 0.1479719277879018, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3146091278603853, + "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": 1.0718654827809488 + }, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5703079922305777, + "pairwise_causal_calibration_ece": 0.07659820557682445, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.05209384637709236, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1324586570262909, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "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": 1.0718654827809488 + }, + "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": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5728588815635123, + "pairwise_causal_calibration_ece": 0.21571602442065527, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.31434308061981353, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19291165471076965, + "selected_candidate_index": 1.0, + "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": 1.0718654827809488 + }, + "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": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5461265642044052, + "pairwise_causal_calibration_ece": 0.020540102462261362, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.8003040878491947, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14406004548072815, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12164360284805298, + "selected_utility_gain_over_base": 0.015173077583312988, + "selector_regret": 0.02241644263267517, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8559399545192719, + "tau": 0.8468573198810986 + }, + "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": 0.6218487394957983, + "pairwise_causal_calibration_confidence": 0.5434572117117997, + "pairwise_causal_calibration_ece": 0.07839152778399854, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.3876858687209335, + "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": 0.11658626794815063, + "selected_utility_gain_over_base": 0.010115742683410645, + "selector_regret": 1.8834137320518494, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.8468573198810986 + }, + "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": 0.509090909090909, + "pairwise_causal_calibration_confidence": 0.5436846632267275, + "pairwise_causal_calibration_ece": 0.03459375413581838, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.7578621304129007, + "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": 0.8468573198810986 + }, + "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": 0.5866666666666667, + "pairwise_causal_calibration_confidence": 0.5192841401886225, + "pairwise_causal_calibration_ece": 0.06738252647804421, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.13235098862844796, + "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": 0.5825884328057473 + }, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5386998051668176, + "pairwise_causal_calibration_ece": 0.03869980516681758, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.48662334960024145, + "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": 0.5825884328057473 + }, + "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": 0.7023809523809523, + "pairwise_causal_calibration_confidence": 0.5315351438061215, + "pairwise_causal_calibration_ece": 0.17084580857483078, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.36262329065526533, + "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": 0.5825884328057473 + }, + "141984615/train0": { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.2833333333333333, + "pairwise_causal_calibration_confidence": 0.5448471499959241, + "pairwise_causal_calibration_ece": 0.2615138166625908, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5415586302800286, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10283424705266953, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "tau": 0.5825884328057473 + }, + "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": 0.6083333333333333, + "pairwise_causal_calibration_confidence": 0.5656089960064071, + "pairwise_causal_calibration_ece": 0.14136934515423885, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.8524230828537844, + "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": 0.5825884328057473 + }, + "141984615/train2": { + "base_success": 0.0, + "base_utility": 0.10283424705266953, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2857772707939148, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.425, + "pairwise_causal_calibration_confidence": 0.5603553061579338, + "pairwise_causal_calibration_ece": 0.1353553061579339, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3015705513281932, + "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": 0.10283424705266953, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "tau": 0.5825884328057473 + }, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.523550311048989, + "pairwise_causal_calibration_ece": 0.14311635561767766, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.06695685840699794, + "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": 0.5825884328057473 + }, + "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": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.5236824574508954, + "pairwise_causal_calibration_ece": 0.19034912411756205, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.24721792540099097, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0939030647277832, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14275743998587132, + "tau": 0.5825884328057473 + }, + "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": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5276472177677529, + "pairwise_causal_calibration_ece": 0.019313884434419615, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2696456255951446, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "selected_candidate_index": 0.0, + "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": 0.5825884328057473 + }, + "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": 0.8282828282828283, + "pairwise_causal_calibration_confidence": 0.5709539167236203, + "pairwise_causal_calibration_ece": 0.2573289115592079, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 2.2046095259547753, + "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": 0.8468573198810986 + }, + "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": 0.5882352941176471, + "pairwise_causal_calibration_confidence": 0.543456481052637, + "pairwise_causal_calibration_ece": 0.04477881306501008, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.735589926228614, + "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": 0.8468573198810986 + }, + "1480856810/train2": { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5370429153145398, + "pairwise_causal_calibration_ece": 0.03438565611403155, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.359251321625822, + "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": 0.21894000470638275, + "selected_utility_gain_over_base": 0.1319059282541275, + "selector_regret": 1.7810599952936172, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.8468573198810986 + }, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.5523718802888111, + "pairwise_causal_calibration_ece": 0.10792743584436665, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.6801689818523341, + "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": 0.8334100246429443, + "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": 1.0718654827809488 + }, + "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": 0.19047619047619047, + "pairwise_causal_calibration_confidence": 0.5517045024701317, + "pairwise_causal_calibration_ece": 0.36122831199394134, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": 0.7807195223473116, + "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": 1.0718654827809488 + }, + "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": 0.5444444444444444, + "pairwise_causal_calibration_confidence": 0.5506485429344871, + "pairwise_causal_calibration_ece": 0.024528598759442233, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": 0.7965612115773528, + "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": 1.0718654827809488 + }, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.550550217481388, + "pairwise_causal_calibration_ece": 0.11611644918527875, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.3705665707855397, + "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": 1.2894892762347252 + }, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5518834340111771, + "pairwise_causal_calibration_ece": 0.13521676734451044, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.8136424781392986, + "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": 1.2894892762347252 + }, + "1522272686/train2": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5635063449255251, + "pairwise_causal_calibration_ece": 0.06840618279050634, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.5686852402790281, + "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": 1.2894892762347252 + }, + "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": 0.43333333333333335, + "pairwise_causal_calibration_confidence": 0.5370476153908349, + "pairwise_causal_calibration_ece": 0.10371428205750155, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.332576441483642, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.060871422290802, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0260902252048254, + "selected_utility_gain_over_base": -0.07842461206018925, + "selector_regret": 0.0347811970859766, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.23658466339111328, + "tau": 0.5825884328057473 + }, + "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": 0.30833333333333335, + "pairwise_causal_calibration_confidence": 0.5437634159530779, + "pairwise_causal_calibration_ece": 0.2354300826197445, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9503772340326878, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0303483996540308, + "selected_utility_gain_over_base": -0.07416643761098385, + "selector_regret": 0.16857311688363552, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "tau": 0.5825884328057473 + }, + "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": 0.4583333333333333, + "pairwise_causal_calibration_confidence": 0.540090634882582, + "pairwise_causal_calibration_ece": 0.08175730154924889, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.4498268017490537, + "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": 0.5825884328057473 + }, + "1578990937/train0": { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7350427350427351, + "pairwise_causal_calibration_confidence": 0.575799443033985, + "pairwise_causal_calibration_ece": 0.15924329200874998, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 1.3204376505810713, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.17586350440979, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.0718654827809488 + }, + "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": 0.4090909090909091, + "pairwise_causal_calibration_confidence": 0.5644259196254204, + "pairwise_causal_calibration_ece": 0.15533501053451115, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.2385033751871737, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12732218205928802, + "selected_utility_gain_over_base": -0.6968143135309219, + "selector_regret": 1.872677817940712, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.0718654827809488 + }, + "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": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5644061750173082, + "pairwise_causal_calibration_ece": 0.13583474644587978, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.1890371636445307, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13359522819519043, + "selected_utility_gain_over_base": -0.6905412673950195, + "selector_regret": 1.8664047718048096, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.0718654827809488 + }, + "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": 0.6807136864797709, + "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": 0.5825884328057473 + }, + "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": 0.7412627528047919, + "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": 0.5825884328057473 + }, + "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": 0.7399153467725452, + "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": 0.5825884328057473 + }, + "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": 0.2818181818181818, + "pairwise_causal_calibration_confidence": 0.5215209331218725, + "pairwise_causal_calibration_ece": 0.23970275130369073, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.1652542261433299, + "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": 0.8468573198810986 + }, + "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": 0.517198535735028, + "pairwise_causal_calibration_ece": 0.16632134275257193, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 1.1907212928313995, + "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": 0.8468573198810986 + }, + "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": 0.43636363636363634, + "pairwise_causal_calibration_confidence": 0.5206471195621108, + "pairwise_causal_calibration_ece": 0.08428348319847451, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 1.0616754451066321, + "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": 0.8468573198810986 + }, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.5678773715560055, + "pairwise_causal_calibration_ece": 0.0719433679479887, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 0.7418095681624295, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06306678056716919, + "selected_utility_gain_over_base": -0.008848525583744049, + "selector_regret": 1.9369332194328308, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.5825884328057473 + }, + "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": 0.6938775510204082, + "pairwise_causal_calibration_confidence": 0.592873970483461, + "pairwise_causal_calibration_ece": 0.10100358053694718, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": 0.7078283118607503, + "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": 0.5825884328057473 + }, + "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": 0.4, + "pairwise_causal_calibration_confidence": 0.5677066182834786, + "pairwise_causal_calibration_ece": 0.16770661828347844, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 0.714483922998961, + "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": 0.5825884328057473 + }, + "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": 0.41304347826086957, + "pairwise_causal_calibration_confidence": 0.51605426349952, + "pairwise_causal_calibration_ece": 0.10301078523865043, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": 0.798908624848506, + "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": 0.5825884328057473 + }, + "1845489013/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7692307692307693, + "pairwise_causal_calibration_confidence": 0.5252519047235767, + "pairwise_causal_calibration_ece": 0.24397886450719253, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": 0.8663205932771525, + "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": 0.5825884328057473 + }, + "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": 0.5274458415050698, + "pairwise_causal_calibration_ece": 0.19255415849493018, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": 0.8229653113797162, + "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": 0.5825884328057473 + }, + "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": 0.543208103565328, + "pairwise_causal_calibration_ece": 0.1860652464224708, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.8291022763943762, + "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": 1.0718654827809488 + }, + "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": 0.6714285714285714, + "pairwise_causal_calibration_confidence": 0.5400358084963361, + "pairwise_causal_calibration_ece": 0.13139276293223526, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.7563954831040236, + "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": 1.0718654827809488 + }, + "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": 0.3815789473684211, + "pairwise_causal_calibration_confidence": 0.5298970718699219, + "pairwise_causal_calibration_ece": 0.14831812450150075, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": 0.8419564487563823, + "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": 1.0718654827809488 + }, + "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": 0.48739495798319327, + "pairwise_causal_calibration_confidence": 0.533600718072752, + "pairwise_causal_calibration_ece": 0.04620576008955871, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.107133532936841, + "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": 0.87335466614541 + }, + "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": 0.5379143734950628, + "pairwise_causal_calibration_ece": 0.07041895983827043, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9306280702125403, + "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": 0.87335466614541 + }, + "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": 0.7717391304347826, + "pairwise_causal_calibration_confidence": 0.5403873973296214, + "pairwise_causal_calibration_ece": 0.2313517331051612, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 1.1151401969224932, + "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": 0.87335466614541 + }, + "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": 0.5879562438719218, + "pairwise_causal_calibration_ece": 0.31209417490640445, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.2008765491740636, + "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": 0.5825884328057473 + }, + "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": 0.24074074074074073, + "pairwise_causal_calibration_confidence": 0.6117747139844449, + "pairwise_causal_calibration_ece": 0.37103397324370413, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.112157225492381, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07674666494131088, + "selected_utility_gain_over_base": -1.9232533350586891, + "selector_regret": 1.9232533350586891, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.5825884328057473 + }, + "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": 0.5954970669139762, + "pairwise_causal_calibration_ece": 0.19145666287357221, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 1.2364955407573137, + "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": 0.5825884328057473 + }, + "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": 0.26666666666666666, + "pairwise_causal_calibration_confidence": 0.5291897348449885, + "pairwise_causal_calibration_ece": 0.13126153408916094, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 1.014453498377106, + "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": 0.5825884328057473 + }, + "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": 0.6754385964912281, + "pairwise_causal_calibration_confidence": 0.5263065559429049, + "pairwise_causal_calibration_ece": 0.07456602027416159, + "pairwise_causal_calibration_pairs": 57.0, + "predicted_margin": 1.074027910955616, + "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": 0.5825884328057473 + }, + "2054899126/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.5, + "execute_generated": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5210084033613446, + "pairwise_causal_calibration_confidence": 0.5536973137638624, + "pairwise_causal_calibration_ece": 0.026703688149182295, + "pairwise_causal_calibration_pairs": 59.5, + "predicted_margin": 1.0710380363686038, + "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": 0.5825884328057473 + }, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.5639694431421344, + "pairwise_causal_calibration_ece": 0.130957329325739, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.39584763737906137, + "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": 1.0718654827809488 + }, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5581364424191588, + "pairwise_causal_calibration_ece": 0.1414697757524921, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.3705725365164993, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07580336928367615, + "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": 0.02355286478996277, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9241966307163239, + "tau": 1.0718654827809488 + }, + "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": 0.5794706736680671, + "pairwise_causal_calibration_ece": 0.13502622922362276, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.345599063353308, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11105132102966309, + "selected_candidate_index": 7.0, + "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": 1.0718654827809488 + }, + "211245842/train0": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6302521008403361, + "pairwise_causal_calibration_confidence": 0.5426955700305087, + "pairwise_causal_calibration_ece": 0.08755653080982741, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.1272780451605109, + "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": 0.8592818975448608, + "selected_utility_gain_over_base": 0.6683734804391861, + "selector_regret": 1.1407181024551392, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.0718654827809488 + }, + "211245842/train1": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.65, + "pairwise_causal_calibration_confidence": 0.5351083386938548, + "pairwise_causal_calibration_ece": 0.12495817852344455, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.2265101367583866, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09285585582256317, + "selected_utility_gain_over_base": -0.09805256128311157, + "selector_regret": 0.12010131776332855, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7870428264141083, + "tau": 1.0718654827809488 + }, + "211245842/train2": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.525, + "pairwise_causal_calibration_confidence": 0.534871348160612, + "pairwise_causal_calibration_ece": 0.009871348160611883, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.1256343724003985, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13431161642074585, + "selected_utility_gain_over_base": -0.056596800684928894, + "selector_regret": 0.09957115352153778, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7661172300577164, + "tau": 1.0718654827809488 + }, + "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": 0.6191919191919192, + "pairwise_causal_calibration_confidence": 0.5909019399202626, + "pairwise_causal_calibration_ece": 0.1344967059497781, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.27572048526322235, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0733563229441643, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05350120924413204, + "selected_utility_gain_over_base": 0.010766219347715378, + "selector_regret": 1.0198551137000322, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.04055231809616089, + "tau": 0.5825884328057473 + }, + "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": 0.5835144927536232, + "pairwise_causal_calibration_confidence": 0.5847988892880861, + "pairwise_causal_calibration_ece": 0.06891187442684295, + "pairwise_causal_calibration_pairs": 106.0, + "predicted_margin": 0.3598249612248885, + "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": 0.5825884328057473 + }, + "278008524/train2": { + "base_success": 0.0, + "base_utility": 0.042734989896416664, + "coverage": 0.5, + "execute_generated": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5271464646464646, + "pairwise_causal_calibration_confidence": 0.59445093065676, + "pairwise_causal_calibration_ece": 0.09739801499095443, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.4752736549448759, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.1359489113092422, + "selected_candidate_index": 14.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.047564467415213585, + "selected_utility_gain_over_base": 0.004829477518796921, + "selector_regret": 1.0883844438940287, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.5825884328057473 + }, + "307068879/train0": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5494505494505495, + "pairwise_causal_calibration_confidence": 0.5728850640530763, + "pairwise_causal_calibration_ece": 0.1199981805489268, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.7747416665744656, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2115575522184372, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7884424477815628, + "tau": 0.87335466614541 + }, + "307068879/train1": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5274725274725275, + "pairwise_causal_calibration_confidence": 0.5704103700900957, + "pairwise_causal_calibration_ece": 0.07309450817078264, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.3600826916101192, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5876073241233826, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "tau": 0.87335466614541 + }, + "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": 0.3956043956043956, + "pairwise_causal_calibration_confidence": 0.5723883106472546, + "pairwise_causal_calibration_ece": 0.1767839150428592, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.979654396381198, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2302134484052658, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2021283358335495, + "selected_utility_gain_over_base": -0.39796222746372223, + "selector_regret": 0.02808511257171631, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7697865515947342, + "tau": 0.87335466614541 + }, + "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": 0.5493461173759878, + "pairwise_causal_calibration_ece": 0.10951912707521982, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 1.0435143603129777, + "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": 1.0718654827809488 + }, + "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": 0.7948717948717948, + "pairwise_causal_calibration_confidence": 0.5593981622164304, + "pairwise_causal_calibration_ece": 0.23547363265536436, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": 0.949568847497014, + "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": 1.0718654827809488 + }, + "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": 0.4880952380952381, + "pairwise_causal_calibration_confidence": 0.5532244830443317, + "pairwise_causal_calibration_ece": 0.0651292449490937, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.9787258921638226, + "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": 1.0718654827809488 + }, + "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": 0.6551724137931034, + "pairwise_causal_calibration_confidence": 0.5624630563402808, + "pairwise_causal_calibration_ece": 0.09270935745282255, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 1.7706684649919293, + "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": 0.8468573198810986 + }, + "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": 0.5471459072582541, + "pairwise_causal_calibration_ece": 0.3939678976832891, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 1.808986225138223, + "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": 0.8468573198810986 + }, + "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": 0.5185185185185185, + "pairwise_causal_calibration_confidence": 0.5475473443233374, + "pairwise_causal_calibration_ece": 0.17413687717265497, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.7310319203842686, + "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": 0.8468573198810986 + }, + "441046872/train0": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.5664245191730715, + "pairwise_causal_calibration_ece": 0.19522652344582997, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.6069194020559678, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.3000448346138, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.87335466614541 + }, + "441046872/train1": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6181818181818182, + "pairwise_causal_calibration_confidence": 0.5652066435973622, + "pairwise_causal_calibration_ece": 0.09055210003018804, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.1629836409186699, + "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": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.023403704166412354, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "tau": 0.87335466614541 + }, + "441046872/train2": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5755444444938004, + "pairwise_causal_calibration_ece": 0.13114432369738743, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.5214657402479244, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7191348671913147, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019179701805114746, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "tau": 0.87335466614541 + }, + "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": 0.6386554621848739, + "pairwise_causal_calibration_confidence": 0.5365694240669033, + "pairwise_causal_calibration_ece": 0.10561970831698192, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.7250273458520244, + "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": 0.8468573198810986 + }, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5326653211281538, + "pairwise_causal_calibration_ece": 0.038763250300417725, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.5703557578982594, + "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": 0.8468573198810986 + }, + "509430392/train2": { + "base_success": 0.0, + "base_utility": 2.2351741790771484e-07, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.5403636942457117, + "pairwise_causal_calibration_ece": 0.14212516552457483, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 1.3794654730008236, + "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": 0.8468573198810986 + }, + "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": 0.36904761904761907, + "pairwise_causal_calibration_confidence": 0.5472546704459165, + "pairwise_causal_calibration_ece": 0.17820705139829746, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.9073249211388359, + "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": 0.5825884328057473 + }, + "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": 0.64, + "pairwise_causal_calibration_confidence": 0.549320518336562, + "pairwise_causal_calibration_ece": 0.19650416069860938, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.8486373501760932, + "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": 0.5825884328057473 + }, + "585904130/train2": { + "base_success": 0.0, + "base_utility": 0.18171554803848267, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7272727272727273, + "pairwise_causal_calibration_confidence": 0.5513841945828967, + "pairwise_causal_calibration_ece": 0.1758885326898305, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.9433176675944015, + "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": 0.5825884328057473 + }, + "608016290/train0": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4857142857142857, + "pairwise_causal_calibration_confidence": 0.594216772727526, + "pairwise_causal_calibration_ece": 0.11412216411454103, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.589765155203597, + "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": 0.707331120967865, + "selected_utility_gain_over_base": -0.018370509147644043, + "selector_regret": 1.292668879032135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.87335466614541 + }, + "608016290/train1": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5047619047619047, + "pairwise_causal_calibration_confidence": 0.5978828235074669, + "pairwise_causal_calibration_ece": 0.20716363895507375, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 1.3392733370703795, + "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": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.87335466614541 + }, + "608016290/train2": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.55, + "pairwise_causal_calibration_confidence": 0.6068596136210345, + "pairwise_causal_calibration_ece": 0.0651220182950748, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.4772393565445312, + "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": 0.6789124011993408, + "selected_utility_gain_over_base": -0.04678922891616821, + "selector_regret": 0.03281521797180176, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "tau": 0.87335466614541 + }, + "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": 0.65, + "pairwise_causal_calibration_confidence": 0.6158650684827733, + "pairwise_causal_calibration_ece": 0.1759930830571895, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9601260167395063, + "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": 0.5825884328057473 + }, + "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": 0.32916666666666666, + "pairwise_causal_calibration_confidence": 0.6045161568297507, + "pairwise_causal_calibration_ece": 0.2753494901630838, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9603159694517454, + "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": 0.5825884328057473 + }, + "626486813/train2": { + "base_success": 0.0, + "base_utility": 0.00509834272088483, + "coverage": 0.5, + "execute_generated": 0.5, + "fallback_rate": 0.5, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.16558238118886948, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.625, + "pairwise_causal_calibration_confidence": 0.6142120016040145, + "pairwise_causal_calibration_ece": 0.14970168842483308, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7735840442684991, + "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": 0.012973218690603971, + "selected_utility_gain_over_base": 0.007874875969719142, + "selector_regret": 0.03192603634670377, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12068312615156174, + "tau": 0.5825884328057473 + }, + "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": 0.425, + "pairwise_causal_calibration_confidence": 0.540894445846163, + "pairwise_causal_calibration_ece": 0.11589444584616293, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 1.0758545824030454, + "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": 0.5825884328057473 + }, + "636680137/train1": { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.625, + "pairwise_causal_calibration_confidence": 0.5301042033378723, + "pairwise_causal_calibration_ece": 0.09489579666212766, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.44432064636845636, + "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": 0.08313504606485367, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10315313935279846, + "tau": 0.5825884328057473 + }, + "636680137/train2": { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": 0.5238615170185813, + "pairwise_causal_calibration_ece": 0.048861517018581324, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5548863156073045, + "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": 0.08313504606485367, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09491270035505295, + "tau": 0.5825884328057473 + }, + "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": 0.5833333333333334, + "pairwise_causal_calibration_confidence": 0.5958635862373126, + "pairwise_causal_calibration_ece": 0.10068633489228643, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.5292851601669687, + "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": 1.0718654827809488 + }, + "655884605/train1": { + "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": 0.5277777777777778, + "pairwise_causal_calibration_confidence": 0.6117746577593474, + "pairwise_causal_calibration_ece": 0.228982599451425, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.8854182759065323, + "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": 0.16300667822360992, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.6498691290616989, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.1871241927146912, + "tau": 1.0718654827809488 + }, + "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": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5899703889806722, + "pairwise_causal_calibration_ece": 0.19094254082377798, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.7967317370820184, + "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": 1.0718654827809488 + }, + "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": 0.8235294117647058, + "pairwise_causal_calibration_confidence": 0.5940991028072462, + "pairwise_causal_calibration_ece": 0.22943030895745967, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": 0.7871393331842744, + "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": 0.5825884328057473 + }, + "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": 0.5428571428571428, + "pairwise_causal_calibration_confidence": 0.5924504110795782, + "pairwise_causal_calibration_ece": 0.07756824536937998, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": 0.7852289914727505, + "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": 0.5825884328057473 + }, + "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": 0.2, + "pairwise_causal_calibration_confidence": 0.6012678269199632, + "pairwise_causal_calibration_ece": 0.4012678269199631, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": 0.6365332388231544, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04013325646519661, + "selected_utility_gain_over_base": -1.9598667435348034, + "selector_regret": 1.9598667435348034, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.5825884328057473 + }, + "701667486/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.38095238095238093, + "pairwise_causal_calibration_confidence": 0.5278746131288938, + "pairwise_causal_calibration_ece": 0.1469222321765129, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 1.1356123561511042, + "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": 1.0718654827809488 + }, + "701667486/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5308683756394741, + "pairwise_causal_calibration_ece": 0.004845910074811617, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 1.234980143424118, + "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": 1.0718654827809488 + }, + "701667486/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.5331361711412006, + "pairwise_causal_calibration_ece": 0.08869172669675614, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 1.3518684532465013, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.8125, + "selected_utility_gain_over_base": -1.1875, + "selector_regret": 1.1875, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 1.0718654827809488 + }, + "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": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5322967641018285, + "pairwise_causal_calibration_ece": 0.046545342019779196, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7403769985406177, + "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": 0.5825884328057473 + }, + "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": 0.5502411276617256, + "pairwise_causal_calibration_ece": 0.2890079292574854, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6210780034951271, + "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": 0.5825884328057473 + }, + "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": 0.7083333333333334, + "pairwise_causal_calibration_confidence": 0.5279460992108118, + "pairwise_causal_calibration_ece": 0.1803872341225216, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.9133990308397049, + "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": 0.5825884328057473 + }, + "750448296/train0": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.6666666666666666, + "execute_generated": 0.6666666666666666, + "fallback_rate": 0.3333333333333333, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.5793650793650794, + "pairwise_causal_calibration_confidence": 0.5359906885657441, + "pairwise_causal_calibration_ece": 0.07322390346905903, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": 0.8003581361777009, + "proposal_oracle_success": 0.3333333333333333, + "proposal_oracle_utility": 0.7526603279014429, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09101381525397301, + "selected_utility_gain_over_base": -0.026579334090153377, + "selector_regret": 0.6616465126474699, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.3333333333333333, + "support_gap": 0.6553361676633358, + "tau": 0.5825884328057473 + }, + "750448296/train1": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4423076923076923, + "pairwise_causal_calibration_confidence": 0.5377836086909143, + "pairwise_causal_calibration_ece": 0.09632988234770806, + "pairwise_causal_calibration_pairs": 101.66666666666667, + "predicted_margin": 0.4055514323654445, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3705032741030057, + "selected_candidate_index": 7.333333333333333, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10720233743389447, + "selected_utility_gain_over_base": -0.010390811910231909, + "selector_regret": 1.2633009366691113, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.03749322146177292, + "tau": 0.5825884328057473 + }, + "750448296/train2": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.6666666666666666, + "execute_generated": 0.6666666666666666, + "fallback_rate": 0.3333333333333333, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.4833333333333334, + "pairwise_causal_calibration_confidence": 0.5376337500179175, + "pairwise_causal_calibration_ece": 0.07076889063044275, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": 0.5039790420368024, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3504486680030823, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09631064782540004, + "selected_utility_gain_over_base": -0.02128250151872635, + "selector_regret": 1.2541380201776822, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.05754782756169637, + "tau": 0.5825884328057473 + }, + "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": 0.49059829059829063, + "pairwise_causal_calibration_confidence": 0.5592639661990686, + "pairwise_causal_calibration_ece": 0.1274080019549793, + "pairwise_causal_calibration_pairs": 119.25, + "predicted_margin": 0.5457716259676435, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0424291878007352, + "selected_candidate_index": 8.25, + "selected_success": 0.0, + "selected_success_gain_over_base": -0.25, + "selected_utility": 0.0617859915655572, + "selected_utility_gain_over_base": -0.5114133504393976, + "selector_regret": 0.980643196235178, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.06562284985557199, + "tau": 0.5825884328057473 + }, + "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": 0.5700877145730087, + "pairwise_causal_calibration_confidence": 0.5526263680260839, + "pairwise_causal_calibration_ece": 0.15613421435842795, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.6028818781511358, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0629530912265182, + "selected_candidate_index": 3.75, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5621845133136958, + "selected_utility_gain_over_base": -0.011014828691259027, + "selector_regret": 0.5007685779128224, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.045098946429789066, + "tau": 0.5825884328057473 + }, + "767747424/train2": { + "base_success": 0.25, + "base_utility": 0.5731993420049548, + "coverage": 0.25, + "execute_generated": 0.25, + "fallback_rate": 0.75, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.5193142641672053, + "pairwise_causal_calibration_confidence": 0.5607639712421639, + "pairwise_causal_calibration_ece": 0.10791269448114733, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5655051629030958, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0671670949086547, + "selected_candidate_index": 8.75, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.567537541501224, + "selected_utility_gain_over_base": -0.005661800503730774, + "selector_regret": 0.49962955340743065, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.04088494274765253, + "tau": 0.5825884328057473 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_task.json b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_task.json new file mode 100644 index 0000000000000000000000000000000000000000..73cc20399067d1c250b60848cc36429977b72b70 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_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": 0.4666666666666667, + "execute_generated": 0.4666666666666667, + "fallback_rate": 0.5333333333333333, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.5333333333333333, + "pairwise_causal_calibration_accuracy": 0.5492625405285252, + "pairwise_causal_calibration_confidence": 0.5651094024103738, + "pairwise_causal_calibration_ece": 0.11881236057424142, + "pairwise_causal_calibration_pairs": 94.8, + "predicted_margin": 0.8862852359573267, + "proposal_oracle_success": 0.6, + "proposal_oracle_utility": 1.4122399787108104, + "selected_candidate_index": 5.333333333333333, + "selected_success": 0.3333333333333333, + "selected_success_gain_over_base": 0.13333333333333333, + "selected_utility": 1.088020184636116, + "selected_utility_gain_over_base": 0.1381496459245682, + "selector_regret": 0.3509542107582092, + "success_selector_gap": 0.26666666666666666, + "success_support_gap": 0.4, + "support_gap": 0.5877600212891897, + "tau": 0.87335466614541 + }, + "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": 0.5277777777777778, + "pairwise_causal_calibration_confidence": 0.5553133321393634, + "pairwise_causal_calibration_ece": 0.10657979977343184, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.5842980964012888, + "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": 1.2894892762347252 + }, + "PullCube-v1": { + "base_success": 0.36, + "base_utility": 0.7751704612234608, + "coverage": 0.6533333333333333, + "execute_generated": 0.6533333333333333, + "fallback_rate": 0.3466666666666667, + "hidden_chart_oracle_success": 0.52, + "hidden_chart_oracle_utility": 1.1467878860235214, + "outcome_ptr": 0.4666666666666667, + "pairwise_causal_calibration_accuracy": 0.5216143238307843, + "pairwise_causal_calibration_confidence": 0.5584521161129652, + "pairwise_causal_calibration_ece": 0.13244853449538532, + "pairwise_causal_calibration_pairs": 92.32, + "predicted_margin": 0.6759761060180944, + "proposal_oracle_success": 0.5333333333333333, + "proposal_oracle_utility": 1.1093241602430741, + "selected_candidate_index": 5.986666666666666, + "selected_success": 0.29333333333333333, + "selected_success_gain_over_base": -0.06666666666666667, + "selected_utility": 0.6341658942727372, + "selected_utility_gain_over_base": -0.14100456695072353, + "selector_regret": 0.4787018513850247, + "success_selector_gap": 0.24, + "success_support_gap": 0.02666666666666667, + "support_gap": 0.1075624164690574, + "tau": 0.5825884328057477 + }, + "PushCube-v1": { + "base_success": 0.16666666666666666, + "base_utility": 0.3924041787783305, + "coverage": 0.7777777777777778, + "execute_generated": 0.7777777777777778, + "fallback_rate": 0.2222222222222222, + "hidden_chart_oracle_success": 0.8333333333333334, + "hidden_chart_oracle_utility": 1.7015800600250561, + "outcome_ptr": 0.8333333333333334, + "pairwise_causal_calibration_accuracy": 0.5465423944585421, + "pairwise_causal_calibration_confidence": 0.5419772241344875, + "pairwise_causal_calibration_ece": 0.11799694975778399, + "pairwise_causal_calibration_pairs": 97.44444444444444, + "predicted_margin": 1.3829754161601442, + "proposal_oracle_success": 0.8888888888888888, + "proposal_oracle_utility": 1.7857814149724112, + "selected_candidate_index": 3.7222222222222223, + "selected_success": 0.5555555555555556, + "selected_success_gain_over_base": 0.3888888888888889, + "selected_utility": 1.162042656706439, + "selected_utility_gain_over_base": 0.7696384779281087, + "selector_regret": 0.6237387582659721, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.1111111111111111, + "support_gap": 0.21421858502758873, + "tau": 0.8468573198810989 + }, + "StackCube-v1": { + "base_success": 0.2727272727272727, + "base_utility": 0.7763947045261209, + "coverage": 0.2727272727272727, + "execute_generated": 0.2727272727272727, + "fallback_rate": 0.7272727272727273, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.6363636363636364, + "pairwise_causal_calibration_accuracy": 0.5043423274143793, + "pairwise_causal_calibration_confidence": 0.5567990222574275, + "pairwise_causal_calibration_ece": 0.13094581844275163, + "pairwise_causal_calibration_pairs": 67.45454545454545, + "predicted_margin": 0.7229237046506615, + "proposal_oracle_success": 0.5151515151515151, + "proposal_oracle_utility": 1.1567951707225856, + "selected_candidate_index": 5.393939393939394, + "selected_success": 0.2727272727272727, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7495684027671814, + "selected_utility_gain_over_base": -0.02682630175893957, + "selector_regret": 0.41081975310137775, + "success_selector_gap": 0.24242424242424243, + "success_support_gap": 0.48484848484848486, + "support_gap": 0.8432048292774142, + "tau": 1.0718654827809493 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/split_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/split_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..865ad41e869c1bfbc57e4db65ed79e91e6d93081 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/table.tex b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/table.tex new file mode 100644 index 0000000000000000000000000000000000000000..7186b70d786ad05e70fc59fa214c5be42dd4e19c --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_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 & 0.5486 & 0.4514 & 0.2917 & 0.3194 & 0.5694 & 0.5486 & 0.2014 & 0.2500 & 0.0302 \\ +\bottomrule +\end{tabular} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/train.log b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_successbonus2_task_envclip_k16_train_to_test/train.log new file mode 100644 index 0000000000000000000000000000000000000000..1c48597a347313250c9c105b1c05dfeda6ec0094 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_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=10.0 +tau={"LiftPegUpright-v1": 0.873355, "PickCube-v1": 1.289489, "PullCube-v1": 0.582588, "PushCube-v1": 0.846857, "StackCube-v1": 1.071865, "__global__": 0.933426} +calibration_input=runs/ctt_base_context_obs_train_cal_envclip_k16_rollout_comparison/combined_measured_candidates.json diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/command.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/command.txt new file mode 100644 index 0000000000000000000000000000000000000000..e7f0b8fb12329026484da8a0f2dc093cb5e295da --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_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_chartcompat_utility_task_envclip_k16_train_to_test --feature-set score_chart_compat --selector-chart-feature-mode base_context_obs --target utility_margin --threshold-scope task --k 16 --bootstrap-samples 300 --no-markdown-report diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/config.yaml b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..810ed38bee4cdf48416dc0659577d8e76fa94be3 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_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_chart_compat +fit_objective: pointwise +k: 16 +no_markdown_report: True +out_dir: runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_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: 1.0 +target: utility_margin +threshold_scope: task diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/data_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/data_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..e65c158c17e61aaf10422563d2ffdd8922735e6a --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/eval.log b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/eval.log new file mode 100644 index 0000000000000000000000000000000000000000..ad44bd37e51f82471f7a17d7372cee5fed00ed3f --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/git_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/git_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..293c6ad4e775f2f937dca574f1b83a5f93f0bdb1 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/git_hash.txt @@ -0,0 +1 @@ +14e81292e2004925f81e0bb4810f6f144bfd87b7 diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/metrics.json b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..47cc3be4121e30ba7ec7ee8c29ce9ead998df45c --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/metrics.json @@ -0,0 +1,9008 @@ +{ + "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": 0.6481481481481481, + "fallback_rate": 0.35185185185185186, + "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": 0.3611111111111111, + "selected_success_gain_over_base": 0.08333333333333333, + "selected_utility": 0.8304847452566236, + "selector_regret": 0.40203449528825674, + "success_selector_gap": 0.2175925925925926, + "success_support_gap": 0.17592592592592593, + "support_gap": 0.35465237537627453 + }, + "calibration_summary": { + "base_success": 0.2777777777777778, + "base_utility": 0.7007981554626086, + "coverage": 0.6481481481481481, + "fallback_rate": 0.35185185185185186, + "hidden_chart_oracle_success": 0.6597222222222222, + "hidden_chart_oracle_utility": 1.4150809271054134, + "outcome_ptr": 0.5462962962962963, + "pairwise_causal_calibration_accuracy": 0.5151666754436779, + "pairwise_causal_calibration_confidence": 0.5305006974422358, + "pairwise_causal_calibration_ece": 0.015334021998557847, + "pairwise_causal_calibration_pairs": 37978.0, + "proposal_oracle_success": 0.5370370370370371, + "proposal_oracle_utility": 1.1440672337592803, + "selected_success": 0.3611111111111111, + "selected_success_gain_over_base": 0.08333333333333333, + "selected_utility": 0.8304847452566236, + "selector_regret": 0.40203449528825674, + "success_selector_gap": 0.2175925925925926, + "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": 0.6388888888888888, + "fallback_rate": 0.3611111111111111, + "hidden_chart_oracle_success": 0.7291666666666666, + "hidden_chart_oracle_utility": 1.5183161981403828, + "outcome_ptr": 0.5486111111111112, + "pairwise_causal_calibration_accuracy": 0.5346630207495576, + "pairwise_causal_calibration_confidence": 0.5293185199263176, + "pairwise_causal_calibration_ece": 0.022107332086325975, + "pairwise_causal_calibration_pairs": 12434.0, + "proposal_oracle_success": 0.5694444444444444, + "proposal_oracle_utility": 1.2159999172436073, + "selected_success": 0.3263888888888889, + "selected_success_gain_over_base": 0.034722222222222224, + "selected_utility": 0.7742872366474735, + "selector_regret": 0.4654409392969683, + "success_selector_gap": 0.25, + "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, + 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", + "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_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, + 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": 0.5151666754436779, + "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": 0.01165962380424257, + "accuracy": 0.5152193497311021, + "confidence": 0.5268789735353446, + "count": 36631, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.09968388489006574, + "accuracy": 0.525908739365816, + "confidence": 0.6255926242558817, + "count": 1293, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.4881582356512103, + "accuracy": 0.2222222222222222, + "confidence": 0.7103804578734325, + "count": 54, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.8, + "upper": 0.9 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.9, + "upper": 1.0 + } + ], + "ece": 0.015334021998557847, + "mean_confidence": 0.5305006974422358, + "n_bins": 10, + "num_pairs": 37978, + "num_rows": 432 + }, + "eval": { + "accuracy": 0.5346630207495576, + "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": 0.014323797330992205, + "accuracy": 0.5394880402853546, + "confidence": 0.5251642429543624, + "count": 11915, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.2007986935695666, + "accuracy": 0.4238921001926782, + "confidence": 0.6246907937622448, + "count": 519, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.8, + "upper": 0.9 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.9, + "upper": 1.0 + } + ], + "ece": 0.022107332086325975, + "mean_confidence": 0.5293185199263176, + "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": 0.3949579831932773, + "pairwise_causal_calibration_confidence": 0.5282333871650522, + "pairwise_causal_calibration_ece": 0.1332754039717749, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.1490312481371141, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train0", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7026570439338684, + "selected_utility_gain_over_base": -0.020948290824890137, + "selector_regret": 1.2973429560661316, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.028235521236802175, + "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": 0.5494505494505495, + "pairwise_causal_calibration_confidence": 0.5260813668824347, + "pairwise_causal_calibration_ece": 0.036558379935610945, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.08783512114247624, + "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": 0.028235521236802175, + "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": 0.6218487394957983, + "pairwise_causal_calibration_confidence": 0.5252122988397412, + "pairwise_causal_calibration_ece": 0.09663644065605714, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.07106166187087547, + "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": 0.028235521236802175, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5619047619047619, + "pairwise_causal_calibration_confidence": 0.5561217648468477, + "pairwise_causal_calibration_ece": 0.03209834914225741, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.5011206164338899, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/train0", + "selected_candidate_index": 11, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.028235521236802175, + "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": 0.36363636363636365, + "pairwise_causal_calibration_confidence": 0.5267148410723766, + "pairwise_causal_calibration_ece": 0.163078477436013, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.046645784157056724, + "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": 0.028235521236802175, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5358486568758676, + "pairwise_causal_calibration_ece": 0.03584865687586758, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.37628277603589594, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18325337767601013, + "seed": "1522272686/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8167466223239899, + "task_id": "PickCube-v1", + "tau": 0.5025584143631503, + "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": 0.04624259356089375, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/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": -0.2012384152043573, + "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": 0.38461538461538464, + "pairwise_causal_calibration_confidence": 0.5087871241212454, + "pairwise_causal_calibration_ece": 0.12417173950586075, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": -0.19362564784596037, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.02627839706838131, + "selected_utility_gain_over_base": -0.13540753163397312, + "selector_regret": 1.9737216029316187, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.30833333333333335, + "pairwise_causal_calibration_confidence": 0.5145099591738125, + "pairwise_causal_calibration_ece": 0.20617662584047913, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.09082588792911983, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06019836664199829, + "seed": "1461069713/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.014822185039520264, + "selected_utility_gain_over_base": -0.07908087968826294, + "selector_regret": 0.04537618160247803, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0991440862417221, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.7416666666666667, + "pairwise_causal_calibration_confidence": 0.5757476456290026, + "pairwise_causal_calibration_ece": 0.22293034880133372, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.775200972367557, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.010183374397456646, + "seed": "626486813/train0", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007048348430544138, + "selected_utility_gain_over_base": 0.005118846776895225, + "selector_regret": 0.003135025966912508, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24323733616620302, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.26666666666666666, + "pairwise_causal_calibration_confidence": 0.5290379603266726, + "pairwise_causal_calibration_ece": 0.26237129366000594, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.02174125236488622, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15456630289554596, + "seed": "2054899126/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.08063507080078125, + "selected_utility_gain_over_base": -1.9193649291992188, + "selector_regret": 0.07393123209476471, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.845433697104454, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.8235294117647058, + "pairwise_causal_calibration_confidence": 0.5525242267379454, + "pairwise_causal_calibration_ece": 0.27100518502676063, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": -0.087075237393225, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4583333333333333, + "pairwise_causal_calibration_confidence": 0.556288803981115, + "pairwise_causal_calibration_ece": 0.09795547064778175, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.160531410420454, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15051491558551788, + "seed": "767747424/train0", + "selected_candidate_index": 11, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08540759235620499, + "selected_utility_gain_over_base": 0.00841909646987915, + "selector_regret": 0.0651073232293129, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08458571135997772, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.6916666666666667, + "pairwise_causal_calibration_confidence": 0.5246693313972304, + "pairwise_causal_calibration_ece": 0.16699733526943622, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.33009554566017457, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14671264588832855, + "seed": "278008524/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05741874501109123, + "selected_utility_gain_over_base": 0.021532438695430756, + "selector_regret": 0.08929390087723732, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08110463619232178, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.32608695652173914, + "pairwise_causal_calibration_confidence": 0.5077729807205135, + "pairwise_causal_calibration_ece": 0.18168602419877433, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": -0.22001414669968217, + "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": -0.2012384152043573, + "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": 0.5556290097511218, + "pairwise_causal_calibration_ece": 0.011037656915544881, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2681415236496052, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07920757681131363, + "seed": "626486813/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.012751738540828228, + "selected_utility_gain_over_base": 0.004484554752707481, + "selector_regret": 0.0664558382704854, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.7166666666666667, + "pairwise_causal_calibration_confidence": 0.5143600301618059, + "pairwise_causal_calibration_ece": 0.2023066365048608, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.31217491974788636, + "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": -0.2012384152043573, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5296207411344719, + "pairwise_causal_calibration_ece": 0.06533502684875758, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.10499618981526784, + "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": -0.2012384152043573, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.523230296021853, + "pairwise_causal_calibration_ece": 0.10656362935518632, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.08363360102690726, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "seed": "141984615/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007176061626523733, + "selected_utility_gain_over_base": -0.09565818542614579, + "selector_regret": 0.008001843933016062, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.49166666666666664, + "pairwise_causal_calibration_confidence": 0.5256123951340491, + "pairwise_causal_calibration_ece": 0.033945728467382474, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.018465700329916374, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21658658981323242, + "seed": "750448296/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21658658981323242, + "selected_utility_gain_over_base": 0.07387763261795044, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7834134101867676, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "chart_id": "ms-9757fc494e79b5543bb6", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5190311348285646, + "pairwise_causal_calibration_ece": 0.10236446816189787, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4324165282764497, + "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": -0.2012384152043573, + "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": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.534999768197681, + "pairwise_causal_calibration_ece": 0.09833356513565228, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.23052873356798131, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.019201835617423058, + "seed": "767747424/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.01041970681399107, + "selected_utility_gain_over_base": -0.04370323661714792, + "selector_regret": 0.008782128803431988, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17790568806231022, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5289748634459925, + "pairwise_causal_calibration_ece": 0.1376918032206741, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": -0.29978209413001944, + "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": -0.2012384152043573, + "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": 0.5740754232851301, + "pairwise_causal_calibration_ece": 0.2982133543196128, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.19762473284895746, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04285776615142822, + "selected_utility_gain_over_base": -1.9571422338485718, + "selector_regret": 1.9571422338485718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.5120242051814605, + "pairwise_causal_calibration_ece": 0.10464246148520617, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.33932393024086993, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.041394393891096115, + "seed": "750448296/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.041394393891096115, + "selected_utility_gain_over_base": -0.05160309746861458, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.18259509280323982, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.32706425588430094, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train0", + "selected_candidate_index": 8, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.49166666666666664, + "pairwise_causal_calibration_confidence": 0.5222884856095076, + "pairwise_causal_calibration_ece": 0.03062181894284094, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.08999271197759756, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10839563608169556, + "seed": "747260437/train0", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.061248183250427246, + "selected_utility_gain_over_base": -1.9387518167495728, + "selector_regret": 0.04714745283126831, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17026761174201965, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.04958367347717285, + "chart_id": "ms-da555a03dd382d93e119", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5404725232935694, + "pairwise_causal_calibration_ece": 0.08592706874811483, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.11381592128190807, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train0", + "selected_candidate_index": 14, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9504163265228271, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5117256505009358, + "pairwise_causal_calibration_ece": 0.15494101616573086, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.495453729546493, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.060871422290802, + "seed": "1549324805/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0260902252048254, + "selected_utility_gain_over_base": -0.07842461206018925, + "selector_regret": 0.0347811970859766, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.23658466339111328, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.549416493223813, + "pairwise_causal_calibration_ece": 0.08571587301053908, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": -0.0008074765493970848, + "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": -0.2012384152043573, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.40476190476190477, + "pairwise_causal_calibration_confidence": 0.5246521315903555, + "pairwise_causal_calibration_ece": 0.11989022682845074, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.3401193749830937, + "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": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.5273358125858131, + "pairwise_causal_calibration_ece": 0.09335384258660073, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.5465320030087847, + "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": 0.07420081675594419, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "chart_id": "ms-2413d1155c87f0442485", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5583333333333333, + "pairwise_causal_calibration_confidence": 0.5085746791327035, + "pairwise_causal_calibration_ece": 0.04975865420062986, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5936226389105781, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14406004548072815, + "seed": "1324642411/train0", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11045536398887634, + "selected_utility_gain_over_base": 0.0039848387241363525, + "selector_regret": 0.03360468149185181, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8559399545192719, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.5077769577736889, + "pairwise_causal_calibration_ece": 0.12595877595550709, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.1484177313105067, + "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": 0.07420081675594419, + "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": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5174492210423474, + "pairwise_causal_calibration_ece": 0.06290376649689283, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": -0.08116638752440412, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train0", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.5129229068912616, + "pairwise_causal_calibration_ece": 0.10052247125999891, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4269940390592024, + "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": 0.07420081675594419, + "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": 0.8383838383838383, + "pairwise_causal_calibration_confidence": 0.5267941522147715, + "pairwise_causal_calibration_ece": 0.3115896861690668, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.8247860886762404, + "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": 0.07420081675594419, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5393032283831621, + "pairwise_causal_calibration_ece": 0.06094260950628667, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.12372212373443373, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09799930453300476, + "seed": "2061635888/train0", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04574880003929138, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9020006954669952, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.5277777777777778, + "pairwise_causal_calibration_confidence": 0.5597351518736766, + "pairwise_causal_calibration_ece": 0.05616248877449882, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.07276876911827704, + "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": 0.13235082030966006, + "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": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5381830912240237, + "pairwise_causal_calibration_ece": 0.10961166265259512, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1171582058773368, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10274861752986908, + "seed": "1113371261/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.897251382470131, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "chart_id": "ms-420280f15a522ce21c72", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7008547008547008, + "pairwise_causal_calibration_confidence": 0.5507814264578395, + "pairwise_causal_calibration_ece": 0.15007327439686133, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.2626760081803321, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train0", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.17586350440979, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.43333333333333335, + "pairwise_causal_calibration_confidence": 0.5388105224755593, + "pairwise_causal_calibration_ece": 0.10547718914222602, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.11725750852840032, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.5882352941176471, + "pairwise_causal_calibration_confidence": 0.517501300906042, + "pairwise_causal_calibration_ece": 0.07073399321160512, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.04424689933210735, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "211245842/train0", + "selected_candidate_index": 8, + "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": 0.13235082030966006, + "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": 0.5296341103177351, + "pairwise_causal_calibration_ece": 0.015348396032020872, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.12464332364425626, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8663521409034729, + "seed": "1151410346/train0", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.6403508484363556, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.133647859096527, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5557404501598795, + "pairwise_causal_calibration_ece": 0.16288330730273656, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.09117531897647907, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.054989203810691833, + "seed": "1324133122/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9450107961893082, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5238095238095238, + "pairwise_causal_calibration_confidence": 0.5307788011897017, + "pairwise_causal_calibration_ece": 0.00696927738017783, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 0.18126309408722283, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/train0", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5444159724840614, + "pairwise_causal_calibration_ece": 0.04441597248406137, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 0.0942295054093702, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/train0", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.2571428571428571, + "pairwise_causal_calibration_confidence": 0.5358621206887306, + "pairwise_causal_calibration_ece": 0.2787192635458735, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.0012762962711883319, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "chart_id": "ms-5799584ed40e211c50a5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.575, + "pairwise_causal_calibration_confidence": 0.5289841828268461, + "pairwise_causal_calibration_ece": 0.04601581717315384, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.036105644427698036, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train1", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7236053347587585, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.2763946652412415, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.028235521236802175, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5218750448469349, + "pairwise_causal_calibration_ece": 0.049553526581636476, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": -0.13881394667704933, + "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": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "task_id": "LiftPegUpright-v1", + "tau": 0.028235521236802175, + "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": 0.6083333333333333, + "pairwise_causal_calibration_confidence": 0.5228008383368301, + "pairwise_causal_calibration_ece": 0.08553249499650317, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.05584757187568534, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/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": "LiftPegUpright-v1", + "tau": 0.028235521236802175, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.49523809523809526, + "pairwise_causal_calibration_confidence": 0.5555371517280593, + "pairwise_causal_calibration_ece": 0.06029905648996403, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.2910589829678913, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/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": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.028235521236802175, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6909090909090909, + "pairwise_causal_calibration_confidence": 0.5234185829111347, + "pairwise_causal_calibration_ece": 0.16749050799795617, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.17358234568930467, + "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": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.023403704166412354, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "task_id": "LiftPegUpright-v1", + "tau": 0.028235521236802175, + "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": 0.2222222222222222, + "pairwise_causal_calibration_confidence": 0.5423563498452709, + "pairwise_causal_calibration_ece": 0.32013412762304877, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.18697224010199912, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07014383375644684, + "seed": "1522272686/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9298561662435532, + "task_id": "PickCube-v1", + "tau": 0.5025584143631503, + "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": 0.00797935515478132, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/train1", + "selected_candidate_index": 7, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.36134453781512604, + "pairwise_causal_calibration_confidence": 0.5104506142141808, + "pairwise_causal_calibration_ece": 0.14910607639905477, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.33419524839225123, + "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": -0.2012384152043573, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5141448492443679, + "pairwise_causal_calibration_ece": 0.09747818257770119, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.1449663100493438, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "seed": "1461069713/train1", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.006729066371917725, + "selected_utility_gain_over_base": -0.08717399835586548, + "selector_regret": 0.009855946525931358, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14275743998587132, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.325, + "pairwise_causal_calibration_confidence": 0.5716128728942771, + "pairwise_causal_calibration_ece": 0.256947612057456, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.6113998221493578, + "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": -0.2012384152043573, + "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": 0.6491228070175439, + "pairwise_causal_calibration_confidence": 0.5248824145830566, + "pairwise_causal_calibration_ece": 0.12424039243448726, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.1555929299105951, + "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": -0.2012384152043573, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-62f8c18a15c566ec4b00", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.45714285714285713, + "pairwise_causal_calibration_confidence": 0.5406287558110948, + "pairwise_causal_calibration_ece": 0.08348589866823782, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": -0.23567605508015027, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/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": -0.2012384152043573, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.5525406960296153, + "pairwise_causal_calibration_ece": 0.05187229094802172, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.14222716429755156, + "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": 0.18194365501403809, + "selected_utility_gain_over_base": 0.10495515912771225, + "selector_regret": 0.027086779475212097, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.026070192456245422, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.4083333333333333, + "pairwise_causal_calibration_confidence": 0.5179096505561795, + "pairwise_causal_calibration_ece": 0.10957631722284622, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.47002328115376657, + "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": -0.2012384152043573, + "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": 1.0, + "pairwise_causal_calibration_confidence": 0.5209647285753305, + "pairwise_causal_calibration_ece": 0.4790352714246695, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": -0.18822734864823312, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1845489013/train1", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.425, + "pairwise_causal_calibration_confidence": 0.5534969698356498, + "pairwise_causal_calibration_ece": 0.1284969698356498, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3519937608406678, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.061759788542985916, + "seed": "626486813/train1", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.010010083205997944, + "selected_utility_gain_over_base": 0.0017428994178771973, + "selector_regret": 0.05174970533698797, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.01598426327109337, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.7606837606837606, + "pairwise_causal_calibration_confidence": 0.5106997024711076, + "pairwise_causal_calibration_ece": 0.24998405821265302, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": -0.16627136796176625, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/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": -0.2012384152043573, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5273196708342051, + "pairwise_causal_calibration_ece": 0.07268032916579492, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.06533471439439086, + "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": -0.2012384152043573, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.5259787015164451, + "pairwise_causal_calibration_ece": 0.049021298483554854, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.1560213376803143, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.037011127918958664, + "seed": "141984615/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.030119141563773155, + "selected_utility_gain_over_base": -0.07271510548889637, + "selector_regret": 0.006891986355185509, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24876614287495613, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.14270895719528198, + "chart_id": "ms-972c5be72d0b24be7cd5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.75, + "pairwise_causal_calibration_confidence": 0.522212848434421, + "pairwise_causal_calibration_ece": 0.22778715156557905, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.30627797855821914, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14270895719528198, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.857291042804718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.7, + "pairwise_causal_calibration_confidence": 0.5160986674159838, + "pairwise_causal_calibration_ece": 0.1839013325840162, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.0489088109005007, + "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": -0.2012384152043573, + "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": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.5313942238827244, + "pairwise_causal_calibration_ece": 0.10193910945060891, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.30496918903874165, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.042781930416822433, + "seed": "767747424/train1", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.042781930416822433, + "selected_utility_gain_over_base": -0.011341013014316559, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15432559326291084, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-ac94b48febe523da15a5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5614035087719298, + "pairwise_causal_calibration_confidence": 0.5257629988292399, + "pairwise_causal_calibration_ece": 0.03564050994268986, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.79693146060142, + "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": -0.2012384152043573, + "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": 0.2962962962962963, + "pairwise_causal_calibration_confidence": 0.5667318952152414, + "pairwise_causal_calibration_ece": 0.27043559891894486, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.3209231912001711, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07674666494131088, + "selected_utility_gain_over_base": -1.9232533350586891, + "selector_regret": 1.9232533350586891, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.375, + "pairwise_causal_calibration_confidence": 0.5124292478571429, + "pairwise_causal_calibration_ece": 0.13742924785714294, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.1084374964792017, + "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": -0.2012384152043573, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.4986531018645557, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/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": -0.2012384152043573, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8fc1568ce66c45c8d8b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6833333333333333, + "pairwise_causal_calibration_confidence": 0.5252521749807111, + "pairwise_causal_calibration_ece": 0.15808115835262226, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.26165809689080527, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06328952312469482, + "seed": "747260437/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": 0.21537372469902039, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "1" + }, + { + "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": 0.6413043478260869, + "pairwise_causal_calibration_confidence": 0.5465697715124626, + "pairwise_causal_calibration_ece": 0.09473457631362422, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.09876914550533503, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train1", + "selected_candidate_index": 8, + "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": -0.2012384152043573, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.5062538341615629, + "pairwise_causal_calibration_ece": 0.06874616583843707, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.18733344993267076, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "seed": "1549324805/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0303483996540308, + "selected_utility_gain_over_base": -0.07416643761098385, + "selector_regret": 0.16857311688363552, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.557604964366195, + "pairwise_causal_calibration_ece": 0.09545626012360084, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": -0.0650108915156051, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train1", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.045472364872694016, + "selected_utility_gain_over_base": -0.026442941278219223, + "selector_regret": 1.954527635127306, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.47692307692307695, + "pairwise_causal_calibration_confidence": 0.5243799390305641, + "pairwise_causal_calibration_ece": 0.047456862107487174, + "pairwise_causal_calibration_pairs": 65.0, + "predicted_margin": -0.39646879322768863, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "chart_id": "ms-05f34a90fa7355262b8e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4, + "pairwise_causal_calibration_confidence": 0.5197720209730798, + "pairwise_causal_calibration_ece": 0.11977202097307982, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 0.629918599045802, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train1", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.680672268907563, + "pairwise_causal_calibration_confidence": 0.5079157579147822, + "pairwise_causal_calibration_ece": 0.1727565109927809, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.3040303172517523, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11658626794815063, + "selected_utility_gain_over_base": 0.010115742683410645, + "selector_regret": 1.8834137320518494, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.37719298245614036, + "pairwise_causal_calibration_confidence": 0.5066944209005328, + "pairwise_causal_calibration_ece": 0.1295014384443925, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 0.13132582655221522, + "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": 0.07420081675594419, + "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": 0.5272727272727272, + "pairwise_causal_calibration_confidence": 0.5193562339693999, + "pairwise_causal_calibration_ece": 0.007916493303327332, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": -0.09547952648462638, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train1", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.5619047619047619, + "pairwise_causal_calibration_confidence": 0.5125583130541626, + "pairwise_causal_calibration_ece": 0.049346448850599356, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.37833433181770104, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/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": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.5180113719626723, + "pairwise_causal_calibration_ece": 0.061820560810436964, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5725138287112228, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21981161832809448, + "selected_utility_gain_over_base": 0.13277754187583923, + "selector_regret": 1.7801883816719055, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5362308522692728, + "pairwise_causal_calibration_ece": 0.06400863004705062, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.049366979703384384, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07580336928367615, + "seed": "2061635888/train1", + "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": 0.02355286478996277, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9241966307163239, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.562343415271088, + "pairwise_causal_calibration_ece": 0.11789897082664356, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.1554209538300596, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8128758072853088, + "seed": "655884605/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11176733672618866, + "selected_utility_gain_over_base": -0.051239341497421265, + "selector_regret": 0.7011084705591202, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.1871241927146912, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.5386619397445, + "pairwise_causal_calibration_ece": 0.06848091739835714, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11952381111596475, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18093222379684448, + "seed": "1113371261/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02250880002975464, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8190677762031555, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "chart_id": "ms-420280f15a522ce21c72", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42727272727272725, + "pairwise_causal_calibration_confidence": 0.5448159860797186, + "pairwise_causal_calibration_ece": 0.11754325880699122, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.17364935131576656, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12732218205928802, + "selected_utility_gain_over_base": -0.6968143135309219, + "selector_regret": 1.872677817940712, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.23809523809523808, + "pairwise_causal_calibration_confidence": 0.5371878138277117, + "pairwise_causal_calibration_ece": 0.29909257573247366, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": -0.10799597937369632, + "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": 0.13235082030966006, + "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": 0.6583333333333333, + "pairwise_causal_calibration_confidence": 0.5160911960797463, + "pairwise_causal_calibration_ece": 0.14224213725358703, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.13077847063672782, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "seed": "211245842/train1", + "selected_candidate_index": 4, + "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": 0.13235082030966006, + "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": 0.44761904761904764, + "pairwise_causal_calibration_confidence": 0.528839557722319, + "pairwise_causal_calibration_ece": 0.08122051010327141, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.10435665617161227, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1151410346/train1", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.7739987075328827, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5544807305695781, + "pairwise_causal_calibration_ece": 0.10154169690861004, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.10108263691246488, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1324586570262909, + "seed": "1324133122/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04025374352931976, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.867541342973709, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5952380952380952, + "pairwise_causal_calibration_confidence": 0.5325935443428202, + "pairwise_causal_calibration_ece": 0.06264455089527499, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.15700516615012938, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/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": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.7692307692307693, + "pairwise_causal_calibration_confidence": 0.5435121854300012, + "pairwise_causal_calibration_ece": 0.22571858380076806, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": 0.029952749201257664, + "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": 0.13235082030966006, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5297587479178145, + "pairwise_causal_calibration_ece": 0.041669823510756876, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": -0.08747088123670077, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train1", + "selected_candidate_index": 8, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.6847826086956522, + "pairwise_causal_calibration_confidence": 0.531948622673026, + "pairwise_causal_calibration_ece": 0.1528339860226262, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.09375170906247592, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train2", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6934840679168701, + "selected_utility_gain_over_base": -0.030121266841888428, + "selector_regret": 1.3065159320831299, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": 0.028235521236802175, + "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": 0.4945054945054945, + "pairwise_causal_calibration_confidence": 0.5232228710238422, + "pairwise_causal_calibration_ece": 0.028717376518347715, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.14397102567158404, + "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": 0.028235521236802175, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5254991602652951, + "pairwise_causal_calibration_ece": 0.14116750640137155, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.07873777798305576, + "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": 0.028235521236802175, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5609759280259827, + "pairwise_causal_calibration_ece": 0.05264259469264974, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4536901796255097, + "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": 0.6789124011993408, + "selected_utility_gain_over_base": -0.04678922891616821, + "selector_regret": 0.03281521797180176, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "task_id": "LiftPegUpright-v1", + "tau": 0.028235521236802175, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5255994659745129, + "pairwise_causal_calibration_ece": 0.07105401142905837, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.022380468606574055, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7191348671913147, + "seed": "441046872/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019179701805114746, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "task_id": "LiftPegUpright-v1", + "tau": 0.028235521236802175, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5411589331210354, + "pairwise_causal_calibration_ece": 0.12449226645436867, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.2402513599376978, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14937306940555573, + "seed": "1522272686/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8506269305944443, + "task_id": "PickCube-v1", + "tau": 0.5025584143631503, + "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": 0.06617060088497922, + "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": -0.2012384152043573, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.16168592870235443, + "chart_id": "ms-0e1ff8d16372796dac97", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6638655462184874, + "pairwise_causal_calibration_confidence": 0.5138205036707686, + "pairwise_causal_calibration_ece": 0.15004504254771878, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.24604226210470226, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "selected_candidate_index": 9, + "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": -0.2012384152043573, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.5185093923599849, + "pairwise_causal_calibration_ece": 0.08149060764001503, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.13654486820462297, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "seed": "1461069713/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09997648000717163, + "selected_utility_gain_over_base": 0.006073415279388428, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05936597287654877, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.7583333333333333, + "pairwise_causal_calibration_confidence": 0.5718595242801999, + "pairwise_causal_calibration_ece": 0.18647380905313332, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5453619418295237, + "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": -0.2012384152043573, + "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": 0.5798319327731093, + "pairwise_causal_calibration_confidence": 0.5308674006096046, + "pairwise_causal_calibration_ece": 0.048964532163504715, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.16497563292480844, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "selected_candidate_index": 14, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.4, + "pairwise_causal_calibration_confidence": 0.5442903896362881, + "pairwise_causal_calibration_ece": 0.1442903896362882, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": -0.14512362014342817, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04013325646519661, + "selected_utility_gain_over_base": -1.9598667435348034, + "selector_regret": 1.9598667435348034, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3, + "pairwise_causal_calibration_confidence": 0.5580073082495534, + "pairwise_causal_calibration_ece": 0.2580073082495535, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.11968923816243313, + "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": 0.0730469822883606, + "selected_utility_gain_over_base": -0.0039415135979652405, + "selector_regret": 0.14581434428691864, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.01623930037021637, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.528326612006066, + "pairwise_causal_calibration_ece": 0.11165994533939946, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2601575872438925, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2718978226184845, + "seed": "278008524/train2", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04554526135325432, + "selected_utility_gain_over_base": 0.009658955037593842, + "selector_regret": 0.22635256126523018, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.44, + "pairwise_causal_calibration_confidence": 0.5128055424573941, + "pairwise_causal_calibration_ece": 0.07280554245739407, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": -0.1547908021155039, + "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": -0.2012384152043573, + "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": 0.525, + "pairwise_causal_calibration_confidence": 0.5554467028317209, + "pairwise_causal_calibration_ece": 0.030446702831720726, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.23546795238077922, + "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": -0.2012384152043573, + "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": 0.5213675213675214, + "pairwise_causal_calibration_confidence": 0.509457332297461, + "pairwise_causal_calibration_ece": 0.011910189070060362, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": -0.2428803998963185, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "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": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.7818181818181819, + "pairwise_causal_calibration_confidence": 0.529609307543077, + "pairwise_causal_calibration_ece": 0.25220887427510486, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.09915967492706912, + "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": -0.2012384152043573, + "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": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5258876236158011, + "pairwise_causal_calibration_ece": 0.007445709717532223, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.046015598870960614, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058951716870069504, + "seed": "141984615/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.02716982364654541, + "selected_utility_gain_over_base": -0.07566442340612411, + "selector_regret": 0.031781893223524094, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.7166666666666667, + "pairwise_causal_calibration_confidence": 0.5298039058264857, + "pairwise_causal_calibration_ece": 0.18686276084018094, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.0072536998424092974, + "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": -0.2012384152043573, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5129089770472632, + "pairwise_causal_calibration_ece": 0.012908977047263237, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.13337534819821673, + "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": -0.2012384152043573, + "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": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.5411620972822431, + "pairwise_causal_calibration_ece": 0.1266896226987229, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.46716305154776483, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04980705305933952, + "seed": "767747424/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04339444637298584, + "selected_utility_gain_over_base": -0.010728497058153152, + "selector_regret": 0.0064126066863536835, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14730047062039375, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.8095238095238095, + "pairwise_causal_calibration_confidence": 0.5337083842579382, + "pairwise_causal_calibration_ece": 0.27581542526587133, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.6377766658396278, + "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": -0.2012384152043573, + "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": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5635629372227472, + "pairwise_causal_calibration_ece": 0.1090174826772927, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.31779873234459755, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.12977690994739532, + "selected_utility_gain_over_base": -1.8702230900526047, + "selector_regret": 1.8702230900526047, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5061085398457259, + "pairwise_causal_calibration_ece": 0.006108539845725902, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.10476507994192752, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.051346004009246826, + "seed": "750448296/train2", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.035547394305467606, + "selected_utility_gain_over_base": -0.05745009705424309, + "selector_regret": 0.01579860970377922, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1726434826850891, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.4919042736821675, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5187728818008999, + "pairwise_causal_calibration_ece": 0.08122711819910011, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.061796090652556146, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.12510187923908234, + "seed": "747260437/train2", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.06438694149255753, + "selected_utility_gain_over_base": -1.9356130585074425, + "selector_regret": 0.06071493774652481, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15356136858463287, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.6262626262626263, + "pairwise_causal_calibration_confidence": 0.543680207968306, + "pairwise_causal_calibration_ece": 0.1027583191032813, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.22274532006538736, + "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": -0.2012384152043573, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.5094928991923253, + "pairwise_causal_calibration_ece": 0.06550710080767463, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3994404854817541, + "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": -0.2012384152043573, + "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": 0.5487264271441268, + "pairwise_causal_calibration_ece": 0.12372642714412677, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": -0.0006465763829108795, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08447941392660141, + "selected_utility_gain_over_base": 0.012564107775688171, + "selector_regret": 1.9155205860733986, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.40476190476190477, + "pairwise_causal_calibration_confidence": 0.5308387675806765, + "pairwise_causal_calibration_ece": 0.12607686281877173, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.563486512521856, + "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": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.2012384152043573, + "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": 0.5240762582462818, + "pairwise_causal_calibration_ece": 0.061113295283318825, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.6241310075694073, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train2", + "selected_candidate_index": 13, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.5181818181818182, + "pairwise_causal_calibration_confidence": 0.5077976736149897, + "pairwise_causal_calibration_ece": 0.010384144566828457, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.07036169250445845, + "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": 0.07420081675594419, + "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": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.5083204445849703, + "pairwise_causal_calibration_ece": 0.12650226276678844, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.12795113455003831, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train2", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10904359817504883, + "selected_utility_gain_over_base": -0.01572386920452118, + "selector_regret": 1.8909564018249512, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.6190476190476191, + "pairwise_causal_calibration_confidence": 0.5192244871616464, + "pairwise_causal_calibration_ece": 0.0998231318859727, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.0006632054109868557, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train2", + "selected_candidate_index": 4, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.5966386554621849, + "pairwise_causal_calibration_confidence": 0.5151111613972951, + "pairwise_causal_calibration_ece": 0.0815274940648898, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.282290647599035, + "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": 0.07420081675594419, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "chart_id": "ms-d4c37ed44fb338c1d551", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5630252100840336, + "pairwise_causal_calibration_confidence": 0.5179749462983955, + "pairwise_causal_calibration_ece": 0.045050263785638145, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4285300550683504, + "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": 0.21894000470638275, + "selected_utility_gain_over_base": 0.1319059282541275, + "selector_regret": 1.7810599952936172, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.07420081675594419, + "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": 0.5555555555555556, + "pairwise_causal_calibration_confidence": 0.5391540330154272, + "pairwise_causal_calibration_ece": 0.016401522540128385, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.015355053119340732, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11105132102966309, + "seed": "2061635888/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.05880081653594971, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.888948678970337, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.5555555555555556, + "pairwise_causal_calibration_confidence": 0.5588372328026256, + "pairwise_causal_calibration_ece": 0.027075654701186606, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.11907170730895571, + "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": 0.13235082030966006, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5353470658344741, + "pairwise_causal_calibration_ece": 0.036081505594097285, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2072636199490521, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "seed": "1113371261/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.36134453781512604, + "pairwise_causal_calibration_confidence": 0.5460279556731249, + "pairwise_causal_calibration_ece": 0.18468341785799877, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.15403622504955924, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15726424753665924, + "selected_utility_gain_over_base": -0.6668722480535507, + "selector_regret": 1.8427357524633408, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.4888888888888889, + "pairwise_causal_calibration_confidence": 0.5391151823956642, + "pairwise_causal_calibration_ece": 0.05022629350677536, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.17204314625253667, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train2", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5172535124316141, + "pairwise_causal_calibration_ece": 0.04941315423505255, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.04386283650419778, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "seed": "211245842/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04297435283660889, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7661172300577164, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.38095238095238093, + "pairwise_causal_calibration_confidence": 0.530463392824977, + "pairwise_causal_calibration_ece": 0.14951101187259608, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.0965746708031926, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "seed": "1151410346/train2", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.32620343565940857, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4477952718734741, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.21428571428571427, + "pairwise_causal_calibration_confidence": 0.5529109212469531, + "pairwise_causal_calibration_ece": 0.33862520696123877, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.05985917108029659, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19291165471076965, + "seed": "1324133122/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.10070674121379852, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8070883452892303, + "task_id": "StackCube-v1", + "tau": 0.13235082030966006, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5185185185185185, + "pairwise_causal_calibration_confidence": 0.5303820261796057, + "pairwise_causal_calibration_ece": 0.011863507661087236, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.1922641707596195, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/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": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.38095238095238093, + "pairwise_causal_calibration_confidence": 0.5454433115811214, + "pairwise_causal_calibration_ece": 0.1644909306287403, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.004657759279381279, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/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": "StackCube-v1", + "tau": 0.13235082030966006, + "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": 0.42105263157894735, + "pairwise_causal_calibration_confidence": 0.5267734463002417, + "pairwise_causal_calibration_ece": 0.10572081472129435, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": -0.10186397965031571, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/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": "StackCube-v1", + "tau": 0.13235082030966006, + "train_seed": "2" + } + ], + "schema_version": 1, + "selected_lambda": 100.0, + "selector_calibration_target_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "selector_calibration_target_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "selector_chart_feature_mode": "base_context_obs", + "selector_eval_target_content_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "selector_eval_target_split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "selector_source_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "selector_source_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "source_content_hash": null, + "source_index": null, + "source_split_hash": null, + "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "success_bonus": 1.0, + "summary": { + "base_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.38958333333333334, + "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": 0.6993055555555557, + "low": 0.5277777777777778, + "mean": 0.6138888888888889, + "n": 120, + "num_samples": 300 + }, + "mean": 0.6138888888888889, + "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": 1.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 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": 0.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 1.0, + "2012211287/train1": 0.0, + "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": 1.0, + "278008524/train1": 1.0, + "278008524/train2": 1.0, + "307068879/train0": 1.0, + "307068879/train1": 0.0, + "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": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 1.0, + "636680137/train0": 1.0, + "636680137/train1": 1.0, + "636680137/train2": 1.0, + "655884605/train0": 0.0, + "655884605/train1": 1.0, + "655884605/train2": 0.0, + "672328262/train0": 1.0, + "672328262/train1": 0.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 1.0, + "747260437/train1": 0.0, + "747260437/train2": 1.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.75, + "767747424/train1": 0.75, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.784, + "low": 0.20509090909090907, + "mean": 0.4939797979797979, + "n": 5, + "num_samples": 300 + }, + "mean": 0.4939797979797979, + "per_task": { + "LiftPegUpright-v1": 0.6666666666666666, + "PickCube-v1": 0.0, + "PullCube-v1": 0.8133333333333334, + "PushCube-v1": 0.7777777777777778, + "StackCube-v1": 0.21212121212121213 + } + }, + "metric": "coverage", + "micro": { + "confidence": 0.95, + "high": 0.7152777777777778, + "low": 0.5625, + "mean": 0.6388888888888888, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "execute_generated": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.6993055555555557, + "low": 0.5277777777777778, + "mean": 0.6138888888888889, + "n": 120, + "num_samples": 300 + }, + "mean": 0.6138888888888889, + "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": 1.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 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": 0.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 1.0, + "2012211287/train1": 0.0, + "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": 1.0, + "278008524/train1": 1.0, + "278008524/train2": 1.0, + "307068879/train0": 1.0, + "307068879/train1": 0.0, + "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": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 1.0, + "636680137/train0": 1.0, + "636680137/train1": 1.0, + "636680137/train2": 1.0, + "655884605/train0": 0.0, + "655884605/train1": 1.0, + "655884605/train2": 0.0, + "672328262/train0": 1.0, + "672328262/train1": 0.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 1.0, + "747260437/train1": 0.0, + "747260437/train2": 1.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.75, + "767747424/train1": 0.75, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.784, + "low": 0.20509090909090907, + "mean": 0.4939797979797979, + "n": 5, + "num_samples": 300 + }, + "mean": 0.4939797979797979, + "per_task": { + "LiftPegUpright-v1": 0.6666666666666666, + "PickCube-v1": 0.0, + "PullCube-v1": 0.8133333333333334, + "PushCube-v1": 0.7777777777777778, + "StackCube-v1": 0.21212121212121213 + } + }, + "metric": "execute_generated", + "micro": { + "confidence": 0.95, + "high": 0.7152777777777778, + "low": 0.5625, + "mean": 0.6388888888888888, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "fallback_rate": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.4722222222222222, + "low": 0.3006944444444445, + "mean": 0.3861111111111111, + "n": 120, + "num_samples": 300 + }, + "mean": 0.3861111111111111, + "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": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 1.0, + "1522272686/train1": 1.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 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": 1.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 0.0, + "2012211287/train1": 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": 0.0, + "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": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 1.0, + "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": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 1.0, + "655884605/train1": 0.0, + "655884605/train2": 1.0, + "672328262/train0": 0.0, + "672328262/train1": 1.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 1.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.3333333333333333, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7949090909090909, + "low": 0.21600000000000003, + "mean": 0.506020202020202, + "n": 5, + "num_samples": 300 + }, + "mean": 0.506020202020202, + "per_task": { + "LiftPegUpright-v1": 0.3333333333333333, + "PickCube-v1": 1.0, + "PullCube-v1": 0.18666666666666668, + "PushCube-v1": 0.2222222222222222, + "StackCube-v1": 0.7878787878787878 + } + }, + "metric": "fallback_rate", + "micro": { + "confidence": 0.95, + "high": 0.4375, + "low": 0.2847222222222222, + "mean": 0.3611111111111111, + "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": 0.5513360014629635, + "low": 0.4987694224782799, + "mean": 0.5238405556765252, + "n": 117, + "num_samples": 300 + }, + "mean": 0.5238405556765252, + "per_seed": { + "1073161778/train0": 0.6218487394957983, + "1073161778/train1": 0.6083333333333333, + "1073161778/train2": 0.6666666666666666, + "1113371261/train0": 0.42857142857142855, + "1113371261/train1": 0.6071428571428571, + "1113371261/train2": 0.5714285714285714, + "1151410346/train0": 0.5142857142857142, + "1151410346/train1": 0.44761904761904764, + "1151410346/train2": 0.38095238095238093, + "1261930456/train0": 0.45454545454545453, + "1261930456/train1": 0.5272727272727272, + "1261930456/train2": 0.6190476190476191, + "1324133122/train0": 0.39285714285714285, + "1324133122/train1": 0.4642857142857143, + "1324133122/train2": 0.21428571428571427, + "1324642411/train0": 0.5583333333333333, + "1324642411/train1": 0.680672268907563, + "1324642411/train2": 0.5181818181818182, + "1375492786/train0": 0.6666666666666666, + "1375492786/train1": 0.5614035087719298, + "1375492786/train2": 0.8095238095238095, + "141984615/train0": 0.4166666666666667, + "141984615/train1": 0.575, + "141984615/train2": 0.5333333333333333, + "1461069713/train0": 0.30833333333333335, + "1461069713/train1": 0.4166666666666667, + "1461069713/train2": 0.6, + "1480856810/train0": 0.8383838383838383, + "1480856810/train1": 0.5798319327731093, + "1480856810/train2": 0.5630252100840336, + "1509980664/train0": 0.43333333333333335, + "1509980664/train1": 0.23809523809523808, + "1509980664/train2": 0.4888888888888889, + "1522272686/train0": 0.5, + "1522272686/train1": 0.2222222222222222, + "1522272686/train2": 0.4166666666666667, + "1549324805/train0": 0.6666666666666666, + "1549324805/train1": 0.575, + "1549324805/train2": 0.575, + "1578990937/train0": 0.7008547008547008, + "1578990937/train1": 0.42727272727272725, + "1578990937/train2": 0.36134453781512604, + "1790188787/train0": 0.38181818181818183, + "1790188787/train1": 0.37719298245614036, + "1790188787/train2": 0.38181818181818183, + "1822210471/train0": 0.575, + "1822210471/train1": 0.6530612244897959, + "1822210471/train2": 0.425, + "1845489013/train0": 0.32608695652173914, + "1845489013/train1": 1.0, + "1845489013/train2": 0.44, + "1999022574/train0": 0.2571428571428571, + "1999022574/train1": 0.5714285714285714, + "1999022574/train2": 0.42105263157894735, + "2012211287/train0": 0.3949579831932773, + "2012211287/train1": 0.575, + "2012211287/train2": 0.6847826086956522, + "2038594191/train0": 0.27586206896551724, + "2038594191/train1": 0.2962962962962963, + "2038594191/train2": 0.45454545454545453, + "2054899126/train0": 0.26666666666666666, + "2054899126/train1": 0.6491228070175439, + "2054899126/train2": 0.5798319327731093, + "2061635888/train0": 0.5, + "2061635888/train1": 0.4722222222222222, + "2061635888/train2": 0.5555555555555556, + "211245842/train0": 0.5882352941176471, + "211245842/train1": 0.6583333333333333, + "211245842/train2": 0.5666666666666667, + "278008524/train0": 0.5731060606060606, + "278008524/train1": 0.5248188405797101, + "278008524/train2": 0.5214646464646465, + "307068879/train0": 0.5494505494505495, + "307068879/train1": 0.5714285714285714, + "307068879/train2": 0.4945054945054945, + "412102563/train0": 0.5, + "412102563/train1": 0.7692307692307693, + "412102563/train2": 0.38095238095238093, + "440577711/train0": 0.6206896551724138, + "440577711/train1": 0.4, + "440577711/train2": 0.46296296296296297, + "441046872/train0": 0.36363636363636365, + "441046872/train1": 0.6909090909090909, + "441046872/train2": 0.45454545454545453, + "509430392/train0": 0.6134453781512605, + "509430392/train1": 0.5619047619047619, + "509430392/train2": 0.5966386554621849, + "585904130/train0": 0.4642857142857143, + "585904130/train1": 0.6, + "585904130/train2": 0.7818181818181819, + "608016290/train0": 0.5619047619047619, + "608016290/train1": 0.49523809523809526, + "608016290/train2": 0.5083333333333333, + "626486813/train0": 0.6541666666666667, + "626486813/train1": 0.375, + "626486813/train2": 0.6416666666666666, + "636680137/train0": 0.4166666666666667, + "636680137/train1": 0.7, + "636680137/train2": 0.5, + "655884605/train0": 0.5277777777777778, + "655884605/train1": 0.4444444444444444, + "655884605/train2": 0.5555555555555556, + "672328262/train0": 0.8235294117647058, + "672328262/train1": 0.45714285714285713, + "672328262/train2": 0.4, + "701667486/train0": 0.5238095238095238, + "701667486/train1": 0.5952380952380952, + "701667486/train2": 0.5185185185185185, + "747260437/train0": 0.49166666666666664, + "747260437/train1": 0.6833333333333333, + "747260437/train2": 0.6, + "750448296/train0": 0.5043650793650793, + "750448296/train1": 0.533974358974359, + "750448296/train2": 0.5404761904761904, + "767747424/train0": 0.5482371794871794, + "767747424/train1": 0.5888404079580549, + "767747424/train2": 0.5254749335631689 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5455222296456455, + "low": 0.43510367170021186, + "mean": 0.5002133038437834, + "n": 5, + "num_samples": 300 + }, + "mean": 0.5002133038437834, + "per_task": { + "LiftPegUpright-v1": 0.5494360697557629, + "PickCube-v1": 0.3796296296296296, + "PullCube-v1": 0.5439315887395845, + "PushCube-v1": 0.5408758312375324, + "StackCube-v1": 0.4871933998564077 + } + }, + "metric": "pairwise_causal_calibration_accuracy", + "micro": { + "confidence": 0.95, + "high": 0.5488772820166298, + "low": 0.503951158120747, + "mean": 0.5269917153318382, + "n": 138, + "num_samples": 300 + }, + "n": 138 + }, + "pairwise_causal_calibration_confidence": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5346481921937558, + "low": 0.5285073781330066, + "mean": 0.5313682932910728, + "n": 117, + "num_samples": 300 + }, + "mean": 0.5313682932910728, + "per_seed": { + "1073161778/train0": 0.5252122988397412, + "1073161778/train1": 0.5228008383368301, + "1073161778/train2": 0.5254991602652951, + "1113371261/train0": 0.5381830912240237, + "1113371261/train1": 0.5386619397445, + "1113371261/train2": 0.5353470658344741, + "1151410346/train0": 0.5296341103177351, + "1151410346/train1": 0.528839557722319, + "1151410346/train2": 0.530463392824977, + "1261930456/train0": 0.5174492210423474, + "1261930456/train1": 0.5193562339693999, + "1261930456/train2": 0.5192244871616464, + "1324133122/train0": 0.5557404501598795, + "1324133122/train1": 0.5544807305695781, + "1324133122/train2": 0.5529109212469531, + "1324642411/train0": 0.5085746791327035, + "1324642411/train1": 0.5079157579147822, + "1324642411/train2": 0.5077976736149897, + "1375492786/train0": 0.5289748634459925, + "1375492786/train1": 0.5257629988292399, + "1375492786/train2": 0.5337083842579382, + "141984615/train0": 0.523230296021853, + "141984615/train1": 0.5259787015164451, + "141984615/train2": 0.5258876236158011, + "1461069713/train0": 0.5145099591738125, + "1461069713/train1": 0.5141448492443679, + "1461069713/train2": 0.5185093923599849, + "1480856810/train0": 0.5267941522147715, + "1480856810/train1": 0.5180113719626723, + "1480856810/train2": 0.5179749462983955, + "1509980664/train0": 0.5388105224755593, + "1509980664/train1": 0.5371878138277117, + "1509980664/train2": 0.5391151823956642, + "1522272686/train0": 0.5358486568758676, + "1522272686/train1": 0.5423563498452709, + "1522272686/train2": 0.5411589331210354, + "1549324805/train0": 0.5117256505009358, + "1549324805/train1": 0.5062538341615629, + "1549324805/train2": 0.5094928991923253, + "1578990937/train0": 0.5507814264578395, + "1578990937/train1": 0.5448159860797186, + "1578990937/train2": 0.5460279556731249, + "1790188787/train0": 0.5077769577736889, + "1790188787/train1": 0.5066944209005328, + "1790188787/train2": 0.5083204445849703, + "1822210471/train0": 0.549416493223813, + "1822210471/train1": 0.557604964366195, + "1822210471/train2": 0.5487264271441268, + "1845489013/train0": 0.5077729807205135, + "1845489013/train1": 0.5209647285753305, + "1845489013/train2": 0.5128055424573941, + "1999022574/train0": 0.5358621206887306, + "1999022574/train1": 0.5297587479178145, + "1999022574/train2": 0.5267734463002417, + "2012211287/train0": 0.5282333871650522, + "2012211287/train1": 0.5289841828268461, + "2012211287/train2": 0.531948622673026, + "2038594191/train0": 0.5740754232851301, + "2038594191/train1": 0.5667318952152414, + "2038594191/train2": 0.5635629372227472, + "2054899126/train0": 0.5290379603266726, + "2054899126/train1": 0.5248824145830566, + "2054899126/train2": 0.5308674006096046, + "2061635888/train0": 0.5393032283831621, + "2061635888/train1": 0.5362308522692728, + "2061635888/train2": 0.5391540330154272, + "211245842/train0": 0.517501300906042, + "211245842/train1": 0.5160911960797463, + "211245842/train2": 0.5172535124316141, + "278008524/train0": 0.5325709273453999, + "278008524/train1": 0.5322397110343211, + "278008524/train2": 0.536003409987186, + "307068879/train0": 0.5260813668824347, + "307068879/train1": 0.5218750448469349, + "307068879/train2": 0.5232228710238422, + "412102563/train0": 0.5444159724840614, + "412102563/train1": 0.5435121854300012, + "412102563/train2": 0.5454433115811214, + "440577711/train0": 0.5273358125858131, + "440577711/train1": 0.5197720209730798, + "440577711/train2": 0.5240762582462818, + "441046872/train0": 0.5267148410723766, + "441046872/train1": 0.5234185829111347, + "441046872/train2": 0.5255994659745129, + "509430392/train0": 0.5129229068912616, + "509430392/train1": 0.5125583130541626, + "509430392/train2": 0.5151111613972951, + "585904130/train0": 0.5296207411344719, + "585904130/train1": 0.5273196708342051, + "585904130/train2": 0.529609307543077, + "608016290/train0": 0.5561217648468477, + "608016290/train1": 0.5555371517280593, + "608016290/train2": 0.5609759280259827, + "626486813/train0": 0.5656883276900622, + "626486813/train1": 0.5625549213649634, + "626486813/train2": 0.5636531135559604, + "636680137/train0": 0.5190311348285646, + "636680137/train1": 0.5160986674159838, + "636680137/train2": 0.5129089770472632, + "655884605/train0": 0.5597351518736766, + "655884605/train1": 0.562343415271088, + "655884605/train2": 0.5588372328026256, + "672328262/train0": 0.5525242267379454, + "672328262/train1": 0.5406287558110948, + "672328262/train2": 0.5442903896362881, + "701667486/train0": 0.5307788011897017, + "701667486/train1": 0.5325935443428202, + "701667486/train2": 0.5303820261796057, + "747260437/train0": 0.5222884856095076, + "747260437/train1": 0.5252521749807111, + "747260437/train2": 0.5187728818008999, + "750448296/train0": 0.5207629106352885, + "750448296/train1": 0.5196740117740427, + "750448296/train2": 0.5222504044176294, + "767747424/train0": 0.5286089316154619, + "767747424/train1": 0.5262713091494071, + "767747424/train2": 0.5306118103750065 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5379868829976331, + "low": 0.5233696556864914, + "mean": 0.5315865513366932, + "n": 5, + "num_samples": 300 + }, + "mean": 0.5315865513366932, + "per_task": { + "LiftPegUpright-v1": 0.5321483671612611, + "PickCube-v1": 0.5397879799473912, + "PullCube-v1": 0.5315713773974821, + "PushCube-v1": 0.5154259344288218, + "StackCube-v1": 0.5389990977485094 + } + }, + "metric": "pairwise_causal_calibration_confidence", + "micro": { + "confidence": 0.95, + "high": 0.5347090519983646, + "low": 0.5285313883446271, + "mean": 0.5314829821239634, + "n": 138, + "num_samples": 300 + }, + "n": 138 + }, + "pairwise_causal_calibration_ece": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.12306181225731887, + "low": 0.09531198783107489, + "mean": 0.10993892641652644, + "n": 120, + "num_samples": 300 + }, + "mean": 0.10993892641652644, + "per_seed": { + "1073161778/train0": 0.09663644065605714, + "1073161778/train1": 0.08553249499650317, + "1073161778/train2": 0.14116750640137155, + "1113371261/train0": 0.10961166265259512, + "1113371261/train1": 0.06848091739835714, + "1113371261/train2": 0.036081505594097285, + "1151410346/train0": 0.015348396032020872, + "1151410346/train1": 0.08122051010327141, + "1151410346/train2": 0.14951101187259608, + "1261930456/train0": 0.06290376649689283, + "1261930456/train1": 0.007916493303327332, + "1261930456/train2": 0.0998231318859727, + "1324133122/train0": 0.16288330730273656, + "1324133122/train1": 0.10154169690861004, + "1324133122/train2": 0.33862520696123877, + "1324642411/train0": 0.04975865420062986, + "1324642411/train1": 0.1727565109927809, + "1324642411/train2": 0.010384144566828457, + "1375492786/train0": 0.1376918032206741, + "1375492786/train1": 0.03564050994268986, + "1375492786/train2": 0.27581542526587133, + "141984615/train0": 0.10656362935518632, + "141984615/train1": 0.049021298483554854, + "141984615/train2": 0.007445709717532223, + "1461069713/train0": 0.20617662584047913, + "1461069713/train1": 0.09747818257770119, + "1461069713/train2": 0.08149060764001503, + "1480856810/train0": 0.3115896861690668, + "1480856810/train1": 0.061820560810436964, + "1480856810/train2": 0.045050263785638145, + "1509980664/train0": 0.10547718914222602, + "1509980664/train1": 0.29909257573247366, + "1509980664/train2": 0.05022629350677536, + "1522272686/train0": 0.03584865687586758, + "1522272686/train1": 0.32013412762304877, + "1522272686/train2": 0.12449226645436867, + "1549324805/train0": 0.15494101616573086, + "1549324805/train1": 0.06874616583843707, + "1549324805/train2": 0.06550710080767463, + "1578990937/train0": 0.15007327439686133, + "1578990937/train1": 0.11754325880699122, + "1578990937/train2": 0.18468341785799877, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.12595877595550709, + "1790188787/train1": 0.1295014384443925, + "1790188787/train2": 0.12650226276678844, + "1822210471/train0": 0.08571587301053908, + "1822210471/train1": 0.09545626012360084, + "1822210471/train2": 0.12372642714412677, + "1845489013/train0": 0.18168602419877433, + "1845489013/train1": 0.4790352714246695, + "1845489013/train2": 0.07280554245739407, + "1999022574/train0": 0.2787192635458735, + "1999022574/train1": 0.041669823510756876, + "1999022574/train2": 0.10572081472129435, + "2012211287/train0": 0.1332754039717749, + "2012211287/train1": 0.04601581717315384, + "2012211287/train2": 0.1528339860226262, + "2038594191/train0": 0.2982133543196128, + "2038594191/train1": 0.27043559891894486, + "2038594191/train2": 0.1090174826772927, + "2054899126/train0": 0.13118564683000297, + "2054899126/train1": 0.06212019621724363, + "2054899126/train2": 0.024482266081752357, + "2061635888/train0": 0.06094260950628667, + "2061635888/train1": 0.06400863004705062, + "2061635888/train2": 0.016401522540128385, + "211245842/train0": 0.07073399321160512, + "211245842/train1": 0.14224213725358703, + "211245842/train2": 0.04941315423505255, + "278008524/train0": 0.12646220200877553, + "278008524/train1": 0.10215544676823522, + "278008524/train2": 0.10720913222134038, + "307068879/train0": 0.036558379935610945, + "307068879/train1": 0.049553526581636476, + "307068879/train2": 0.028717376518347715, + "412102563/train0": 0.04441597248406137, + "412102563/train1": 0.22571858380076806, + "412102563/train2": 0.1644909306287403, + "440577711/train0": 0.09335384258660073, + "440577711/train1": 0.11977202097307982, + "440577711/train2": 0.061113295283318825, + "441046872/train0": 0.163078477436013, + "441046872/train1": 0.16749050799795617, + "441046872/train2": 0.07105401142905837, + "509430392/train0": 0.10052247125999891, + "509430392/train1": 0.049346448850599356, + "509430392/train2": 0.0815274940648898, + "585904130/train0": 0.06533502684875758, + "585904130/train1": 0.07268032916579492, + "585904130/train2": 0.25220887427510486, + "608016290/train0": 0.03209834914225741, + "608016290/train1": 0.06029905648996403, + "608016290/train2": 0.05264259469264974, + "626486813/train0": 0.1169840028584393, + "626486813/train1": 0.19272229094655288, + "626486813/train2": 0.10846025594242703, + "636680137/train0": 0.10236446816189787, + "636680137/train1": 0.1839013325840162, + "636680137/train2": 0.012908977047263237, + "655884605/train0": 0.05616248877449882, + "655884605/train1": 0.11789897082664356, + "655884605/train2": 0.027075654701186606, + "672328262/train0": 0.27100518502676063, + "672328262/train1": 0.08348589866823782, + "672328262/train2": 0.1442903896362882, + "701667486/train0": 0.00696927738017783, + "701667486/train1": 0.06264455089527499, + "701667486/train2": 0.011863507661087236, + "747260437/train0": 0.03062181894284094, + "747260437/train1": 0.15808115835262226, + "747260437/train2": 0.08122711819910011, + "750448296/train0": 0.08615947226034647, + "750448296/train1": 0.13755775384340305, + "750448296/train2": 0.10634938783489285, + "767747424/train0": 0.1306918529485389, + "767747424/train1": 0.13822538375258459, + "767747424/train2": 0.13666304064151388 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.13873127819057124, + "low": 0.0957185382719192, + "mean": 0.1139495708835625, + "n": 5, + "num_samples": 300 + }, + "mean": 0.1139495708835625, + "per_task": { + "LiftPegUpright-v1": 0.08779692862966537, + "PickCube-v1": 0.16015835031776168, + "PullCube-v1": 0.1202240575596694, + "PushCube-v1": 0.09497784791093052, + "StackCube-v1": 0.10659066999978553 + } + }, + "metric": "pairwise_causal_calibration_ece", + "micro": { + "confidence": 0.95, + "high": 0.12486695399157724, + "low": 0.09782132803662658, + "mean": 0.1113981018733552, + "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": 0.12714647100500526, + "low": 0.045656196216331324, + "mean": 0.08627731575280823, + "n": 120, + "num_samples": 300 + }, + "mean": 0.08627731575280823, + "per_seed": { + "1073161778/train0": 0.07106166187087547, + "1073161778/train1": -0.05584757187568534, + "1073161778/train2": 0.07873777798305576, + "1113371261/train0": -0.1171582058773368, + "1113371261/train1": -0.11952381111596475, + "1113371261/train2": -0.2072636199490521, + "1151410346/train0": 0.12464332364425626, + "1151410346/train1": 0.10435665617161227, + "1151410346/train2": 0.0965746708031926, + "1261930456/train0": -0.08116638752440412, + "1261930456/train1": -0.09547952648462638, + "1261930456/train2": -0.0006632054109868557, + "1324133122/train0": -0.09117531897647907, + "1324133122/train1": 0.10108263691246488, + "1324133122/train2": -0.05985917108029659, + "1324642411/train0": 0.5936226389105781, + "1324642411/train1": 0.3040303172517523, + "1324642411/train2": 0.07036169250445845, + "1375492786/train0": -0.29978209413001944, + "1375492786/train1": -0.79693146060142, + "1375492786/train2": -0.6377766658396278, + "141984615/train0": 0.08363360102690726, + "141984615/train1": 0.1560213376803143, + "141984615/train2": -0.046015598870960614, + "1461069713/train0": 0.09082588792911983, + "1461069713/train1": 0.1449663100493438, + "1461069713/train2": 0.13654486820462297, + "1480856810/train0": 0.8247860886762404, + "1480856810/train1": 0.5725138287112228, + "1480856810/train2": 0.4285300550683504, + "1509980664/train0": -0.11725750852840032, + "1509980664/train1": -0.10799597937369632, + "1509980664/train2": -0.17204314625253667, + "1522272686/train0": -0.37628277603589594, + "1522272686/train1": -0.18697224010199912, + "1522272686/train2": -0.2402513599376978, + "1549324805/train0": 0.495453729546493, + "1549324805/train1": 0.18733344993267076, + "1549324805/train2": 0.3994404854817541, + "1578990937/train0": 0.2626760081803321, + "1578990937/train1": 0.17364935131576656, + "1578990937/train2": 0.15403622504955924, + "1747178030/train0": 0.04624259356089375, + "1747178030/train1": 0.00797935515478132, + "1747178030/train2": 0.06617060088497922, + "1790188787/train0": 0.1484177313105067, + "1790188787/train1": 0.13132582655221522, + "1790188787/train2": 0.12795113455003831, + "1822210471/train0": -0.0008074765493970848, + "1822210471/train1": -0.0650108915156051, + "1822210471/train2": -0.0006465763829108795, + "1845489013/train0": -0.22001414669968217, + "1845489013/train1": -0.18822734864823312, + "1845489013/train2": -0.1547908021155039, + "1999022574/train0": 0.0012762962711883319, + "1999022574/train1": -0.08747088123670077, + "1999022574/train2": -0.10186397965031571, + "2012211287/train0": 0.1490312481371141, + "2012211287/train1": -0.036105644427698036, + "2012211287/train2": 0.09375170906247592, + "2038594191/train0": 0.19762473284895746, + "2038594191/train1": 0.3209231912001711, + "2038594191/train2": 0.31779873234459755, + "2054899126/train0": 0.15266150175970736, + "2054899126/train1": 0.1715300859769803, + "2054899126/train2": 0.16346432037867953, + "2061635888/train0": 0.12372212373443373, + "2061635888/train1": 0.049366979703384384, + "2061635888/train2": 0.015355053119340732, + "211245842/train0": 0.04424689933210735, + "211245842/train1": 0.13077847063672782, + "211245842/train2": 0.04386283650419778, + "278008524/train0": 0.2219557334710413, + "278008524/train1": 0.2843962133295508, + "278008524/train2": 0.24145145365463994, + "307068879/train0": 0.08783512114247624, + "307068879/train1": -0.13881394667704933, + "307068879/train2": 0.14397102567158404, + "412102563/train0": 0.0942295054093702, + "412102563/train1": 0.029952749201257664, + "412102563/train2": -0.004657759279381279, + "440577711/train0": 0.5465320030087847, + "440577711/train1": 0.629918599045802, + "440577711/train2": 0.6241310075694073, + "441046872/train0": 0.046645784157056724, + "441046872/train1": -0.17358234568930467, + "441046872/train2": -0.022380468606574055, + "509430392/train0": 0.4269940390592024, + "509430392/train1": 0.37833433181770104, + "509430392/train2": 0.282290647599035, + "585904130/train0": 0.10499618981526784, + "585904130/train1": 0.06533471439439086, + "585904130/train2": 0.09915967492706912, + "608016290/train0": 0.5011206164338899, + "608016290/train1": 0.2910589829678913, + "608016290/train2": 0.4536901796255097, + "626486813/train0": 0.5216712480085811, + "626486813/train1": 0.4816967914950128, + "626486813/train2": 0.39041494710515146, + "636680137/train0": 0.4324165282764497, + "636680137/train1": 0.0489088109005007, + "636680137/train2": 0.13337534819821673, + "655884605/train0": 0.07276876911827704, + "655884605/train1": 0.1554209538300596, + "655884605/train2": 0.11907170730895571, + "672328262/train0": -0.087075237393225, + "672328262/train1": -0.23567605508015027, + "672328262/train2": -0.14512362014342817, + "701667486/train0": 0.18126309408722283, + "701667486/train1": 0.15700516615012938, + "701667486/train2": 0.1922641707596195, + "747260437/train0": -0.08999271197759756, + "747260437/train1": -0.26165809689080527, + "747260437/train2": -0.061796090652556146, + "750448296/train0": 0.005890085195897528, + "750448296/train1": -0.19810309176890203, + "750448296/train2": -0.15048924424583973, + "767747424/train0": -0.028685105901352856, + "767747424/train1": -0.013317565754431063, + "767747424/train2": 0.02448240692729431 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.22447325377947683, + "low": -0.13330904884281763, + "mean": 0.05126313589726559, + "n": 5, + "num_samples": 300 + }, + "mean": 0.05126313589726559, + "per_task": { + "LiftPegUpright-v1": 0.09934494198504117, + "PickCube-v1": -0.2678354586918643, + "PullCube-v1": 0.0587216273047277, + "PushCube-v1": 0.32846837901195985, + "StackCube-v1": 0.03761618987646351 + } + }, + "metric": "predicted_margin", + "micro": { + "confidence": 0.95, + "high": 0.1248668767428929, + "low": 0.0427827948842033, + "mean": 0.08503163117842487, + "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": 6.555555555555556, + "low": 4.948611111111111, + "mean": 5.677777777777778, + "n": 120, + "num_samples": 300 + }, + "mean": 5.677777777777778, + "per_seed": { + "1073161778/train0": 6.0, + "1073161778/train1": 7.0, + "1073161778/train2": 6.0, + "1113371261/train0": 2.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 9.0, + "1151410346/train1": 3.0, + "1151410346/train2": 8.0, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 4.0, + "1324133122/train0": 6.0, + "1324133122/train1": 2.0, + "1324133122/train2": 1.0, + "1324642411/train0": 13.0, + "1324642411/train1": 4.0, + "1324642411/train2": 2.0, + "1375492786/train0": 3.0, + "1375492786/train1": 4.0, + "1375492786/train2": 2.0, + "141984615/train0": 6.0, + "141984615/train1": 15.0, + "141984615/train2": 3.0, + "1461069713/train0": 1.0, + "1461069713/train1": 13.0, + "1461069713/train2": 3.0, + "1480856810/train0": 0.0, + "1480856810/train1": 9.0, + "1480856810/train2": 4.0, + "1509980664/train0": 10.0, + "1509980664/train1": 9.0, + "1509980664/train2": 9.0, + "1522272686/train0": 1.0, + "1522272686/train1": 2.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 1.0, + "1578990937/train0": 9.0, + "1578990937/train1": 0.0, + "1578990937/train2": 7.0, + "1747178030/train0": 3.0, + "1747178030/train1": 7.0, + "1747178030/train2": 15.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 9.0, + "1822210471/train0": 5.0, + "1822210471/train1": 3.0, + "1822210471/train2": 4.0, + "1845489013/train0": 5.0, + "1845489013/train1": 1.0, + "1845489013/train2": 8.0, + "1999022574/train0": 5.0, + "1999022574/train1": 8.0, + "1999022574/train2": 7.0, + "2012211287/train0": 13.0, + "2012211287/train1": 13.0, + "2012211287/train2": 14.0, + "2038594191/train0": 5.0, + "2038594191/train1": 9.0, + "2038594191/train2": 3.0, + "2054899126/train0": 4.5, + "2054899126/train1": 7.0, + "2054899126/train2": 8.0, + "2061635888/train0": 3.0, + "2061635888/train1": 2.0, + "2061635888/train2": 7.0, + "211245842/train0": 8.0, + "211245842/train1": 4.0, + "211245842/train2": 4.0, + "278008524/train0": 10.0, + "278008524/train1": 8.0, + "278008524/train2": 14.0, + "307068879/train0": 11.0, + "307068879/train1": 6.0, + "307068879/train2": 3.0, + "412102563/train0": 10.0, + "412102563/train1": 8.0, + "412102563/train2": 7.0, + "440577711/train0": 1.0, + "440577711/train1": 10.0, + "440577711/train2": 13.0, + "441046872/train0": 7.0, + "441046872/train1": 6.0, + "441046872/train2": 1.0, + "509430392/train0": 15.0, + "509430392/train1": 3.0, + "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": 11.0, + "626486813/train1": 10.0, + "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": 5.0, + "672328262/train1": 3.0, + "672328262/train2": 4.0, + "701667486/train0": 10.0, + "701667486/train1": 11.0, + "701667486/train2": 7.0, + "747260437/train0": 14.0, + "747260437/train1": 4.0, + "747260437/train2": 6.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 8.333333333333334, + "750448296/train2": 10.333333333333334, + "767747424/train0": 7.0, + "767747424/train1": 7.0, + "767747424/train2": 5.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 6.673212121212122, + "low": 3.1532121212121216, + "mean": 5.104323232323233, + "n": 5, + "num_samples": 300 + }, + "mean": 5.104323232323233, + "per_task": { + "LiftPegUpright-v1": 7.2, + "PickCube-v1": 1.3333333333333333, + "PullCube-v1": 6.16, + "PushCube-v1": 5.222222222222222, + "StackCube-v1": 5.606060606060606 + } + }, + "metric": "selected_candidate_index", + "micro": { + "confidence": 0.95, + "high": 6.548611111111111, + "low": 5.208333333333333, + "mean": 5.923611111111111, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.42916666666666664, + "low": 0.2604166666666667, + "mean": 0.33958333333333335, + "n": 120, + "num_samples": 300 + }, + "mean": 0.33958333333333335, + "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": 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": 1.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.5, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.5, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 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": 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": 0.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 0.0, + "747260437/train1": 1.0, + "747260437/train2": 0.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.41393939393939394, + "low": 0.12460606060606061, + "mean": 0.27793939393939393, + "n": 5, + "num_samples": 300 + }, + "mean": 0.27793939393939393, + "per_task": { + "LiftPegUpright-v1": 0.26666666666666666, + "PickCube-v1": 0.0, + "PullCube-v1": 0.32, + "PushCube-v1": 0.5, + "StackCube-v1": 0.30303030303030304 + } + }, + "metric": "selected_success", + "micro": { + "confidence": 0.95, + "high": 0.3888888888888889, + "low": 0.2569444444444444, + "mean": 0.3263888888888889, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_success_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.1, + "low": -0.029166666666666667, + "mean": 0.03333333333333333, + "n": 120, + "num_samples": 300 + }, + "mean": 0.03333333333333333, + "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": 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": 1.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": -1.0, + "2038594191/train1": -1.0, + "2038594191/train2": -1.0, + "2054899126/train0": -0.5, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.5, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 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": 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": -1.0, + "747260437/train1": 0.0, + "747260437/train2": -1.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.0, + "767747424/train1": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.21333333333333332, + "low": -0.010666666666666668, + "mean": 0.07806060606060605, + "n": 5, + "num_samples": 300 + }, + "mean": 0.07806060606060605, + "per_task": { + "LiftPegUpright-v1": 0.06666666666666667, + "PickCube-v1": 0.0, + "PullCube-v1": -0.04, + "PushCube-v1": 0.3333333333333333, + "StackCube-v1": 0.030303030303030304 + } + }, + "metric": "selected_success_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.09722222222222222, + "low": -0.034722222222222224, + "mean": 0.034722222222222224, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.9767636695663291, + "low": 0.6694472551869696, + "mean": 0.8134982612139235, + "n": 120, + "num_samples": 300 + }, + "mean": 0.8134982612139235, + "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": 0.11045536398887634, + "1324642411/train1": 0.11658626794815063, + "1324642411/train2": 0.10647052526473999, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.007176061626523733, + "141984615/train1": 0.030119141563773155, + "141984615/train2": 0.02716982364654541, + "1461069713/train0": 0.014822185039520264, + "1461069713/train1": 0.006729066371917725, + "1461069713/train2": 0.09997648000717163, + "1480856810/train0": 2.0, + "1480856810/train1": 0.21981161832809448, + "1480856810/train2": 0.21894000470638275, + "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": 0.0260902252048254, + "1549324805/train1": 0.0303483996540308, + "1549324805/train2": 0.04690708592534065, + "1578990937/train0": 2.0, + "1578990937/train1": 0.12732218205928802, + "1578990937/train2": 0.15726424753665924, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 2.0, + "1790188787/train1": 0.10585354268550873, + "1790188787/train2": 0.10904359817504883, + "1822210471/train0": 0.06247224286198616, + "1822210471/train1": 0.045472364872694016, + "1822210471/train2": 0.08447941392660141, + "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": 0.7026570439338684, + "2012211287/train1": 0.7236053347587585, + "2012211287/train2": 0.6934840679168701, + "2038594191/train0": 0.04285776615142822, + "2038594191/train1": 0.07674666494131088, + "2038594191/train2": 0.12977690994739532, + "2054899126/train0": 1.0403175354003906, + "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": 1.0287093725055456, + "278008524/train1": 1.0095977680757642, + "278008524/train2": 1.0227726306766272, + "307068879/train0": 0.19761089980602264, + "307068879/train1": 0.6000905632972717, + "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": 0.6999551653862, + "441046872/train2": 0.6999551653862, + "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": 2.0, + "608016290/train1": 2.0, + "608016290/train2": 0.6789124011993408, + "626486813/train0": 0.009900043485686183, + "626486813/train1": 0.006450424436479807, + "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": 0.04013325646519661, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 0.061248183250427246, + "747260437/train1": 2.0, + "747260437/train2": 0.06438694149255753, + "750448296/train0": 0.125017994393905, + "750448296/train1": 0.109618179500103, + "750448296/train2": 0.09164678926269214, + "767747424/train0": 0.5305264240596443, + "767747424/train1": 0.5966028785333037, + "767747424/train2": 0.5695318393409252 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.9610522832274435, + "low": 0.6580356173443072, + "mean": 0.8072238209012783, + "n": 5, + "num_samples": 300 + }, + "mean": 0.8072238209012783, + "per_task": { + "LiftPegUpright-v1": 0.9396337101856868, + "PickCube-v1": 0.5839591026306152, + "PullCube-v1": 0.6822500004371007, + "PushCube-v1": 1.0611259018381436, + "StackCube-v1": 0.769150389414845 + } + }, + "metric": "selected_utility", + "micro": { + "confidence": 0.95, + "high": 0.9093973265650371, + "low": 0.6554252143218441, + "mean": 0.7742872366474735, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_utility_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.14305571060378697, + "low": -0.09107819030889205, + "mean": 0.023380198916129302, + "n": 120, + "num_samples": 300 + }, + "mean": 0.023380198916129302, + "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": 0.0039848387241363525, + "1324642411/train1": 0.010115742683410645, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": -0.09565818542614579, + "141984615/train1": -0.07271510548889637, + "141984615/train2": -0.07566442340612411, + "1461069713/train0": -0.07908087968826294, + "1461069713/train1": -0.08717399835586548, + "1461069713/train2": 0.006073415279388428, + "1480856810/train0": 1.9129659235477448, + "1480856810/train1": 0.13277754187583923, + "1480856810/train2": 0.1319059282541275, + "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": -0.07842461206018925, + "1549324805/train1": -0.07416643761098385, + "1549324805/train2": -0.057607751339673996, + "1578990937/train0": 1.17586350440979, + "1578990937/train1": -0.6968143135309219, + "1578990937/train2": -0.6668722480535507, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.87523253262043, + "1790188787/train1": -0.01891392469406128, + "1790188787/train2": -0.01572386920452118, + "1822210471/train0": -0.009443063288927078, + "1822210471/train1": -0.026442941278219223, + "1822210471/train2": 0.012564107775688171, + "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": -0.020948290824890137, + "2012211287/train1": 0.0, + "2012211287/train2": -0.030121266841888428, + "2038594191/train0": -1.9571422338485718, + "2038594191/train1": -1.9232533350586891, + "2038594191/train2": -1.8702230900526047, + "2054899126/train0": -0.9596824645996094, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.985974382609129, + "278008524/train1": 0.9668627781793475, + "278008524/train2": 0.9800376407802105, + "307068879/train0": -0.4024796634912491, + "307068879/train1": 0.0, + "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": 0.0, + "441046872/train2": 0.0, + "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": 1.274298369884491, + "608016290/train1": 1.274298369884491, + "608016290/train2": -0.04678922891616821, + "626486813/train0": 0.004801700764801353, + "626486813/train1": 0.0013520817155949771, + "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": -1.9598667435348034, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": -1.9387518167495728, + "747260437/train1": 0.0, + "747260437/train2": -1.9356130585074425, + "750448296/train0": 0.007424845049778621, + "750448296/train1": -0.007974969844023386, + "750448296/train2": -0.02594636008143425, + "767747424/train0": -0.04267291794531047, + "767747424/train1": 0.023403536528348923, + "767747424/train2": -0.0036675026640295982 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.39773680510846043, + "low": -0.05779964217698822, + "mean": 0.11166402372726322, + "n": 5, + "num_samples": 300 + }, + "mean": 0.11166402372726322, + "per_task": { + "LiftPegUpright-v1": -0.010236828525861105, + "PickCube-v1": 0.0, + "PullCube-v1": -0.09292046078636, + "PushCube-v1": 0.6687217230598131, + "StackCube-v1": -0.007244315111275875 + } + }, + "metric": "selected_utility_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.15572734094651727, + "low": -0.09504848822366006, + "mean": 0.03246765020513623, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selector_regret": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5917341341859557, + "low": 0.35403546871255254, + "mean": 0.4665542311901744, + "n": 120, + "num_samples": 300 + }, + "mean": 0.4665542311901744, + "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": 0.03360468149185181, + "1324642411/train1": 1.8834137320518494, + "1324642411/train2": 1.89352947473526, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.008001843933016062, + "141984615/train1": 0.006891986355185509, + "141984615/train2": 0.031781893223524094, + "1461069713/train0": 0.04537618160247803, + "1461069713/train1": 0.009855946525931358, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 1.7801883816719055, + "1480856810/train2": 1.7810599952936172, + "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": 0.0347811970859766, + "1549324805/train1": 0.16857311688363552, + "1549324805/train2": 0.06619475409388542, + "1578990937/train0": 0.0, + "1578990937/train1": 1.872677817940712, + "1578990937/train2": 1.8427357524633408, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.8941464573144913, + "1790188787/train2": 1.8909564018249512, + "1822210471/train0": 1.9375277571380138, + "1822210471/train1": 1.954527635127306, + "1822210471/train2": 1.9155205860733986, + "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": 1.2973429560661316, + "2012211287/train1": 1.2763946652412415, + "2012211287/train2": 1.3065159320831299, + "2038594191/train0": 1.9571422338485718, + "2038594191/train1": 1.9232533350586891, + "2038594191/train2": 1.8702230900526047, + "2054899126/train0": 0.036965616047382355, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "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": 0.04464695043861866, + "278008524/train1": 0.028076826594769955, + "278008524/train2": 0.11317628063261509, + "307068879/train0": 0.01394665241241455, + "307068879/train1": 0.0, + "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": 0.023403704166412354, + "441046872/train2": 0.019179701805114746, + "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": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.03281521797180176, + "626486813/train0": 0.034795432118698955, + "626486813/train1": 0.029377450235188007, + "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": 1.9598667435348034, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.04714745283126831, + "747260437/train1": 0.0, + "747260437/train2": 0.06071493774652481, + "750448296/train0": 0.6276423335075378, + "750448296/train1": 1.2608850946029027, + "750448296/train2": 1.25880187874039, + "767747424/train0": 0.5119027637410909, + "767747424/train1": 0.4663502126932144, + "767747424/train2": 0.4976352555677295 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.6077285580653132, + "low": 0.17293525018475273, + "mean": 0.40844574604446837, + "n": 5, + "num_samples": 300 + }, + "mean": 0.40844574604446837, + "per_task": { + "LiftPegUpright-v1": 0.47343848447004955, + "PickCube-v1": 0.0, + "PullCube-v1": 0.4528969661643108, + "PushCube-v1": 0.7246555131342676, + "StackCube-v1": 0.3912377664537141 + } + }, + "metric": "selector_regret", + "micro": { + "confidence": 0.95, + "high": 0.6050060989574477, + "low": 0.3520631702493928, + "mean": 0.4654409392969683, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "success_selector_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.3256944444444444, + "low": 0.18680555555555556, + "mean": 0.2534722222222222, + "n": 120, + "num_samples": 300 + }, + "mean": 0.2534722222222222, + "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": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 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": 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": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.3424242424242424, + "low": 0.10909090909090909, + "mean": 0.23220202020202022, + "n": 5, + "num_samples": 300 + }, + "mean": 0.23220202020202022, + "per_task": { + "LiftPegUpright-v1": 0.3333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": 0.22666666666666666, + "PushCube-v1": 0.3888888888888889, + "StackCube-v1": 0.21212121212121213 + } + }, + "metric": "success_selector_gap", + "micro": { + "confidence": 0.95, + "high": 0.3263888888888889, + "low": 0.18055555555555555, + "mean": 0.25, + "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": 0.006988199735446857, + "low": -0.04978091939463336, + "mean": -0.02190632784962469, + "n": 120, + "num_samples": 300 + }, + "mean": -0.02190632784962469, + "per_seed": { + "1073161778/train0": 0.028235521236802175, + "1073161778/train1": 0.028235521236802175, + "1073161778/train2": 0.028235521236802175, + "1113371261/train0": 0.13235082030966006, + "1113371261/train1": 0.13235082030966006, + "1113371261/train2": 0.13235082030966006, + "1151410346/train0": 0.13235082030966006, + "1151410346/train1": 0.13235082030966006, + "1151410346/train2": 0.13235082030966006, + "1261930456/train0": 0.07420081675594419, + "1261930456/train1": 0.07420081675594419, + "1261930456/train2": 0.07420081675594419, + "1324133122/train0": 0.13235082030966006, + "1324133122/train1": 0.13235082030966006, + "1324133122/train2": 0.13235082030966006, + "1324642411/train0": 0.07420081675594419, + "1324642411/train1": 0.07420081675594419, + "1324642411/train2": 0.07420081675594419, + "1375492786/train0": -0.2012384152043573, + "1375492786/train1": -0.2012384152043573, + "1375492786/train2": -0.2012384152043573, + "141984615/train0": -0.2012384152043573, + "141984615/train1": -0.2012384152043573, + "141984615/train2": -0.2012384152043573, + "1461069713/train0": -0.2012384152043573, + "1461069713/train1": -0.2012384152043573, + "1461069713/train2": -0.2012384152043573, + "1480856810/train0": 0.07420081675594419, + "1480856810/train1": 0.07420081675594419, + "1480856810/train2": 0.07420081675594419, + "1509980664/train0": 0.13235082030966006, + "1509980664/train1": 0.13235082030966006, + "1509980664/train2": 0.13235082030966006, + "1522272686/train0": 0.5025584143631503, + "1522272686/train1": 0.5025584143631503, + "1522272686/train2": 0.5025584143631503, + "1549324805/train0": -0.2012384152043573, + "1549324805/train1": -0.2012384152043573, + "1549324805/train2": -0.2012384152043573, + "1578990937/train0": 0.13235082030966006, + "1578990937/train1": 0.13235082030966006, + "1578990937/train2": 0.13235082030966006, + "1747178030/train0": -0.2012384152043573, + "1747178030/train1": -0.2012384152043573, + "1747178030/train2": -0.2012384152043573, + "1790188787/train0": 0.07420081675594419, + "1790188787/train1": 0.07420081675594419, + "1790188787/train2": 0.07420081675594419, + "1822210471/train0": -0.2012384152043573, + "1822210471/train1": -0.2012384152043573, + "1822210471/train2": -0.2012384152043573, + "1845489013/train0": -0.2012384152043573, + "1845489013/train1": -0.2012384152043573, + "1845489013/train2": -0.2012384152043573, + "1999022574/train0": 0.13235082030966006, + "1999022574/train1": 0.13235082030966006, + "1999022574/train2": 0.13235082030966006, + "2012211287/train0": 0.028235521236802175, + "2012211287/train1": 0.028235521236802175, + "2012211287/train2": 0.028235521236802175, + "2038594191/train0": -0.2012384152043573, + "2038594191/train1": -0.2012384152043573, + "2038594191/train2": -0.2012384152043573, + "2054899126/train0": -0.2012384152043573, + "2054899126/train1": -0.2012384152043573, + "2054899126/train2": -0.2012384152043573, + "2061635888/train0": 0.13235082030966006, + "2061635888/train1": 0.13235082030966006, + "2061635888/train2": 0.13235082030966006, + "211245842/train0": 0.13235082030966006, + "211245842/train1": 0.13235082030966006, + "211245842/train2": 0.13235082030966006, + "278008524/train0": -0.2012384152043573, + "278008524/train1": -0.2012384152043573, + "278008524/train2": -0.2012384152043573, + "307068879/train0": 0.028235521236802175, + "307068879/train1": 0.028235521236802175, + "307068879/train2": 0.028235521236802175, + "412102563/train0": 0.13235082030966006, + "412102563/train1": 0.13235082030966006, + "412102563/train2": 0.13235082030966006, + "440577711/train0": 0.07420081675594419, + "440577711/train1": 0.07420081675594419, + "440577711/train2": 0.07420081675594419, + "441046872/train0": 0.028235521236802175, + "441046872/train1": 0.028235521236802175, + "441046872/train2": 0.028235521236802175, + "509430392/train0": 0.07420081675594419, + "509430392/train1": 0.07420081675594419, + "509430392/train2": 0.07420081675594419, + "585904130/train0": -0.2012384152043573, + "585904130/train1": -0.2012384152043573, + "585904130/train2": -0.2012384152043573, + "608016290/train0": 0.028235521236802175, + "608016290/train1": 0.028235521236802175, + "608016290/train2": 0.028235521236802175, + "626486813/train0": -0.2012384152043573, + "626486813/train1": -0.2012384152043573, + "626486813/train2": -0.2012384152043573, + "636680137/train0": -0.2012384152043573, + "636680137/train1": -0.2012384152043573, + "636680137/train2": -0.2012384152043573, + "655884605/train0": 0.13235082030966006, + "655884605/train1": 0.13235082030966006, + "655884605/train2": 0.13235082030966006, + "672328262/train0": -0.2012384152043573, + "672328262/train1": -0.2012384152043573, + "672328262/train2": -0.2012384152043573, + "701667486/train0": 0.13235082030966006, + "701667486/train1": 0.13235082030966006, + "701667486/train2": 0.13235082030966006, + "747260437/train0": -0.2012384152043573, + "747260437/train1": -0.2012384152043573, + "747260437/train2": -0.2012384152043573, + "750448296/train0": -0.20123841520435728, + "750448296/train1": -0.20123841520435728, + "750448296/train2": -0.20123841520435728, + "767747424/train0": -0.2012384152043573, + "767747424/train1": -0.2012384152043573, + "767747424/train2": -0.2012384152043573 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.32202231621643945, + "low": -0.08862578081332213, + "mean": 0.10722143149223981, + "n": 5, + "num_samples": 300 + }, + "mean": 0.10722143149223981, + "per_task": { + "LiftPegUpright-v1": 0.028235521236802175, + "PickCube-v1": 0.5025584143631503, + "PullCube-v1": -0.2012384152043576, + "PushCube-v1": 0.0742008167559442, + "StackCube-v1": 0.13235082030965997 + } + }, + "metric": "tau", + "micro": { + "confidence": 0.95, + "high": -0.024000803025241515, + "low": -0.07546951652137965, + "mean": -0.05179500907541345, + "n": 144, + "num_samples": 300 + }, + "n": 144 + } + }, + "target": "utility_margin", + "tau": { + "LiftPegUpright-v1": 0.028235521236802175, + "PickCube-v1": 0.5025584143631503, + "PullCube-v1": -0.2012384152043573, + "PushCube-v1": 0.07420081675594419, + "StackCube-v1": 0.13235082030966006, + "__global__": 0.018528842023845943 + }, + "threshold_scope": "task", + "weights": [ + -0.06419078293938928, + 0.07864473400199715, + 0.05670874450563532, + 0.030682030137095147, + -0.017874347910056848, + -0.02068898371618812, + -0.14209815512665164, + -0.15088977540454895, + 0.058742278143481846, + 0.04013778871314058, + 0.029322121431418394, + 0.029091810230969208, + -0.04043658921086313, + -0.0310373634235519, + 0.00947510210158854, + -0.0061723529946101805, + -0.015421451059575333, + 0.03658962429356194, + -1.5425149460816038e-16, + -0.011433546961072683, + -0.09105471904483498, + 0.24636175694603346, + 0.2830253454973892, + 0.08308278332924052, + -0.18110157639978264, + -0.11400591941440252, + 0.006962277279499496, + -0.07089795409078059, + 0.23677529599504066, + 0.2830446516154337, + 5.828605755163547e-16, + 0.20725888095715322, + 0.21619497138712734, + 0.01867862192678118, + -0.09424784020821385, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/metrics_by_seed.json b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/metrics_by_seed.json new file mode 100644 index 0000000000000000000000000000000000000000..b3357bc80f2bef7327c10c6a793e72233115f334 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_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": 0.6218487394957983, + "pairwise_causal_calibration_confidence": 0.5252122988397412, + "pairwise_causal_calibration_ece": 0.09663644065605714, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.07106166187087547, + "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": 0.028235521236802175 + }, + "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": 0.6083333333333333, + "pairwise_causal_calibration_confidence": 0.5228008383368301, + "pairwise_causal_calibration_ece": 0.08553249499650317, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.05584757187568534, + "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": 0.028235521236802175 + }, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5254991602652951, + "pairwise_causal_calibration_ece": 0.14116750640137155, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.07873777798305576, + "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": 0.028235521236802175 + }, + "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": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5381830912240237, + "pairwise_causal_calibration_ece": 0.10961166265259512, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1171582058773368, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10274861752986908, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.897251382470131, + "tau": 0.13235082030966006 + }, + "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": 0.5386619397445, + "pairwise_causal_calibration_ece": 0.06848091739835714, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11952381111596475, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18093222379684448, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02250880002975464, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8190677762031555, + "tau": 0.13235082030966006 + }, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5353470658344741, + "pairwise_causal_calibration_ece": 0.036081505594097285, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2072636199490521, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "tau": 0.13235082030966006 + }, + "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": 0.5296341103177351, + "pairwise_causal_calibration_ece": 0.015348396032020872, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.12464332364425626, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8663521409034729, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.6403508484363556, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.133647859096527, + "tau": 0.13235082030966006 + }, + "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": 0.44761904761904764, + "pairwise_causal_calibration_confidence": 0.528839557722319, + "pairwise_causal_calibration_ece": 0.08122051010327141, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.10435665617161227, + "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": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.7739987075328827, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.13235082030966006 + }, + "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": 0.38095238095238093, + "pairwise_causal_calibration_confidence": 0.530463392824977, + "pairwise_causal_calibration_ece": 0.14951101187259608, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.0965746708031926, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.32620343565940857, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4477952718734741, + "tau": 0.13235082030966006 + }, + "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": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5174492210423474, + "pairwise_causal_calibration_ece": 0.06290376649689283, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": -0.08116638752440412, + "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": 0.07420081675594419 + }, + "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": 0.5272727272727272, + "pairwise_causal_calibration_confidence": 0.5193562339693999, + "pairwise_causal_calibration_ece": 0.007916493303327332, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": -0.09547952648462638, + "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": 0.07420081675594419 + }, + "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": 0.6190476190476191, + "pairwise_causal_calibration_confidence": 0.5192244871616464, + "pairwise_causal_calibration_ece": 0.0998231318859727, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.0006632054109868557, + "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": 0.07420081675594419 + }, + "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": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5557404501598795, + "pairwise_causal_calibration_ece": 0.16288330730273656, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.09117531897647907, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.054989203810691833, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9450107961893082, + "tau": 0.13235082030966006 + }, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5544807305695781, + "pairwise_causal_calibration_ece": 0.10154169690861004, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.10108263691246488, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1324586570262909, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04025374352931976, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.867541342973709, + "tau": 0.13235082030966006 + }, + "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": 0.21428571428571427, + "pairwise_causal_calibration_confidence": 0.5529109212469531, + "pairwise_causal_calibration_ece": 0.33862520696123877, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.05985917108029659, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19291165471076965, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.10070674121379852, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8070883452892303, + "tau": 0.13235082030966006 + }, + "1324642411/train0": { + "base_success": 0.0, + "base_utility": 0.10647052526473999, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5583333333333333, + "pairwise_causal_calibration_confidence": 0.5085746791327035, + "pairwise_causal_calibration_ece": 0.04975865420062986, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5936226389105781, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14406004548072815, + "selected_candidate_index": 13.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11045536398887634, + "selected_utility_gain_over_base": 0.0039848387241363525, + "selector_regret": 0.03360468149185181, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8559399545192719, + "tau": 0.07420081675594419 + }, + "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": 0.680672268907563, + "pairwise_causal_calibration_confidence": 0.5079157579147822, + "pairwise_causal_calibration_ece": 0.1727565109927809, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.3040303172517523, + "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": 0.11658626794815063, + "selected_utility_gain_over_base": 0.010115742683410645, + "selector_regret": 1.8834137320518494, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.07420081675594419 + }, + "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": 0.5181818181818182, + "pairwise_causal_calibration_confidence": 0.5077976736149897, + "pairwise_causal_calibration_ece": 0.010384144566828457, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.07036169250445845, + "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": 0.07420081675594419 + }, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5289748634459925, + "pairwise_causal_calibration_ece": 0.1376918032206741, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": -0.29978209413001944, + "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": -0.2012384152043573 + }, + "1375492786/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5614035087719298, + "pairwise_causal_calibration_confidence": 0.5257629988292399, + "pairwise_causal_calibration_ece": 0.03564050994268986, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.79693146060142, + "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": -0.2012384152043573 + }, + "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": 0.8095238095238095, + "pairwise_causal_calibration_confidence": 0.5337083842579382, + "pairwise_causal_calibration_ece": 0.27581542526587133, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.6377766658396278, + "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": -0.2012384152043573 + }, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.523230296021853, + "pairwise_causal_calibration_ece": 0.10656362935518632, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.08363360102690726, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007176061626523733, + "selected_utility_gain_over_base": -0.09565818542614579, + "selector_regret": 0.008001843933016062, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "tau": -0.2012384152043573 + }, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.5259787015164451, + "pairwise_causal_calibration_ece": 0.049021298483554854, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.1560213376803143, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.037011127918958664, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.030119141563773155, + "selected_utility_gain_over_base": -0.07271510548889637, + "selector_regret": 0.006891986355185509, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24876614287495613, + "tau": -0.2012384152043573 + }, + "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": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5258876236158011, + "pairwise_causal_calibration_ece": 0.007445709717532223, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.046015598870960614, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058951716870069504, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.02716982364654541, + "selected_utility_gain_over_base": -0.07566442340612411, + "selector_regret": 0.031781893223524094, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "tau": -0.2012384152043573 + }, + "1461069713/train0": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.30833333333333335, + "pairwise_causal_calibration_confidence": 0.5145099591738125, + "pairwise_causal_calibration_ece": 0.20617662584047913, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.09082588792911983, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06019836664199829, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.014822185039520264, + "selected_utility_gain_over_base": -0.07908087968826294, + "selector_regret": 0.04537618160247803, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0991440862417221, + "tau": -0.2012384152043573 + }, + "1461069713/train1": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5141448492443679, + "pairwise_causal_calibration_ece": 0.09747818257770119, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.1449663100493438, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "selected_candidate_index": 13.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.006729066371917725, + "selected_utility_gain_over_base": -0.08717399835586548, + "selector_regret": 0.009855946525931358, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14275743998587132, + "tau": -0.2012384152043573 + }, + "1461069713/train2": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.5185093923599849, + "pairwise_causal_calibration_ece": 0.08149060764001503, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.13654486820462297, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09997648000717163, + "selected_utility_gain_over_base": 0.006073415279388428, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05936597287654877, + "tau": -0.2012384152043573 + }, + "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": 0.8383838383838383, + "pairwise_causal_calibration_confidence": 0.5267941522147715, + "pairwise_causal_calibration_ece": 0.3115896861690668, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.8247860886762404, + "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": 0.07420081675594419 + }, + "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": 0.5180113719626723, + "pairwise_causal_calibration_ece": 0.061820560810436964, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5725138287112228, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21981161832809448, + "selected_utility_gain_over_base": 0.13277754187583923, + "selector_regret": 1.7801883816719055, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.07420081675594419 + }, + "1480856810/train2": { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5630252100840336, + "pairwise_causal_calibration_confidence": 0.5179749462983955, + "pairwise_causal_calibration_ece": 0.045050263785638145, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4285300550683504, + "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": 0.21894000470638275, + "selected_utility_gain_over_base": 0.1319059282541275, + "selector_regret": 1.7810599952936172, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.07420081675594419 + }, + "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": 0.43333333333333335, + "pairwise_causal_calibration_confidence": 0.5388105224755593, + "pairwise_causal_calibration_ece": 0.10547718914222602, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.11725750852840032, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.13235082030966006 + }, + "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": 0.23809523809523808, + "pairwise_causal_calibration_confidence": 0.5371878138277117, + "pairwise_causal_calibration_ece": 0.29909257573247366, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": -0.10799597937369632, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.13235082030966006 + }, + "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": 0.4888888888888889, + "pairwise_causal_calibration_confidence": 0.5391151823956642, + "pairwise_causal_calibration_ece": 0.05022629350677536, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.17204314625253667, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.13235082030966006 + }, + "1522272686/train0": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5358486568758676, + "pairwise_causal_calibration_ece": 0.03584865687586758, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.37628277603589594, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18325337767601013, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8167466223239899, + "tau": 0.5025584143631503 + }, + "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": 0.2222222222222222, + "pairwise_causal_calibration_confidence": 0.5423563498452709, + "pairwise_causal_calibration_ece": 0.32013412762304877, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.18697224010199912, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07014383375644684, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9298561662435532, + "tau": 0.5025584143631503 + }, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5411589331210354, + "pairwise_causal_calibration_ece": 0.12449226645436867, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.2402513599376978, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14937306940555573, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8506269305944443, + "tau": 0.5025584143631503 + }, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5117256505009358, + "pairwise_causal_calibration_ece": 0.15494101616573086, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.495453729546493, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.060871422290802, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0260902252048254, + "selected_utility_gain_over_base": -0.07842461206018925, + "selector_regret": 0.0347811970859766, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.23658466339111328, + "tau": -0.2012384152043573 + }, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.5062538341615629, + "pairwise_causal_calibration_ece": 0.06874616583843707, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.18733344993267076, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0303483996540308, + "selected_utility_gain_over_base": -0.07416643761098385, + "selector_regret": 0.16857311688363552, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "tau": -0.2012384152043573 + }, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.5094928991923253, + "pairwise_causal_calibration_ece": 0.06550710080767463, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3994404854817541, + "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": -0.2012384152043573 + }, + "1578990937/train0": { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7008547008547008, + "pairwise_causal_calibration_confidence": 0.5507814264578395, + "pairwise_causal_calibration_ece": 0.15007327439686133, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.2626760081803321, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.17586350440979, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.13235082030966006 + }, + "1578990937/train1": { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42727272727272725, + "pairwise_causal_calibration_confidence": 0.5448159860797186, + "pairwise_causal_calibration_ece": 0.11754325880699122, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.17364935131576656, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12732218205928802, + "selected_utility_gain_over_base": -0.6968143135309219, + "selector_regret": 1.872677817940712, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.13235082030966006 + }, + "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": 0.36134453781512604, + "pairwise_causal_calibration_confidence": 0.5460279556731249, + "pairwise_causal_calibration_ece": 0.18468341785799877, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.15403622504955924, + "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": 0.15726424753665924, + "selected_utility_gain_over_base": -0.6668722480535507, + "selector_regret": 1.8427357524633408, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.13235082030966006 + }, + "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": 0.04624259356089375, + "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": -0.2012384152043573 + }, + "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": 0.00797935515478132, + "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": -0.2012384152043573 + }, + "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": 0.06617060088497922, + "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": -0.2012384152043573 + }, + "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": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.5077769577736889, + "pairwise_causal_calibration_ece": 0.12595877595550709, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.1484177313105067, + "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": 0.07420081675594419 + }, + "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": 0.37719298245614036, + "pairwise_causal_calibration_confidence": 0.5066944209005328, + "pairwise_causal_calibration_ece": 0.1295014384443925, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 0.13132582655221522, + "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": 0.07420081675594419 + }, + "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": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.5083204445849703, + "pairwise_causal_calibration_ece": 0.12650226276678844, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.12795113455003831, + "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": 0.10904359817504883, + "selected_utility_gain_over_base": -0.01572386920452118, + "selector_regret": 1.8909564018249512, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.07420081675594419 + }, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.549416493223813, + "pairwise_causal_calibration_ece": 0.08571587301053908, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": -0.0008074765493970848, + "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": -0.2012384152043573 + }, + "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": 0.557604964366195, + "pairwise_causal_calibration_ece": 0.09545626012360084, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": -0.0650108915156051, + "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": 0.045472364872694016, + "selected_utility_gain_over_base": -0.026442941278219223, + "selector_regret": 1.954527635127306, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.2012384152043573 + }, + "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": 0.5487264271441268, + "pairwise_causal_calibration_ece": 0.12372642714412677, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": -0.0006465763829108795, + "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": 0.08447941392660141, + "selected_utility_gain_over_base": 0.012564107775688171, + "selector_regret": 1.9155205860733986, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.2012384152043573 + }, + "1845489013/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.32608695652173914, + "pairwise_causal_calibration_confidence": 0.5077729807205135, + "pairwise_causal_calibration_ece": 0.18168602419877433, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": -0.22001414669968217, + "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": -0.2012384152043573 + }, + "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": 1.0, + "pairwise_causal_calibration_confidence": 0.5209647285753305, + "pairwise_causal_calibration_ece": 0.4790352714246695, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": -0.18822734864823312, + "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": -0.2012384152043573 + }, + "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": 0.44, + "pairwise_causal_calibration_confidence": 0.5128055424573941, + "pairwise_causal_calibration_ece": 0.07280554245739407, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": -0.1547908021155039, + "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": -0.2012384152043573 + }, + "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": 0.2571428571428571, + "pairwise_causal_calibration_confidence": 0.5358621206887306, + "pairwise_causal_calibration_ece": 0.2787192635458735, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.0012762962711883319, + "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": 0.13235082030966006 + }, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5297587479178145, + "pairwise_causal_calibration_ece": 0.041669823510756876, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": -0.08747088123670077, + "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": 0.13235082030966006 + }, + "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": 0.42105263157894735, + "pairwise_causal_calibration_confidence": 0.5267734463002417, + "pairwise_causal_calibration_ece": 0.10572081472129435, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": -0.10186397965031571, + "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": 0.13235082030966006 + }, + "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": 0.3949579831932773, + "pairwise_causal_calibration_confidence": 0.5282333871650522, + "pairwise_causal_calibration_ece": 0.1332754039717749, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.1490312481371141, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7026570439338684, + "selected_utility_gain_over_base": -0.020948290824890137, + "selector_regret": 1.2973429560661316, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.028235521236802175 + }, + "2012211287/train1": { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.575, + "pairwise_causal_calibration_confidence": 0.5289841828268461, + "pairwise_causal_calibration_ece": 0.04601581717315384, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.036105644427698036, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7236053347587585, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.2763946652412415, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.028235521236802175 + }, + "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": 0.6847826086956522, + "pairwise_causal_calibration_confidence": 0.531948622673026, + "pairwise_causal_calibration_ece": 0.1528339860226262, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.09375170906247592, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 14.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6934840679168701, + "selected_utility_gain_over_base": -0.030121266841888428, + "selector_regret": 1.3065159320831299, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.028235521236802175 + }, + "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": 0.5740754232851301, + "pairwise_causal_calibration_ece": 0.2982133543196128, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.19762473284895746, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04285776615142822, + "selected_utility_gain_over_base": -1.9571422338485718, + "selector_regret": 1.9571422338485718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.2012384152043573 + }, + "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": 0.2962962962962963, + "pairwise_causal_calibration_confidence": 0.5667318952152414, + "pairwise_causal_calibration_ece": 0.27043559891894486, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.3209231912001711, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07674666494131088, + "selected_utility_gain_over_base": -1.9232533350586891, + "selector_regret": 1.9232533350586891, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.2012384152043573 + }, + "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": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5635629372227472, + "pairwise_causal_calibration_ece": 0.1090174826772927, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.31779873234459755, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.12977690994739532, + "selected_utility_gain_over_base": -1.8702230900526047, + "selector_regret": 1.8702230900526047, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.2012384152043573 + }, + "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": 0.26666666666666666, + "pairwise_causal_calibration_confidence": 0.5290379603266726, + "pairwise_causal_calibration_ece": 0.13118564683000297, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 0.15266150175970736, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.077283151447773, + "selected_candidate_index": 4.5, + "selected_success": 0.5, + "selected_success_gain_over_base": -0.5, + "selected_utility": 1.0403175354003906, + "selected_utility_gain_over_base": -0.9596824645996094, + "selector_regret": 0.036965616047382355, + "success_selector_gap": 0.0, + "success_support_gap": 0.5, + "support_gap": 0.922716848552227, + "tau": -0.2012384152043573 + }, + "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": 0.6491228070175439, + "pairwise_causal_calibration_confidence": 0.5248824145830566, + "pairwise_causal_calibration_ece": 0.06212019621724363, + "pairwise_causal_calibration_pairs": 57.0, + "predicted_margin": 0.1715300859769803, + "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": -0.2012384152043573 + }, + "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": 0.5798319327731093, + "pairwise_causal_calibration_confidence": 0.5308674006096046, + "pairwise_causal_calibration_ece": 0.024482266081752357, + "pairwise_causal_calibration_pairs": 59.5, + "predicted_margin": 0.16346432037867953, + "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": -0.2012384152043573 + }, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5393032283831621, + "pairwise_causal_calibration_ece": 0.06094260950628667, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.12372212373443373, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09799930453300476, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04574880003929138, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9020006954669952, + "tau": 0.13235082030966006 + }, + "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": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5362308522692728, + "pairwise_causal_calibration_ece": 0.06400863004705062, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.049366979703384384, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07580336928367615, + "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": 0.02355286478996277, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9241966307163239, + "tau": 0.13235082030966006 + }, + "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": 0.5555555555555556, + "pairwise_causal_calibration_confidence": 0.5391540330154272, + "pairwise_causal_calibration_ece": 0.016401522540128385, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.015355053119340732, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11105132102966309, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.05880081653594971, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.888948678970337, + "tau": 0.13235082030966006 + }, + "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": 0.5882352941176471, + "pairwise_causal_calibration_confidence": 0.517501300906042, + "pairwise_causal_calibration_ece": 0.07073399321160512, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.04424689933210735, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "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": 0.13235082030966006 + }, + "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": 0.6583333333333333, + "pairwise_causal_calibration_confidence": 0.5160911960797463, + "pairwise_causal_calibration_ece": 0.14224213725358703, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.13077847063672782, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "selected_candidate_index": 4.0, + "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": 0.13235082030966006 + }, + "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": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5172535124316141, + "pairwise_causal_calibration_ece": 0.04941315423505255, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.04386283650419778, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04297435283660889, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7661172300577164, + "tau": 0.13235082030966006 + }, + "278008524/train0": { + "base_success": 0.0, + "base_utility": 0.042734989896416664, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5731060606060606, + "pairwise_causal_calibration_confidence": 0.5325709273453999, + "pairwise_causal_calibration_ece": 0.12646220200877553, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.2219557334710413, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0733563229441643, + "selected_candidate_index": 10.0, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.5, + "selected_utility": 1.0287093725055456, + "selected_utility_gain_over_base": 0.985974382609129, + "selector_regret": 0.04464695043861866, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.04055231809616089, + "tau": -0.2012384152043573 + }, + "278008524/train1": { + "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": 0.5248188405797101, + "pairwise_causal_calibration_confidence": 0.5322397110343211, + "pairwise_causal_calibration_ece": 0.10215544676823522, + "pairwise_causal_calibration_pairs": 106.0, + "predicted_margin": 0.2843962133295508, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0376745946705341, + "selected_candidate_index": 8.0, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.5, + "selected_utility": 1.0095977680757642, + "selected_utility_gain_over_base": 0.9668627781793475, + "selector_regret": 0.028076826594769955, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.07623404636979103, + "tau": -0.2012384152043573 + }, + "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": 0.5214646464646465, + "pairwise_causal_calibration_confidence": 0.536003409987186, + "pairwise_causal_calibration_ece": 0.10720913222134038, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.24145145365463994, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.1359489113092422, + "selected_candidate_index": 14.0, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.5, + "selected_utility": 1.0227726306766272, + "selected_utility_gain_over_base": 0.9800376407802105, + "selector_regret": 0.11317628063261509, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.2012384152043573 + }, + "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": 0.5494505494505495, + "pairwise_causal_calibration_confidence": 0.5260813668824347, + "pairwise_causal_calibration_ece": 0.036558379935610945, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.08783512114247624, + "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": 0.028235521236802175 + }, + "307068879/train1": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5218750448469349, + "pairwise_causal_calibration_ece": 0.049553526581636476, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": -0.13881394667704933, + "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": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "tau": 0.028235521236802175 + }, + "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": 0.4945054945054945, + "pairwise_causal_calibration_confidence": 0.5232228710238422, + "pairwise_causal_calibration_ece": 0.028717376518347715, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.14397102567158404, + "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": 0.028235521236802175 + }, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5444159724840614, + "pairwise_causal_calibration_ece": 0.04441597248406137, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 0.0942295054093702, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.13235082030966006 + }, + "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": 0.7692307692307693, + "pairwise_causal_calibration_confidence": 0.5435121854300012, + "pairwise_causal_calibration_ece": 0.22571858380076806, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": 0.029952749201257664, + "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": 0.13235082030966006 + }, + "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": 0.38095238095238093, + "pairwise_causal_calibration_confidence": 0.5454433115811214, + "pairwise_causal_calibration_ece": 0.1644909306287403, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.004657759279381279, + "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": 0.13235082030966006 + }, + "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": 0.5273358125858131, + "pairwise_causal_calibration_ece": 0.09335384258660073, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.5465320030087847, + "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": 0.07420081675594419 + }, + "440577711/train1": { + "base_success": 0.0, + "base_utility": 0.036152780055999756, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4, + "pairwise_causal_calibration_confidence": 0.5197720209730798, + "pairwise_causal_calibration_ece": 0.11977202097307982, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 0.629918599045802, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.07420081675594419 + }, + "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": 0.5240762582462818, + "pairwise_causal_calibration_ece": 0.061113295283318825, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.6241310075694073, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.07420081675594419 + }, + "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": 0.36363636363636365, + "pairwise_causal_calibration_confidence": 0.5267148410723766, + "pairwise_causal_calibration_ece": 0.163078477436013, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": 0.046645784157056724, + "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": 0.028235521236802175 + }, + "441046872/train1": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6909090909090909, + "pairwise_causal_calibration_confidence": 0.5234185829111347, + "pairwise_causal_calibration_ece": 0.16749050799795617, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.17358234568930467, + "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": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.023403704166412354, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "tau": 0.028235521236802175 + }, + "441046872/train2": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.45454545454545453, + "pairwise_causal_calibration_confidence": 0.5255994659745129, + "pairwise_causal_calibration_ece": 0.07105401142905837, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.022380468606574055, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7191348671913147, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019179701805114746, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "tau": 0.028235521236802175 + }, + "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": 0.6134453781512605, + "pairwise_causal_calibration_confidence": 0.5129229068912616, + "pairwise_causal_calibration_ece": 0.10052247125999891, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4269940390592024, + "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": 0.07420081675594419 + }, + "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": 0.5619047619047619, + "pairwise_causal_calibration_confidence": 0.5125583130541626, + "pairwise_causal_calibration_ece": 0.049346448850599356, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.37833433181770104, + "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": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.07420081675594419 + }, + "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": 0.5966386554621849, + "pairwise_causal_calibration_confidence": 0.5151111613972951, + "pairwise_causal_calibration_ece": 0.0815274940648898, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.282290647599035, + "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": 0.07420081675594419 + }, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5296207411344719, + "pairwise_causal_calibration_ece": 0.06533502684875758, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.10499618981526784, + "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": -0.2012384152043573 + }, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5273196708342051, + "pairwise_causal_calibration_ece": 0.07268032916579492, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.06533471439439086, + "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": -0.2012384152043573 + }, + "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": 0.7818181818181819, + "pairwise_causal_calibration_confidence": 0.529609307543077, + "pairwise_causal_calibration_ece": 0.25220887427510486, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.09915967492706912, + "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": -0.2012384152043573 + }, + "608016290/train0": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5619047619047619, + "pairwise_causal_calibration_confidence": 0.5561217648468477, + "pairwise_causal_calibration_ece": 0.03209834914225741, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.5011206164338899, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 11.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.028235521236802175 + }, + "608016290/train1": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.49523809523809526, + "pairwise_causal_calibration_confidence": 0.5555371517280593, + "pairwise_causal_calibration_ece": 0.06029905648996403, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.2910589829678913, + "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": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.028235521236802175 + }, + "608016290/train2": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5083333333333333, + "pairwise_causal_calibration_confidence": 0.5609759280259827, + "pairwise_causal_calibration_ece": 0.05264259469264974, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4536901796255097, + "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": 0.6789124011993408, + "selected_utility_gain_over_base": -0.04678922891616821, + "selector_regret": 0.03281521797180176, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "tau": 0.028235521236802175 + }, + "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": 0.6541666666666667, + "pairwise_causal_calibration_confidence": 0.5656883276900622, + "pairwise_causal_calibration_ece": 0.1169840028584393, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5216712480085811, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04469547560438514, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.009900043485686183, + "selected_utility_gain_over_base": 0.004801700764801353, + "selector_regret": 0.034795432118698955, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12161866808310151, + "tau": -0.2012384152043573 + }, + "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": 0.375, + "pairwise_causal_calibration_confidence": 0.5625549213649634, + "pairwise_causal_calibration_ece": 0.19272229094655288, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4816967914950128, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.035827874671667814, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.006450424436479807, + "selected_utility_gain_over_base": 0.0013520817155949771, + "selector_regret": 0.029377450235188007, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12975450651720166, + "tau": -0.2012384152043573 + }, + "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": 0.5636531135559604, + "pairwise_causal_calibration_ece": 0.10846025594242703, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.39041494710515146, + "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": -0.2012384152043573 + }, + "636680137/train0": { + "base_success": 0.0, + "base_utility": 0.08313504606485367, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.17056432366371155, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5190311348285646, + "pairwise_causal_calibration_ece": 0.10236446816189787, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4324165282764497, + "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": -0.2012384152043573 + }, + "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": 0.7, + "pairwise_causal_calibration_confidence": 0.5160986674159838, + "pairwise_causal_calibration_ece": 0.1839013325840162, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.0489088109005007, + "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": -0.2012384152043573 + }, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5129089770472632, + "pairwise_causal_calibration_ece": 0.012908977047263237, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.13337534819821673, + "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": -0.2012384152043573 + }, + "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": 0.5277777777777778, + "pairwise_causal_calibration_confidence": 0.5597351518736766, + "pairwise_causal_calibration_ece": 0.05616248877449882, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.07276876911827704, + "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": 0.13235082030966006 + }, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.562343415271088, + "pairwise_causal_calibration_ece": 0.11789897082664356, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.1554209538300596, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8128758072853088, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11176733672618866, + "selected_utility_gain_over_base": -0.051239341497421265, + "selector_regret": 0.7011084705591202, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.1871241927146912, + "tau": 0.13235082030966006 + }, + "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": 0.5555555555555556, + "pairwise_causal_calibration_confidence": 0.5588372328026256, + "pairwise_causal_calibration_ece": 0.027075654701186606, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.11907170730895571, + "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": 0.13235082030966006 + }, + "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": 0.8235294117647058, + "pairwise_causal_calibration_confidence": 0.5525242267379454, + "pairwise_causal_calibration_ece": 0.27100518502676063, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": -0.087075237393225, + "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": -0.2012384152043573 + }, + "672328262/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.45714285714285713, + "pairwise_causal_calibration_confidence": 0.5406287558110948, + "pairwise_causal_calibration_ece": 0.08348589866823782, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": -0.23567605508015027, + "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": -0.2012384152043573 + }, + "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": 0.4, + "pairwise_causal_calibration_confidence": 0.5442903896362881, + "pairwise_causal_calibration_ece": 0.1442903896362882, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": -0.14512362014342817, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04013325646519661, + "selected_utility_gain_over_base": -1.9598667435348034, + "selector_regret": 1.9598667435348034, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.2012384152043573 + }, + "701667486/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5238095238095238, + "pairwise_causal_calibration_confidence": 0.5307788011897017, + "pairwise_causal_calibration_ece": 0.00696927738017783, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 0.18126309408722283, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.13235082030966006 + }, + "701667486/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5952380952380952, + "pairwise_causal_calibration_confidence": 0.5325935443428202, + "pairwise_causal_calibration_ece": 0.06264455089527499, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.15700516615012938, + "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": 0.13235082030966006 + }, + "701667486/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5185185185185185, + "pairwise_causal_calibration_confidence": 0.5303820261796057, + "pairwise_causal_calibration_ece": 0.011863507661087236, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.1922641707596195, + "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": 0.13235082030966006 + }, + "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": 0.49166666666666664, + "pairwise_causal_calibration_confidence": 0.5222884856095076, + "pairwise_causal_calibration_ece": 0.03062181894284094, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.08999271197759756, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10839563608169556, + "selected_candidate_index": 14.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.061248183250427246, + "selected_utility_gain_over_base": -1.9387518167495728, + "selector_regret": 0.04714745283126831, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17026761174201965, + "tau": -0.2012384152043573 + }, + "747260437/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6833333333333333, + "pairwise_causal_calibration_confidence": 0.5252521749807111, + "pairwise_causal_calibration_ece": 0.15808115835262226, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.26165809689080527, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06328952312469482, + "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": 0.21537372469902039, + "tau": -0.2012384152043573 + }, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5187728818008999, + "pairwise_causal_calibration_ece": 0.08122711819910011, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.061796090652556146, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.12510187923908234, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.06438694149255753, + "selected_utility_gain_over_base": -1.9356130585074425, + "selector_regret": 0.06071493774652481, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15356136858463287, + "tau": -0.2012384152043573 + }, + "750448296/train0": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.6666666666666666, + "execute_generated": 0.6666666666666666, + "fallback_rate": 0.3333333333333333, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.5043650793650793, + "pairwise_causal_calibration_confidence": 0.5207629106352885, + "pairwise_causal_calibration_ece": 0.08615947226034647, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": 0.005890085195897528, + "proposal_oracle_success": 0.3333333333333333, + "proposal_oracle_utility": 0.7526603279014429, + "selected_candidate_index": 0.6666666666666666, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.125017994393905, + "selected_utility_gain_over_base": 0.007424845049778621, + "selector_regret": 0.6276423335075378, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.3333333333333333, + "support_gap": 0.6553361676633358, + "tau": -0.20123841520435728 + }, + "750448296/train1": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.533974358974359, + "pairwise_causal_calibration_confidence": 0.5196740117740427, + "pairwise_causal_calibration_ece": 0.13755775384340305, + "pairwise_causal_calibration_pairs": 101.66666666666667, + "predicted_margin": -0.19810309176890203, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3705032741030057, + "selected_candidate_index": 8.333333333333334, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.109618179500103, + "selected_utility_gain_over_base": -0.007974969844023386, + "selector_regret": 1.2608850946029027, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.03749322146177292, + "tau": -0.20123841520435728 + }, + "750448296/train2": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.6666666666666666, + "execute_generated": 0.6666666666666666, + "fallback_rate": 0.3333333333333333, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.5404761904761904, + "pairwise_causal_calibration_confidence": 0.5222504044176294, + "pairwise_causal_calibration_ece": 0.10634938783489285, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": -0.15048924424583973, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3504486680030823, + "selected_candidate_index": 10.333333333333334, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09164678926269214, + "selected_utility_gain_over_base": -0.02594636008143425, + "selector_regret": 1.25880187874039, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.05754782756169637, + "tau": -0.20123841520435728 + }, + "767747424/train0": { + "base_success": 0.25, + "base_utility": 0.5731993420049548, + "coverage": 0.75, + "execute_generated": 0.75, + "fallback_rate": 0.25, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.5482371794871794, + "pairwise_causal_calibration_confidence": 0.5286089316154619, + "pairwise_causal_calibration_ece": 0.1306918529485389, + "pairwise_causal_calibration_pairs": 119.25, + "predicted_margin": -0.028685105901352856, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0424291878007352, + "selected_candidate_index": 7.0, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5305264240596443, + "selected_utility_gain_over_base": -0.04267291794531047, + "selector_regret": 0.5119027637410909, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.06562284985557199, + "tau": -0.2012384152043573 + }, + "767747424/train1": { + "base_success": 0.25, + "base_utility": 0.5731993420049548, + "coverage": 0.75, + "execute_generated": 0.75, + "fallback_rate": 0.25, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.5888404079580549, + "pairwise_causal_calibration_confidence": 0.5262713091494071, + "pairwise_causal_calibration_ece": 0.13822538375258459, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.013317565754431063, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0629530912265182, + "selected_candidate_index": 7.0, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5966028785333037, + "selected_utility_gain_over_base": 0.023403536528348923, + "selector_regret": 0.4663502126932144, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.045098946429789066, + "tau": -0.2012384152043573 + }, + "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": 0.5254749335631689, + "pairwise_causal_calibration_confidence": 0.5306118103750065, + "pairwise_causal_calibration_ece": 0.13666304064151388, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.02448240692729431, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0671670949086547, + "selected_candidate_index": 5.5, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5695318393409252, + "selected_utility_gain_over_base": -0.0036675026640295982, + "selector_regret": 0.4976352555677295, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.04088494274765253, + "tau": -0.2012384152043573 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/metrics_by_task.json b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/metrics_by_task.json new file mode 100644 index 0000000000000000000000000000000000000000..3b3cccead676ca1682cc802928cca9b21a5d3d9c --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_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": 0.6666666666666666, + "execute_generated": 0.6666666666666666, + "fallback_rate": 0.3333333333333333, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.5333333333333333, + "pairwise_causal_calibration_accuracy": 0.5494360697557629, + "pairwise_causal_calibration_confidence": 0.5321483671612611, + "pairwise_causal_calibration_ece": 0.08779692862966537, + "pairwise_causal_calibration_pairs": 94.8, + "predicted_margin": 0.09934494198504117, + "proposal_oracle_success": 0.6, + "proposal_oracle_utility": 1.4122399787108104, + "selected_candidate_index": 7.2, + "selected_success": 0.26666666666666666, + "selected_success_gain_over_base": 0.06666666666666667, + "selected_utility": 0.9396337101856868, + "selected_utility_gain_over_base": -0.010236828525861105, + "selector_regret": 0.47343848447004955, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.4, + "support_gap": 0.5877600212891897, + "tau": 0.028235521236802175 + }, + "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": 0.3796296296296296, + "pairwise_causal_calibration_confidence": 0.5397879799473912, + "pairwise_causal_calibration_ece": 0.16015835031776168, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.2678354586918643, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13425676027933756, + "selected_candidate_index": 1.3333333333333333, + "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": 0.5025584143631503 + }, + "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": 0.5439315887395845, + "pairwise_causal_calibration_confidence": 0.5315713773974821, + "pairwise_causal_calibration_ece": 0.1202240575596694, + "pairwise_causal_calibration_pairs": 92.32, + "predicted_margin": 0.0587216273047277, + "proposal_oracle_success": 0.5333333333333333, + "proposal_oracle_utility": 1.1093241602430741, + "selected_candidate_index": 6.16, + "selected_success": 0.32, + "selected_success_gain_over_base": -0.04, + "selected_utility": 0.6822500004371007, + "selected_utility_gain_over_base": -0.09292046078636, + "selector_regret": 0.4528969661643108, + "success_selector_gap": 0.22666666666666666, + "success_support_gap": 0.02666666666666667, + "support_gap": 0.1075624164690574, + "tau": -0.2012384152043576 + }, + "PushCube-v1": { + "base_success": 0.16666666666666666, + "base_utility": 0.3924041787783305, + "coverage": 0.7777777777777778, + "execute_generated": 0.7777777777777778, + "fallback_rate": 0.2222222222222222, + "hidden_chart_oracle_success": 0.8333333333333334, + "hidden_chart_oracle_utility": 1.7015800600250561, + "outcome_ptr": 0.8333333333333334, + "pairwise_causal_calibration_accuracy": 0.5408758312375324, + "pairwise_causal_calibration_confidence": 0.5154259344288218, + "pairwise_causal_calibration_ece": 0.09497784791093052, + "pairwise_causal_calibration_pairs": 97.44444444444444, + "predicted_margin": 0.32846837901195985, + "proposal_oracle_success": 0.8888888888888888, + "proposal_oracle_utility": 1.7857814149724112, + "selected_candidate_index": 5.222222222222222, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.3333333333333333, + "selected_utility": 1.0611259018381436, + "selected_utility_gain_over_base": 0.6687217230598131, + "selector_regret": 0.7246555131342676, + "success_selector_gap": 0.3888888888888889, + "success_support_gap": 0.1111111111111111, + "support_gap": 0.21421858502758873, + "tau": 0.0742008167559442 + }, + "StackCube-v1": { + "base_success": 0.2727272727272727, + "base_utility": 0.7763947045261209, + "coverage": 0.21212121212121213, + "execute_generated": 0.21212121212121213, + "fallback_rate": 0.7878787878787878, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.6363636363636364, + "pairwise_causal_calibration_accuracy": 0.4871933998564077, + "pairwise_causal_calibration_confidence": 0.5389990977485094, + "pairwise_causal_calibration_ece": 0.10659066999978553, + "pairwise_causal_calibration_pairs": 67.45454545454545, + "predicted_margin": 0.03761618987646351, + "proposal_oracle_success": 0.5151515151515151, + "proposal_oracle_utility": 1.1567951707225856, + "selected_candidate_index": 5.606060606060606, + "selected_success": 0.30303030303030304, + "selected_success_gain_over_base": 0.030303030303030304, + "selected_utility": 0.769150389414845, + "selected_utility_gain_over_base": -0.007244315111275875, + "selector_regret": 0.3912377664537141, + "success_selector_gap": 0.21212121212121213, + "success_support_gap": 0.48484848484848486, + "support_gap": 0.8432048292774142, + "tau": 0.13235082030965997 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/split_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/split_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..865ad41e869c1bfbc57e4db65ed79e91e6d93081 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/table.tex b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/table.tex new file mode 100644 index 0000000000000000000000000000000000000000..24b4bc35514f25fc28bd05c4dd73d3398434dc78 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_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 & 0.6389 & 0.3611 & 0.2917 & 0.3264 & 0.5694 & 0.5486 & 0.2014 & 0.2500 & 0.0221 \\ +\bottomrule +\end{tabular} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/train.log b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_task_envclip_k16_train_to_test/train.log new file mode 100644 index 0000000000000000000000000000000000000000..34298d0999a299a6e4af72fc2eaa12f327267986 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_chartcompat_utility_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=100.0 +tau={"LiftPegUpright-v1": 0.028236, "PickCube-v1": 0.502558, "PullCube-v1": -0.201238, "PushCube-v1": 0.074201, "StackCube-v1": 0.132351, "__global__": 0.018529} +calibration_input=runs/ctt_base_context_obs_train_cal_envclip_k16_rollout_comparison/combined_measured_candidates.json diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/command.txt b/workspace/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/workspace/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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/config.yaml b/workspace/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/workspace/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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/data_hash.txt b/workspace/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/workspace/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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/eval.log b/workspace/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/workspace/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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/git_hash.txt b/workspace/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/workspace/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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics.json b/workspace/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/workspace/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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_seed.json b/workspace/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/workspace/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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/metrics_by_task.json b/workspace/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/workspace/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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/split_hash.txt b/workspace/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/workspace/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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/table.tex b/workspace/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/workspace/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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_successbonus2_task_envclip_k16_train_to_test/train.log b/workspace/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/workspace/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 diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/command.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/command.txt new file mode 100644 index 0000000000000000000000000000000000000000..532f1a4eef70cd0ae5376a6cf3780cf9cb08e0f0 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_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_utility_task_envclip_k16_train_to_test --feature-set score_context_chart_compat --selector-chart-feature-mode base_context_obs --target utility_margin --threshold-scope task --k 16 --bootstrap-samples 300 --no-markdown-report diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/config.yaml b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d8657cb93936588102fdb7624c250d1640c72920 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_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_utility_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: 1.0 +target: utility_margin +threshold_scope: task diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/data_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/data_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..e65c158c17e61aaf10422563d2ffdd8922735e6a --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/eval.log b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/eval.log new file mode 100644 index 0000000000000000000000000000000000000000..ad44bd37e51f82471f7a17d7372cee5fed00ed3f --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/git_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/git_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..293c6ad4e775f2f937dca574f1b83a5f93f0bdb1 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/git_hash.txt @@ -0,0 +1 @@ +14e81292e2004925f81e0bb4810f6f144bfd87b7 diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/metrics.json b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..7f15763653411c009cfc7594a4f7923ec007e1f5 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_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": 0.6365740740740741, + "fallback_rate": 0.36342592592592593, + "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": 0.36342592592592593, + "selected_success_gain_over_base": 0.08564814814814815, + "selected_utility": 0.8364669688942089, + "selector_regret": 0.39494272902048116, + "success_selector_gap": 0.2152777777777778, + "success_support_gap": 0.17592592592592593, + "support_gap": 0.35465237537627453 + }, + "calibration_summary": { + "base_success": 0.2777777777777778, + "base_utility": 0.7007981554626086, + "coverage": 0.6365740740740741, + "fallback_rate": 0.36342592592592593, + "hidden_chart_oracle_success": 0.6597222222222222, + "hidden_chart_oracle_utility": 1.4150809271054134, + "outcome_ptr": 0.5462962962962963, + "pairwise_causal_calibration_accuracy": 0.5172468271104324, + "pairwise_causal_calibration_confidence": 0.5316986569212452, + "pairwise_causal_calibration_ece": 0.014451829810813084, + "pairwise_causal_calibration_pairs": 37978.0, + "proposal_oracle_success": 0.5370370370370371, + "proposal_oracle_utility": 1.1440672337592803, + "selected_success": 0.36342592592592593, + "selected_success_gain_over_base": 0.08564814814814815, + "selected_utility": 0.8364669688942089, + "selector_regret": 0.39494272902048116, + "success_selector_gap": 0.2152777777777778, + "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": 0.6458333333333334, + "fallback_rate": 0.3541666666666667, + "hidden_chart_oracle_success": 0.7291666666666666, + "hidden_chart_oracle_utility": 1.5183161981403828, + "outcome_ptr": 0.5486111111111112, + "pairwise_causal_calibration_accuracy": 0.5299179668650474, + "pairwise_causal_calibration_confidence": 0.5299324847302089, + "pairwise_causal_calibration_ece": 0.01884675795915894, + "pairwise_causal_calibration_pairs": 12434.0, + "proposal_oracle_success": 0.5694444444444444, + "proposal_oracle_utility": 1.2159999172436073, + "selected_success": 0.3263888888888889, + "selected_success_gain_over_base": 0.034722222222222224, + "selected_utility": 0.7742405242590595, + "selector_regret": 0.46548765168538214, + "success_selector_gap": 0.25, + "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": 0.5172468271104324, + "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": 0.009888920147429925, + "accuracy": 0.5176925608953259, + "confidence": 0.5275814810427558, + "count": 36456, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.10613206470874337, + "accuracy": 0.5200276243093923, + "confidence": 0.6261596890181357, + "count": 1448, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.4684039184066699, + "accuracy": 0.24324324324324326, + "confidence": 0.7116471616499132, + "count": 74, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.8, + "upper": 0.9 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.9, + "upper": 1.0 + } + ], + "ece": 0.014451829810813084, + "mean_confidence": 0.5316986569212452, + "n_bins": 10, + "num_pairs": 37978, + "num_rows": 432 + }, + "eval": { + "accuracy": 0.5299179668650474, + "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": 0.009902735064229229, + "accuracy": 0.5350587837266345, + "confidence": 0.5251560486624053, + "count": 11823, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.19191579672635034, + "accuracy": 0.43044189852700493, + "confidence": 0.6223576952533553, + "count": 611, + "lower": 0.6, + "upper": 0.7 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.7, + "upper": 0.8 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.8, + "upper": 0.9 + }, + { + "abs_gap": 0.0, + "accuracy": 0.0, + "confidence": 0.0, + "count": 0, + "lower": 0.9, + "upper": 1.0 + } + ], + "ece": 0.01884675795915894, + "mean_confidence": 0.5299324847302089, + "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": 0.40336134453781514, + "pairwise_causal_calibration_confidence": 0.5298480396428488, + "pairwise_causal_calibration_ece": 0.12648669510503363, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.06399336728918298, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train0", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7026570439338684, + "selected_utility_gain_over_base": -0.020948290824890137, + "selector_regret": 1.2973429560661316, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": -0.08577718626706465, + "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": 0.4945054945054945, + "pairwise_causal_calibration_confidence": 0.5277478798612748, + "pairwise_causal_calibration_ece": 0.03324238535578037, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.008148511169388736, + "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": -0.08577718626706465, + "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": 0.5258276499617055, + "pairwise_causal_calibration_ece": 0.1128278122231684, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.02282134674243222, + "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": -0.08577718626706465, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5428571428571428, + "pairwise_causal_calibration_confidence": 0.5574158017414859, + "pairwise_causal_calibration_ece": 0.02535410150173202, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.40694674604777425, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/train0", + "selected_candidate_index": 11, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": -0.08577718626706465, + "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": 0.36363636363636365, + "pairwise_causal_calibration_confidence": 0.5291035564844788, + "pairwise_causal_calibration_ece": 0.16546719284811517, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.034318319001119896, + "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": -0.08577718626706465, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.53869991771853, + "pairwise_causal_calibration_ece": 0.038699917718529986, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.5860479678966253, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18325337767601013, + "seed": "1522272686/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8167466223239899, + "task_id": "PickCube-v1", + "tau": 0.4225255908772142, + "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": 0.015531805591637943, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/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": -0.1741162794419627, + "train_seed": "0" + }, + { + "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": 0.3418803418803419, + "pairwise_causal_calibration_confidence": 0.5087204156069876, + "pairwise_causal_calibration_ece": 0.16684007372664567, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": -0.1809542016196376, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train0", + "selected_candidate_index": 4, + "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": -0.1741162794419627, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.5097507053831263, + "pairwise_causal_calibration_ece": 0.17641737204979296, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.027181194046941026, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06019836664199829, + "seed": "1461069713/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.014822185039520264, + "selected_utility_gain_over_base": -0.07908087968826294, + "selector_regret": 0.04537618160247803, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0991440862417221, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.001929501653648913, + "chart_id": "ms-34726202b475d29cdf38", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.25342071056365967, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.725, + "pairwise_causal_calibration_confidence": 0.5744615236334294, + "pairwise_causal_calibration_ece": 0.20561875516069625, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.7230414029723224, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.010183374397456646, + "seed": "626486813/train0", + "selected_candidate_index": 12, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.007048348430544138, + "selected_utility_gain_over_base": 0.005118846776895225, + "selector_regret": 0.003135025966912508, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24323733616620302, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.5278373424628687, + "pairwise_causal_calibration_ece": 0.27783734246286873, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.011851042440313583, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15456630289554596, + "seed": "2054899126/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.08063507080078125, + "selected_utility_gain_over_base": -1.9193649291992188, + "selector_regret": 0.07393123209476471, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.845433697104454, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.8235294117647058, + "pairwise_causal_calibration_confidence": 0.5545676976677929, + "pairwise_causal_calibration_ece": 0.26896171409691294, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": -0.10240791384393264, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5541688916621307, + "pairwise_causal_calibration_ece": 0.08750222499546396, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.11409547735145165, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15051491558551788, + "seed": "767747424/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06666018813848495, + "selected_utility_gain_over_base": -0.010328307747840881, + "selector_regret": 0.08385472744703293, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08458571135997772, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.7, + "pairwise_causal_calibration_confidence": 0.5198578981073542, + "pairwise_causal_calibration_ece": 0.1801421018926458, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2974954997476227, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14671264588832855, + "seed": "278008524/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05741874501109123, + "selected_utility_gain_over_base": 0.021532438695430756, + "selector_regret": 0.08929390087723732, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08110463619232178, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.30434782608695654, + "pairwise_causal_calibration_confidence": 0.5082603489371776, + "pairwise_causal_calibration_ece": 0.2039125228502211, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": -0.2719633789755977, + "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": -0.1741162794419627, + "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": 0.5568570093495174, + "pairwise_causal_calibration_ece": 0.029393371256699805, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2790350112593491, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07920757681131363, + "seed": "626486813/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.012751738540828228, + "selected_utility_gain_over_base": 0.004484554752707481, + "selector_regret": 0.0664558382704854, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5153330648095253, + "pairwise_causal_calibration_ece": 0.1513336018571413, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.3351214852227477, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/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": -0.1741162794419627, + "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": 0.4880952380952381, + "pairwise_causal_calibration_confidence": 0.5333471334307192, + "pairwise_causal_calibration_ece": 0.045251895335481074, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.10907919344361995, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "585904130/train0", + "selected_candidate_index": 14, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8182844519615173, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.36666666666666664, + "pairwise_causal_calibration_confidence": 0.523727380049807, + "pairwise_causal_calibration_ece": 0.15706071338314037, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.09190846980596201, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "seed": "141984615/train0", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.006696522235870361, + "selected_utility_gain_over_base": -0.09613772481679916, + "selector_regret": 0.008481383323669434, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5276589138181056, + "pairwise_causal_calibration_ece": 0.0056744195152277355, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.025632048429654414, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21658658981323242, + "seed": "750448296/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21658658981323242, + "selected_utility_gain_over_base": 0.07387763261795044, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7834134101867676, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.38333333333333336, + "pairwise_causal_calibration_confidence": 0.5210271884357356, + "pairwise_causal_calibration_ece": 0.13769385510240223, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4860706638275267, + "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": -0.1741162794419627, + "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": 0.65, + "pairwise_causal_calibration_confidence": 0.5372285037967022, + "pairwise_causal_calibration_ece": 0.12615592265185224, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.25787960659551035, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.019201835617423058, + "seed": "767747424/train0", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.011628031730651855, + "selected_utility_gain_over_base": -0.04249491170048714, + "selector_regret": 0.007573803886771202, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17790568806231022, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.68, + "pairwise_causal_calibration_confidence": 0.529960870770317, + "pairwise_causal_calibration_ece": 0.1500391292296831, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": -0.3263045462730741, + "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": -0.1741162794419627, + "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": 0.3103448275862069, + "pairwise_causal_calibration_confidence": 0.5818783523078931, + "pairwise_causal_calibration_ece": 0.27153352472168624, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.23900074944006316, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04285776615142822, + "selected_utility_gain_over_base": -1.9571422338485718, + "selector_regret": 1.9571422338485718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.575, + "pairwise_causal_calibration_confidence": 0.5119355800805098, + "pairwise_causal_calibration_ece": 0.06306441991949019, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3441128312125427, + "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": -0.1741162794419627, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.34548062112785, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train0", + "selected_candidate_index": 8, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5223126728351409, + "pairwise_causal_calibration_ece": 0.022312672835140934, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.11408826824860728, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10839563608169556, + "seed": "747260437/train0", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.061248183250427246, + "selected_utility_gain_over_base": -1.9387518167495728, + "selector_regret": 0.04714745283126831, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17026761174201965, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.04958367347717285, + "chart_id": "ms-da555a03dd382d93e119", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.3838383838383838, + "pairwise_causal_calibration_confidence": 0.5403481857052789, + "pairwise_causal_calibration_ece": 0.15650980186689503, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.06152926325612136, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07767494767904282, + "selected_utility_gain_over_base": 0.028091274201869965, + "selector_regret": 1.9223250523209572, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.509674291684611, + "pairwise_causal_calibration_ece": 0.12365904164872232, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4651587653308018, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.060871422290802, + "seed": "1549324805/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0260902252048254, + "selected_utility_gain_over_base": -0.07842461206018925, + "selector_regret": 0.0347811970859766, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.23658466339111328, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.5517394604755897, + "pairwise_causal_calibration_ece": 0.08897015573763616, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 0.01957929235012683, + "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": -0.1741162794419627, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5271374770274253, + "pairwise_causal_calibration_ece": 0.11047081036075862, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.3270796693682031, + "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": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.5862068965517241, + "pairwise_causal_calibration_confidence": 0.529336326158639, + "pairwise_causal_calibration_ece": 0.056870570393085074, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.5294705452065069, + "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": 0.059256325887868444, + "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": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.5083549905395566, + "pairwise_causal_calibration_ece": 0.008311676127110146, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5904813494260328, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14406004548072815, + "seed": "1324642411/train0", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11045536398887634, + "selected_utility_gain_over_base": 0.0039848387241363525, + "selector_regret": 0.03360468149185181, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8559399545192719, + "task_id": "PushCube-v1", + "tau": 0.059256325887868444, + "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": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.507883924063129, + "pairwise_causal_calibration_ece": 0.12606574224494715, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.10954971541551203, + "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": 0.059256325887868444, + "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": 0.5167751391406238, + "pairwise_causal_calibration_ece": 0.052128674494159144, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": -0.12059592120185664, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train0", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.059256325887868444, + "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": 0.6302521008403361, + "pairwise_causal_calibration_confidence": 0.5148097853516517, + "pairwise_causal_calibration_ece": 0.11544231548868444, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4358418255784948, + "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": 0.059256325887868444, + "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": 0.8282828282828283, + "pairwise_causal_calibration_confidence": 0.5274298724184954, + "pairwise_causal_calibration_ece": 0.3008529558643329, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.8212684057863501, + "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": 0.059256325887868444, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "chart_id": "ms-146ff6c3f22b3a49af3a", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5414447509454381, + "pairwise_causal_calibration_ece": 0.06262112758251581, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.23896323450290802, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09799930453300476, + "seed": "2061635888/train0", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09799930453300476, + "selected_utility_gain_over_base": 0.04574880003929138, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9020006954669952, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.5277777777777778, + "pairwise_causal_calibration_confidence": 0.5624770111779376, + "pairwise_causal_calibration_ece": 0.03469923340015974, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.17094865469784903, + "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": 0.22913040029030945, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5395033820587356, + "pairwise_causal_calibration_ece": 0.07521766777302136, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.0030470281399136434, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10274861752986908, + "seed": "1113371261/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.897251382470131, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "chart_id": "ms-420280f15a522ce21c72", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7094017094017094, + "pairwise_causal_calibration_confidence": 0.5530254674859004, + "pairwise_causal_calibration_ece": 0.15637624191580896, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.35124403976358615, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train0", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.17586350440979, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.4111111111111111, + "pairwise_causal_calibration_confidence": 0.5409580694546181, + "pairwise_causal_calibration_ece": 0.12984695834350699, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.046710380447063826, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train0", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.5546218487394958, + "pairwise_causal_calibration_confidence": 0.518448754321454, + "pairwise_causal_calibration_ece": 0.03617309441804184, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.13866925649969028, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "211245842/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8090915828943253, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.5238095238095238, + "pairwise_causal_calibration_confidence": 0.531157925956513, + "pairwise_causal_calibration_ece": 0.007348402146989175, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.20840851240351038, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8663521409034729, + "seed": "1151410346/train0", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.6403508484363556, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.133647859096527, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.5618284314570119, + "pairwise_causal_calibration_ece": 0.13325700288558334, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.04291783793925408, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.054989203810691833, + "seed": "1324133122/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9450107961893082, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5476190476190477, + "pairwise_causal_calibration_confidence": 0.5311290004714053, + "pairwise_causal_calibration_ece": 0.01649004714764235, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 0.27106178902715317, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/train0", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.5475641766574334, + "pairwise_causal_calibration_ece": 0.0700132758404072, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 0.18513254256821268, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/train0", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5386338994647198, + "pairwise_causal_calibration_ece": 0.2529196137504342, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.1009604457083228, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train0", + "selected_candidate_index": 5, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "chart_id": "ms-5799584ed40e211c50a5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5310578566731103, + "pairwise_causal_calibration_ece": 0.03560880999355631, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.10501261504636256, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train1", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7236053347587585, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.2763946652412415, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": -0.08577718626706465, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "chart_id": "ms-715db95b13ebc6c977c8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5054945054945055, + "pairwise_causal_calibration_confidence": 0.5229087111015109, + "pairwise_causal_calibration_ece": 0.017414205607005462, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": -0.19456398757464405, + "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": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "task_id": "LiftPegUpright-v1", + "tau": -0.08577718626706465, + "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": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.5233895260511188, + "pairwise_causal_calibration_ece": 0.10994380728221453, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.1309941525333758, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/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": "LiftPegUpright-v1", + "tau": -0.08577718626706465, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.49523809523809526, + "pairwise_causal_calibration_confidence": 0.5568088448070179, + "pairwise_causal_calibration_ece": 0.06157074956892268, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.22198156017266713, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "608016290/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": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": -0.08577718626706465, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6363636363636364, + "pairwise_causal_calibration_confidence": 0.5234537989267526, + "pairwise_causal_calibration_ece": 0.11290983743688376, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.23082979751591187, + "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": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.023403704166412354, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "task_id": "LiftPegUpright-v1", + "tau": -0.08577718626706465, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.19444444444444445, + "pairwise_causal_calibration_confidence": 0.544614872631952, + "pairwise_causal_calibration_ece": 0.3501704281875075, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.3307920673952962, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07014383375644684, + "seed": "1522272686/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9298561662435532, + "task_id": "PickCube-v1", + "tau": 0.4225255908772142, + "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": 0.02856820772223631, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/train1", + "selected_candidate_index": 7, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.35294117647058826, + "pairwise_causal_calibration_confidence": 0.5091414959691674, + "pairwise_causal_calibration_ece": 0.1562003194985791, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.35576003727573224, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train1", + "selected_candidate_index": 11, + "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": -0.1741162794419627, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.425, + "pairwise_causal_calibration_confidence": 0.5100392790091248, + "pairwise_causal_calibration_ece": 0.08503927900912484, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.13909218118483407, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "seed": "1461069713/train1", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.006729066371917725, + "selected_utility_gain_over_base": -0.08717399835586548, + "selector_regret": 0.009855946525931358, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14275743998587132, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.325, + "pairwise_causal_calibration_confidence": 0.5702518593503015, + "pairwise_causal_calibration_ece": 0.255537139219144, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5693907727992329, + "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": -0.1741162794419627, + "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": 0.6842105263157895, + "pairwise_causal_calibration_confidence": 0.5259923842692189, + "pairwise_causal_calibration_ece": 0.1582181420465706, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.13618783086462727, + "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": -0.1741162794419627, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-62f8c18a15c566ec4b00", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.45714285714285713, + "pairwise_causal_calibration_confidence": 0.5403743645771392, + "pairwise_causal_calibration_ece": 0.0832315074342818, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": -0.23155797487397917, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train1", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5833333333333334, + "pairwise_causal_calibration_confidence": 0.5542021748100352, + "pairwise_causal_calibration_ece": 0.03420667383282964, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.15456429426010548, + "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": 0.18194365501403809, + "selected_utility_gain_over_base": 0.10495515912771225, + "selector_regret": 0.027086779475212097, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.026070192456245422, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.49166666666666664, + "pairwise_causal_calibration_confidence": 0.5154680354688722, + "pairwise_causal_calibration_ece": 0.02380136880220557, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4443203974084463, + "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": -0.1741162794419627, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 1.0, + "pairwise_causal_calibration_confidence": 0.5242226800008922, + "pairwise_causal_calibration_ece": 0.4757773199991078, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": -0.22208543687845692, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1845489013/train1", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.39166666666666666, + "pairwise_causal_calibration_confidence": 0.555425837235733, + "pairwise_causal_calibration_ece": 0.16375917056906675, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3517130031569937, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.061759788542985916, + "seed": "626486813/train1", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.010010083205997944, + "selected_utility_gain_over_base": 0.0017428994178771973, + "selector_regret": 0.05174970533698797, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.01598426327109337, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.7435897435897436, + "pairwise_causal_calibration_confidence": 0.5130916989229173, + "pairwise_causal_calibration_ece": 0.23049804466682633, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": -0.1576112307914954, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/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": -0.1741162794419627, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5285102521950974, + "pairwise_causal_calibration_ece": 0.07148974780490258, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.05173460799750952, + "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": -0.1741162794419627, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5258782846930987, + "pairwise_causal_calibration_ece": 0.07412171530690126, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.15467096667199068, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.037011127918958664, + "seed": "141984615/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.030119141563773155, + "selected_utility_gain_over_base": -0.07271510548889637, + "selector_regret": 0.006891986355185509, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24876614287495613, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.14270895719528198, + "chart_id": "ms-972c5be72d0b24be7cd5", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7583333333333333, + "pairwise_causal_calibration_confidence": 0.5243992772435688, + "pairwise_causal_calibration_ece": 0.23393405608976447, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.31758435202283863, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 15, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14270895719528198, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.857291042804718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.675, + "pairwise_causal_calibration_confidence": 0.5170756938619067, + "pairwise_causal_calibration_ece": 0.15792430613809338, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.11583401883298983, + "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": -0.1741162794419627, + "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": 0.6833333333333333, + "pairwise_causal_calibration_confidence": 0.5340903255555935, + "pairwise_causal_calibration_ece": 0.14924300777773983, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.34311330403908313, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.042781930416822433, + "seed": "767747424/train1", + "selected_candidate_index": 10, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.042781930416822433, + "selected_utility_gain_over_base": -0.011341013014316559, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15432559326291084, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.5259219860596367, + "pairwise_causal_calibration_ece": 0.00039380341404748886, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.8215150285449934, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/train1", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.2962962962962963, + "pairwise_causal_calibration_confidence": 0.5700204949171992, + "pairwise_causal_calibration_ece": 0.27372419862090286, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.3330720074528296, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train1", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07674666494131088, + "selected_utility_gain_over_base": -1.9232533350586891, + "selector_regret": 1.9232533350586891, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.4666666666666667, + "pairwise_causal_calibration_confidence": 0.5115971455700387, + "pairwise_causal_calibration_ece": 0.044930478903371984, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.10549043137976366, + "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": -0.1741162794419627, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.5175746414403108, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/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": -0.1741162794419627, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8fc1568ce66c45c8d8b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.5248794662055539, + "pairwise_causal_calibration_ece": 0.10845386712777938, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.27266683578349143, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06328952312469482, + "seed": "747260437/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": 0.21537372469902039, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "1" + }, + { + "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": 0.6195652173913043, + "pairwise_causal_calibration_confidence": 0.5454064824752691, + "pairwise_causal_calibration_ece": 0.0741587349160352, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.06103172307010252, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train1", + "selected_candidate_index": 8, + "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": -0.1741162794419627, + "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": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5050545552431692, + "pairwise_causal_calibration_ece": 0.028278778090164103, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.17043131073244489, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "seed": "1549324805/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0303483996540308, + "selected_utility_gain_over_base": -0.07416643761098385, + "selector_regret": 0.16857311688363552, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.5595321641601091, + "pairwise_causal_calibration_ece": 0.09352906032968679, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": -0.03627695714246994, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train1", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.045472364872694016, + "selected_utility_gain_over_base": -0.026442941278219223, + "selector_regret": 1.954527635127306, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.46153846153846156, + "pairwise_causal_calibration_confidence": 0.5259641242479297, + "pairwise_causal_calibration_ece": 0.06442566270946815, + "pairwise_causal_calibration_pairs": 65.0, + "predicted_margin": -0.3811137243373446, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "750448296/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.26666666666666666, + "pairwise_causal_calibration_confidence": 0.5208750509234744, + "pairwise_causal_calibration_ece": 0.25420838425680775, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 0.6282525745787317, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train1", + "selected_candidate_index": 10, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.059256325887868444, + "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": 0.5074297248847603, + "pairwise_causal_calibration_ece": 0.19845262805641617, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.29159110923611015, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11658626794815063, + "selected_utility_gain_over_base": 0.010115742683410645, + "selector_regret": 1.8834137320518494, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.059256325887868444, + "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": 0.37719298245614036, + "pairwise_causal_calibration_confidence": 0.506702038669967, + "pairwise_causal_calibration_ece": 0.12950905621382663, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 0.1144364826187316, + "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": 0.059256325887868444, + "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": 0.5197662577584645, + "pairwise_causal_calibration_ece": 0.010675348667555484, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": -0.11462503285369541, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train1", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.059256325887868444, + "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": 0.5619047619047619, + "pairwise_causal_calibration_confidence": 0.5134314316187146, + "pairwise_causal_calibration_ece": 0.04847333028604728, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.3747876596460843, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/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": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.059256325887868444, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5168654047616045, + "pairwise_causal_calibration_ece": 0.05456316666696692, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5782627775556237, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.1545013189315796, + "selected_utility_gain_over_base": 0.06746724247932434, + "selector_regret": 1.8454986810684204, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.059256325887868444, + "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": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5381973885533406, + "pairwise_causal_calibration_ece": 0.06597516633111851, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.18138954354814413, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07580336928367615, + "seed": "2061635888/train1", + "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": 0.02355286478996277, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9241966307163239, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.5653526989123352, + "pairwise_causal_calibration_ece": 0.1209082544678908, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.2625572115443, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8128758072853088, + "seed": "655884605/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11176733672618866, + "selected_utility_gain_over_base": -0.051239341497421265, + "selector_regret": 0.7011084705591202, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.1871241927146912, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.54070868503236, + "pairwise_causal_calibration_ece": 0.06643417211049711, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.02463520530320129, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18093222379684448, + "seed": "1113371261/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02250880002975464, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8190677762031555, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.44545454545454544, + "pairwise_causal_calibration_confidence": 0.5470265023854355, + "pairwise_causal_calibration_ece": 0.10157195693088991, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.29218243151021844, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12732218205928802, + "selected_utility_gain_over_base": -0.6968143135309219, + "selector_regret": 1.872677817940712, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.31746031746031744, + "pairwise_causal_calibration_confidence": 0.538847046384772, + "pairwise_causal_calibration_ece": 0.22138672892445443, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": -0.03542479330582579, + "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": 0.22913040029030945, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.625, + "pairwise_causal_calibration_confidence": 0.5162865066172287, + "pairwise_causal_calibration_ece": 0.10871349338277125, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.25579648935341137, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "seed": "211245842/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19948488473892212, + "selected_utility_gain_over_base": 0.008576467633247375, + "selector_regret": 0.013472288846969604, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7870428264141083, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.45714285714285713, + "pairwise_causal_calibration_confidence": 0.5307398847955603, + "pairwise_causal_calibration_ece": 0.0735970276527032, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.22134577714486947, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1151410346/train1", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.7739987075328827, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "chart_id": "ms-9dd1ff4c297a81b13071", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5611245570508201, + "pairwise_causal_calibration_ece": 0.09683884276510585, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.23171951530495727, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1324586570262909, + "seed": "1324133122/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07714252173900604, + "selected_utility_gain_over_base": -0.015062391757965088, + "selector_regret": 0.05531613528728485, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.867541342973709, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5333844210863801, + "pairwise_causal_calibration_ece": 0.07375843605647703, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.2505788547291904, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/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": 0.22913040029030945, + "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": 0.7692307692307693, + "pairwise_causal_calibration_confidence": 0.546479468539056, + "pairwise_causal_calibration_ece": 0.22275130069171328, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": 0.14951478428388437, + "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": 0.22913040029030945, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5322044449316384, + "pairwise_causal_calibration_ece": 0.06779555506836145, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.03337865189847511, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train1", + "selected_candidate_index": 8, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.6739130434782609, + "pairwise_causal_calibration_confidence": 0.533819076214467, + "pairwise_causal_calibration_ece": 0.14009396726379386, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.01094329025435617, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train2", + "selected_candidate_index": 14, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6934840679168701, + "selected_utility_gain_over_base": -0.030121266841888428, + "selector_regret": 1.3065159320831299, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": -0.08577718626706465, + "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": 0.4725274725274725, + "pairwise_causal_calibration_confidence": 0.5244461582060973, + "pairwise_causal_calibration_ece": 0.05191868567862479, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.0840500728219218, + "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": -0.08577718626706465, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5263938706546207, + "pairwise_causal_calibration_ece": 0.1402727960120459, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.01090870267033292, + "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": -0.08577718626706465, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "chart_id": "ms-99f263cb3f383b1f6243", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.56209932221557, + "pairwise_causal_calibration_ece": 0.0620993222155701, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3766943789955354, + "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": 0.6789124011993408, + "selected_utility_gain_over_base": -0.04678922891616821, + "selector_regret": 0.03281521797180176, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "task_id": "LiftPegUpright-v1", + "tau": -0.08577718626706465, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "chart_id": "ms-ec2e4afb34b439731136", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.43636363636363634, + "pairwise_causal_calibration_confidence": 0.5275776417866733, + "pairwise_causal_calibration_ece": 0.09121400542303698, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.08843117974127858, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7191348671913147, + "seed": "441046872/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019179701805114746, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "task_id": "LiftPegUpright-v1", + "tau": -0.08577718626706465, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5443083211760574, + "pairwise_causal_calibration_ece": 0.12764165450939086, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.36394636603715164, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14937306940555573, + "seed": "1522272686/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8506269305944443, + "task_id": "PickCube-v1", + "tau": 0.4225255908772142, + "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": 0.042189669044126965, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/train2", + "selected_candidate_index": 9, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.16168592870235443, + "chart_id": "ms-0e1ff8d16372796dac97", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6638655462184874, + "pairwise_causal_calibration_confidence": 0.5130436609619252, + "pairwise_causal_calibration_ece": 0.15082188525656215, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.2769216856539635, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "selected_candidate_index": 9, + "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": -0.1741162794419627, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "chart_id": "ms-223b0a8319beb051a1ee", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5916666666666667, + "pairwise_causal_calibration_confidence": 0.5122176002496256, + "pairwise_causal_calibration_ece": 0.07944906641704108, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.08608232017406181, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "seed": "1461069713/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09997648000717163, + "selected_utility_gain_over_base": 0.006073415279388428, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05936597287654877, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.7416666666666667, + "pairwise_causal_calibration_confidence": 0.5690517638677979, + "pairwise_causal_calibration_ece": 0.1726149027988688, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4771141058809072, + "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": -0.1741162794419627, + "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": 0.5296355479804102, + "pairwise_causal_calibration_ece": 0.033389662103623374, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.1657280539424397, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "selected_candidate_index": 14, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.43333333333333335, + "pairwise_causal_calibration_confidence": 0.5452442045537841, + "pairwise_causal_calibration_ece": 0.1119108712204508, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": -0.166628216752181, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "672328262/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04013325646519661, + "selected_utility_gain_over_base": -1.9598667435348034, + "selector_regret": 1.9598667435348034, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.07698849588632584, + "chart_id": "ms-6661b157c41ad4d6ab85", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2351006269454956, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.25833333333333336, + "pairwise_causal_calibration_confidence": 0.5588555004887193, + "pairwise_causal_calibration_ece": 0.3005221671553856, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.09795945728062418, + "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": 0.0730469822883606, + "selected_utility_gain_over_base": -0.0039415135979652405, + "selector_regret": 0.14581434428691864, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.01623930037021637, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.4083333333333333, + "pairwise_causal_calibration_confidence": 0.5244708097774345, + "pairwise_causal_calibration_ece": 0.11613747644410122, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.2556069316285349, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2718978226184845, + "seed": "278008524/train2", + "selected_candidate_index": 13, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04554526135325432, + "selected_utility_gain_over_base": 0.009658955037593842, + "selector_regret": 0.22635256126523018, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7923b8196b397863cf4f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4, + "pairwise_causal_calibration_confidence": 0.5133463004507663, + "pairwise_causal_calibration_ece": 0.11334630045076632, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": -0.19739643651997574, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1845489013/train2", + "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": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.55, + "pairwise_causal_calibration_confidence": 0.5559887220247433, + "pairwise_causal_calibration_ece": 0.03483139047849486, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.22569787155333226, + "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": -0.1741162794419627, + "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": 0.48717948717948717, + "pairwise_causal_calibration_confidence": 0.5104724908508356, + "pairwise_causal_calibration_ece": 0.0232930036713484, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": -0.25313070764934636, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "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": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.7727272727272727, + "pairwise_causal_calibration_confidence": 0.5338099460898557, + "pairwise_causal_calibration_ece": 0.23891732663741705, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.10442126551685532, + "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": -0.1741162794419627, + "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": 0.525, + "pairwise_causal_calibration_confidence": 0.5265869328502524, + "pairwise_causal_calibration_ece": 0.0015869328502523539, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.07577538797561234, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058951716870069504, + "seed": "141984615/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.02716982364654541, + "selected_utility_gain_over_base": -0.07566442340612411, + "selector_regret": 0.031781893223524094, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.7083333333333334, + "pairwise_causal_calibration_confidence": 0.5336259935852464, + "pairwise_causal_calibration_ece": 0.17470733974808694, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.01930870437379159, + "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": -0.1741162794419627, + "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": 0.48333333333333334, + "pairwise_causal_calibration_confidence": 0.5151375785964762, + "pairwise_causal_calibration_ece": 0.031804245263142905, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.18578510943339843, + "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": -0.1741162794419627, + "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": 0.6166666666666667, + "pairwise_causal_calibration_confidence": 0.5399665318110574, + "pairwise_causal_calibration_ece": 0.13805641745564656, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4780975757936197, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04980705305933952, + "seed": "767747424/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.035472434014081955, + "selected_utility_gain_over_base": -0.018650509417057037, + "selector_regret": 0.014334619045257568, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14730047062039375, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.8095238095238095, + "pairwise_causal_calibration_confidence": 0.5355938240012034, + "pairwise_causal_calibration_ece": 0.27392998552260617, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.6500022540507773, + "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": -0.1741162794419627, + "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": 0.43434343434343436, + "pairwise_causal_calibration_confidence": 0.5648483849875251, + "pairwise_causal_calibration_ece": 0.1305049506440908, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.3386689344106315, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2038594191/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.12977690994739532, + "selected_utility_gain_over_base": -1.8702230900526047, + "selector_regret": 1.8702230900526047, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5077578387506657, + "pairwise_causal_calibration_ece": 0.05890882791600094, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.112820669009982, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.051346004009246826, + "seed": "750448296/train2", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.035547394305467606, + "selected_utility_gain_over_base": -0.05745009705424309, + "selector_regret": 0.01579860970377922, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1726434826850891, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": NaN, + "pairwise_causal_calibration_confidence": NaN, + "pairwise_causal_calibration_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": 0.5158423609204148, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/train2", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.5916666666666667, + "pairwise_causal_calibration_confidence": 0.5203117574231598, + "pairwise_causal_calibration_ece": 0.07135490924350685, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.08183742266671763, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.12510187923908234, + "seed": "747260437/train2", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.06438694149255753, + "selected_utility_gain_over_base": -1.9356130585074425, + "selector_regret": 0.06071493774652481, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15356136858463287, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.5353535353535354, + "pairwise_causal_calibration_confidence": 0.5438651040720597, + "pairwise_causal_calibration_ece": 0.011448236732114864, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.2207670754919378, + "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": -0.1741162794419627, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5086088218670698, + "pairwise_causal_calibration_ece": 0.09139117813293018, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3864742495974729, + "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": -0.1741162794419627, + "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": 0.5509339900966759, + "pairwise_causal_calibration_ece": 0.12593399009667586, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 0.00519928677928394, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1822210471/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08447941392660141, + "selected_utility_gain_over_base": 0.012564107775688171, + "selector_regret": 1.9155205860733986, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.11707299947738647, + "chart_id": "ms-fa8b4ba41ecac8030271", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5308327709157353, + "pairwise_causal_calibration_ece": 0.13797562805859248, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.5897469498805397, + "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": 0.11707299947738647, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.8829270005226135, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.1741162794419627, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.525156875392907, + "pairwise_causal_calibration_ece": 0.08071243094846259, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.599974637285598, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train2", + "selected_candidate_index": 13, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.059256325887868444, + "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": 0.5272727272727272, + "pairwise_causal_calibration_confidence": 0.5073524497785574, + "pairwise_causal_calibration_ece": 0.019920277494169802, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.056361885016702556, + "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": 0.059256325887868444, + "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": 0.35454545454545455, + "pairwise_causal_calibration_confidence": 0.5080851163654574, + "pairwise_causal_calibration_ece": 0.15353966182000284, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.08944061124079167, + "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": 0.059256325887868444, + "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": 0.6190476190476191, + "pairwise_causal_calibration_confidence": 0.5178085323750468, + "pairwise_causal_calibration_ece": 0.10123908667257231, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.046092622673385486, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1261930456/train2", + "selected_candidate_index": 4, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.059256325887868444, + "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": 0.5630252100840336, + "pairwise_causal_calibration_confidence": 0.5164530817228409, + "pairwise_causal_calibration_ece": 0.04657212836119273, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.2716524269409146, + "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": 0.059256325887868444, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "chart_id": "ms-d4c37ed44fb338c1d551", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5630252100840336, + "pairwise_causal_calibration_confidence": 0.517349995500313, + "pairwise_causal_calibration_ece": 0.04567521458372059, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4155042382010001, + "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": 0.21894000470638275, + "selected_utility_gain_over_base": 0.1319059282541275, + "selector_regret": 1.7810599952936172, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": 0.059256325887868444, + "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": 0.5833333333333334, + "pairwise_causal_calibration_confidence": 0.5409068778231741, + "pairwise_causal_calibration_ece": 0.05426508199292427, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.14213746526284532, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11105132102966309, + "seed": "2061635888/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.05880081653594971, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.888948678970337, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "chart_id": "ms-15fc8cc6f74cf589056f", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5555555555555556, + "pairwise_causal_calibration_confidence": 0.56240741403742, + "pairwise_causal_calibration_ece": 0.03641674663594687, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.23713238846053236, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21447330713272095, + "seed": "655884605/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09131735563278198, + "selected_utility_gain_over_base": -0.07168932259082794, + "selector_regret": 0.12315595149993896, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.785526692867279, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5357178921702087, + "pairwise_causal_calibration_ece": 0.03571067925836269, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.07970004189125712, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "seed": "1113371261/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.36134453781512604, + "pairwise_causal_calibration_confidence": 0.5477244061705656, + "pairwise_causal_calibration_ece": 0.18637986835543957, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.26493380406470457, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1578990937/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15726424753665924, + "selected_utility_gain_over_base": -0.6668722480535507, + "selector_regret": 1.8427357524633408, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.4888888888888889, + "pairwise_causal_calibration_confidence": 0.5413409105541477, + "pairwise_causal_calibration_ece": 0.05245202166525893, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.08315965672063733, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train2", + "selected_candidate_index": 9, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "chart_id": "ms-55e78c04643bd06fd5f8", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.575, + "pairwise_causal_calibration_confidence": 0.5180070692001414, + "pairwise_causal_calibration_ece": 0.056992930799858565, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.15934077937193158, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "seed": "211245842/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04297435283660889, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7661172300577164, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.3523809523809524, + "pairwise_causal_calibration_confidence": 0.5321764777196925, + "pairwise_causal_calibration_ece": 0.17979552533874008, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.2005950618303269, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "seed": "1151410346/train2", + "selected_candidate_index": 8, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.32620343565940857, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4477952718734741, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.21428571428571427, + "pairwise_causal_calibration_confidence": 0.5603240341069567, + "pairwise_causal_calibration_ece": 0.3460383198212425, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.08474741252260032, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19291165471076965, + "seed": "1324133122/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.10070674121379852, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8070883452892303, + "task_id": "StackCube-v1", + "tau": 0.22913040029030945, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-d7254aa722729aa1a031", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5185185185185185, + "pairwise_causal_calibration_confidence": 0.5311189080220702, + "pairwise_causal_calibration_ece": 0.01260038950355169, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.31082243661779413, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/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": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.38095238095238093, + "pairwise_causal_calibration_confidence": 0.5484312023799853, + "pairwise_causal_calibration_ece": 0.16747882142760429, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.1263043297940751, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/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": "StackCube-v1", + "tau": 0.22913040029030945, + "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": 0.40789473684210525, + "pairwise_causal_calibration_confidence": 0.5287541242780439, + "pairwise_causal_calibration_ece": 0.12085938743593866, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": 0.027745817665301403, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/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": "StackCube-v1", + "tau": 0.22913040029030945, + "train_seed": "2" + } + ], + "schema_version": 1, + "selected_lambda": 100.0, + "selector_calibration_target_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "selector_calibration_target_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "selector_chart_feature_mode": "base_context_obs", + "selector_eval_target_content_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "selector_eval_target_split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "selector_source_content_hash": "67f7b4a692f7c7e71da24378dc71e3399c5d35afc14c494869eba6087b765c42", + "selector_source_split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372", + "source_content_hash": null, + "source_index": null, + "source_split_hash": null, + "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "success_bonus": 1.0, + "summary": { + "base_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.38958333333333334, + "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": 0.7041666666666667, + "low": 0.5479166666666667, + "mean": 0.6284722222222222, + "n": 120, + "num_samples": 300 + }, + "mean": 0.6284722222222222, + "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": 1.0, + "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": 1.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 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": 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": 0.0, + "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": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 1.0, + "211245842/train2": 0.0, + "278008524/train0": 1.0, + "278008524/train1": 1.0, + "278008524/train2": 1.0, + "307068879/train0": 1.0, + "307068879/train1": 0.0, + "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": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 1.0, + "636680137/train0": 1.0, + "636680137/train1": 1.0, + "636680137/train2": 1.0, + "655884605/train0": 0.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 1.0, + "672328262/train1": 0.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 1.0, + "747260437/train1": 0.0, + "747260437/train2": 1.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.5, + "767747424/train1": 0.75, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7546666666666667, + "low": 0.22133333333333333, + "mean": 0.5102222222222222, + "n": 5, + "num_samples": 300 + }, + "mean": 0.5102222222222222, + "per_task": { + "LiftPegUpright-v1": 0.6666666666666666, + "PickCube-v1": 0.0, + "PullCube-v1": 0.7733333333333333, + "PushCube-v1": 0.7777777777777778, + "StackCube-v1": 0.3333333333333333 + } + }, + "metric": "coverage", + "micro": { + "confidence": 0.95, + "high": 0.7222222222222222, + "low": 0.5763888888888888, + "mean": 0.6458333333333334, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "execute_generated": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.7041666666666667, + "low": 0.5479166666666667, + "mean": 0.6284722222222222, + "n": 120, + "num_samples": 300 + }, + "mean": 0.6284722222222222, + "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": 1.0, + "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": 1.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 1.0, + "1578990937/train1": 1.0, + "1578990937/train2": 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": 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": 0.0, + "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": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 1.0, + "211245842/train2": 0.0, + "278008524/train0": 1.0, + "278008524/train1": 1.0, + "278008524/train2": 1.0, + "307068879/train0": 1.0, + "307068879/train1": 0.0, + "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": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 1.0, + "636680137/train0": 1.0, + "636680137/train1": 1.0, + "636680137/train2": 1.0, + "655884605/train0": 0.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 1.0, + "672328262/train1": 0.0, + "672328262/train2": 1.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 1.0, + "747260437/train1": 0.0, + "747260437/train2": 1.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.5, + "767747424/train1": 0.75, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7546666666666667, + "low": 0.22133333333333333, + "mean": 0.5102222222222222, + "n": 5, + "num_samples": 300 + }, + "mean": 0.5102222222222222, + "per_task": { + "LiftPegUpright-v1": 0.6666666666666666, + "PickCube-v1": 0.0, + "PullCube-v1": 0.7733333333333333, + "PushCube-v1": 0.7777777777777778, + "StackCube-v1": 0.3333333333333333 + } + }, + "metric": "execute_generated", + "micro": { + "confidence": 0.95, + "high": 0.7222222222222222, + "low": 0.5763888888888888, + "mean": 0.6458333333333334, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "fallback_rate": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.45208333333333334, + "low": 0.29583333333333334, + "mean": 0.3715277777777778, + "n": 120, + "num_samples": 300 + }, + "mean": 0.3715277777777778, + "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": 0.0, + "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": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 1.0, + "1522272686/train1": 1.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "1578990937/train2": 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": 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": 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": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 1.0, + "211245842/train1": 0.0, + "211245842/train2": 1.0, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 1.0, + "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": 1.0, + "441046872/train2": 1.0, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.0, + "626486813/train0": 0.0, + "626486813/train1": 0.0, + "626486813/train2": 0.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.0, + "655884605/train0": 1.0, + "655884605/train1": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 0.0, + "672328262/train1": 1.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 1.0, + "747260437/train2": 0.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.3333333333333333, + "767747424/train0": 0.5, + "767747424/train1": 0.25, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7786666666666666, + "low": 0.24533333333333332, + "mean": 0.4897777777777777, + "n": 5, + "num_samples": 300 + }, + "mean": 0.4897777777777777, + "per_task": { + "LiftPegUpright-v1": 0.3333333333333333, + "PickCube-v1": 1.0, + "PullCube-v1": 0.22666666666666666, + "PushCube-v1": 0.2222222222222222, + "StackCube-v1": 0.6666666666666666 + } + }, + "metric": "fallback_rate", + "micro": { + "confidence": 0.95, + "high": 0.4236111111111111, + "low": 0.2777777777777778, + "mean": 0.3541666666666667, + "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": 0.5461415482273919, + "low": 0.49457833483332414, + "mean": 0.5193293830155317, + "n": 117, + "num_samples": 300 + }, + "mean": 0.5193293830155317, + "per_seed": { + "1073161778/train0": 0.6386554621848739, + "1073161778/train1": 0.6333333333333333, + "1073161778/train2": 0.6666666666666666, + "1113371261/train0": 0.4642857142857143, + "1113371261/train1": 0.6071428571428571, + "1113371261/train2": 0.5714285714285714, + "1151410346/train0": 0.5238095238095238, + "1151410346/train1": 0.45714285714285713, + "1151410346/train2": 0.3523809523809524, + "1261930456/train0": 0.46464646464646464, + "1261930456/train1": 0.509090909090909, + "1261930456/train2": 0.6190476190476191, + "1324133122/train0": 0.42857142857142855, + "1324133122/train1": 0.4642857142857143, + "1324133122/train2": 0.21428571428571427, + "1324642411/train0": 0.5166666666666667, + "1324642411/train1": 0.7058823529411765, + "1324642411/train2": 0.5272727272727272, + "1375492786/train0": 0.68, + "1375492786/train1": 0.5263157894736842, + "1375492786/train2": 0.8095238095238095, + "141984615/train0": 0.36666666666666664, + "141984615/train1": 0.6, + "141984615/train2": 0.525, + "1461069713/train0": 0.3333333333333333, + "1461069713/train1": 0.425, + "1461069713/train2": 0.5916666666666667, + "1480856810/train0": 0.8282828282828283, + "1480856810/train1": 0.5714285714285714, + "1480856810/train2": 0.5630252100840336, + "1509980664/train0": 0.4111111111111111, + "1509980664/train1": 0.31746031746031744, + "1509980664/train2": 0.4888888888888889, + "1522272686/train0": 0.5, + "1522272686/train1": 0.19444444444444445, + "1522272686/train2": 0.4166666666666667, + "1549324805/train0": 0.6333333333333333, + "1549324805/train1": 0.5333333333333333, + "1549324805/train2": 0.6, + "1578990937/train0": 0.7094017094017094, + "1578990937/train1": 0.44545454545454544, + "1578990937/train2": 0.36134453781512604, + "1790188787/train0": 0.38181818181818183, + "1790188787/train1": 0.37719298245614036, + "1790188787/train2": 0.35454545454545455, + "1822210471/train0": 0.6, + "1822210471/train1": 0.6530612244897959, + "1822210471/train2": 0.425, + "1845489013/train0": 0.30434782608695654, + "1845489013/train1": 1.0, + "1845489013/train2": 0.4, + "1999022574/train0": 0.2857142857142857, + "1999022574/train1": 0.6, + "1999022574/train2": 0.40789473684210525, + "2012211287/train0": 0.40336134453781514, + "2012211287/train1": 0.5666666666666667, + "2012211287/train2": 0.6739130434782609, + "2038594191/train0": 0.3103448275862069, + "2038594191/train1": 0.2962962962962963, + "2038594191/train2": 0.43434343434343436, + "2054899126/train0": 0.25, + "2054899126/train1": 0.6842105263157895, + "2054899126/train2": 0.5630252100840336, + "2061635888/train0": 0.5, + "2061635888/train1": 0.4722222222222222, + "2061635888/train2": 0.5833333333333334, + "211245842/train0": 0.5546218487394958, + "211245842/train1": 0.625, + "211245842/train2": 0.575, + "278008524/train0": 0.5419191919191919, + "278008524/train1": 0.5556159420289855, + "278008524/train2": 0.47184343434343434, + "307068879/train0": 0.4945054945054945, + "307068879/train1": 0.5054945054945055, + "307068879/train2": 0.4725274725274725, + "412102563/train0": 0.5166666666666667, + "412102563/train1": 0.7692307692307693, + "412102563/train2": 0.38095238095238093, + "440577711/train0": 0.5862068965517241, + "440577711/train1": 0.26666666666666666, + "440577711/train2": 0.4444444444444444, + "441046872/train0": 0.36363636363636365, + "441046872/train1": 0.6363636363636364, + "441046872/train2": 0.43636363636363634, + "509430392/train0": 0.6302521008403361, + "509430392/train1": 0.5619047619047619, + "509430392/train2": 0.5630252100840336, + "585904130/train0": 0.4880952380952381, + "585904130/train1": 0.6, + "585904130/train2": 0.7727272727272727, + "608016290/train0": 0.5428571428571428, + "608016290/train1": 0.49523809523809526, + "608016290/train2": 0.5, + "626486813/train0": 0.6458333333333333, + "626486813/train1": 0.35833333333333334, + "626486813/train2": 0.6458333333333334, + "636680137/train0": 0.38333333333333336, + "636680137/train1": 0.675, + "636680137/train2": 0.48333333333333334, + "655884605/train0": 0.5277777777777778, + "655884605/train1": 0.4444444444444444, + "655884605/train2": 0.5555555555555556, + "672328262/train0": 0.8235294117647058, + "672328262/train1": 0.45714285714285713, + "672328262/train2": 0.43333333333333335, + "701667486/train0": 0.5476190476190477, + "701667486/train1": 0.6071428571428571, + "701667486/train2": 0.5185185185185185, + "747260437/train0": 0.5, + "747260437/train1": 0.6333333333333333, + "747260437/train2": 0.5916666666666667, + "750448296/train0": 0.5083333333333334, + "750448296/train1": 0.5621794871794873, + "750448296/train2": 0.5559523809523809, + "767747424/train0": 0.5313034188034188, + "767747424/train1": 0.5907993966817496, + "767747424/train2": 0.5065112583494937 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5351217624780109, + "low": 0.4288224360452929, + "mean": 0.492973171068228, + "n": 5, + "num_samples": 300 + }, + "mean": 0.492973171068228, + "per_task": { + "LiftPegUpright-v1": 0.5353055242569309, + "PickCube-v1": 0.3703703703703704, + "PullCube-v1": 0.5394044361388536, + "PushCube-v1": 0.5261888915984856, + "StackCube-v1": 0.49359663297649975 + } + }, + "metric": "pairwise_causal_calibration_accuracy", + "micro": { + "confidence": 0.95, + "high": 0.5431970839321437, + "low": 0.4996715805830799, + "mean": 0.5226064420691535, + "n": 138, + "num_samples": 300 + }, + "n": 138 + }, + "pairwise_causal_calibration_confidence": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5359009794841617, + "low": 0.5293666416536561, + "mean": 0.532499603279374, + "n": 117, + "num_samples": 300 + }, + "mean": 0.532499603279374, + "per_seed": { + "1073161778/train0": 0.5258276499617055, + "1073161778/train1": 0.5233895260511188, + "1073161778/train2": 0.5263938706546207, + "1113371261/train0": 0.5395033820587356, + "1113371261/train1": 0.54070868503236, + "1113371261/train2": 0.5357178921702087, + "1151410346/train0": 0.531157925956513, + "1151410346/train1": 0.5307398847955603, + "1151410346/train2": 0.5321764777196925, + "1261930456/train0": 0.5167751391406238, + "1261930456/train1": 0.5197662577584645, + "1261930456/train2": 0.5178085323750468, + "1324133122/train0": 0.5618284314570119, + "1324133122/train1": 0.5611245570508201, + "1324133122/train2": 0.5603240341069567, + "1324642411/train0": 0.5083549905395566, + "1324642411/train1": 0.5074297248847603, + "1324642411/train2": 0.5073524497785574, + "1375492786/train0": 0.529960870770317, + "1375492786/train1": 0.5259219860596367, + "1375492786/train2": 0.5355938240012034, + "141984615/train0": 0.523727380049807, + "141984615/train1": 0.5258782846930987, + "141984615/train2": 0.5265869328502524, + "1461069713/train0": 0.5097507053831263, + "1461069713/train1": 0.5100392790091248, + "1461069713/train2": 0.5122176002496256, + "1480856810/train0": 0.5274298724184954, + "1480856810/train1": 0.5168654047616045, + "1480856810/train2": 0.517349995500313, + "1509980664/train0": 0.5409580694546181, + "1509980664/train1": 0.538847046384772, + "1509980664/train2": 0.5413409105541477, + "1522272686/train0": 0.53869991771853, + "1522272686/train1": 0.544614872631952, + "1522272686/train2": 0.5443083211760574, + "1549324805/train0": 0.509674291684611, + "1549324805/train1": 0.5050545552431692, + "1549324805/train2": 0.5086088218670698, + "1578990937/train0": 0.5530254674859004, + "1578990937/train1": 0.5470265023854355, + "1578990937/train2": 0.5477244061705656, + "1790188787/train0": 0.507883924063129, + "1790188787/train1": 0.506702038669967, + "1790188787/train2": 0.5080851163654574, + "1822210471/train0": 0.5517394604755897, + "1822210471/train1": 0.5595321641601091, + "1822210471/train2": 0.5509339900966759, + "1845489013/train0": 0.5082603489371776, + "1845489013/train1": 0.5242226800008922, + "1845489013/train2": 0.5133463004507663, + "1999022574/train0": 0.5386338994647198, + "1999022574/train1": 0.5322044449316384, + "1999022574/train2": 0.5287541242780439, + "2012211287/train0": 0.5298480396428488, + "2012211287/train1": 0.5310578566731103, + "2012211287/train2": 0.533819076214467, + "2038594191/train0": 0.5818783523078931, + "2038594191/train1": 0.5700204949171992, + "2038594191/train2": 0.5648483849875251, + "2054899126/train0": 0.5278373424628687, + "2054899126/train1": 0.5259923842692189, + "2054899126/train2": 0.5296355479804102, + "2061635888/train0": 0.5414447509454381, + "2061635888/train1": 0.5381973885533406, + "2061635888/train2": 0.5409068778231741, + "211245842/train0": 0.518448754321454, + "211245842/train1": 0.5162865066172287, + "211245842/train2": 0.5180070692001414, + "278008524/train0": 0.5301030419063165, + "278008524/train1": 0.5304372589720707, + "278008524/train2": 0.5341679569247471, + "307068879/train0": 0.5277478798612748, + "307068879/train1": 0.5229087111015109, + "307068879/train2": 0.5244461582060973, + "412102563/train0": 0.5475641766574334, + "412102563/train1": 0.546479468539056, + "412102563/train2": 0.5484312023799853, + "440577711/train0": 0.529336326158639, + "440577711/train1": 0.5208750509234744, + "440577711/train2": 0.525156875392907, + "441046872/train0": 0.5291035564844788, + "441046872/train1": 0.5234537989267526, + "441046872/train2": 0.5275776417866733, + "509430392/train0": 0.5148097853516517, + "509430392/train1": 0.5134314316187146, + "509430392/train2": 0.5164530817228409, + "585904130/train0": 0.5333471334307192, + "585904130/train1": 0.5285102521950974, + "585904130/train2": 0.5338099460898557, + "608016290/train0": 0.5574158017414859, + "608016290/train1": 0.5568088448070179, + "608016290/train2": 0.56209932221557, + "626486813/train0": 0.5656592664914735, + "626486813/train1": 0.5628388482930173, + "626486813/train2": 0.5625202429462706, + "636680137/train0": 0.5210271884357356, + "636680137/train1": 0.5170756938619067, + "636680137/train2": 0.5151375785964762, + "655884605/train0": 0.5624770111779376, + "655884605/train1": 0.5653526989123352, + "655884605/train2": 0.56240741403742, + "672328262/train0": 0.5545676976677929, + "672328262/train1": 0.5403743645771392, + "672328262/train2": 0.5452442045537841, + "701667486/train0": 0.5311290004714053, + "701667486/train1": 0.5333844210863801, + "701667486/train2": 0.5311189080220702, + "747260437/train0": 0.5223126728351409, + "747260437/train1": 0.5248794662055539, + "747260437/train2": 0.5203117574231598, + "750448296/train0": 0.5222439903086803, + "750448296/train1": 0.5206535156871791, + "750448296/train2": 0.5240722010838824, + "767747424/train0": 0.5288627189688364, + "767747424/train1": 0.5276314238144284, + "767747424/train2": 0.5305845460281344 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5401894019322679, + "low": 0.5240603128744336, + "mean": 0.5329968012689369, + "n": 5, + "num_samples": 300 + }, + "mean": 0.5329968012689369, + "per_task": { + "LiftPegUpright-v1": 0.5334598489552488, + "PickCube-v1": 0.5425410371755132, + "PullCube-v1": 0.532007783280179, + "PushCube-v1": 0.5156592220791224, + "StackCube-v1": 0.5413161148546212 + } + }, + "metric": "pairwise_causal_calibration_confidence", + "micro": { + "confidence": 0.95, + "high": 0.5358428244849915, + "low": 0.5292508202467114, + "mean": 0.5324880846363355, + "n": 138, + "num_samples": 300 + }, + "n": 138 + }, + "pairwise_causal_calibration_ece": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.12194642575208414, + "low": 0.09473437972456429, + "mean": 0.10909280010740333, + "n": 120, + "num_samples": 300 + }, + "mean": 0.10909280010740333, + "per_seed": { + "1073161778/train0": 0.1128278122231684, + "1073161778/train1": 0.10994380728221453, + "1073161778/train2": 0.1402727960120459, + "1113371261/train0": 0.07521766777302136, + "1113371261/train1": 0.06643417211049711, + "1113371261/train2": 0.03571067925836269, + "1151410346/train0": 0.007348402146989175, + "1151410346/train1": 0.0735970276527032, + "1151410346/train2": 0.17979552533874008, + "1261930456/train0": 0.052128674494159144, + "1261930456/train1": 0.010675348667555484, + "1261930456/train2": 0.10123908667257231, + "1324133122/train0": 0.13325700288558334, + "1324133122/train1": 0.09683884276510585, + "1324133122/train2": 0.3460383198212425, + "1324642411/train0": 0.008311676127110146, + "1324642411/train1": 0.19845262805641617, + "1324642411/train2": 0.019920277494169802, + "1375492786/train0": 0.1500391292296831, + "1375492786/train1": 0.00039380341404748886, + "1375492786/train2": 0.27392998552260617, + "141984615/train0": 0.15706071338314037, + "141984615/train1": 0.07412171530690126, + "141984615/train2": 0.0015869328502523539, + "1461069713/train0": 0.17641737204979296, + "1461069713/train1": 0.08503927900912484, + "1461069713/train2": 0.07944906641704108, + "1480856810/train0": 0.3008529558643329, + "1480856810/train1": 0.05456316666696692, + "1480856810/train2": 0.04567521458372059, + "1509980664/train0": 0.12984695834350699, + "1509980664/train1": 0.22138672892445443, + "1509980664/train2": 0.05245202166525893, + "1522272686/train0": 0.038699917718529986, + "1522272686/train1": 0.3501704281875075, + "1522272686/train2": 0.12764165450939086, + "1549324805/train0": 0.12365904164872232, + "1549324805/train1": 0.028278778090164103, + "1549324805/train2": 0.09139117813293018, + "1578990937/train0": 0.15637624191580896, + "1578990937/train1": 0.10157195693088991, + "1578990937/train2": 0.18637986835543957, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.12606574224494715, + "1790188787/train1": 0.12950905621382663, + "1790188787/train2": 0.15353966182000284, + "1822210471/train0": 0.08897015573763616, + "1822210471/train1": 0.09352906032968679, + "1822210471/train2": 0.12593399009667586, + "1845489013/train0": 0.2039125228502211, + "1845489013/train1": 0.4757773199991078, + "1845489013/train2": 0.11334630045076632, + "1999022574/train0": 0.2529196137504342, + "1999022574/train1": 0.06779555506836145, + "1999022574/train2": 0.12085938743593866, + "2012211287/train0": 0.12648669510503363, + "2012211287/train1": 0.03560880999355631, + "2012211287/train2": 0.14009396726379386, + "2038594191/train0": 0.27153352472168624, + "2038594191/train1": 0.27372419862090286, + "2038594191/train2": 0.1305049506440908, + "2054899126/train0": 0.13891867123143437, + "2054899126/train1": 0.0791090710232853, + "2054899126/train2": 0.016694831051811687, + "2061635888/train0": 0.06262112758251581, + "2061635888/train1": 0.06597516633111851, + "2061635888/train2": 0.05426508199292427, + "211245842/train0": 0.03617309441804184, + "211245842/train1": 0.10871349338277125, + "211245842/train2": 0.056992930799858565, + "278008524/train0": 0.1683259518797704, + "278008524/train1": 0.04898005185912038, + "278008524/train2": 0.06379285658810804, + "307068879/train0": 0.03324238535578037, + "307068879/train1": 0.017414205607005462, + "307068879/train2": 0.05191868567862479, + "412102563/train0": 0.0700132758404072, + "412102563/train1": 0.22275130069171328, + "412102563/train2": 0.16747882142760429, + "440577711/train0": 0.056870570393085074, + "440577711/train1": 0.25420838425680775, + "440577711/train2": 0.08071243094846259, + "441046872/train0": 0.16546719284811517, + "441046872/train1": 0.11290983743688376, + "441046872/train2": 0.09121400542303698, + "509430392/train0": 0.11544231548868444, + "509430392/train1": 0.04847333028604728, + "509430392/train2": 0.04657212836119273, + "585904130/train0": 0.045251895335481074, + "585904130/train1": 0.07148974780490258, + "585904130/train2": 0.23891732663741705, + "608016290/train0": 0.02535410150173202, + "608016290/train1": 0.06157074956892268, + "608016290/train2": 0.0620993222155701, + "626486813/train0": 0.11750606320869803, + "626486813/train1": 0.20964815489410538, + "626486813/train2": 0.10372314663868182, + "636680137/train0": 0.13769385510240223, + "636680137/train1": 0.15792430613809338, + "636680137/train2": 0.031804245263142905, + "655884605/train0": 0.03469923340015974, + "655884605/train1": 0.1209082544678908, + "655884605/train2": 0.03641674663594687, + "672328262/train0": 0.26896171409691294, + "672328262/train1": 0.0832315074342818, + "672328262/train2": 0.1119108712204508, + "701667486/train0": 0.01649004714764235, + "701667486/train1": 0.07375843605647703, + "701667486/train2": 0.01260038950355169, + "747260437/train0": 0.022312672835140934, + "747260437/train1": 0.10845386712777938, + "747260437/train2": 0.07135490924350685, + "750448296/train0": 0.05973654993182551, + "750448296/train1": 0.1144300659008682, + "750448296/train2": 0.12386393190756012, + "767747424/train0": 0.13295795580777578, + "767747424/train1": 0.14253701144399372, + "767747424/train2": 0.15317336838473566 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.14504401671121855, + "low": 0.09524685046648067, + "mean": 0.11614647880266613, + "n": 5, + "num_samples": 300 + }, + "mean": 0.11614647880266613, + "per_task": { + "LiftPegUpright-v1": 0.08576162490103227, + "PickCube-v1": 0.17217066680514279, + "PullCube-v1": 0.1182675802568201, + "PushCube-v1": 0.10017848048000332, + "StackCube-v1": 0.10435404157033219 + } + }, + "metric": "pairwise_causal_calibration_ece", + "micro": { + "confidence": 0.95, + "high": 0.1254588065829863, + "low": 0.09690946083383803, + "mean": 0.11055486745592664, + "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": 0.14361744151797284, + "low": 0.052973391836692606, + "mean": 0.09832945719580487, + "n": 120, + "num_samples": 300 + }, + "mean": 0.09832945719580487, + "per_seed": { + "1073161778/train0": -0.02282134674243222, + "1073161778/train1": -0.1309941525333758, + "1073161778/train2": -0.01090870267033292, + "1113371261/train0": 0.0030470281399136434, + "1113371261/train1": 0.02463520530320129, + "1113371261/train2": -0.07970004189125712, + "1151410346/train0": 0.20840851240351038, + "1151410346/train1": 0.22134577714486947, + "1151410346/train2": 0.2005950618303269, + "1261930456/train0": -0.12059592120185664, + "1261930456/train1": -0.11462503285369541, + "1261930456/train2": -0.046092622673385486, + "1324133122/train0": 0.04291783793925408, + "1324133122/train1": 0.23171951530495727, + "1324133122/train2": 0.08474741252260032, + "1324642411/train0": 0.5904813494260328, + "1324642411/train1": 0.29159110923611015, + "1324642411/train2": 0.056361885016702556, + "1375492786/train0": -0.3263045462730741, + "1375492786/train1": -0.8215150285449934, + "1375492786/train2": -0.6500022540507773, + "141984615/train0": 0.09190846980596201, + "141984615/train1": 0.15467096667199068, + "141984615/train2": -0.07577538797561234, + "1461069713/train0": 0.027181194046941026, + "1461069713/train1": 0.13909218118483407, + "1461069713/train2": 0.08608232017406181, + "1480856810/train0": 0.8212684057863501, + "1480856810/train1": 0.5782627775556237, + "1480856810/train2": 0.4155042382010001, + "1509980664/train0": -0.046710380447063826, + "1509980664/train1": -0.03542479330582579, + "1509980664/train2": -0.08315965672063733, + "1522272686/train0": -0.5860479678966253, + "1522272686/train1": -0.3307920673952962, + "1522272686/train2": -0.36394636603715164, + "1549324805/train0": 0.4651587653308018, + "1549324805/train1": 0.17043131073244489, + "1549324805/train2": 0.3864742495974729, + "1578990937/train0": 0.35124403976358615, + "1578990937/train1": 0.29218243151021844, + "1578990937/train2": 0.26493380406470457, + "1747178030/train0": 0.015531805591637943, + "1747178030/train1": 0.02856820772223631, + "1747178030/train2": 0.042189669044126965, + "1790188787/train0": 0.10954971541551203, + "1790188787/train1": 0.1144364826187316, + "1790188787/train2": 0.08944061124079167, + "1822210471/train0": 0.01957929235012683, + "1822210471/train1": -0.03627695714246994, + "1822210471/train2": 0.00519928677928394, + "1845489013/train0": -0.2719633789755977, + "1845489013/train1": -0.22208543687845692, + "1845489013/train2": -0.19739643651997574, + "1999022574/train0": 0.1009604457083228, + "1999022574/train1": 0.03337865189847511, + "1999022574/train2": 0.027745817665301403, + "2012211287/train0": 0.06399336728918298, + "2012211287/train1": -0.10501261504636256, + "2012211287/train2": 0.01094329025435617, + "2038594191/train0": 0.23900074944006316, + "2038594191/train1": 0.3330720074528296, + "2038594191/train2": 0.3386689344106315, + "2054899126/train0": 0.1668147893437682, + "2054899126/train1": 0.1906934052878418, + "2054899126/train2": 0.17505715348898754, + "2061635888/train0": 0.23896323450290802, + "2061635888/train1": 0.18138954354814413, + "2061635888/train2": 0.14213746526284532, + "211245842/train0": 0.13866925649969028, + "211245842/train1": 0.25579648935341137, + "211245842/train2": 0.15934077937193158, + "278008524/train0": 0.17951238150187201, + "278008524/train1": 0.2526760602392744, + "278008524/train2": 0.23818700356023634, + "307068879/train0": 0.008148511169388736, + "307068879/train1": -0.19456398757464405, + "307068879/train2": 0.0840500728219218, + "412102563/train0": 0.18513254256821268, + "412102563/train1": 0.14951478428388437, + "412102563/train2": 0.1263043297940751, + "440577711/train0": 0.5294705452065069, + "440577711/train1": 0.6282525745787317, + "440577711/train2": 0.599974637285598, + "441046872/train0": -0.034318319001119896, + "441046872/train1": -0.23082979751591187, + "441046872/train2": -0.08843117974127858, + "509430392/train0": 0.4358418255784948, + "509430392/train1": 0.3747876596460843, + "509430392/train2": 0.2716524269409146, + "585904130/train0": 0.10907919344361995, + "585904130/train1": 0.05173460799750952, + "585904130/train2": 0.10442126551685532, + "608016290/train0": 0.40694674604777425, + "608016290/train1": 0.22198156017266713, + "608016290/train2": 0.3766943789955354, + "626486813/train0": 0.5010382071158358, + "626486813/train1": 0.4605518879781133, + "626486813/train2": 0.3514059887171197, + "636680137/train0": 0.4860706638275267, + "636680137/train1": 0.11583401883298983, + "636680137/train2": 0.18578510943339843, + "655884605/train0": 0.17094865469784903, + "655884605/train1": 0.2625572115443, + "655884605/train2": 0.23713238846053236, + "672328262/train0": -0.10240791384393264, + "672328262/train1": -0.23155797487397917, + "672328262/train2": -0.166628216752181, + "701667486/train0": 0.27106178902715317, + "701667486/train1": 0.2505788547291904, + "701667486/train2": 0.31082243661779413, + "747260437/train0": -0.11408826824860728, + "747260437/train1": -0.27266683578349143, + "747260437/train2": -0.08183742266671763, + "750448296/train0": 0.014221736757998019, + "750448296/train1": -0.19773588166013986, + "750448296/train2": -0.15253919216558867, + "767747424/train0": -0.03602515072385583, + "767747424/train1": -0.003923417442009747, + "767747424/train2": 0.0115011599427335 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.21452552969477603, + "low": -0.21582586541468346, + "mean": 0.022183610963800605, + "n": 5, + "num_samples": 300 + }, + "mean": 0.022183610963800605, + "per_task": { + "LiftPegUpright-v1": 0.023658521728357904, + "PickCube-v1": -0.42692880044302434, + "PullCube-v1": 0.05246869761637144, + "PushCube-v1": 0.31253125927801373, + "StackCube-v1": 0.14918837663928425 + } + }, + "metric": "predicted_margin", + "micro": { + "confidence": 0.95, + "high": 0.13699651155446962, + "low": 0.048784757782340066, + "mean": 0.09415293640225546, + "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": 6.4847222222222225, + "low": 4.847916666666667, + "mean": 5.597916666666666, + "n": 120, + "num_samples": 300 + }, + "mean": 5.597916666666666, + "per_seed": { + "1073161778/train0": 6.0, + "1073161778/train1": 7.0, + "1073161778/train2": 6.0, + "1113371261/train0": 2.0, + "1113371261/train1": 1.0, + "1113371261/train2": 1.0, + "1151410346/train0": 9.0, + "1151410346/train1": 3.0, + "1151410346/train2": 8.0, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 4.0, + "1324133122/train0": 6.0, + "1324133122/train1": 2.0, + "1324133122/train2": 1.0, + "1324642411/train0": 13.0, + "1324642411/train1": 4.0, + "1324642411/train2": 2.0, + "1375492786/train0": 3.0, + "1375492786/train1": 1.0, + "1375492786/train2": 2.0, + "141984615/train0": 15.0, + "141984615/train1": 15.0, + "141984615/train2": 3.0, + "1461069713/train0": 1.0, + "1461069713/train1": 13.0, + "1461069713/train2": 3.0, + "1480856810/train0": 0.0, + "1480856810/train1": 2.0, + "1480856810/train2": 4.0, + "1509980664/train0": 10.0, + "1509980664/train1": 9.0, + "1509980664/train2": 9.0, + "1522272686/train0": 1.0, + "1522272686/train1": 2.0, + "1522272686/train2": 1.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 1.0, + "1578990937/train0": 9.0, + "1578990937/train1": 0.0, + "1578990937/train2": 7.0, + "1747178030/train0": 3.0, + "1747178030/train1": 7.0, + "1747178030/train2": 9.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 3.0, + "1822210471/train0": 5.0, + "1822210471/train1": 3.0, + "1822210471/train2": 4.0, + "1845489013/train0": 5.0, + "1845489013/train1": 1.0, + "1845489013/train2": 10.0, + "1999022574/train0": 5.0, + "1999022574/train1": 8.0, + "1999022574/train2": 7.0, + "2012211287/train0": 13.0, + "2012211287/train1": 13.0, + "2012211287/train2": 14.0, + "2038594191/train0": 5.0, + "2038594191/train1": 9.0, + "2038594191/train2": 3.0, + "2054899126/train0": 4.5, + "2054899126/train1": 7.0, + "2054899126/train2": 8.0, + "2061635888/train0": 3.0, + "2061635888/train1": 2.0, + "2061635888/train2": 7.0, + "211245842/train0": 7.0, + "211245842/train1": 4.0, + "211245842/train2": 4.0, + "278008524/train0": 8.0, + "278008524/train1": 8.0, + "278008524/train2": 14.0, + "307068879/train0": 11.0, + "307068879/train1": 6.0, + "307068879/train2": 3.0, + "412102563/train0": 10.0, + "412102563/train1": 8.0, + "412102563/train2": 7.0, + "440577711/train0": 1.0, + "440577711/train1": 10.0, + "440577711/train2": 13.0, + "441046872/train0": 7.0, + "441046872/train1": 6.0, + "441046872/train2": 1.0, + "509430392/train0": 15.0, + "509430392/train1": 3.0, + "509430392/train2": 1.0, + "585904130/train0": 14.0, + "585904130/train1": 15.0, + "585904130/train2": 1.0, + "608016290/train0": 11.0, + "608016290/train1": 3.0, + "608016290/train2": 1.0, + "626486813/train0": 11.0, + "626486813/train1": 10.0, + "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": 5.0, + "672328262/train1": 1.0, + "672328262/train2": 4.0, + "701667486/train0": 10.0, + "701667486/train1": 5.0, + "701667486/train2": 7.0, + "747260437/train0": 14.0, + "747260437/train1": 3.0, + "747260437/train2": 6.0, + "750448296/train0": 4.666666666666667, + "750448296/train1": 6.0, + "750448296/train2": 10.333333333333334, + "767747424/train0": 7.5, + "767747424/train1": 8.5, + "767747424/train2": 5.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 6.676121212121212, + "low": 3.14, + "mean": 4.962787878787879, + "n": 5, + "num_samples": 300 + }, + "mean": 4.962787878787879, + "per_task": { + "LiftPegUpright-v1": 7.2, + "PickCube-v1": 1.3333333333333333, + "PullCube-v1": 6.386666666666667, + "PushCube-v1": 4.5, + "StackCube-v1": 5.393939393939394 + } + }, + "metric": "selected_candidate_index", + "micro": { + "confidence": 0.95, + "high": 6.527777777777778, + "low": 5.215277777777778, + "mean": 5.902777777777778, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.4395833333333333, + "low": 0.26875, + "mean": 0.34375, + "n": 120, + "num_samples": 300 + }, + "mean": 0.34375, + "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": 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": 1.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 1.0, + "1845489013/train1": 1.0, + "1845489013/train2": 1.0, + "1999022574/train0": 1.0, + "1999022574/train1": 1.0, + "1999022574/train2": 1.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.0, + "2038594191/train2": 0.0, + "2054899126/train0": 0.5, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "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": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 0.0, + "509430392/train1": 1.0, + "509430392/train2": 0.0, + "585904130/train0": 1.0, + "585904130/train1": 0.0, + "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": 0.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 0.0, + "747260437/train1": 1.0, + "747260437/train2": 0.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.41393939393939394, + "low": 0.12460606060606061, + "mean": 0.27793939393939393, + "n": 5, + "num_samples": 300 + }, + "mean": 0.27793939393939393, + "per_task": { + "LiftPegUpright-v1": 0.26666666666666666, + "PickCube-v1": 0.0, + "PullCube-v1": 0.32, + "PushCube-v1": 0.5, + "StackCube-v1": 0.30303030303030304 + } + }, + "metric": "selected_success", + "micro": { + "confidence": 0.95, + "high": 0.3888888888888889, + "low": 0.25, + "mean": 0.3263888888888889, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_success_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.10416666666666667, + "low": -0.025, + "mean": 0.0375, + "n": 120, + "num_samples": 300 + }, + "mean": 0.0375, + "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": 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": 1.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 1.0, + "1578990937/train1": 0.0, + "1578990937/train2": 0.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 0.0, + "1822210471/train0": 0.0, + "1822210471/train1": 0.0, + "1822210471/train2": 0.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": -1.0, + "2038594191/train1": -1.0, + "2038594191/train2": -1.0, + "2054899126/train0": -0.5, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "278008524/train1": 0.5, + "278008524/train2": 0.5, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 0.0, + "509430392/train1": 1.0, + "509430392/train2": 0.0, + "585904130/train0": 1.0, + "585904130/train1": 0.0, + "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": 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": 0.0, + "747260437/train2": -1.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": 0.0, + "767747424/train0": 0.0, + "767747424/train1": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.21333333333333332, + "low": -0.010666666666666668, + "mean": 0.07806060606060605, + "n": 5, + "num_samples": 300 + }, + "mean": 0.07806060606060605, + "per_task": { + "LiftPegUpright-v1": 0.06666666666666667, + "PickCube-v1": 0.0, + "PullCube-v1": -0.04, + "PushCube-v1": 0.3333333333333333, + "StackCube-v1": 0.030303030303030304 + } + }, + "metric": "selected_success_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.09722222222222222, + "low": -0.034722222222222224, + "mean": 0.034722222222222224, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.9943820327475098, + "low": 0.679139230467586, + "mean": 0.8208872987106184, + "n": 120, + "num_samples": 300 + }, + "mean": 0.8208872987106184, + "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": 0.07714252173900604, + "1324133122/train2": 0.09220491349697113, + "1324642411/train0": 0.11045536398887634, + "1324642411/train1": 0.11658626794815063, + "1324642411/train2": 0.10647052526473999, + "1375492786/train0": 2.0, + "1375492786/train1": 2.0, + "1375492786/train2": 2.0, + "141984615/train0": 0.006696522235870361, + "141984615/train1": 0.030119141563773155, + "141984615/train2": 0.02716982364654541, + "1461069713/train0": 0.014822185039520264, + "1461069713/train1": 0.006729066371917725, + "1461069713/train2": 0.09997648000717163, + "1480856810/train0": 2.0, + "1480856810/train1": 0.1545013189315796, + "1480856810/train2": 0.21894000470638275, + "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": 0.0260902252048254, + "1549324805/train1": 0.0303483996540308, + "1549324805/train2": 0.04690708592534065, + "1578990937/train0": 2.0, + "1578990937/train1": 0.12732218205928802, + "1578990937/train2": 0.15726424753665924, + "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": 0.045472364872694016, + "1822210471/train2": 0.08447941392660141, + "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": 0.7026570439338684, + "2012211287/train1": 0.7236053347587585, + "2012211287/train2": 0.6934840679168701, + "2038594191/train0": 0.04285776615142822, + "2038594191/train1": 0.07674666494131088, + "2038594191/train2": 0.12977690994739532, + "2054899126/train0": 1.0403175354003906, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 0.09799930453300476, + "2061635888/train1": 0.05225050449371338, + "2061635888/train2": 0.05225050449371338, + "211245842/train0": 0.19090841710567474, + "211245842/train1": 0.19948488473892212, + "211245842/train2": 0.19090841710567474, + "278008524/train0": 0.06754684634506702, + "278008524/train1": 1.0095977680757642, + "278008524/train2": 1.0227726306766272, + "307068879/train0": 0.19761089980602264, + "307068879/train1": 0.6000905632972717, + "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": 0.6999551653862, + "441046872/train2": 0.6999551653862, + "509430392/train0": 0.11310504376888275, + "509430392/train1": 2.0, + "509430392/train2": 2.682209014892578e-07, + "585904130/train0": 2.0, + "585904130/train1": 0.04792279005050659, + "585904130/train2": 2.0, + "608016290/train0": 2.0, + "608016290/train1": 2.0, + "608016290/train2": 0.6789124011993408, + "626486813/train0": 0.009900043485686183, + "626486813/train1": 0.006450424436479807, + "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": 0.09131735563278198, + "672328262/train0": 2.0, + "672328262/train1": 2.0, + "672328262/train2": 0.04013325646519661, + "701667486/train0": 2.0, + "701667486/train1": 2.0, + "701667486/train2": 2.0, + "747260437/train0": 0.061248183250427246, + "747260437/train1": 2.0, + "747260437/train2": 0.06438694149255753, + "750448296/train0": 0.11766380071640015, + "750448296/train1": 0.109618179500103, + "750448296/train2": 0.09164678926269214, + "767747424/train0": 0.5599935371428728, + "767747424/train1": 0.5966028785333037, + "767747424/train2": 0.5675513362511992 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.9598884660042822, + "low": 0.6576425695058071, + "mean": 0.806756487143273, + "n": 5, + "num_samples": 300 + }, + "mean": 0.806756487143273, + "per_task": { + "LiftPegUpright-v1": 0.9396337101856868, + "PickCube-v1": 0.5839591026306152, + "PullCube-v1": 0.6831284697043399, + "PushCube-v1": 1.0588933833771281, + "StackCube-v1": 0.7681677698185949 + } + }, + "metric": "selected_utility", + "micro": { + "confidence": 0.95, + "high": 0.8983550137264602, + "low": 0.6434929672686849, + "mean": 0.7742405242590595, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selected_utility_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.1567380258231424, + "low": -0.08708108724838691, + "mean": 0.030769236412824186, + "n": 120, + "num_samples": 300 + }, + "mean": 0.030769236412824186, + "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": -0.015062391757965088, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0039848387241363525, + "1324642411/train1": 0.010115742683410645, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": -0.09613772481679916, + "141984615/train1": -0.07271510548889637, + "141984615/train2": -0.07566442340612411, + "1461069713/train0": -0.07908087968826294, + "1461069713/train1": -0.08717399835586548, + "1461069713/train2": 0.006073415279388428, + "1480856810/train0": 1.9129659235477448, + "1480856810/train1": 0.06746724247932434, + "1480856810/train2": 0.1319059282541275, + "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": -0.07842461206018925, + "1549324805/train1": -0.07416643761098385, + "1549324805/train2": -0.057607751339673996, + "1578990937/train0": 1.17586350440979, + "1578990937/train1": -0.6968143135309219, + "1578990937/train2": -0.6668722480535507, + "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": -0.026442941278219223, + "1822210471/train2": 0.012564107775688171, + "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": -0.020948290824890137, + "2012211287/train1": 0.0, + "2012211287/train2": -0.030121266841888428, + "2038594191/train0": -1.9571422338485718, + "2038594191/train1": -1.9232533350586891, + "2038594191/train2": -1.8702230900526047, + "2054899126/train0": -0.9596824645996094, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.04574880003929138, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.008576467633247375, + "211245842/train2": 0.0, + "278008524/train0": 0.02481185644865036, + "278008524/train1": 0.9668627781793475, + "278008524/train2": 0.9800376407802105, + "307068879/train0": -0.4024796634912491, + "307068879/train1": 0.0, + "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": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 0.11310482025146484, + "509430392/train1": 1.999999776482582, + "509430392/train2": 4.470348358154297e-08, + "585904130/train0": 1.8182844519615173, + "585904130/train1": -0.13379275798797607, + "585904130/train2": 1.8182844519615173, + "608016290/train0": 1.274298369884491, + "608016290/train1": 1.274298369884491, + "608016290/train2": -0.04678922891616821, + "626486813/train0": 0.004801700764801353, + "626486813/train1": 0.0013520817155949771, + "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": -0.07168932259082794, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": -1.9598667435348034, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": -1.9387518167495728, + "747260437/train1": 0.0, + "747260437/train2": -1.9356130585074425, + "750448296/train0": 7.065137227376302e-05, + "750448296/train1": -0.007974969844023386, + "750448296/train2": -0.02594636008143425, + "767747424/train0": -0.013205804862082005, + "767747424/train1": 0.023403536528348923, + "767747424/train2": -0.0056480057537555695 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.39620077011260113, + "low": -0.05727256061664472, + "mean": 0.11119668996925795, + "n": 5, + "num_samples": 300 + }, + "mean": 0.11119668996925795, + "per_task": { + "LiftPegUpright-v1": -0.010236828525861105, + "PickCube-v1": 0.0, + "PullCube-v1": -0.09204199151912083, + "PushCube-v1": 0.6664892045987977, + "StackCube-v1": -0.008226934707526005 + } + }, + "metric": "selected_utility_gain_over_base", + "micro": { + "confidence": 0.95, + "high": 0.14060055556021528, + "low": -0.09054212117593528, + "mean": 0.03242093781672237, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "selector_regret": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5815499054584912, + "low": 0.33973770285245136, + "mean": 0.4591651936934796, + "n": 120, + "num_samples": 300 + }, + "mean": 0.4591651936934796, + "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": 0.05531613528728485, + "1324133122/train2": 0.10070674121379852, + "1324642411/train0": 0.03360468149185181, + "1324642411/train1": 1.8834137320518494, + "1324642411/train2": 1.89352947473526, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.008481383323669434, + "141984615/train1": 0.006891986355185509, + "141984615/train2": 0.031781893223524094, + "1461069713/train0": 0.04537618160247803, + "1461069713/train1": 0.009855946525931358, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 1.8454986810684204, + "1480856810/train2": 1.7810599952936172, + "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": 0.0347811970859766, + "1549324805/train1": 0.16857311688363552, + "1549324805/train2": 0.06619475409388542, + "1578990937/train0": 0.0, + "1578990937/train1": 1.872677817940712, + "1578990937/train2": 1.8427357524633408, + "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": 1.954527635127306, + "1822210471/train2": 1.9155205860733986, + "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": 1.2973429560661316, + "2012211287/train1": 1.2763946652412415, + "2012211287/train2": 1.3065159320831299, + "2038594191/train0": 1.9571422338485718, + "2038594191/train1": 1.9232533350586891, + "2038594191/train2": 1.8702230900526047, + "2054899126/train0": 0.036965616047382355, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.02355286478996277, + "2061635888/train2": 0.05880081653594971, + "211245842/train0": 1.8090915828943253, + "211245842/train1": 0.013472288846969604, + "211245842/train2": 0.04297435283660889, + "278008524/train0": 1.0058094765990973, + "278008524/train1": 0.028076826594769955, + "278008524/train2": 0.11317628063261509, + "307068879/train0": 0.01394665241241455, + "307068879/train1": 0.0, + "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": 0.023403704166412354, + "441046872/train2": 0.019179701805114746, + "509430392/train0": 1.8868949562311172, + "509430392/train1": 0.0, + "509430392/train2": 5.155801773071289e-06, + "585904130/train0": 0.0, + "585904130/train1": 1.9520772099494934, + "585904130/train2": 0.0, + "608016290/train0": 0.0, + "608016290/train1": 0.0, + "608016290/train2": 0.03281521797180176, + "626486813/train0": 0.034795432118698955, + "626486813/train1": 0.029377450235188007, + "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": 0.12315595149993896, + "672328262/train0": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 1.9598667435348034, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.04714745283126831, + "747260437/train1": 0.0, + "747260437/train2": 0.06071493774652481, + "750448296/train0": 0.6349965271850427, + "750448296/train1": 1.2608850946029027, + "750448296/train2": 1.25880187874039, + "767747424/train0": 0.4824356506578624, + "767747424/train1": 0.4663502126932144, + "767747424/train2": 0.49961575865745544 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.6092645930611725, + "low": 0.17313177410400277, + "mean": 0.4089130798024737, + "n": 5, + "num_samples": 300 + }, + "mean": 0.4089130798024737, + "per_task": { + "LiftPegUpright-v1": 0.47343848447004955, + "PickCube-v1": 0.0, + "PullCube-v1": 0.4520184968970716, + "PushCube-v1": 0.726888031595283, + "StackCube-v1": 0.39222038604996423 + } + }, + "metric": "selector_regret", + "micro": { + "confidence": 0.95, + "high": 0.6026968432965482, + "low": 0.34726376117194174, + "mean": 0.46548765168538214, + "n": 144, + "num_samples": 300 + }, + "n": 144 + }, + "success_selector_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.325, + "low": 0.17916666666666664, + "mean": 0.24930555555555556, + "n": 120, + "num_samples": 300 + }, + "mean": 0.24930555555555556, + "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": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "1522272686/train2": 0.0, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": 0.0, + "1578990937/train1": 1.0, + "1578990937/train2": 1.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "1822210471/train0": 1.0, + "1822210471/train1": 1.0, + "1822210471/train2": 1.0, + "1845489013/train0": 0.0, + "1845489013/train1": 0.0, + "1845489013/train2": 0.0, + "1999022574/train0": 0.0, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 1.0, + "2012211287/train1": 1.0, + "2012211287/train2": 1.0, + "2038594191/train0": 1.0, + "2038594191/train1": 1.0, + "2038594191/train2": 1.0, + "2054899126/train0": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.5, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 0.0, + "307068879/train1": 0.0, + "307068879/train2": 0.0, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 0.0, + "440577711/train0": 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": 0.0, + "585904130/train1": 1.0, + "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": 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": 0.6666666666666666, + "750448296/train2": 0.6666666666666666, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.3424242424242424, + "low": 0.10909090909090909, + "mean": 0.23220202020202022, + "n": 5, + "num_samples": 300 + }, + "mean": 0.23220202020202022, + "per_task": { + "LiftPegUpright-v1": 0.3333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": 0.22666666666666666, + "PushCube-v1": 0.3888888888888889, + "StackCube-v1": 0.21212121212121213 + } + }, + "metric": "success_selector_gap", + "micro": { + "confidence": 0.95, + "high": 0.3263888888888889, + "low": 0.18055555555555555, + "mean": 0.25, + "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": 0.029607004640731814, + "low": -0.03261993689211708, + "mean": -0.0022591183112715177, + "n": 120, + "num_samples": 300 + }, + "mean": -0.0022591183112715177, + "per_seed": { + "1073161778/train0": -0.08577718626706465, + "1073161778/train1": -0.08577718626706465, + "1073161778/train2": -0.08577718626706465, + "1113371261/train0": 0.22913040029030945, + "1113371261/train1": 0.22913040029030945, + "1113371261/train2": 0.22913040029030945, + "1151410346/train0": 0.22913040029030945, + "1151410346/train1": 0.22913040029030945, + "1151410346/train2": 0.22913040029030945, + "1261930456/train0": 0.059256325887868444, + "1261930456/train1": 0.059256325887868444, + "1261930456/train2": 0.059256325887868444, + "1324133122/train0": 0.22913040029030945, + "1324133122/train1": 0.22913040029030945, + "1324133122/train2": 0.22913040029030945, + "1324642411/train0": 0.059256325887868444, + "1324642411/train1": 0.059256325887868444, + "1324642411/train2": 0.059256325887868444, + "1375492786/train0": -0.1741162794419627, + "1375492786/train1": -0.1741162794419627, + "1375492786/train2": -0.1741162794419627, + "141984615/train0": -0.1741162794419627, + "141984615/train1": -0.1741162794419627, + "141984615/train2": -0.1741162794419627, + "1461069713/train0": -0.1741162794419627, + "1461069713/train1": -0.1741162794419627, + "1461069713/train2": -0.1741162794419627, + "1480856810/train0": 0.059256325887868444, + "1480856810/train1": 0.059256325887868444, + "1480856810/train2": 0.059256325887868444, + "1509980664/train0": 0.22913040029030945, + "1509980664/train1": 0.22913040029030945, + "1509980664/train2": 0.22913040029030945, + "1522272686/train0": 0.4225255908772142, + "1522272686/train1": 0.4225255908772142, + "1522272686/train2": 0.4225255908772142, + "1549324805/train0": -0.1741162794419627, + "1549324805/train1": -0.1741162794419627, + "1549324805/train2": -0.1741162794419627, + "1578990937/train0": 0.22913040029030945, + "1578990937/train1": 0.22913040029030945, + "1578990937/train2": 0.22913040029030945, + "1747178030/train0": -0.1741162794419627, + "1747178030/train1": -0.1741162794419627, + "1747178030/train2": -0.1741162794419627, + "1790188787/train0": 0.059256325887868444, + "1790188787/train1": 0.059256325887868444, + "1790188787/train2": 0.059256325887868444, + "1822210471/train0": -0.1741162794419627, + "1822210471/train1": -0.1741162794419627, + "1822210471/train2": -0.1741162794419627, + "1845489013/train0": -0.1741162794419627, + "1845489013/train1": -0.1741162794419627, + "1845489013/train2": -0.1741162794419627, + "1999022574/train0": 0.22913040029030945, + "1999022574/train1": 0.22913040029030945, + "1999022574/train2": 0.22913040029030945, + "2012211287/train0": -0.08577718626706465, + "2012211287/train1": -0.08577718626706465, + "2012211287/train2": -0.08577718626706465, + "2038594191/train0": -0.1741162794419627, + "2038594191/train1": -0.1741162794419627, + "2038594191/train2": -0.1741162794419627, + "2054899126/train0": -0.1741162794419627, + "2054899126/train1": -0.1741162794419627, + "2054899126/train2": -0.1741162794419627, + "2061635888/train0": 0.22913040029030945, + "2061635888/train1": 0.22913040029030945, + "2061635888/train2": 0.22913040029030945, + "211245842/train0": 0.22913040029030945, + "211245842/train1": 0.22913040029030945, + "211245842/train2": 0.22913040029030945, + "278008524/train0": -0.1741162794419627, + "278008524/train1": -0.1741162794419627, + "278008524/train2": -0.1741162794419627, + "307068879/train0": -0.08577718626706465, + "307068879/train1": -0.08577718626706465, + "307068879/train2": -0.08577718626706465, + "412102563/train0": 0.22913040029030945, + "412102563/train1": 0.22913040029030945, + "412102563/train2": 0.22913040029030945, + "440577711/train0": 0.059256325887868444, + "440577711/train1": 0.059256325887868444, + "440577711/train2": 0.059256325887868444, + "441046872/train0": -0.08577718626706465, + "441046872/train1": -0.08577718626706465, + "441046872/train2": -0.08577718626706465, + "509430392/train0": 0.059256325887868444, + "509430392/train1": 0.059256325887868444, + "509430392/train2": 0.059256325887868444, + "585904130/train0": -0.1741162794419627, + "585904130/train1": -0.1741162794419627, + "585904130/train2": -0.1741162794419627, + "608016290/train0": -0.08577718626706465, + "608016290/train1": -0.08577718626706465, + "608016290/train2": -0.08577718626706465, + "626486813/train0": -0.1741162794419627, + "626486813/train1": -0.1741162794419627, + "626486813/train2": -0.1741162794419627, + "636680137/train0": -0.1741162794419627, + "636680137/train1": -0.1741162794419627, + "636680137/train2": -0.1741162794419627, + "655884605/train0": 0.22913040029030945, + "655884605/train1": 0.22913040029030945, + "655884605/train2": 0.22913040029030945, + "672328262/train0": -0.1741162794419627, + "672328262/train1": -0.1741162794419627, + "672328262/train2": -0.1741162794419627, + "701667486/train0": 0.22913040029030945, + "701667486/train1": 0.22913040029030945, + "701667486/train2": 0.22913040029030945, + "747260437/train0": -0.1741162794419627, + "747260437/train1": -0.1741162794419627, + "747260437/train2": -0.1741162794419627, + "750448296/train0": -0.1741162794419627, + "750448296/train1": -0.1741162794419627, + "750448296/train2": -0.1741162794419627, + "767747424/train0": -0.1741162794419627, + "767747424/train1": -0.1741162794419627, + "767747424/train2": -0.1741162794419627 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.27251366164458324, + "low": -0.09210612110603722, + "mean": 0.090203770269273, + "n": 5, + "num_samples": 300 + }, + "mean": 0.090203770269273, + "per_task": { + "LiftPegUpright-v1": -0.08577718626706465, + "PickCube-v1": 0.42252559087721425, + "PullCube-v1": -0.17411627944196265, + "PushCube-v1": 0.059256325887868444, + "StackCube-v1": 0.22913040029030962 + } + }, + "metric": "tau", + "micro": { + "confidence": 0.95, + "high": -0.0003851964424351642, + "low": -0.057694787503295264, + "mean": -0.03090197849971999, + "n": 144, + "num_samples": 300 + }, + "n": 144 + } + }, + "target": "utility_margin", + "tau": { + "LiftPegUpright-v1": -0.08577718626706465, + "PickCube-v1": 0.4225255908772142, + "PullCube-v1": -0.1741162794419627, + "PushCube-v1": 0.059256325887868444, + "StackCube-v1": 0.22913040029030945, + "__global__": -0.10099867826651904 + }, + "threshold_scope": "task", + "weights": [ + -0.06419078293939162, + 0.06581856642764146, + 0.06783757711377926, + 0.03583996030335042, + -0.021216149729034267, + -0.02150342787769646, + -0.15255679441090167, + -0.1306977314199636, + 0.03033969059163233, + 0.0417956341362353, + 0.029205509066893037, + 0.03169051701270447, + -0.04594148182579705, + -0.03688042414096415, + 0.010413683043998716, + -0.006162495427208757, + -0.017132426196872275, + 0.03703576928844393, + 0.008489290117141501, + -0.008491474229473939, + 0.002057393188079626, + 0.0010061120588211736, + -0.002983205425164559, + 0.005458188167173361, + 0.002538212645373239, + -0.00023205219498523277, + 0.008489290117138726, + -0.008491474229482154, + 0.0020573931880782936, + 0.0010061120588173988, + -0.0029832054251656137, + 0.005458188167174249, + 0.0025382126453701304, + -0.00023205219498612095, + 0.008688035410416406, + 0.002299792484484975, + -0.0063026156125245425, + 0.011566558392106518, + 0.011771296294692868, + 0.011386329580009846, + -3.3204989188462264e-05, + 0.008430405576806077, + -1.8627115564592392e-15, + -0.008956444760984894, + 0.005313011302490622, + -1.0621925501104184e-15, + -0.023727733702906484, + -0.09940519559184402, + 0.2557645705908722, + 0.2739557899160474, + 0.10262567490198815, + -0.12715605569545474, + -0.06963324252170722, + -0.00498403758683752, + -0.08026015728484337, + 0.2441090125676173, + 0.2739952156037364, + -1.785141671384296e-16, + 0.21431350474406585, + 0.23309302085346284, + 0.0017031473551851473, + -0.0994660079287843, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/metrics_by_seed.json b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/metrics_by_seed.json new file mode 100644 index 0000000000000000000000000000000000000000..ec9341a0694fdf00d2670defffd410ee026e0eb5 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_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": 0.5258276499617055, + "pairwise_causal_calibration_ece": 0.1128278122231684, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.02282134674243222, + "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": -0.08577718626706465 + }, + "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": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.5233895260511188, + "pairwise_causal_calibration_ece": 0.10994380728221453, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.1309941525333758, + "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": -0.08577718626706465 + }, + "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": 0.6666666666666666, + "pairwise_causal_calibration_confidence": 0.5263938706546207, + "pairwise_causal_calibration_ece": 0.1402727960120459, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.01090870267033292, + "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": -0.08577718626706465 + }, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5395033820587356, + "pairwise_causal_calibration_ece": 0.07521766777302136, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.0030470281399136434, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10274861752986908, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.897251382470131, + "tau": 0.22913040029030945 + }, + "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": 0.54070868503236, + "pairwise_causal_calibration_ece": 0.06643417211049711, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.02463520530320129, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18093222379684448, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02250880002975464, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8190677762031555, + "tau": 0.22913040029030945 + }, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5357178921702087, + "pairwise_causal_calibration_ece": 0.03571067925836269, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.07970004189125712, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19527167081832886, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15842342376708984, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.036848247051239014, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8047283291816711, + "tau": 0.22913040029030945 + }, + "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": 0.5238095238095238, + "pairwise_causal_calibration_confidence": 0.531157925956513, + "pairwise_causal_calibration_ece": 0.007348402146989175, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.20840851240351038, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8663521409034729, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.6403508484363556, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.133647859096527, + "tau": 0.22913040029030945 + }, + "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": 0.45714285714285713, + "pairwise_causal_calibration_confidence": 0.5307398847955603, + "pairwise_causal_calibration_ece": 0.0735970276527032, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.22134577714486947, + "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": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.7739987075328827, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.22913040029030945 + }, + "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": 0.3523809523809524, + "pairwise_causal_calibration_confidence": 0.5321764777196925, + "pairwise_causal_calibration_ece": 0.17979552533874008, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.2005950618303269, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5522047281265259, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2260012924671173, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.32620343565940857, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4477952718734741, + "tau": 0.22913040029030945 + }, + "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": 0.5167751391406238, + "pairwise_causal_calibration_ece": 0.052128674494159144, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": -0.12059592120185664, + "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": 0.059256325887868444 + }, + "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": 0.5197662577584645, + "pairwise_causal_calibration_ece": 0.010675348667555484, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": -0.11462503285369541, + "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": 0.059256325887868444 + }, + "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": 0.6190476190476191, + "pairwise_causal_calibration_confidence": 0.5178085323750468, + "pairwise_causal_calibration_ece": 0.10123908667257231, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.046092622673385486, + "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": 0.059256325887868444 + }, + "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": 0.5618284314570119, + "pairwise_causal_calibration_ece": 0.13325700288558334, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.04291783793925408, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.054989203810691833, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9450107961893082, + "tau": 0.22913040029030945 + }, + "1324133122/train1": { + "base_success": 0.0, + "base_utility": 0.09220491349697113, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5611245570508201, + "pairwise_causal_calibration_ece": 0.09683884276510585, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.23171951530495727, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1324586570262909, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07714252173900604, + "selected_utility_gain_over_base": -0.015062391757965088, + "selector_regret": 0.05531613528728485, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.867541342973709, + "tau": 0.22913040029030945 + }, + "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": 0.21428571428571427, + "pairwise_causal_calibration_confidence": 0.5603240341069567, + "pairwise_causal_calibration_ece": 0.3460383198212425, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.08474741252260032, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19291165471076965, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09220491349697113, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.10070674121379852, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8070883452892303, + "tau": 0.22913040029030945 + }, + "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": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.5083549905395566, + "pairwise_causal_calibration_ece": 0.008311676127110146, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5904813494260328, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14406004548072815, + "selected_candidate_index": 13.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11045536398887634, + "selected_utility_gain_over_base": 0.0039848387241363525, + "selector_regret": 0.03360468149185181, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8559399545192719, + "tau": 0.059256325887868444 + }, + "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": 0.5074297248847603, + "pairwise_causal_calibration_ece": 0.19845262805641617, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.29159110923611015, + "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": 0.11658626794815063, + "selected_utility_gain_over_base": 0.010115742683410645, + "selector_regret": 1.8834137320518494, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.059256325887868444 + }, + "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": 0.5272727272727272, + "pairwise_causal_calibration_confidence": 0.5073524497785574, + "pairwise_causal_calibration_ece": 0.019920277494169802, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.056361885016702556, + "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": 0.059256325887868444 + }, + "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": 0.68, + "pairwise_causal_calibration_confidence": 0.529960870770317, + "pairwise_causal_calibration_ece": 0.1500391292296831, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": -0.3263045462730741, + "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": -0.1741162794419627 + }, + "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": 0.5259219860596367, + "pairwise_causal_calibration_ece": 0.00039380341404748886, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": -0.8215150285449934, + "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": -0.1741162794419627 + }, + "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": 0.8095238095238095, + "pairwise_causal_calibration_confidence": 0.5355938240012034, + "pairwise_causal_calibration_ece": 0.27392998552260617, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": -0.6500022540507773, + "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": -0.1741162794419627 + }, + "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": 0.36666666666666664, + "pairwise_causal_calibration_confidence": 0.523727380049807, + "pairwise_causal_calibration_ece": 0.15706071338314037, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.09190846980596201, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.015177905559539795, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.006696522235870361, + "selected_utility_gain_over_base": -0.09613772481679916, + "selector_regret": 0.008481383323669434, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.270599365234375, + "tau": -0.1741162794419627 + }, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5258782846930987, + "pairwise_causal_calibration_ece": 0.07412171530690126, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.15467096667199068, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.037011127918958664, + "selected_candidate_index": 15.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.030119141563773155, + "selected_utility_gain_over_base": -0.07271510548889637, + "selector_regret": 0.006891986355185509, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24876614287495613, + "tau": -0.1741162794419627 + }, + "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": 0.525, + "pairwise_causal_calibration_confidence": 0.5265869328502524, + "pairwise_causal_calibration_ece": 0.0015869328502523539, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.07577538797561234, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058951716870069504, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.02716982364654541, + "selected_utility_gain_over_base": -0.07566442340612411, + "selector_regret": 0.031781893223524094, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2268255539238453, + "tau": -0.1741162794419627 + }, + "1461069713/train0": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.5097507053831263, + "pairwise_causal_calibration_ece": 0.17641737204979296, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.027181194046941026, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06019836664199829, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.014822185039520264, + "selected_utility_gain_over_base": -0.07908087968826294, + "selector_regret": 0.04537618160247803, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0991440862417221, + "tau": -0.1741162794419627 + }, + "1461069713/train1": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.425, + "pairwise_causal_calibration_confidence": 0.5100392790091248, + "pairwise_causal_calibration_ece": 0.08503927900912484, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.13909218118483407, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.016585012897849083, + "selected_candidate_index": 13.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.006729066371917725, + "selected_utility_gain_over_base": -0.08717399835586548, + "selector_regret": 0.009855946525931358, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14275743998587132, + "tau": -0.1741162794419627 + }, + "1461069713/train2": { + "base_success": 0.0, + "base_utility": 0.0939030647277832, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.1593424528837204, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5916666666666667, + "pairwise_causal_calibration_confidence": 0.5122176002496256, + "pairwise_causal_calibration_ece": 0.07944906641704108, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.08608232017406181, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997648000717163, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09997648000717163, + "selected_utility_gain_over_base": 0.006073415279388428, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05936597287654877, + "tau": -0.1741162794419627 + }, + "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": 0.8282828282828283, + "pairwise_causal_calibration_confidence": 0.5274298724184954, + "pairwise_causal_calibration_ece": 0.3008529558643329, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.8212684057863501, + "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": 0.059256325887868444 + }, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5168654047616045, + "pairwise_causal_calibration_ece": 0.05456316666696692, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.5782627775556237, + "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": 0.1545013189315796, + "selected_utility_gain_over_base": 0.06746724247932434, + "selector_regret": 1.8454986810684204, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.059256325887868444 + }, + "1480856810/train2": { + "base_success": 0.0, + "base_utility": 0.08703407645225525, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5630252100840336, + "pairwise_causal_calibration_confidence": 0.517349995500313, + "pairwise_causal_calibration_ece": 0.04567521458372059, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4155042382010001, + "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": 0.21894000470638275, + "selected_utility_gain_over_base": 0.1319059282541275, + "selector_regret": 1.7810599952936172, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.059256325887868444 + }, + "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": 0.4111111111111111, + "pairwise_causal_calibration_confidence": 0.5409580694546181, + "pairwise_causal_calibration_ece": 0.12984695834350699, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.046710380447063826, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.22913040029030945 + }, + "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": 0.31746031746031744, + "pairwise_causal_calibration_confidence": 0.538847046384772, + "pairwise_causal_calibration_ece": 0.22138672892445443, + "pairwise_causal_calibration_pairs": 63.0, + "predicted_margin": -0.03542479330582579, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.22913040029030945 + }, + "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": 0.4888888888888889, + "pairwise_causal_calibration_confidence": 0.5413409105541477, + "pairwise_causal_calibration_ece": 0.05245202166525893, + "pairwise_causal_calibration_pairs": 90.0, + "predicted_margin": -0.08315965672063733, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8334100246429443, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1665899753570557, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.22913040029030945 + }, + "1522272686/train0": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.53869991771853, + "pairwise_causal_calibration_ece": 0.038699917718529986, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.5860479678966253, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18325337767601013, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8167466223239899, + "tau": 0.4225255908772142 + }, + "1522272686/train1": { + "base_success": 0.0, + "base_utility": 0.5839591026306152, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.19444444444444445, + "pairwise_causal_calibration_confidence": 0.544614872631952, + "pairwise_causal_calibration_ece": 0.3501704281875075, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.3307920673952962, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07014383375644684, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9298561662435532, + "tau": 0.4225255908772142 + }, + "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": 0.4166666666666667, + "pairwise_causal_calibration_confidence": 0.5443083211760574, + "pairwise_causal_calibration_ece": 0.12764165450939086, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.36394636603715164, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14937306940555573, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5839591026306152, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8506269305944443, + "tau": 0.4225255908772142 + }, + "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": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.509674291684611, + "pairwise_causal_calibration_ece": 0.12365904164872232, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4651587653308018, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.060871422290802, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0260902252048254, + "selected_utility_gain_over_base": -0.07842461206018925, + "selector_regret": 0.0347811970859766, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.23658466339111328, + "tau": -0.1741162794419627 + }, + "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": 0.5333333333333333, + "pairwise_causal_calibration_confidence": 0.5050545552431692, + "pairwise_causal_calibration_ece": 0.028278778090164103, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.17043131073244489, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19892151653766632, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0303483996540308, + "selected_utility_gain_over_base": -0.07416643761098385, + "selector_regret": 0.16857311688363552, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09853456914424896, + "tau": -0.1741162794419627 + }, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5086088218670698, + "pairwise_causal_calibration_ece": 0.09139117813293018, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3864742495974729, + "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": -0.1741162794419627 + }, + "1578990937/train0": { + "base_success": 0.0, + "base_utility": 0.82413649559021, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7094017094017094, + "pairwise_causal_calibration_confidence": 0.5530254674859004, + "pairwise_causal_calibration_ece": 0.15637624191580896, + "pairwise_causal_calibration_pairs": 117.0, + "predicted_margin": 0.35124403976358615, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.17586350440979, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.22913040029030945 + }, + "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": 0.44545454545454544, + "pairwise_causal_calibration_confidence": 0.5470265023854355, + "pairwise_causal_calibration_ece": 0.10157195693088991, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.29218243151021844, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12732218205928802, + "selected_utility_gain_over_base": -0.6968143135309219, + "selector_regret": 1.872677817940712, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.22913040029030945 + }, + "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": 0.36134453781512604, + "pairwise_causal_calibration_confidence": 0.5477244061705656, + "pairwise_causal_calibration_ece": 0.18637986835543957, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.26493380406470457, + "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": 0.15726424753665924, + "selected_utility_gain_over_base": -0.6668722480535507, + "selector_regret": 1.8427357524633408, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.22913040029030945 + }, + "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": 0.015531805591637943, + "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": -0.1741162794419627 + }, + "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": 0.02856820772223631, + "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": -0.1741162794419627 + }, + "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": 0.042189669044126965, + "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": -0.1741162794419627 + }, + "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": 0.38181818181818183, + "pairwise_causal_calibration_confidence": 0.507883924063129, + "pairwise_causal_calibration_ece": 0.12606574224494715, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.10954971541551203, + "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": 0.059256325887868444 + }, + "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": 0.37719298245614036, + "pairwise_causal_calibration_confidence": 0.506702038669967, + "pairwise_causal_calibration_ece": 0.12950905621382663, + "pairwise_causal_calibration_pairs": 114.0, + "predicted_margin": 0.1144364826187316, + "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": 0.059256325887868444 + }, + "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": 0.35454545454545455, + "pairwise_causal_calibration_confidence": 0.5080851163654574, + "pairwise_causal_calibration_ece": 0.15353966182000284, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.08944061124079167, + "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": 0.059256325887868444 + }, + "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": 0.5517394604755897, + "pairwise_causal_calibration_ece": 0.08897015573763616, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 0.01957929235012683, + "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": -0.1741162794419627 + }, + "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": 0.5595321641601091, + "pairwise_causal_calibration_ece": 0.09352906032968679, + "pairwise_causal_calibration_pairs": 49.0, + "predicted_margin": -0.03627695714246994, + "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": 0.045472364872694016, + "selected_utility_gain_over_base": -0.026442941278219223, + "selector_regret": 1.954527635127306, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.1741162794419627 + }, + "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": 0.5509339900966759, + "pairwise_causal_calibration_ece": 0.12593399009667586, + "pairwise_causal_calibration_pairs": 40.0, + "predicted_margin": 0.00519928677928394, + "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": 0.08447941392660141, + "selected_utility_gain_over_base": 0.012564107775688171, + "selector_regret": 1.9155205860733986, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.1741162794419627 + }, + "1845489013/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.30434782608695654, + "pairwise_causal_calibration_confidence": 0.5082603489371776, + "pairwise_causal_calibration_ece": 0.2039125228502211, + "pairwise_causal_calibration_pairs": 46.0, + "predicted_margin": -0.2719633789755977, + "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": -0.1741162794419627 + }, + "1845489013/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": 1.0, + "pairwise_causal_calibration_confidence": 0.5242226800008922, + "pairwise_causal_calibration_ece": 0.4757773199991078, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": -0.22208543687845692, + "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": -0.1741162794419627 + }, + "1845489013/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": 0.4, + "pairwise_causal_calibration_confidence": 0.5133463004507663, + "pairwise_causal_calibration_ece": 0.11334630045076632, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": -0.19739643651997574, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.1741162794419627 + }, + "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": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5386338994647198, + "pairwise_causal_calibration_ece": 0.2529196137504342, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.1009604457083228, + "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": 0.22913040029030945 + }, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5322044449316384, + "pairwise_causal_calibration_ece": 0.06779555506836145, + "pairwise_causal_calibration_pairs": 70.0, + "predicted_margin": 0.03337865189847511, + "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": 0.22913040029030945 + }, + "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": 0.40789473684210525, + "pairwise_causal_calibration_confidence": 0.5287541242780439, + "pairwise_causal_calibration_ece": 0.12085938743593866, + "pairwise_causal_calibration_pairs": 76.0, + "predicted_margin": 0.027745817665301403, + "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": 0.22913040029030945 + }, + "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": 0.40336134453781514, + "pairwise_causal_calibration_confidence": 0.5298480396428488, + "pairwise_causal_calibration_ece": 0.12648669510503363, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.06399336728918298, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7026570439338684, + "selected_utility_gain_over_base": -0.020948290824890137, + "selector_regret": 1.2973429560661316, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08577718626706465 + }, + "2012211287/train1": { + "base_success": 0.0, + "base_utility": 0.7236053347587585, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5666666666666667, + "pairwise_causal_calibration_confidence": 0.5310578566731103, + "pairwise_causal_calibration_ece": 0.03560880999355631, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.10501261504636256, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7236053347587585, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.2763946652412415, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08577718626706465 + }, + "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": 0.6739130434782609, + "pairwise_causal_calibration_confidence": 0.533819076214467, + "pairwise_causal_calibration_ece": 0.14009396726379386, + "pairwise_causal_calibration_pairs": 92.0, + "predicted_margin": 0.01094329025435617, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 14.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6934840679168701, + "selected_utility_gain_over_base": -0.030121266841888428, + "selector_regret": 1.3065159320831299, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08577718626706465 + }, + "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": 0.3103448275862069, + "pairwise_causal_calibration_confidence": 0.5818783523078931, + "pairwise_causal_calibration_ece": 0.27153352472168624, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.23900074944006316, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04285776615142822, + "selected_utility_gain_over_base": -1.9571422338485718, + "selector_regret": 1.9571422338485718, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.1741162794419627 + }, + "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": 0.2962962962962963, + "pairwise_causal_calibration_confidence": 0.5700204949171992, + "pairwise_causal_calibration_ece": 0.27372419862090286, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.3330720074528296, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 9.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.07674666494131088, + "selected_utility_gain_over_base": -1.9232533350586891, + "selector_regret": 1.9232533350586891, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.1741162794419627 + }, + "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": 0.43434343434343436, + "pairwise_causal_calibration_confidence": 0.5648483849875251, + "pairwise_causal_calibration_ece": 0.1305049506440908, + "pairwise_causal_calibration_pairs": 99.0, + "predicted_margin": 0.3386689344106315, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.12977690994739532, + "selected_utility_gain_over_base": -1.8702230900526047, + "selector_regret": 1.8702230900526047, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.1741162794419627 + }, + "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": 0.5278373424628687, + "pairwise_causal_calibration_ece": 0.13891867123143437, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 0.1668147893437682, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.077283151447773, + "selected_candidate_index": 4.5, + "selected_success": 0.5, + "selected_success_gain_over_base": -0.5, + "selected_utility": 1.0403175354003906, + "selected_utility_gain_over_base": -0.9596824645996094, + "selector_regret": 0.036965616047382355, + "success_selector_gap": 0.0, + "success_support_gap": 0.5, + "support_gap": 0.922716848552227, + "tau": -0.1741162794419627 + }, + "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": 0.6842105263157895, + "pairwise_causal_calibration_confidence": 0.5259923842692189, + "pairwise_causal_calibration_ece": 0.0791090710232853, + "pairwise_causal_calibration_pairs": 57.0, + "predicted_margin": 0.1906934052878418, + "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": -0.1741162794419627 + }, + "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": 0.5296355479804102, + "pairwise_causal_calibration_ece": 0.016694831051811687, + "pairwise_causal_calibration_pairs": 59.5, + "predicted_margin": 0.17505715348898754, + "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": -0.1741162794419627 + }, + "2061635888/train0": { + "base_success": 0.0, + "base_utility": 0.05225050449371338, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.5414447509454381, + "pairwise_causal_calibration_ece": 0.06262112758251581, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.23896323450290802, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09799930453300476, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09799930453300476, + "selected_utility_gain_over_base": 0.04574880003929138, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9020006954669952, + "tau": 0.22913040029030945 + }, + "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": 0.4722222222222222, + "pairwise_causal_calibration_confidence": 0.5381973885533406, + "pairwise_causal_calibration_ece": 0.06597516633111851, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.18138954354814413, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07580336928367615, + "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": 0.02355286478996277, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9241966307163239, + "tau": 0.22913040029030945 + }, + "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": 0.5833333333333334, + "pairwise_causal_calibration_confidence": 0.5409068778231741, + "pairwise_causal_calibration_ece": 0.05426508199292427, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.14213746526284532, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11105132102966309, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05225050449371338, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.05880081653594971, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.888948678970337, + "tau": 0.22913040029030945 + }, + "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": 0.5546218487394958, + "pairwise_causal_calibration_confidence": 0.518448754321454, + "pairwise_causal_calibration_ece": 0.03617309441804184, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.13866925649969028, + "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": 0.19090841710567474, + "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": 0.22913040029030945 + }, + "211245842/train1": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.625, + "pairwise_causal_calibration_confidence": 0.5162865066172287, + "pairwise_causal_calibration_ece": 0.10871349338277125, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.25579648935341137, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21295717358589172, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19948488473892212, + "selected_utility_gain_over_base": 0.008576467633247375, + "selector_regret": 0.013472288846969604, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7870428264141083, + "tau": 0.22913040029030945 + }, + "211245842/train2": { + "base_success": 0.0, + "base_utility": 0.19090841710567474, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.575, + "pairwise_causal_calibration_confidence": 0.5180070692001414, + "pairwise_causal_calibration_ece": 0.056992930799858565, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.15934077937193158, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23388276994228363, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19090841710567474, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04297435283660889, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7661172300577164, + "tau": 0.22913040029030945 + }, + "278008524/train0": { + "base_success": 0.0, + "base_utility": 0.042734989896416664, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5419191919191919, + "pairwise_causal_calibration_confidence": 0.5301030419063165, + "pairwise_causal_calibration_ece": 0.1683259518797704, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.17951238150187201, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0733563229441643, + "selected_candidate_index": 8.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06754684634506702, + "selected_utility_gain_over_base": 0.02481185644865036, + "selector_regret": 1.0058094765990973, + "success_selector_gap": 0.5, + "success_support_gap": 0.0, + "support_gap": 0.04055231809616089, + "tau": -0.1741162794419627 + }, + "278008524/train1": { + "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": 0.5556159420289855, + "pairwise_causal_calibration_confidence": 0.5304372589720707, + "pairwise_causal_calibration_ece": 0.04898005185912038, + "pairwise_causal_calibration_pairs": 106.0, + "predicted_margin": 0.2526760602392744, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0376745946705341, + "selected_candidate_index": 8.0, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.5, + "selected_utility": 1.0095977680757642, + "selected_utility_gain_over_base": 0.9668627781793475, + "selector_regret": 0.028076826594769955, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.07623404636979103, + "tau": -0.1741162794419627 + }, + "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": 0.47184343434343434, + "pairwise_causal_calibration_confidence": 0.5341679569247471, + "pairwise_causal_calibration_ece": 0.06379285658810804, + "pairwise_causal_calibration_pairs": 109.5, + "predicted_margin": 0.23818700356023634, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.1359489113092422, + "selected_candidate_index": 14.0, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.5, + "selected_utility": 1.0227726306766272, + "selected_utility_gain_over_base": 0.9800376407802105, + "selector_regret": 0.11317628063261509, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.1741162794419627 + }, + "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": 0.4945054945054945, + "pairwise_causal_calibration_confidence": 0.5277478798612748, + "pairwise_causal_calibration_ece": 0.03324238535578037, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.008148511169388736, + "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": -0.08577718626706465 + }, + "307068879/train1": { + "base_success": 0.0, + "base_utility": 0.6000905632972717, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5054945054945055, + "pairwise_causal_calibration_confidence": 0.5229087111015109, + "pairwise_causal_calibration_ece": 0.017414205607005462, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": -0.19456398757464405, + "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": 0.6000905632972717, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4123926758766174, + "tau": -0.08577718626706465 + }, + "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": 0.4725274725274725, + "pairwise_causal_calibration_confidence": 0.5244461582060973, + "pairwise_causal_calibration_ece": 0.05191868567862479, + "pairwise_causal_calibration_pairs": 91.0, + "predicted_margin": 0.0840500728219218, + "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": -0.08577718626706465 + }, + "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": 0.5166666666666667, + "pairwise_causal_calibration_confidence": 0.5475641766574334, + "pairwise_causal_calibration_ece": 0.0700132758404072, + "pairwise_causal_calibration_pairs": 60.0, + "predicted_margin": 0.18513254256821268, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.22913040029030945 + }, + "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": 0.7692307692307693, + "pairwise_causal_calibration_confidence": 0.546479468539056, + "pairwise_causal_calibration_ece": 0.22275130069171328, + "pairwise_causal_calibration_pairs": 39.0, + "predicted_margin": 0.14951478428388437, + "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": 0.22913040029030945 + }, + "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": 0.38095238095238093, + "pairwise_causal_calibration_confidence": 0.5484312023799853, + "pairwise_causal_calibration_ece": 0.16747882142760429, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.1263043297940751, + "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": 0.22913040029030945 + }, + "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": 0.5862068965517241, + "pairwise_causal_calibration_confidence": 0.529336326158639, + "pairwise_causal_calibration_ece": 0.056870570393085074, + "pairwise_causal_calibration_pairs": 29.0, + "predicted_margin": 0.5294705452065069, + "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": 0.059256325887868444 + }, + "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": 0.26666666666666666, + "pairwise_causal_calibration_confidence": 0.5208750509234744, + "pairwise_causal_calibration_ece": 0.25420838425680775, + "pairwise_causal_calibration_pairs": 15.0, + "predicted_margin": 0.6282525745787317, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.059256325887868444 + }, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.525156875392907, + "pairwise_causal_calibration_ece": 0.08071243094846259, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.599974637285598, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 13.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.9638472199440002, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.059256325887868444 + }, + "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": 0.36363636363636365, + "pairwise_causal_calibration_confidence": 0.5291035564844788, + "pairwise_causal_calibration_ece": 0.16546719284811517, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.034318319001119896, + "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": -0.08577718626706465 + }, + "441046872/train1": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6363636363636364, + "pairwise_causal_calibration_confidence": 0.5234537989267526, + "pairwise_causal_calibration_ece": 0.11290983743688376, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.23082979751591187, + "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": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.023403704166412354, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2766411304473877, + "tau": -0.08577718626706465 + }, + "441046872/train2": { + "base_success": 0.0, + "base_utility": 0.6999551653862, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.43636363636363634, + "pairwise_causal_calibration_confidence": 0.5275776417866733, + "pairwise_causal_calibration_ece": 0.09121400542303698, + "pairwise_causal_calibration_pairs": 55.0, + "predicted_margin": -0.08843117974127858, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7191348671913147, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.6999551653862, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019179701805114746, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2808651328086853, + "tau": -0.08577718626706465 + }, + "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": 0.6302521008403361, + "pairwise_causal_calibration_confidence": 0.5148097853516517, + "pairwise_causal_calibration_ece": 0.11544231548868444, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.4358418255784948, + "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": 0.059256325887868444 + }, + "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": 0.5619047619047619, + "pairwise_causal_calibration_confidence": 0.5134314316187146, + "pairwise_causal_calibration_ece": 0.04847333028604728, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.3747876596460843, + "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": 1.999999776482582, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.059256325887868444 + }, + "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": 0.5630252100840336, + "pairwise_causal_calibration_confidence": 0.5164530817228409, + "pairwise_causal_calibration_ece": 0.04657212836119273, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.2716524269409146, + "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": 0.059256325887868444 + }, + "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": 0.4880952380952381, + "pairwise_causal_calibration_confidence": 0.5333471334307192, + "pairwise_causal_calibration_ece": 0.045251895335481074, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.10907919344361995, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 14.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.8182844519615173, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.1741162794419627 + }, + "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": 0.6, + "pairwise_causal_calibration_confidence": 0.5285102521950974, + "pairwise_causal_calibration_ece": 0.07148974780490258, + "pairwise_causal_calibration_pairs": 75.0, + "predicted_margin": 0.05173460799750952, + "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": -0.1741162794419627 + }, + "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": 0.7727272727272727, + "pairwise_causal_calibration_confidence": 0.5338099460898557, + "pairwise_causal_calibration_ece": 0.23891732663741705, + "pairwise_causal_calibration_pairs": 110.0, + "predicted_margin": 0.10442126551685532, + "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": -0.1741162794419627 + }, + "608016290/train0": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5428571428571428, + "pairwise_causal_calibration_confidence": 0.5574158017414859, + "pairwise_causal_calibration_ece": 0.02535410150173202, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.40694674604777425, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 11.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 1.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08577718626706465 + }, + "608016290/train1": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.49523809523809526, + "pairwise_causal_calibration_confidence": 0.5568088448070179, + "pairwise_causal_calibration_ece": 0.06157074956892268, + "pairwise_causal_calibration_pairs": 105.0, + "predicted_margin": 0.22198156017266713, + "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": 1.274298369884491, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08577718626706465 + }, + "608016290/train2": { + "base_success": 0.0, + "base_utility": 0.725701630115509, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5, + "pairwise_causal_calibration_confidence": 0.56209932221557, + "pairwise_causal_calibration_ece": 0.0620993222155701, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3766943789955354, + "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": 0.6789124011993408, + "selected_utility_gain_over_base": -0.04678922891616821, + "selector_regret": 0.03281521797180176, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2882723808288574, + "tau": -0.08577718626706465 + }, + "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": 0.6458333333333333, + "pairwise_causal_calibration_confidence": 0.5656592664914735, + "pairwise_causal_calibration_ece": 0.11750606320869803, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.5010382071158358, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04469547560438514, + "selected_candidate_index": 11.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.009900043485686183, + "selected_utility_gain_over_base": 0.004801700764801353, + "selector_regret": 0.034795432118698955, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12161866808310151, + "tau": -0.1741162794419627 + }, + "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": 0.35833333333333334, + "pairwise_causal_calibration_confidence": 0.5628388482930173, + "pairwise_causal_calibration_ece": 0.20964815489410538, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4605518879781133, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.035827874671667814, + "selected_candidate_index": 10.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.006450424436479807, + "selected_utility_gain_over_base": 0.0013520817155949771, + "selector_regret": 0.029377450235188007, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.12975450651720166, + "tau": -0.1741162794419627 + }, + "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": 0.6458333333333334, + "pairwise_causal_calibration_confidence": 0.5625202429462706, + "pairwise_causal_calibration_ece": 0.10372314663868182, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.3514059887171197, + "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": -0.1741162794419627 + }, + "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": 0.38333333333333336, + "pairwise_causal_calibration_confidence": 0.5210271884357356, + "pairwise_causal_calibration_ece": 0.13769385510240223, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.4860706638275267, + "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": -0.1741162794419627 + }, + "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": 0.675, + "pairwise_causal_calibration_confidence": 0.5170756938619067, + "pairwise_causal_calibration_ece": 0.15792430613809338, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.11583401883298983, + "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": -0.1741162794419627 + }, + "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": 0.48333333333333334, + "pairwise_causal_calibration_confidence": 0.5151375785964762, + "pairwise_causal_calibration_ece": 0.031804245263142905, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": 0.18578510943339843, + "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": -0.1741162794419627 + }, + "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": 0.5277777777777778, + "pairwise_causal_calibration_confidence": 0.5624770111779376, + "pairwise_causal_calibration_ece": 0.03469923340015974, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.17094865469784903, + "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": 0.22913040029030945 + }, + "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": 0.4444444444444444, + "pairwise_causal_calibration_confidence": 0.5653526989123352, + "pairwise_causal_calibration_ece": 0.1209082544678908, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.2625572115443, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8128758072853088, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11176733672618866, + "selected_utility_gain_over_base": -0.051239341497421265, + "selector_regret": 0.7011084705591202, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.1871241927146912, + "tau": 0.22913040029030945 + }, + "655884605/train2": { + "base_success": 0.0, + "base_utility": 0.16300667822360992, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5555555555555556, + "pairwise_causal_calibration_confidence": 0.56240741403742, + "pairwise_causal_calibration_ece": 0.03641674663594687, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": 0.23713238846053236, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21447330713272095, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09131735563278198, + "selected_utility_gain_over_base": -0.07168932259082794, + "selector_regret": 0.12315595149993896, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.785526692867279, + "tau": 0.22913040029030945 + }, + "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": 0.8235294117647058, + "pairwise_causal_calibration_confidence": 0.5545676976677929, + "pairwise_causal_calibration_ece": 0.26896171409691294, + "pairwise_causal_calibration_pairs": 17.0, + "predicted_margin": -0.10240791384393264, + "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": -0.1741162794419627 + }, + "672328262/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.45714285714285713, + "pairwise_causal_calibration_confidence": 0.5403743645771392, + "pairwise_causal_calibration_ece": 0.0832315074342818, + "pairwise_causal_calibration_pairs": 35.0, + "predicted_margin": -0.23155797487397917, + "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": -0.1741162794419627 + }, + "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": 0.43333333333333335, + "pairwise_causal_calibration_confidence": 0.5452442045537841, + "pairwise_causal_calibration_ece": 0.1119108712204508, + "pairwise_causal_calibration_pairs": 30.0, + "predicted_margin": -0.166628216752181, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.04013325646519661, + "selected_utility_gain_over_base": -1.9598667435348034, + "selector_regret": 1.9598667435348034, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.1741162794419627 + }, + "701667486/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5476190476190477, + "pairwise_causal_calibration_confidence": 0.5311290004714053, + "pairwise_causal_calibration_ece": 0.01649004714764235, + "pairwise_causal_calibration_pairs": 42.0, + "predicted_margin": 0.27106178902715317, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 10.0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": 0.22913040029030945 + }, + "701667486/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5333844210863801, + "pairwise_causal_calibration_ece": 0.07375843605647703, + "pairwise_causal_calibration_pairs": 84.0, + "predicted_margin": 0.2505788547291904, + "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": 0.22913040029030945 + }, + "701667486/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5185185185185185, + "pairwise_causal_calibration_confidence": 0.5311189080220702, + "pairwise_causal_calibration_ece": 0.01260038950355169, + "pairwise_causal_calibration_pairs": 54.0, + "predicted_margin": 0.31082243661779413, + "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": 0.22913040029030945 + }, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5223126728351409, + "pairwise_causal_calibration_ece": 0.022312672835140934, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.11408826824860728, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10839563608169556, + "selected_candidate_index": 14.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.061248183250427246, + "selected_utility_gain_over_base": -1.9387518167495728, + "selector_regret": 0.04714745283126831, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.17026761174201965, + "tau": -0.1741162794419627 + }, + "747260437/train1": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6333333333333333, + "pairwise_causal_calibration_confidence": 0.5248794662055539, + "pairwise_causal_calibration_ece": 0.10845386712777938, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.27266683578349143, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06328952312469482, + "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": 0.21537372469902039, + "tau": -0.1741162794419627 + }, + "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": 0.5916666666666667, + "pairwise_causal_calibration_confidence": 0.5203117574231598, + "pairwise_causal_calibration_ece": 0.07135490924350685, + "pairwise_causal_calibration_pairs": 120.0, + "predicted_margin": -0.08183742266671763, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.12510187923908234, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.06438694149255753, + "selected_utility_gain_over_base": -1.9356130585074425, + "selector_regret": 0.06071493774652481, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15356136858463287, + "tau": -0.1741162794419627 + }, + "750448296/train0": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.6666666666666666, + "execute_generated": 0.6666666666666666, + "fallback_rate": 0.3333333333333333, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.5083333333333334, + "pairwise_causal_calibration_confidence": 0.5222439903086803, + "pairwise_causal_calibration_ece": 0.05973654993182551, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": 0.014221736757998019, + "proposal_oracle_success": 0.3333333333333333, + "proposal_oracle_utility": 0.7526603279014429, + "selected_candidate_index": 4.666666666666667, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11766380071640015, + "selected_utility_gain_over_base": 7.065137227376302e-05, + "selector_regret": 0.6349965271850427, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.3333333333333333, + "support_gap": 0.6553361676633358, + "tau": -0.1741162794419627 + }, + "750448296/train1": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5621794871794873, + "pairwise_causal_calibration_confidence": 0.5206535156871791, + "pairwise_causal_calibration_ece": 0.1144300659008682, + "pairwise_causal_calibration_pairs": 101.66666666666667, + "predicted_margin": -0.19773588166013986, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3705032741030057, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.109618179500103, + "selected_utility_gain_over_base": -0.007974969844023386, + "selector_regret": 1.2608850946029027, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.03749322146177292, + "tau": -0.1741162794419627 + }, + "750448296/train2": { + "base_success": 0.0, + "base_utility": 0.11759314934412639, + "coverage": 0.6666666666666666, + "execute_generated": 0.6666666666666666, + "fallback_rate": 0.3333333333333333, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.5559523809523809, + "pairwise_causal_calibration_confidence": 0.5240722010838824, + "pairwise_causal_calibration_ece": 0.12386393190756012, + "pairwise_causal_calibration_pairs": 108.0, + "predicted_margin": -0.15253919216558867, + "proposal_oracle_success": 0.6666666666666666, + "proposal_oracle_utility": 1.3504486680030823, + "selected_candidate_index": 10.333333333333334, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09164678926269214, + "selected_utility_gain_over_base": -0.02594636008143425, + "selector_regret": 1.25880187874039, + "success_selector_gap": 0.6666666666666666, + "success_support_gap": 0.0, + "support_gap": 0.05754782756169637, + "tau": -0.1741162794419627 + }, + "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": 0.5313034188034188, + "pairwise_causal_calibration_confidence": 0.5288627189688364, + "pairwise_causal_calibration_ece": 0.13295795580777578, + "pairwise_causal_calibration_pairs": 119.25, + "predicted_margin": -0.03602515072385583, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0424291878007352, + "selected_candidate_index": 7.5, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5599935371428728, + "selected_utility_gain_over_base": -0.013205804862082005, + "selector_regret": 0.4824356506578624, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.06562284985557199, + "tau": -0.1741162794419627 + }, + "767747424/train1": { + "base_success": 0.25, + "base_utility": 0.5731993420049548, + "coverage": 0.75, + "execute_generated": 0.75, + "fallback_rate": 0.25, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.5907993966817496, + "pairwise_causal_calibration_confidence": 0.5276314238144284, + "pairwise_causal_calibration_ece": 0.14253701144399372, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": -0.003923417442009747, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0629530912265182, + "selected_candidate_index": 8.5, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5966028785333037, + "selected_utility_gain_over_base": 0.023403536528348923, + "selector_regret": 0.4663502126932144, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.045098946429789066, + "tau": -0.1741162794419627 + }, + "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": 0.5065112583494937, + "pairwise_causal_calibration_confidence": 0.5305845460281344, + "pairwise_causal_calibration_ece": 0.15317336838473566, + "pairwise_causal_calibration_pairs": 119.0, + "predicted_margin": 0.0115011599427335, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0671670949086547, + "selected_candidate_index": 5.25, + "selected_success": 0.25, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5675513362511992, + "selected_utility_gain_over_base": -0.0056480057537555695, + "selector_regret": 0.49961575865745544, + "success_selector_gap": 0.25, + "success_support_gap": 0.0, + "support_gap": 0.04088494274765253, + "tau": -0.1741162794419627 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/metrics_by_task.json b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/metrics_by_task.json new file mode 100644 index 0000000000000000000000000000000000000000..85fceb4376650f797e19c739f1a5154f360bef54 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_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": 0.6666666666666666, + "execute_generated": 0.6666666666666666, + "fallback_rate": 0.3333333333333333, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.5333333333333333, + "pairwise_causal_calibration_accuracy": 0.5353055242569309, + "pairwise_causal_calibration_confidence": 0.5334598489552488, + "pairwise_causal_calibration_ece": 0.08576162490103227, + "pairwise_causal_calibration_pairs": 94.8, + "predicted_margin": 0.023658521728357904, + "proposal_oracle_success": 0.6, + "proposal_oracle_utility": 1.4122399787108104, + "selected_candidate_index": 7.2, + "selected_success": 0.26666666666666666, + "selected_success_gain_over_base": 0.06666666666666667, + "selected_utility": 0.9396337101856868, + "selected_utility_gain_over_base": -0.010236828525861105, + "selector_regret": 0.47343848447004955, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.4, + "support_gap": 0.5877600212891897, + "tau": -0.08577718626706465 + }, + "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": 0.3703703703703704, + "pairwise_causal_calibration_confidence": 0.5425410371755132, + "pairwise_causal_calibration_ece": 0.17217066680514279, + "pairwise_causal_calibration_pairs": 36.0, + "predicted_margin": -0.42692880044302434, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13425676027933756, + "selected_candidate_index": 1.3333333333333333, + "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": 0.42252559087721425 + }, + "PullCube-v1": { + "base_success": 0.36, + "base_utility": 0.7751704612234608, + "coverage": 0.7733333333333333, + "execute_generated": 0.7733333333333333, + "fallback_rate": 0.22666666666666666, + "hidden_chart_oracle_success": 0.52, + "hidden_chart_oracle_utility": 1.1467878860235214, + "outcome_ptr": 0.4666666666666667, + "pairwise_causal_calibration_accuracy": 0.5394044361388536, + "pairwise_causal_calibration_confidence": 0.532007783280179, + "pairwise_causal_calibration_ece": 0.1182675802568201, + "pairwise_causal_calibration_pairs": 92.32, + "predicted_margin": 0.05246869761637144, + "proposal_oracle_success": 0.5333333333333333, + "proposal_oracle_utility": 1.1093241602430741, + "selected_candidate_index": 6.386666666666667, + "selected_success": 0.32, + "selected_success_gain_over_base": -0.04, + "selected_utility": 0.6831284697043399, + "selected_utility_gain_over_base": -0.09204199151912083, + "selector_regret": 0.4520184968970716, + "success_selector_gap": 0.22666666666666666, + "success_support_gap": 0.02666666666666667, + "support_gap": 0.1075624164690574, + "tau": -0.17411627944196265 + }, + "PushCube-v1": { + "base_success": 0.16666666666666666, + "base_utility": 0.3924041787783305, + "coverage": 0.7777777777777778, + "execute_generated": 0.7777777777777778, + "fallback_rate": 0.2222222222222222, + "hidden_chart_oracle_success": 0.8333333333333334, + "hidden_chart_oracle_utility": 1.7015800600250561, + "outcome_ptr": 0.8333333333333334, + "pairwise_causal_calibration_accuracy": 0.5261888915984856, + "pairwise_causal_calibration_confidence": 0.5156592220791224, + "pairwise_causal_calibration_ece": 0.10017848048000332, + "pairwise_causal_calibration_pairs": 97.44444444444444, + "predicted_margin": 0.31253125927801373, + "proposal_oracle_success": 0.8888888888888888, + "proposal_oracle_utility": 1.7857814149724112, + "selected_candidate_index": 4.5, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.3333333333333333, + "selected_utility": 1.0588933833771281, + "selected_utility_gain_over_base": 0.6664892045987977, + "selector_regret": 0.726888031595283, + "success_selector_gap": 0.3888888888888889, + "success_support_gap": 0.1111111111111111, + "support_gap": 0.21421858502758873, + "tau": 0.059256325887868444 + }, + "StackCube-v1": { + "base_success": 0.2727272727272727, + "base_utility": 0.7763947045261209, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.6363636363636364, + "pairwise_causal_calibration_accuracy": 0.49359663297649975, + "pairwise_causal_calibration_confidence": 0.5413161148546212, + "pairwise_causal_calibration_ece": 0.10435404157033219, + "pairwise_causal_calibration_pairs": 67.45454545454545, + "predicted_margin": 0.14918837663928425, + "proposal_oracle_success": 0.5151515151515151, + "proposal_oracle_utility": 1.1567951707225856, + "selected_candidate_index": 5.393939393939394, + "selected_success": 0.30303030303030304, + "selected_success_gain_over_base": 0.030303030303030304, + "selected_utility": 0.7681677698185949, + "selected_utility_gain_over_base": -0.008226934707526005, + "selector_regret": 0.39222038604996423, + "success_selector_gap": 0.21212121212121213, + "success_support_gap": 0.48484848484848486, + "support_gap": 0.8432048292774142, + "tau": 0.22913040029030962 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/split_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/split_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..865ad41e869c1bfbc57e4db65ed79e91e6d93081 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_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/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/table.tex b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/table.tex new file mode 100644 index 0000000000000000000000000000000000000000..64cb521b9a731fcc4595c87b24e5b4e3a7ebda02 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_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 & 0.6458 & 0.3542 & 0.2917 & 0.3264 & 0.5694 & 0.5486 & 0.2014 & 0.2500 & 0.0188 \\ +\bottomrule +\end{tabular} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/train.log b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_task_envclip_k16_train_to_test/train.log new file mode 100644 index 0000000000000000000000000000000000000000..6d8fbd7ea557aaa341224dd1d38c95f5ed722163 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_score_context_chartcompat_utility_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=100.0 +tau={"LiftPegUpright-v1": -0.085777, "PickCube-v1": 0.422526, "PullCube-v1": -0.174116, "PushCube-v1": 0.059256, "StackCube-v1": 0.22913, "__global__": -0.100999} +calibration_input=runs/ctt_base_context_obs_train_cal_envclip_k16_rollout_comparison/combined_measured_candidates.json